UFDC Home myUFDC Home  |   Help
<%BANNER%>

# Conductances in the Two-Impurity Anderson Model

## Material Information

Title: Conductances in the Two-Impurity Anderson Model
Physical Description: 1 online resource (143 p.)
Language: english
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2008

## Subjects

Subjects / Keywords: anderson, competition, computational, condensed, conductance, dot, dqd, function, impurity, kondo, magnetic, matter, metal, microscope, microscopy, nrg, parallel, parallelized, phase, physics, qpt, quantum, scalapack, scanning, solid, spectral, state, stm, susceptibility, theory, transition, transport, tunneling, wilson
Physics -- Dissertations, Academic -- UF
Genre: Physics thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

## Notes

Abstract: In a number of systems of interest that involve magnetic atoms and their analogous quantum dot manifestations, there arises a competition between Kondo screening and various types of magnetic ordering (direct and induced). This competition can be studied in detail using scanning tunneling microscopy to probe clusters of magnetic adatoms on metallic surfaces and has direct implications for systems of double quantum dots. In both of these cases, an observable quantity of interest is the electrical conductance, which can be calculated by applying the numerical renormalization group to the two-impurity Anderson model. Depending on their separation and the strength of their exchange interaction, pairs of magnetic adatoms may exhibit ferromagnetic or antiferromagnetic alignment of the impurity local moments, in some cases leading to a two-stage Kondo screening process, effectively isolated impurity screening, or a complete suppression of the Kondo effect. These behaviors have different signatures in the differential conductance. A class of double quantum dot devices composed of a Kondo-like dot and a weakly interacting dot is predicted to display a splitting of the Kondo resonance and a pair of quantum phase transitions. These behaviors introduce unique signatures in the device conductance when the level energy on either dot is varied by tuning the appropriate gate voltage. This work demonstrates that double quantum dots can provide a controlled experimental setting in which to study quantum phase transitions in a strongly correlated system.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Thesis: Thesis (Ph.D.)--University of Florida, 2008.

## Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2008
System ID: UFE0022090:00001

## Material Information

Title: Conductances in the Two-Impurity Anderson Model
Physical Description: 1 online resource (143 p.)
Language: english
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2008

## Subjects

Subjects / Keywords: anderson, competition, computational, condensed, conductance, dot, dqd, function, impurity, kondo, magnetic, matter, metal, microscope, microscopy, nrg, parallel, parallelized, phase, physics, qpt, quantum, scalapack, scanning, solid, spectral, state, stm, susceptibility, theory, transition, transport, tunneling, wilson
Physics -- Dissertations, Academic -- UF
Genre: Physics thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

## Notes

Abstract: In a number of systems of interest that involve magnetic atoms and their analogous quantum dot manifestations, there arises a competition between Kondo screening and various types of magnetic ordering (direct and induced). This competition can be studied in detail using scanning tunneling microscopy to probe clusters of magnetic adatoms on metallic surfaces and has direct implications for systems of double quantum dots. In both of these cases, an observable quantity of interest is the electrical conductance, which can be calculated by applying the numerical renormalization group to the two-impurity Anderson model. Depending on their separation and the strength of their exchange interaction, pairs of magnetic adatoms may exhibit ferromagnetic or antiferromagnetic alignment of the impurity local moments, in some cases leading to a two-stage Kondo screening process, effectively isolated impurity screening, or a complete suppression of the Kondo effect. These behaviors have different signatures in the differential conductance. A class of double quantum dot devices composed of a Kondo-like dot and a weakly interacting dot is predicted to display a splitting of the Kondo resonance and a pair of quantum phase transitions. These behaviors introduce unique signatures in the device conductance when the level energy on either dot is varied by tuning the appropriate gate voltage. This work demonstrates that double quantum dots can provide a controlled experimental setting in which to study quantum phase transitions in a strongly correlated system.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Thesis: Thesis (Ph.D.)--University of Florida, 2008.

## Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2008
System ID: UFE0022090:00001

Full Text
xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID E20101204_AAAAAT INGEST_TIME 2010-12-04T11:13:40Z PACKAGE UFE0022090_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES
FILE SIZE 22686 DFID F20101204_AAAQZT ORIGIN DEPOSITOR PATH lane_w_Page_006.QC.jpg GLOBAL false PRESERVATION BIT MESSAGE_DIGEST ALGORITHM MD5
225044827aaef4b5eebca2fc0b4cb0df
SHA-1
3fd8fed7a848ae453a8becc3d4279ed6fa29b9c5
24348 F20101204_AAAQDA lane_w_Page_116.QC.jpg
b5f4fc3389b79d3a73cc59f84426cd8d
23924e15d30ec3f617e93c9f0cb12eafbec4391d
864581 F20101204_AAAQCM lane_w_Page_099.jp2
2022163129081832727e9e223c9097a2
9d9e3e18463fc596d8fd7741853cb748e7c997ac
62265 F20101204_AAAPXG lane_w_Page_018.jpg
e7d7170edd2c6b38d543f5851e2026c1
cce7a11e164501292373b943080102bf9aa2ec56
873786 F20101204_AAAQBY lane_w_Page_045.jp2
bf75a7f9739f6cf5ff8d48e94fbd9d5f
d81f6f08f5a771960ae78a3be50e5e74ee7b54e3
806722 F20101204_AAAPWR lane_w_Page_070.jp2
2cd458d89863004828854c86b2de9355c9a3eb63
4640 F20101204_AAAQZU lane_w_Page_006thm.jpg
a0d84ee0b695104d519a217b09705261
72330 F20101204_AAAQDB lane_w_Page_060.jpg
3cbef085d4e3978361c21b7e9e324124
43f9b4207cbc09a47fd92412ccc320a281dd0ee4
31887 F20101204_AAAQCN lane_w_Page_030.pro
db549668ceb22daa2a0cceca6e576c25
27f21531b9d2ff0748173a94452837bc1b1dcdfc
5687 F20101204_AAAPXH lane_w_Page_062thm.jpg
3381d83ca8ecfe29ab54fe330b513944
25e59fafff07e84a79865d0fb616f211fc7449c5
6210 F20101204_AAAQBZ lane_w_Page_114thm.jpg
5d4884148358e7a4945cbbd9f2fd35d41ca509ac
25271604 F20101204_AAAPWS lane_w_Page_089.tif
21033 F20101204_AAAQZV lane_w_Page_007.QC.jpg
05ba269b6b2867afa62711884f78495f
15901 F20101204_AAAQDC lane_w_Page_067.QC.jpg
404227065e45762629ae2917eb4ee466
639e8dbecb5788c497388511dd6bdf6b539a1baf
90212 F20101204_AAAQCO lane_w_Page_054.jpg
d832fd64606a63d9dfda4bf4529b2440
204c89a32e6616434f3ea905cbfa6e00d7e3edb7
1756 F20101204_AAAPXI lane_w_Page_070.txt
6fa6d0aa3561c88a3124546b12df1fa202e20d38
40660 F20101204_AAAPWT lane_w_Page_119.pro
b1a9424d81be13451a2b3c3728d4bbcf
bc37f28715f719529491f17fff8857c96726a7c5
4168 F20101204_AAAQZW lane_w_Page_007thm.jpg
69562 F20101204_AAAQDD lane_w_Page_026.jpg
c2f21c3947006088ef79cc0b561bca8a
b06ac4cb347663e2806f8e7a32aa96b16d4dccbd
1848 F20101204_AAAQCP lane_w_Page_058.txt
280fd96a17768deff09e45f541746b23
f60237fddaec04d2f599cd7298358d9a7c163344
4477 F20101204_AAAPXJ lane_w_Page_085thm.jpg
4c83049d6ef359e920cc4ebb7c93d9de
be71dbab0afa27621221de19bf05e129507bc223
90227 F20101204_AAAPWU lane_w_Page_080.jpg
ac7c2ceec7ee4f59d608ab8ff9e0d97e
c418c46a5fdee5f894918960268a58d3195c91b9
5134 F20101204_AAAQZX lane_w_Page_008thm.jpg
86e3167bcedfe1de5787d949740a2d34
db84f8783ce97b9a054016132d40bf7694f5b0d1
90833 F20101204_AAAQDE lane_w_Page_006.jpg
2dd86f735e5e4190bde846cbc596a9ae76422eda
1051943 F20101204_AAAQCQ lane_w_Page_134.jp2
83f05cf2f871f055e39fbc5f4d5de27c
a11bc35222919bd8fd0a8938abbbb8ef9620d948
67041 F20101204_AAAPXK lane_w_Page_138.jpg
c6b7da31c7f088529a66b537d0755708
a874b36a9e90029e7ce02b9e2307ac01f798f11b
57307 F20101204_AAAPWV lane_w_Page_008.pro
fdbe0ebe6b0ef16e05d56f5f349a57f8
26408 F20101204_AAAQZY lane_w_Page_009.QC.jpg
2699f5f30ca09195faca663f83403a34926bddc6
F20101204_AAAQDF lane_w_Page_007.tif
8417d92e0a4be2c6866c8a8437912c35
f76fbc2108b94f0a8a360057533f43b01f9c0fbe
346781 F20101204_AAAQCR lane_w_Page_100.jp2
6705a0e25212be6e9327ced0cc1449ce
6854682e708a50ecd44360aba4707938c4a83055
4840 F20101204_AAAPXL lane_w_Page_026thm.jpg
6a55b0825495313dcd9003a46a92f12a
17383d3250cb745fbdde423e2e132128f70e1a0c
764469 F20101204_AAAPWW lane_w_Page_058.jp2
b10bd6bcfa86d549c7ccd9367a245187
2461364e6bf8ab4b20718bb272198083387539de
5255 F20101204_AAAQZZ lane_w_Page_009thm.jpg
797984fbbea7611f0689598249212dee
b83e1ac849664542e8059dffc20c027538710d6c
26328 F20101204_AAAQDG lane_w_Page_099.pro
2d2eb62e6f63ffb076b59c47ab4c400e
f694bde753dc8078c75665a7111cae7b44a922e6
2335 F20101204_AAAPYA lane_w_Page_134.txt
dd848214a880d1387e260017ae6ed953
21ea3691892f60aa369cb777fa2538e7aeddeccc
50644 F20101204_AAAQCS lane_w_Page_140.pro
468f2428d4081e28405bc5009320588d
0c8fd20f2896c9ed0c6e193513eac372cfb0abe4
1003508 F20101204_AAAPXM lane_w_Page_046.jp2
a97b1bdbd52cd89390a0252112b1d22c
47cc07e55f81656e97c98e1c9b8105e4102de727
1051983 F20101204_AAAPWX lane_w_Page_106.jp2
732fb14ca0539870c364f37948f9ebc6e2360cd8
843120 F20101204_AAAQDH lane_w_Page_059.jp2
caac537bf99226730159cd874565c284
8638560a9f393f1242ee01e95f8dcde717064670
57480 F20101204_AAAPYB lane_w_Page_054.pro
f31409310a4e4f56cbb0f6d6023e84e2
1743 F20101204_AAAQCT lane_w_Page_025.txt
eb7493be6b299fd48bdac32ca56da0c3
1937 F20101204_AAAPXN lane_w_Page_084.txt
f76450e4481156259d1a8954f1c41344
5d0a727d82896bbeef5aca5f4545b9054e55d0c3
1051968 F20101204_AAAPWY lane_w_Page_048.jp2
16647001a7294b4c46502ac6afb5f247
a677727dbff30035a7ac31bc4ddb32a8d87bcbc8
1989 F20101204_AAAQDI lane_w_Page_073.txt
44c2f7abfe4ea54103fa525778eeeb5f
3942 F20101204_AAAPYC lane_w_Page_081thm.jpg
e132b5b1c3483f598e6eb2397a42a446
e5fab7211f33f431870414c6e2d5323ebc9e9f06
2220 F20101204_AAAQCU lane_w_Page_137.QC.jpg
89a6004216fb8fd97bbb062b4ab49d29
76b66d048b53070eeb787de358c1c5b99336b1cf
22754 F20101204_AAAPXO lane_w_Page_011.QC.jpg
dd0bfd9e0ee6d2ef883cbd025c9e365b54c54c48
5169 F20101204_AAAPWZ lane_w_Page_024thm.jpg
1f05f549401ec0cdf2abcff0d72d337d
ffe1fdd3d1135494b52d4578c125971214df67fe
5290 F20101204_AAAPYD lane_w_Page_099thm.jpg
681264ca4fb8caab124b0c9c240bb92f
9fa79a173164814086f8df609a11abfdd361b27c
846182 F20101204_AAAQCV lane_w_Page_117.jp2
f6cfbdb83df4ed18641ccd263d8e0069
944650c8b3ff8fc2360444975ecef655a62312ba
18776 F20101204_AAAPXP lane_w_Page_036.QC.jpg
28759152cdd2c016feaa1aeffa024568
5365654026a96a68c2de0ab4068807d4333e1a46
122743 F20101204_AAAQDJ lane_w_Page_080.jp2
400f79e2abd1ac76472c54c77d4c876a
2a4e926087b4e1846c73969fd58f741d65ac9436
56502 F20101204_AAAPYE lane_w_Page_103.pro
d173626d45876079491fafce333cf43e78f972fa
2095 F20101204_AAAQCW lane_w_Page_064.txt
40d4b5e03889ee7f741917ffe30c7f64
8c7c14df5118c72f07e6b2af46dc2c85fa5d9b91
60001 F20101204_AAAPXQ lane_w_Page_122.jpg
c4f45bc567c842fbd1353f080acc97d5
bc23c4187ab94acf80eab8647cb4f48dd9b1d1a2
51726 F20101204_AAAQDK lane_w_Page_064.pro
a07f385a3a5a50b1cda67b6148a1986d
40a0cfcc97de44ab0bb88a7123d502b21a0d14f9
5632 F20101204_AAAPYF lane_w_Page_014thm.jpg
55b84b272359b976ae7906fc139eb10e9b5bf4af
F20101204_AAAQCX lane_w_Page_022.tif
de7254ccb9587f00a0ce625c7c2268bc
98633ceef807c24bf29c128bd7021249164fb83b
F20101204_AAAQDL lane_w_Page_136.tif
6a3df3225d1136624ce2dafd3900699cabea4f08
23060 F20101204_AAAPYG lane_w_Page_130.QC.jpg
3549f65c9abbba38a7e3308d9b0ccf10
840e17545fd54fa0876c0f49ed7c3280e83d0423
55200 F20101204_AAAQCY lane_w_Page_086.jpg
fe43f5c05092fa8374843109b6f1642a
dc6f9608e47621e764390a76e586836273f50c26
77680 F20101204_AAAPXR lane_w_Page_007.jpg
5d144074d004f42191fedd1809061b23
836ef2403b738a1610e8363389d75dabfafb2eb7
28356 F20101204_AAAQEA lane_w_Page_077.QC.jpg
e11b3683ee098166e4fe393912a82507
1051950 F20101204_AAAQDM lane_w_Page_114.jp2
1c90f1be608554101aa6c68915cb0b74
956dd13745e5c17ca519dd8425277b3b157b188d
7470 F20101204_AAAPYH lane_w_Page_001.pro
4507111e28d94f5a7bdcb4f917314b8a
81310168af62fa59cffdd9e55dd17bc2f6b81c00
1057 F20101204_AAAQCZ lane_w_Page_033.txt
2379655af00b961522b7642b0763cd9d
62014002ae5bfc89be04bb10c348d817b4e0dcec
F20101204_AAAPXS lane_w_Page_057.tif
2c625e1e041cbdb22be26da4d9963305
1051966 F20101204_AAAQEB lane_w_Page_043.jp2
8a8941de2e171922c10f269e049a9902
35929532d8dce09c92751af9eb2a1d1aa1da8646
40294 F20101204_AAAQDN lane_w_Page_018.pro
240268c23197d479831204358923dcf3
14a7af9a549d53ce25f62666d9d96a37c2178920
F20101204_AAAPYI lane_w_Page_113.tif
ffef2f82bac80b688a46acdf392e2089
56403 F20101204_AAAPXT lane_w_Page_121.jpg
22132cbdb8071567e72cfc20c35a9dcdfdca9543
5715 F20101204_AAAQEC lane_w_Page_053thm.jpg
69cb98a0fd4195ab639fa3fc44b6470b76f24a83
24913 F20101204_AAAQDO lane_w_Page_048.QC.jpg
f71ca455b081fd0a726acf26618a7f75
d563026a25f29d154e235a07e156caec594fa10a
15929 F20101204_AAAPYJ lane_w_Page_132.QC.jpg
9410d035cc437c9c79f0ec6a450f63ac
fd20ebc007331804e546613c33a153a2a174723f
1051957 F20101204_AAAPXU lane_w_Page_039.jp2
69ea938c887e818b62a23ef1c176ab07
5e6214f46f7a192b01bb751082a0b48cfa2c9d27
20207 F20101204_AAAQED lane_w_Page_024.QC.jpg
0352a62133b219e7f5f444696dd3cca5c07e9ee9
1053954 F20101204_AAAQDP lane_w_Page_080.tif
e747e2352d204b37d2904526b1ef8977
857f7226a1c6c0e65c69c519a3de04cc6c13d319
19720 F20101204_AAAPYK lane_w_Page_099.QC.jpg
9ee8ee3a5ae3484e9c4477b1ca3c0c9df820a3ac
F20101204_AAAPXV lane_w_Page_103.tif
22f604868536d38387ed5fd03b137401
8fae2edb0fa8ce3944287cbf6617c64dcd04bdf3
36230 F20101204_AAAQEE lane_w_Page_123.pro
357fd716ea121c0b0cb5c352913a7a11
ba8f6a3f6251a1f95db4d0bdae992c802bb2d918
F20101204_AAAQDQ lane_w_Page_054.jp2
eb63b981c0d8542a527ae6071262b7d4
279d56b637a758247ed5c2a64d6f88e99d19900b
87096 F20101204_AAAPYL lane_w_Page_016.jpg
c7e47b2e623cd34fa816e3b5e0dddd0b
085537ed470918991e33b64a51ce9797e443b1a8
2196 F20101204_AAAPXW lane_w_Page_016.txt
3284728a605bce10fc0978a3885d38bb
8cfb746da8fa68eb8ba2898dc64f2a16ac637d24
1553 F20101204_AAAQEF lane_w_Page_115.txt
f793511c79da3074ee6b628ca726c020
5699cc43a50950a9aac8f21036723de6921e6018
81690 F20101204_AAAQDR lane_w_Page_141.jpg
23ffa6ee46377dff635bbe4eb4fde86e
87d626b48ded90852163cd49458cebe0d281b59a
1297 F20101204_AAAPYM lane_w_Page_030.txt
56839 F20101204_AAAPXX lane_w_Page_044.pro
4fe4ceda81b3e8af55b9f0925348caee
6ec1c5e2c3693cf77567bca23b70b4bab4877192
44412 F20101204_AAAQEG lane_w_Page_060.pro
21fe77c94b6ac852c33141e30168d98a
393b530222300232a1f9c1532d1db3857109f083
32150 F20101204_AAAPZA lane_w_Page_051.pro
e8ae3997a6589a1000823668edf2ea0f
aa595eff27a533468496356a9b796387dd74151c
F20101204_AAAQDS lane_w_Page_027.tif
14c591ea55c595af78d118c66040e8b3
4c8352231dd62f87a8a383584d8aa4118f558491
26941 F20101204_AAAPYN lane_w_Page_035.QC.jpg
8605e8554ccd671d88524d621760e66d
3a21cb2069a6b65a10e37be0435aa6ce0c276ea1
1387 F20101204_AAAPXY lane_w_Page_093.txt
279d62e97772bf0befee68d131ec10a1
5b68066b4c9836d24eb9bd4a7483cb64dff8d13b
6596 F20101204_AAAQEH lane_w_Page_037thm.jpg
751c0b40577b667673c75b351f7153bd
27742 F20101204_AAAPZB lane_w_Page_069.pro
fba4ce0fd33df1cb4bc10d7f4bc15a05
ecaaf044e67dd8ff167c3d0e4f661f20787c3cf9
670483 F20101204_AAAQDT lane_w_Page_085.jp2
3c7fab733ffa35ef887d3c451ffb4938
4b93505d07c261ffbffd0cbe1a0efb0e308c46eb
13516 F20101204_AAAPYO lane_w_Page_071.QC.jpg
3a2a2c81139d8d0c82f7f27ed3754350
2202 F20101204_AAAPXZ lane_w_Page_020.txt
c430cba0f28f3b71371eb9897297d99e
0b82f02bc104b61ca0d30a4a75562c17ca17a067
4413 F20101204_AAAQEI lane_w_Page_089thm.jpg
3294e14c4e376a5904887fe0d34b2c8e
31178ddf7de626d75dc8f4f9a0d5dc89f2d167f8
73440 F20101204_AAAPZC lane_w_Page_118.jpg
cae94a4eaeca65ac622749a039594ced
51f8cc49749e865a3e78ef01653c323f23e490f8
27747 F20101204_AAAQDU lane_w_Page_086.pro
981b7f1a276513033caabc38c7070fe9
1051980 F20101204_AAAPYP lane_w_Page_008.jp2
a8be50016d08d264088c70e513edb92784613330
F20101204_AAAQEJ lane_w_Page_015.tif
b0e8ef639d52b5f6cba00fedea137ee0
e0f203b7bcb50c4c7ba9b572c65189e799f838c1
F20101204_AAAPZD lane_w_Page_003.tif
fe56c8255ca2b1c32a3db0a81a8ee771
d8cb6124d98e18edb22647c97bcf21ba496cea84
4645 F20101204_AAAQDV lane_w_Page_068thm.jpg
8fd3fabf02f15d724730135ea1eb8c00
84869 F20101204_AAAPYQ lane_w_Page_040.jpg
3529ec034ab303970dbfb5debea12019851aaf59
F20101204_AAAQEK lane_w_Page_082.tif
36bbc15da2b14d5f63a420a10650a316
4fee9e40c4b0b9da9060881d9267a4caaffdc38d
17734 F20101204_AAAPZE lane_w_Page_079.QC.jpg
ca27f7847989921506eaa36e77552cfd43b4d079
F20101204_AAAQDW lane_w_Page_116.tif
a45174ffaf0574da778eed17d0c8632a
9cbb2e0efb86ccaa2b766178f729e7f3b29e9e8c
4753 F20101204_AAAPYR lane_w_Page_018thm.jpg
4af6954456904bb47684515176651bbf
9bd88bf6b5ba382a86696b4875fbbdf8dd9d6016
23846 F20101204_AAAQEL lane_w_Page_129.QC.jpg
546159563e302a58a931736fc29d5531
dd0e7c2ccc9fd7599af32b4a43576dbfba5f433b
38425 F20101204_AAAPZF lane_w_Page_084.pro
d3e59c73dd6e47a1aef088276c631db8
4fff69e4260d5376201b84a134ffe2b076869ea3
57979 F20101204_AAAQDX lane_w_Page_078.pro
bcccc3e7cd86fd8e943d103b85a484d8
abd01378009f02b95de46ba3f73143f5945a9b33
1051971 F20101204_AAAQFA lane_w_Page_020.jp2
cee74c7813eea14fcb3e4ef0e65bca91
6dbd023aac045e202422a6d57a8613294dea8e1c
1606 F20101204_AAAQEM lane_w_Page_131.txt
cb457935666ab5a71df59d5b544f1039
1ab1125028904345fee3bf1ddb813f6178e9fe03
F20101204_AAAPZG lane_w_Page_050.tif
faa65299788c09f99da4195f3394f3bd
b0432d6b4921e5149fb8aa487789800530db24ab
90103 F20101204_AAAQDY lane_w_Page_133.jpg
3d6b13a36589805ca05fe6c78e6c348c
79aeedaaa879bbdba404ed13c4991e3b4242f40b
24505 F20101204_AAAPYS lane_w_Page_062.QC.jpg
ef75fa7da010ba2fa2208ca64946bca9
40cd29e05fcc7c501fb9db85ce6967330f607776
F20101204_AAAQFB lane_w_Page_063.jp2
152d49accc8f659e4b9b20b05b8352df
9af71a07789d673d4515eb10a10ffbb602626379
27307 F20101204_AAAQEN lane_w_Page_095.QC.jpg
a9a9b317eb4dc52de9afaa491d7d07d3
126f8b62c54bcb6d84380c68cb471d9bab3b4e8e
4763 F20101204_AAAPZH lane_w_Page_021thm.jpg
4d5705e2716394a58140b2ffd52d2c66
28d4467a764c02656328d9ba39cbe56925c6ea4d
84003 F20101204_AAAQDZ lane_w_Page_008.jpg
4f017126767aeb98c6268335c2d9ef82
2f0b7693f117a3a736e4331db2f7fe54c08b293f
5325 F20101204_AAAPYT lane_w_Page_039thm.jpg
afe0980ca2f7439f1eab755206a7a94854c35a12
20890 F20101204_AAAQFC lane_w_Page_105.QC.jpg
ffacc39a804aea75228a8a73be8d9718
aa34835275045a509a849771f267ace34bcec320
5842 F20101204_AAAQEO lane_w_Page_041thm.jpg
556a57d6cd8b469a3230f3337151cf8d
ab44a5012edd49bf9ec9aa3cd10414e14ea29518
20868 F20101204_AAAPZI lane_w_Page_132.pro
acc742b9d0507e26d765106d846c1712
2ac5765ae6bfb732883640c073cc25af17368580
F20101204_AAAPYU lane_w_Page_074.tif
07a3a4da9336f140be5e170e1d32cea6
218acbf5903e6c323c10f5589379e242951647d0
22319 F20101204_AAAQFD lane_w_Page_001.jp2
89bb2780ace46373791cb1410d28e009
58b9490979bb884b1f43e793bef46a411fd2349a
43707 F20101204_AAAQEP lane_w_Page_069.jpg
7e9d0ea97c85b977cb3a44220f4ecde8
7c8518fa14ae52bff128677b47f3bf3582171733
76159 F20101204_AAAPZJ lane_w_Page_139.jpg
873ae68b7d4143c53eef5ac3bfe952c1
c73f970ec6cb311662d9f57cd15dff821040fa6b
64088 F20101204_AAAPYV lane_w_Page_131.jpg
9f11daeed3666818be8a1f6e78c563b5af5203d8
26286 F20101204_AAAQFE lane_w_Page_012.QC.jpg
d548231304b089378f07cd7760a7b038
702c5777b765c8034d45253af249e85ca45038be
1051967 F20101204_AAAQEQ lane_w_Page_009.jp2
1b23d879effafbf5995b510d6c8ccdd9
862184 F20101204_AAAPZK lane_w_Page_013.jp2
0955e2b0431aab4de1cf714986fa5de9
2850f3d1df4c2f2ea0017cb864e6f836232e2a74
1889 F20101204_AAAPYW lane_w_Page_125.txt
0eb8b400d3c8f175c02156f3c3debb4d
73f1a73cc7d85446cf23047fa007c19ac3ac4333
F20101204_AAAQFF lane_w_Page_087.tif
c962018f6099488fce1ac3b8ed4d81aa
bef60739b672786084e6cc85c99f58e01e063658
5943 F20101204_AAAQER lane_w_Page_016thm.jpg
1d81bbc197a46175f3b4e5484482546cdea31c05
1599 F20101204_AAAPZL lane_w_Page_046.txt
3b718748331f8419ed50ca3f78f8b31e
85b6551ac3ce0a7446440faa2deb076a8247d87d
1252 F20101204_AAAPYX lane_w_Page_085.txt
c2fe32d4027c5aa7da1c7072f6764147
e85bfe707058966464a7f09e662b9aeca5fb8549
18851 F20101204_AAAQFG lane_w_Page_059.QC.jpg
2d86201bd56c259aed3c94f1b4137164
c7f8aa94f3e1138eb7bb6bc8f69d482d7cf96654
27448 F20101204_AAAQES lane_w_Page_135.QC.jpg
00f7b7206d410604b1712e559c5c06f3
c090fde3616788ae881c7bb9c1a3501a51466faa
33804 F20101204_AAAPZM lane_w_Page_127.pro
b931fde199309f7faa9c95c4023df259
F20101204_AAAPYY lane_w_Page_086.tif
885cf7970faf603262d6e754f709e9db
43d8955a58b6c0a7a1e93f2b5c3226c8e15f0917
1393 F20101204_AAAQFH lane_w_Page_072.txt
0bb6caa77c7e7a0368c8fec93dc922a9
ac132bc544a4bc5174a7369b02ff076a4dc9e81f
5386 F20101204_AAAQET lane_w_Page_096thm.jpg
4cc97db5e511132bf6ee12b182a2de3d
109ee1602db71b98781b33cd13c44c777a34d501
70966 F20101204_AAAPZN lane_w_Page_019.jpg
bc26d47517f8b944bffe6854a8dfed32
F20101204_AAAPYZ lane_w_Page_040.jp2
826d26f0dc99d6915ccb480d731d222d
a4e3ebd50883b928d78b98fc58b1bc8eff1bac5e
10064 F20101204_AAAQFI lane_w_Page_137.jp2
567eea4849ce02ed49434522bf6a70da141bac28
48609 F20101204_AAAQEU lane_w_Page_104.pro
c5e1951ab0068bd7d0423af8f2ed42d4
449ecfba425a8d077dda87684e3c50e1726ec847
1377 F20101204_AAAPZO lane_w_Page_013.txt
24a53cddabe04fae2df5a81022bd3e01
69540 F20101204_AAAQFJ lane_w_Page_047.jpg
f8ccf6e65b43650128596199da5a1fe8185d505b
5716 F20101204_AAAQEV lane_w_Page_098thm.jpg
cab50c3b33ba0c3037f6542672f49f03
06025db81e3e2f0f04989af8178b74e39e747c80
18923 F20101204_AAAPZP lane_w_Page_028.QC.jpg
b5ee22878cba81339cfdebce22b1cb96
22133fdb69625ebd4a03a3f4d9303d0a29acfe34
2308 F20101204_AAAQFK lane_w_Page_095.txt
b44a53d1ae0e23f67c71c88202d505014e1077fa
6286 F20101204_AAAQEW lane_w_Page_116thm.jpg
bd091f4e929f6669a02cd7c821a7e7c1f265aac3
F20101204_AAAPZQ lane_w_Page_143.tif
2114d7a0fc70e92e149c91c463da7ca0
930be9c911c5114279ed3318f741b2e52e6833b1
766274 F20101204_AAAQFL lane_w_Page_121.jp2
ba81b7dd335b47351eebe39454f150750c012f4b
8239 F20101204_AAAQEX lane_w_Page_075.QC.jpg
dd8810ab2464f0789265db99be5f9a21
47a6db6a14f995fa95b81efe08b5b424461fe2ae
35684 F20101204_AAAPZR lane_w_Page_124.pro
3c106b96931951210f9163a0ea5c1255
7c89c59dbe7451ac53403e2682193d983a6e7098
5248 F20101204_AAAQFM lane_w_Page_030thm.jpg
1920 F20101204_AAAQEY lane_w_Page_017.txt
3fb3925427f36018129765a1d5c7ceca
a8db1124e4e35154905a8ceaa64560c431ab80eb
F20101204_AAAPZS lane_w_Page_135.tif
0569731eb07e33ec96939d61b101822f
49575967cfaaae76ab814a7cc0b8be9884890fc6
66103 F20101204_AAAQGA lane_w_Page_024.jpg
f23d4cb3ab5d7884ea4d73272dab170f
46f4b76b44432bae6916b0ccb1ea55db420be280
21124 F20101204_AAAQFN lane_w_Page_026.QC.jpg
727d7505b8fa814673eda0c620d0bf4d
fd1793baa791178253ff06b54dcf6174bce8c567
314517 F20101204_AAAQEZ lane_w_Page_056.jp2
11445a4168dd29bda1686917d99aaf19
4055 F20101204_AAAQGB lane_w_Page_003.jpg
7573432fa7a0c4c261475e10be091817
4546ebcb469ac95ff2e0645d677dbc7dfd3c29cf
52918 F20101204_AAAQFO lane_w_Page_010.jpg
1c97c28a30081202924a857272aa975c
2e629e64e51ed67a9d6f0bef347b591713a30d8f
27124 F20101204_AAAPZT lane_w_Page_078.QC.jpg
9b314798e07eaf8e28b27d6d00509d4a1b74b432
9647 F20101204_AAAQGC lane_w_Page_143.QC.jpg
ecc364998be0b91d86365df990cd8c65
d1b09dc46fcb761b94eebc356799eab354757bd1
22554 F20101204_AAAQFP lane_w_Page_091.QC.jpg
e6bfeabe6ee0ba42e64c7d87c6dc60ec
42b8627956776f5ea51a82d7e9cf9415080fa805
54592 F20101204_AAAPZU lane_w_Page_036.jpg
3fe9d79299d7ca5bfb01795cbd9c780f
eb220b35db684de5619e62eb52400d5dba69572a
6008 F20101204_AAAQGD lane_w_Page_020thm.jpg
d825cb81fa8a534f8a938283e6e0205f
17876 F20101204_AAAQFQ lane_w_Page_143.pro
8cc36b0e78f78b3c466fcd849671e551
fe6838da4b5f74f91ca18f7a483e8d3a7939ecf3
52347 F20101204_AAAPZV lane_w_Page_032.pro
b6b9ee47bbc3fbdc8e95282f27a66b45
3286b7cd6ffaa1aebf72bd49f64f831fd63354b2
5298 F20101204_AAAQGE lane_w_Page_120thm.jpg
159ffc2a9b05cea79d786c862383679b
b43c75310dc8c72edb5efab79dfcb44b4e2ab631
5825 F20101204_AAAQFR lane_w_Page_118thm.jpg
71f358b396407d97525ea94872cd98e6
2a78b61c56a2d5ef6c02e9cc352c77eee416e8be
18158 F20101204_AAAPZW lane_w_Page_021.QC.jpg
dc0eec309a14bc1f17c98cc083c8cdef
18249 F20101204_AAAQGF lane_w_Page_090.QC.jpg
8904c9b631d68f6bf58ac7c665dd9c57
7820fc238bbabf16ffde1cd27bba28ddc58a848f
4629 F20101204_AAAQFS lane_w_Page_058thm.jpg
fd070389b48a8d221f783490cc07e447
24f5cd6d63444a773d0e2223d91715e843c2f3aa
5798 F20101204_AAAPZX lane_w_Page_005.jpg
3fb0a77f40b42938e36a55c7beffe959
37721 F20101204_AAAQGG lane_w_Page_059.pro
d15d28168ae69f1873c89640b807a35d
6e659e0c03e410543b4898052026a5ba90f438a9
25214 F20101204_AAAQFT lane_w_Page_071.pro
841186302517bfcfe33caf555b451fc0
9179e49d7f5f3e2ef09b3caf5bbfe4f163a2bbd7
54474 F20101204_AAAPZY lane_w_Page_041.pro
c3047fbe0d8053dbbd1a1c84d1e8a67b
23411ddcc41ccfdfa52f3c12591714e287f1b26a
24385 F20101204_AAAQGH lane_w_Page_074.QC.jpg
38366 F20101204_AAAQFU lane_w_Page_142.jpg
02346c878828c5685f05488eb0b8890e
deb60436224e8cbe35effa1b20ff663dc76208b8
34565 F20101204_AAAPZZ lane_w_Page_046.pro
5c14d28df4c9885fc3303c2b21bb0106
0d737c18390e29d15917e378822e018935f90246
F20101204_AAAQGI lane_w_Page_006.tif
d76605e563a9042e08f3ea7053e7502e
79f20a312e793919428f8638166a39ffd49d07ed
5955 F20101204_AAAQFV lane_w_Page_031thm.jpg
3296abd82543049d9dd116e43306c461
a9c3d37b068efba59606677653a9b976463f1497
4663 F20101204_AAAQGJ lane_w_Page_086thm.jpg
ce33e492e7f866c5d8c85ac7ea46518e
85d1f5d77ab8f063a4bcf4ce4ef2b60bd390a9a2
88982 F20101204_AAAQFW lane_w_Page_136.jpg
2597489300eafb83bdd2cf4a218800d8
40912ac83032f002339d4d1d884ef76a44486cd8
51877 F20101204_AAAQGK lane_w_Page_141.pro
d878a361d0116a1c21b57f136d846a93
5feee4f16014570ac429d1c43d4df7ceb1f53769
2152 F20101204_AAAQFX lane_w_Page_143thm.jpg
68f149a03e383ce027f35191e5f672f8
73104 F20101204_AAAQGL lane_w_Page_110.jpg
b8db5b0d8ed33916d67754f132f69a1f
F20101204_AAAQFY lane_w_Page_140.tif
bfb36d7b8f6b3349b6e4df9cf3f7bc89307fb725
F20101204_AAAQHA lane_w_Page_058.tif
2778a9c5ae6344cc896ee51cbf97db98
df14e445ca90f560acffdf56f5b242f1ae464598
6320 F20101204_AAAQGM lane_w_Page_055thm.jpg
9c09da251d0260c3f53cae8abbc6bdef
d60c360b3c3d7d192e028d359e67aeb468857fd7
73664 F20101204_AAAQFZ lane_w_Page_091.jpg
71c72598b35cb31a30dc098fda51d4df
1f220f1559063ff04e33f4fa333bbd4881a3f54b
59606 F20101204_AAAQHB lane_w_Page_043.pro
c30675473d043b51601b08cdd10807938b1e1d8b
32804 F20101204_AAAQGN lane_w_Page_120.pro
7779c8e96efe2d72e334eb326bd284b9
7fe7337cdcb444a9a872791ff855569ea8c916ae
21150 F20101204_AAAQHC lane_w_Page_142.pro
6b2cf7dabbbae118e2bb5f37c1a79c75
83d817eb0520201ba6c794229499e444b80cf7d0
82110 F20101204_AAAQGO lane_w_Page_064.jpg
3976f22f38dc8a4f100372ae362a5db8
4f46077e4b9c9f8afa8128b12e5de1ebdf0b1b66
F20101204_AAAQHD lane_w_Page_099.tif
e577fa53344006f96e77c0e140c338cc
b38d2b81abf15face2a3e96966ed79bec86c3d42
F20101204_AAAQGP lane_w_Page_118.tif
6b86a946bbecb4d360ab72a7620123f3
c67113b5081bd0cb66ea1e4e77fe206e4a5f1d1c
1051952 F20101204_AAAQHE lane_w_Page_029.jp2
67568 F20101204_AAAQGQ lane_w_Page_126.jpg
9cd5cb21ce00253f2d9fb3feecce6af8
bf5042003df7aa78485856e69e7a61b944bb3d38
21140 F20101204_AAAQHF lane_w_Page_093.QC.jpg
09007efc4dab0065413391c240f98a99
736d73ddb38d9b736962e9e4f5585d28ef0d9247
23322 F20101204_AAAQGR lane_w_Page_092.QC.jpg
02e09bc620d3463af3392bc1a50357be
F20101204_AAAQHG lane_w_Page_104.tif
4dd635d6b95485fb91135301f8a183f5
eba61b88cd08b5b79a30802a2dc4a4032c920b54
2287 F20101204_AAAQGS lane_w_Page_078.txt
6ba5a6ea76453ca894df19a8c83e7517
247f9dddbba1ca5e3b3203a1b6f7b9641881a95c
741571 F20101204_AAAQHH lane_w_Page_122.jp2
7ec6764859baf3e79913920aed2ef42e
4937 F20101204_AAAQGT lane_w_Page_028thm.jpg
f07bc237e541c2fcec191a7b65753852
1051935 F20101204_AAAQHI lane_w_Page_078.jp2
0c6ac6daa79215a709ac400a59ac9b91
942c667df9111cec1168b77f7424dbefbc4250fb
75710 F20101204_AAAQGU lane_w_Page_023.jpg
408458f06093cf0280f1cc7a0d09423383f13193
F20101204_AAAQHJ lane_w_Page_011.tif
8b30e16b74f4ab985555cd5a1c69bb07
c3c14fec7f528a88ca7b672b409122642f0199a7
1991 F20101204_AAAQGV lane_w_Page_018.txt
7dd48f253bbd062c3a36b441dcc138f7
366e0e9b3039af84bed77592fb680fd14cc91e3c
1847 F20101204_AAAQHK lane_w_Page_139.txt
bbb84ec1f22b1657df3781240308202a
2639 F20101204_AAAQGW lane_w_Page_005.pro
f49fea44299d4e468c7cf1c38fb9d097
804 F20101204_AAAQIA lane_w_Page_100.txt
9fe92006f94f1b34bd35466a7bee5957
9f9a40b04d41bb0ed5249d0066be4bed9d0286d7
14646 F20101204_AAAQHL lane_w_Page_075.pro
4cdbbb057493a70ec355502199989426
b0039781fa212a15d7438ae5c96acefd33dcb849
31349 F20101204_AAAQGX lane_w_Page_098.pro
b97e3c22a45a0a2168c7b6c14725c886
3083aa3f63624460543dd5cce2a361bdbe1ccc32
24653 F20101204_AAAQHM lane_w_Page_056.jpg
a7e733773df858d4e37eb37534ee1a90
b08c143873f9939703e2ac86be9cf3f34d0d661b
2217 F20101204_AAAQGY lane_w_Page_031.txt
10afa8c080b561869fb8610c77a59441
b6271027a00a850afa0a16c99c054c14c1140563
5752 F20101204_AAAQIB lane_w_Page_126thm.jpg
e0e28b99439b8f014cb3fa711fd56018
841232eb6b175c2e5d56906d0b2f93f787de2919
5851 F20101204_AAAQHN lane_w_Page_040thm.jpg
e89d34d8eb772a9baea0db9e4259562b
e9449b1746fa5e1e24c81aa356bbb6e628f3a020
92484 F20101204_AAAQGZ lane_w_Page_095.jpg
485160cfd90b0ff1429db94b4d66d0e307f0ceb0
F20101204_AAAQIC lane_w_Page_031.tif
1d719500e66988f1fd613a4663f72e66
21241 F20101204_AAAQHO lane_w_Page_039.pro
733589476b93fd4f928fd0805db02700
7a0320e2e7fd552e62d180a17beda67bb210dc84
2361 F20101204_AAAQID lane_w_Page_063.txt
F20101204_AAAQHP lane_w_Page_048.tif
a228d6b20b6a4889ec07ce91f11b2deb
af4d5af9f05b9bbc41b90d95f1e8908b8ee6372a
55673 F20101204_AAAQIE lane_w_Page_020.pro
760702 F20101204_AAAQHQ lane_w_Page_090.jp2
45a9c2a045c570f88d5cd1f204eafd51
acff99722926b57629f2f961d09d9a4c9b9cd55c
F20101204_AAAQIF lane_w_Page_126.tif
20c7ef9fc5ff65f7af794fdc53988f57
6d45b92240abdc79c6f3f4cdc86e4514cb808874
F20101204_AAAQHR lane_w_Page_123.tif
4fc109901f0037b33359255a663d5249
9befc3a671a4eeb378b801724937c0affd8bc906
F20101204_AAAQHS lane_w_Page_038.tif
ce254db56f92a8c9bd1c2ce05c5fcebb
cc6450daa627510c8bb402b0a5db61d7ecfed689
F20101204_AAAQIG lane_w_Page_026.tif
4e71c7ac1177f7790bc08742b7166b80
cb699dd47640b95a61a15c950b4bdd4203876188
24558 F20101204_AAAQHT lane_w_Page_064.QC.jpg
8b02894a4c3f9acc4115d024966de35e
a9a31272fd672f3a6521db3eed3911714599fed4
84887 F20101204_AAAQIH lane_w_Page_048.jpg
32114b2ceab121c86dea4cd1b31e16c9
70629d652e5e74e56517a6fff58696f7b15c8bb8
33338 F20101204_AAAQHU lane_w_Page_090.pro
290fa03fd44082566f586e6b4792ec6b
0f2d941b9d4eab5f55d1b9eb075b56949245cd6b
20600 F20101204_AAAQII lane_w_Page_108.QC.jpg
0edcb3ccd5eff2137554ed3dd0953e9c
f647bcece7c7b51e69cf5ac16146362076b3666d
20913 F20101204_AAAQHV lane_w_Page_045.QC.jpg
6375db6efffd4a00d816f3bdc0a00978c4f27a98
28238 F20101204_AAAQIJ lane_w_Page_085.pro
52dcacb9fe576faa1c52f07f1a51177e
307831a3363f9f6cef47210becc81f655a16bff9
1559 F20101204_AAAQHW lane_w_Page_067.txt
e9a9c41bc34046c010751bf3ca972ecf
ea69a52ba24ff6f9172f2d188c261f3b285a2cb6
20439 F20101204_AAAQIK lane_w_Page_065.QC.jpg
9db4dc6319b5737e5f1b595bb1f643c3
0d707314999fab4525a64eefac611240f71c7c61
4918 F20101204_AAAQHX lane_w_Page_122thm.jpg
5443da0c383e845092766cbe8643e418
02d1a2df6f9649e55f911af569974b6ba2ee2c81
2266 F20101204_AAAQJA lane_w_Page_062.txt
83c3a5756688d11482a88a9f346798bf
089aee75694f39a0583bf6046605046458dd4cc0
62153 F20101204_AAAQIL lane_w_Page_039.jpg
24999bc7974dbf704b6ce6ac3659c814
3058997cd99338b0e6a0264ebcb6d2055c6b2579
19119 F20101204_AAAQHY lane_w_Page_070.QC.jpg
c40da77b7c682e20c5a6c037b2c695a1
78d78e7f957b096b7b29e0b3f815fb7c4552e6d2
31465 F20101204_AAAQJB lane_w_Page_067.pro
84903ea0cfe7219339686d1850e5e3f1
5f6447f4098d1e94a4e0f957a82a85d92341b7c0
31164 F20101204_AAAQIM lane_w_Page_096.pro
7e0bee6b3ec330e5b299538fa9ca1779
b7abc4fb318bf82b996424ecaba4bf72488f35b7
18821 F20101204_AAAQHZ lane_w_Page_087.QC.jpg
a0707c5bfaffb7207fefd117f8e8c78f
e6a1541b1ae519528b473214a8989fd4e21fd4c8
2100 F20101204_AAAQIN lane_w_Page_106.txt
7a7cf8af4bc82147750383b869f74279
6189 F20101204_AAAQJC lane_w_Page_128thm.jpg
9bed17cb4f10ddd1d6afecd799f957e74dd67c2e
771847 F20101204_AAAQIO lane_w_Page_086.jp2
0298b94222eb184b0c3e1a5d3a26cbbd
49ce0d5e3c009e8ae46325d6e712605a8a8c98ec
F20101204_AAAQJD lane_w_Page_023.tif
b4f8f1e4e366b2232d78b3018fdc5dd2
9cbda7d5a8a2bc8c622664a7e61bb68b12b1b4fc
67893 F20101204_AAAQIP lane_w_Page_097.jpg
f6a77fe2b966a3e8c08b9e415d93fa46
b0f8e557cf48586ac065492ccd71ae21ffeafee8
987656 F20101204_AAAQJE lane_w_Page_105.jp2
8e13b69dfd4303cfb8cef36698713769
f63b17c9f8d03d3c6548c54b10a9817f020b4eef
2078 F20101204_AAAQIQ lane_w_Page_005.QC.jpg
fdd07f1a22107c5b143879b5b33393f1
ce5a0e03b17027ae90b4b096bc842be6378eb4ce
823068 F20101204_AAAQJF lane_w_Page_087.jp2
05f7044bef17995c4b69ed9c4930c1c7
efc48547a9f1e32d9ba48d025a0a41c436b1d2bc
26300 F20101204_AAAQIR lane_w_Page_100.jpg
5af96eae28e77e9459eeac9f9b055a4a
7bc57ae230836df27cb3b9af65b1c09439407e80
57793 F20101204_AAAQJG lane_w_Page_136.pro
e670011d98fd7076eb16113541d625e7
3561a598dd5715dcfbbb0ebd91170a47d5473426
45167 F20101204_AAAQIS lane_w_Page_081.jpg
d7e4801abddae3cedc6e906260fe43e6
73984 F20101204_AAAQJH lane_w_Page_027.jpg
fc1da84f11132a418bcd739dc6fd46e1
65b95fa6878806c87474fb34c6a28ee1b28bbf57
413284 F20101204_AAAQIT lane_w_Page_102.jp2
b07949a89b9cef482e6e51b04ba0c0c5d9303087
13019 F20101204_AAAQJI lane_w_Page_069.QC.jpg
c1b433483573d19a47e55748a085329a
35f8462f69b2bafe08cfe3ae765497e19dd1bc90
27693 F20101204_AAAQIU lane_w_Page_050.QC.jpg
0f1f1d5414d08ed7a9cb57e428638038cb61212f
20556 F20101204_AAAQJJ lane_w_Page_030.QC.jpg
f5790c517a6c8f28772dbccf8082dbb9
afa19798472a6bb4ac60f594dd14c3203f8bb86b
97266 F20101204_AAAQIV lane_w_Page_139.jp2
85875e0308302cc888526508a8f7ce7f
18366 F20101204_AAAQJK lane_w_Page_083.QC.jpg
f3b6e5f90d84befcaa9d5152e67aa93b
b648201695c259b87cb2caa6022c87726036993f
51396 F20101204_AAAQIW lane_w_Page_014.pro
96393a1c10234ecf9e01e0f2d6cdae25
be10efec471697f372cd09e70b33fcb6561884c0
1476 F20101204_AAAQKA lane_w_Page_098.txt
843b89a12028bdf6a2ea644171029e76
654e11d8d913a748c35f6f5958c7d4556cfe232e
1180 F20101204_AAAQJL lane_w_Page_066.txt
8a9be2a658270bcdbd2407975729904e28b93c24
4338 F20101204_AAAQIX lane_w_Page_132thm.jpg
c1a84cf40e6ea105b1fcbbbb36a67f9f
f602ef06c5959e7a1618d37d974deee740b0c11d
25707 F20101204_AAAQKB lane_w_Page_042.QC.jpg
b0840a2565612abf34baaa3c9c3216ae
f0f5e465a4347f049f2813e289b3c83b4902cb3b
51219 F20101204_AAAQJM lane_w_Page_050.pro
d54b831d9a6c36ddd45ea8569012d282
799 F20101204_AAAQIY lane_w_Page_075.txt
24d2f5d316366c3f4e820abbc1f66e3e
bf738e3f2920dbea4d87b64b7b85a81af95a065e
F20101204_AAAQKC lane_w_Page_017.tif
fff5ba7ea6d2cb874108583cca2dc805
39202ba7648a7630254e047d19166bf0a87c1a93
24284 F20101204_AAAQJN lane_w_Page_008.QC.jpg
0224c8f7657b2e352fd16b21d3505a57
95f5b7ba82d4c0726e6ef825cd63135c03c50dd2
F20101204_AAAQIZ lane_w_Page_012.tif
f1fb67cf4e593dc0a46dc839f4363c78
916331f8b06234da1e7b0856d4e542f532f7677c
1729 F20101204_AAAQJO lane_w_Page_121.txt
6edccaf9947696dc5baafc28bcd31af5
2a8b2016c3e8c37e590266c0be635a60f02f29fc
F20101204_AAAQKD lane_w_Page_130.tif
2bd63b3076e93986de4b4c8a2955bd0f
2c31a22ce01a372f0d6d6653efabc4c9df16ab6a
1051963 F20101204_AAAQJP lane_w_Page_014.jp2
dcf1c158dccfdb77c03af01e8222b7dc
94d553d7bb7e269bc8195ac8ca28dd5b5a40b6ae
951136 F20101204_AAAQKE lane_w_Page_127.jp2
a3455b0a4d408eff6065cefe18e0387a675c5f11
5005 F20101204_AAAQJQ lane_w_Page_015thm.jpg
3fc7b2049b4be50a25fd74868e10fa9c
a4a37ff5505a28ac59321fd58c408c3a90f2bbc5
70125 F20101204_AAAQKF lane_w_Page_022.jpg
5e9dde236f35e12182947de23459257c
108a9f5fe203a3fcbfce136abcdd6f6fb93feb6e
90140 F20101204_AAAQJR lane_w_Page_020.jpg
120754d5091bb21c908c92571c22187b
419 F20101204_AAAQKG lane_w_Page_001.txt
8738711e7c2eeb4d44b16ef739362b49113f8736
F20101204_AAAQJS lane_w_Page_076.tif
c3d6f4e2d2b351f4e402e35f4e17565a
8676e3f0b067de4382fef982beb5eac4e6f64a09
22071 F20101204_AAAQKH lane_w_Page_047.QC.jpg
d9dcd3c987450d6e8be7fb54c17e729d
1fc9b6309d0a708182e30a879828c1464c2720b8
30854 F20101204_AAAQJT lane_w_Page_045.pro
e45d54e617a6d259965b78f584a6d14b
47c984261347d33849cab692504979288c413c6d
864068 F20101204_AAAQKI lane_w_Page_034.jp2
45e3d7c6d551e234570abde9f6260d8e
cfd9521ca5fbf0150fa8d8045163fefb231b044b
95858 F20101204_AAAQJU lane_w_Page_063.jpg
cdf2f6d7bdbfa77c85e9d316727bccaa
978602ee06ecebbf3a07911d62482dbb97f1bb8c
5865 F20101204_AAAQKJ lane_w_Page_140thm.jpg
4f3751cb2460594021aa324c4fa27784
4865bf59271a2265aaac8b5416a14c96f71c164b
F20101204_AAAQJV lane_w_Page_112.jp2
f5427152fa7e727370a5ec2b50c07796
e5f07da892f97090f4f126f89fc848aea6c4b303
74995 F20101204_AAAQKK lane_w_Page_082.jpg
baca3a59d4e4c01eabe2e1ee8cf90b03
993bd727944a73c6a8eb51f5202bca730d944eda
73601 F20101204_AAAQJW lane_w_Page_051.jpg
f41ca4aa614c6986d25a938acddff1fe
2315 F20101204_AAAQKL lane_w_Page_107.txt
174ee022cddc39636b0088a5bb0396b4
6315 F20101204_AAAQJX lane_w_Page_063thm.jpg
f012d0f992ff286c3e25e9002c6601ab
272bd5b64e4d51c24fa79478fea85ee11d3a4c1f
59466 F20101204_AAAQLA lane_w_Page_021.jpg
8d5c63a9cf8f027466883aabfcac53ce
35eb5546ba4eaa9542c543050327af679f232d0e
45364 F20101204_AAAQKM lane_w_Page_004.pro
3ec1febfe0f896ba463c00e39c766152
ac030e3d231befa517b25c9a861277649a989ff6
F20101204_AAAQJY lane_w_Page_133.tif
a2bc135fdc0e938dec3612bca0de2dc9
045bbf0aa89b9d3ddf3438547c68a899d55c71fc
58319 F20101204_AAAQLB lane_w_Page_025.jpg
b6e6f9e95f16d0c4f7e792e074fe065c
82d5d35e00f35750fc26a45b0cf9075912045392
54821 F20101204_AAAQKN lane_w_Page_038.pro
3a266c2185997c4608fa162efb4d64ba
1051985 F20101204_AAAQJZ lane_w_Page_049.jp2
89a739f41fc35c1d5efaa608a675f8c3
63458 F20101204_AAAQLC lane_w_Page_028.jpg
981b41d71f5b1aa14c489ba8bce608b5
b94b4ec1a2b82dcd5a7dae5249061c6717246322
F20101204_AAAQKO lane_w_Page_137.tif
e0fd1651950def83ce892325ffe6fb73
c7a96accfd5ed21d86125b7fb405f4d8b3aa6539
82843 F20101204_AAAQLD lane_w_Page_029.jpg
1bd88929a67ddf2cb40d040c6a323e2a
995b7b12658e6877f7db12c1a6eee153d8390912
60167 F20101204_AAAQKP lane_w_Page_090.jpg
9d29960130f17943a8552a0e4c5e8509
dfe668dd07da4cfbeb7f404b01a7715ba7bf294b
211941 F20101204_AAAQKQ UFE0022090_00001.xml FULL
10bf100a4637296fa874659cfc909990
41f5dd8f745520285bc2484223a9306dbcd66dee
67570 F20101204_AAAQLE lane_w_Page_030.jpg
74fa0fe95de003522c03445c027741de
90719 F20101204_AAAQLF lane_w_Page_031.jpg
097d88fb87b37cff4f6456ee002f109c
aa35264f1d27d5c9a94b512bab54a904e95cab0b
80823 F20101204_AAAQLG lane_w_Page_032.jpg
d4d5cded95935bc1262a2674f6de113a63949fa3
3755 F20101204_AAAQKT lane_w_Page_002.jpg
d3a61dd562ccdba50fe8b307f43f211b
dda3235351ed12d953bfb6a7fb3b0704a2dd085b
51550 F20101204_AAAQLH lane_w_Page_033.jpg
f743dc84d1d35373f52b104f1d8c8197
3180cbeb21acbcbbef11d2e30898d5f4191aeef4
74367 F20101204_AAAQKU lane_w_Page_004.jpg
9072342d2c607c24f98a467bab43a790
919333e473cd82fae42bf32163fdd4330f9ea1f9
89895 F20101204_AAAQLI lane_w_Page_035.jpg
1e8f19e7030cfa7693f4fb32fd7af863
7da4ded22d56f3978eb026d093a52fa6d43abc32
87759 F20101204_AAAQKV lane_w_Page_012.jpg
da277e30e6841503cb13a688e7f17008
8692ba654e619d634552ec79ea02a27697effa0d
74318 F20101204_AAAQLJ lane_w_Page_037.jpg
18bba8d7250b7783e907f752f1ea13ab
60856 F20101204_AAAQKW lane_w_Page_013.jpg
0cdcaef7280e9504cd4d72f8ec8d122a513ff2a6
85516 F20101204_AAAQLK lane_w_Page_042.jpg
f0dac02f7146f7bc8595a4e476d645c1
80172 F20101204_AAAQKX lane_w_Page_014.jpg
ba6638ae57da655b3ae9a9283845df92
c31f928999ee573fcafdb245112dc2c08fda17ab
42773 F20101204_AAAQMA lane_w_Page_071.jpg
811f8cf8c3c5966d38591b39c4a75c32
95904 F20101204_AAAQLL lane_w_Page_043.jpg
8c9f362831a376ff29d06d4359425780
1202a38cf3ace2f6545dd7b121f4c204df4bb4da
71519 F20101204_AAAQKY lane_w_Page_015.jpg
81e4326b47170098ce140a2dc8f78489
175be86e150b2a6ee317b9b7f237baec1b4e8d8c
47139 F20101204_AAAQMB lane_w_Page_072.jpg
749a83fc26d67695f8f8df293bd014d6
59e59b5fe1effe31ef23fa621397c6469f705d3b
90564 F20101204_AAAQLM lane_w_Page_044.jpg
af5c6799ba6ae1c696266503ee8aff62
69919 F20101204_AAAQKZ lane_w_Page_017.jpg
28173d94bcb19fb9e92f474e3afff03e
3a25f4184df5fbc10c261a5ddd775e16c5139488
78479 F20101204_AAAQMC lane_w_Page_074.jpg
2f1a08b735ba4e09d03897dd8e25fc2e
c78a914b5367c6dd67713d2bc942d11b37226817
66053 F20101204_AAAQLN lane_w_Page_045.jpg
b7186bb8e31fe19dec17f247e767fb89
e549ab1e04878aa9c3310a34297db792e66f9cbf
28631 F20101204_AAAQMD lane_w_Page_075.jpg
2e7ff50962fcf2e8f801ea2827c252a1
90871 F20101204_AAAQLO lane_w_Page_050.jpg
9b8a1e698bfd8d46bd1dfba4eb226448
3da1cd9564c252a99c773fd2656b64e41509be6e
93617 F20101204_AAAQME lane_w_Page_076.jpg
330836159a1291a77fb954ce6aacc20b
9a12f8664de22b3c1237a319843f2b09bd381d99
77772 F20101204_AAAQLP lane_w_Page_052.jpg
412e0d84970dc47d58eb843e7445ed01
d0179a32122114db35b3e2bc9d48539474a1eb7f
82482 F20101204_AAAQLQ lane_w_Page_053.jpg
e6109b1655cf41030be0bf97d5109083
e008525beb0fdb8714780dbbd378f4651d74b185
95090 F20101204_AAAQMF lane_w_Page_077.jpg
8bb79ba0f38a580cdb54565435cc1814
8330b529228e4c4393d9a7d2595459d1d36ff6c1
84096 F20101204_AAAQLR lane_w_Page_057.jpg
364bb7983a7f545d713581cc8113132c42d6c7e2
89990 F20101204_AAAQMG lane_w_Page_078.jpg
a3c4e47818d7a43ab46a28e35819d697
84063d8bf50d7bb148bcfbecf1b3b2a116b69d80
56083 F20101204_AAAQLS lane_w_Page_058.jpg
e4659df1420dd893727b9db7f88e1f3e
497c95d7bc306dc25a9d97122175c5e7b0eaffb3
56737 F20101204_AAAQMH lane_w_Page_079.jpg
3a9a47b535f774c18ae312eca68e6c28
2ac120592800155a2eaa18726d8e01a9e37407d0
61986 F20101204_AAAQLT lane_w_Page_059.jpg
e55436e325a57d78fa06975a5ecf0a27
6eefe7118c863b7b833a53782cc8485251578465
61477 F20101204_AAAQMI lane_w_Page_083.jpg
93edf1562a9f30f2b002b70bd8e35f0a
0b1b7f90bf7baa81d149ce8cf6d68323d4cd3587
74099 F20101204_AAAQLU lane_w_Page_061.jpg
7e7c7cd586f9301492ce2e42da9fdee3
60b5bccb26bf98d2f126e99bff25f5eb2966fb00
72098 F20101204_AAAQMJ lane_w_Page_084.jpg
59c3b41db30b7680b4cb33a622460d90
349d16d57a0094814fbf3b0aae8e55a202fcedde
82672 F20101204_AAAQLV lane_w_Page_062.jpg
bcd9d4630542e5581145288f452a6ae4
3804d82e9a89d3b62f1ba20343dcf8e3be28ac1f
54154 F20101204_AAAQMK lane_w_Page_085.jpg
496a2abeaab261b9f079dbafc88764ce
492ea04ed694b601e8cd80ba20132459807f59d5
62873 F20101204_AAAQLW lane_w_Page_065.jpg
762aecd40f2617dbd4c4e8af1e07853d
10311041b239e952c1edfeecb5ec2bd3aa90ecd6
76162 F20101204_AAAQNA lane_w_Page_111.jpg
464955a34fb1db56d7644a0d7a04e03c
68f0c816720c6888e91803fa55605963fe667329
59185 F20101204_AAAQML lane_w_Page_087.jpg
af5ac85b867640b5314a219923845966
c9772b862fbaaa0fd04e51d45406acea1039be61
50427 F20101204_AAAQLX lane_w_Page_066.jpg
65014a6ec8cae1055f9320160e22f054
b18af0c282bfa6378d5f08a6c06c3ce675b8e849
64588 F20101204_AAAQNB lane_w_Page_113.jpg
400cc01249182da88767fbf88108a2be27182db9
56798 F20101204_AAAQMM lane_w_Page_088.jpg
c9755467481c03019520d73fed58e584
dc2e27d382dce98fea42c3414d956ccfda178b39
50901 F20101204_AAAQLY lane_w_Page_067.jpg
db2f35d14a2a382b6bd3f84ee4be4395
337808df57315ea1a8cb52da2bbb5e71b137e139
73857 F20101204_AAAQNC lane_w_Page_114.jpg
39492 F20101204_AAAQMN lane_w_Page_089.jpg
f427f7e8df0814089edc1f3ca0721f29
61588 F20101204_AAAQLZ lane_w_Page_070.jpg
09a1be14cc03e489e3951b447db319cd
6e434bcd4bb1f21eb14caebcc3cd8fff0b09b176
63452 F20101204_AAAQND lane_w_Page_115.jpg
e8bcd0f69f0ecb6931c60be266865d1e
e1a9b8f1ba8b0734cb11249000996b305e0c0da2
73123 F20101204_AAAQMO lane_w_Page_092.jpg
b303bea2d346dab657483dd536161d7e
206684c580a5086fce087e62a08a5b56457998cc
66748 F20101204_AAAQMP lane_w_Page_093.jpg
5bf1ee537cb11cb91773ae5749ab0ae8
f51ffc0e96ac31199c22d9edeef4eb164ed16753
64756 F20101204_AAAQNE lane_w_Page_117.jpg
ea2e48a6da3064420380b5e92070d8fe
a22834374b07bd268c9650ecabd0e409a90b9fb0
91948 F20101204_AAAQMQ lane_w_Page_094.jpg
c1cbfd702b5eb60989f101f1c88e5b85
67788 F20101204_AAAQNF lane_w_Page_120.jpg
951d4c7fa7a6a5a76840da1425b13a32
994e54def1c8f6dd896040e00b3cf1714ecd5598
58370 F20101204_AAAQMR lane_w_Page_099.jpg
1614d5c414a8da9c6215097910f58b74
70f16e3739799d4f76b93f28817091f2622d19bf
26659 F20101204_AAAQMS lane_w_Page_101.jpg
075768b9703f77e0699c9228b416e9de
ec563b4874a2d3474f46d465601e86a51010bd7f
74476 F20101204_AAAQNG lane_w_Page_125.jpg
b3cbe7c5351970a92493c2eecb4447c8
b121478caf71632a89a308bdcd25ec226c022eed
32158 F20101204_AAAQMT lane_w_Page_102.jpg
d7a0a7d5578ea1a37ae0d9a81d0578c7
5358877246d0d1e6969d9f45fa2a39848f56ba81
68488 F20101204_AAAQNH lane_w_Page_127.jpg
52b62bb89a9435b92125353e20b071c3
ab57697c6839399346cf6c76c6ec4e47e464de22
91143 F20101204_AAAQMU lane_w_Page_103.jpg
65ce5f0cc28a35529a35b854a25bdf34
76884 F20101204_AAAQNI lane_w_Page_128.jpg
ed43051a7c1ebdac0ff9c9bc841320d9
a18b00dbd6057c91f6a7f915a2b20740529f1e36
76983 F20101204_AAAQMV lane_w_Page_104.jpg
f7d1bef8fd947d895c313bd93e61ef18
2e1e6d332b9975ea8865c1cfee44e3bffcc02ab3
76489 F20101204_AAAQNJ lane_w_Page_129.jpg
fedef8b26b9bb8b2d7b87a2d9b0f030f
a29e4531851b5ff56a43c45e494ed7e5908d7539
71237 F20101204_AAAQMW lane_w_Page_105.jpg
8b7d17d00e7c3ddfec2a8b002cdfa13c
4c10c76123f85d3445f98973832eda9059277609
73927 F20101204_AAAQNK lane_w_Page_130.jpg
a10a877d10ba803208fe8548147ac9bc
a1d8b233504f6df8658b3880948a10c8f5031759
76072 F20101204_AAAQMX lane_w_Page_106.jpg
60daa5e3b25092c61bb812ffc7ce4e87
832635 F20101204_AAAQOA lane_w_Page_025.jp2
76a66277190293cdf67e6079f0ea3fa8
b57f876381d40e1970d111a0246182b533867f89
6538 F20101204_AAAQNL lane_w_Page_137.jpg
3a4d19b3e199b1e23d10d1abf4f8b3f4
642fb2b2ab30cbf9771e61fe0b1b2a024d8741cd
66326 F20101204_AAAQMY lane_w_Page_108.jpg
50d945dddaf434af1c9f85583bc15dae
8273993e10f179ed6ff59c3da34cfb441c3e6b16
90893 F20101204_AAAQOB lane_w_Page_026.jp2
1824790e03cc95370ed4bf254a757d03
cf6bf6967f1d66f9da0973c09e0bf31f54afd2e9
5533 F20101204_AAAQNM lane_w_Page_002.jp2
3ec053ce5f15da721d5e9bfb4f8b77f8
f5d42744d290ec9ae1c4b6f045a7bed3f4366055
69657 F20101204_AAAQMZ lane_w_Page_109.jpg
3d8fdaaeba0f9a7414008f039a8af5c0
80ddd8389d3ffc7c762f266dca67e660017f566d
1004421 F20101204_AAAQOC lane_w_Page_027.jp2
f7d97b59356116db3067da2ca450c4e0
ccc1214e280971aa56a92791f70745ba031d734f
6367 F20101204_AAAQNN lane_w_Page_003.jp2
86c05734d474f4603d9c1932f9f33500
55aa8aa95e817a6d49f1be46958ca6f7ce71ea7c
973917 F20101204_AAAQOD lane_w_Page_030.jp2
d864aa95344ae6144f0062d97ac0d5cf
98269 F20101204_AAAQNO lane_w_Page_004.jp2
ff41696d4e99b7cfba6a649c8076e6e7
1051979 F20101204_AAAQOE lane_w_Page_032.jp2
f9aa54c40f3dd85b6b6d3cebbd4f1129
fbf37e91216ba006348dc4a20cfd0a3d78f6cfc2
8878 F20101204_AAAQNP lane_w_Page_005.jp2
596281 F20101204_AAAQOF lane_w_Page_033.jp2
9d59defff9e02bd48b270b5559882ea5
02636818dd016ec3aa43d9cf59d5022f1d0b8f6e
1051882 F20101204_AAAQNQ lane_w_Page_006.jp2
095749b8596f4814c4dc6baae9bae983
4a77a95a187fa8145fc69cab651956dac3841878
1051930 F20101204_AAAQOG lane_w_Page_035.jp2
33aaab2b18707e315328e2f84c28e86f
467e7bf5186945cc17d66a59b711c7cb9afe9738
F20101204_AAAQNR lane_w_Page_007.jp2
0e330bf01d4604091799b5fe90327482
17ab53c1ce7faffd886afa155129107591f82ea2
F20101204_AAAQNS lane_w_Page_010.jp2
7e3c8e480d09f3c39288446f646752b6a131d43b
961136 F20101204_AAAQOH lane_w_Page_037.jp2
cf2d7338b4f06278269fc5f64d30f322
21a0d3086d062c3c763022513657551a8a1f2286
968017 F20101204_AAAQNT lane_w_Page_015.jp2
1051976 F20101204_AAAQOI lane_w_Page_038.jp2
540235a290e425ed909553c9a0e4883e
F20101204_AAAQNU lane_w_Page_016.jp2
afecdf60bc035c993177be035f27c7fd
1051961 F20101204_AAAQOJ lane_w_Page_041.jp2
0a96d5749f6ef43e8abfd60a24b7d156
ca29cf1ea537832db4d4b09ee3112233e6c93481
926925 F20101204_AAAQNV lane_w_Page_017.jp2
711f0b8f462aef341ee5a1bc67dd826d
a18a0b860905cd742a4081a68f7ddc80055ba9f2
1051977 F20101204_AAAQOK lane_w_Page_044.jp2
09c5f44b2491fb947987510fe8ffdbc3
3a3411c30ea9a00a0e2a5e429c311d3bbc116a82
823679 F20101204_AAAQNW lane_w_Page_018.jp2
f26e91fb0cdb7a7647980f0cc04ce81d
1089c603778fb3d5eb381c9b33fbd2dd0e1ff338
737548 F20101204_AAAQPA lane_w_Page_079.jp2
88550b945c3b93db5a9e334625639000
2b0d2f24d05fd30ae03377cefdf9f9a4cee58456
1051978 F20101204_AAAQOL lane_w_Page_050.jp2
f1cb71fb438cebd11a11fcd4148cb0b0b24d4d88
950533 F20101204_AAAQNX lane_w_Page_022.jp2
9bd80e4dd3880fd80584f5c73d7150f1
557356 F20101204_AAAQPB lane_w_Page_081.jp2
ce8d13c0a23c9f659106ceae6bee9f07
85b238f79de84c54f8939588be0ea0143638c7b8
974276 F20101204_AAAQOM lane_w_Page_052.jp2
c3a86f29c51b3a3434375b0e44cf821c
879be579de2e933d665aee61436fd3b38a4a7cef
1050902 F20101204_AAAQNY lane_w_Page_023.jp2
027bded2318dfb0243f0de7cd1364b7c
4aaa1d9f7c8f4ee5f91e4c32c5c786682d96793e
985651 F20101204_AAAQPC lane_w_Page_082.jp2
bf63e706e063f74d38d71b909fddce03
86437fa2d772be5f735804c53d05f1fa953870f7
1051986 F20101204_AAAQON lane_w_Page_055.jp2
9ca905e15a6f6be96f1415cb8084e942
899244 F20101204_AAAQNZ lane_w_Page_024.jp2
d6362b6280e7d920bb3518e218bb6c17
d87c68a8d59db01678613b3abdf975cdaeac3587
799802 F20101204_AAAQPD lane_w_Page_083.jp2
9778e4bce7b96985fb08f912e52680b3
63e4d3d3d382ca7394398fa76eb60904f893740f
1051951 F20101204_AAAQOO lane_w_Page_057.jp2
8f5ae17df1c5ecea59c67d6fbde42dda
ece7e14c9a40e411c372dcb120f48c2cc608923e
1012565 F20101204_AAAQPE lane_w_Page_084.jp2
435cda2abd5098cfdcf47663250caa0d8bb6ecf9
988569 F20101204_AAAQOP lane_w_Page_060.jp2
71ef9958ddbe51a2bbc04db3967e3cdd
b427998d291c567c18040a78f1a47f01c3416fbf
588375 F20101204_AAAQPF lane_w_Page_089.jp2
ab2fa9d385b89eb8e954b31a28975685
92f87ecbc0186e5dca5e8e23facd8dcd7f4eebb3
974659 F20101204_AAAQOQ lane_w_Page_061.jp2
8847c90403220eaacd39164e71c08f06
846db9c8e8625b571a5b1a7edd1d0c5a748396e1
1015561 F20101204_AAAQPG lane_w_Page_091.jp2
657650963059515fed4929680a867dd9
67e07af4c5e48155408c63dd54a823c77ea77822
1051934 F20101204_AAAQOR lane_w_Page_064.jp2
9e6e1962cff0e02ac875040f12a5ce9c
e65e5d95787e5ec7de7e1c70021e680a27b240dd
934767 F20101204_AAAQPH lane_w_Page_093.jp2
03e5828277570529f0019f4612255839
bdde81694f50d280aaaa25ce70693d7716504b22
822866 F20101204_AAAQOS lane_w_Page_065.jp2
17395cea37593e92cc1b2f27f40e2268
d18cd25ef6011bf38491938a9cca806e7997dc5c
630067 F20101204_AAAQOT lane_w_Page_066.jp2
e5341e3bdcf6913d4c7edecf5ce9fd5b
a6be26e78e8428c647854b05524863a5778afa26
F20101204_AAAQPI lane_w_Page_094.jp2
3afc4184084fcdb84a78f2c47c1520ab
554429 F20101204_AAAQOU lane_w_Page_071.jp2
f6c7a745cb3645b77eae8d276d65ebcf
cf9a927a24b8173111e91aba263476c3bd85aa51
1051972 F20101204_AAAQPJ lane_w_Page_095.jp2
617f8ec41088670f1bf5e895b3bdd708
c2127510fbcdab542b51e504a03ed3d0a34f7cab
57665 F20101204_AAAQOV lane_w_Page_072.jp2
7bd62052da6260b5f5dea67da1aa30e8
7eef5fcc0352706b3319e277e7c3378ec11f2490
930807 F20101204_AAAQPK lane_w_Page_097.jp2
dc49b1c08ce6ea776ec296e59af47149ec3cb8b4
1051953 F20101204_AAAQOW lane_w_Page_073.jp2
86e9fd7834c72620341f180e273b2f2434182626
909773 F20101204_AAAQPL lane_w_Page_098.jp2
6dc8eddc770d12a21bc22181a5a6afc4a09aa981
F20101204_AAAQOX lane_w_Page_074.jp2
e36032e8da4d6d759bdbb4ecefc67475
e0d5412c5e6834c0656526f955bdafc9013cd425
979772 F20101204_AAAQQA lane_w_Page_126.jp2
cdedfea4798fdf62b0ce089d5abd2ae5
3fe6228ef5fe643ffeb29494bda80bc2f1ebb7a7
343687 F20101204_AAAQPM lane_w_Page_101.jp2
206887b0e3e72e1cbe4da4c81e89d3f5
39d3380c45fd5496de683845e558270a79eb402f
335937 F20101204_AAAQOY lane_w_Page_075.jp2
7b386ae8d3413e652189b2ac0da57fd3
14b7b6150c57a3c1bb6c75d2831a03abdfd1b632
F20101204_AAAQQB lane_w_Page_128.jp2
63ea989aa98603d2f11eca448e03947a
f867ca9f6dda4e635f49b033ca5bf69cdb560627
1051974 F20101204_AAAQPN lane_w_Page_103.jp2
d6cd85a12e66c6f8037dcc6ca56eea27
21043b5f41b62549449f616c779524da050bae9a
F20101204_AAAQOZ lane_w_Page_077.jp2
0d4a395ecc8a317002f50cfcd81a862c
f7f163f23ffd37c23a6e8333bc718ed66edcc805
F20101204_AAAQQC lane_w_Page_130.jp2
7506f2cd8e0ffb283425b354252739fa
c37f45158f351f9d0516d05b3d0b811022199e7a
F20101204_AAAQPO lane_w_Page_104.jp2
e15d1ae1278034cb6a4d4c6af347fa4e
645783 F20101204_AAAQQD lane_w_Page_132.jp2
d7637707c29c4920844b2f190853fb89
c72a168431438f49d3eb509d19245f75946aec9e
911234 F20101204_AAAQPP lane_w_Page_108.jp2
3693550d0da1857ee0d7a9cacdd6d09f
a0e7b0ed5ac06a32c4213b484d339b0f2018f15d
1051939 F20101204_AAAQQE lane_w_Page_133.jp2
819dec02be67bd6b8e9d24385db7ecec
ac3bc9d80bb9b7a798b64f49d749a9ea07782e08
955421 F20101204_AAAQPQ lane_w_Page_109.jp2
c1cf4939c09a12128f04cb0271c112d9
F20101204_AAAQQF lane_w_Page_135.jp2
c4dc36756556dbb915edac5bb39bef43
1051947 F20101204_AAAQPR lane_w_Page_110.jp2
91d3ec22fb351ae1f17b83f15b6b1af2
F20101204_AAAQQG lane_w_Page_136.jp2
3088fdb1fb99170d7fae4df2e20a2beb
c969178835dc4f8b7c655639cc83229c3a5f68d1
1051929 F20101204_AAAQPS lane_w_Page_111.jp2
9378d2c117a2595a84330cba12c0dd52
730a6955464d3bdbd738e3476ed963f633b25d1b
83514 F20101204_AAAQQH lane_w_Page_138.jp2
ff3d8de9fcb6eeb55a81a9bab162dcf1
5c8fc4d423c96092971416b4c5f9fcc68c1f6c0f
808340 F20101204_AAAQPT lane_w_Page_115.jp2
106690 F20101204_AAAQQI lane_w_Page_141.jp2
b9e2a1979561655c2e151ff4ca33267d
98b9f3fda4b045a9950aa24431332237b080a245
1027250 F20101204_AAAQPU lane_w_Page_116.jp2
c66d618b8cf918a048b9a1b2860655bd
1e72c569418b6419696102f2d2ca4a2ac8353638
966131 F20101204_AAAQPV lane_w_Page_118.jp2
561b7861ac32b48e8f2bc72b1fbf0671
6e17ea61b6129d7c1e405620c6d4337bd6aab533
41345 F20101204_AAAQQJ lane_w_Page_143.jp2
e471f6bcc99cdbe1cb17d2a306946731
9f0b4727741fbf66cd73aa355460dfc352a89c01
991290 F20101204_AAAQPW lane_w_Page_119.jp2
4784c50ac5a14a328184279271ac4ea6
2ef850679dc7e7da622a4ee6bfc3aa2c0fa737fe
F20101204_AAAQQK lane_w_Page_001.tif
687735bef2d85afee1885aec3fcb6653
837232 F20101204_AAAQPX lane_w_Page_120.jp2
8bd54befccbe7bcd1d646a93f39b33dd
f8b3e418161b9559938b952a1123acab00e0b5c2
F20101204_AAAQRA lane_w_Page_034.tif
12927a8d70cdcd12bd1caef6606d818b
052c984216a8ff5bc7ceddd0356c3d2269b713cb
F20101204_AAAQQL lane_w_Page_002.tif
b4c7a8049a620292be2eb0068430e11e
44e2cc2cff15fcafeeb984f851c750f87b7a8a34
768155 F20101204_AAAQPY lane_w_Page_124.jp2
2308b983fb3dcdff92a77653ac26ed7f
1d49abe11e14ca450000f2d368e8d76df09e8549
F20101204_AAAQRB lane_w_Page_035.tif
66f3d4614be7770deaa301367a06004d
55fdaa1afb6858e1ec843d8ca4ac28f419558842
F20101204_AAAQQM lane_w_Page_004.tif
a1d716014d6c69001d12eb2df73b280d
991324 F20101204_AAAQPZ lane_w_Page_125.jp2
bbe6c87ea7eb388302719acd06ef3806
5cc71d74a55bfc6c3b769dcf35d88f707dc256bb
F20101204_AAAQRC lane_w_Page_036.tif
c1608babfeaf0419b7d76099c7cbf98d
F20101204_AAAQQN lane_w_Page_008.tif
3dfd0289fe7f3ee23b276e97c53c0835
7226b3f0d8a583597631df8ac833237610ce8f76
F20101204_AAAQRD lane_w_Page_037.tif
4f0f2b6604c490ce27a55f3451198a1a
947a29e908f6a1c21004ed4c974a9121c4d2eccf
F20101204_AAAQQO lane_w_Page_010.tif
b170c242733478182a695c8d69454637
76b00710f1ef45d2f757fd566a89761a833f159d
F20101204_AAAQRE lane_w_Page_039.tif
950662cbe9d9b95eed5fc2bb2e7b548f
994d737e557ec73da67b2b3f9060af0b9e32b0cf
F20101204_AAAQQP lane_w_Page_013.tif
b8c258acb0d5a089469d93d62c1c5210
fe7705fe6e5b2cfc7309bc52528c5d8c1caabfa9
F20101204_AAAQRF lane_w_Page_040.tif
d31be793c04fdc6fcebd3a10d3e54382
13e1a2271b26d179de29aa290cbfcb8380d68fdb
F20101204_AAAQQQ lane_w_Page_014.tif
0f150f1c88b305491d31d6e1deb6dd79
d558a83b3ba3c5dc57a7cd31dd2979fde162cc55
F20101204_AAAQRG lane_w_Page_041.tif
2570573dac5262db08c44078822f9805
fe00736afccbd3702b07622cf17ff3219a8642bd
F20101204_AAAQQR lane_w_Page_016.tif
be0ea7d311504e02700a436b4e4f2c0d
2d53bb7efb2650203f0732fcd966bc09158e2383
F20101204_AAAQRH lane_w_Page_042.tif
21684e21307018a0f854106de83b1840
5ec0edfb9b86f6a85c0571107e15236f2204576a
F20101204_AAAQQS lane_w_Page_018.tif
f1e06ca9c1a62f12ac4461b58e993fc0
14300e111431d40f2ebb701bf771a427ed6481fd
F20101204_AAAQRI lane_w_Page_043.tif
0d25664dac3a0efc109cdce12cbbba5f7eba1e90
F20101204_AAAQQT lane_w_Page_019.tif
2047d317932eb2ae2d8b8494fd280c51
be8248b7ba1acf06a0f25940bce59b619f09df4a
F20101204_AAAQRJ lane_w_Page_044.tif
cc637201f24d97359890dd0eef7b68fd
24789fe8279ca4e57ee5325b2885ee7cae071b8e
F20101204_AAAQQU lane_w_Page_021.tif
040777c453c73ba7df020bcf301168a2666e554e
F20101204_AAAQQV lane_w_Page_024.tif
33a4bfd7682388e15b8c55c86d386012
09db87d9361b831cb59b22a9b3e8067c44a865cf
F20101204_AAAQRK lane_w_Page_045.tif
b11fabe1a6df9cce5d6e48ba27253d52
c72aa39de28fef6c010c5fa4787e7b1b5a1176c8
F20101204_AAAQQW lane_w_Page_025.tif
a2956d821f9ec59463e87cf76a84bae4
1e39bc651ec89f8f14e0243f3b5b89e2a56402d3
F20101204_AAAQSA lane_w_Page_068.tif
670c871ed6cea2553a2bdf87924bca97
da61d519051cd65c449f1ae401d64c9a46a1e505
F20101204_AAAQRL lane_w_Page_046.tif
b83018953ff623db3d692cc80cc01c00
F20101204_AAAQQX lane_w_Page_029.tif
5d1d5458f29c5e5838f1254960419837
746602702b13c44c66ac24826ed726f3398e870d
F20101204_AAAQRM lane_w_Page_047.tif
1892e75f71ebc8a4e6af584942511975
f2f4c69cfb6730ca0fae30b04a1ded379772dea1
F20101204_AAAQQY lane_w_Page_030.tif
3bb6995de7411103eb9cb450a7ab25956b277519
F20101204_AAAQSB lane_w_Page_069.tif
0ed054daf28a076afef4c220f47e8e83
36b48438ee18aedfe0f6bf209330b7b180bbc5a0
F20101204_AAAQRN lane_w_Page_049.tif
a3a3937ac8369193068c4e9b57f8c57a
c5f2897bfbaff189337a2b83230c91e215217a12
F20101204_AAAQQZ lane_w_Page_033.tif
f9233edc2a6a13aee707bbd6aaf7f5df
283f36fdeb45afd1ffe850eb81104033009ed3c9
F20101204_AAAQSC lane_w_Page_070.tif
80960aa6388807126bc893b61dbfab2c
d0e9231844c8ffcaedb1b3790a4bf0ceb65072cf
F20101204_AAAQRO lane_w_Page_051.tif
6dec6eaa57c62ec6337e97f472906f21
b229d506ff5c3f6855896e250143c3a4c55ae80d
F20101204_AAAQSD lane_w_Page_071.tif
e33d5762086814f58226f16d51984180
3e5822ed35dc7cbdb4d12de9a0ab727b63cd43cd
F20101204_AAAQRP lane_w_Page_052.tif
616d0ddb556d7a5a79d6f981154775cc8658decc
F20101204_AAAQSE lane_w_Page_072.tif
1a2d8b8381b5b340c94ee444a4e5f8be
2a8056976b2d4c77343da5bc32e034e0ff15b094
F20101204_AAAQRQ lane_w_Page_054.tif
0e40af2d0364b7c198d2c07e6f3e71d3
e9ef0fa99641dbc342d15d9b2d408ae31b1a910a
F20101204_AAAQSF lane_w_Page_073.tif
4ff320900a9e81a0f597798aa0fa95db
642dc6451f7073a5a66cbd6c6587a629778d6059
F20101204_AAAQRR lane_w_Page_055.tif
e975fb3a1a57099a35925014ac798512
F20101204_AAAQSG lane_w_Page_075.tif
f319cce1417e181bab45d6c0e5b22c99
cbea423a729112b06c6dc63cdb11a26d24437af5
F20101204_AAAQRS lane_w_Page_056.tif
2252b552202826e91c96cd5514f3259e44630a2d
F20101204_AAAQSH lane_w_Page_077.tif
dd98352d402c1bcfc407b1eb3bfca7272342b56e
F20101204_AAAQRT lane_w_Page_059.tif
53ab4579d67eb565cbc32184452fec4713799c3e
F20101204_AAAQSI lane_w_Page_078.tif
d4f359d615af278f8cd1e1736bff5b7e
464782574b40b72eb22c1d4fa544a28bec5d1a25
F20101204_AAAQRU lane_w_Page_060.tif
f9d0d056d68962b410398d6e9e45da86
1ed17b57842e74fb8c123e18fb5e5a0284aed1b8
F20101204_AAAQSJ lane_w_Page_079.tif
829cdd4ac080419c8775f0320997a8a5
35b836217db874bd44af855e17c50fa495daac40
F20101204_AAAQRV lane_w_Page_062.tif
56746905a6a3a3a6da2de3fed114a2ee
df5ff21baab105bda24bea60785317d31ea03c38
F20101204_AAAQSK lane_w_Page_083.tif
e4fb799cc648eac9da9618bf62fb2283
db6d05740632871742dd2d95513c6b250e07423c
F20101204_AAAQRW lane_w_Page_063.tif
d684835d07632565bc484b8e4b48af7c
F20101204_AAAQRX lane_w_Page_064.tif
8946697b1f48dbbd1715588de1740a3e
F20101204_AAAQTA lane_w_Page_109.tif
cbcd633625ca44da6c17562f0928e8a6
6de7285351f91636ac3a8a4fec189c578a4582f4
F20101204_AAAQSL lane_w_Page_085.tif
46a5f4527218331ea1dcdecc15cf9b2a
9735b1dfe898cbcd7525a85a44cf25e5ebdf9ba9
F20101204_AAAQRY lane_w_Page_065.tif
ec47a15182925a85bdd70783af31d08a
7a5eda55218d108ec7d6b238d8f70feab25566ed
F20101204_AAAQTB lane_w_Page_110.tif
ee288bd9799a1d33fba504c4aaacfea9a4b0145c
F20101204_AAAQSM lane_w_Page_088.tif
d140eb54da15be8f2d8949a96323bcbf
4131055214c24fb03eecf9cb0d38830566509f4d
F20101204_AAAQRZ lane_w_Page_067.tif
cfe6bd6b69f08ef4cf48c49ecb5fccf7
8a8d0136a9c0df76c6090dfcb1368771d72f3746
F20101204_AAAQTC lane_w_Page_111.tif
46ea2ca676e16d7823557db84f25eb0a
255b723b665ffc14fcd2ea6bbe9ce1d4d0f4071b
F20101204_AAAQSN lane_w_Page_090.tif
7657eda03a9dae98dfd12eff072987c6
744fb8985e0e97036b137e8fc9f68c317e649641
F20101204_AAAQTD lane_w_Page_112.tif
c0e9bf55ef507065c9ce165004a9dc37
00889e1c43b146e6a20387ee0f3bf6599916af8d
F20101204_AAAQSO lane_w_Page_092.tif
ed14fdd02cf756ea7ab3c605dfbe3e31
47c6136bb69634e715607f9c7e6fa9b7ee1ed71f
F20101204_AAAQTE lane_w_Page_115.tif
F20101204_AAAQSP lane_w_Page_094.tif
261c47f263e53dccf5fa6191427423aa435592a6
F20101204_AAAQTF lane_w_Page_117.tif
e538fdd54dbaa24677bd5545bc282e02
a7f0ae0325ef0fa10dbf1af5c6576ac1e3161a76
F20101204_AAAQSQ lane_w_Page_095.tif
e4a4f69cac0a08e294db336bf962f95e
09a052016aef483731590af659a4db9eaf236c7b
F20101204_AAAQTG lane_w_Page_120.tif
60d847976fc93ce38c15062fc849e452
ac470c40cecc71592a7f85892bd0e33e71ce0fe0
F20101204_AAAQSR lane_w_Page_096.tif
1f6f7f68ffd4e6b0317e31a3124870ea
3ca7e1c31e96c0999f4cdf5c7ce2a537ddc07ce5
F20101204_AAAQTH lane_w_Page_121.tif
70371612f14f2a5fe971895a17811752
f62610350d127c3f2715cb9ca00b24d9dc3a2057
F20101204_AAAQSS lane_w_Page_097.tif
734de025f3aec407bf43cc2fd76a056a
476f40d123572b73cdf971fe8aa9b258a2d2a97b
F20101204_AAAQTI lane_w_Page_122.tif
a430d5e8599e0599ce6293ca7100a88f
b3e3a4c360dd33423abfc4783713cd4a769d1f3f
F20101204_AAAQST lane_w_Page_098.tif
e3725d2dd8215b4c6366168969ba2dce
e438a609ce7daab2f62531caff88ae8a4e560cd1
F20101204_AAAQTJ lane_w_Page_124.tif
f2354efe301c0c0400ced8cee66f342f
F20101204_AAAQSU lane_w_Page_100.tif
74cd76e075267fd7e3920c60b3911ee9
a1c2efc420815a2e41c1454aa4916fb770f83fd8
F20101204_AAAQTK lane_w_Page_125.tif
0d4153af07eba65332d552e6d857e8a6
c6bb5b7b028b60355d3b10e32b442d9c060419f2
F20101204_AAAQSV lane_w_Page_101.tif
1da2ca0752a5e9b7e9a43cfac5c26f5e
F20101204_AAAQTL lane_w_Page_128.tif
c89a493e6d5b04ebf3dba175f6dde9df
afd71d1808ba0681829a05b1780bb6501092a42f
F20101204_AAAQSW lane_w_Page_102.tif
0515b50ecd8d14e9bf7200029d741bde
ba62c8cc7a1ce1563eb7c3c44e350596047ee5cd
43787 F20101204_AAAQUA lane_w_Page_017.pro
1e662e4c46effb4cfbf083e6b00e296c
41559481ae47ae66837dbdbaa356b1550ef084ce
F20101204_AAAQSX lane_w_Page_105.tif
fb0181dcbb8d67c7516c18df2c5c6c3c
72414939f70f815125b54fa51677880497341924
46544 F20101204_AAAQUB lane_w_Page_019.pro
63b7dd91f563a78b001f389c1892c34a
F20101204_AAAQTM lane_w_Page_129.tif
9af2178b0513745c9a40fe2001fcc675
2dd359cd6f3ac7608d658c246eb8bb9727909340
F20101204_AAAQSY lane_w_Page_106.tif
e19581ac25e8ce2a31fa96e072aa9880648a2aa6
36083 F20101204_AAAQUC lane_w_Page_021.pro
15408cc752eacc753a37c54a16146102
56d30e4fb6dfb0d91b9e3150c9d7b130593d6635
F20101204_AAAQTN lane_w_Page_131.tif
f4c6c5e1b4e505184183837c344a6f49
F20101204_AAAQSZ lane_w_Page_108.tif
91c9b9d714de18d0392ca3e3527c1b13
f2827b4db0d0d94d4f0a1c81e5c5c955efca69f1
48060 F20101204_AAAQUD lane_w_Page_023.pro
49d5a0ac29f30d2a7d38a49ea53266f7
2f55289ac19d564417d902f426ac74ddc40423c0
F20101204_AAAQTO lane_w_Page_132.tif
957c3a68d7ee959e134a2858a75db31c0d2a4256
40795 F20101204_AAAQUE lane_w_Page_024.pro
b3753e907636184646149f75f206a0ac8509c43e
F20101204_AAAQTP lane_w_Page_138.tif
6cc485353c48695dc8221be73a79c67c
35603 F20101204_AAAQUF lane_w_Page_025.pro
884302d60dceeec9636d72a8e3561eb9
F20101204_AAAQTQ lane_w_Page_139.tif
14ae8ee5d5bf2e0881ba91dc854e984b
86eff9ccb3c2873b198e60d16b36f0284ed4c9a2
15663 F20101204_AAARAA lane_w_Page_010.QC.jpg
4727b5ea6a606bb32e743588900e9965
3c1f218469bae583db78183e35201128e4697d06
43211 F20101204_AAAQUG lane_w_Page_026.pro
c5cb0b00c0eed7d73c9b156c6bce464bef850c0a
F20101204_AAAQTR lane_w_Page_141.tif
824c6a96b973e38f38561af6908a3eef
5073 F20101204_AAARAB lane_w_Page_011thm.jpg
2387260dcd3bfed7e6b15f2d2fc8b5a8
01e40663be11dd75e1cf1b997fe4f90be947c029
36197 F20101204_AAAQUH lane_w_Page_028.pro
85d611cf7c24573a04458b89677a94fc
d9f9988dbe3dc0c42c97be3303995032e48bd741
F20101204_AAAQTS lane_w_Page_142.tif
60630189fa8e3783f5ffea46c5755208
5896 F20101204_AAARAC lane_w_Page_012thm.jpg
82b9edd98dd9a2f9916d848ba02849e3
6cd1db9923d09c1766416977159658cc984e41bb
21547 F20101204_AAAQUI lane_w_Page_033.pro
c2d2045e54b52e8c186edcc286c49a73
79dceb58a098e1f7e70b838b20b73331bf760a6d
1506 F20101204_AAAQTT lane_w_Page_003.pro
1cd3e94245f588df7fa98e5d21184ff5
a714c6c2c3ffa6b72faf98edbed5a107fc3f12b6
106391a1fd13beceec6624f7acde44a6
62bab6d23831efe6c343e530ea35eca85d66ccf5
31982 F20101204_AAAQUJ lane_w_Page_034.pro
61a733130a2fe0d95334bee2dc3ac0a5
5d2a12d56083d7667e09f11be069397fbcf8db9d
53684 F20101204_AAAQTU lane_w_Page_007.pro
2090eca6c7ef533e6a25df08b4a9a7fb
451daf3bf97093cb660b004459ffdc1e3c07a678
21560 F20101204_AAARAE lane_w_Page_015.QC.jpg
5f523579690abea808c69a9745b093ef
19c0ceb9d1bfb18a554fda6bafd7820f0d30945d
57340 F20101204_AAAQUK lane_w_Page_035.pro
f2a90a95f03b6da12ca28c3bd4ecb050
30965 F20101204_AAAQTV lane_w_Page_010.pro
94fed799a00cfe7d23333b9711250ea2bc464c44
25869 F20101204_AAARAF lane_w_Page_016.QC.jpg
074e55f5ca2b92a3c0e5f7dc952429df
51924a32c94623e4e556bd22a6b8084cfbebaa1a
15686 F20101204_AAAQUL lane_w_Page_036.pro
68e8eb25352d5a87e946f93217fda7c3
49a2f0a322ec6cc7792f8548e2b9d2e6786e0180
49607 F20101204_AAAQTW lane_w_Page_011.pro
c3f543e9d106aa765b239599d418061d1fabe949
20955 F20101204_AAARAG lane_w_Page_017.QC.jpg
8377c2679ec10b90fbfd2f6066d11430
bbe5cdd3e87d429d36b0fe4fd0121eef251c1844
52639 F20101204_AAAQUM lane_w_Page_040.pro
12e431728cbe0b4e8f10a3ce525a33a8
25d0cc78b35b3506e616a3909e10685f36e27c3d
55402 F20101204_AAAQTX lane_w_Page_012.pro
e307ab0d5502407c4abf1a7c473dc3715509d6ac
49954 F20101204_AAAQVA lane_w_Page_073.pro
133eb7d0b27fffa3c4e8c6de77fcdacd
4806b4a6cb6397ff4b48db9eed44015c64e77fb6
5206 F20101204_AAARAH lane_w_Page_017thm.jpg
7e50eb585bbd0ab307cdfa6af6749850
e4946db2a26ce6d3dba206b2ba93e44f433ee2c6
29932 F20101204_AAAQTY lane_w_Page_013.pro
f6f4b533d28083513eae9d25585a330d
43a98b85838ec85974cae4b22a16585cae3795be
49757 F20101204_AAAQVB lane_w_Page_074.pro
f9d8610390931a004da4323072d7fb9c
9ac8fa56bdb5f1ac7eb10638e0e292596452c8c3
5321 F20101204_AAARAI lane_w_Page_019thm.jpg
a0a4594ec8349467ff9308c3637d0c24
88632262401f65c1a99c72b65ebaaa7f05554969
54012 F20101204_AAAQUN lane_w_Page_042.pro
19a3f28d52e1ef4261e1b16659bd6608
53531 F20101204_AAAQTZ lane_w_Page_016.pro
222c1065198a0dc83ac2275b2dfec0bc
724aece4aa12c145c5daba6d49e2d9f1c7461f23
57197 F20101204_AAAQVC lane_w_Page_076.pro
3912e20e5591d29a44da9327a04930707342fc14
26737 F20101204_AAARAJ lane_w_Page_020.QC.jpg
5ccea7c3b6258a3dae67082668902762
9ba35820f98e1b7719ded89660162cbbe122e56b
34567 F20101204_AAAQUO lane_w_Page_047.pro
a844d782b9e9d7ac819e06fb985e80f4
3ebbe19ce646ce2b6972fc28e59714a3687c9ee4
60210 F20101204_AAAQVD lane_w_Page_077.pro
5b7685d27e8304635a3b511524ffde72
d458884883bac6dc76f5dfe1691f27a68f3d0049
21831 F20101204_AAARAK lane_w_Page_022.QC.jpg
19ecdc20ca784413e6aece587c8b2aec
0a997baf1e1a4e4593ac8d66d18512930183dd08
32530 F20101204_AAAQUP lane_w_Page_049.pro
06947b4b7081f9c719eb6c2b5130b5b8
24d0100b7dfe1a1b07eb17935e436bf3d6c18a17
34189 F20101204_AAAQVE lane_w_Page_079.pro
998ed4743b6235e3146262b15e51cef8
49b7423639d838e945421cafac68d98d2c825140
5266 F20101204_AAARAL lane_w_Page_022thm.jpg
0234d27ebccd5a75e58c7b39f6149f5f
ca53c3701354478f639d4392b169f591183a4a81
27368 F20101204_AAAQUQ lane_w_Page_052.pro
605005fbe375483461dedc24e19628a9
a9aa2078b865c1f8a910ebe0e879e23acd860982
60225 F20101204_AAAQVF lane_w_Page_080.pro
d63c352e09f82b666febb2c2a669a28b
8653670b14895012d6c339ee6ab827e4c7bcbb47
18958 F20101204_AAARBA lane_w_Page_039.QC.jpg
5b61abedda421f088fe211bbb5486ec6
89138361094e0ffcc7f26f3780a28d0d40152259
22958 F20101204_AAARAM lane_w_Page_023.QC.jpg
9e72b9e9e2d5d06abaa305e6d9e6129c
37599fe16b471d53f209fe117b2a4510dec588c7
50812 F20101204_AAAQUR lane_w_Page_053.pro
a7ea311cb71bc68df9b3cf2d4c828562
a9a8454e3ddb10db1eebe6a0afaba4eced7da4ab
45470 F20101204_AAAQVG lane_w_Page_082.pro
ac8a97b64d0a3da6d57a24975eaba3be
b4ac98240e6b06dae7f5dff2051f07a8be9810ae
26133 F20101204_AAARBB lane_w_Page_041.QC.jpg
04c08d479f7eeedd3f69fd75e1464497
2ef7ed67ec3e56f8c1ee455db2b6098297d91823
5496 F20101204_AAARAN lane_w_Page_023thm.jpg
13a17372d1fb4a4b5cd60feb1a64d6b5
cf3a4a3fd72b22e3bb85e4447c6ff94cfa3658de
62355 F20101204_AAAQUS lane_w_Page_055.pro
750963e5afde4d79203cd61ae400a2d9b8cb3fe3
28304 F20101204_AAAQVH lane_w_Page_087.pro
a273de958767d687629a99aa0b0c4501
ded49705f38bc54549ba38e606f7fd18a095a656
6285 F20101204_AAARBC lane_w_Page_043thm.jpg
8ea7397e2ca26801fb5be6b89e0dcff5
724fa6cc619172e97a681ab92152f1a40375a0d0
4668 F20101204_AAARAO lane_w_Page_025thm.jpg
c9f6febbc1555ef4792867e8e691323de2af26d2
13686 F20101204_AAAQUT lane_w_Page_056.pro
be2c6b0bc8de6ff0660c262dec57a90026af743d
12585 F20101204_AAAQVI lane_w_Page_089.pro
cc1d200fdb62e940e96a5fcdb7c4f23a
a5be0172f523a7274e85dcb045beffab2491508d
26653 F20101204_AAARBD lane_w_Page_044.QC.jpg
22129 F20101204_AAARAP lane_w_Page_027.QC.jpg
d955217306b761fb828b1b5826a53dfe
b4a772ca55c00c5f64a26dd2d4b573068dd38404
33480 F20101204_AAAQUU lane_w_Page_058.pro
ecd7a20cea430bd95c8c7e0c6199090d
c448b1cdd873daab966267d39413b882cd131f0b
36452 F20101204_AAAQVJ lane_w_Page_092.pro
77ae9784005bc5f84bac3387dc79d2c1
61cf65a37739652bf29ed2ee1222d62fe083ebbc
6069 F20101204_AAARBE lane_w_Page_044thm.jpg
3d20591e5cd4b41748041a3b635bdb09
24777 F20101204_AAARAQ lane_w_Page_029.QC.jpg
c3739e184d928c61cf7228a5ec87684a
44027 F20101204_AAAQUV lane_w_Page_061.pro
fdb6cdea44309914ee39f5f0a54f164d
feaf6cd54f31cfa06cc63b5298867f366d6d1a3c
58699 F20101204_AAAQVK lane_w_Page_095.pro
c92fd5bd65185c68825dd998323d94f1
c170065f9146433be8cdce2e79fdcf0ceba95811
5770 F20101204_AAARBF lane_w_Page_045thm.jpg
dc036fe9b85aba2bd15016860d1f40e7
6bd52f7f41de240f9967b6c16c13b5589736fe27
5745 F20101204_AAARAR lane_w_Page_029thm.jpg
3bdfd71681b56eff5c6d3a783a2d502a
52207 F20101204_AAAQUW lane_w_Page_062.pro
56d57ea38c5b9a5b6765f5d9128fe1ae
52c88497a95a0bcd311eef420513bf35a03dd459
31986 F20101204_AAAQVL lane_w_Page_097.pro
f82d1fd69a7553a89ab0a80193fabf80
8d8693a5b2785a68f289b35118d52bafae623e0b
23599 F20101204_AAARBG lane_w_Page_046.QC.jpg
8e10331881517aa308cf8b36c3c0fa5f
e73df1592f9a6d1a049e5fb22b3ebb0d8fb46ae6
27256 F20101204_AAARAS lane_w_Page_031.QC.jpg
c90afc55edf195e529ba73b2afe0dd5f
34f1e7b406665564d6dc3d7518b7ab6d567576cd
30233 F20101204_AAAQUX lane_w_Page_065.pro
a86e124f909454ddb85ba13595639241
8d5f80ca63a01662d3f557ee3f27bf5f87b9a55a
31959 F20101204_AAAQWA lane_w_Page_128.pro
8a2f7b0106a3e0e5a5bbdb9acc263bec
8363a9e2e4812bfb4aa28b514d5818bcd5040481
12878 F20101204_AAAQVM lane_w_Page_100.pro
3606a501ae74b0835e98ab6a7e36bcfb
d47cdf18a320b15ddf4c68039eacf3d2fc5f9687
6109 F20101204_AAARBH lane_w_Page_046thm.jpg
65d17a5fbdb4d2a26ddbf1244c0c3b1d
86db0f0b72225c1ce5d8d0c0d92e3f7bde38d76a
24492 F20101204_AAARAT lane_w_Page_032.QC.jpg
33b86393f32f79594cc4496be7a42147
89e38c2682fbeecd8f42573404fd49dfc268a956
27474 F20101204_AAAQUY lane_w_Page_066.pro
167d379f2449b264a69028847c2b684a
66479fe5fd96582198e2064173a4f9d55a96374f
41614 F20101204_AAAQWB lane_w_Page_129.pro
d7f16aa50053b47ddeddbca5a0249b78
5162fa1692ec8b18740583e4fffabece9b9e0cce
40093 F20101204_AAAQVN lane_w_Page_105.pro
bc21d8cbb67b2b7d9b7001a2641ab789
74f375108ecf3cd12f1f99cab8fc4d16be72cb9b
F20101204_AAARBI lane_w_Page_047thm.jpg
5c0e6bd598a6548b2ddfb0a5e47937d1
e4b1367891170863058b71b46e9ed48b42109a62
20076 F20101204_AAARAU lane_w_Page_034.QC.jpg
b5999dc97843be0be289482cc7830734
45ca31d051375c93f0b1fb6347f4c3ef994b03e5
20080 F20101204_AAAPTA lane_w_Page_131.QC.jpg
f85766b0f04f96d8baf18c9ee2071aa6
6e899f4ee89faf799274b51e7fc8897ab7a7d5d5
36261 F20101204_AAAQWC lane_w_Page_130.pro
eb60d4a7406ebd85febab749b021f6e58ecef967
33820 F20101204_AAAQUZ lane_w_Page_068.pro
53393eb456e81ed1e0c49cc379b5f906
670cfd10377082c054898d58abd9710875fc4f99
6111 F20101204_AAARBJ lane_w_Page_048thm.jpg
441322961f8ea5cfbefb127d898fa657
e16a50034e9dcfb8a965172d34c9a49af3b9983a
5711 F20101204_AAARAV lane_w_Page_034thm.jpg
89e082b5e3c29442d2cda71d32185f45
b2b66d38a20dbb0163678a9c5f7a63a38b288179
50145 F20101204_AAAPTB lane_w_Page_132.jpg
d7cd5fface29c4dee0819c02970ba585
2f8d951324648f8fe9d374b984d319187d668575
34318 F20101204_AAAQWD lane_w_Page_131.pro
a12e362f591130122133b219aea19cbb8ea1a39e
48946 F20101204_AAAQVO lane_w_Page_106.pro
73913ec8a6f06298e3eb9e67148a89d4
2c710081ca45e29ce454afb374aec6e52aecf6b2
24147 F20101204_AAARBK lane_w_Page_049.QC.jpg
667522a6210c3187ebd43f92a6dd4a7a
0bda3b4563242cd495f51fe41d4c2a3a49f3d15d
5971 F20101204_AAARAW lane_w_Page_035thm.jpg
4c406f142534d4044c7263f7693221b2
aa0164afacf25ff38acdea08c3ba4f8ea65776ac
1280761 F20101204_AAAPTC lane_w.pdf
4f061c75d7d02daef610102159b1cf46
a1dabd7d21e09a53b703802b04d565a4b542cb39
56270 F20101204_AAAQWE lane_w_Page_133.pro
6ea1f20b1e6046f298d53ec8dee3b4f6
8bc089155c103df0f56b0e8d22d57ba3ff7f3d5a
58934 F20101204_AAAQVP lane_w_Page_107.pro
2d8a389e01595b5b55a7deab5aaa78ac
9fa614442148b2f418d518b36fa6d840f45e1b9c
6495 F20101204_AAARBL lane_w_Page_049thm.jpg
e27f724a35cfe8e9802172effb10ee8d
5bfab82ec4ee6499caf80c516a059da9608ea480
5399 F20101204_AAARAX lane_w_Page_036thm.jpg
4e523168c710a344b1e0158f9e6a3c17
7ea4dd7040011273cb8ca10fd327452cc3298c1a
1783 F20101204_AAAPTD lane_w_Page_114.txt
6c8854e27e529848d7b8fa246562d727
57550 F20101204_AAAQWF lane_w_Page_135.pro
84e750fb32ff8d216f03bbd6c951bce0
33143 F20101204_AAAQVQ lane_w_Page_108.pro
2f658a77ddf118458affc1023b8af17e
1999d49d939c6e126ee043584bac278ebd3b107a
21946 F20101204_AAARCA lane_w_Page_060.QC.jpg
27be8320d119dfaa9424f2488d95ec7d
6887 F20101204_AAARBM lane_w_Page_050thm.jpg
5be449a5ffe53269e0629e389e34017b
89b9ef10482cf99506f158f4e653399f74f18653
88802 F20101204_AAAPTE lane_w_Page_038.jpg
4902be885de44be5d6df868c44a308da
3107 F20101204_AAAQWG lane_w_Page_137.pro
13e72a961bf8395f67c27d622a53db6a
da72733b4893b739303a029e7182d7f29f1b28eb
34248 F20101204_AAAQVR lane_w_Page_109.pro
eb05e416ce52af9c1cdc265858821d52
5b1630173ac2a5df425d83db124b10b7dd17b1e1
5676 F20101204_AAARCB lane_w_Page_060thm.jpg
46c39198df316328d3cb643921f3d7f4
8af454f941734ef34bd29265ea58ab9cf35920a7
22075 F20101204_AAARBN lane_w_Page_051.QC.jpg
7347b95377f1de20f9036a16114856dd
26497 F20101204_AAARAY lane_w_Page_038.QC.jpg
363b0b7418bb08329b6e2d65c89136e9
8a596287775962dacbc9276725731d32aa86fa3d
18929 F20101204_AAAPTF lane_w_Page_081.pro
1f169d22e4fd1a382f5e135753453936
5384513123c0f02285cdc92c4646673b6559c5b9
116 F20101204_AAAQWH lane_w_Page_003.txt
83bd9c4f09d3b1ec3f85a88722b343de
de8679a0e4832c3d49ce82be605a4b93b0a5f4ab
35463 F20101204_AAAQVS lane_w_Page_110.pro
db0da5020f35c03ef99050bf8d138b31
c487bc167c9ea7229cd60c722380e8b459f3e6d2
22272 F20101204_AAARCC lane_w_Page_061.QC.jpg
8cf528006cedf3996c372d8a96300fed
b8aa96a997b45a90d261ed7a17a2b4a4b0e9cdd3
6114 F20101204_AAARBO lane_w_Page_051thm.jpg
a7f5a5ae4578e6d1d9a301f7488aa255
f25c1d653421f6c7054b34c926706cba1f1438e3
5941 F20101204_AAARAZ lane_w_Page_038thm.jpg
b0f4043d5c9687d400069db81ac01899
0a63c309742e70ccd66f2ffe7b93e93c0922d34e
59137 F20101204_AAAPTG lane_w_Page_124.jpg
4b1580260f8470f09c1a96bee0c0ff47
1917 F20101204_AAAQWI lane_w_Page_004.txt
4e4b9afa2ca3fa870c2985c4da6d125b
bcd358984085092a1c59af63006acbe2ebd33cc9
30010 F20101204_AAAQVT lane_w_Page_111.pro
383faf12d4eb1ba7eb64738e82839a00
5464 F20101204_AAARCD lane_w_Page_061thm.jpg
d871a09f7a7711f9fab3678bb160a30d
25868 F20101204_AAARBP lane_w_Page_052.QC.jpg
dbf024696b1c725da143a7456fa2e0fd2e4e455c
2159 F20101204_AAAPTH lane_w_Page_057.txt
f8aac7efb74e5431c07fc87cb7c5f3fb
f1e8c59340d85eb2c07340ac3c5784e12d291d53
112 F20101204_AAAQWJ lane_w_Page_005.txt
8895c185251dbfc48ee60967e2859fb2
33395 F20101204_AAAQVU lane_w_Page_112.pro
da00379081983a11e9e63c3129791daf
243761e70c6691bceab7e6d694fce4ecd6b38b3d
28646 F20101204_AAARCE lane_w_Page_063.QC.jpg
00b5f23acbdc8dbd223e3fae9091890f
522c88139318b76dc87f0974bee58ff9611bca25
6655 F20101204_AAARBQ lane_w_Page_052thm.jpg
464479692f48b38ecf255d9c45cd8f4b
1f2357e97d1ed4e5a4503c8a5ff1e637445701b6
73316 F20101204_AAAPTI lane_w_Page_119.jpg
34ff0a8ea0c3db86b7b4dfcff57c4096
863343d310d6a5cd9c46c8bcc19663f0c3db0fe6
2768 F20101204_AAAQWK lane_w_Page_006.txt
8ef9bec7e26afdf0c5d6f033b56aa76e9c9f22c1
38358 F20101204_AAAQVV lane_w_Page_113.pro
560b52908833a0f4026c09e0e0120f14
F20101204_AAARCF lane_w_Page_064thm.jpg
6f5140bb8d1f79e3089b6f75d8db6325
47f234645679378befbcf075546f359beeb5a5b4
24577 F20101204_AAARBR lane_w_Page_053.QC.jpg
9790c13009c829749a65b0d5e5146f6b
16075 F20101204_AAAPTJ lane_w_Page_037.pro
ed42cc3f5f9b6713ea083596963ec2d0
f457dd8f459d954e0225d70305edf77ae8ee7ce4
2549 F20101204_AAAQWL lane_w_Page_009.txt
db925c5b6ca8ed2c7bc6da1d395a7cf6
c91cf76e959f00dc9cffaae5d80c66fec4336fac
39334 F20101204_AAAQVW lane_w_Page_114.pro
3a6a1d5969afa24122cc288fa5059ec1
d2d551d8d44dc038423b3826cb455a687e119e38
5471 F20101204_AAARCG lane_w_Page_065thm.jpg
d59a6954e85bebf1888a37f906be18e7
41f141f1ce7da6067fc02f1033ae557c05f43683
26968 F20101204_AAARBS lane_w_Page_054.QC.jpg
312c3aef489e1c9dd780c838729d9a5c
c92e7b08a8c1534932159da8dd06e68a4bd8a85a
F20101204_AAAPTK lane_w_Page_032.tif
3d046da0091bf49a6e376c935a5ae856
28a602dcfd1d9299130ef7a121e4090d6172467d
1294 F20101204_AAAQWM lane_w_Page_010.txt
c28b408f4a79610f747d8debcdbb0fa3
35728 F20101204_AAAQVX lane_w_Page_115.pro
1554 F20101204_AAAQXA lane_w_Page_034.txt
744fda7d03341aa7d4fd3250295a7e16
F20101204_AAARCH lane_w_Page_066thm.jpg
070fcc062d4d8c6cc9811f11ab80c7bd
665b7e77a3155f31ea9056d4906cbdd3d2ea932e
29037 F20101204_AAARBT lane_w_Page_055.QC.jpg
27742564758452babaaebe4898230311
4626 F20101204_AAAPTL lane_w_Page_090thm.jpg
de6b71bb89cd960b8e3f9574c33e5242
bc6b08d9591decd65c681a8e0d3d7dea24c08eb3
2121 F20101204_AAAQWN lane_w_Page_011.txt
b6d1b52d41e4d1c5d368fd87653fdfbc
bb723d94c655357e5a6f0f8dedc9ed2909071c35
20688 F20101204_AAAPSW lane_w_Page_001.jpg
892c86551387a67413c22fa4264d015c
70814ebb0634f84a1b3f3772d4420c705cbe93c8
41848 F20101204_AAAQVY lane_w_Page_118.pro
dff7105b06acb1c2dcc29de6e8f3003b
700af4cf049d528e0b8d30884a48b35ea5342c83
2253 F20101204_AAAQXB lane_w_Page_035.txt
5492c585c0a83474bc0a49a2630a17ca
8e0acc9928c566deb1200901a95e96150bece16d
4119 F20101204_AAARCI lane_w_Page_067thm.jpg
aca268428e6881c8793fa9267a2322c7
bbfcb6dc7381f7258349bbdfe8066f915ce97e7c
7882 F20101204_AAARBU lane_w_Page_056.QC.jpg
844bfa0dd24b7cf688761fa13d591fb7
02e5b39a65895959d2d39e10e6c1539063242291
874 F20101204_AAAQXC lane_w_Page_036.txt
6e538ed101611c73ea48aa6eb253c58d
73ff08467915dab04cab7560f225122b8b7fbcc6
437 F20101204_AAAPTM lane_w_Page_002thm.jpg
0e6a54231279f16a98b040a6aa6be019
68dafc8baa5c993bffde9dce67aa25b5423a287c
2282 F20101204_AAAQWO lane_w_Page_012.txt
2def498aa1b6d633ebf2f0dff272689e
d5300ea9d6c38cf93e0169339d5a5283033e5d1d
17936 F20101204_AAAPSX lane_w_Page_025.QC.jpg
e1f156331696b1f70b83eb4e52e56e30
1d6576874718df66ea68773ac5ea9c8388fdc674
38392 F20101204_AAAQVZ lane_w_Page_126.pro
65613 F20101204_AAAPUA lane_w_Page_009.pro
e3da9b1447943c5d6a5905c8ae2abc3d
b596f13153c16678421a8348cac444565566147b
17119 F20101204_AAARCJ lane_w_Page_068.QC.jpg
df0bc534230de6c4cd2a1f0317bb21ba
39cf06b0cb149b487f0a16ae7b02db8754ab3bf2
1952 F20101204_AAARBV lane_w_Page_056thm.jpg
f52806984481ee9e79453ddec20db476
690 F20101204_AAAQXD lane_w_Page_037.txt
196eeb356549dfcc490bc70b49d84a89
57f9bfa64c5fb9918792deebe8e54886ea915286
F20101204_AAAPSY lane_w_Page_093.tif
3cd90f431530aab8565678477546ab3e
F20101204_AAAPUB lane_w_Page_092.jp2
4a9c9cdf603dced9d54d15af1bc36ae9
3553 F20101204_AAARCK lane_w_Page_071thm.jpg
4680266dbbcfd3a7caeb31367f1e7c45
7e7ec3da502ae429461c5d85e6639dcfa956dc98
24849 F20101204_AAARBW lane_w_Page_057.QC.jpg
94af04566bd41f67b94d889239b61e26
97b9281c1575c7d8bbc2da9e414b6598e7fa0189
1205 F20101204_AAAQXE lane_w_Page_039.txt
53e162e8d2a3f22277b5a21357e23b70
a10f8b300d79c5d870f2ee803a7b94334ea61c04
F20101204_AAAPTN lane_w_Page_081.tif
c7a8b221408a8527c1b5a3da2ec9f765
810c555d500e50c780526b949552d022e9d1d43a
2230 F20101204_AAAQWP lane_w_Page_014.txt
a0c4e04eac921e501fc11cd59b0caeb1
a58b814704a1c36798c7716512ea6e50974e4f6d
769065 F20101204_AAAPSZ lane_w_Page_021.jp2
fa9c41f2626f413e759cab858112b22c
f328d71ed9829527c76286ffcb680032af23413e
6446 F20101204_AAAPUC lane_w_Page_123thm.jpg
07d5a8c7dca0bddf1e4d04cbe53fb15d
79ca1a9dfb750236f52ed35f42e959f15a1fa3ec
14537 F20101204_AAARCL lane_w_Page_072.QC.jpg
ddffeda96cfa1022db0d077169360069
5809 F20101204_AAARBX lane_w_Page_057thm.jpg
dba28a1d711360584a9ba35f28a88a1a
efd320f58bd7cc64c2e285348cb015fd5900c173
2111 F20101204_AAAQXF lane_w_Page_040.txt
d56517179d10aedc315ee8a98978f4b0
604f1f438120727ec812ef6574ff7d3bd9d0fdae
52329 F20101204_AAAPTO lane_w_Page_029.pro
825cff7c92017441d9f17579212c7986
b031efe87a3cd2db3e6256ac8b81e9cdd15c9bdd
1946 F20101204_AAAQWQ lane_w_Page_015.txt
a0c53281b149b7f7a8b5c0ebf277c4ba44c88b28
105817 F20101204_AAAPUD lane_w_Page_011.jp2
8461625920e4702b73779e333a107ebd
a2ee5c94132031c82c28b994c5ffac0f0ecf7740
5193 F20101204_AAARDA lane_w_Page_084thm.jpg
cc07fd4ae42115f4565df1c40ebea0a4
520e439b3f1dd9cc058c79387ef8800115c83daf
3871 F20101204_AAARCM lane_w_Page_072thm.jpg
d0fe7ac35a455c3e11865157d02f2a1e
17934 F20101204_AAARBY lane_w_Page_058.QC.jpg
a538597eda06289bfab02f32454e4900
1ca880f1c2bd1612f73fa8ebd21992b98eed000d
F20101204_AAAQXG lane_w_Page_041.txt
2eac15490d6c732e6475598cd0265f6f
F20101204_AAAPTP lane_w_Page_117.pro
4860023b3132e68daa53f1bbc5a906a2e74ca248
2102 F20101204_AAAQWR lane_w_Page_019.txt
d9310cb48ac14109815768a13900e1a80ba1bc71
5658 F20101204_AAAPUE lane_w_Page_032thm.jpg
82e45965b4722847b6bfdef891965f80
cb259409b95a69bdc8f0c06d1fb5a0c72728b179
16806 F20101204_AAARDB lane_w_Page_085.QC.jpg
7f57f93e1dd1d67fe3af456b49fd8dd2
f63520942ab4c4deff9d48cbee22bf42b376d40f
24367 F20101204_AAARCN lane_w_Page_073.QC.jpg
a20504c32f1ecff9c8b842acc03bf5e6
e05e203114550707412b2402b9535b3ec589480c
2241 F20101204_AAAQXH lane_w_Page_044.txt
ba74ce1f41d10dfb5d9c215b8aa110a7
6e3e54dd97af7be5177afa1b8e814210cc8ff2b3
46310 F20101204_AAAPTQ lane_w_Page_091.pro
b331e9c264170faa4ebc0f2f661a07ce
76d22fbe43d43c49d3353d69c765d5c9dbf1e962
1811 F20101204_AAAQWS lane_w_Page_021.txt
e9c3675f5e5c695772ae12e460b553fc
caa500770e4cea7ab0fff40f526b4f7410916db5
82013 F20101204_AAAPUF lane_w_Page_140.jpg
cfc79605fdc5569592391256e33f775d
5000 F20101204_AAARDC lane_w_Page_087thm.jpg
60429d79826a0c426ecfbafde6b24d57
d151a5a433e05aa6a4a037255343981ed924d1b8
5239 F20101204_AAARCO lane_w_Page_073thm.jpg
383ab0645eb8545878987311ef35c71d
363ebc5cea7911a84b72736bbe3dc90ced9665f3
4718 F20101204_AAARBZ lane_w_Page_059thm.jpg
fcaf8f3f5367f81dcde601a5dcf193f1
0da9298963ffb6b22e43dceee0658dee8b039486
1700 F20101204_AAAQXI lane_w_Page_047.txt
1b423ff03e9ff4d97894ef6c23c7c7e9
336e36ba231f428d5346aeae7bb8c9b7afac23b7
34874 F20101204_AAAPTR lane_w_Page_121.pro
33170a824d26c2bd463d476be8837f10
1997 F20101204_AAAQWT lane_w_Page_022.txt
3a2256f3836e950bd6ce2a8754fba7ec
91e342342935fb86f90dd42fcc0e9bbbf9d364fd
74212 F20101204_AAAQAA lane_w_Page_046.jpg
0f1c25eb79bdce367ae9b63efe8aec75c9196896
14460 F20101204_AAAPUG lane_w_Page_081.QC.jpg
2db4aa9e28f3f05c0105db3f282729ea
270a1c820461267b593174a77f84cef9dec4a475
17943 F20101204_AAARDD lane_w_Page_088.QC.jpg
431cfd345c54d92db6650a4e7fa22f63
6081b8d42fe9a970c70193a9748057da400f0f0f
5517 F20101204_AAARCP lane_w_Page_074thm.jpg
922714d47c9a8490f3b7c2654a7749ba
5814f25b4c2d636248059842176861d7d420f743
1805 F20101204_AAAQXJ lane_w_Page_048.txt
15bac9f0d27435310fceb147a37dc717
21a44f3907172f984cb689dfcd816e17f3336606
77482 F20101204_AAAPTS lane_w_Page_112.jpg
76f7353607342e71a4945b86276b8ff4
3bc0de3c468b4e69ffde03055fb76a4b312af96e
1990 F20101204_AAAQWU lane_w_Page_023.txt
09c680bd0f037da87e90a76c3aa93b97
e85d21d348d045554f7de445d609e115850d3fa0
35323 F20101204_AAAQAB lane_w_Page_070.pro
e879299c4847efc99d8b45d537f38d02
9c39e22b7869511a7f1e8d6fc2db13b280dc82f6
F20101204_AAAPUH lane_w_Page_076.jp2
4b6cca1e40ff431b42652cd26b73f7c0
de4e1b8cb8828929ba7d68ce60af5c5368936c2a
4711 F20101204_AAARDE lane_w_Page_088thm.jpg
b29f2abbaeca10cf3dbc701280a820d6
29c9ceb35178c1b5852ce46b2afcea10b6667d43
27297 F20101204_AAARCQ lane_w_Page_076.QC.jpg
c7e75ac6c60b7b99a48b6413c1671090
1514 F20101204_AAAQXK lane_w_Page_049.txt
44b980c5733ea048d55fa2bf21831c22284ba679
2163 F20101204_AAAPTT lane_w_Page_038.txt
a6a676a46fe6f426c4a7bef7217d3eb3
c5736022bc1be4e2b1f1b39b9104e5c687d71b9a
1760 F20101204_AAAQWV lane_w_Page_024.txt
cbb84a27dcda78e8ffb6c40aff0fe8fa
71c47ace367803f4b65a9d226d53753a0a0fb257
20569 F20101204_AAAQAC lane_w_Page_113.QC.jpg
0e49b26e37a0497645974890c0cac669
7321e718599edd4b721e8a169dc0f117b6c63ce4
F20101204_AAAPUI lane_w_Page_028.tif
0888771b2aff8703c5f9d519b265c7cf
ea29813500d7da5fe1c81588e098d7d874cd5cd1
13128 F20101204_AAARDF lane_w_Page_089.QC.jpg
57dae84b1e75e46f3859799af8880f60
6001 F20101204_AAARCR lane_w_Page_076thm.jpg
b21518f704ef99be033f553dff7062bb
7b168859ecbb91092aa9fd8631d63f0e568e86d1
2516 F20101204_AAAQXL lane_w_Page_050.txt
7c795e9fc5e332760cb623c00228a2e5dd1af10d
4589 F20101204_AAAPTU lane_w_Page_138thm.jpg
7c4549230e355b85c94c97c327b2c5e3
b0ac02c775f17511502d42cef595476a502a0c4b
1930 F20101204_AAAQWW lane_w_Page_026.txt
daa5629c7128317129797c8d852477ba
e96c1d97e6378944a6bd61953a214077486d9bb6
3543997c489a34734f3ac23cfb43fd5c
9abc41c1fd3ae09865218c4bb7230454f283330c
3941 F20101204_AAAPUJ lane_w_Page_069thm.jpg
d18d4bc1331e2f44c1931c74a0d12f9e
5379 F20101204_AAARDG lane_w_Page_091thm.jpg
96a55b9f6b5b404940ee963546fd8789
6106 F20101204_AAARCS lane_w_Page_078thm.jpg
980176ea0327d6156eec76f2bf07c70c
310ff800d0999de1ded7c4e74dd0da9a1e708e07
990 F20101204_AAAQYA lane_w_Page_081.txt
57289ca6a87aba28e065eb156715e177
48d6a9cbcaf143c1078af4000da03599c1dfd0d1
1239 F20101204_AAAQXM lane_w_Page_051.txt
570ba040e9a1c0d9a2611b2dd02eb749
b2fed0651768577d71560b3bf7f3667d0c649b73
58740 F20101204_AAAPTV lane_w_Page_134.pro
499fd990db5ac3860a8dd444625c71af
67ab6e9ebfca3be6c2c14bbde4ae66cb41dc51ac
1943 F20101204_AAAQWX lane_w_Page_027.txt
24058a96ca34ae4d74baeb9b84930f29
879115 F20101204_AAAQAE lane_w_Page_131.jp2
8f8e6735356109ffa7b9c3a2215692b0
bbe815530efa5635665976d6e9ef0b8717305655
94 F20101204_AAAPUK lane_w_Page_002.txt
e959ecd85f2ed98d692ea2b9153371ef
b79164215012c273165b288df01806814b5f69d7
5978 F20101204_AAARDH lane_w_Page_092thm.jpg
dbc10d771ab46427e610ab5bfdf013bdf2d6fc4d
F20101204_AAARCT lane_w_Page_079thm.jpg
faaa79d48a576c59b404bd15781c6d89
15e5fb19387c161676b4e25d67ffd5e6ce0e5d07
2050 F20101204_AAAQYB lane_w_Page_082.txt
27100b38c1833207a23ed94377a10f61
5484a7b122371b9337cd489eedce84647b9c03be
2004 F20101204_AAAQXN lane_w_Page_053.txt
8e078e1d3098c37be0aeac78e88aaca0
cab2510643dd2f3495a8c8355f2e009a602cc5c9
18641 F20101204_AAAPTW lane_w_Page_124.QC.jpg
570be33fba7d92c5886140274d585cd5
f805ff4c9b0ca2f4d013f2ec5c03a589d2e494ce
1695 F20101204_AAAQWY lane_w_Page_028.txt
5355f7e79bbaaef2ecac76260c4eee0d
30fe34f396d6f2ee7889ca397536f55bfbda7e49
43016 F20101204_AAAQAF lane_w_Page_048.pro
c4919ae9aa48a12723e2a47f3019b67f
18195 F20101204_AAAPUL lane_w_Page_013.QC.jpg
f1a23c115e2717a2ea22f53a9ba53252
7e9471fa3660819fbb15ab7d27c4c95c3156674a
27621 F20101204_AAARDI lane_w_Page_094.QC.jpg
834ef8b4ce8c7a4117b053e21bb918ff
2e43be73cdfe4defed0810b9dfd04dcb9e91e57d
26774 F20101204_AAARCU lane_w_Page_080.QC.jpg
5b15600af71fb918cb175ed1e8a6836d
c5e448e3597549fbf2ba099b87c8e2a99fe95f2c
1487 F20101204_AAAQYC lane_w_Page_083.txt
da6c1eb3ef94db51d0099511d6cb3b0ce125b389
2294 F20101204_AAAQXO lane_w_Page_054.txt
553ed00270574c7e501621cfb8aafc43
6573da2b8c7d309fb4167d804b245b232ef0c581
86451 F20101204_AAAPTX lane_w_Page_041.jpg
98d4b886a84f0a4b59baf767e517eb9d
9c1edb3deea3c5fa44867acc54f2b23e5eb0cc70
2307 F20101204_AAAQWZ lane_w_Page_032.txt
fb149fbe4c3e46e5b0631f8713f05434
a27e5b6482d86a0ea3ca048d3dc11ddb177625c7
28474 F20101204_AAAQAG lane_w_Page_043.QC.jpg
c042b3e58c84bb45430a42639e7e7ebc
3f7dd267bc446d6d527543b276a90a142f6fb6f1
F20101204_AAAPVA lane_w_Page_129.jp2
440277cce0d95a41d4f3c5326fc200b3
fb77eed0ee776320ccfce4f18dbb3b7e1d00bc25
2564 F20101204_AAAPUM lane_w_Page_075thm.jpg
35f3285cf2d1efcee36ec324854e9a71
6027 F20101204_AAARDJ lane_w_Page_095thm.jpg
174da24e03bf05a84d108b780e531aa9
38c701aa45fe24d412be7e86d23cbe580c6f88dd
5935 F20101204_AAARCV lane_w_Page_080thm.jpg
74bf1a7a0ed0c00f10d87e4009b2802f
1504 F20101204_AAAQYD lane_w_Page_086.txt
86515a43d5b4d09de4e3071cc9b174b6
667244267a7c12450f2646f38bfc366e2ba1038e
551 F20101204_AAAQXP lane_w_Page_056.txt
3c85bf60981eed81ab52e0b1b41746d4
8c5ec5aa8a60dc2390ab5cfa84cbf05d29eee5a5
1675 F20101204_AAAPTY lane_w_Page_068.txt
837db4cb5f36001286fbcd0f197fa027
778ff39678593ca65fb9697f4184e7a43548c05e
F20101204_AAAQAH lane_w_Page_012.jp2
94a787f08e198357813d16e61b693983
df1a73fcd3e6b2b21b6b54bd39141d2ebe782454
51426 F20101204_AAAPVB lane_w_Page_057.pro
d37fd760bf40a184b01fc41c7d4d97c4
6a13076b301d0ceac7ff2278740b4d08577e2c1e
F20101204_AAAPUN lane_w_Page_009.tif
497fde0dc59c643aeb9e15365fcea861
84bc17c551ee552f974ac34bb2cd8fb7c85177bd
21257 F20101204_AAARDK lane_w_Page_097.QC.jpg
8564fe800c414e068b99fa2dccd3e38f
79f34334b34a440760229291a28a5b9f4be62b7e
21942 F20101204_AAARCW lane_w_Page_082.QC.jpg
a77173c57c1859294dba45d5e70dc4c86d249212
1269 F20101204_AAAQYE lane_w_Page_087.txt
ac0407b790874242babf748212b3752c
073ecaca0ebf9951cf29c6dd64c7a8400dfa70e1
17830 F20101204_AAAPTZ lane_w_Page_086.QC.jpg
3aabd1e80d41b05c75147c12826f7d10
28cddbbb8553be9aed713f8400d62ae23e2683dd
1684 F20101204_AAAQAI lane_w_Page_105.txt
1997925518de05f0857126b34247f7a8
f2858fa020587b4898fc75eb3a2d2ab544584982
6124 F20101204_AAAPVC lane_w_Page_141thm.jpg
50d7a34715418448f9e904db6efb20f8
5202ff0cff349c454eb74bd5e0b391b36327373a
5305 F20101204_AAARDL lane_w_Page_097thm.jpg
09372f9accbab019bf73e5d10d2c270e
aec53c599952d163eaee7c4cfc2877174cecb451
5088 F20101204_AAARCX lane_w_Page_082thm.jpg
96dd9ddc890d7bba7ab630075fc32121
9a68e6a163d878213b3bd12be8009af5315defaa
1417 F20101204_AAAQYF lane_w_Page_088.txt
48e99d11b027d46d358b10c024cd0766
967295c5a6c850b9ac7458dd62d3f5cf8a4b97d7
1734 F20101204_AAAQXQ lane_w_Page_059.txt
424614c2480620023c8e45b96510293f
57cc01faa5f735e540d3f79fbfc48008a1360697
8512 F20101204_AAAQAJ lane_w_Page_102.pro
418c5c055c554e534ddf45214f03ac26
e5a4609b7e1994cea334890344af6050b57e2116
F20101204_AAAPVD lane_w_Page_134.tif
bca4c1abd3b4b0237bdfb901da26ab09
fac84ffc1f17d2f075620a850c1988cc2c601758
1415 F20101204_AAAPUO lane_w_Page_045.txt
52215e3745ff04804732c0e1aa1ecf01
8746914f24d23f7d80a5ddd038dbfef292fcd686
22879 F20101204_AAAREA lane_w_Page_109.QC.jpg
e82a55269f1a0c354ba93ec4093d0e2a
a9683bcb69d380c8b8f90fba80033a022ee8e15a
22064 F20101204_AAARDM lane_w_Page_098.QC.jpg
cc43f75964340853a9f00304b5196c25
62b965c50f32d58f9b5708839d52aa10f4c1f696
4511 F20101204_AAARCY lane_w_Page_083thm.jpg
40d5c608bcfee6a5ed8f17db6a5f51af
658 F20101204_AAAQYG lane_w_Page_089.txt
28a5b6dd7426aaf6e4d93947fb63d814
55fb0cff0162ecb851b388c2c2e9d83c3ae5d9d4
2096 F20101204_AAAQXR lane_w_Page_060.txt
601eb3df95247b5e7cab61e5eaaa93c1
24021e6c4dd86d64ae725348bd1e68ee6136c6a1
15611 F20101204_AAAQAK lane_w_Page_066.QC.jpg
75199c9a27203b2e1e9640a059526aba
1805c018b65c5b3d1aa3b271ea203ec1d5e4622f
1902 F20101204_AAAPVE lane_w_Page_069.txt
ae938ed717c6517f1b779d20bd035793
2351 F20101204_AAAPUP lane_w_Page_043.txt
a14cfb4b501cbdae71896dc1f2e74c10
0ee6d9579de34d2e275eb0a42507941feeffba52
5997 F20101204_AAAREB lane_w_Page_109thm.jpg
ea0a9901d45c2c407c4ab05e5de2e827
2878a69bd47268f0f974e74fa3e946fdc83742ed
2852 F20101204_AAARDN lane_w_Page_100thm.jpg
cdcf86a0fe9489e95dae79a2be8bebcd
cf3eb88cec8a0468f54a11d6d334a7c83e95490e
21539 F20101204_AAARCZ lane_w_Page_084.QC.jpg
468e8eab9c032e08464d85305135786d
42b43ef4c3d2baa7b56b3412d6424f41ae5c7243
2010 F20101204_AAAQYH lane_w_Page_091.txt
139cd18f195854961f14d26bf26322c81a31632e
1896 F20101204_AAAQXS lane_w_Page_061.txt
f9486a1b10ca56ca8169be32b63e5b95fcf713a4
F20101204_AAAQAL lane_w_Page_054thm.jpg
c5b58405ac4f756340d675825a687d32
8831f4050b7c2fbbd27e848f9b8601be22276f89
5993 F20101204_AAAPVF lane_w_Page_094thm.jpg
763b9cea0da3923ecd018947e95d77ca
3348 F20101204_AAAPUQ lane_w_Page_010thm.jpg
14abeee4d8c75c1a90ac4000fa469773
a7e711c4737394d92fe14d7a9f873e23dd70442b
22354 F20101204_AAAREC lane_w_Page_110.QC.jpg
6aebe8f211722cf18ae3a56b776816cb
efcf1b5fb76e93e5f3a869b20a16a1e95c90e057
8573 F20101204_AAARDO lane_w_Page_101.QC.jpg
b676a7b1235a3ea4b6818b4ae344acbe
3b96b7ed5ab81dc7433e4d5b128bab817523622a
1753 F20101204_AAAQYI lane_w_Page_092.txt
76357412bfffcb91b22aa2e55c8dfbd8
dd8978afa528cfe2b3f04b9e66513e0dbb33f423
1349 F20101204_AAAQXT lane_w_Page_065.txt
ef930ac2402a3ecbbcb07687d6c84e60716d155f
F20101204_AAAQAM lane_w_Page_042.jp2
d2dfa36d17b7998bf690ae35d6fe1ced9bb8d6b6
1598 F20101204_AAAPVG lane_w_Page_096.txt
6e9f9b9ec10e927ca6c16d979d933f2a
5768 F20101204_AAAPUR lane_w_Page_093thm.jpg
20d70b8a39df3c07a3a7dc4fe945631c
ef25567b3ed6ae4150db2ff02323cfe69ded1679
9793 F20101204_AAAQBA lane_w_Page_101.pro
b9e67d234773e9b6a42b9b07ca4a8edd
8d32b3352f1cc709f32d7e77b1cd576fdd621d09
5539 F20101204_AAARED lane_w_Page_110thm.jpg
6854a16c718d17883873bc1a710483db
a761c50a356180b9ae368eeec5c074273be31a3e
2838 F20101204_AAARDP lane_w_Page_101thm.jpg
9b20737b6687c94f38ba9673ec341229
6d67168749d8c6173d135efdec19425b4896242c
596 F20101204_AAAQYJ lane_w_Page_101.txt
3321290af4bc31604643580a0d18407c
5ec3a890a19c32dee1aa2a84222d02f13c8ea089
1335 F20101204_AAAQXU lane_w_Page_071.txt
132eca562e2a56ae0e40dbae4dd6876d
ffe5fbbc66595949f117413877793f4c20459712
1547 F20101204_AAAQAN lane_w_Page_097.txt
6ce747bd779fbdffcc7d62a86b0c8edd
d32c2136f3fdc084d696af4d0cac28981d79be21
2314 F20101204_AAAPVH lane_w_Page_007.txt
2075b4c93beef0e9805b813a36f43be0
ac42108068ac70c7a68e672c7b353da3b8298418
F20101204_AAAPUS lane_w_Page_091.tif
a71b7480f861978f32217f9bf5cc016e
9c591040be2873fa37b240558c67b8f06affa611
16400 F20101204_AAAQBB lane_w_Page_033.QC.jpg
c605495a4d96d307241cb022e6b7ff2f
167a168031866890ccec479d1f46f35f136c046a
24522 F20101204_AAAREE lane_w_Page_111.QC.jpg
393371428816fbb05c84805851890cb3
5b5331a261cd46b94752af898b5c234f38a52da3
10956 F20101204_AAARDQ lane_w_Page_102.QC.jpg
87fec24b7786af681f50889dd986eb12
ae7a4a75df96c4362fe42afaa308286283a33374
506 F20101204_AAAQYK lane_w_Page_102.txt
ec0d3ee7b50dab4044a6da6156533409
e8cd43f6c204cd8b5551f94da7903d47c0bbfafa
2022 F20101204_AAAQXV lane_w_Page_074.txt
078ef2dc3a0c39b0f97917ea98340ee6
47fa412040547f9623015a3f42c3062ee8211f5a
F20101204_AAAQAO lane_w_Page_005.tif
1a53ce983614b5fb7e26df3a98c8d730
a830ee9e2a88f74f1a14c600b7af79162b03d43c
951032 F20101204_AAAPVI lane_w_Page_047.jp2
6e4dc96635d5166170532846efeb8a97
6fa8e873d2f25d8661503eceaabeb770495e27eb
42947 F20101204_AAAPUT lane_w_Page_022.pro
e191538c1b17d3f02bb97438538113dc
5b0eac515b1cf52faa0d12f0c8bc052b0326bf7c
1224 F20101204_AAAQBC lane_w_Page_099.txt
819163f9e374672633b26687c4483f8a
6462 F20101204_AAAREF lane_w_Page_111thm.jpg
6bcc66caee80512bda4bd8390632d3f3
1efefe28b945e518e291049490e0e86a3ee86abc
3463 F20101204_AAARDR lane_w_Page_102thm.jpg
8a1483c1b90d3c544b038e64e19d8c87
2279 F20101204_AAAQYL lane_w_Page_103.txt
7015f05ced4644a9fcd05d819bef1d2c
2193ccd4ef7210b70a64d3a24a9607825846e0bb
2321 F20101204_AAAQXW lane_w_Page_076.txt
b2e128cc504dbbf7ca86dfda7ea7b339
b95ac62c22d5eb89aea2775c9baff0dd8753b0a5
25524 F20101204_AAAQAP lane_w_Page_040.QC.jpg
d2a8fc0cf4ca74ff2fc800e21d77b30f
60b2358156c976b2200a57a435792949301a61c4
32277 F20101204_AAAPVJ lane_w_Page_093.pro
4c7c075d5cc2ee5f1a0423cdf11f2725
7594370b377829818c459cb69be161f7b3fc7273
599100 F20101204_AAAPUU lane_w_Page_036.jp2
14f8839f462ef1b9a858c1d4a56aa334
42b80c9c4165a365fbdc0e6260cf1f1cd5913ef0
46761 F20101204_AAAQBD lane_w_Page_139.pro
b9449b6a4db8aafb85e7d44a7d5067af
6d4143c993694f5cd71e6035465071f621c75ee0
25619 F20101204_AAAREG lane_w_Page_112.QC.jpg
a18e725c8fa0b9bf206ee8d3794fec36e0ef210a
27356 F20101204_AAARDS lane_w_Page_103.QC.jpg
dff0703db4dd698e92b23d669295ee2e
1770 F20101204_AAAQZA lane_w_Page_127.txt
cbb0a8d755338a1149a06afc3b62d5b2
7834d833b99d0743e618b70984b999d00f592219
2026 F20101204_AAAQYM lane_w_Page_104.txt
1b7ef563095be783b91a83c1a2a9f167
51b7b7681f8dff37e41e61a0b4a6d06dc634dc46
2363 F20101204_AAAQXX lane_w_Page_077.txt
633c4c7d0b411c73d9d4508595c87f4a
c5c91bcd531b43de81a4d686b72ea5b3aa830a46
62703 F20101204_AAAQAQ lane_w_Page_006.pro
52856a42692b25a8e46350953b13c298
26730 F20101204_AAAPVK lane_w_Page_133.QC.jpg
7900624083c2f1cc979f0478c99a9347d1d94c1f
F20101204_AAAPUV lane_w_Page_107thm.jpg
3987e26d843ab07596866187c4a48930caa75ce4
24093 F20101204_AAAQBE lane_w_Page_119.QC.jpg
cb4eb564fb680c0f46113713232b1725
fefc7dc727acd2475b755cddb467da4e2825560c
6732 F20101204_AAAREH lane_w_Page_112thm.jpg
ac955da7633d2378e6d07f7e5b8cae0c585b5a9c
6136 F20101204_AAARDT lane_w_Page_103thm.jpg
34a4893756c236af9b0aa039efdfc78b
31926cbe6e6170879a462a2ccb28f597ae28b0cc
1367 F20101204_AAAQZB lane_w_Page_128.txt
f25ca2e78a358677e8d2e50564e646ff830bb360
1591 F20101204_AAAQYN lane_w_Page_108.txt
e1beb0b2c36cce3d0f0f7a5096eecf55
b3e67f3c1eea33943f097ab25f7cdae2804c6bf1
1513 F20101204_AAAQXY lane_w_Page_079.txt
040487b0144ab5a911d590a058939ced
26499477b1b3d6c27a5a90c97f07b8ffab5ca112
1616 F20101204_AAAQAR lane_w_Page_090.txt
aa7f75bf4206e1a30ef6ef241dfda9f7
64f2076a1a4baa0debbe49d0c167e8a30d95b945
92166 F20101204_AAAPVL lane_w_Page_123.jpg
641c6e43601c32d9df3b3e09cc1dff82
23980 F20101204_AAAPUW lane_w_Page_104.QC.jpg
3e662a48bd29f3aaba974d3982d4a3d0
0fe42a147cc98814cb81896d3540ce4920d12ab5
4907 F20101204_AAAQBF lane_w_Page_113thm.jpg
3fca1134527951df0dce356840675e7d
f1f57a2402709fa44e6542c636f59d2e5354e0ab
5323 F20101204_AAAREI lane_w_Page_115thm.jpg
da1378aa4578ed458dc7f037004144d7
cf0d2bea76683b77f673ca2356cd1c567836c916
5696 F20101204_AAARDU lane_w_Page_104thm.jpg
2299be7224ccfac73049af940f6ba008
f6e283b9472733a68a7ba0c2724c0eb94780625e
2013 F20101204_AAAQZC lane_w_Page_129.txt
87049f88cc4350824566f27e47910188
241f809dcbd7d41ec831f07edb015eca5b929d02
1473 F20101204_AAAQYO lane_w_Page_109.txt
1bf0f560dd62b918d734f3b2720d7e123fefa1b5
F20101204_AAAQXZ lane_w_Page_080.txt
82870795d6594dfe454c1a7c04f4d16e
f5438c4ded76c7881fee16a0c6ab761fbcdfa624
2166 F20101204_AAAPWA lane_w_Page_029.txt
0a51d2ea1aa807028888af9cae3f5b6e
2a45449494910444b186909b00e2b29a7a9d9f5b
23575 F20101204_AAAQAS lane_w_Page_114.QC.jpg
a9489b510d4d5e5bbac3e86ccf7aba59
93889 F20101204_AAAPVM lane_w_Page_107.jpg
d5a849d35aa76c2027e6e5f4f42ea09e8e4fe285
1557 F20101204_AAAPUX lane_w_Page_112.txt
a6258475d3ed5acbfb184049e4b16408
47f80f33d05f08809afe7b6b52d248bf17c1abaa
39831 F20101204_AAAQBG lane_w_Page_116.pro
dc5712f937f8445c9d33cb1c99f43e1539ecbb9e
21486 F20101204_AAAREJ lane_w_Page_117.QC.jpg
9780d7cc6ddb5dc402112004ec7a4e51
9742a32ba157e0a9b8edd59b71ecd7ec322a0fb0
5115 F20101204_AAARDV lane_w_Page_105thm.jpg
a1d191f44c8b289be8e4e4d0c5d72b89264fed86
944 F20101204_AAAQZD lane_w_Page_132.txt
6596c3e577af946b4ef504522b5731c7
580aa9964726fcfc4a68a4bd6689d525c161a9c9
1603 F20101204_AAAQYP lane_w_Page_110.txt
6fa94a452697644c869d448b990354da
8799a9ab9ac2fd7a6e305b90994b2bc43ca7606a
18993 F20101204_AAAPWB lane_w_Page_018.QC.jpg
b74678f76e11e686ed253f56cec6cf7d
441d4038a932afde9e1901c83a02362983c4406d
32386 F20101204_AAAQAT lane_w_Page_143.jpg
312acd749ca4e45c930155bebdabc1d1
4f39922cb15f376264bddb0508e2dd0b6429eb4f
56252 F20101204_AAAPVN lane_w_Page_031.pro
16e54247e053d1c446231089731a742b
0f6186b4ceabd2baca0db6291025f4dabd894dea
F20101204_AAAPUY lane_w_Page_008.txt
712d53c891777283e54b909730868b7d
24803 F20101204_AAAQBH lane_w_Page_014.QC.jpg
9cba5ecdbdacd9da3fcd429f010dcae2
b1ea622e2bc79e706d49b9a3c12154f53ae5ea4a
5362 F20101204_AAAREK lane_w_Page_117thm.jpg
09665e699d51fccd00ed110aa09e93f1
57215f2e96c1e6cf0219608b16c3f5c13e523eb6
23582 F20101204_AAARDW lane_w_Page_106.QC.jpg
38694f080f8f94b2015112410cc9c574
7712cd690f95c239d8dcc9cc0e57574fcf0868f0
2305 F20101204_AAAQZE lane_w_Page_133.txt
20bdce42d44ba6f837986a86a613fdbd
8a8cf64e2a38210c6c721eccbef83517ece25350
1445 F20101204_AAAQYQ lane_w_Page_111.txt
6d4947532865b376f7e21d1cf28fed33
cd824d4404ca2fa21f7e245969d041d4885f24ff
80615 F20101204_AAAPWC lane_w_Page_049.jpg
d55884ea13028cdc1fcf30ecff9dcb1d
7e00666f60a7bf2b72409fbd2168e4ae31b8208a
F20101204_AAAQAU lane_w_Page_062.jp2
fe90e77b8459b032c0016562efc69204
5a422423f5da62e3b2a3689d3c480986e25b4311
9119 F20101204_AAAPVO lane_w_Page_100.QC.jpg
b3d60e3ec7c941806d8f0efff7bfd5806328fcd9
44201 F20101204_AAAPUZ lane_w_Page_015.pro
a5bdbf71c2f687183764e384a6959544
98af2af4f2c5df0dbf138190fa2924cdf0084643
30443 F20101204_AAAQBI lane_w_Page_088.pro
160acfca11563dbcaf1321040b27aae3
2f6fbbca45ef26ca1a13fb9eba989e4edf9d6f58
23903 F20101204_AAAREL lane_w_Page_118.QC.jpg
0373263e77a834acd2747e5bb0a3838e
fd84e1caf949fa0cfcd9488bdcca5c2547e7ffb6
5405 F20101204_AAARDX lane_w_Page_106thm.jpg
025b124935996e2f3e6846bf2941ed12
950cb95d9127c5ea525574ccdcd7b69dc5248004
2265 F20101204_AAAQZF lane_w_Page_135.txt
385dc48dd1467a399bf510749b085296
1fbb59818c66b84106ce5dfd5deb4693cfa3e237
45749 F20101204_AAAPWD lane_w_Page_027.pro
c91017100af08c39c10310dbaa071a41682fa496
2441 F20101204_AAAQAV lane_w_Page_055.txt
866cb4ffb770162d664beb4e9e02bd8f
b82264119cc254590a231d06b5bbd9c03854c504
F20101204_AAAQBJ lane_w_Page_127.tif
b321187933d13bc3ea4d8aa8af0f40f7
31fc983c0f13f8d36a0afcf83e6790808447bff7
5468 F20101204_AAARFA lane_w_Page_131thm.jpg
c6743fc25a6dfbd1f9ded866073ea97f
cb0825ceb5a579944a9b43ed682073a46d8d62fc
6158 F20101204_AAAREM lane_w_Page_119thm.jpg
428c4dfd9c9fb91770b978532a3d6794
f2535dfb1af17e469c557968cb1e8b1e12807b90
28082 F20101204_AAARDY lane_w_Page_107.QC.jpg
cd57887771a9e4071ef5f05da184cdb7
f9f3301a29c9fc603b59e6156f4492d1b2ba6799
F20101204_AAAQZG lane_w_Page_136.txt
fc39d579afca01c1bb5a5e655dbc05f3
cacdd49b208358fae98aef126070b40f02301958
1792 F20101204_AAAQYR lane_w_Page_113.txt
54838 F20101204_AAAPWE lane_w_Page_068.jpg
9f73daba4607b60fd9e32b8b10dbb097
2305692c9d1aaa16b2d91d9eb667eb04b8430636
811192 F20101204_AAAQAW lane_w_Page_028.jp2
3166c5e8f3d2cf1ac07a4e3ba4d7cbdd
00feb3d4744b0e299aa2500c2195109fa81eab31
F20101204_AAAPVP lane_w_Page_053.tif
cfb6171e4cc9619c77159df3d773ae0d
ebce2fe17b87283d39b81246028c357aa11d3850
F20101204_AAAQBK lane_w_Page_119.tif
9b15453e29f20b09af98e2ee898d8331
d92f70f35a0a0bb8a2f48b862f53ef4406a5d6ce
5838 F20101204_AAARFB lane_w_Page_133thm.jpg
df308f8a9e13e14207105876ccc26120
dba41d5956596aa6b468c47e62d26125def09e87
20162 F20101204_AAAREN lane_w_Page_120.QC.jpg
0bd2a09419c441b1a0fc46a42c0e8054
54bbf4a1929dffd11f7e7516cc0d29d76e2fd932
5428 F20101204_AAARDZ lane_w_Page_108thm.jpg
450274478036b51f662cf73aec279537
b8bb4d11893fea377e7ae0b497ae37e2ffb4af04
128 F20101204_AAAQZH lane_w_Page_137.txt
e100feea12e74de45732c91faa0e3ec8
ed82238592ae8d9d235146602db7e41d6a289091
2040 F20101204_AAAQYS lane_w_Page_116.txt
4993deca3f7844b64b1c274172a7bb78
4fa598025db92b48040d703096e986df997bef72
21633 F20101204_AAAPWF lane_w_Page_019.QC.jpg
3637f05c15dbceb1540db7c5bf26daa1
c1507f8b1f08482523a55de681d58b6fc1863eae
77415 F20101204_AAAQAX lane_w_Page_011.jpg
2a58a218eb2535e6c02b05d6cd993a3e
7a7724f17c25702f0c40272b6846520b43a4543d
67679 F20101204_AAAPVQ lane_w_Page_098.jpg
ef671ed5352506458a4a7f612d32088c
0aefe796848491bb05563826c46e1e70f83dc66a
F20101204_AAAQBL lane_w_Page_066.tif
4a6e1c1d5eb7b28b69061f0548f59ea2
8366a17ac3fde292d6c01f114f7eac08027753fd
28034 F20101204_AAARFC lane_w_Page_134.QC.jpg
ea1d8f3ccd43e9d37c5d093639e47eaf
c8aef2e7eb8e80e1b55d1c4c62a1eec6eb22c76b
18139 F20101204_AAAREO lane_w_Page_121.QC.jpg
f012516eb0f566e2938d479c48cb70a4
1622 F20101204_AAAQZI lane_w_Page_138.txt
7eeee33d4d51776e430664bfe94012c9
b9474013ddcd5bab9c427f11f32f026e74a6c482
1712 F20101204_AAAQYT lane_w_Page_117.txt
8f60f97c7270315b40fca05d8e55016a
c79c0e658966416d4cba59f2857e99899eb52ff4
98713 F20101204_AAAPWG lane_w_Page_055.jpg
c08d4e83d00518e56c8f80855b651b55
fc064c2eabcc24c991a228e8c3f12730a2c17bc4
945827 F20101204_AAAQAY lane_w_Page_051.jp2
4def5f3d4a43fc8927042503ebd5c636
14007a7f672568f96a5f7b9e6c5b0ae6c507c937
F20101204_AAAPVR lane_w_Page_020.tif
281d1ece5d2d74cd151487a188c4cf4dd20e4f97
F20101204_AAAQCA lane_w_Page_107.tif
4959ac9fea608c35897b844b6f812094
232c3653f8f55495a2d5c16d8c082f0aac07145d
68139 F20101204_AAAQBM lane_w_Page_067.jp2
7bf4fd179ef917336a600a7ed6594e9a
b253653c5276d01a0e9b101c998841fe0f5ea882
F20101204_AAARFD lane_w_Page_134thm.jpg
8cfc9f89332e5a3f56bb012b345de6b9
93410e76441ca5544dc79380930becf4192566ab
4650 F20101204_AAAREP lane_w_Page_121thm.jpg
a2357e8062d921cb892657f708f675eb
a772df3f164b4e0dc38c6aa41eb471b395c545d9
2068 F20101204_AAAQZJ lane_w_Page_141.txt
2cc3ab3d825a400063d252421545dda3
5fb2bfb3007b1009cfe9b59fff58f19c114d086f
2028 F20101204_AAAQYU lane_w_Page_118.txt
a9cf5effe5949326c7cf9f435ce31dc2
cb99f23d1226a41ffb7047dc60c30516428cf2c7
589150 F20101204_AAAPWH lane_w_Page_142.jp2
4e56fa7187534990267cedf4f0628e6b
179550b8bdfeff18823f62f44c285344db7a48fc
F20101204_AAAPVS lane_w_Page_053.jp2
55eb6e9e7742705e36e809ccffcba8ca
84d099606f1dbe296de1922675874a1fa1a788cb
20005 F20101204_AAAQCB lane_w_Page_096.QC.jpg
07970d53315f59ae6e79bd70b5c1fdb7
26853c6912309d85ecfbdd849b20f1858c83b10f
F20101204_AAAQBN lane_w_Page_119.txt
a49b60664feeb04c83a1178c951ce9e1
2b1d5a1daeb93d5ca58823d3269f7c9c7d8b7908
5981 F20101204_AAARFE lane_w_Page_135thm.jpg
d225b6c28d692090731013617076491a
44d1a7b2230545cf0d30322158cfd3ccc7343f40
26782 F20101204_AAAREQ lane_w_Page_123.QC.jpg
23c0ee002d640bcdcbdca42bf4f24ac2
e8432b48a091a5052913ab3ee56c7f1bbf64cc9d
857 F20101204_AAAQZK lane_w_Page_142.txt
b0318d626d3304825223d72404a6112f
9ef4d321cea0318d9db56f070e4a91154d65bf2f
1398 F20101204_AAAQYV lane_w_Page_120.txt
cd4dbf982f59bb81da821c769637bd33
2a59ebc7fa95ec5c5847caae665c2dfba7e846af
5821 F20101204_AAAPWI lane_w_Page_042thm.jpg
a28be2ee2c7bae13cc4204fda8f97240
868211b89a74420ec5e3926c418a7bb9c6c6e054
62049 F20101204_AAAQAZ lane_w_Page_034.jpg
0d26f30dd24d6e1c4c14cff7130ea368
c0cf8e4d6f1362c20cef1fb6d2f57a095d5e6f99
89624 F20101204_AAAPVT lane_w_Page_009.jpg
97654cc48858f1a20cbdd2f68ddd497f
75fa45fb893852ef10d98a5a1e25858fea3e7881
82012 F20101204_AAAQCC lane_w_Page_073.jpg
d1a3d32c66c77e74df4523b48f17a1c4
f756536db9995362bb03bb8d81968c9a1bfce74b
4659 F20101204_AAAQBO lane_w_Page_033thm.jpg
961fa5da26bbcab3700d0639c63f1399
846d62449e111135c7a53e5b0ef245ddaf2c0011
26759 F20101204_AAARFF lane_w_Page_136.QC.jpg
52a90c70308cc9c26d364c74817a00c8
c79154deb6f81e1c74fef7c3bcba521d86a77df5
4793 F20101204_AAARER lane_w_Page_124thm.jpg
2d0313707653e0bafb0c92d951007e3b
062275ef3417282611c31fb57c8516e3d71010f4
746 F20101204_AAAQZL lane_w_Page_143.txt
a6930d24fcb05b191bff17b73870f30ae0ac09c9
2044 F20101204_AAAQYW lane_w_Page_122.txt
0238a58b6c29de1c30f43a6e901cf437
b1e7ae0ab46c68013d1b316d1cbaf8ac0bb602a3
F20101204_AAAPWJ lane_w_Page_084.tif
790cbe891bc66cd1aebf997b04d2fff4
f1509f485ac1c03b6044340049f6863d8efb0f0e
F20101204_AAAPVU lane_w_Page_061.tif
5e4c6afac2a037382c10b5790c6c96ed
603ca022bace596f665d10f2541ea57c217df299
F20101204_AAAQCD lane_w_Page_094.txt
39191bb51c32f2cba5a62838f0e53c77
F20101204_AAAQBP lane_w_Page_123.jp2
1c90a998f4d54fc98d0d4f19e2500bc2
621c776a1aac413fa15a24b8c0a6ddf4ba41569d
6107 F20101204_AAARFG lane_w_Page_136thm.jpg
1d3a0a9465febd992d8e86f3fe5b5d71
4f5f05605ec6f7953d3fc7721b568b04147f1037
23487 F20101204_AAARES lane_w_Page_125.QC.jpg
ee3e862febb4716d8a1390ef46829f07
4ee333fa43e5a363be4854048a1f94224a84f075
1388 F20101204_AAAQZM lane_w_Page_001thm.jpg
c8953243a6614b976fd934b4077e88c1
07e9bfbe5b32e979b67dc37eaf3b61366a027a75
1657 F20101204_AAAQYX lane_w_Page_123.txt
dea3c7c7439dda9a1a401f95acd0d28c
cd15bcbd3c6ddec83514151f74806626dc0a94ac
692160 F20101204_AAAPWK lane_w_Page_088.jp2
b509c44c94725824e84334efb046a108ae42598a
61521 F20101204_AAAPVV lane_w_Page_096.jpg
f7ebdf5e3ea69a19e590f57ff9a949cfcac15882
39890 F20101204_AAAQCE lane_w_Page_125.pro
c5bb577f459133540d2f3795bba65976
008ea163d51a31bf8012b65dfb096b89aabf7e8e
1051949 F20101204_AAAQBQ lane_w_Page_107.jp2
20421 F20101204_AAARFH lane_w_Page_138.QC.jpg
684dbf7565aa10abc3feb162507a9fd7
8f722c28e801e31a43085b0c7a3de30c9a717f4b
5915 F20101204_AAARET lane_w_Page_125thm.jpg
22f252e4cf40e0d6231dcaf21753b88f
8528eca74bd47d971043ea87cb95428c79086a47
1254 F20101204_AAAQZN lane_w_Page_002.QC.jpg
717f8b93ae4727b1a93413ee80511f5d
1596 F20101204_AAAQYY lane_w_Page_124.txt
3d413892dfd42a6eeed5e43fa0d4e782833dc8ed
22845 F20101204_AAAPWL lane_w_Page_139.QC.jpg
a23bda072780f45014eb0c7ae5506c3b
60106 F20101204_AAAPVW lane_w_Page_063.pro
5da3e70883c013118fe9c7dbf66bdbdc
0b6ccd701480f4e23044536421a728e617168ebe
39914 F20101204_AAAQCF lane_w_Page_138.pro
aa0fb1eecebd42c6299cae330876a1d4
2183 F20101204_AAAQBR lane_w_Page_042.txt
acdf54bc44ff0fc6a5d969245a8c5af0
499e3ef977f71ed3dc430ec520cc0711ee339470
5346 F20101204_AAARFI lane_w_Page_139thm.jpg
6441f88b91f5c0202f492986b659438b
3b0513914bffd704b6982006638623462fb104e7
22149 F20101204_AAAREU lane_w_Page_126.QC.jpg
9b16b1e2e3cf63da28decb0aaa605521
d08fe767bfeeceab0cfb051126934e509f67930a
F20101204_AAAQZO lane_w_Page_003.QC.jpg
24aa734c66f65832f8b43749693bc19d
dbb4e8703bd94615ab9d8d6bda7a7458a314ba59
2039 F20101204_AAAQYZ lane_w_Page_126.txt
155cdbe5ab80f41bc1e00b37ba2bcb03
910491c33ea7604828688b3e5bcc3d3d3af016e6
5902 F20101204_AAAPVX lane_w_Page_001.QC.jpg
a6147341828f931ac5b091141f2b238f
65cdf4d47ec4ab12c83e1b53b157944f9e20493f
57097 F20101204_AAAQCG lane_w_Page_069.jp2
800c69dfa6a278daf2bae237a3d55cd0
16639ebc2369f8d3ea531b335cda91d8ece91c3e
4761 F20101204_AAAPXA lane_w_Page_070thm.jpg
0306135f9025d66b1bcaac65a17415f0
5102c04a4592cfe7183fdf81f861d1450cbbf2bd
6241 F20101204_AAAQBS lane_w_Page_077thm.jpg
83d6fc2620d28f7e97c9fc00e95ffd8f
1cfc970f6a654a4d1aa641f069668ae5cf1454ba
94518 F20101204_AAAPWM lane_w_Page_134.jpg
16c2b47357a8ec68164079eda8c3e9b7
24513 F20101204_AAARFJ lane_w_Page_140.QC.jpg
8f82d60bf139b9b5231a8aa064e798fb
20954 F20101204_AAAREV lane_w_Page_127.QC.jpg
f48b7688ffd6a7145f35ae5ec412ce16
9caacbef56894e76ee6ccb81ef9fa7c0cd0958e7
610 F20101204_AAAQZP lane_w_Page_003thm.jpg
fdc5476ff7f4b3c53faac4f569faef54
92229 F20101204_AAAPVY lane_w_Page_135.jpg
b14e5792e306dae5958bdc43b536568e
982204 F20101204_AAAQCH lane_w_Page_019.jp2
a56c383b6c39ec4d579ae2a689d7b2c5
d9668e90dd288b0ac689f704bb778033fcba6be9
861035 F20101204_AAAPXB lane_w_Page_113.jp2
c2a9a530571e3a58a71bc8f864d7b38a
2f08061ec1d61e110347cdd3e0eaf4d87a67837f
1253 F20101204_AAAQBT lane_w_Page_052.txt
bb7552dfc220b2c3a60d57d95860b3e8
b6921db4e67019fb1817bcb3d7978be3c6cdb0c0
18936 F20101204_AAAPWN lane_w_Page_122.QC.jpg
68bcf695d8e378df390380daa5c50088
c74dc55d32fa73136e07970a6510dc309f71846c
25058 F20101204_AAARFK lane_w_Page_141.QC.jpg
b7fba7c205b98552490624117ae4a77b
d055df8de77cc615b13abb6515244483f174a69c
5158 F20101204_AAAREW lane_w_Page_127thm.jpg
f44e8122ce144349b64bb1f5be6468b84b2b224f
22600 F20101204_AAAQZQ lane_w_Page_004.QC.jpg
a010fcff585ab2231a8761a2f070114a
489c89020d36a3d9eb3cb28c857dd3289930b5ed
2020 F20101204_AAAPVZ lane_w_Page_140.txt
b1b45a67e0dab681c68dabba8c68b3ba
9e6384ec9df7e1fcecf3a05e5597e49e34827c80
107432 F20101204_AAAQCI lane_w_Page_140.jp2
14727d82f39ffee6d27f40ce517276fe
464facc541e59716d78e655aa7ee17dfdfd08b4a
F20101204_AAAPXC lane_w_Page_114.tif
9168cd42b4618f6fd61e71d7a7077e8d
4d45b4fa65218a7b6952b6f39df4c1a2a3cff194
32843 F20101204_AAAQBU lane_w_Page_083.pro
28481ee1fd15f47b6770310cb21c3d68
32d6708668af1b13cb38c464ce219fe14bc4c555
1661 F20101204_AAAPWO lane_w_Page_130.txt
29dff43f7364f94eb2db87b622464666
3b37ff56bcffe4e8591fc00e42332bd5e51279c4
11710 F20101204_AAARFL lane_w_Page_142.QC.jpg
551ecb9f1d41a904b865d729569f2a95
bc7f46aff7f3e7aaf36e29eede7c434b386280e9
24142 F20101204_AAAREX lane_w_Page_128.QC.jpg
dec9f0864d7c14dc1da3f932c960d3f0
df162dd30e0921cd9f3a13ccd75a9635e17e4d08
5408 F20101204_AAAQZR lane_w_Page_004thm.jpg
401b9eeea2b3308abd84cd723696f662eda02392
24307 F20101204_AAAQCJ lane_w_Page_037.QC.jpg
41509ce538d3f1f2289b5c5287b2cffb
1f917f742c3e902d198c8ff54040a372f06a9874
671 F20101204_AAAPXD lane_w_Page_137thm.jpg
cc8e1dfd71b0ae71ddcfc72756dc4a6420f90161
880700 F20101204_AAAQBV lane_w_Page_096.jp2
cdc9ff577205170e4a62b37d37805b64b51e4a67
20232 F20101204_AAAPWP lane_w_Page_115.QC.jpg
d9cfc422a81d3865eddaebdda4331859
e4b90a9dd0f0c20e13bfabaae0ce9352f156c618
3013 F20101204_AAARFM lane_w_Page_142thm.jpg
8df99deef799bb9d7e98044855a0d8b9
528cc6c3fc542e7762ec58981d9c6988ec7385ae
F20101204_AAAREY lane_w_Page_129thm.jpg
2763bd12a5f334cc415c21514b5973b3
bff6c6630a53c43139eba54dd8a1ea3bae58815a
75877 F20101204_AAAQCK lane_w_Page_116.jpg
38fa206263d2885f13a855f0da290608
f8412e8dbf9c9b0a7ee87607d317d506491f1c6c
25618 F20101204_AAAPXE lane_w_Page_072.pro
33f891666c7172c7d5f218677c67c1dc
59f7334552c95c5ca8f028f58f603f7be933808b
57389 F20101204_AAAQBW lane_w_Page_094.pro
50af269e461550fc35e27da9327632de
163801 F20101204_AAARFN UFE0022090_00001.mets
d96a980ac0e9a9cfee545cdfd1c1b5e2
256bf0f1d94108637826ebf93c457c7573a92b94
5946 F20101204_AAAREZ lane_w_Page_130thm.jpg
cf4a69c273fa2a61e8a368be424b2dd4
9e5ee472c4327078a4c9b484c94ed9f96f0f7879
628 F20101204_AAAQZS lane_w_Page_005thm.jpg
a3367a4faec3ce8747c7df54e40c0a91
7462c134417172a16e73329ea0392919f5f4a811
936 F20101204_AAAQCL lane_w_Page_002.pro
905f8199f3434515d6629ed616372e1f
ac6a92146bca63f62c169cd7b8f19bc1b10ca055
39593 F20101204_AAAPXF lane_w_Page_122.pro
3f6fb1f165bc1d2b542d6ee79cf20efc
42d08cf503aeef7f55bce4e918cba4a7c78481b6
5434 F20101204_AAAQBX lane_w_Page_027thm.jpg
7c5e1b25e3a569fc74194bae00df73d9
38a9f5e7d3f903e92e02c9bceeb1b7aeb3dd08af
1051982 F20101204_AAAPWQ lane_w_Page_031.jp2
f02cbf8f2cec6f0d53bd52c9873b0e99
ef24d1289ed81a15a3d2538e45aeaa5cfff64a75

CONDUCTANCES IN THE TWO-IMPURITY ANDERSON MODEL

By

WILLIAM BRIAN LANE

A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY

UNIVERSITY OF FLORIDA

2008

2008 William Brian Lane

For Amy: my wife, friend, editor, and treasure.

ACKNOWLEDGMENTS

There are many people to thank, perhaps more than my memory and words can do

justice:

The members of my supervisory committee for their insight and understanding:

Kevin Ingersent, ('! i-I, 1ip!, r Stanton, Arthur Hebard, Robert Coldwell, and James

Keesling.

John Klauder, Kevin Ingersent, James Fry, James Dufty, Arthur Hebard, Dimitri

Maslov, Konstantin Matchev, and Pradeep Kumar for their classroom instruction at

the University of Florida.

Paul Simony, Steve Browder, William Mendoza, Bashir I ,-r, Robert Hollister, Pam

Crawford, S 'ii 'y Rai, Marcelle Bessman, and Marilyn Repsher for their classroom

instruction and professional collaboration at Jacksonville University.

Mark Meisel and Steve Hill for their dedication, wisdom, and care.

Darlene Latimer, Nathan Williams, Kristin Nichola, Donna Balkcom, and Yvonne

Dixon for their faithful service to the Department of Physics.

Many of the calculations that went into this dissertation were performed on the UF

HPC Cluster; many thanks go to C!( i 1l Taylor and the HPC Staff.

Matt Glossop, Mengxing ('!C iw- Luis Dias da Silva, Nancy Sandler, and Sergio Ulloa

for their collaboration and insight.

On a more personal note, I would like to issue these thanks, as well:

My mother, for her love, support, belief in me, and words of kindness and discipline.

My brother, for his love and friendship.

Richard Parker, Steve Gre-__- James Walden, Tobey Sorrels, Richard Horner, Rick

Borque, Dan MacDonald, Keith Jackson, Dan Brinkmann, Scott Moffatt, Gardner

Gordon, Ed Barnard, Ken Kurdziel, Dana Focks, and Ralph Coleman for their

spiritual care and wisdom.

My church families at Eastside and Creekside.

Lastly, I would like to thank my lovely wife, Amy, who has seen me through this

page

ACKNOW LEDGMENTS ................................. 4

LIST OF FIGURES .................................... 8

A BSTRA CT . . . . . . . . . . 11

CHAPTER

1 INTRODUCTION ................................ 12

1.1 The Kondo Problem ................ .......... 12
1.1.1 Resistivity Minimum and the Success of the Kondo Model ..... 12
1.1.2 The Anderson Model ................... ...... 16
1.1.3 Further Attempts at Perturbative Techniques . . 19
1.2 The Numerical Renormalization Group .... . . 22
1.2.1 The Renormalization Group Concept .. ............ 22
1.2.2 Application to the Kondo Problem ..... . ... 23
1.2.3 Iterative Procedure ............... ....... .. 26
1.2.4 Fixed Points and Results .... . . ..... .. 29
1.3 Surface and Quantum Dot Realizations of the One-Impurity Kondo Effect 31
1.3.1 Scanning Tunneling Microscopy Studies . . ..... 31
1.3.2 Quantum Dot Studies .................. ....... .. 35
1.4 Systems of Multiple Impurities . . ......... ..... 40
1.4.1 Theoretical Studies of Two-Impurity Models . . 40
1.4.2 Multiple-Impurity ST\i Studies ............... .. 44
1.4.3 Double Quantum Dot Studies ............... .. .. .. 46
1.5 Study Overview .................. .............. .. 54

2 BACKGROUND MATERIAL .............. . .. 57

2.1 Application of the NRG to the Anderson Model . . ..... 57
2.1.1 Discretization and Eigensolution .... . . 57
2.1.2 Calculation of Thermodynamic Properties . . ...... 59
2.1.3 Calculation of Spectral Functions .................. .. 61
2.1.4 Fixed Points and Results .................. ... .. 62
2.2 Extension to Two-Impurity Systems ................. . .. 66
2.2.1 Transformation to One-Dimensional Form . . ..... 67
2.2.2 Discretization and Eigensolution .... . . 70
2.2.3 Special Cases: Identical Impurities and R = 0 . 73
2.2.4 Calculation of Thermodynamic and Spectral Properties ...... ..74

3 PARALLELIZATION OF THE NRG PROCEDURE .

3.1 Parallelization of the NRG Eigensolution .. .... .......... 76
3.2 Parallelization of the Matrix Element Calculation . . 78

4 ST\i STUDIES .......... .......... ................ 82

4.1 Review of Single-Impurity Behavior ................. . .. 82
4.1.1 Single-Impurity ST:\ Setup ................... . 82
4.1.2 Results for Single-Impurity ST:\ .............. .. .. 84
4.2 Two-Impurity ST:\I Studies .................. ..... .. 87
4.2.1 Two-Impurity Set-up .................. ..... .. 87
4.2.2 Thermodynamic and Spectral Results . . ..... 91
4.2.3 Two-Impurity ST:\I Conductance .................. .. 94
4.2.4 Varying Impurity Parameters ................. . .. 97

5 ASYMMETRIC DOUBLE QUANTUM-DOT DEVICES . . 103

5.1 Double Quantum Dot Setup ................... .. .... .. 104
5.1.1 Model and Simplifications ................ ... 104
5.1.2 Special Case: 2 = 0 .................. ..... .. 106
5.2 Side-Coupled DQD .................. ............ .. 107
5.2.1 Special Case: U = 0 .................. ..... .. 107
5.2.2 Extended Case: U2 > 0 .................. .... 108
5.3 Parallel DQD .................. ............... .. 111
5.3.1 Special Case: 2 = 0 .................. .... 112
5.3.2 Extended Case: U2 > 0 Phase Diagram and Susceptibility ..... 116
5.3.3 Extended Case: U2 > 0 Spectral Function and Conductance . 124

6 CONCLUSIONS .................. ................ .. 133

6.1 Scanning Tunneling Microscopy Studies .............. .. 133
6.2 Double Quantum Dot Studies .................. .... 134
6.3 Epilogue .................. .................. .. 136

REFERENCES . .......... ................. .. .. 138

BIOGRAPHICAL SKETCH .................. ............. 143

LIST OF FIGURES

Figure page

1-1 Resistivity minimum ............... ........... .. .. 13

1-2 Logarithmic discretization of the energy space .................. .. 24

1-3 Impurity coupled to a chain of electron states .................. .. 25

1-4 Wall-clock time for the NRG procedure .................. .. 28

1-5 Evolution of energy spectra during NRG process ................. .. 30

1-6 Differential conductance for a single Co atom ......... . ..... 33

1-7 Differential conductance for a single Ce atom .................. .. 34

1-8 Differential conductance for a single Co atom with varying tip position . 36

1-9 Differential conductance for a single Ce atom with varying tip position . 37

1-10 Conductance of a quantum dot for various dot occupancies . ..... 39

1-11 Differential conductance for Co atoms .................. ...... 45

1-12 Differential conductance for a pair of Ni atoms .................. .. 46

1-13 Differential conductance for various Ce configurations . . 47

1-14 Setup for coupled double quantum dot experiment ................ ..48

1-15 Coulomb blockade valleys for a DQD .................. ..... 49

1-16 Differential conductance vs. interdot coupling .................. .. 50

1-17 Side-coupled DQD displaying two-stage Kondo screening behavior . . 51

1-18 Conductance vs. gate voltage for a side-coupled DQD .............. .52

2-1 Single-impurity TXimp(T) ............... ...... 65

2-2 Single-impurity Ad(Lo) ................ ...... 66

3-1 Wall-clock time vs. Np for iterative eigensolution with Nkeep = 3000. . 79

3-2 Wall-clock time vs. Np for calculation of operator matrix elements ...... ..81

4-1 Single-Impurity STM Setup .................. .......... .. 83

4-2 Single-Impurity STM Conductance .................. ..... .. 84

4-3 Single-Impurity STM Conductance, tc = 0 .................. .. 85

4-4

4-5

4-6

4-7

4-8

4-9

4-10

4-11

4-12

4-13

4-14

4-15

4-16

4-17

4-18

5-1

5-2

5-3

5-4

5-5

5-6

5-7

5-8

5-9

5-10

5-11

One-Impurity ST \i Conductance positive voltages . .

One-Impurity ST\ \ Conductance negative voltages . .

Fit of G(V) with td/tc = 0 . ..............

Fit of G(V) with td/tc = 0.3 . .............

Two-impurity STM setup . ...............

Strength of RKKY interaction . ............

Two-impurity magnetic susceptibility . ........

Two-impurity spectral function . ............

Two-impurity conductance, td/tc =0.1, positive voltages .

Two-impurity conductance, tdl/t =0.1, negative voltages .

Two-impurity conductance, tdl/t = 0.4, positive voltages .

Two-impurity conductance, tdl/t = 0.4, negative voltages .

Fitted two-impurity conductance, positive voltages . .

Fitted STM\i two-impurity conductance, negative voltages .

Differential conductance for various two-impurity systems .

Double quantum dot schematic . ............

Dot 1 spectral function for side-coupled DQD with U2 = 0 .

Zero-T conductance for side-coupled DQD with U2 = 0 .

Zero-T Anl(w) for side-coupled DQD with U2 > 0 . ..

Zero-T conductance for side-coupled DQD with U2 > 0 .

Zero-T conductance for side-coupled DQD with e2 > 0 .

Parallel-dot phase diagram for U2 = 0 . ........

Observation of upper QPT in TXimp vs. T for U2 = 0 . .

Linear relationship between TK and . .......

Zero-T Anl(w) vs. w > 0 for Kondo-phase parallel DQD, U2

Zero-T All(w) vs. w < 0 for Kondo-phase parallel DQD, U2

5-12 Zero-T All(w) vs. w > 0 for local-moment-phase parallel DQD, U2

. 86

. 87

. 88

. 89

. 89

. 90

. 92

. 93

. . 96

. . 97

. . 98

. . 99

. . 100

. . 101

. . 102

.. 105

. . 108

.. . 09

. . 110

. . 111

. . 112

. . 113

. . 114

. . 115

S . . 116

) . . 117

5-13 Zero-T Ain(w) vs. w < 0 for local-moment-phase parallel DQD, U2 = 0

5-14 Approximate phase diagram for parallel DQD, U2 > 0 .

Critical point vs. U2 .................... . .....

Scaled ec vs. U 2 . . . . . . . . .

Traces of TXimp vs. T, U2 = 10-3 ................ . .....

Zero-T limit of Ximp vs. 6c U2 > 0. . . .....

Kondo temperature vs. 6c+ for parallel DQD with U2 > 0 . . ..

Kondo temperature TK vs. c1 for parallel DQD without local-moment phase .

Zero-T An(w)) vs. w > 0 for Kondo-phase parallel DQD, U2 10-3 . .

Zero-T conductance G/Go vs. c1 for parallel DQD . . .

Zero-T conductance G/Go vs. 6c+ for parallel DQD . . .

5-15

5-16

5-17

5-18

5-19

5-20

5-21

5-22

5-23

5-24

5-25

5-26

conductance

conductance

conductance

G/Go vs. e, for parallel DQD with

G/Go vs. C2 for parallel DQD with

G/Go vs. C2 for parallel DQD .

120

121

122

123

124

125

126

127

128

129

no local-moment phase .

1 -UI/2 .......

Zero-T

Zero-T

Zero-T

Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy

CONDUCTANCES IN THE TWO-IMPURITY ANDERSON MODEL

By

William Brian Lane

May 2008

C('! i': James Kevin Ingersent
Major: Physics

In a number of systems of interest that involve magnetic atoms and their analogous

quantum dot manifestations, there arises a competition between Kondo screening and

various types of magnetic ordering (direct and induced). This competition can be studied

in detail using scanning tunneling microscopy to probe clusters of magnetic adatoms

on metallic surfaces and has direct implications for systems of double quantum dots. In

both of these cases, an observable quantity of interest is the electrical conductance, which

can be calculated by applying the numerical renormalization group to the two-impurity

Anderson model. Depending on their separation and the strength of their exchange

interaction, pairs of magnetic adatoms may exhibit ferromagnetic or antiferromagnetic

alignment of the impurity local moments, in some cases leading to a two-stage Kondo

screening process, effectively isolated impurity screening, or a complete suppression of the

Kondo effect. These behaviors have different signatures in the differential conductance.

A class of double quantum dot devices composed of a Kondo-like dot and a weakly

interacting dot is predicted to display a splitting of the Kondo resonance and a pair of

quantum phase transitions. These behaviors introduce unique signatures in the device

conductance when the level energy on either dot is varied by tuning the appropriate gate

voltage. This work demonstrates that double quantum dots can provide a controlled

experimental setting in which to study quantum phase transitions in a strongly correlated

system.

CHAPTER 1
INTRODUCTION

The Kondo effect is an emergent in ,ii v- body phenomenon that, even after years of

study, continues to unfold newer levels of insight and application in the realm of condensed

matter physics [54]. In its most basic form, the Kondo effect is based upon the interaction

between a localized magnetic impurity and the spins of the conduction electrons of the

host metal. As the impurity engages in a spin-exchange "d oi, with the conduction

electrons, the electrons become strongly correlated through the impurity. Below a low

temperature scale TK (called the Kondo temperature) the higher-energy conduction

electrons can be thought to "cluster" around the impurity, forming a net spin singlet with

the impurity, masking it from the electrons that lie closest to the Fermi energy, which then

Today, the Kondo effect is being investigated experimentally by scanning tunneling

microscope studies of surface impurities and in the context of tunable quantum dot

devices. In this work, I present theoretical results that demonstrate how a number of

interesting behaviors may be identified in conductance measurements in such experiments.

Scanning tunneling microscope studies of a pair of surface impurities are shown to exhibit

a variety of signatures that permit the study of the competition between the Kondo effect

and magnetic ordering that arises between the two impurities. Certain highly .i-vmmetric

quantum-dot devices are shown to exhibit a splitting of the Kondo resonance and a pair of

quantum phase transitions, with intriguing manifestations in the device conductance.

1.1 The Kondo Problem

1.1.1 Resistivity Minimum and the Success of the Kondo Model

Evidence of the Kondo effect was first observed in the 1930's in measurements of

the resistance of nominally pure samples of Au. It was expected that the resistance of

such samples would decrease monotonically with temperature as T5 (due to the scattering

of electrons from phonons; see [3]). However, in some samples, results showed that the

T (arb. units)

Figure 1-1. Schematic plot of resistance R vs. temperature T in experiments (black
circles) that first observed what is now called the Kondo effect. The minimum
in R is unexplainable by the original prediction (green curve). Kondo's result
(red curve) successfully described the minimum, but produced a diverging R as
T 0.

resistance reached a minimum value at a certain temperature Tmin (on the order of 4 to

40 K), and then increased as temperature was further decreased [1, 9]. This behavior is

summarized schematically in Figure 1-1. The unexpected minimum was attributed to the

presence of impurities, as when the impurity concentration cimp was varied, it was found

that Tmin oc 15 and the depth of the minimum R(T = 0) -R(T = Tmin) oc cimp. Study was

then undertaken to understand what aspect of these impurities was causing this minimum.

One possible explanation was that the resistance minimum might be attributed to the

potential scattering of the conduction electrons off the impurities. Such a process can be

modeled by the Hamiltonian

H H, + exp (i RJ)K, c+ c,, (1-1)
k,qj

where

CHe Z qkk, (1 2)
k
is the impurity-free conduction band Hamiltonian. The cn (c,) operators are standard

fermion operators that create (annihilate) conduction band electrons of momentum hk and

spin a. The conduction band has a bandwidth of 2D (such that Ie| < D), with a density

of states p(c) = EY (e e). (In sections to follow, it will often be assumed that p(c) = po
k
for I|e < D and zero elsewhere.) Repeated spin indices (a in this case) are implicitly

summed over, Rj is the position vector of impurity j, and Kq is the impurity scattering

potential at wavevector q.

To calculate the impurities' contribution to the resistivity Rimp, an average is

performed over the impurity positions, such that only K = Kq=o is important. A

perturbation expansion in K is then employ, 1 using the Sommerfeld expansion [24],

producing the result [54]

3wTTCi m IK2
Rimp 3Cimpe2 [ + 0 ((kBT/F)2)] (1 3)
2e2hCF

where kB is Boltzmann's constant, me is the mass of an electron, e is the magnitude of the

electronic charge, and pF is the Fermi energy of the host metal. A combined resistivity of

R(T) = aT5 + Rimp does successfully explain a non-zero R(T = 0); however, because of

the minuscule magnitude of (kBT/eC)2 (typically of order 108- in the temperature range

of interest) in Eq. (1-3), R(T) does not display the minimum seen in the experiments

[54]. It was thus concluded that the resistance minimum must be due to some other

impurity-related mechanism.

This i r,--I i. us mechanism was not unveiled until 1964, when Kondo made a

number of observations [10]: First, samples that exhibited a resistivity minimum at

low temperatures also di-plih i t a zero-field magnetic susceptibility X (defined as X

dM/dHH=,o, where M is the magnetization of a material and H is an applied magnetic

field) that obeyed a Curie-Weiss law of the form C/(T + 0) at higher temperatures,

indicating the presence of localized magnetic moments (hence the use of the term

inii ,.' tic impurity"). Second, as the size of the magnetic moments increased, so did the

depth R(T = 0) R(T = Tmin) and the location Tmin of the minimum [9], -,-. -1ii,-; that

the minimum was caused by an interaction between the spins of the impurities and the

spins of the conduction electrons. Lastly, Kondo noted that while R(T = 0) R(T = Tmin)

was proportional to cimp, the ratio (R(T = 0) R(T = Tmin))/R(T = 0) did not vary with

Cimp, indicating that this spin interaction was a single-impurity effect.
Based on the conclusion that the resistivity minimum was caused by the spin

interaction between the conduction electrons and individual localized magnetic moments,

Kondo applied the s-d exchange model developed by Zener [2], now more commonly

referred to as the Kondo Hamiltonian:

HK = H, + sYcZ % c,,. (1-4)
k,k'

Here, r-,, = x, y, z, are the Pauli spin matrices and S is the impurity spin. The

quantity J is the strength of the exchange interaction between the spins of the impurity

and the conduction electrons; if J is negative (positive), the interaction is ferromagnetic

(antiferromagnetic). (Note that this Hamiltonian may also be expanded to include

potential scattering, but this feature is not necessary to obtain Kondo's results.) With this

Hamiltonian, Kondo carried out a perturbative expansion to third order in J to obtain the

resistivity contributed by a single impurity: [10]

Rimp 37mJS(S + ) [1 2Jpo(eC) ln(kBT/D)]. (1-5)
8e2hCF

Kondo thus arrived at the total resistivity

R(T) = aT5 + cimpRo CimpRi In kBT/D, (1-6)

which, if J > 0 (the antiferromagnetic case), di- pl'l a minimum that satisfies Rmin -

R(T = 0) oc cimp at a temperature Tmin = (Ricimp/5a)1/5, matching the experimental

observations (see red curve in Figure 1-1).

Kondo's solution was a phenomenal success. He used his perturbative expression for

R(T) to fit the results from a wide variety of experiments, including some that implied

that J < 0 (ferromagnetic case) [10]. His model was also used to successfully produce

perturbative expressions for a magnetic impurity's contributions Ximp(T) and Cimp(T) to

the zero-field magnetic susceptibility and the specific heat [54]:

(gpsB)2S(S + 1)
Ximp(T) + [1 poJ+ (po )2 In(kBT/D) + c2(poJ)2] (1 7)
3kUT
Cimp(T) = kB2S(S + 1)(po4 [- 4poJln(kBT/D)], (1-8)

where c2 is a constant that depends on the conduction band density of states.

However, there were a number of remaining questions to be addressed: How do

these local moments arise within the host metal? How does the spin exchange interaction

take place? What happens at very low temperatures T < Tmin? The last question was

prompted by the irony that though the In T term in Kondo's expansion explained the

resistance minimum, this term diverges as T -- 0, and Kondo's solution was not valid for

very low temperatures. Thus arose a quest to find a non-perturbative solution, dubbed

"the Kondo problem."

1.1.2 The Anderson Model

The answer to the first two questions-how the local moments arise and the

mechanism of spin interaction-is answered rather straightforwardly by examining the

Anderson model [7], which was developed before Kondo employ ,1 the s-d exchange model.

The Anderson model considers the scattering of conduction electrons off transition metal

or rare earth impurities whose d levels lie within the host metal's conduction band.

The hybridization Vd between an impurity's d-state Od (defined in a coordinate system

that places the impurity at the origin) and the conduction electron states (Wannier

functions centered at the impurity site) is given by:

Vyd= (OdH |I ,}. (1-9)

A pair of electrons in the impurity's d-state experience a Coulomb interaction

Se2ri a
J *dkr1l)*dk(ri) 11 1)d(r2)drldr2, (1 10)

which is typically on the order of 1 to 7 eV.

The Anderson Hamiltonian is

HA = H, + CddAd + Udd dd + (V4 c d, + V*dc,), (1-11)
k

where d4 (de) is a standard fermion operator that creates (annihilates) an electron in the

impurity's d-state of energy ed. All energies are defined such that eF = 0.

The shape of the density of states p(c) away from the Fermi level and the k-dependence

of V are features not crucial to the model [7, 54]. Thus, for the remainder of this section,

the density of states will be taken as p(c) = po for |c < D and 0 elsewhere, and Vjd will be

taken as a constant Vd. To gain an understanding of the general features of the Anderson

model, it is helpful to first consider three simplified cases: (1) U, Vd = 0, (2) U = 0, Vd > 0

and (3) U > 0, Vd = 0. For each case, we examine the change Ap in the electronic density

states caused by an Anderson impurity.

(1) In the case of U = Vd = 0 (an isolated, non-interacting impurity), Ap(e) is a Dirac

delta-function at C = ed. This indicates the localized nature of the impurity electron, which

for this case is isolated from the conduction electrons.

(2) In the case of U = 0, Vd > 0 (a non-interacting impurity), the Anderson

Hamiltonian is quadratic in fermionic operators and can be diagonalized exactly. The

effect of Vd > 0 is to give the impurity state a finite lifetime, broadening the delta-function

for the Vd = 0 case to a Lorentzian given by

Ap(O) (-12)
(c Cd)2 + A2(

where

A =- TPVd|2, (113)

and Ed satisfies

D Ed
Ea-- ea-- po|V121n = 0. (1-14)

With the result (1-12), the impurity's contribution to the magnetic susceptibility is

Ximp = 2, Ap(eC), (1-15)

which does not lead to a Curie-Weiss behavior in x(T). Using Eq. (1-12) to calculate Rimp

does not result in a resistance minimum at lower temperatures, either. Thus, an impurity

model with U = 0 cannot explain the Kondo effect.

(3) Next, consider the case of U > 0, Vd = 0 (an isolated interacting impurity). Such

an impurity is described by an empty state |0) of energy 0, two singly occupied states I T)

and 1) of energy ed, and a doubly occupied state I| T) of energy 2Cd + U. Thus, if ed < 0

and ed + U > 0, the singly occupied states will be energetically favored, producing a spin-2

localized on the impurity.

For the general case of U, Va > 0, the conditions for local-moment formation are

modified because the impurity levels acquire a width of order A. For example, the singly

occupied level is broadened to the range from ed cA to ed + cA, where c ~ 4. Thus, in

general, a local moment arises if --d > A and ed + U > A. In such a case, the Anderson

model maps onto the Kondo model using the Schrieffer-Wolff transformation [12] to obtain

the spin coupling J and potential scattering K:

J= 2 + V (116)
Cd + U + Cd )Vd (1 1)

K I ( 1 ) vl (17

The change in the density of states Ap(e) then takes a form similar to Eq. (1-12), but
with A replaced by a resonance width of order

kBTK m De-1/(po). (1-18)

(This expression for the Kondo temperature TK will be explained in more detail below.)
This resonance is called the Abrikosov-Suhl resonance or the Kondo resonance. Thus, the
Kondo model is actually a limiting case of the more general Anderson model.
1.1.3 Further Attempts at Perturbative Techniques

The low-temperature divergence of the perturbative results in Eqs. (1-5) through
(1-8) remained a longstanding problem. In 1965, Abrikosov [11] tried to eliminate the
divergence by carrying out Kondo's perturbative expansion further and summing all
higher-order terms proportional to [poJln(kBT/D)]". For the impurity contribution to the
magnetic susceptibility, he arrived at [cf. Eq. (1-7)]

Ximp(T) (gpB)2S(S + 1) poJ C ( )
XimpT 2 0 2 _19)
3kBT 1 + pooJln (kBT/D) (1

This result represented progress, as the ferromagnetic case (J < 0) remains well-behaved
as T -+ 0. Unfortunately, in the antiferromagnetic case (J > 0), this expression diverges at
the Kondo temperature TK [Eq. (1-18)] instead of at T = 0.
In 1966, Yosida approached the Kondo problem with a variational method, and found
the ground state energies for a singlet (labeled s) and triplet (labeled t) configuration [13]:

E ) = -De-4/(3po0), (1 20)

EO) = -De4/(pJ). (1-21)

Thus, for the antiferromagnetic case, a ground state in which the impurity and conduction
electrons form a singlet is energetically preferred. For the case of a spin-2 impurity, similar

variational calculations predicted a finite Ximp(T = 0) = (gpB)2/4kBTK [14], holding out

the promise of a low-temperature solution.

From 1967 to 1970, Anderson and collaborators [16, 17, 19-21] made a series of

developments that eventually resulted in an approach known as "poor man's scaling."

This method predicted that as T -i 0, the "effective interaction" J (described below) goes

to oo, such that the local moment is in a fully compensated singlet state (as predicted by

Yosida above), producing non-magnetic behavior and a finite Ximp(T = 0).

The poor man's scaling approach begins with the observation that the divergence

problem arises from terms involving In (kBT/D), in which each decade of energy

approaching eF contributes equally to the properties at T = 0. Thus, no energy scales

may be ignored in perturbation theory (as might be permitted if the divergences grew as

D/kBT or (D/kBT)2). However, the contributions of the energy scales much greater than

kBT can be handled by considering how they renormalize the exchange coupling J.

Anderson considered states that were near the band edge-i.e., states of energy

D 16DI < Ie1 < D, where 6D < 0. He then constructed an effective Hamiltonian by

considering states of the form b = I',, + Qi + ,'_, where 1i represents states in which

no electrons are in the upper 16DI of the band and no holes are in the lower 16DI of the

band, ,',, represents states in which at least one hole is in the lower 16DI of the band, and

62 represents states in which at least one electron is in the upper 16DI of the band. This

effective Hamiltonian consists of blocks such as H21, which scatters an electron into the

upper 16DI of the band, and Hol, which scatters a hole into the lower 1SDI of the band.

Working to leading order in |6DI/D, Anderson arrived at an effective Hamiltonian

H1 with the same form as HK, but with the spin coupling J replaced by J' -- J + JkI,.

The renormalization 6Jk,, accounting for scattering of an electron between two states in

the low-energy sector 1 via one or more transitions involving a virtual intermediate state

in a high-energy sector 0 or 2, is given to second order in J by

pk' oJ D 1 1 (1 22)
2T E E-D+e+ E-D+c,'

where E is the energy eigenvalue of the effective Hamiltonian. This effective Hamiltonian

is valid over a reduced bandwidth D D |6D|.

For low-energy excitations, one can set E, ke, and e, to zero in Eq. (1-22) to arrive

at the second-order scaling equation

dJ
S- poJ2. (123)

Reducing D to D by absorbing the energy levels near the band edge into the renormalization

of J to J will cause J to increase. Thus, the higher-energy excitations still contribute to

the solution of the Kondo model, but they do so by renormalizing J.

In this rescaling of D and J, if 0 < poJ < 1, the Kondo temperature TK(J, D)

functions as a scaling invariant:

De-1/(poJ) De-1/(poJ) kBTK. (1 24)

This property is extremely useful; for example, the impurity contribution to the magnetic

susceptibility can be written

Ximp(T) F (1-25)

where F(X) is a universal function.

Thermodynamic results based on Eq. (1-23) still exhibit a divergence at T = TK.

Carrying the rescaling out to third order in J produces

d(poJ) _(poJ)2 + (o)3 + O((poJ)4), (126)
d In D 2

resulting in a more accurate expression for the Kondo temperature:

kBTK M D poJe-1 (p"J)+O(pOJ). (1 27)

It is justified to integrate out high-energy band degrees of freedom only until D has

been reduced to D = akBT, where a is of order 10. (At temperature T, real thermal

excitations span all energies Ile| akBT.) In effect, J becomes a function of T, given by

0- + t In In a In (T (-28)
poJ 2 PoJ K

Applying Eq. (1-28) to the magnetic susceptibility gives

Ximp(T) B [- poJ(T) + 0[poJ(T)]] (1 29)

1 In[(n (TTK))] + 1. 30)
4kBT ln(T/TK) 2(ln[(T/TK)])2 [In (T/TK)12j (

Equation (1-30) still contains a logarithmic divergence at T = TK. Thus, even after

Anderson's scaling methods, the low-temperature behavior remained a mystery.

The poor man's scaling method provided vast progress toward solving the Kondo

and Anderson problems, but still broke down due to its perturbative approach. Anderson

had shown that it was reasonable for J -- o as T -i 0 and for Ximp(T -+ 0) to
be finite, though he was unable to arrive at a well-defined low-temperature expression

for Ximp. However, the concept of scaling would prove to be essential to the successful

non-perturbative solution found in the numerical renormalization group.
1.2 The Numerical Renormalization Group

1.2.1 The Renormalization Group Concept

The goal of the renormalization group (RG) is the study of a Hamiltonian H(KI)

characterized by a set of couplings (K1, K2,...). An RG transformation R maps

H(K) into another Hamiltonian H(K') of the same form, but with a different set of

couplings:

R[H(K)] H(K'), (1-31)

or, more compactly,

R(K) K'. (1-32)

This new Hamiltonian is then valid over a reduced energy scale. The transformation R is

usually characterized by a, the ratio of the new energy scale to the old, such that

R3 (R',(K)) R { (K). (1-33)

Making a sequence of transformations can be thought to generate a trajectory or flow line

in K-space (somewhat analogous to slope fields in coupled differential equations). These

trajectories begin and end at fixed points K* that satisfy

R.(K*) K*. (1-34)

A fixed point is classified as stable or unstable according to whether the local direction

of RG flow is toward or away from K* (again, analogous to fixed points in coupled

differential equations).

As a simple example of the RG concept, in the poor man's scaling approach to the

Kondo problem of the previous section, the elimination of higher-energy levels transforms

J into J. The new Hamiltonian has the same form, but is defined for a reduced energy

scale D and has a renormalized J. Anderson demonstrated that the ferromagnetic Kondo

model had a stable fixed point at J = 0, and made the reasonable extrapolation that

J -- o was the stable fixed point of the antiferromagnetic Kondo model, but was unable

to prove it.

1.2.2 Application to the Kondo Problem

In 1975, Wilson [23] combined the renormalization group with the poor man's scaling

method to examine the S = Kondo model. His two most important results were proof

that J -- o is the only stable fixed point for the antiferromagnetic Kondo model and

an expression for the effective Hamiltonian Hj=, near this fixed point which he used to

calculate thermodynamic behavior at T < TK.

To arrive at these results, Wilson considered the problem of divergence that had

plagued perturbative studies of the Kondo model. He reasoned that the ln(kBT/D)

-1 A- -A2 0 A-2 A-1 1

Figure 1-2. Logarithmic discretization of the energy space -1 < e/D < 1, where the Fermi
energy CF = 0.

divergences occurred in expressions for thermodynamic quantities because all energy scales

were contributing to the calculation-e.g., scales of order D, D/10, D/100 .... To account

for this energy cascade, Wilson divided the conduction band -D < < D (where CF is

taken to be 0) into intervals (see Figure 1-2) given by

DA-(m+l) < |Ic < DA-' (1-35)

for m =0, 1, 2,.... Here, A > 1 is a chosen quantity termed the discretization parameter.

With this energy discretization in place, the conduction electron basis is then changed

from the set of operators {c i|e < D} to a set {at b~t}, ,}, where aq,,(b,a)

creates an electron of positive (negative) energy in a state ,~,q,(q(/D) given by

A /2
m,q(E) (- -1)1/2 exp(iwUq), A-(z-1) < Ie < A-, (1-36)

where w, = 27rA'/(1 A-1) and the integer q is a Fourier harmonic index. One can

picture the ~y,,q =o() states as concentric shells centered around the impurity, each of

radius r,- kFlAA/2, where kF is the Fermi wavevector.

Upon rewriting HK with the new an,q, and bq,, operators, it can be seen that the

impurity couples directly only to states with q = 0. In Wilson's numerical renormalization

group (NRG) approach, states with q / 0 are ignored, incurring an error proportional

to (1 A-1)/27, which is negligible as long as A is not much greater than 1. (Typically,

1 < A < 3 produces acceptable results.)

Impurity Conduction Electrons
'ko". X 0 kl 2 3 ) 4 k 5

Figure 1-3. Impurity coupled to a chain of electron states. In Wilson's discretization, the
coupling AX ~ A-"/2

After dropping the states with q / 0, a Lanczos transformation is applied to the set

of operators {at ,q o,, b -,,0,r}, resulting in a set {f ,.}, defined as linear combinations of

a q0,, and bt ,.

With the new ft,, operators, the Kondo Hamiltonian becomes

HK = C [ nf,,f, + fi n+, fr +lfn,] + 2JS s'o, (1-37)
n=o

This Hamiltonian can be pictured as forming a linear chain (see Figure 1-3), in which the

impurity is connected only to site 0 (whose spin operator is written as soT,T = fot,-fo,')

and each site is coupled only to its nearest neighbors.

Working with a density of states p(c) = 1/(2D), -D < c < D, Wilson arrived at

expressions for the couplings

= 0 (1-38)

A D( + A-/2, (1 39)
2

where is a set of constants of order unity, given by

1 A-"-1
[(1 A-2n-) (1 A-2n-1)]1/2'

such that t 1 as n -- oo.

Here is where Wilson's procedure differs from Anderson's perturbative scaling

technique: The NRG scheme does not attempt to follow the coupling J as D is reduced,

nor does it rely on poJ to be small. Instead (as we shall see in the next section), it tracks

the evolution of the low-lying in i ,,v-1 ody eigenstates of HK on a characteristic energy

scale D DA-"/2 as n -- oo.

1.2.3 Iterative Procedure

To apply the techniques of the RG, Wilson considered the Hamiltonian HN for a

partial chain consisting of the impurity plus the N + 1 innermost electron sites,

HN D(1+A 1) '/2 t + + .,,) + 2J. s'o,,,,/ (1-41)
n=0

such that the sequence of Hamiltonians H, H1, H2,... obeys the recursion relation

HN+1 = HN + 2(1 + A- )A -N N(f NfN+1, + h.c.). (1-42)

This relationship enables an iterative solution of the chain. First, the simple Hamiltonian

Ho = 2JS. s'o,,,, is diagonalized, obtaining eigenstates IN, Q,SS, Sm) labeled by

the quantum numbers Q (charge measured from half-filling), S (total spin), Sz (spin

z-component), and m (which distinguishes states with all other quantum numbers the

same). These eigenstates are then multiplied by the states of site 1 (10), 1Z),| 1), and I T1))

to construct a basis of states in which HI is written. The new Hamiltonian H1 is then

diagonalized, obtaining eigenstates that are multiplied by the states of site 2 to write H2,

etc. This iterative procedure provides the basis of the RG transformation.

One of the primary goals of this procedure is to compare the set of energy eigenvalues

(also referred to as the energy spectrum) of HN to that of HN+I to find a fixed point

of the RG transformation. As the characteristic energy scale of each HN decreases as

DA-N2, in order to make the comparison between successive spectra, a dimensionless

Hamiltonian HN is used:
2A(N-1)/2
HN -1)HN (1-43)
D(1 + A-1)

Transforming the recursion relation Eq. (1-42) in this manner leads to the RG transformation

HN+1 = R(HN) (1-44)

A1/2HN + N(f,,fN,+ h.c.). (1-45)

Diagonalizing HN yields the set of energy eigenvalues EN(ri) and their eigenvectors

IN, M) (where IN, ,) is a shorthand for IN, Q, S, Sz, m)). Computationally, the RG
transformation can be carried out by calculating

HN+1 = EN(f')IN,i)(N,fn (1-46)

+&Y [KN,MJf7 \N, M') N, M}) (N, Mr' fN1,, + h.+c.

Thus, the RG transformation of Eq. (1-44) can be thought of as acting on the set of

eigenvalues of HN and matrix elements of fN,a:

R[(EN(J7), (N, m| fN,,~N, i'))] = (EN+(1n), (N + 1, I fN+1,,~N + 1, i')). (1-47)

The iterative procedure is carried out on a computer-hence the name "Numerical

Renormalization Group." Each Hamiltonian HN is block diagonal in the conserved

quantum numbers of the model. For example, it can be shown that [HK, S] = 0, and so

there will be no non-zero matrix elements between states of different total spin or different

spin z-component. This block-diagonal nature permits a faster eigensolution. As described

in C'! lpter 2, it is straightforward to obtain thermodynamic and spectral quantities as

functions of temperature by keeping track of the energy eigenvalues and operator matrix

elements at each iteration.

The NRG method's most significant detriment is that the dimension of the Hamiltonian

HN is 22N+3. Since the time to diagonalize a matrix grows as the cube of its dimension,

the computational effort to obtain a complete solution of HN becomes prohibitive beyond

N 5. Thus, in practice, the summations over mi and mi' are restricted to the Nkeep states

10000- o runtime / runtime(Nkp = 10)
SEkeep keep(Nkeep 10)

1000

100 -

10- 0 -

0 0000

10 100 1000
N
keep

Figure 1-4. Wall-clock time for the NRG procedure and the highest energy level retained
Ekeep VS. Nkeep. The wall-clock time and energy levels are scaled by their
values for Nkeep = 10. Note how the runtime (the cost of the calculation) grows
much more quickly than the energy level retention (the benefit).

of lowest energy, or to all states with energy lower than some level Ekeep, thus reducing the

subsequent Hamiltonian to a manageable size.

To illustrate the state truncation approximation, Figure 1-4 shows the runtime

required for the NRG procedure to be carried out on a 2.2-GHz AMD Opteron processor

for various values of Nkeep and the resulting truncated value of Ekeep. The runtime and

Ekeep results are scaled by their values for Nkeep = 10, which are 0.3 seconds and 1.68,

respectively. Proportionately, the runtime grows faster than Ekeep. (These results are

actually for the NRG study of the Anderson model, as described in Section 2.1.1, but they

illustrate the point well.)

This state truncation is an acceptable approximation, since the low-temperature

behavior is largely determined by the lower-lying energy eigenstates (EN(fi) < 10), and a

compromise can usually be reached between computational costs and maintaining an Nkeep

sufficiently high to ensure accuracy. Generally, more complex problems (e.g., multiple

impurities models or models with fewer conserved quantum numbers) have a richer energy

spectrum and thus require a larger Nkeep to maintain a sufficiently high Ekeep to obtain

acceptable results.

1.2.4 Fixed Points and Results

A fixed point of the RG transformation has been found when the renormalized energy

eigenvalues EN(f) remain unchanged upon increasing N (as seen in Figure 1-5). It

turns out that the transformation R as given in Eq. (1-47) has no fixed points, since the

energy spectra for odd and even N are manifestly different in form [30]. Thus, it is the

transformation R2 that yields fixed points, and comparisons are made between EN(fn) and

EN+2 (f).
Wilson demonstrated that the ferromagnetic Kondo model has a stable fixed point

H*=o (confirming the results from poor man's scaling), and that the antiferromagnetic

Kondo model has an unstable fixed point H*=o and a stable fixed point H*j,. (Technically,

each of these "fixed points" is a pair of fixed points-one for even N and one for odd

N-but the thermodynamic properties are the same.)

Wilson constructed Hj*j to successfully arrive at a finite low-temperature expression

for Ximp in the antiferromagnetic case [23]:

(gp9)2(0.4128 0.002)
Ximp(T) =4T T < TK. (148)
4kBTK

This constant result showed definitively that the impurity local moment disappears as

T -+ 0 due to the impurity spin being fully compensated. The number 0.4128, dubbed

the Wilson number w, has since been obtained exactly from methods based on the Bethe

Ansatz [28, 29], in which the Kondo model is reduced to a subsidiary spin problem as

solved by Yang [15], producing the exact result w = ec+1/4/ 2 (where C 0.577216 is

Euler's constant), which agrees with the NRG result to within the printed accuracy.

N (even)

Figure 1-5.

Energy levels E(N) vs. iteration number N (even) for the NRG procedure
applied to the Anderson model. The energy eigenstates are labeled by spin
and charge. (Due to spin and particle hole symmetries, states of opposite spin
or opposite charge are degenerate.) The energy levels are measured from the
ground state energy, such that the lowest energy level at each iteration is
al-iv--, 0. Plateaus in the spectrum indicate the presence of a fixed point of
the renormalization group transformation. At iteration 6, an unstable fixed
point is approached, but the levels flow away from it by iteration 96. A stable
fixed point is reached at iteration 110, signifying the onset of the Kondo effect.

Wilson also reproduced the high-temperature (T > TK) results from poor man's

scaling [Eq. (1-30)], and obtained an expression for the crossover regime [23]

Ximp(T) -

0.68(gpB)2
4B(+ ) 0.5TK < T < 16TK,
4kB(T + STK )

(1-49)

which is identified as a Curie-Weiss law for a reduced (or partially screened) local moment.

The Numerical Renormailzation Group is similarly applied to the Anderson model; I

will defer that description until C'! lpter 2.

1.3 Surface and Quantum Dot Realizations of the One-Impurity Kondo Effect

Wilson's RG approach to the Kondo and Anderson problems had answered many

unresolved questions, confirmed the screening of the localized magnetic moment at low

temperatures, and explained the observed effects of magnetic impurities in bulk materials.

The question then arose of whether it was possible to observe a single impurity (rather

than a dilute set of impurities) exhibiting the Kondo effect. It was not until more recently

that such study became possible with development of scanning tunneling microscopy and

quantum dots.

1.3.1 Scanning Tunneling Microscopy Studies

Scanning tunneling microscopy (ST\ i) [46] permits the study of metallic surfaces at

the atomic level. As a small metal tip is swept over the surface to be studied, the current

in the tip caused by electrons tunneling to the surface is monitored, allowing analysis of

the surface's structure and the local electron density of states. One of the most interesting

applications of ST\ i has been the direct observation of single magnetic impurities.

By using ST-\i to study individual isolated magnetic atoms adsorbed onto a

non-magnetic surface, several experiments [57, 59, 62, 89, 90] have observed signatures

of the Kondo effect in the differential conductance G(V) = dI/dV, where V is the bias

voltage of the sample relative to the tip. The differential conductance is proportional

to the rate at which electrons tunnel from the tip into a state of energy pF + eV on the

surface [57]. This tunneling rate is related to the impurity spectral function Ad(eV/h, T),

which, is defined such that Ad(wu)du is the probability that an electron in the impurity

level has an energy between hu and h(u + dw). (For convenience, in the remainder of

this work, h will be taken to be 1.) For T < TK and u < kBTK, Ad(T, w) exhibits the

aforementioned Kondo resonance: a Lorentzian [similar to the shape of Eq. (1-12)] of

width TK. Thus, when an STM tip probes a magnetic impurity, G(V) is expected to show

a physical manifestation of the Kondo resonance at temperatures T < TK.

Figure 1-6 shows the results of an STu i study of an isolated Co atom adsorbed onto

a gold (111) surface [57]. When the ST \i tip was positioned far away (12 A) from the

Co atom, a featureless G(V) was observed; when the ST \i tip was positioned directly

above the Co atom, G(V) exhibited a resonance around zero bias (corresponding to

energy excitations in the sample around CF). The structure of the resonance is unlike the

Lorentzian shape of the Kondo resonance in Ad(UL). Figure 1-7 shows similar results for

an STA\i study of an isolated Ce atom on Ag(lll) [59]. The feature at zero bias is more

symmetric than for Co, but still contains a dip (sometimes called an il i. i ")

rather than a peak.

The form of G(V) for both of these cases is the lineshape predicted by Fano [8] who

worked with a model similar to the Anderson model in the field of atomic ionization.

Fano explained the lineshape by considering interference between tunneling into a discrete

state and a continuum of states. Similarly, the lineshape of G(V) in the ST\ i studies has

been attributed to quantum mechanical interference between tunneling of electrons from

the tip directly to the surface (a continuum of states for all positions F on the surface)

and tunneling from the tip to the impurity (the localized state associated with dt) and

thereafter to the surface through the Anderson model's hybridization Vd.

Applying Fano's study to the U = 0 Anderson model, the rate of transitions from the
tip into a final state of energy c is given by [57]

[q + f(c){2
R(c) = Ro(C) [1 f(2 (1 50)
1 + f (C)2

where Ro is the transition rate for an impurity-free system, and

f(C) (1-51)
A/2

As G(V) is proportional to R(c = eV), it di-~p' this Fano lineshape, with a resonance

of width TK. The lineshape is characterized by the Fano parameter q < 0, which depends

on the surface's electronic structure and the relative probabilities of the two tunneling

1,0

0.8 6

0,6

0,4

0,2

0.0

-100

-50

50

100

Sample voltage (mV)

Figure 1-6. STM \ differential conductance dl/dV for a tip located over bare Au(lll) and
over a single Co atom. The Co trace is fitted with a Fano lineshape of width
2kBTK [57]. From Madhavan et al., Science 24 April 1998 280: 567-569.
Reprinted with permission from AAAS.

paths [73], such that q

0 indicates no tunneling to the impurity, q

no tunneling directly to the surface, and q

o0 indicates

1 indicates equal tunneling through both

channels.

To compare with the experimental results, Eq. (1-50) can be applied to the U / 0
Anderson model with the modification

C Ed ReZ(e)
f ( Im ()

-- Experiment
Off Cobalt Atom eor
------- Thleory

On Cobalt Atom

-\

(1-52)

32 -
ILI

30- .ag
Ce/Ag(111)

-50 -25 0 25 50

Sample Voltage (mV)

Figure 1-7. ST\ I differential conductance dI/dV for a tip located over a single Ce atom on
an Ag(lll) surface. The data are fitted with a Fano lineshape of width 2kBTK
[59]. Reprinted with permission from J. Li, W.-D. Schneider, R. Berndt, and
B. Delley, Phys. Rev. Lett. 80, 2893 (1998). Copyright 1998 by the American
Physical Society.

where E(e) is the impurity electron's self-energy. For energies near the Kondo resonance
and for T < TK, f() is

f(c) =- (1-53)
kBTK'
where a is a constant.
Figures 1-6 and 1-7 contain fits for the Co and Ce data using Eqs. (1-50) and (1-53).
For the Co data, the fit gives q = -0.7 [57], indicating comparable tunneling through
both channels; for the Ce data, the fit gives q m 0 [59], indicating almost no tunneling
to the impurity. What is fascinating about the q = 0 result is that, even though no
electrons tunnel directly to the impurity, the impurity's presence is still observed in the
differential conductance. This behavior is evidenced by the fact that, even when q = 0,

the tunneling transmission rate [Eq. (1-50)] still acquires a non-uniform shape due to the

presence of the impurity electron's self-energy. Physically, this property can be understood

by considering second-order processes, in which electrons that have tunneled from the tip

to the surface then tunnel from the surface to the impurity and back again to the surface

before returning to the ST\ i apparatus. The extra tunneling causes the electrons to pick

up a phase shift relative to the electrons that do not tunnel, and destructive interference

takes place.

The Fano resonance has been observed in the differential conductance in many

other STM experiments [62, 89, 90], including magnetic impurities adsorbed onto carbon

nanotubes, which exhibit a one-dimensional density of states, further demonstrating the

robustness of the Fano-resonance behavior. We will observe that the NRG confirms the

Fano-resonance behavior in ('C Ilpter 4.

ST\ i studies also permit an examination of the spatial extent 1K of the screening

cloud that causes the Kondo effect. Such a quantity is not directly calculated in the

standard NRG procedure, but can be estimated experimentally by moving the ST\i

tip laterally away from the impurity. Figures 1-8 and 1-9 show G(V) for various lateral

distances between the tip and impurity. The magnitude and .,i- i I ry of the Fano-shape

feature decrease as the lateral distance increases, indicating a reduction of q caused

by a drop in the rate of electrons tunneling to the impurity. Sufficiently far from the

impurity, the featureless conductance is recovered. Section 6.1 describes a novel approach

to calculate 1K with an extension of the NRG [96].

1.3.2 Quantum Dot Studies

A quantum dot (QD) is a nanostructure that spatially confines a specified number

of electrons. The electronic properties of a QD are similar to those of atoms, featuring

discrete energy levels, internal Coulomb repulsion, and a coupling to the sea of electrons

in the external host. When attention is focused on the highest unfilled (or partially filled)

level in the QD, these parameters are akin to ed, U, and Vd from the Anderson model,

2.5

-'
^c
3

I-

-o

2.0

1.5

1.0

0.5

0.0

I I I I I

12A -

9A.

5A

sA

2A

oA

-3A-

-4A

-7A

-12A-

I I I I I I I i I

-100

-50

50

100

Sample voltage (mV)

Figure 1-8. STA\ differential conductance dI/dV for tip located at various lateral distances
from a single Co atom on Au(lll) [57]. The resonance's .,-vmmetry and
amplitude decrease as the distance increases. From Madhavan et al., Science
24 April 1998 280: 567-569. Reprinted with permission from AAAS.

r,

-100 0 100
Sample Voltage (mV)

(a) oA

(b) 1oA

(c)2oA

(d) 3oA

(e) 40A

(f) 50A

(g) eoA

(h) soA

(i) looA

() 120A

(k) 140A

200

Figure 1-9.

ST\ i differential conductance dI/dV for tip located at various lateral distances
from a single Ce atom on Ag(lll) [59]. The feature at V = -70mV is
attributed to the Ag surface state, and is modified by the proximity of the
impurity. Reprinted with permission from J. Li, W.-D. Schneider, R. Berndt,
and B. Delley, Phys. Rev. Lett. 80, 2893 (1998). Copyright 1998 by the
American Physical Society.

-200

respectively. The values of Cd, U, and Vd can be controlled by tunable gate voltages,

permitting manipulation of this artificial atom. Thus, QD devices have become a versatile

tool for studying magnetic impurity systems, permitting the experimental observation of

previously inaccessible regimes and behaviors.

Tuning ed permits control over the number of electrons in the dot. When a QD is

tuned to have an odd number of electrons (i.e., it meets the conditions for local-moment

formation in the Anderson model as described in Section 1.1.2), it must have a non-zero

total spin, and therefore can display the Kondo effect. Because of the two-dimensional

geometry of most QDs (such as semiconductor QDs formed at the interface between GaAs

and AlGaAs), the Kondo effect causes an increase in the conductance G(V) through the

dot [55, 64] as T decreases.

These predictions are confirmed in the results of a QD experiment [64] shown in

Figure 1-10. Part (a) of the figure shows G(V) vs. the gate voltage V corresponding to

dd, for temperatures ranging from 25 mK (light blue curve) to 1 K (orange curve). As

the gate voltage is increased over the range shown in the figure, the occupancy of the dot

increases from N to N + 4 (where N is even). These regions form a Coulomb blockade

pattern, indicated by peaks and valleys in the conductance. For even occupancy, the

conductance decreases with temperature, indicating that no Kondo effect takes place; for

odd occupancy, the conductance increases with temperature, indicating the occurrence of

the Kondo effect and thus the presence of a local moment on the dot. Similar results were

found in [55], where the enhanced conductance was shown to disappear for temperatures

above the Kondo temperature.

Figure 1-10(b) shows G(V) vs. T for three values of V that produce odd occupancy

N + 1. As V (i.e., Cd) changes, so does TK. If the data are instead plotted as G(V) vs.

T/TK, as in part (c), the results lie on top of each other, indicating that G(V) is given by

a universal function F(T/TK) as seen in Eq. (1-25).

a 2

N+2

0
gate voltage

b c

o 4
2 2 s->

o ,O \\

1 I I
0.01 0.1 1 0.1 1
T (K) T/TK

Figure 1-10. (a) Conductance G(V) vs. gate voltage V for temperatures ranging from 25
mK (light blue curve) to 1 K (orange curve). Coulomb blockade valleys are
labeled by electron occupancy, where N is even. (b) For odd occupancy
N + 1, G(V) vs. T for three values of V (corresponding to arrows in a). (c)
G(V) vs. T/TK for the same three values of V, indicating the universal
scaling of the conductance in the Kondo regime [64]. Reprinted with
permission. Copyright 2001 by Physics World (2001January pp33-38).

It is fascinating to note that, when the Kondo effect takes place, this quantum dot

becomes nearly I1i i-pI ient," as G(V) nearly reaches the quantum mechanical maximum

value of 2e2/h.

Quantum dots are thus an exciting arena of study, offering a vast range of possibilities

for the creation of tunable atom-like structures. For example, another possible application

of QDs to the Kondo effect is the study of a dot with spin S > 1/2 [63].

1.4 Systems of Multiple Impurities

Since Kondo's breakthrough in 1964 and Wilson's application of the RG in 1975,

the physics of a single isolated magnetic impurity in a simple metal host has become

well understood. However, matters become more subtle when one considers a system of

multiple magnetic impurities that are permitted to interact with each other. The direct

or indirect exchange of electrons among multiple impurities can lead to the ferromagnetic

(parallel) or antiferromagnetic antiparallell) alignment of the impurities' spins, modifying

the Kondo screening behavior. The competition between magnetic ordering and Kondo

screening has important implications for heavy fermion systems, for small magnetic

devices, and for future quantum computers that will utilize magnetic moments as quantum

bits.

As a first approach to understanding systems of multiple impurities, one can gain a

great deal of insight by considering a pair of magnetic impurities. There has been much

theoretical and numerical work performed to studying the two-impurity Kondo model and

the two-impurity Anderson model. Experimentally, the study of multiple-impurity systems

has been pursued in the venues of ST_ i and double quantum dots (DQDs).

1.4.1 Theoretical Studies of Two-Impurity Models

As modeling a system of an arbitrary number of impurities is a very demanding task,

much attention has been devoted to two-impurity models, which, as we shall see, still

capture much of the essential physics of a many-impurity system. The Kondo Hamiltonian

[Eq. (1-4)] is extended into its two-impurity version:

HK2 Hc + 2 -S e 6i( Ta 6k',. (154)
k,k'

Here, the new index i 1= 2 labels the impurities and r is the position vector of impurity

i. Potential scattering terms have been omitted from Eq. (1-54) for simplicity.

As with the single-impurity Kondo model, the two-impurity Kondo model can be

thought of as a special case of the two-impurity Anderson model, when each impurity

favors the formation of a local moment. The two-impurity Anderson model will be

described in greater detail in Chapter 2.

Various theoretical and numerical methods have been emploi-, 1 to study the

two-impurity Kondo and Anderson models, including perturbative scaling [32], the NRG

[34-36, 41, 42, 49, 74, 80], quantum Monte Carlo methods [33, 37], variational methods

[40], and conformal field theory [43, 45]. These studies have highlighted the importance of

a number features that are not present in the single-impurity Kondo model: the presence

of two effective conduction-band channels, a greater complexity of the electron density of

states, and a competition between the Kondo effect and magnetic ordering effects.

The first new feature, the presence of two effective conduction-band channels arising

from the coupling to the impurities at locations ri and r2, will be explored in more detail

in Section 2.2.1. Essentially, it is computationally more convenient to define conduction

electron states that are symmetric and antisymmetric about the midpoint between the two

impurities, called the 1i,' 1 channel and "odd" channel, respectively. While this division

of the conduction band is a theoretical construct, it does have physical implications, such

as the two-stage Kondo screening process described below and found in [27].

The second new feature, a greater complexity of the electron density of states, will

also be illustrated in more detail in Section 2.2.1. The variable impurity separation

R = tr r2 means that one cannot simply ignore the k-dependence arising from the

form of Ck. As described below, confusion and disagreement have resulted in the past from

ignoring this complexity. I therefore lp li careful attention to this issue in the present work.

The final new feature in the two-impurity Kondo model is the competition between

the Kondo effect (which seeks to lock each impurity into a net spin singlet with the

conduction electrons) and magnetic interactions between the local moments of the

impurities (which seek to lock the impurities into a singlet or triplet with each other). The

two impurities may directly interact with each other, reflected in a term such as

H = -IS1 S (1-55)

being added to the Hamiltonian in Eq. (1 54). Here, I is the strength of the magnetic

interaction, with I > 0 causing an ferromagnetic (parallel) alignment of the impurity

moments and I < 0 causing an antiferromagnetic antiparallell) alignment of the impurity

moments. In order to examine how this interaction competes with the Kondo effect, I||

must be compared with kBTK' m where T1i"p is the single-impurity Kondo temperature.

Even if such a direct interaction is not present, there is still an indirect Ruderman-Kittel-

Kasuya-Yosida (RKKY) interaction to consider [4, 5]. The RKKY interaction is mediated

by the conduction electrons, and arises from the Friedel oscillations in the conduction

electron density around each impurity [24]. As we shall see in greater detail in C'!i pter 4,

the RKKY interaction IRKKY varies in strength and magnetic nature (ferromagnetic for

IRKKY > 0 or antiferromagnetic for IRKKY < 0) depending on the separation between the

two impurities. The interaction strength is, again, to be compared with the single-impurity

Kondo temperature.

The majority of the previous theoretical work has focused on identical impurities

(J = J2) without a direct magnetic interaction. The competition between the RKKY

interaction and the Kondo effect has been predicted to cause a number of behaviors that

were not present in the single-impurity Kondo model.

For example, when IRKKY > TI-ilp, at higher temperatures the ferromagnetic

RKKY interaction causes the two impurity moments to form a combined spin-1 system

[32, 35, 37, 49]. This spin-1 is then screened in a two-stage Kondo effect. In this two-stage

Kondo effect, the spin-1 reduces to a spin-, which then reduces to a spin-0. (For examples

of this behavior, see Figures 1-17 and 4-10.) These two stages can be explained by the

presence of the even and odd conduction-band channels (see above discussion and [27]),

each of which exhibits its own Kondo temperature. ('!i Ilter 4 will explore how this

two-stage screening process may be observed in an ST \i experiment.

Another class of behavior arises when IIRKKY < TI -imp For this regime, the RKKY

interaction is not strong enough to combine the impurity moments into a spin-1 before the

Kondo effect screens each moment individually. Thus, the two impurities are effectively

isolated from each other [32, 49]. This behavior will also be observed in the context of

STA\I experiments in ('!i lpter 4.

A very different behavior arises when -IRKKY Tj- imp For this regime, the RKKY

interaction causes the impurity spins to antiferromagnetically align into a spin singlet

state, such that no Kondo effect can take place [32, 35, 37, 49]. This behavior will also be

observed in the context of ST,\i experiments in ('!i lpter 4.

One last feature of the model that has been explored in great detail is the transition

between the independent-impurity regime (|IRKKYI < Tj- `imp) and the impul i--int

regime (-IRKKY T--imp). The original poor man's scaling analysis [32] predicted a

smooth transition between the two regimes. This smooth transition was later supported

by quantum Monte Carlo methods [33, 37]. However, an NRG study [35] predicted the

presence of an unstable critical point separating the two regimes. At first, the discussion

of the discrepancy centered around the differences in the methods used (for example,

the NRG work had not used --dependent couplings between the impurities and the

conduction-band channels), but finally a connection between the two results was verified

by variational methods [40], conformal field theory [43, 45], and a modified version of

the NRG (which preserved the k-dependence of the couplings) [49]. The later bodies of

work explain that the unstable critical point only occurs when the two-impurity model

employ, ,1 is symmetric under a very specific particle-hole transformation (regardless of

whether the couplings are taken to be k-dependent). While this symmetry is physically

unlikely, a system close to particle-hole symmetry still exhibits a signature of the critical

point.

As noted above, the majority of the previous work has focused on the special case of

identical impurities, which is both computationally simpler than non-identical impurities

and physically relevant, since one often deals with impurities of the same element (as in

the ST\ I experiments described below). In the newer context of double quantum dots,

however, one can tune the dots to be non-identical, revealing a rich tapestry of new

behaviors, as will be explored in C'i lpter 5.

1.4.2 Multiple-Impurity STM Studies

Using ST\i techniques, it is possible to arrange multiple impurities in various

configurations. The simplest case to study is that of two impurities at various separations.

Figure 1-11 plots the results of an extension of the single Co atom study [60] in

which a pair of Co atoms are examined. The top two traces are the same as the results

in Figure 1-6. The bottom trace shows G(V) when a second Co atom is added, revealing

the disappearance of the Kondo resonance. The authors of the study attribute this

disappearance to the value of Ed shifting away from the Fermi level due to the addition

of the second impurity. This shift of ed causes a drastic reduction of J [see Eq. (1-16)],

thereby lowering TK below 6 K, which is the temperature at which the experiment took

place, making the Kondo resonance vanish from the data.

Figure 1-12 shows the results of a similar experiment [70] involving two magnetic

Ni atoms adsorbed onto Au(lll). The four traces show G(V) for different impurity

separations. The Kondo resonance (labeled as peak "A" in each trace) remains unchanged

until the atoms are moved to within 3.4 A of each other, at which point the resonance

1.2

S0.8

3 Co Atom

> 0.4
o Co Dimer

0.0-

-0.10 -0.05 0.00 0.05 0.10 0.15
Sample Voltage (V)

Figure 1-11. STM I differential conductance for Co atoms on a non-magnetic Au(lll)
surface [60]. Reprinted with permission from W. C'!, i1, T. Jamneala, V.
Madhavan, and M. F. Crommie, Phys. Rev. B 60, R8529 (1999). Copyright
1999 by the American Physical Society.

shrinks, indicating a drop in TK, again thought to be caused by a modification of ed. (This
time, the resonance does not disappear, since the new Kondo temperature is still above
the temperature at which the experiment took place.)
Returning to the Ce study described in Section 1.3.1, Figure 1-13 [59] shows the
differential conductance for a single Ce atom, a cluster of Ce atoms, and a film of Ce
atoms deposited onto the same Ag(lll) surface. While the general lineshape is the
same, the width 2kBTK of the dip and the .,-vmmetry parameter q change, indicating a
modification of the tunneling path interference pattern.

I p I I I I '

Spectroscopy of Co Atom vs. Co Dimer on Au(111)

SSeparation Dependence
of Ni Dimer Spectra

A
B
St

A
B St

A
B A

L 8t"

-100

age

age

-200 -100 0 100 200
Sample Voltage (mV)

#1
- ^ Sep.>12A

#2

o Sep.= 7A

age #3
SSep.= 4A

4 Sep.= 3.4A

-50 0 50
Sample Voltage (mV)

Figure 1-12. STM\ differential conductance for a pair of Ni atoms on a non-magnetic
Au(lll) surface [70]. Reprinted with permission from V. Madhavan, T.
Jamneala, K. N I, ., 1;:, W. C'!i, i J. L. Li, S. G. Louie, and M. F. Crommie,
Phys. Rev. B 66, 212411 (2002). Copyright 2002 by the American Physical
Society.

1.4.3 Double Quantum Dot Studies

Similar modification to the Kondo effect has been observed in DQD devices (two

quantum dots connected to each other and to the same external structure), which have

been proposed as a possible two-qubit system to be used in quantum computation. Figure

1-14 shows the setup of a DQD device [76] composed of dots L and R coupled by a

conducting region C. The dot energy parameters a,, U,, and A, (where a = L, R labels

the dots) and the inter-dot tunneling rate are tuned by the gate voltages VgL, Vgc, and

VR. The differential conductance of each dot G,(V) = dl/dV, is measured with the left

r

"0

100 150

__________

Ce Atom Ce/Ag(111)

f...* .'
:'

"" "a)

Ce Cluster

'' *b)

Ce Film

-.- J 4 /-
c)

-200 0 200
Sample Voltage (mV)

Figure 1-13. STM\ differential conductance dI/dV for various Ce configurations on Ag(lll)
[59]. Reprinted with permission from J. Li, W.-D. Schneider, R. Berndt, and
B. Delley, Phys. Rev. Lett. 80, 2893 (1998). Copyright 1998 by the
American Physical Society.

As in the study described in Section 1.3.2, the occupancy of each dot can be tuned by
the gate voltages VgL and VgR that control the resonant energies eL and 6R, respectively.
Coulomb blockade valleys were found in the differential conductance plots of Figure 1-15,
parts A and C. The valleys are labeled by the number of electrons in dot R, where M is
odd. Parts A and B correspond to dot L having an even number (N 1) of electrons, and
thus no local moment. As in the single QD study, the Kondo resonance is present around
zero bias only when dot R has an odd number of electrons, forming a local moment on dot
R. However, when dot L contains an odd number of electrons (parts C and D), the Kondo
resonance on dot R is suppressed. The authors of the study attribute this suppression

VgL V9C VgR

I 'l ilIR -

Figure 1-14. Setup for coupled double quantum dot experiment [76]. Dots L and R are
coupled by the conducting region C. Gate voltages determine energy levels,
occupancies, and coupling of the dots; leads measure the conductance of each
dot. From Craig et al., Science 23 April 2004 304: 565-567. Reprinted with
permission from AAAS.

to the exchange of electrons across the central region, leading to either a spin-0 state

(antiferromagnetic alignment) between the two dots with no Kondo effect, or to a spin-1
state (ferromagnetic alignment) with a weaker Kondo effect having a TK smaller than the

temperature of the experiment. Whichever behavior is the cause, the magnetic interaction

between the dots significantly modifies the Kondo effect.

This modification can be observed to develop in Figure 1-16, which plots the

conductance of the left dot GL(VL) for various strengths of the coupling between dot

R and the conducting region. When dot L and dot R both have an odd number of

electrons (part A), increasing the coupling (and therefore increasing the magnetic ordering

effect) causes the Kondo resonance to become suppressed and to split into two peaks.

This suppression and splitting behavior agrees with theoretical predictions [71, 82, 88] and

has also been observed in parallel-coupled double quantum dot experiments [77]. When

dot L has an odd number of electrons and dot R an even number (part B), increasing

the coupling causes slight quantitative modification to the Kondo resonance, but the

qualitative features remain the same.

dl/dVR (e2/h)
0.3 0.4 0.5 0.6

-710 -705 -700 -695
VgR (mV)

-690 -685 -680

------

-200 -100 0
V, 'pV

-100- -"q
M-1
-200
-710

-705 -700 -695
Vg, (mYV)

M+1

-690 -685 -680

-200 -100 0 100
VR (pV)

Figure 1-15. Differential conductance of dot R for the DQD setup shown in Figure 1-14.
(A) Coulomb blockade valleys for dot R when dot L has an even number
(N 1) of electrons. Valleys are labeled by the electron occupancy number of
dot R, where M is odd. (B) Differential conductance dl/dVR when dot L has
N 1 electrons and dot R has M 1, M, and M + 1 electrons. The Kondo
resonance is observed around zero bias when dot R has an odd number of
electrons. (C) Coulomb blockade valleys for dot R when dot L has an odd
number (N) of electrons. Valleys are labeled by the electron occupancy
number of dot R, where M is odd. (D) Differential conductance dI/dVR when
dot L has N electrons and dot R has M 1, M, and M + 1 electrons. The
Kondo resonance previously observed on dot R has been suppressed [76].
From Craig et al., Science 23 April 2004 304: 565-567. Reprinted with
permission from AAAS.

0.7 0.8

200 C

100 200

200

I 1 '1111

0.9, 0.9

0.8 0.8 N. N -M -': -U

0.7 0.7
weak
S0coupling .
ov 0.6- ul 0.6
weak
-a -
0.5> coupling
0.5 o.5-

0.4 0.4-

0.3- coupling 0.3 strong
A B coupling
0.2 I 1 0.2
-200 0 200 -200 0 200
VL( V) VL (IV)

Figure 1-16. Differential conductance of dot L for the DQD setup shown in Figure 1-14,
for various strengths of the coupling between dot R and the conducting
region, (A) when dots L and R both have an odd number of electrons and
(B) when dot L has an odd number of electrons and dot R an even number
[76]. From Craig et al., Science 23 April 2004 304: 565-567. Reprinted with
permission from AAAS.

By constructing DQD devices of different geometries and fine-tuning the device

parameters with gate voltages, one may study a great variety of physical behaviors

previously inaccessible in experiments. This new breadth of possibilities has stimulated a

large number of theoretical and numerical studies.

For example, a study utilizing the NRG and slave-boson mean-field theory [85]

working with a side-coupled DQD (for which a central dot is connected to the leads and

to a side dot) predicts a two-stage Kondo screening effect (cf. Section 1.4.1), as seen in

Figures 1-17 and 1-18. The upper panel of Figure 1-17 shows TXimp vs. T/TK, indicating

the screening of the DQD spin in two stages. The lower panel shows how this two-stage

process is evidenced in the conductance (given in units of Go = 2e2/h). The screening

of the central dot at T ~ TK enhances the conductance, which rises nearly to Go. The

screening of the side dot at a much lower T ~ 10-4TK leads to a suppression of the

conductance. This suppression is attributed to a splitting of the Kondo resonance in the

central dot spectral function.

10-2
T/'<

Figure 1-17.

Side-coupled DQD displaying two-stage Kondo screening behavior [85].
Model parameters are C1 = C2 = -0.25, U1 = U2 = 0.5, A1 = 0.035, and
inter-dot tunneling amplitude A 0.003. The upper panel shows the square
of the impurity magnetic moment p2 [kBTXimp/(g B/)2 in the language of the
present work] vs. T/TK. The lower panel shows how the two-stage process is
evidenced in the conductance (given in units of Go 2e2/h). Reprinted with
permission from P. S. Cornaglia and D. R. Grempel, Phys. Rev. B 71,
075305 (2005). Copyright 2005 by the American Physical Society.

Figure 1-18 [85] shows the conductance for this side-coupled device as e = Ce = C2

is varied at various temperatures. For T = 0 (the bottom panel), at the particle-hole

symmetric point c = -U/2 where the total device occupancy is even, the conductance

vanishes, whereas there is perfect conductance at c 0 and c -U, where the total

occupancy of the device is odd. Thus, here we see the behavior of a Coulomb blockade

valley, as observed in experimental studies.

-2 -1 0 1
e/U
Figure 1-18. Conductance vs. C = t 2 = C2 (controlled by gate voltage) for a side-coupled
DQD [85] at various temperatures. Model parameters are U1 = U2 = 0.25,
A = 0.125, and inter-dot tunneling amplitude A = 0.025. Note that the
conductance features vanish when T > TK. Reprinted with permission from
P. S. Cornaglia and D. R. Grempel, Phys. Rev. B 71, 075305 (2005).
Copyright 2005 by the American Physical Society.

In the middle panels of Figure 1-18, the second stage of the Kondo screening effect

is not taking place, and so the conductance at c = -U/2 is enhanced. In the top

panel, the temperature is on the order of TK, and so the features of the conductance

spectrum disappear. (This agrees with the experiment in [55], in which the conductance

enhancement disappears when T exceeds TK.)

Other experimental and theoretical studies have used DQD devices to examine

Coulomb-blockade behavior [72], observe interference between electron paths [75],

construct Aharonov-Bohm interferometers [58, 68, 69, 87], observe the two-channel

Kondo effect [95], construct two-level QD devices [75], probe the length of the Kondo

screening cloud [91], and observe the Kondo effect in a QD side-coupled to a quantum wire

[66, 91].

While there has been much theoretical work done, the 1 i ii ii iy of these studies has

focused on devices composed of identical quantum dots or devices for which both dots

are in the Kondo regime. Recently, various efforts [88, 91, 94, 98] have been made to

study a class of DQD devices in which one dot is in the Kondo regime and the other is

weakly interacting (i.e., has a small Coulomb energy U2, and therefore no local moment).

Attention has been paid to the special limit of U2 = 0 which, though difficult to achieve

experimentally, can be modeled as a single Anderson impurity (Dot 1) connected to a

conduction band via a nonconstant hybridization [88]

A(c)

[AvA 2 +(c ) 2
(c )2 + (A )2

(1-56)

(1-57)

where A is the inter-dot tunneling amplitude and Ai = rp2 is the effective hybridization

width of dot i with the leads.

The effects of this nonconstant hybridization have been studied [88, 94, 98] in the

specific geometries of the side-coupled DQD (in which case the Kondo-like Dot 1 couples

to the leads only through the non-interacting Dot 2) and the parallel DQD (in which

case both dots connect directly to the leads, but not to each other). In the side-coupled

configuration, for sufficiently strong inter-impurity tunneling A, the Kondo resonance

in Dot 1 splits, producing noticeable changes in the device conductance. In the parallel

configuration, there arises a pair of quantum phase transitions separating Kondo-screened

phases and a local-moment phase in which the impurity remains unscreened down to zero

temperature. Since the condition U2 = 0 is experimentally unreasonable to obtain, it is

necessary to explore the effects of small positive U2 on these behaviors. In ('!i Ilter 5, I

will employ the two-impurity Anderson model (since the effective one-impurity model is

limited to U2 0) to study how these behaviors are modified for U2 > 0 and eventually

destroyed for sufficiently large U2.

1.5 Study Overview

The data presented in Section 1.4 show that systems of coupled magnetic impurities

and their DQD analogs display a rich v ,, I' i of behaviors warranting deeper investigation.

A system of two coupled Kondo-like impurities has been shown to exhibit interesting

properties, but it is difficult to observe those properties experimentally in bulk hosts.

Studying a two-impurity system on a surface with STI\i techniques should allow one

to probe effects such as two-stage Kondo screening and impurity singlet formation.

Also, while systems of two identical (or at least two Kondo-like) impurities have been

explored in depth, systems of highly inequivalent impurities have received little attention.

This broader range of impurity configurations is achievable in systems of DQDs. The

remainder of this study will follow along these two paths of interest: The manifestation of

known two-impurity behavior in ST\ i studies and the properties of systems of two highly

inequivalent impurities in the form of DQDs.

Before embarking upon these paths, in C'!i lpter 2, I will review the transformation

of the single-impurity and two-impurity Anderson Hamiltonians into forms that can be

studied using NRG techniques. I will describe the NRG procedures used to diagonalize

these Hamiltonians and to calculate the thermodynamic and spectral properties of

interest. In the case of the two-impurity Anderson Hamiltonian, I will focus on the special

limits of identical impurities and zero impurity separation, as these computationally

simplified cases will be of use in ('!i lpters 4 and 5.

In C'!i lpter 3, I will give an overview of my efforts to improve the computational

efficiency of the NRG iterative procedure by applying parallel processing techniques to

the diagonalization of the iterative Hamiltonians HN and the calculation of operator

matrix elements, such as (N, hM dt lN, M ). To parallelize the diagonalization process, I

have employ, ,1 two methods that take advantage of the block-diagonal nature of HN: In

the first method, larger matrix blocks are diagonalized by all processors working together

using Scalable LAPACK [101] routines; in the second method, smaller matrix blocks

are diagonalized by individual processors using LAPACK [100] routines. For the sample

calculation shown in ('! Ilpter 3, the two methods result in a minimum relative wall-clock

time of 7.' for 4 processors, reflecting a mediocre scalability of wall-clock time with

increasing the number of processors. To parallelize the calculation of operator matrix

elements, I have emploil a method in which each processor calculates a "chunk" of

matrix elements at a time, again taking advantage of the block-diagonal nature of HN.

For the sample calculations shown in C'! lpter 3, this method results in a much better

scalability of wall-clock time with increasing number of processors.

In C'!i lpter 4, I will apply the NRG methods of C'!i lpter 2 to model STM studies of

surface impurities. I will begin by recapitulating the setup and the results for a single

surface impurity [73], and then present new results for a pair of identical surface impurities

without a direct interaction. By varying the impurity separation, I will demonstrate how

many of the behaviors described in Section 1.4.1 are manifested in the STM\i conductance

spectrum as the magnitude and sign of the RKKY interaction change. In particular, we

will see that impurity configurations that display a two-stage Kondo effect (IRKKY > TK)

produce a conductance spectrum given by a sum of two Fano lineshapes (each having its

own q and TK); effectively isolated impurities (ILRKKYI < TK) produce a conductance

spectrum given by a single Fano lineshape; and impurities locked into a spin singlet state

(-IRKKY > TK) produce a generally featureless conductance spectrum.
In C'!i lpter 5, I will apply the NRG methods of C'!i lpter 2 to model the aforementioned

class of highly- -i-iiiiii., 1 ii. DQD devices, beginning with the U2 = 0 special case described

above and expanding to include U2 > 0. This new range of device parameters will

require the use the two-impurity model of Chapter 2, instead of the effective one-impurity

model [88] discussed above. I will focus on the side-dot and parallel DQD configurations

described above. In the side-coupled case, I will explore the effects of positive U2 on the

splitting of the Kondo resonance and the resulting changes in the conductance minimum

[98]. In the parallel-dot case, I will explain how the pair of QPTs evolve and eventually

disappear as U2 is increased from zero.

In ('!i lpter 6, I will summarize the behaviors observed in ('!i lpters 4 and 5, and

present ideas for future work.

CHAPTER 2
BACKGROUND MATERIAL

2.1 Application of the NRG to the Anderson Model

In this chapter I will review the single-impurity Anderson Hamiltonian and show

how it is developed into a form that can be studied using NRG techniques [30, 31]. I will

then outline the procedure for using these NRG techniques to calculate thermodynamic

and spectral properties such as the impurity magnetic susceptibility and operator spectral

functions. I will then describe the fixed points that arise in the NRG treatment and how

they are manifested in the thermodynamic and spectral properties. Lastly, I will describe

the extension of the NRG techniques to the two-impurity Anderson model.

2.1.1 Discretization and Eigensolution

The single-impurity Anderson Hamiltonian [7] consists of (as ordered below) a

conduction-band term, an impurity term, and a hybridization term:

HA Z- C Clo + cddtdo + U(jdd)(dtd1) + Z(Vd4t d, + Vcdth). (2-1)
k k
All energies are measured with respect to the Fermi level (E = e eF, II < D) and

repeated spin indices (a, in this case) are implicitly summed over. The co (cq,) operators

are standard fermion operators that create (annihilate) conduction band electrons of

momentum k and spin a, and the d) (do) operators create (annihilate) impurity electrons

of energy ed, spin a, and on-site Coulomb repulsion U.

It is standard to assume for simplicity that the hybridization parameter Vjd is

a constant, Vd, and that the conduction band density of states is a constant over a

bandwidth of 2D: p(E) = po for II < D and 0 elsewhere. With these assumptions,

ed = -U/2 is a point of particle-hole symmetry, i.e., invariance under the transformation
S-dand c c In the case -d > A =TpoV and Cd + U > A, this model
d,- -dt and c .,In th ae-d Pi

describes an impurity with a well-defined spin-- for kBT < A, and exhibits the Kondo

effect at sufficiently low temperatures.

The study of the Anderson model using NRG methods [30] employs the same

discretization procedure as was described for the Kondo model in Section 1.2.2 and

[23]. The Anderson Hamiltonian is thus transformed into a linear chain form similar to

Eq. (1-37):

HA [cnfn,, + f + Afn+\,afn,t (22)
n=0
+ Cddt d + U(dtd)(dt d) + V2Vd (d fo, + h.c.).

Again choosing a density of states p(c) = po = 1/2D, I|e < D, the couplings T, and A, are

the same as in Eqs. (1-38) and (1-39).

As with the NRG treatment of the Kondo problem, the chain Hamiltonian (2-2) is

solved via an iterative diagonalization procedure. The iterative dimensionless Hamiltonian

is
N-1
HN A(N-1)/2 A-n'/2 (ft,,fiz+l, + h.c.) (2-3)
n=0
+ 6Ndtd, + UN (dd, 1)2 t+ A, + h..) ,

where is given by Eq. (1-40), and

S (2A(N12) 2 ) (24)

UN (2A(N-1)2) (25)
Ut +A-1' 2D

(2A(N-1)/2 2 2A(26)
A 1 +A-1 7D"

The eigenstates of HN, again abbreviated IN, fi), are labeled by charge Q, spin S, and

spin z-compoment S,, as HN commutes with the charge and spin operators, defined as [30]
N
QN = (ff,, -1) + (dd 1) (2-7)
n=0
N 1
SN = 2 f n, + 0d1 ,,,d. (2-8)
n=0

As described in Section 1.2.3, the NRG procedure begins by diagonalizing the initial
Hamiltonian Ho:

Ho = 6odd, + Uo (dd 1)2 + A2 (dfo,a + h.c.) (2-9)

Using the eigenvalues Eo(f7) and eigenstates |0, fJ), the RG transformation

HN+1 = R(HN) (2-10)

A12HN + vN(fv, fN+1,, + h.c.) (2-11)

is employ, 1, to construct H1, which is then diagonalized, etc.
2.1.2 Calculation of Thermodynamic Properties

At each iteration N in the NRG procedure, one can calculate certain properties of
the impurity system along an exponentially decreasing energy or temperature scale. The

two properties of primary interest in this dissertation are the impurity contribution to the
magnetic susceptibility Ximp(T) and the impurity spectral function Ad(uw).
For the calculation of Ximp(T), there is an exponential temperature scale given by [30]

TN(3) -(1 + A-1)A-(N-1)2, (2-12)
2kBs3

where 3 is a chosen parameter. (We will see the criteria for choosing 3 below.) If during a
run of the NRG algorithm one utilizes two values of /3 such that TN(Q31) = TN+(2) T,
one can compare the two resulting values of Ximp(T) as a measure of the NRG algorithm's

accuracy (and thereby evaluate the choices of A and the number of states kept).
The calculation of Ximp(TN) is given by [30]

BTN mp(TN)/(g/) ~ TrS, exp(-3Hy) Tr(SNo)2 exp(- 3H 3)
p Trexp(-3HN) Trexp(-O3H) (2

in which SN, is the z-component of the spin operator defined in (2-8), and the superscript

"0" refers to an operator for an impurity-free system, i.e.,
N-1
H A( N-1)/2 -[LA /2 (ift f, i + f, +1,f (2-14)

N
2 = ffct fn,fc. (2 15)
n=0

To evaluate the traces in Eq. (2-13), one utilizes the fact that the energy eigenvalues

of HN (H) do not depend on Sz. Thus, the traces reduce to sums over k and S, where

k represents all quantum numbers other than S and Sz. Carrying out the sums over Sz

produces the results

Trexp(- HN) (2S + 1) exp (- 3EN(k, S)), (2 16)
k,S

TrSN exp(- HN) (2S + 1) [(2S + I)2 ] exp (-EN(k, S)) (2-17)
k,S
Trexp(- H) (2S + 1) exp (- 3E(k, S)), (2-18)
k,S

Tr(So)2 exp(- 3H) (2 + 1) [(2S + 1)2 ] exp (-3E (k, S)) (2-19)
k,S

Thus, at the end of each iteration N, TNXimp(TN) may be evaluated using only the energy

eigenvalues of HN and H.

Here is where a trade-off arises in the NRG procedure. Using Eq. (2-13) incurs an

error on the order of 3/A (due to the contribution from chain sites with index n > N

that are not incorporated until subsequent iterations). Thus, choosing a small value of

3 would seem to be preferred. However, in order for Eqs. (2-16) through (2-19) to be

evaluated accurately, all energy eigenvalues up to several times 1// must be included (so

that exp (- 3EN) < 1 for the omitted states). Because the higher energy eigenstates at

iteration N 1 are truncated (see Section 1.2.3), these larger eigenvalues at iteration N

may not be calculated accurately. Thus, 3 is chosen to keep the O(3/A) error reasonable

without requiring too many eigenstates to be kept. For A = 2.5 or 3 (the values used in

the calculations for this study) 0.3 < P < 0.6 works well.

2.1.3 Calculation of Spectral Functions

As mentioned in Section 1.3.1, the impurity spectral function Ad(U, T) is needed

in order to calculate the impurity's contribution to transport properties such as the

conductance. The impurity spectral function can be evaluated using the NRG procedure

by employing the formula [44]

Ad(,(wT) Z ) (N, d N, 17n') 2 (3E(M) + 3E(m') x
,,'
6 (w (EN(') E(N())), (2-20)

where h has been taken to be 1, Q satisfies Eq. (2-12) with TN -- T, 6(x) is the Dirac

delta-function, and ZN(Q) is the partition function for an N-site chain [73]:

ZN(/) = e-3EN(m). (2-21)

The matrix elements (N, f| d) N, fi') are evaluated recursively by a transformation

similar to Eq. (2-10):

(N, J|d|N, 7') = U(, iN)UN((', R'i')6SN,{(N 1, iI d N 1, i'). (2-22)
,fl,i' ,iNi

Here, UN(T1, niN) is a unitary matrix composed of the eigenvectors IN, fi) of HN

expressed in the basis of IN 1, n) 0 1iN), where iN = 0, 1, 1, T11 labels the basis state

of site N. Thus, Ad(ow) can be evaluated at the end of each iteration N using the energy

eigenvalues EN(fi), the eigenvectors UN(fn, niN), and the recursively-calculated dt matrix

elements. It turns out that the matrix-element calculation is vastly more time consuming

than the diagonalization process.

Since the energy spectrum EN(ri) ranges from around 1 (due to the finite value of N)

to K (due to the truncation of eigenstates), Ad(o, T) can be evaluated accurately only at

U = WN(q) (similar to TN in Section 2.1.2), given by

(1 + A-')A1/2
SN(q) A-N/2, (2-23)
2

where q ~ 1. For the results in this study, I have used q = 1.25.

In practice, the Dirac delta-functions in Eq. (2-20) are replaced with logarithmic

Gaussian functions [93]:

-b2/4
( (E(') E(N))) e- [ -- -ln(/(wN(Mn)-EN(M)))/b] 2, (2-24)
J(Lw (EN(Hn7 EN(ba))) b 4(

where b is width parameter, typically chosen between 0.3 and 0.7. (The calculations in

this work have used b = 0.5.) The set of exponentially-spaced points Ad(;N) is then

connected by a spline curve [51]: a piecewise polynomial function with a continuous second

derivative. All of the spectral function results in this dissertation are for T = 0, for which

case the summation in Eq. (2-20) is restricted to terms in which M and/or i' are/is a

ground state of HN, and ZN(3) reduces to the number of ground states.

2.1.4 Fixed Points and Results

As the parameter space of the Anderson model is larger than that of the Kondo

model, the transformation R2 [see Eq. (2-10)] for the Anderson model has more fixed

points than for the Kondo model. Each of these fixed points is described by an effective

Hamiltonian Hf obtained by inserting special values (0 or oo) for ed, U, and A into the

general HA [Eq. (2-2)]. At each fixed point, the excitation spectrum can be related by

suitable phase shifts to that at the free-electron fixed point Ho, whose iterative form

Ho is given in Eq. (2-14). (As mentioned in Section 1.2.4, each of these fixed points is

technically a pair of fixed points, one for even N and one for odd N.)

The free-orbital fixed point HFo can be obtained by setting d = U = A = 0 in Eq.

(2-2). The impurity is thus decoupled from the metal, and has a set of four degenerate

states: |0), I T), { 1), TI). The iterative Hamiltonian HIO,N Ho, and the fixed point

energy spectrum is that of Ho with an additional four-fold degeneracy arising from

the four possible states of the decoupled impurity. This fixed point is characterized by

TXimp = 1/8 [30, 31]. [Henceforth, I set kB/(gpL)2 1.]

The mixed-valence fixed point HQv can be obtained by setting d = A = 0 and

U = oo (notice that this fixed point only occurs for particle-hole .'i-mmetric cases). The

doubly-occupied state I| t) is thus inaccessible, and the states 10), | ), I1) are degenerate

and decoupled from the conduction band. The iterative Hamiltonian HyV,N = HH, and

the fixed point energy spectrum is that of Hk with an additional three-fold degeneracy

arising from the three possible states of the decoupled impurity. This fixed point is

characterized by TXimp = 1/6 [30, 31].

The local-moment fixed point HLM can be obtained by setting A = 0 and -d=

U = oo (cf. conditions for local-moment formation in Section 1.1.2). Thus only the

singly-occupied states I ), 1) are accessible, and the impurity behaves as a free local

moment. The iterative Hamiltonian HLM,N = H and the fixed point energy spectrum is

that of H with an additional two-fold degeneracy arising from the two possible states of

the decoupled impurity. This fixed point is thus analogous to the J = 0 fixed point from

the RG solution of the Kondo problem and is characterized by TXimp = 1/4 [30, 31].

The strong-coupling fixed point HIs can be obtained by keeping Qd and U fixed and

finite and setting A = oo. The iterative Hamiltonian H C,N is given by a modified version

of Hk in which the sum on n is taken from 1 to N 1 (instead of from 0 to N 1).

Thus, the energy spectrum of HcN is given by Hk_ In this case, the impurity and the

ft=o, states are locked into a spin singlet and decouple from the rest of the conduction
band. This fixed point is thus analogous to the J = oo fixed point from the RG solution

of the Kondo problem and is characterized by TXimp = 0 [30, 31]. Physically, switching

from the excitation spectrum of Hk to that of Hk_1 amounts to a r/2 phase shift of the

low-energy conduction electrons. This phase shift reflects the fact that such conduction

electrons cannot hop on or off site 0 of the NRG chain because that site is locked into a

singlet with the impurity spin. It is this maximal (unitary) scattering that is responsible

for the impurity contribution to the resistance reaching its maximum value at T = 0. It is

found that, as long as A, U > 0, Hic is the only stable fixed point.

Lastly, there is a pair of similar fixed points: the empty-impurity (or full-impurity)

fixed point can be obtained by setting A = U = 0 and cd = oo (or ed = -oo), in which

case only the state |0) (or I T)) is allowed. The iterative Hamiltonian HI,N (or HJI,N) is

equal to Hk, and the fixed point spectrum is that of Hk with a shift of -1 (or +1) in the

charge quantum number Q of each state. Both of these fixed points are characterized by

TXimp = 0 and are related to the strong-coupling fixed point [30, 31].

As the NRG procedure is carried out, the Hamiltonians HN (N = 0,1, 2,...) will

follow a trajectory governed by their proximity to these fixed points. This trajectory

picture is useful in understanding the thermodynamic results of the NRG method by

comparing the temperature TN to id U, and TK. Whenever, as TN decreases (with

increasing N), TXimp passes near one of its special fixed-point values, the iterative

Hamiltonian HN is near the corresponding fixed point.

For example, Figure 2-1 plots the magnetic susceptibility for a set of NRG calculations

for three values of ed, U, and A. These calculations were performed using A = 3.0 and

keeping Nkeep = 500 states at the end of each iteration. Following convention [30], I will

identify the Kondo temperature by the condition

TKXimp(TK) = 0.0701, (2-25)

which agrees with Eq. (1-27).

Curve A demonstrates a case in which U > --d > A, displaying a number of the

RG fixed points over various temperature regimes. At higher temperatures T > U, there

is a free-orbital regime in which TXimp t 1/8. As temperature decreases to |Idl < T < U,

there is a mixed-valence regime in which TXimp t 1/6. When T decreases further to

TK < T IEdl, there is a local-moment regime in which TXimp t 1/4. (If A were

0.25

oB
0.15 B

0.1

0.05

0
le-10 le-08 le-06 0.0001 0.01 1
kBT/D

Figure 2-1. TXimp(T) vs. T for an Anderson impurity with (A) Cd = -0.01D, U = D,
A 0.001D, (B) Cd = -U/2, U = D, A = 0.05D, and (C) Cd -U/2, U = D,
A D.

zero, this regime would persist down to zero temperature.) Lastly, when T nears TK

(kBT/D w 2.6 x 10-10), the system enters the strong-coupling regime and TXimp 0.

Curve B demonstrates a case in which particle-hole symmetry (Cd = -U/2) is obeyed,

and U > A. Again, there is a free-orbital regime at T > U, a local-moment regime at

TK < T < Id1, and a strong-coupling regime at T < TK (kBT/D t 1.0 x 10-5), but the

mixed-valence regime has disappeared due to the particle-hole symmetry.

Curve C demonstrates another particle-hole-symmetric case, but with A = U.

Here, the free-orbital regime is followed directly by the strong-coupling regime, with no

local-moment behavior in between.

Figure 2-2 plots the T = 0 impurity spectral function corresponding to the parameters

for curve B in Figure 2-1. The main features are the Hubbard bands at hw = Cd and

I I I I I I I I I

6

4

2

-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8
0)

Figure 2-2. Zero-temperature spectral function Ad(wj) vs. hw/D for the parameters of
curve (B) of Figure 2-1.

Cd + U and the Lorentzian-shaped resonance of width kBTK centered around hw = 0,

signaling the occurrence of the Kondo effect. Note that, because of particle-hole symmetry,

2.2 Extension to Two-Impurity Systems

As seen in Section 1.4, a v ii, I. i of new physics arises when one considers systems of

multiple magnetic impurities. While it is challenging to consider an arbitrary number

of impurities, much of this new physics is manifested in the less computationally

demanding case of a pair of impurities. In this section, I will outline the extension of

the NRG procedure to the two-impurity Anderson Model [39, 41, 42], and then explain the

computation of thermodynamic and spectral properties.

2.2.1 Transformation to One-Dimensional Form

To study a system of two magnetic impurities, we begin with the two-impurity

Anderson Hamiltonian:

HA2 = H + Himps + Hhyb, (2-26)

ZH cC %(2-27)
k
Himps >- [eA, + U(d i t)(dl i)j + A(dd2, + h.c.), (2-28)

Hhyb (V ic'r'dc + h.c.) (2-29)
k,i

where i = 1, 2 labels the impurities, r T Rz is the position of impurity i, and A

represents a direct electron tunneling between the two impurities. Let us assume that

Vi = V, where V1 and V2 are real constants. Now the conditions for particle-hole
symmetry (invariance under di,, -- -da, and c, ct_ ) are i -U/2, A 0, and that

the density of states satisfies p(c) = p(-e).

The first step in developing HA2 into a form suitable for NRG study is to rewrite

it in a one-dimensional form. This rewriting will replace the vectors k and r with a

dimensionless energy E = e/D and the impurity separation R = Ir' rl respectively.

Let us begin by focusing on the hybridization term,

Hhyb ii (d eikicC, + h.c.) (2-30)
k,i

First, the summation over k is replaced by an integral using

Qe d3k, (2-31)
(273C

where Qo is the volume of the unit cell. This replacement will also require a transformation
of the conduction electron operators

6c ( c ) (2-32)

giving

Hhyb V= dV d3 ci c,) + h.c. (2-33)

The quantity in parentheses is identified as the field operator which annihilates a
conduction electron at site i, and whose Hermitian conjugate creates a conduction
electron at site r:

'(0) Jd3ke',. (2 34)

Thus, the hybridization term is given simply by

Hhyb Z= V [d .(K) + h.c.. (2-35)

In order to rewrite Hhyb in a one-dimensional form, we define even (e) and odd (o)
parity operators

1
2>e = (Ti) + (~2)] (2 36)

o, = (T1) -' *(~2)]. (2-37)

Replacing the '' -(r) operators with the ',,'; operators (p = e, o), Hhyb becomes

Hhyb [ ,' + o,) + h.c.] + (', + h.c. (2-38)

Jones [34] showed that, if one considers an isotropic conduction band e = eI and
transforms the basis over k into a basis over the dimensionless energy E, only four linear
combinations of the c, operators couple to the impurities at each energy E. These linear
combinations are denoted c ,,,,, and obey

{ c ,, c,Ep' 6 (E ') 6,,6,. (2-39)

If the decoupled linear combinations of c,- are neglected, the parity field operators

become

,, dE W (R, ) c,p,,, (2-40)
1-
where

(R, ) (DE) + kR (2-41)
po K kR

w /(R, E) pD sin k, (242)
po kR

and
o dk
p() o k2d (243)
27r2D dc

is the density of states per unit cell per spin direction.

Thus, the conduction band is pictured as consisting of two effective channels of

electrons, labeled by parity p. To complete the picture, even and odd impurity operators

are defined:

1
de,( (d1, + d2,), (2-44)
v2
1
do,a = /(d,1 d2,a). (2-45)

Using the c,,~, and dp,, operators, Hhyb arrives at its final one-dimensional form

Hhyb p f dw (R, )x (2-46)
2 _

[(V. + V2) dl,,c,p,, + (Vi V2) dt-,pac + h.c. ,

where -p indicates the parity opposite to p.

Using the new set of parity operators, the conduction-band term is similarly into

H = ,, P + (2-47)
S-1

where "..." represents the contribution of states that are completely decoupled from the
impurities and are neglected. The impurity term transforms into

Himps = +2 U + 62 + -U2 dde, + t,da) +

S(c. + U1 2 U2) (dt-d,, + d2,)
11 1
2 2Ut- 2 d C do 0 + +]

1U
+2 Ul [d e,7 + d ,do,a + ddo,a + dde,, 1]2

+A (dtde, dtdo,)) t(UI + U2). (2 48)

2.2.2 Discretization and Eigensolution

To transform HA2 [defined in Eqs. (2-26) through (2-29)] into a discretized form, it is
necessary to define a new basis of conduction electron wavefunctions ,' ,(E) (associated
with energies E > 0) and I', ,() (associated with energies E < 0). These states, which
are centered around the impurity sites, are generalizations of the states bm,,q(E) defined
in Eq. (1-36). Again, m is the index of an energy bin, as depicted in Figure 1-2 and q
is the Fourier harmonic index. As in the single-impurity case, states with q / 0 do not
couple to the impurity, and can be ignored as a reasonable approximation as long as the
discretization parameter A is not too much greater than 1.
The states with q = 0 are given by

,,() Awp(R, ) A-(n- ) < < A-", (2-49)

', (,) B.,p (R, E), -A-" < < -A-(n-1), (2-50)

where

A [ dE I,(R, ) (2-51)

Sd-E I(R, E). (2-52)
J-A-"

As in the discreitzation of the single-impurity problem, the states ,, ,=o(E) and

', i=o(E) are associated with the set of operators f,,,,. The innermost state f0,p,,

(i.e., the one that couples directly to the impurity) is given by

1 +1
fo,pa 7 dE Wp (R, ) cp,, (2-53)
V'F_ 2_

where

F =dE d ,(, E) (2-54)
2 11

Using the fn,p,, operators, HA2 is thus transformed into a linear chain form, with the

conduction band and hybridization terms given by

H,, = p,, ,'+ P'p + A p,n ( ,p,,7n.+,p,i + h.c.)] (2-55)
n,p

Hhyb D p [(VI + V2) d,f0o,p, + (VI V2) dp, o,p, + h.c. (2 56)
P

and Himps given by Eq. (2-48). Here, the couplings cp,T and XAp, are more complicated

than their single-impurity forms in Eqs. (1-38) and (1-39), and depend on R through

wp (R, E).
Finally, an iterative dimensionless Hamiltonian analogous to Eq. (2-9) is defined:

HN = H,N + Himps,N + Hhyb,N, (2-57)

where

A(N-1)/2 N
Hc,N = 1+ A-)D S Y [ pf fn + pn (fn,p,+fn+l,p, + h.c.)] (2-58)
n=0 p

Himps,N 2 sum ,d,, dt,,d,,,,) + ff (ddo, + dt,de,,)
+2 sum[ d e,id t + Vdo,td do,j
+ + t1 id,1, d, + 0,t

+dI,1 de dt ,gdot +d ,do,Id do,t + d doid det

+d, de,T ,l do, + ddo, tddeo, + dTdo,t d, de,,
't tl c t ctI c t tl

+d, d,T~ddo,i + d o, td, T d d + dT,i, d,tdt, d ,

d d
( sum 1/2 sum 1/2
H ,N d d p,, op, d+ h.c. (2-60)

where
1A1 D 2) (

2A(N-1)/2 1 U1 2 (261)
1+ A-1 D D d
dif 2A(N-1)/2 l+ -1 ) (262)
N 1+ A- 1 D '2
A(N-1)/2 c1l 2 (2 63)
AN 1 + A-1 D '
S A(N-1)/2- (2
/sm 1 + A-1 D2

(2A(N-1)/2 22po (VI V2)2 (266)
Asum + D (2-65)
N 1 + A-1 D

N t +A-1 D

Thus, by applying a renormalization group transformation similar to Eq. (2-10), the

two-impurity Anderson model may be solved iteratively.

2.2.3 Special Cases: Identical Impurities and R= 0

The presence of the two electron channels illustrates the greater computational

complexity of the two-impurity Anderson model over the single-impurity Anderson

model. The bases of the iterative Hamiltonians HN square in size compared to the

single-impurity model, requiring more states to be retained after each NRG iteration in

order to achieve the same degree of accuracy. Another complication is the modification of

the conduction-electron density of states by the addition of the factor sin (kRR)/(kRR) in

Eqs. (2-41) and (2-42). Due to this factor, the densities of states of even- and odd-parity

states are inequivalent, and the two-impurity model becomes particle-hole .i-vmmetric

even if p(c) = p(-e). In addition to the more complex nature of the calculations, there are

now eight parameters to investigate (ci, Ui, Vi, A, and R). However, these complications can

be simplified by considering one of two special cases: identical impurities or R = 0.

In the special case of identical impurities (c1 = 2, U1 U2, V1 = V2), Eq. (2-26)

exhibits parity symmetry-i.e., it remains unchanged under the transformations di d2

(or de de, do -do in parity-operator language) and 'i -+ r2. As such, parity is a

conserved quantum number, as there are no matrix elements of HN that connect states

of opposite parity (cf. the "sum" and "diff" terms in HN). This property causes HN

to be broken up into smaller matrix blocks. In addition to this simplification, there are

three fewer parameters to investigate. Because one often considers impurities of the same

element, this special case is very relevant. For example, I will utilize parity symmetry in

the context of surface impurities in C'i plter 4.

Another simplified case is the limit of zero impurity separation. For rP = r- = 0,
w2 (0, ) =2p(D)/po and w2 (0, ) = 0. Thus, the hybridization term simplifies to

Hhyb -= JI d 2p(DE)/po x (2-67)

[(VI + V2) Kd,,ce, + (VI V2) dt,,ce,, + h.c.] .

Therefore, the odd conduction electron states decouple from the impurity. Because of

this decoupling, the odd conduction electron states (i.e., all of the fn,p=o, states) may be

ignored completely, thereby reducing the basis of the iterative Hamiltonians HN.

Even though there is now effectively only one conduction band, the model is still

much richer than the one-impurity Anderson model: The impurity part of the Hamiltonian

Himp,N in Eq. (2-59) contains four (rather than two) distinct impurity operators and seven

(rather than three) coupling parameters. While the limit of R = 0 may seem physically
unreasonable for magnetic impurities in a metal, it is a good approximation whenever the

two impurities are separated by a distance R satisfying kFR < 1, where kF is the Fermi

wavevector. This condition is assumed to apply in the studies of double quantum dots in

C'! ipter 5.

2.2.4 Calculation of Thermodynamic and Spectral Properties

The methods for obtaining the magnetic susceptibility and the impurity spectral

function outlined in Sections 2.1.2 and 2.1.3 can also be applied to the two-impurity

model. Applying Eq. (2-13) to the two-impurity model will calculate the contribution of

both impurities to the magnetic susceptibility.

To calculate spectral functions using a formula analogous to Eq. (2-20), the matrix

elements of dt are replaced by some combination of the matrix elements of dt, and dS,,

permitting the calculation of a variety of spectral functions. For example, if one wishes to

calculate the linear conductance through a double quantum dot device, it is necessary to

employ the Landauer formula [6, 18]

/+o0
G(T)- Go ( df/dw) [ Im1(w)], (2-68)
-oo

where Go = 2e2/h is the conductance quantum, f(, T) = [exp (w/T) + 1]- is the

Fermi-Dirac function, and T(w) is the transmission matrix, whose imaginary part is given

by

Im(w) {AIAII(wL ) + A2A22() + A2 [A12(w) + A21(U)] (2-69)

The spectral functions of interest may be calculated using the NRG formula

Aij( w,T) 1- _N-, N d\N, i') (N, MI d IN, m') x
ZN m) m f -
(E- (m) + -EN(m') (_ (EN( ) EN(f))). (2 70)

CHAPTER 3
PARALLELIZATION OF THE NRG PROCEDURE

As described in Section 2.2.3, one of the drawbacks of using the NRG to study

the rich physics of the two-impurity Anderson model is the scale of the computational

requirements. The bases of the iterative Hamiltonians HN are squared in size compared

to the single-impurity model. Because of the greater complexity of the system (such as

the presence of two inequivalent channels), a larger number of states must be retained

at the end of each iteration in order to achieve an acceptable level of accuracy. The

computational rigor becomes even greater when studying three-impurity systems [61, 84].

In order to improve on the computational efficiency of two-impurity studies (and

more complicated systems), I have sought to adapt the iterative NRG algorithm to

utilize parallel processors, implemented using the Message Passing Interface (\!PI, [99]).

The two most time-consuming aspects of the algorithm are the diagonalization of the

matrices HN and the recursive calculation of operator matrix elements [see Eq. (2-22)].

As mentioned above, the matrix element calculations require much more computer time

than the Hamiltonian diagonalization. It is shown below that the calculation of operator

matrix elements also reaps a greater benefit from the use of parallelization.

3.1 Parallelization of the NRG Eigensolution

I have achieved parallelization in the eigensolution of HN in two v-.-,- The first

method takes advantage of the block-diagonal nature of HN by assigning each of the

individual blocks of the Hamiltonian to a different processor to diagonalize individually

using standard LAPACK routines [100]. This method employs a master-slave arrangement

of the processors, in which one node (the master node) assigns the matrix blocks to

the other nodes (the slave nodes) and manages the results; since the master node does

not perform any calculations itself, it sits idle for most of the run. Because of the

straightforward division of the independent diagonalization tasks, this procedure falls

under the category of ilI i 1i --oij,!y parallel" algorithms, meaning that as long as there

is no significant overhead, the wall-clock time should scale roughly as 1/Np, where Np is

the number of processors in use. This method does require a good deal of inter-processor

communication: At the beginning of each iteration, the nodes must be updated with

the energy eigenvalues and eigenstates of the previous iteration [in order to evaluate the

recursive relation in Eq. (2-10)]; during each iteration, the master node must combine the

results from the diagonalization of the matrix blocks and communicate up-to-date status

on which states to truncate.

The second method involves using Scalable LAPACK (ScaLAPACK, [101]) routines to

diagonalize a single large matrix on multiple computer nodes. ScaLAPACK requires that

the matrix be divided up among the processors, resulting in a great deal of communication

between the processors during the calls to ScaLAPACK routines. For the case considered

below, this communication quickly creates a noticeable overhead as Np is increased.

In practice, I utilize these two methods together to minimize the computer wall-clock

time, employing the first for smaller matrix blocks and the second for larger matrix blocks.

The most efficient balance of the two methods depends on the computer hardware being

used. On clusters with slower inter-processor communication (such as the UF Physics

Department's dragon cluster), it is more advantageous to primarily use the first method,

while on clusters with faster inter-processor communication (such as the UF HPC cluster),

it is more advantageous to apply the second method.

Figure 3-1 illustrates the improvement in performance for iteration N = 5 for an

NRG calculation that retained 3000 states at the end of each iteration. These runs were

performed using different numbers of 2.2-GHz AMD Opteron processors connected by a

gigabyte-ethernet network. The vertical axis is the computer wall-clock time, scaled by

the wall-clock time for a single-processor run, and the horizontal axis is the number of

processors Np involved in the calculation. Initially (Np = 2, 3), there is a steep decrease

in the amount of wall-clock time. There is markedly better improvement between 2

and 3 processors than there is between 1 and 2. This difference is due to the fact that,

during the Np = 2 run, the master node sits idle while the one slave node diagonalizes

the matrix blocks of dimension less than 100, whereas during the Np = 3 run, two

slave nodes diagonalize the matrix blocks of dimension less than 100. After the initial

improvement in performance, the wall-clock time saturates, remaining nearly constant

for 4 < Np < 7. Upon using Np > 7, the wall-clock time begins to increase, due to

the increasing overhead of communication between the processors. This interpretation

is evidenced by the linear-like increase in the wall-clock time with increasing Np. The

inefficient performance of this calculation is attributed to inefficiencies in ScaLAPACK,

which rarely sees remarkable improvement for matrices with dimensions less than several

thousand [101].

In general, it turns out that the iterative eigensolution process does not benefit

significantly from the application of parallel processing techniques, as evidenced by

the minimum relative wall-clock time of i.'. in Figure 3-1. While there is improved

performance for larger values of Nkeep, it is rarely necessary to keep more than 3000

states (in current NRG endeavors). Also, the benefit accrued by using multiple processors

is usually outweighed by the amount of time spent waiting for many processors to all

become available at once in a typical high-performance cluster environment in which many

users' jobs compete for computational resources. However, future studies utilizing the

NRG procedure that require a higher number of states may benefit from the preceding

parallelization techniques.

3.2 Parallelization of the Matrix Element Calculation

The calculation of the matrix elements of operator such as dt [see Eq. (2-22)] is

extremely computationally demanding. The necessary computer time grows as the cube of

the Hamiltonian dimension, and is even larger than the time required for the eigensolution

process. However, I have succeeded in greatly reducing the wall-clock time for the matrix

element calculation by, again, adapting the algorithm to run on parallel processors. In

this implementation, each processor works on a "chunk" of matrix elements independently

W I I I I I I

S0.9

S0.8

0.7
B -

-

0

0

0

0 o o 0

_

0.6
0

Figure 3-1.

NP

Wall-clock time vs. number of processors Np for eigensolution of HN 5.
Matrix blocks with dimension less than 100 were diagonalized by individual
processors using LAPACK routines; matrix blocks with dimension greater than
100 were diagonalized by all processors using ScaLAPACK routines. The
vertical axis is scaled by the wall-clock time for a calculation utilizing a single
processor.

(made possible by the block-diagonal nature of HN and of the matrix element arrays),

which causes the wall-clock time to decrease rather quickly as the number of processors

increases. This method employs a master-slave arrangement of the processors, in which

one node (the master node) assigns the chunks to the other nodes (the slave nodes) and

manages the results. Since the master node does not perform any calculations itself, it sits

idle for most of the run.

Figure 3-2 shows the benefits of applying the above procedure to the calculation

of the matrix elements of dt at the end of a single iteration. The vertical axis is the

wall-clock time scaled by the wall-clock time for a single processor run. The horizontal

axis is the number of processors Np and the traces are labeled by Nkeep. Here there is a

rather significant improvement in performance, with 6 processors reducing the wall-clock

time by 91 '- The data are plotted on a log-log scale and are fitted with power laws to

illustrate the nature of the improvement in performance.

For Nkeep = 2000, the wall-clock time scales as 1/Ny, with a t 1.5. For Nkeep = 4000,

the wall-clock time does not scale as smoothly as in the keep = 2000 case, but the

wall-clock time still follows the general trend of 1/Naeep, with a t 1.2. These values

of a would seem to exceed the optimum value of a = 1 that one can expect from an

-. ii in .i~--ii!i!y parallel" algorithm. However, the variation in the wall-clock data for the

Keep = 4000 runs indicate the amount of error associated with analyzing this performance

behavior. For example, a run that utilizes processors located on the same node will

have much faster inter-processor communication than a run that utilizes processors

located on different nodes. Another possible factor is the difficulty in guaranteeing that

a parallel calculation has access to 10(I' of the computational power of all Np nodes; if

other jobs utilize the same processor, the computational performance will decrease and

wall-clock time will increase. This situation is typically guarded against by job scheduling

software, but is not ahv--, avoided. This second error is more likely to occur during

a run that takes a longer amount of time, which may explain why the fluctuations in

the wall-clock time are so much greater for the larger value of Nkeep. It would therefore

be a worthwhile endeavor to see how this algorithm scales in a more controlled parallel

processing environment.

In spite of such errors and difficulties, these results represent a much larger

p li-off than was seen for the eigensolution phase of the NRG procedure. This superior

improvement is attributed to the fact that the parallelized matrix-element algorithm

requires much less inter-processor communication than the parallelized eigensolution

does. Notwithstanding these improvements, the benefits achieved by utilizing many

processors are often outweighed by the amount of time spent waiting for those processors

U. I

0.1-

0O

0

0.01
1 10
NP

Figure 3-2. Wall-clock time vs. Np for calculation of dt operator matrix elements. Traces
are labeled by number of states kept and fitted with power laws. The vertical
axis is scaled by the wall-clock time for a calculation utilizing a single
processor.

to become available. When many independent NRG calculations are required (such as

for the conductance plots in C'! lpter 5), it is typically more efficient to perform these

calculations simultaneously on independent processors. However, future NRG endeavors

or improvements in cluster management may bring about a greater benefit from the

parallelization techniques described above.

CHAPTER 4
STM STUDIES

4.1 Review of Single-Impurity Behavior

4.1.1 Single-Impurity STM Setup

As already seen in C'! ipter 1, magnetic impurities produce interesting effects when

studied using scanning tunneling microscopy. In Figure 4-1, we see the setup for the

study of a single magnetic impurity adsorbed onto a non-magnetic surface with the ST\ i

tip situated directly over the impurity. As depicted schematically by the dashed arrows,

electrons in the STM tip can either (1) tunnel (with matrix element td) into the impurity,

and then tunnel (with matrix element Vd) from the impurity into the surface, or (2) tunnel

(with matrix element tc) directly into the surface at a location .,idi ient to the impurity.

Tunneling to other locations on the surface is ignored in this model as the tunneling

current density decreases exponentially with tip-surface separation.

The ST\ \ differential conductance can be calculated using the iterative NRG method.

Assuming that the tip is centered directly over the impurity, the essential physics of the

conductance is di-pli' '. 1 in the formula [73]

dl 41Fe2
G(V)= pt-2A,(eV/h), (4-1)
dV h

where p is defined below, e is the magnitude of the electron charge, V is the voltage of the

surface relative to the tip (such that V = 0 corresponds to alignment of the two Fermi

energies), pt is the tip density of states (taken as a constant about ep), and A,(w) is the

spectral function of the tunneling operator

a = p [tddL + tcr (tmp)]. (4-2)

The field operator ?4(rtmp) creates an electron in a surface state .,.i] i'ent to the impurity

and p = V/ltd2 + tS 2 is a normalization factor introduced to maintain f_ Aa(w)du 1.

STM Tip

td

Impurity to
d/

Non-magnetic Metal
Figure 4-1. Schematic STM setup indicating possible tunneling paths from the STM tip to
the metal surface.

Following [73], I assume that the surface-state operator /(rimp) is equivalent

to the innermost Wilson-shell operator fot,. While this equivalency is a significant

assumption, relaxing it should only change the quantitative details of the results, and not

the qualitative features. Using the NRG procedure, we can calculate the matrix elements

of the operators dt and fot in the basis of the eigenstates of each HN and evaluate Aa at

U = WN (see Section 2.1.3 for details) using a formula analogous to Eq. (2-20):

A,(w, T)

2
P- 2 Y 'I,| (td + tf,) IN, n,'/) |2 x
ZN(/3) (N' fin
_3EyOin) + 3EOWin)) w (N(~' ~))

(4 3)

Here, we see that interference terms proportional to tdtc will come into pl i v, producing

the .,i-vi'i I ic Fano lineshape observed experimentally (see Sections 1.3.1 and 1.4.2). As

mentioned in Section 2.1.3, all of the results in this chapter are for T = 0, in which case

the summation in Eq. (4-3) is restricted to terms in which fn and/or fi' is a ground state

of HN, and ZN(3) reduces to the number of ground states.

4.1.2 Results for Single-Impurity STM

Here I will review NRG results for an ST\ i study of a single isolated magnetic surface

impurity. To ensure that the impurity develops a well-defined spin-1/2 local moment and

a noticeable Kondo temperature, I have worked with model parameters (measured in units

of D) d = -U/2 = -0.5 and A = 0.051, with a top-hat density of states p(c) = po for

Iec < D and 0 otherwise. In the NRG procedure, I have used a discretization parameter

A = 3.0 and kept the lowest-lying 500 energy eigenstates at the end of each iteration.

0.8

t-tc = 0.0
0.1
0.2
0.6 0.3

S0.4 -

0.2 -

0 I I I
-0.004 -0.002 0 0.002 0.004
eVID

Figure 4-2. Conductance (arbitrary units) vs. bias voltage V for an ST:\I tip located
directly over a single magnetic impurity. Model parameters are (in units of D)
d = -U/2 = -0.5, A = 0.051, with td/te as labeled in the legend. The NRG
calculations were performed for A = 3, retaining Nkeep = 500 states after each
iteration. The lineshapes are similar to those in Figures 1-6 and 1-7.

Figure 4-2 plots the ST\i differential conductance (in arbitrary units) for this system

for various values of tdl/t (the absolute values of td, t, and pt only change the vertical

scale), essentially reproducing the results of [73]. (The NRG data points have been

4

2

0
-0.004 -0.002 0 0.002 0.004
eVID

Figure 4-3. Conductance (arbitrary units) vs. vias voltage V for an STM tip located
directly over a single magnetic impurity. Model parameters are as for Figure
4-2, except tc = 0. In this case, the conductance is directly proportional to the
impurity spectral function Ad(U) (cf. Figure 2-2).

fitted with spline curves.) As tdl/t increases, and the fraction of electrons that tunnel

via the impurity becomes larger, the conductance lineshape becomes .,i-ii .i I1 ic due to

interference between the two tunneling paths. Figure 4-3 shows the conductance for t, = 0,

in which case there is no direct tunneling to the surface and G(V) has the same form as

It is important to note that all of the interesting features of the lineshape occur on

a small energy scale determined by kBTK, which in this case is 1.2 x 10 -D. This small

energy scale is even more visible in Figures 4-4 and 4-5, which show the same conductance

spectra on a logarithmic voltage scale.

0.6 U- | -3
\

0.4 -

0.2 /

0 I
le-08 le-06 0.0001 0.01
eVID

Figure 4-4. Positive-bias data from Figure 4-2, replotted on a logarithmic voltage scale.

As examples, the NRG results for td/t, = 0.0 and 0.3 are successfully fitted with a

Fano lineshape in Figures 4-6 and 4-7, respectively. The fits were obtained using [73]

q2 1 + 2xq
G(V) = o + a + 2xq(44)
1+> (44)

where go is a background conductance, a is a constant, x = eV/kTK, and q is Fano's

.i- ii 1! I ry parameter [cf. Eq. (1-50)]. (Note that the NRG results show a broader

resonance than the fits; this is typical of NRG spectral results, which tend to be less

accurate at higher energy scales.) The fitted values of q for td/t = 0.0 and 0.3 are 0.0 and

-0.84, respectively. These values are similar to those found for the STM studies of single

Ce and Co atoms in Section 1.3.1, which showed similar lineshapes.

0.6 |I __ .3 -

0.4

0.2

8.01 0.0001 le-06 le-08
-eVID

Figure 4-5. Negative-bias data from Figure 4-2, replotted on a logarithmic voltage scale.

4.2 Two-Impurity STM Studies

4.2.1 Two-Impurity Set-up

Figure 4-8 shows the setup for an STM study of two impurities separated by a

distance R. I take these impurities to be identical (i.e., they have the same values of

Ci = td, Ui = U, and Vi = V). I also assume that these values do not change with

R (as ed did in the experiments described in Section 1.4.2), and that there is no direct

tunneling from one impurity to the other [i.e., A = 0 in Eq. (2-28)]. Thus, the impurities

only exchange electrons via the conduction band. I also assume that the ST,\ [tip is

situated directly above one of the impurities (impurity 1 in the diagram), such that the

tip's electrons tunnel only into impurity 1 and the surface surrounding it, and not into

impurity 2. This assumption means that I can use the same tunneling parameters td and

0.6

0.4 O O

0.2

-8.004 -0.002 0.002 0.004
eVID

Figure 4-6. NRG results for the single-impurity STM conductance G(V) for the
parameters shown in Fig. 4-2 with tdl/t = 0 (squares) fitted to a Fano
lineshape (line). The best-fit value of the Fano parameter q in Eq. (4-4) is
q =0.

t, for comparison with the one-impurity results. (C'!i lpter 6 discusses a possible method

for considering a tip at ,;,' location.)

The Ruderman-Kittel-Kasuya-Yosida (RKKY) interaction [4, 5] is an indirect

coupling that arises between local magnetic moments in a metal, and will p1 li a role

in the two-impurity system described above. The RKKY interaction, which leads to an

HRKKY = -IRKKY 1- 2 (4-5)

in the effective low-energy Hamiltonian, is mediated by the conduction electrons. Even

when there is no direct exchange interaction between the impurities, their spins can

become aligned or anti-aligned, depending on the magnitude and sign of IRKKY. For

the two-impurity Anderson model with ed = -U/2, a constant density of states, and a

0.6

0.4 b

0.2

-8.004

Figure 4-7.

-0.002 0 0.002 0.004
eVID
NRG results for the single-impurity STM conductance G(V) for the
parameters shown in Fig. 4-2 with td/tc =0.3 (squares) fitted to a Fano
lineshape (line). The best-fit value of the Fano parameter q in Eq. (4-4) is
q -0.84.

STM Tip

Impurity

1 ;
v I
I (

Impurity 2 0

VIl
op.

Non-magnetic Metal

Figure 4-8. Schematic STM setup for a study of two impurities.

0.015

0.01

0.005

FM
AFM FM
0. I I
0 2 4 6
k,

Figure 4-9. Plot of f(kFR) entering Eq. (4-6), calculated for the top-hat density of states
and band dispersion e ~ k3 employ, ,1 in this chapter. Regions of
ferromagnetic and antiferromagnetic interaction are labeled FM and AFM,
respectively.

dispersion e ~ k3, the strength of the RKKY interaction is given by [52]

IRKKY = 21n2 D(poJ)2f(kFR), (4-6)

where poJ = 8A/rU < 1 is the spin-exchange coupling entering the Kondo model and

f(kFR) is an oscillating function of the separation R multiplied by the Fermi wavevector

kF, plotted in Figure 4-9.

When IRKKY > 0 the impurities exhibit a ferromagnetic interaction (favoring the

formation of an impurity spin triplet), and when IRKKY < 0 the impurities exhibit an

antiferromagnetic interaction (favoring the formation of an impurity spin singlet). In

an ST\ i experiment, changes in the magnitude and sign of the RKKY interaction may

be observed by moving the impurities using the ST \i tip. We shall see that the change

between ferromagnetic (FM) and antiferromagnetic (AFM) RKKY interactions exerts a

key influence on the shape of the ST\ i conductance spectrum.

As a guide to the ST i study, I follow previous work [32, 35] and compare IRKKY with

the characteristic energy scale of the Kondo effect, given by kB times the (R-independent)

one-impurity Kondo temperature [54]:

kBTJ-imp D VoJexp(1.58poJ -1/poJ). (4-7)

To plot the ST\ i conductance, I use an extension of Eq. (4-1)

(V) dl 41re2
G(V) = p -2A, (eV/h), (4-8)
dV h

where Aa, (w) is the spectral function of the impurity-1 tunneling operator

a,, = p [d, + i)(r )]. (4-9)

As in Section 4.1.1, where I took ?4(rimp) o f0t,,, here, I take the operator it(ri) to be

(up to a prefactor) fo,1,, \ Foea + Fooa, where Fp is defined in Eq. (2-54).

Again, this assumption will only change the quantitative details of the conductance

results, but not the qualitative features.

4.2.2 Thermodynamic and Spectral Results

To give an idea of the behavior of a system of two impurities with varying separation,
Figures 4-10 and 4-11 plot the susceptibility and the d>,-spectral function Ad, for several

values of kFR, along with the one-impurity results for comparison. (Note that the

one-impurity susceptibility results have been multiplied by 2 to represent two impurities

at infinite separation.) As in Section 4.1.2, these results are for (measured in units of

D) ed = -U/2 = -0.5 and A = 0.051, with a top-hat density of states p(c) = po for
ce| < D and 0 otherwise, and a dispersion ke ~ k3. With these model parameters held

constant, I have evaluated Eqs. (4-6) and (4-7) for five different impurity separations:

kFR = 1.5, 2.0, 2.5, 2.84 and 3.3. These values give IRKKY/T'1ip = 400, 190, 50, -0.23,
and -19, respectively. As before, I have used discretization parameter A = 3.0, but for

0.4 ~ 2.5 -
-284 /
3.3
1-imp. results

TID
S /

S0.12 /

III /

1e-14 le-12 le-10 le-08 le-06 0.0001 0.01 1
k T/D

Figure 4-10. Impurity susceptibility TXimp vs. temperature for two impurities, each having
U = -2ed = D and A = 0.051D, and five different impurity separations R
specified in the legend.

these calculations, I kept 2000 states at the end of each iteration (since the two-impurity

problem's Hamiltonians have a much larger basis).

For a separation kFR = 1.5, the RKKY interaction is strongly ferromagnetic, as

shown by the ratio IRKKY/T-mp = 400. At relatively high temperatures (kT/D ~ 10-4),

the ferromagnetic RKKY interaction causes the two impurity spins to point in the same

direction, creating a spin-1 singlet state for which kBTXimp/(gp9B)2 ,= as seen in Figure

4-10. This combined spin-1 is then screened out by the Kondo effect in two stages, seen

as the two separate falls in TXimp (occurring at kBT/D ~ 10-5 and kBT/D ~ 10-12)

separated by a relatively flat region. This two-stage screening is also manifested in Figure

4-11, where there are two rises in the spectral function, occurring over the same energy

ranges at which the drops in TXimp occurred, again indicating a two-stage Kondo effect.

8 -- i- -imp. results

-

4 -

0
le-14 le-12 le-10 le-08 le-06 0.0001 0.01 1

Figure 4-11. Impurity-1 spectral function Ad, vs. frequency U for the same cases as shown
in Figure 4-10.

The temperature and energy scales over which this two-stage screening process occurs

correspond to two Kondo temperatures TK, where p = e, o labels the even and odd

conduction band channels [32].

For a separation kFR = 2.0, the susceptibility indicates that the ferromagnetic RKKY

interaction again creates a spin-1 singlet state at relatively high temperatures, but now

the two energy scales over which the Kondo screening occurs are much closer to each

other, and there is no flat region between the two falls in TXimp. (As kFR increases from

1.5 to 2.0, this flat region progressively narrows in width.) This proximity of the two

energy scales is also seen in the spectral function for kFR = 2.0, in which it is difficult to

distinguish two separate rises toward the zero-frequency peak.

For a separation kFR = 2.5, the susceptibility resembles the one-impurity results

(which have been scaled up by a factor of 2). This resemblance indicates that the RKKY

interaction is too weak to align the impurity spins, such that the two spin-I moments

are Kondo-screened independently. This independent-screening effect is also evidenced in

the spectral function, which matches up very closely with the one-impurity results. (This

independent-impurity behavior is surprising given the estimated ratio of IRKKY/Tk1imp

50; see below for possible explanations.)

Finally, the two antiferromagnetic cases (kFR = 2.84 and 3.3) show very different

behavior than the ferromagnetic cases did. The susceptibility plot shows that the

antiferromagnetic RKKY interaction causes the two impurity spins to point in opposite

directions, creating a spin-0 state for which TXimp falls with decreasing temperature

even faster than it does in the one-impurity Kondo effect. Because of this dominant

development of a spin-0 system, the Kondo resonance in the spectral function is very

strongly suppressed for the strongly antiferromagnetic cases. As we shall see in the next

section, this suppression enters into the STM\I conductance spectrum, as well.

As mentioned earlier, the independent-impurity behavior found for kFR = 2.5 seems

inconsistent with the estimated ratio of IRKKY/T1-mp = 50. A similar inconsistency occurs

in the case for kFR 2.84, which has a predicted ratio of IRKKY/T1im = -0.23, yet

shows behavior in which the the antiferromagnetic RKKY interaction dominates over the

Kondo effect. It is most likely that these inconsistencies (which occur as one approaches

IRKKY = 0) indicate (1) an inaccuracy in the perturbative expressions for IRKKY and

TI"imp in Eqs. (4-6) and (4-7), and/or (2) errors arising from the NRG discretization.

Also, such errors may be exacerbated by the proximity of the unstable quantum critical

point described in Section 1.4.1. Whatever the reason, the five cases studied in this section

do effectively display the range of qualitative behaviors expected in the two-impurity

Anderson model.

4.2.3 Two-Impurity STM Conductance

Figures 4-12 through 4-15 plot the STM differential conductance for each of the

five values of kFR considered in the previous section, along with the results for a single

impurity. Figures 4-12 and 4-13 show results for td/t = 0.1, which corresponds to

approximately 9' of the electrons that leave the tip tunneling directly into the surface

and 1 tunneling into the impurity; Figures 4-14 and 4-15 show the results for td/tc = 0.4,

which corresponds to approximately I.' of the electrons that leave the tip tunneling

directly into the surface and 1,!' tunneling into the impurity. For the ferromagnetic cases,

increasing td/te causes an increase in the zero-bias value of G, as was seen in the case of a

one-impurity study (cf. Figures 4-4 and 4-5). For the antiferromagnetic cases, increasing

td/te causes almost no change in the conductance spectrum, indicating the isolated nature

of the spin singlet state of the two impurities.

For the cases with kFR = 1.5, G(V) exhibits similar features to the one-impurity

case, but, as was seen in the susceptibility and spectral function, the changes in G(V)

occur over two easily distinguished energy scales. For the cases of kFR = 2.0, G(V) shows

lineshapes similar to those for kFR = 1.5, but the two energy scales are more difficult to

distinguish. As seen in the susceptibility and spectral function, the results for kFR = 2.5

closely resemble those for the one-impurity case. For the two strongly antiferromagnetic

cases, the conductance spectrum is essentially featureless for |eV/D| < 10-5 due to the

suppression of the Kondo resonance seen in the spectral function.

These results indicate that the competition between Kondo screening and the

RKKY interaction is clearly revealed in the conductance spectrum. Thus, it is possible

to study this competition in STA\i experiments simply by moving the two impurities. We

also see that it is possible to study one- and two-stage Kondo screening effects in such

experiments. For a relatively strong ferromagnetic RKKY coupling, the effective Kondo

temperature [as defined in Eq. (2-25)] drops with separation, and (as a result) the lowest

energy scale of the conductance lineshape also decreases with separation. For a relatively

strong antiferromagnetic RKKY coupling, the suppression of the Kondo resonance causes

the lineshape to remain featureless for smaller energies.

0.4 2.5
2.84

1-imp. results I
0.3

0.2

0.1- 1/ -

0 1 I I I I
le-12 le-10 le-08 le-06 0.0001 0.01
eVID

Figure 4-12. ST:\i differential conductance vs. positive bias voltage for the same
two-impurity cases as shown in Figure 4-10, with tunneling into impurity 1
described by td/tc 0.1.

The two-impurity conductance results may also be fitted with Fano lineshapes,

though the two-stage Kondo screening effect requires a sum of two Fano lineshapes to be

used. For example, Figures 4-16 and 4-17 show the NRG results for td/tc = 0.4, kFR = 1.5

fitted with the formula [cf. Eq. (4-4)]

ql 1+ 2x1ql q1 1+ 2x2q2
G(V) go+ ai 1 -+ a2 1 x2 t (4-10)
1 + x2 1 + x2

where xi = eV/kTK,, with TK,1 and TK,2 being the temperatures at which the two stages

of the screening process occur. The best fits to the data in Figures 4-16 and 4-17 are

achieved with kBTK,1/D = 3.75 x 10-5, kBTK,2/D = 3.40 x 10-12, q1 1.0, and q2 -0.2.

(As with the single-impurity results, the NRG produces resonances slightly broader than

the fit does.)

0.4 2.5
2.84
\3.3
0.3 ---1-imp. results

0.2

0.1 1

0.01 0.0001 le-06 le-08 le-10 le-12
-e VD

Figure 4-13. STM:\ differential conductance vs. negative bias voltage for the same
two-impurity cases as shown in Figure 4-10, with tunneling into impurity 1
described by td/tc 0.1.

4.2.4 Varying Impurity Parameters

In addition to the impurity separation R, it is important to consider the variation of

the above behaviors with other impul ii ,-- ,-1. i:: parameters which influence the energy

scales T imp and IRKKY. For example, as A increases, TK increases more strongly than

IRKKY [cf. Eqs. (4-7) and (4-6)]. This behavior is illustrated in Figure 4-18, which plots

the differential conductance for a two-impurity system with A given in the legend. For

A = 0.05 and 0.1, IRKKY > T-"imp and the conductance has the form of a sum of two

Fano lineshapes, indicating a two-stage Kondo screening effect. For A = 0.15 and 0.2,

IRKKY Tk-imp and the conductance has the form of a single Fano lineshape, indicating

independent-impurity screening. (Note that varying A cannot change the sign of IRKKY-)

----- ---------
1 \\

0.5 r k = 1.5
2.0
2.5
2.84
3.3
-l m 1-imp. results

le-14 le-12 le-10 le-08 le-06 0.0001 0.01
eV/D

Figure 4-14. ST \i differential conductance vs. positive bias voltage for the same
two-impurity cases as shown in Figure 4-10, with tunneling into impurity 1
described by td/t, = 0.4.

Similar changes may be found by decreasing U, since T1-imp and IRKKY depend on

poJ 8A/TU.

It is also important to consider the effects of allowing electrons to tunnel directly

between the two impurities. For A > 0, impurity states created by dt<, and dt,, have

energies ed + A and ed A, respectively. The resulting changes to the energy spectrum of

HN lead to a modification of the RKKY interaction IRKKY --- IRKKY 4A2/U [92]. Thus,

in the cases in which the Kondo effect is observed (IRKKY > 0 or -IRKKY < T-imp), a A

of sufficiently large magnitude (for example, |AI m A in the case of U < D and R = 0 [92])

causes impuritfv--i-o,:.t formation and destroys the Kondo effect. Thus, the conductance

signatures in Figures 4-12 through 4-15 should still be observed in the general case A / 0,

1 -

0.5

0-
0.01 0.0001 le-06 le-08 le-10 le-12 le-14
-eVID

Figure 4-15. STA\ differential conductance vs. negative bias voltage for the same
two-impurity cases as shown in Figure 4-10, with tunneling into impurity 1
described by td/tc = 0.4.

with the modification that the comparison of IRKKY- 4A2/U with Tr-imp determines which

of the three types of behavior is observed.

The results in this chapter demonstrate that ST\i devices may be used to study

the competition between magnetic ordering (induced directly or indirectly) with the

Kondo effect in a simple two-impurity system. The different behaviors of two-stage Kondo

screening, independent-impurity screening, and impurity-singlet formation yield different

signatures in the ST\i conductance spectrum. Such signatures permit the experimental

observation of these behaviors that have been studied in depth with various theoretical

methods [32-37, 40-43, 45, 49, 74, 80].

1 -

0.5 E
I---------OD
0.5 EE=1\ ss n> P

0 I I I I I
le-14 le-12 le-10 le-08 le-06 0.0001 0.01

eVID

Figure 4-16. NRG results for the two-impurity ST \i conductance G vs. positive bias
voltage for the parameters shown in Figure 4-2 with kFR 1.5 and
td/te = 0.1 (squares), fitted to the sum of two Fano lineshapes [Eq. (4-13)].

1 -

0.5 kR = 1.5

O I I I I I

0.01 0.0001 le-06 le-08 le-10 le-12 le-14

-eV/D

Figure 4-17. NRG results for the two-impurity ST \i conductance G vs. negative bias
voltage for the parameters shown in Figure 4-2 with kFR 1.5 and
td/te = 0.1 (squares), fitted to the sum of two Fano lineshapes [Eq. (4-13)1.

0.3

0.2

0.1

0
le-14

Figure 4-18.

le-12 le-10 le-08 le-06 0.0001 0.01 1
eVID

STA \ differential conductance vs. positive bias voltage for a two-impurity
system with U = -2d = D, kFR 1.5, tunneling into impurity 1 described
by td/te =0.1, and values of A (in units of D) given in the legend.

CHAPTER 5
ASYMMETRIC DOUBLE QUANTUM-DOT DEVICES

As seen in Sections and 1.3.2 and 1.4.3, quantum dot (QD) devices permit the study

of a rich variety of physical behaviors thanks to the experimental control of their geometry

and energy level configuration. Due to their discrete energy levels, QD devices have rightly

been described as artificial atoms (or artificial molecules in the case of coupled QDs).

In keeping with this comparison, when a QD's energy levels are tuned such that it is

energetically favorable for the dot to have an odd number of electrons (see Section 1.3.2),

the dot interacts with the lead electrons in much the same way that a magnetic impurity

interacts with conduction electrons, thus displaying the Kondo effect [55, 64, 76]. When

multiple Kondo-regime dots are connected to each other, an artificial multiple-impurity

system is created, permitting the careful study of many behaviors that were previously

difficult to access experimentally.

As described in Section 1.4.3, most of the previous theoretical work on DQDs has

focused on devices for which both dots are in the Kondo regime. A different class of

highly .,i-viii. i ii DQD devices has begun to be explored [88, 91, 94, 98], revealing novel

properties, including a splitting of the Kondo resonance and a pair of quantum phase

transitions. In this new class of devices, one of the dots (hereafter referred to as Dot 1)

is tuned to be in the Kondo regime, while the other (hereafter referred to as Dot 2) is

effectively non-interacting and is tuned near a Coulomb blockade peak [i.e., at a maximum

in G(V) such as in Figure 1-10], such that it does not display a Kondo resonance. (Thus,

this device exhibits properties very different from the two-impurity behaviors discussed in

Section 1.4.1.)

In terms of the two-impurity Anderson model, this description means that Dot 1 will

meet the conditions for local-moment formation (-ei > A1, el + U1 > Ai), while Dot

2 must be tuned such that C2 w 0 (to be near resonance with the leads) and constructed

such that U2 = 0 (to be non-interacting and not develop a local moment). This last

condition, while difficult to achieve exactly, can be approximately met by considering a

large dot, since the Coulomb interaction U2 will vary inversely with the size of the dot. A

similar construction was developed to study the two-channel Kondo effect in a quantum

dot [95].

As we shall see below, the condition U2 = 0 also reduces the computational

rigor of studying this .,-i- i in, 1 ii: DQD numerically, permitting the use of an effective

single-impurity model [88, 94, 98]. However, as it is impossible to realize a perfectly

non-interacting dot in experiments, it is important to consider the effects of a small but

non-zero interaction on Dot 2. This chapter will focus on the limit in which U2 remains

small compared to other bare parameters of the two-impurity Anderson model, in which

case no significant local moment is expected to form on Dot 2.

In this chapter, I will first describe the application of the Anderson model to the

DQD device of interest. I will then focus on two device configurations-the "side-dot"

and p i 11, i" geometries-that exhibit physical properties of particular interest. For each

configuration, I will summarize the main features of the U2 = 0 special case, and then

demonstrate how these features are modified by a non-zero U2.

5.1 Double Quantum Dot Setup

5.1.1 Model and Simplifications

The .i-,iiiii,. I ic DQD device is represented schematically in Figure 5-1 and is

described by the Hamiltonian [88]

~tH + E^^L A,, + E t d)i()] +

k,i,a
> (Vi e."r dt a + h.c.) + A(dd2,, + dS,,dl,,). (5 1)

The left and right leads are labeled by a = L, R, respectively, and the dots are labeled by

i = 1, 2. I assume that the impurity separation R = I' r21 satisfies kFR < 1 (since kF

tends to be relatively small in semiconductor heterostructures), such that the dots may

Dot 1

Dot 2

Figure 5-1. Schematic of an .,i-viiii. I lic double quantum dot. Attention will be focused on
the side-coupled configuration V1 = 0 and the parallel configuration A = 0.
Each configuration will first be explored in the limit of U2 = 0 (Dot 2 large),
and then U2 small but non-zero will be considered.

be treated as Anderson impurities with zero separation (hence, the odd conduction band

channel may be dropped from the NRG analysis as described in Section 2.2.3).

To simplify the model, it is assumed (as in previous chapters) that Vj,~i is real and

independent of k. It is also assumed that the two leads are identical (ek, = e,R) and

couple symmetrically to the quantum dots (iL = V,R), in which case each dot hybridizes
with the operators ct and ct only in the linear combination ct = (cL +ct )/2
k,L,a k,R,a k,c k,L,a k,R,oI
[88]. In effect, Dot i couples to a single lead with a hybridization Vi = V/2V,. Thus, the

Hamiltonian in Eq. (5-1) reduces to Eq. (2-26) with R = 0. Finally, in the work to follow,

the density of states in each of the leads is taken to be p(c) = po for I|e < D and zero

otherwise. Henceforth, I take D = 1 as the unit of energy and for notational simplicity set

h= 1.

In an experimental realization of this DQD device, ce and 62 would be controlled by

tuning the voltages of plunger gates on Dot 1 and 2 (respectively), while A, V1, and V2

would be controlled by tuning the voltages on gates that define the tunneling barriers

separating the dots from one another and from the leads. In the sections that follow, I

focus on fixing U/1, U2, V2, and A, and examining the changes in the zero-temperature

linear conductance G of the DQD device as ec or C2 is varied. The linear conductance is

calculated using the Landauer formula [6, 18], given in Eq. (2-68). This formula requires

four impurity spectral functions All(w), A22(w), A12(w), and A21(w), which are calculated

using the NRG formula in Eq. (2-70). All calculations below were performed for A = 3.0,

retaining Nkeep = 500 states after each iteration.

5.1.2 Special Case: U2 = 0

Even with the above simplifications, the preceding Hamiltonian is quite difficult to

study. As a further simplification, previous work [88, 94, 98] has begun by considering the

special case of U2 = 0. It has been shown [88] that for U2 = 0, the above two-impurity

Anderson model can be mapped onto that of a single impurity (in this case, Dot 1)

described by Eq. (1-13) with a nonconstant hybridization function [88]

A(c) Qr Vd16{d(c- ik) (5-2)
k
[AA + ( ) (53)
S- )+ (A2)2](53)

where Ai rpoVi2. Because this special limit maps onto an effective one-impurity

problem, it may be studied with relative computational ease.

This nonconstant hybridization gives rise to a number of novel effects, including

zero-field splitting of the Kondo resonance and a pair of quantum phase transitions. As we

shall see, both of these effects are observable in the linear conductance of the DQD device.

These effects are most straightforwardly observed in two special cases: the side-coupled

DQD (A1 = 0) and the parallel DQD (A = 0). I shall now examine these two special

cases, beginning each by considering the limit of U2 = 0 and then expanding to discuss the

effects of small but finite U2.

5.2 Side-Coupled DQD

5.2.1 Special Case: U2 =

In the special case A1 = 0 and A > 0, Dot 1 is connected to the leads only through

Dot 2. In the U2 = 0 limit, Eq. (5-3) produces a Lorentzian hybridization function of

width A2 centered on w = 62. Figure 5-2 shows that, when 62 = 0 (i.e., Dot 2 is precisely

on resonance), for sufficiently small values of A, the Dot-1 spectral function Al1(w)

di-p,~ i' a shape similar to that seen in the case of a constant hybridization: the Hubbard

bands at w tl and el + U1, and the Kondo resonance of width 2kBTK (cf. Figure 2-2). If

A is increased, however, the Kondo resonance splits. This splitting is attributed to the fact

that the relatively low value of A(w) for I1w S kBTK causes the Kondo resonance to rise,

but the enhanced value of A(w = 0) causes All to fall as w -i 0 to satisfy the Fermi-liquid

relation An(w = 0) < 1/[7A(0)] [88]. For still higher values of A, the splitting increases,

indicating that TK grows as A is increased.

Such a splitting of the Kondo resonance is similar to that observed for a Kondo

impurity under the influence of a magnetic field, which causes a suppression of the Kondo

effect due to a destruction of the Kondo ground state at lower temperatures. In the case

at hand, however, one finds upon examining the energy spectra calculated by the NRG

procedure [88] that the Kondo ground state is preserved; therefore, the Kondo effect is

still taking place, at a TK that increases with A. A similar splitting effect has also been

predicted to occur in a side-coupled DQD device with identical impurities [85].

The zero-field splitting of the Kondo resonance is also evidenced in the linear

conductance of the side-coupled DQD device, as shown in Figure 5-3. As also seen in

Figure 1-18, the zero-temperature conductance vanishes at the particle-hole symmetric

point cl = -U//2, regardless of the value of A; this feature can be explained by considering

the fact that G(T = 0) o sin(T/22), where T722 is the phase shift of electrons scattering from

Dot 2 [98]. At the particle-hole symmetric point, T722 makes a discontinuous jump from -T

to 0 (regardless of the value of A), such that G(T = 0) = 0. Away from the particle-hole

' I I I I

X = 0.03
6 0.063
0.078

4-

2
<0

-0.4 -0.2 0 0.2 0.4
0)
Figure 5-2. Dot 1 spectral function for a side-coupled DQD device with
c1 -Ui/2 = -0.25, C2 U2= 0, A2 = 0.02, and A as labeled in the legend.
For sufficiently large A, we observe a novel splitting of the Kondo resonance
that does not destroy the Kondo ground state.

symmetric point, Dot 1 no longer has a Kondo resonance, and G approaches its maximum

value Go = 2e2/h. Increasing A causes the Kondo effect to strengthen (indicated by an
increase in TK as predicted by the zero-field splitting of the Kondo resonance), pushing the

upturns in G away from the particle-hole symmetric point [98].
5.2.2 Extended Case: U2 > 0

If we now consider U2 > 0 while holding C2 = 0, we see in Figure 5-4 that for

sufficiently small U2, there is still a visible splitting of the Kondo resonance, with a

new particle-hole ..i-mmetry introduced by the fact that the DQD no longer satisfies

2 = -U2/2. As U2 increases, this .,i-v iiii. I1 y becomes more pronounced, shifting spectral
weight from the positive-w side to the negative-w side. Once U2 becomes strong enough

0.5 /-
0 \ Ii
N /

-1 -0.5 0 0.5

Figure 5-3. Zero-temperature conductance vs. Dot 1 energy level for a side-coupled DQD
with U1 = 0.5, 62 = 2 = 0, A2 = 0.02, and A as labeled in the legend.

(U2 ~ 0.1 in the case at hand), the peak on the positive-cw side disappears, leaving a single

peak that moves toward w = 0.

The .-i-v ilii. I ry in the Dot-1 spectral function manifests itself in the conductance

traces in Figure 5-5. For U2 < 10-4 (not shown), there is no noticeable deviation from

the U2 = 0 behavior, indicating the robustness of the effective one-impurity model.

For U2 = 0.01, where the Dot-2 interaction has become comparable to other energy

scales of the device (such as U1, A1, and A2), a noticeable .-i-,ii::i. r1 y develops in G vs.

ec. Upon further increase in U2, the conductance develops a peak on the order of Go

on the right-hand side. This is comparable to the peaks seen in the identical-impurity

side-coupled device in Figure 1-18. The absence of a left-hand peak is most likely due to

the inequivalence of the two dots in this device. Note that for all cases with U2 > 0 shown

in Figure 5-5, the conductance drops all the way to zero at a single value of cl. Just as for

4 |I u |

2- -

0I

-0.2 -0.1 0 0.1 0.2

0)

Figure 5-4. Zero-temperature Dot-1 spectral function for a side-coupled DQD with
e1 = -UI/2 = -0.25, Al = 0, C2-= 0, A2 = 0.02, A = 0.067, and the values of
U2 specified in the legend. The interaction on Dot 2 introduces a particle-hole
-v i:111. I 1y that becomes more pronounced as U2 increases. For sufficiently
large U2, the right-hand peak in All(aw) disappears.

U2 = 0, the zero in G occurs at the point where the Dot-2 phase shift 7722 jumps between

-Tr and 0.

An important question to consider is whether the deviations from the U2 = 0

behavior are unique to devices with U2 > 0 or are merely the result of broken particle-hole

symmetry on Dot 2. This issue can be investigated by comparing Figure 5-5 with Figure

5-6, which plots the conductance for a side-coupled DQD with U2 = 0 and C2 / 0 (for

which Dot 2 breaks particle-hole symmetry even though it is still noninteracting). For

relatively small deviations from particle-hole symmetry (i.e., U2 and I|2| smaller than

the other device parameters), there is no qualitative difference between the device with

U2 > 0 and C2 = 0 and the device with U2 = 0 and s2 / 0. For more significant deviations

0.5 \ I

I /

0
-1 -0.5 0 0.5
L1

Figure 5-5. Zero-temperature conductance vs. Dot 1 energy level for a side-coupled DQD
with U1 = 0.5, 62 = 0, A2 = 0.02, A = 0.063, and the values of U2 specified in
the legend.

from particle-hole symmetry (i.e., when U2 and 62 are comparable to the other device

parameters), there is still qualitative agreement on some features (the peak of G C Go and

the minimum of G = 0) but the conductance for the device with U2 = 0 and E2 / 0 shows

a plateau that is not present in the device with U2 > 0 and E2 = 0. Thus, while there is

some similarity in the behavior of the two devices, they do produce noticeably different

conductance spectra, signifying that the U2 > 0 side-coupled DQD device offers a new

realm of behavior to explore.

5.3 Parallel DQD

In the special case of A = 0 and Ai / 0, the dots are not directly connected to each

other, but, as was the case in Section 4.2.1, the dots still interact with each other via the

conduction electrons. The primary feature of this special case is a pair of quantum phase

II / I f

0.5 r -

\ I /

\I I / 1
-------_ _

-1 -0.5 0 0.5
L1

Figure 5-6. Zero-temperature conductance vs. Dot 1 energy level for a side-coupled DQD
with U1 = 0.5, U2 = 0, A2 = 0.02, A = 0.063, and the values of 62 specified in
the legend.

transitions (QPTs) that can be observed by tuning ei and C2. As in the side-coupled case,
first the behavior of the dots in the U2 = 0 limit will be reviewed, and then the general

case of U2 > 0 will be explored.
5.3.1 Special Case: U2 = 0

In the special limit of U2 = 0, the effective hybridization for Dot 1 [Eq. (5-3)]

vanishes at c = C2 with a power-law A(c) oc (c 22)2. Thus, when Dot 2 is in resonance

with the leads (C2 = 0), the hybridization vanishes at the Fermi energy, C = 0. The
presence of such a power-law vanishing of the hybridization (or pseudogap) in the

Anderson impurity model is known to introduce a pair of quantum phase transitions

between Kondo-screened phases and a local-moment phase in which the impurity degree of
freedom remains unquenched at absolute zero [47, 48, 53, 56, 67, 81]. In the DQD device

being considered, the finite-temperature manifestations of these QPTs may be studied
experimentally by tuning e1 for fixed values of U1, A1, 62, and A2. (Later, it will be shown
that these QPTs can be observed by tuning 62, as well.) The QPTs are summarized in the
phase diagram in Figure 5-7.

0- El
MV TXimp 4 0
K
E+lc- - TXimp 4 1/6

-U/2 __ LM TXimp 1/4

1c- -- TXimp 1/6
K
MV TXimp 0
FI
Figure 5-7. Schematic zero-temperature phase diagram for the U2 = 0 parallel-dot system
as a function of ei.

If one begins with a large, positive 1e and moves through e1 = 0 to negative values,
the behavior of the DQD device exhibits an evolution from an empty-impurity regime (EI)
to a mixed-valence regime (MV) and then to a Kondo regime (K). Each of these regimes
(together termed the -I i i-coupling regime") possesses the property that the impurity
magnetic susceptibility Ximp obeys TXimp -i 0 as T -+ 0. [However, in contrast to the
conventional Anderson model, for which Ximp(T = 0) generally remains finite-approaching
0.103/TK in the Kondo limit [30]-the power-law form of the effective hybridization results
in a strong-coupling value Ximp(T = 0) = 0 [56]. This aspect of the strong-coupling regime
will be important below when U2 > 0 is considered.] As can be seen in Figures 5-8 and
5-9, the Kondo temperature TK defined via the condition TKXimp(TK) = 0.0701 is linearly

proportional to JC t Ce, where e = e1 + denotes the location of the quantum critical

point, at which there is no Kondo effect and TXimp 1/6 all the way to T = 0.

0.25 -- /

0.2 -\ *
\ \ + -2
-- 6e1 = 10
-- 10-4
0.15- 1
/ 11)-6

le-06
T

Figure 5-8. TXimp vs. T for a
e calculated for
values of +c = e1

U2 = 0 parallel-coupled DQD near its upper QPT at ec
U1 0.5, A1 0.05, c2 0, A2 0.02, A 0, and the
- eC specified in the legend.

If e1 is decreased below e+, the device enters a local-moment regime (LM) in which

TXimp approaches 1/4 as T 0, indicating the presence of an unscreened local moment

on Dot 1. In this LM regime, we can define a crossover temperature T* (analogous to TK

in the strong-coupling regime) via the (somewhat arbitrary) criterion T*Ximp(T*) = 1/5.

In the vicinity of the QPT, it is found that T* oc 6e+. (The same linearity is observed for

other choices of the crossover value of TXimp.)

Once eC is decreased past the particle-hole symmetric point eC = -UL/2, however,

T* begins to decrease, until it becomes zero at a second critical point, e1 = Ce. Similar

to the behavior in the upper half of the local-moment regime, in the vicinity of the QPT,

T* varies linearly with 6c, = e1 Ec. If e1 is decreased past this second critical point,

0.01 -

0.0001 -

le-06

le-08 -

le-10 I I I I I I I
le-08 le-07 le-06 le-05 0.0001 0.001 0.01 0.1 1
I1 + I

Figure 5-9. Kondo temperature TK vs. J+ -- e1 e+ for a U2 = 0 parallel-coupled DQD.
Other dot parameters are U1 = 0.5, A1 = 0.05, 62 = 0, A2 = 0.02, A = 0.

Kondo screening again takes place, and the device enters a second Kondo regime (K).

Further decrease of e1 will bring Dot 1 into a mixed valence regime (MV) and finally a

full-impurity regime (FI), all of which possess the property that TXimp -i 0 as T -+ 0.

This lower strong-coupling regime is similar to the upper one, with TK oc b6.

The QPT at ci = e which was not discussed in [88], is related to the QPT at

ci = e+ by particle-hole inversion, as evidenced by the fact that te + U1/21 = ce + UI/21

when 2 = -U2/2. (This rule applies to the U2 > 0 case, as well.) Because these two

quantum critical points are related via such a simple symmetry, they exhibit many of the

same features. Thus, I shall focus my discussion on the behavior around Cet.

Figures 5-10 through 5-13 show the spectral function An(aw) of Dot 1 for various

values of e1 in the upper strong-coupling regime and the upper half of the local-moment

regime. In the strong-coupling regime (Figures 5-10 and 5-11), the main feature of the

le+08

le+06- 1
S f -- 10-4
1 \ .- 10o
S*I */ -6
10000- \ 10
S. 10

100

1
I \ \

0.01 *
/ /

0.0001
le-10 le-08 le-06 0.0001 0.01
0)

Figure 5-10. Zero-temperature Dot 1 spectral function All(w) vs. U > 0 for a
parallel-coupled DQD with U1 = 0.5, A1 = 0.05, 2 U2 = 0, A2 = 0.02, A = 0,
and various values of e1 located in the strong-coupling regime, specified in the
legend through 6et = 1l Ce.

spectral function is a resonance or quasiparticle peak centered at U -_ kBTK. This peak,

which is the pseudogap generalization of the Kondo resonance, has a width proportional to

TK, a height inversely proportional to TK, and therefore an integrated area that remains

constant as TK vanishes at the QPT. For wIu < TK, A11(w) vanishes as U2. This power-law

behavior is attributed to the pseudogap nature of the hybridization, which also vanishes

as w2. There are similar features in A11 in the local-moment regime (Figures 5-12 and

5-13), except that now the resonance occurs on the negative frequency side, centered at

a w -kBT*.

5.3.2 Extended Case: U2 > 0 Phase Diagram and Susceptibility

I now examine the effects of allowing U2 > 0 in the parallel DQD device, focusing on

the limit of small U2 (meaning 0 < U2 < U1, A, A2) and (initially) keeping e2 = 0. This

le+08

le+06 IU
10.4
--- 10 5

-o8
.~~ """ 10-s
S 100 -

0.01 -

0.0001 II I
0.1 0.01 0.001 0.0001 le-05 le-06 le-07 le-08 le-09 le-10
CIl
Figure 5-11. Negative-frequency spectral functions corresponding to the data shown in
Figure 5-10.

more general system contains a richer phase diagram, given the larger parameter space to

explore. We will begin by considering the schematic phase diagram in Figure 5-14, which

focuses on varying e1 and U2.

The first effect of non-zero U2 that should be noted is a change in the locations ct

and e of the critical points. Figure 5-14 shows that the critical points (at which TK -+ 0)

shift toward each other as U2 increases. The special property le + UI/21 = |ec + UI/21

does not apply, due to the fact that the DQD device breaks the particle-hole symmetry

condition 2a -(2/2.

For U2 U1, A, A2, the phase boundaries follow the form

Ec = c(U2 = 0) + A'(Ui,l 2,A2)U2 (5-4)

where A+ (U A1, 62, A2) is negative (positive). This relationship is demonstrated in

Figures 5-15 and 5-16, which show the relative change in ct with increasing U2. These

le+08

- oe1
"n3
18E
- -10-2
S I 1 0-3

-10-1

-I
le+06 --
-10-4
-- -10-4
10000 -10-6

t 100-

0.01 / /

0.0001 I I I I I
le-10 le-09 le-08 le-07 le-06 le-05 0.0001 0.001 0.01 0.1

Figure 5-12. Zero-temperature Dot 1 spectral function Anl(w) vs. w > 0 for a
parallel-coupled DQD with U1 = 0.5, A1 = 0.05, 2 U2 = 0, A2 = 0.02, A = 0,
and various values of e1 located in the local-moment regime, specified in the
legend through 6e+ = e1 c.

plots show that A+ is proportional to (A2)-1 but only weakly dependent on A1. Such

behavior is also found for A-, confirming the similarity of the two QPTs.

As observed in Figure 5-14, when U2 becomes comparable to the other energy scales

(U1, A1, and A2), the critical points approach each other more quickly than seen in the

linear behavior for smaller U2. At a certain value of U2 (approximately 0.046, for the

case illustrated in Figure 5-14), the critical points merge, and the local moment regime

disappears as U2 is increased further. In the discussion to follow, I will first focus on values

of U2 that are small enough to preserve the local-moment regime, and then discuss the

behavior of the system when the local-moment regime has disappeared.

Considerable insight can be gained into the behavior of the U2 > 0 parallel DQD

device near its quantum critical points by examining the temperature variation of TXimp.

Figure 5-17 shows TXimp vs. T for a sequence of e1 values and a fixed U2 = 10-3. (This

le+08

le+06 -
-- -10-4
/ V -10

t o\ -6
10000- .

S -oo \ / \ -

S. ...\.

0.01 \ -
\ *

IIl
0.00011
0.1 0.01 0.001 0.0001 le-05 le-06 le-07 le-08 le-09 le-10

Figure 5-13. Negative-frequency spectral functions corresponding to the data shown in
Figure 5-12.

value of U2 has been chosen since it gives rise to very well-separated energy scales that will

provide insight into behavior for larger U2.)

First, as ec approaches e+ from the upper strong-coupling regime, TXimp remains

near 1/6 (as in the U2 0 case) until Dot 1 becomes Kondo-screened and TXimp falls to

zero at TK, with TK oc Jc as before. As in the case of U2 0, imp --- 0 as T --- 0.

Around a crossover value e+ > et, a region of new behavior unfolds. This new region is

characterized by a rise in TXimp from 1/6 toward 1/4. However, as long as e1 > e+, TXimp

never reaches 1/4, and is eventually screened down to zero.

Another new development in this region is that Ximp -- 0.103/TK (instead of 0)

as T -+ 0, mimicking the conventional single-impurity Kondo effect [30]. This shift in

Ximp(T = 0) is shown in Figure 5-18 and is also evidenced by how the red traces in Figure

5-17 fall to zero less steeply than the black traces. It is important to note that the change

in Ximp(T = 0), while very rapid, is a continuous evolution as ce passes below ez+. Thus,

-0.3

-0.4

-0.5

0

Figure 5-14.

0.01 0.02 0.03 0.04 0.05
U2

Approximate phase diagram on the U2-1l plane for a parallel-coupled DQD
device with Ui = 0.5, A1 = 0.05, C2 = 0, and A2 = 0.02. Solid lines indicate
the quantum phase transitions separating the strong-coupling and
local-moment phases. Note the shift of the critical energies e, and e- as U2
increases, resulting in the disappearance of the local-moment regime for
U2 > 0.046. Dashed lines (labeled e+ and c -) represent crossovers from
regions of behavior essentially identical to that for U2 0 (ei > c1+ and
CI < C ) to regions of novel behavior located between the solid and dashed
lines. Arrows indicate values of U2 for which TK and G are plotted against eC
in later figures.

e+ is to be considered a crossover, and not a new critical point. (There is some degree of

arbitrariness in the definition of C+.) Figure 5-19 shows that this new region c+ > eC > e

exhibits a very rapid drop of TK with 6c+, departing from the universal behavior of the

U2 = 0 limit.

Introduction of an interaction on Dot 2 not only changes the behavior in the

strong-coupling phase near the transition, but also changes the behavior at the transition

m y I I I I I I I

0 A = 10-4 A2 10-4
1 10-4, 10-3
-4 -2
10 10
SA 10-4, 10-1
0.01 o 10-^1-4 10
-2 -4
10, 10-4
I I
S0.0001 -
FA

^ le-06
A

le-08 -

le-10 le-09 le-08 le-07 le-06 le-05 0.0001 0.001 0.01

U2

Figure 5-15. Critical value ct vs. U2 for a parallel-coupled DQD with Ui = 0.5, 62 = 0,
A 0, and A1 and A2 as specified in the legend. Note the linear variation of
ect with U2, and the weak dependence on Ai.

CI = e+. As seen in Figure 5-17, instead of remaining at 1/6 down to zero temperature (as

it did in the U2 = 0 case), TXimp rises up to 1/4 and remains there as T -+ 0.

As ci is lowered below ct, Dot 1 exhibits local-moment behavior in which TXimp

1/4 down to zero temperature. In contrast to the case U2 = 0, for U2 > 0, it is not

possible to define a crossover scale T* that vanishes continuously as c --- t. Because

of this lack of scale, the phase transition for U2 > 0 is identified as a Kosterlitz-Thouless

type transition [22]. This identification is supported by the fact that there is such a

strong drop in TK on the strong-coupling side of the transition, but no such feature

on the local-moment side. Such a highly ..i-i'i. I1 ic divergence is characteristic of a

Kosterlitz-Thouless type transition.

I I I

0
O

-1 -

S-2 A
II O
+ o O
c- -3

to
--4
A

0 0 A2=0.1
A O 0.02
O A 0.01

-6 0 I I I I I
-6 -5 -4 -3 -2 -1 0
log(U2/A2)

Figure 5-16. Critical value eI vs. U2/A2 for a parallel-coupled DQD device with
U1i 0.5, 2 = 0, A = 0, and Ai = 0.05.

Approaching c. from the local-moment phase, there is still no characteristic

temperature scale at which TXimp 1/4, indicating that the lower quantum phase

transition is also a Kosterlitz-Thouless type. For c. > e1 > c there is again a region

of new behavior in which Timpp initially rises with decreasing T from 1/6 towards 1/4,

but is eventually Kondo-screened to TXimp = 0. This screening sets in around a Kondo

temperature TK that drops to zero very rapidly as ec -i -c from below and exhibits the

conventional Kondo behavior of Xinp(T = 0) = 0.103/TK. For ec < l the universal

U2 = 0 behavior is recovered, with TK oc 6be = c, e1 and Ximp(T = 0) = 0. (As with

crossing el+, this change is continuous, and so c- is to be regarded as a crossover and not a

critical point.)

As illustrated in Figure 5-14, at a certain value of U2 (approximately 0.046, in the

case at hand), the two critical points merge and the local-moment phase disappears.

I I

' I

0.25

0.2

0.15

0.1 + -

0.05

0
le-20 le-18 le-16 le-14 le-12 le-10 le-08 le-06 0.0001 0.01
T

Figure 5-17. TXimp vs. T for a parallel-coupled DQD with U2 10-3. Other dot
parameters are Ui 0.5, A1 = 0.05, A2 0.02, e2 0, and A 0 Each trace
represents a different value of ei near et. The thick blue line corresponds to
the critical level energy et, and the dashed curve corresponds to the
crossover (corresponding to the upper dashed line in Figure 5-14) from the
U2 0 pseudogap behavior (black traces in the region labeled "ei > cl") to
novel behavior (red traces in the region labeled "e+ < eC < eK").

This disappearance is demonstrated in Figure 5-20, which shows that TK has a non-zero

minimum that increases with U2. (The minimum value of TK for U2 = 0.05 is on the

order of 10-90, and not visible in the plot.) It is noted that the width of the dip in TK

vs. el remains roughly constant with increasing U2 beyond the disappearance of the

local-moment regime. Indeed, Figure 5-14 shows that the crossover energies e, defined as

the locations of the downturns in TK vs. c- seem to become independent of U2 for large

U2. Thus, even though the quantum phase transitions have disappeared for sufficiently

large U2, there remains a signature of their proximity in the dip in TK vs c1. It should

0.1k0 0
0.1

0.05

0L
-8

Figure 5-18.

-7.98 -7.96 -7.94 -7.92
log(Ei E1 )

TKXimp(T = 0)/ vs. b~ = e ej for a parallel-coupled DQD device with
U1 0.5, A1 0.05, U2 10-6, A2 0.02, and C2 A 0. The rapid upturn
in Ximp(T 0 ) at c1 e-c (around 6e+ = -7.96 on this graph) signifies the
departure from the U2 0 universal behavior. (Note that the value of c+ in
this graph does not correspond to its value in Figure 5-19 as a greater
number of energy eigenstates were kept at the end of each NRG iteration to
obtain these results.)

therefore be feasible to probe the quantum phase transitions in an experiment, even when

U2 w 0 is not satisfied.

5.3.3 Extended Case: U2 > 0 Spectral Function and Conductance

The evolution of the DQD device's behavior with increasing U2 exhibits signatures in

the dot spectral functions and linear conductance, as well. Figure 5-21 shows the spectral

function All(w) for a parallel DQD device with U2 = 10-3. (As in Figure 5-17, this value

of U2 was chosen because it creates well-separated energy scales.) For ec > e+ (seen in the

black traces), the main feature of the spectral function is a quasiparticle peak centered at

uw kBTK, just as it was in the case of the U2 = 0 device. However, in the U2 > 0 device,

0 0 0O 0O
0

0

0

SI a, I a

0.01 05
10-4
10.4
0.001
0.01
0.0001- 0.02
0.03
K 0.04
0.045
le-06 -

le-08 -

le-10 I I
le-lO ..-- --- 1 -- --- J -- L .....
le-08 le-07 le-06 le-05 0.0001 0.001
e1 e, |

Figure 5-19. Kondo temperature TK vs. Je = e1 ec for U =
E2 A 0 For U2 0, TK decreases linearly with
departure from linearity, which sets in around c =
novel U2 > 0 behavior.

0.5, A1 0.05, A2 0.02,
6c+. The downward
e+, signals the onset of

All(a = 0) has a non-zero value, whereas All(au 0) = 0 in the U2 = 0 device. As ec

crosses e+ from above, All(aw 0) rises to a height comparable to the quasiparticle peak

(as in the solid blue trace). For e1 between et and e+ (seen in the red and green traces),

the quasiparticle peak is absorbed into the resonance centered around U = 0, and the

Kondo temperature TK may be determined from the width of the zero-frequency resonance

(similar to how it may be identified in the conventional Anderson model). Qualitatively

similar behavior-a quasiparticle peak for e1 > c+ which is absorbed into the U = 0

resonance for el < ec < e -is observed in the spectral function A22(u) (not shown).

It will now be seen that the .,i-,iiiii, I ic nature of the quantum critical points for

U2 > 0 will have interesting effects on the conductance. Even when U2 is sufficiently large

to remove the local-moment phase, signatures of the critical points are visible. Figure 5-22

0.01

0.0001 --

\\ /
le-06 \ /

\ \ /
le-08 \ /
\^ /

\ U2 = 0.045
le-10 \ -- 0.05
f- ---- 0.1
0.5
0.75
le-12
-0.5 -0.4 -0.3 -0.2 -0.1 0

E1

Figure 5-20. Kondo temperature TK vs. ec for a parallel DQD with U2 as specified in the
legend. After the critical points eIC meet at U2 w 0.046, the local-moment
regime disappears, and TK reaches a non-zero minimum. Other device
parameters are Ui = 0.5, Ai = 0.05, C2 = 0, A2 = 0.02, and A = 0.

shows the zero-temperature conductance G (measured in units of Go = 2e2/h) as e1 is

swept for various values of U2 that exhibit the quantum critical points et. The general

effect of increasing U2 is to suppress the conductance, as seen in a similar theoretical

study of a Kondo-like dot connected to a large grain [91]. When e1 passes e there is

a dip in G. At each critical point, there is a discontinuous jump in G. The thinning of

the local-moment regime with increasing U2 can be seen by the narrowing of the middle

plateau of each trace.

The behavior near et and e1t can be more clearly understood from Figure 5-23, which

plots G/Go vs. 16ct = 1 et I for ei in the upper strong-coupling and local-moment

phases. (There is similar behavior for el near c1-, which is omitted for simplicity.) By

10000

S. '
s ..._....................... + '':''".
S- 5eI=0.01 "" '.. I 0
-- 0.002
100 0.00092
0.00082
0.00072
-- 0.00067
S* 0.00062
0.00057 I
-- 0.00052 -------------------------

11
le-30 le-25 le-20 le-15 le-10 le-05
(0

Figure 5-21. Zero-temperature Dot 1 spectral function All(w) vs. w > 0 for a
parallel-coupled DQD with U1 0.5, A1 0.05, 62 0, U2 10-3
A2 0.02, A 0, and various values of ec located in the strong-coupling
regime, specified in the legend through &+ = c1 ec.

comparing the traces marked by e1 > ec in Figure 5-23 with their corresponding traces

of TK in Figure 5-19, it is seen that the onset of the drop in TK corresponds to the dip in

G/Go. Note that a decrease in TK leads to a suppression of the conductance.

The discontinuous conductance jumps begin to close as U2 is increased (leading to

their disappearance when the local-moment regime vanishes). The limiting values on

opposite sides of the conductance jumps exhibit an interesting property; defining the left-

and right-hand limit of G at each critical point as

GL lim G(ci T) (5-5)

GR lim G(ci + T), (5-6)
rwo+

C

0.5

0

Figure 5-22.

-0.5 -0.4 -0.3 -0.2 -0.1 0
E1

Zero-temperature conductance G/Go vs. ct for a parallel-coupled DQD device
with U2 as specified in the legend. At each critical point, there is a
discontinuous jump in G. The limiting values of each discontinuity obey
GL + GR Go [definitions in Eqs. (5-5) and (5-6)]. When ei passes c:,
there is a dip in G. Other device parameters are U = 0.5, A1 = 0.05, 62 = 0,
A2 0.02, and A = 0.

respectively, it is found that GLI + GR = Go. This property-perhaps the result of some

remaining symmetry across each critical point-will be seen later when e2 is varied for

fixed e1 and U2 > 0.

These results show that it is possible to identify the critical points el and the new

boundaries et from the conductance in a DQD device that does not satisfy U2 = 0.

Even when U2 is large enough to close the local-moment regime entirely, the quantum

phase transitions leave a signature, as seen previously in the non-zero minimum of TK

(see Figure 5-20). A signature is also found in the zero-temperature conductance, plotted

in Figure 5-24. When U2 is large enough to close the local-moment regime, there is still

--= U.UI, E E
--- 0.01, l < 1 I I
0.5 -- 0.02, P1 > c1 I I -
SI I
0.02, P1 < 1 + I
-- 0.03, Pe > Fs I/
f 0.03, P1 < +- -"
0.03,< - --

0.001 0.01 0.1
P1 +le
Figure 5-23. Zero-temperature conductance G/Go vs. 16e+l 1 e'+ for a parallel DQD
device with U2 as specified in the legend. Traces marked by e1 > ec are in
the strong-coupling regime and may be compared with traces in Figure 5-19.
Other device parameters are as in Figure 5-22.

a dip (peak) at el = ) with a plateau region corresponding to the dip in TK vs. 1i.

(The rough region around e1 = -0.35 in the U2 = 0.05 trace is the result of a numerical

instability associated with being near a Kosterlitz-Thouless transition.) It is interesting

to note that the shapes of the conductance traces for U2 > A1, A2 resemble those for the

side-coupled case with U2 > A2, A in Figure 5-5.

Finally, as mentioned earlier, it is also possible to observe these quantum phase

transitions by varying c2. While a complete phase diagram of C1, c2, and U2 is beyond the

scope of the present study, it is insightful to examine a couple of specific cases.

Figure 5-25 shows the zero-temperature conductance as c2 is swept in a parallel DQD

device with a particle-hole symmetric Dot 1 (c1 = -UL/2). Because of this particle-hole

symmetry on Dot 1, the traces in Figure 5-25 are symmetric about c2 -(2/2. For

0.5 / -

0 I
//

-0.5 -0.4 -0.3 -0.2 -0.1 0
e1

Figure 5-24. Zero-temperature conductance G/Go vs. et for a parallel DQD device with U2
as specified in the legend. The traces may be compared with those in Figure
5-20. Other device parameters are U/ = 0.5, A =- 0.05, C2 = 0, A2 = 0.02, and
A = 0. The rough feature in the solid red trace is due to a numerical
instability associated with the Kosterlitz-Thouless transition.

U2 = 0, the device is alhv-l in the strong-coupling phase except at C2 = 0, where there is

an infinitesimally thin local-moment phase and a discontinuous point in the conductance

trace. When U2 is positive, however, the local-moment phase acquires a non-zero width,

and there is a pair of discontinuous jumps in G, signifying the presence of two quantum

critical points e It is important to note that the left- and right-hand limiting values of

G again add to Go, confirming that the same quantum phase transitions are observed by

varying C2 as were observed when el was varied.

Similar behavior is also found when Dot 1 is particle-hole .i-vmmetric. Figure 5-26

shows G vs. C2 with e1 et- for each value of U2. There is a very pronounced ..-i-i.i.. ii, ry

in these results, but the same general features apply as in the particle-hole symmetric

case: For U2 = 0, there is an infinitesimally thin local-moment regime, marked by a

discontinuous jump in G; for U2 > 0, there is a pair of quantum critical points Ce, each

marked by a discontinuous jump in G whose left- and right-hand limiting values add to

Go.

1 -
U2 = 0
SU2=O
0.01

0.5

-0.04 -0.02 0 0.02 0.04

Figure 5-25.

Zero-temperature conductance G/GO vs. 62 for a parallel DQD device with
c1i -U1/2 -0.25, A1 0.05, A2 0.02, A 0, and U2 as specified in the
legend. For U2 > 0, there is a pair of critical points e each causing a
discontinuous jump in G. The limiting values of each discontinuity obey
GL + GR Go, as in the case of G vs. e1. Note that the traces are
symmetric about C2 -U2/2 due to the particle-hole symmetry of Dot 1.

The results seen in this chapter indicate that these highly .,-viiiiii, i ic DQD devices

exhibit a rich variety of interesting physics. In the U2 = 0 special limit, the side-coupled

DQD configuration exhibits a splitting of the Kondo resonance and the parallel DQD

configuration exhibits a pair of quantum phase transitions. We have seen that, for

sufficiently small U2, these behaviors persist, although their qualitative features are

0.5

0

Figure 5-26.

-0.04 -0.02 0 0.02 0.04
C 2

Zero-temperature conductance G/GO vs. 62 for a parallel DQD device with U2
as specified in the legend and ei near ct for each value of U2. For U2 > 0,
there is again a pair of critical points e each causing a discontinuous jump
in G. The limiting values of each discontinuity obey GCL + GR Go, as in
the case of G vs. el. Other device parameters are as in Figure 5-25.

modified. It should be possible to realize these behaviors experimentally by varying the

dot resonant levels by tuning the appropriate gate voltages. For sufficiently large U2, these

behaviors disappear. Thus, such experiments can also observe the evolution of the DQD

device from the U2 = 0 limit to a device with two local-moment dots.

CHAPTER 6
CONCLUSIONS

It is clear that systems of multiple magnetic impurities and their analogous

manifestations in QD systems are not understood exhaustively. While much theoretical

work has been performed and novel experiments have been designed to probe two-impurity

systems, there are still many more possibilities to consider. In this work, I have demonstrated

that a number of interesting behaviors-such as the competition between magnetic

ordering and the Kondo effect, a splitting of the Kondo resonance, and a pair of quantum

phase transitions-should be experimentally observable in ST\ [ conductance studies of a

pair of magnetic adatoms, and in electrical transport through highly .i-iii,, ii. iiH DQDs.

6.1 Scanning Tunneling Microscopy Studies

In ST:\i studies, the competition between magnetic ordering and the Kondo effect

is manifested in the differential conductance, displaying a variety of behaviors as the

impurity separation is varied. For a sufficiently large ferromagnetic RKKY interaction,

the two impurities form a net spin singlet that experiences a two-stage Kondo effect,

producing a conductance spectrum given by a sum of two Fano lineshapes. For a weaker

RKKY interaction, the two impurities do not become aligned, and are Kondo-screened

independently, producing a conductance spectrum given by a single Fano lineshape,

qualitatively matching the conductance spectrum of a single impurity. For a sufficiently

strong antiferromagnetic RKKY interaction, the Kondo effect is suppressed, leading

to a disappearance of the Kondo resonance and a featureless conductance spectrum.

While these results were obtained for a two-impurity system with several simplifications,

qualitatively similar behaviors should be found in more general cases.

For example, the results of Section 4.2.3 considered the limiting case of an ST\ i

tip positioned directly over one of the impurities. To consider an STM study performed

with the tip at a variable position, it should be possible to utilize an extension of the

method developed by Borda [96] to study spatial correlations in the single-impurity

Kondo problem: This method involves setting up an NRG calculation of a two-impurity

system (cf. Section 2.2) with the second impurity site "empty" (i.e., C2, U2,2 = 0),

permitting the computation of properties at any distance from the impurity of study (by

varying the impurity separation R), thereby enabling a theoretical prediction of the Kondo

screening length 1K. The extension of this method to a two-impurity ST\ i study with a

generalized tip position would thus involve setting up a three-impurity calculation with

the third site "empty" and calculating the matrix elements of the impurity-3 operator

dt at successive iterations. Given the great complexity of three-impurity models [61, 84],

such NRG calculations would benefit from the parallel-processing techniques described in

C'! lpter 3. Studies of single impurities-such as depicted in Figures 1-8 and 1-9-indicate

that the Fano lineshape in the conductance only changes quantitatively as the ST \i tip is

moved. In contrast, in a generalized two-impurity STM study, it is necessary to consider

three tunneling paths, which may give rise to novel interference effects, producing a more

complicated version of the Fano lineshape.

In C'! lpter 4, the impurities under consideration also exhibited no dissipative

effects, as would be caused by the presence of phonons or noise which would introduce

decoherence to the electron tunneling paths, thereby modifying the Fano lineshape. These

scenarios may be modeled by coupling the impurities to a bosonic bath. Such coupling

has been studied in single-impurity systems using the NRG [50, 78, 97], and is known

to produce a number of novel effects, including a quantum phase transition accessible

by tuning the coupling between the impurity and the bosonic bath. The study of a

two-impurity system coupled to a bosonic bath has not been approached with the NRG

due to the computational complexity; thus, such NRG studies would benefit from the

parallelization techniques described in Chapter 3.

6.2 Double Quantum Dot Studies

Systems of DQDs offer a tunable two-impurity system, sparking a great deal of

theoretical interest. To take advantage of this tunability, a class of highly- -i- i ii, ii ic

DQD devices has been studied in which Dot 1 possesses a well-defined local magnetic

moment and Dot 2 is near-resonance with the leads (e2 w 0) and is essentially non-interacting

(U/2 0). It has been shown [88, 94, 98] that, in the special limit of U2 = 0, such a device

may be modeled as a single Anderson impurity with a nonconstant hybridization to the

conduction band. In the side-coupled DQD configuration (A1 = 0), the nonconstant

hybridization produces a zero-field splitting of the Kondo resonance in which the

Kondo effect is preserved. In the parallel DQD configuration (A = 0), the nonconstant

hybridization gives rise to a pair of continuous QPTs separating strong-coupling and

local-moment phases. These QPTs may be observed experimentally in the device

conductance G by tuning the gate voltages that control el and e2 [98].

Even though the special condition of U2 = 0 may be impossible to achieve

experimentally, the results of C'! lpter 5 indicate that the resonance splitting and quantum

phase transitions may be observed when U2 is non-zero but sufficiently small, with

intriguing modifications to these behaviors. In the side-coupled DQD configuration, a U2

that is non-zero but sufficiently smaller than the other device parameters produces merely

a slight ..i-mmetry in the splitting of the Kondo resonance and in the variation of G with

el. As U2 becomes larger than the other device parameters, qualitatively new behavior

develops, including a peak in G vs. el.

In the parallel DQD configuration, as U2 increases from zero, the locations ect of the

QPTs shift toward each other and develop an ..i-mmetry indicative of a Kosterlitz-Thouless

type transition. Also, regions of new behavior unfold in the strong-coupling phases,

characterized by a dramatic drop in the Kondo temperature TK as the QPTs are

approached. The critical points e c are manifested in the zero-temperature conductance

G by discontinuous jumps that obey an intriguing sum rule GL + GR = Go, where

Go = 2e2/h is the quantum-mechanical maximum value of G. These same QPTs may be

observed by varying e2, giving rise to critical points e:f which also exhibit discontinuous

jumps in G that obey the sum rule. Finally, for sufficiently large U2, the QPTs merge and

the local-moment phase disappears entirely, producing a continuous G vs. e1.

Further study of the parallel DQD configuration is called for, including the development

of phase diagrams that illustrates the effects of changes in U1, A1, and A2, which were left

fixed in ('!i lpter 5. It will be useful to see the effects that such changes produce on the

behavior of the critical points et and ec, and also the effects that they produce on the

value of U2 for which the local-moment phase disappears.

Further work is also needed to understand the physics behind the sum rule exhibited

by the discontinuities in the conductance at the critical points. To develop this unil, l i, lii .

it will be necessary to examine other device properties, such as the phase shifts Trij

(i,j = 1,2) of electrons that tunnel through the DQD device. It will also be necessary to

understand the conditions under which this sum rule is ob, i, '1

To make predictions that are experimentally relevant, it will also be necessary to

explore the effects of non-zero temperature on the conductance signatures reported for

the two DQD configurations in Chapter 5. Calculating the spectral functions in the

Landauer formula at finite temperature is not a straightforward task, but can be handled

to reasonable accuracy [44, 93]. In general, it has been found that the conductance

spectrum remains qualitatively the same for 0 < T < TK and is significantly different

for T > TK [85, 98]. In the parallel configuration, it will be particularly interesting to see

the effects of finite temperature on the discontinuities observed in the zero-temperature

conductance G.

6.3 Epilogue

Even after years of progress, the study of magnetic impurity systems continues

to yield newer levels of insight and application. With the ongoing development of

theoretical methods such as the numerical renormalization group and parallel computation

techniques, and the continuing progression of experimental innovations such as scanning

tunneling microscopy and quantum dot devices, we will continue to push the boundaries of

our understanding of these systems and develop novel applications. Truly, the only limit is

our imagination.

REFERENCES

[1] W. J. de Haas, J. de Boer, and G. J. van Den Berg, Physica 1, 1115 (1933).

[2] C. Zener, Phys. Rev. 81, 440 (1951).

[3] A. H. Wilson, The Theory of Metals (Cambridge University Press, Cambridge,
England, 1932).

[4] M. A. Ruderman and C. Kittel, Phys. Rev. 96, 99 (1954).

[5] K. Yosida, Phys. Rev. 106, 5 (1957).
[6] R. Landauer, IBM J. Res. Develop. 1, 233 (1957).

[7] P. W. Anderson, Phys. Rev. 124, 41 (1961).
[8] U. Fano, Phys. Rev. 124, 1866 (1961).

[9] M. Sarachik, E. Corenzwit, and L. D. Longinotti, Phys, Rev. A 135, 1041 (1964).
[10] J. Kondo, Prog. Theor. Phys. 32, 1 (1964).

[11] A. A. Abrikosov, Physics 2, 5 (1965).

[12] J. R. Schrieffer and P. A. Wolff, Phys. Rev. 149, 491 (1966).

[13] K. Yosida, Phys. Rev. 147, 223 (1966).

[14] H. Ishii and K. Yosida, Prog. Theor. Phys. 38, 61 (1967).

[15] C. N. Yang, Prog. Theor. Phys. 55, 67 (1967).

[16] P. W. Anderson, Phys. Rev. 164, 352 (1967).

[17] P. W. Anderson and G. Yuval, Phys. Rev. Lett. 23, 89 (1969).

[18] R. Landauer, Phil. Mag. 21, 863 (1970).

[19] P. W. Anderson, G. Yuval, and D. R. Hamann, Phys. Rev. B 1, 4464 (1970).

[20] G. Yuval and P. W. Anderson, Phys. Rev. B 1, 1522 (1970).

[21] P. W. Anderson, J. Phys. C 3, 2439 (1970).

[22] J. M. Kosterlitz and D. J. Thouless, J. Phys. C 6, 1181 (1973).

[23] K. G. Wilson, Rev. Mod. Phys. 47, 773 (1975).

[24] N. W. Ashcroft and N. D. Mermin, Solid State Ph;-,. (Saunders College,

[25] J. H. Jefferson, J. Phys. C 10, 3589 (1977).

[26] F. D. M. Haldane, Phys. Rev. Lett. 40, 416 (1978).

[27] P. Nozi6res and A. Blandin, J. Phys. (Paris) 41, 193 (1980).

[28] N. Andrei, Phys. Rev. Lett. 45, 379 (1980).

[29] P. B. Wiegmann, Sov. Phys. JETP Lett. 31, 392 (1980).

[30] H. R. Krishna-murthy, J. W. Wilkins, and K. G. Wilson, Phys. Rev. B 21, 1003
(1980).

[31] H. R. Krishna-murthy, J. W. Wilkins, and K. G. Wilson, Phys. Rev. B 21, 1044
(1980).

[32] C. Jayaprakash, H. R. Krishna-murthy, and J. W. Wilkins, Phys. Rev. Lett. 47, 737
(1981).

[33] J. E. Hirsch and R. M. Fye, Phys. Rev. Lett. 56, 2521 (1986).

[34] B. A. Jones, Ph.D. thesis, Cornell University, 1987 (unpublished).

[35] B. A. Jones and C. M. Varma, Phys. Rev. Lett. 58, 843 (1987).
[36] B. A. Jones, C. M. Varma, and J. W. Wilkins, Phys. Rev. Lett. 61, 125 (1988).

[37] R. M. Fye and J. E. Hirsch, Phys. Rev. B 40, 4780 (1989).
[38] A. J. Millis, B. G. Kotliar, and B. A. Jones, in Field Theories in Condensed Matter
Ph/,/-. edited by Z. Tesanovic (Addison-Wesley, Redwood City, CA, 1990), pp.
159-166.

[39] 0. Sakai, Y. Shimizu, and T. Kasuya, Solid State Commun. 75, 81 (1990).
[40] T. Saso, Phys. Rev. B 44, 450 (1991).

[41] 0. Sakai and Y. Shimizu, J. Phys. Soc. Jap. 61, 2333 (1992).

[42] 0. Sakai and Y. Shimizu, J. Phys. Soc. Jap. 61, 2348 (1992).

[43] I. Affleck and A. W. W. Ludwig, Phys. Rev. Lett. 68, 1046 (1992).

[44] T. A. Costi, A. C. Hewson, and V. Zlati6, J. Phys.: Condens. Matter 6, 2519 (1994).

[45] I. Affleck, A. W. W. Ludwig, and B. A. Jones, Phys. Rev. B 52, 9528 (1995).
[46] C. J. C'!li, Introduction to S., i,:.,..':, Tunneling Microscopy (Springer, Berlin, 1996).

[47] C. Gonzalez-Buxton and K. Ingersent, Phys. Rev. B 54, R15614 (1996).
[48] K. Ingersent, Phys. Rev. B 57, 11936 (1996).

[49] J. B. Silva, W. L. C. Lima, W. C. Oliveira, J. L. N. Mello, L. N. Oliveira, and J. W.
Wilkins, Phys. Rev. Lett. 76, 275 (1996).

[50] T. A. Costi and C. Kieffer, Phys. Rev. Lett. 76, 1683 (1996).

[51] W. H. Press, S. A. Teukolsky, W. T. V. I. ili:. B. P. Flannery, Numerical Recipes
in Fortran 77 (Cambridge University, Cambridge, England, 1996).

[52] K. Yoshida, Theory of Magnetism (Springer, New York, 1996).

[53] R. Bulla, Th. Pruschke, and A. C. Hewson, J. Phys. Condens. Matter 9, 10463
(1997).

[54] A. C. Hewson, The Kondo Problem to H .,;; Fermions (Cambridge University Press,
Cambridge, England, 1997).

[55] D. Goldhaber-Gordon, H. Shtrikman, D. Mahalu, D. Abusch-Magder, U. Meirav,
and M. A. Kastner, Nature 391, 156 (1998).

[56] C. Gonzalez-Buxton and K. Ingersent, Phys. Rev. B 57, 14254 (1998).

[57] V. Madhavan, W. C(! in T. Jamneala, M. F. Crommie, and N. S. Wingreen, Science
280, 567 (1998).

[58] J. Wu, B. Gu, H. C('! in W. Duan, and Y. Kawazoe, Phys. Rev. Lett. 80, 1952
(1998).

[59] J. Li, W.-D. Schneider, R. Berndt, and B. Delley, Phys. Rev. Lett. 80, 2893 (1998).

[60] W. C('! i, T. Jamneala, V. Madhavan, and M. F. Crommie, Phys. Rev. B 60, R8529
(1999).

[61] B. C. Paul, Ph.D. Dissertation, University of Florida, 2000 (unpublished).

[62] J. Nygird, D. H. Cobden, and P. E. Lindelof, Nature 408, 342 (2000).

[63] S. Sasaki, S. de Franceschi, J. M. Elzerman, W. G. van der Wiel, L. P.
Kouwenhoven, M. Eto, and S. Tarucha, Nature 405, 6788 (2000).

[64] L. Kouwenhoven and L. Glazman, Phys. World 14, 33 (2001).

[65] H. Jeong, A. M. Chang, and M. R. Melloch, Science 293, 2221 (2001).

[66] K. Kang, S. Y. Cho, J.-J. Kim, and S. C. Shin, Phys. Rev. B 63, 113304 (2001).

[67] M. Vojta and R. Bulla, Phys. Rev. B 65, 014511 (2001).

[68] W. Hofstetter, J. Knig, and H. Schoeller, Phys. Rev. Lett. 87, 156803 (2001).

[69] B. Kubala and J. Knig, Phys. Rev. B 65, 245301 (2002).

[70] V. Madhavan, T. Jamneala, K. N .,, .1: W. C'!. i, J. L. Li, S. G. Louie, and M. F.
Crommie, Phys. Rev. B 66, 212411 (2002).

[71] R. Aguado and D. C. Langreth, Phys. Rev. B 67, 245307 (2003).

[72] V. M. Apel, M. A. Davidovich, E. Anda, C. A. Biisser, and G. Chiappe,
Microelectronics Journal 34, 729 (2003).

[73] P. S. Cornaglia and C. A. Balseiro, Phys. Rev. B 67, 205420 (2003).

[74] M. Garst, S. Kehrein, T. Pruschke, A. Rosch, and M. Vojta, Phys. Rev. B 69,
214413 (2004).

[75] C. A. Biisser, G. B. Martins, K. A. Al-Hassanieh, A. Moreo, and E. Dagotto, Phys.
Rev. B 70, 245303 (2004).

[76] N. J. Ci i.- J. M. Taylor, E. A. Lester, C. M. Marcus, M. P. Hanson, and A. C.
Gossard, Science 304, 565 (2004).

[77] J. C. C'!I i, A. M. Chang, and M. R. Melloch, Phys. Rev. Lett. 92, 176801 (2004).
[78] P. S. Cornaglia, H. Ness, and D. R. Grempel, Phys. Rev. Lett. 93, 147201 (2004).

[79] A. Fuhrer, T. Ihn, K. Ensslin, W. Wegscheider, and M. Bichler, Phys. Rev. Lett. 93,
176803 (2004).

[80] V. L. Campo and L. N. Oliveira, Phys. Rev. B 70, 153401 (2004).

[81] L. Fritz and M. Vojta, Phys. Rev. B 70, 214427 (2004).

[82] P. Simon, R. Lopez, and Y. Oreg, Phys. Rev. Lett. 94, 086602 (2005).

[83] R. Leturcq, L. Schmid, K. Ensslin, Y. Meir, D. C. Driscoll, and A. C. Gossard, Phys.
Rev. Lett. 95, 126603 (2005).

[84] K. Ingersent, A. W. W. Ludwig, I. Affleck, Phys. Rev. Lett. 95, 257204 (2005).

[85] P. S. Cornaglia and D. R. Grempel, Phys. Rev. B 71, 075305 (2005).

[86] G. H. Ding, C. K. Kim, and K. Nahm, Phys. Rev. B 71, 205313 (2005).

[87] L. G. G. V. Dias da Silva, S. E. Ulloa, and T. V. Shahbazyan, Phys. Rev. B 72,
125327 (2005).

[88] L. G. G. V. Dias da Silva, N. P. Sandler, K. Ingersent, and S. E. Ulloa, Phys. Rev.
Lett. 97, 096603 (2006).

[89] A. J. Heinrich, J. A. Gupta, C. P. Lutz, and D. M. Eigler, Science 298, 1381 (2006).

[90] A. J. Heinrich, C. P. Lutz, and C. F. Hirjibehedin, e-J. Surf. Sci. Nanotech. 4, 384
(2006).

[91] P. Simon, J. Salomez, and D. Feinberg, Phys. Rev. B 73, 205325 (2006).
[92] R. Zitko and J. Bonca, Phys. Rev. B 74, 045312 (2006).

[93] R. Bulla, T. Costi, and T. Pruschke, arXiv:cond-mat/0701105vl (2007).

[94] L. G. G. V. Dias da Silva, N. P. Sander, K. Ingersent, and S. E. Ulloa, Phys. Rev.
Lett. 99, 209702 (2007).

[95] R. M. Potok, I. G. Rau, H. Shtrikman, Y. Oreg, and D. Goldhaber-Gordon, Nature
446, 167 (2007).
[96] L. Borda, Phys. Rev. B 75, 041307(R) (2007).

[97] M. T. Glossop and K. Ingersent, Phys. Rev. B 75, 104410 (2007).
[98] L. G. G. V. Dias da Silva, K. Ingersent, N. P. Sander, and S. E. Ulloa,
arXiv:cond-mat/0804.0805 (2008).

[99] http://www-unix.mcs.anl.gov/mpi/index.htm
[100] http://www.netlib.org/lapack/index.html

[101] http://www.netlib.org/scalapack/index.html

BIOGRAPHICAL SKETCH

Brian Lane has been enjoying the study of physics since his junior year in high school.

In 2003, he graduated with his B.S. in physics from Jacksonville University (summa cum

laude). In 2005, he received his M.S. in physics from the University of Florida and was

wed to Amy Knight at Creekside Community C('!ii !i Since 2004, he has been developing

this dissertation under the supervision of Professor Kevin Ingersent, while developing

himself as a physics instructor. Upon successful defense of this dissertation, he will

continue to pursue his interests in condensed matter theory and physics education as a

faculty member at Jacksonville University.

PAGE 1

1

PAGE 2

2

PAGE 3

3

PAGE 4

Therearemanypeopletothank,perhapsmorethanmymemoryandwordscandojustice: 4

PAGE 5

5

PAGE 6

page ACKNOWLEDGMENTS ................................. 4 LISTOFFIGURES .................................... 8 ABSTRACT ........................................ 11 CHAPTER 1INTRODUCTION .................................. 12 1.1TheKondoProblem .............................. 12 1.1.1ResistivityMinimumandtheSuccessoftheKondoModel ..... 12 1.1.2TheAndersonModel .......................... 16 1.1.3FurtherAttemptsatPerturbativeTechniques ............. 19 1.2TheNumericalRenormalizationGroup .................... 22 1.2.1TheRenormalizationGroupConcept ................. 22 1.2.2ApplicationtotheKondoProblem ................... 23 1.2.3IterativeProcedure ........................... 26 1.2.4FixedPointsandResults ........................ 29 1.3SurfaceandQuantumDotRealizationsoftheOne-ImpurityKondoEect 31 1.3.1ScanningTunnelingMicroscopyStudies ................ 31 1.3.2QuantumDotStudies .......................... 35 1.4SystemsofMultipleImpurities ......................... 40 1.4.1TheoreticalStudiesofTwo-ImpurityModels ............. 40 1.4.2Multiple-ImpuritySTMStudies .................... 44 1.4.3DoubleQuantumDotStudies ..................... 46 1.5StudyOverview ................................. 54 2BACKGROUNDMATERIAL ............................ 57 2.1ApplicationoftheNRGtotheAndersonModel ............... 57 2.1.1DiscretizationandEigensolution .................... 57 2.1.2CalculationofThermodynamicProperties .............. 59 2.1.3CalculationofSpectralFunctions ................... 61 2.1.4FixedPointsandResults ........................ 62 2.2ExtensiontoTwo-ImpuritySystems ...................... 66 2.2.1TransformationtoOne-DimensionalForm ............... 67 2.2.2DiscretizationandEigensolution .................... 70 2.2.3SpecialCases:IdenticalImpuritiesandR=0 ............ 73 2.2.4CalculationofThermodynamicandSpectralProperties ....... 74 6

PAGE 7

............... 76 3.1ParallelizationoftheNRGEigensolution ................... 76 3.2ParallelizationoftheMatrixElementCalculation .............. 78 4STMSTUDIES .................................... 82 4.1ReviewofSingle-ImpurityBehavior ...................... 82 4.1.1Single-ImpuritySTMSetup ...................... 82 4.1.2ResultsforSingle-ImpuritySTM .................... 84 4.2Two-ImpuritySTMStudies .......................... 87 4.2.1Two-ImpuritySet-up .......................... 87 4.2.2ThermodynamicandSpectralResults ................. 91 4.2.3Two-ImpuritySTMConductance ................... 94 4.2.4VaryingImpurityParameters ...................... 97 5ASYMMETRICDOUBLEQUANTUM-DOTDEVICES ............. 103 5.1DoubleQuantumDotSetup .......................... 104 5.1.1ModelandSimplications ....................... 104 5.1.2SpecialCase:U2=0 .......................... 106 5.2Side-CoupledDQD ............................... 107 5.2.1SpecialCase:U2=0 .......................... 107 5.2.2ExtendedCase:U2>0 ......................... 108 5.3ParallelDQD .................................. 111 5.3.1SpecialCase:U2=0 .......................... 112 5.3.2ExtendedCase:U2>0-PhaseDiagramandSusceptibility ..... 116 5.3.3ExtendedCase:U2>0-SpectralFunctionandConductance .... 124 6CONCLUSIONS ................................... 133 6.1ScanningTunnelingMicroscopyStudies .................... 133 6.2DoubleQuantumDotStudies ......................... 134 6.3Epilogue ..................................... 136 REFERENCES ....................................... 138 BIOGRAPHICALSKETCH ................................ 143 7

PAGE 8

Figure page 1-1Resistivityminimum ................................. 13 1-2Logarithmicdiscretizationoftheenergyspace ................... 24 1-3Impuritycoupledtoachainofelectronstates ................... 25 1-4Wall-clocktimefortheNRGprocedure ....................... 28 1-5EvolutionofenergyspectraduringNRGprocess .................. 30 1-6DierentialconductanceforasingleCoatom ................... 33 1-7DierentialconductanceforasingleCeatom ................... 34 1-8DierentialconductanceforasingleCoatomwithvaryingtipposition ..... 36 1-9DierentialconductanceforasingleCeatomwithvaryingtipposition ..... 37 1-10Conductanceofaquantumdotforvariousdotoccupancies ............ 39 1-11DierentialconductanceforCoatoms ....................... 45 1-12DierentialconductanceforapairofNiatoms ................... 46 1-13DierentialconductanceforvariousCecongurations ............... 47 1-14Setupforcoupleddoublequantumdotexperiment ................. 48 1-15CoulombblockadevalleysforaDQD ........................ 49 1-16Dierentialconductancevs.interdotcoupling ................... 50 1-17Side-coupledDQDdisplayingtwo-stageKondoscreeningbehavior ........ 51 1-18Conductancevs.gatevoltageforaside-coupledDQD ............... 52 2-1Single-impurityTimp(T) .............................. 65 2-2Single-impurityAd(!) ................................ 66 3-1Wall-clocktimevs.NPforiterativeeigensolutionwithNkeep=3000. ...... 79 3-2Wall-clocktimevs.NPforcalculationofoperatormatrixelements ....... 81 4-1Single-ImpuritySTMSetup ............................. 83 4-2Single-ImpuritySTMConductance ......................... 84 4-3Single-ImpuritySTMConductance,tc=0 ..................... 85 8

PAGE 9

................ 86 4-5One-ImpuritySTMConductance-negativevoltages ............... 87 4-6FitofG(V)withtd=tc=0 .............................. 88 4-7FitofG(V)withtd=tc=0:3 ............................. 89 4-8Two-impuritySTMsetup .............................. 89 4-9StrengthofRKKYinteraction ............................ 90 4-10Two-impuritymagneticsusceptibility ........................ 92 4-11Two-impurityspectralfunction ........................... 93 4-12Two-impurityconductance,td=tc=0:1,positivevoltages ............. 96 4-13Two-impurityconductance,td=tc=0:1,negativevoltages ............. 97 4-14Two-impurityconductance,td=tc=0:4,positivevoltages ............. 98 4-15Two-impurityconductance,td=tc=0:4,negativevoltages ............. 99 4-16Fittedtwo-impurityconductance,positivevoltages ................ 100 4-17FittedSTMtwo-impurityconductance,negativevoltages ............. 101 4-18Dierentialconductanceforvarioustwo-impuritysystems ............. 102 5-1Doublequantumdotschematic ........................... 105 5-2Dot1spectralfunctionforside-coupledDQDwithU2=0 ............ 108 5-3Zero-Tconductanceforside-coupledDQDwithU2=0 .............. 109 5-4Zero-TA11(!)forside-coupledDQDwithU2>0 ................. 110 5-5Zero-Tconductanceforside-coupledDQDwithU2>0 .............. 111 5-6Zero-Tconductanceforside-coupledDQDwith2>0 .............. 112 5-7Parallel-dotphasediagramforU2=0 ........................ 113 5-8ObservationofupperQPTinTimpvs.TforU2=0 ............... 114 5-9LinearrelationshipbetweenTKand+1 115 5-10Zero-TA11(!)vs.!>0forKondo-phaseparallelDQD,U2=0 ......... 116 5-11Zero-TA11(!)vs.!<0forKondo-phaseparallelDQD,U2=0 ......... 117 5-12Zero-TA11(!)vs.!>0forlocal-moment-phaseparallelDQD,U2=0 ..... 118 9

PAGE 10

..... 119 5-14ApproximatephasediagramforparallelDQD,U2>0 ............... 120 5-15Criticalpoint+1cvs.U2 121 5-16Scaled+1cvs.U2 122 5-17TracesofTimpvs.T,U2=103 123 5-18Zero-Tlimitofimpvs.+1,U2>0 ......................... 124 5-19Kondotemperaturevs.+1forparallelDQDwithU2>0 ............ 125 5-20KondotemperatureTKvs.1forparallelDQDwithoutlocal-momentphase .. 126 5-21Zero-TA11(!)vs.!>0forKondo-phaseparallelDQD,U2=103 127 5-22Zero-TconductanceG=G0vs.1forparallelDQD ................. 128 5-23Zero-TconductanceG=G0vs.+1forparallelDQD ................ 129 5-24Zero-TconductanceG=G0vs.1forparallelDQDwithnolocal-momentphase 130 5-25Zero-TconductanceG=G0vs.2forparallelDQDwith1=U1=2 ....... 131 5-26Zero-TconductanceG=G0vs.2forparallelDQD ................. 132 10

PAGE 11

PAGE 12

PAGE 13

SchematicplotofresistanceRvs.temperatureTinexperiments(blackcircles)thatrstobservedwhatisnowcalledtheKondoeect.TheminimuminRisunexplainablebytheoriginalprediction(greencurve).Kondo'sresult(redcurve)successfullydescribedtheminimum,butproducedadivergingRasT!0. resistancereachedaminimumvalueatacertaintemperatureTmin(ontheorderof4to40K),andthenincreasedastemperaturewasfurtherdecreased[ 1 9 ].ThisbehaviorissummarizedschematicallyinFigure 1-1 .Theunexpectedminimumwasattributedtothepresenceofimpurities,aswhentheimpurityconcentrationcimpwasvaried,itwasfoundthatTmin/c1=5impandthedepthoftheminimumR(T=0)R(T=Tmin)/cimp.Studywasthenundertakentounderstandwhataspectoftheseimpuritieswascausingthisminimum. Onepossibleexplanationwasthattheresistanceminimummightbeattributedtothepotentialscatteringoftheconductionelectronsotheimpurities.SuchaprocesscanbemodeledbytheHamiltonian 13

PAGE 14

istheimpurity-freeconductionbandHamiltonian.Thecy~k(c~k)operatorsarestandardfermionoperatorsthatcreate(annihilate)conductionbandelectronsofmomentum~~kandspin.Theconductionbandhasabandwidthof2D(suchthatj~kj
PAGE 15

9 ],suggestingthattheminimumwascausedbyaninteractionbetweenthespinsoftheimpuritiesandthespinsoftheconductionelectrons.Lastly,KondonotedthatwhileR(T=0)R(T=Tmin)wasproportionaltocimp,theratio(R(T=0)R(T=Tmin))=R(T=0)didnotvarywithcimp,indicatingthatthisspininteractionwasasingle-impurityeect. Basedontheconclusionthattheresistivityminimumwascausedbythespininteractionbetweentheconductionelectronsandindividuallocalizedmagneticmoments,Kondoappliedthes-dexchangemodeldevelopedbyZener[ 2 ],nowmorecommonlyreferredtoastheKondoHamiltonian: Here,j;0;j=x;y;z;arethePaulispinmatricesand~Sistheimpurityspin.ThequantityJisthestrengthoftheexchangeinteractionbetweenthespinsoftheimpurityandtheconductionelectrons;ifJisnegative(positive),theinteractionisferromagnetic(antiferromagnetic).(NotethatthisHamiltonianmayalsobeexpandedtoincludepotentialscattering,butthisfeatureisnotnecessarytoobtainKondo'sresults.)WiththisHamiltonian,KondocarriedoutaperturbativeexpansiontothirdorderinJtoobtaintheresistivitycontributedbyasingleimpurity:[ 10 ] 8e2~F[12J0(F)ln(kBT=D)]:(1{5) Kondothusarrivedatthetotalresistivity 15

PAGE 16

1-1 ). Kondo'ssolutionwasaphenomenalsuccess.HeusedhisperturbativeexpressionforR(T)tottheresultsfromawidevarietyofexperiments,includingsomethatimpliedthatJ<0(ferromagneticcase)[ 10 ].Hismodelwasalsousedtosuccessfullyproduceperturbativeexpressionsforamagneticimpurity'scontributionsimp(T)andCimp(T)tothezero-eldmagneticsusceptibilityandthespecicheat[ 54 ]: 3kBT10J+(0J)2ln(kBT=D)+c2(0J)2; wherec2isaconstantthatdependsontheconductionbanddensityofstates. However,therewereanumberofremainingquestionstobeaddressed:Howdotheselocalmomentsarisewithinthehostmetal?Howdoesthespinexchangeinteractiontakeplace?WhathappensatverylowtemperaturesTTmin?ThelastquestionwaspromptedbytheironythatthoughthelnTterminKondo'sexpansionexplainedtheresistanceminimum,thistermdivergesasT!0,andKondo'ssolutionwasnotvalidforverylowtemperatures.Thusaroseaquesttondanon-perturbativesolution,dubbed\theKondoproblem." 7 ],whichwasdevelopedbeforeKondoemployedthes-dexchangemodel.TheAndersonmodelconsidersthescatteringofconductionelectronsotransitionmetalorrareearthimpuritieswhosedlevelsliewithinthehostmetal'sconductionband.ThehybridizationVd~kbetweenanimpurity'sd-stated(denedinacoordinatesystemthatplacestheimpurityattheorigin)andtheconductionelectronstatesd(Wannier 16

PAGE 17

PAGE 18

()== where =0jVdj2;(1{13) and~dsatises ~dd0jVdj2lnD+~d Withtheresult( 1{12 ),theimpurity'scontributiontothemagneticsusceptibilityis whichdoesnotleadtoaCurie-Weissbehaviorin(T).UsingEq.( 1{12 )tocalculateRimpdoesnotresultinaresistanceminimumatlowertemperatures,either.Thus,animpuritymodelwithU=0cannotexplaintheKondoeect. (3)Next,considerthecaseofU>0;Vd=0(anisolatedinteractingimpurity).Suchanimpurityisdescribedbyanemptystatej0iofenergy0,twosinglyoccupiedstatesj"iandj#iofenergyd,andadoublyoccupiedstatej"#iofenergy2d+U.Thus,ifd<0andd+U>0,thesinglyoccupiedstateswillbeenergeticallyfavored,producingaspin-1 2localizedontheimpurity. ForthegeneralcaseofU;Vd>0,theconditionsforlocal-momentformationaremodiedbecausetheimpuritylevelsacquireawidthoforder.Forexample,thesinglyoccupiedlevelisbroadenedtotherangefromdctod+c,wherec4.Thus,ingeneral,alocalmomentarisesifdandd+U.Insuchacase,theAndersonmodelmapsontotheKondomodelusingtheSchrieer-Woltransformation[ 12 ]toobtainthespincouplingJandpotentialscatteringK: 18

PAGE 19

21 Thechangeinthedensityofstates()thentakesaformsimilartoEq.( 1{12 ),butwithreplacedbyaresonancewidthoforder (ThisexpressionfortheKondotemperatureTKwillbeexplainedinmoredetailbelow.)ThisresonanceiscalledtheAbrikosov-SuhlresonanceortheKondoresonance.Thus,theKondomodelisactuallyalimitingcaseofthemoregeneralAndersonmodel. 1{5 )through( 1{8 )remainedalongstandingproblem.In1965,Abrikosov[ 11 ]triedtoeliminatethedivergencebycarryingoutKondo'sperturbativeexpansionfurtherandsummingallhigher-ordertermsproportionalto[0Jln(kBT=D)]n.Fortheimpuritycontributiontothemagneticsusceptibility,hearrivedat[cf.Eq.( 1{7 )] 3kBT10J Thisresultrepresentedprogress,astheferromagneticcase(J<0)remainswell-behavedasT!0.Unfortunately,intheantiferromagneticcase(J>0),thisexpressiondivergesattheKondotemperatureTK[Eq.( 1{18 )]insteadofatT=0. In1966,YosidaapproachedtheKondoproblemwithavariationalmethod,andfoundthegroundstateenergiesforasinglet(labeleds)andtriplet(labeledt)conguration[ 13 ]: Thus,fortheantiferromagneticcase,agroundstateinwhichtheimpurityandconductionelectronsformasingletisenergeticallypreferred.Forthecaseofaspin-1 2impurity,similar 19

PAGE 20

PAGE 21

whereEistheenergyeigenvalueoftheeectiveHamiltonian.ThiseectiveHamiltonianisvalidoverareducedbandwidth~DDjDj. Forlow-energyexcitations,onecansetE,~k,and~k0tozeroinEq.( 1{22 )toarriveatthesecond-orderscalingequation dlnD=0J2:(1{23) ReducingDto~DbyabsorbingtheenergylevelsnearthebandedgeintotherenormalizationofJto~JwillcauseJtoincrease.Thus,thehigher-energyexcitationsstillcontributetothesolutionoftheKondomodel,buttheydosobyrenormalizingJ. InthisrescalingofDandJ,if0<0J1,theKondotemperatureTK(J;D)functionsasascalinginvariant: Thispropertyisextremelyuseful;forexample,theimpuritycontributiontothemagneticsusceptibilitycanbewritten TK;(1{25) whereF(X)isauniversalfunction. ThermodynamicresultsbasedonEq.( 1{23 )stillexhibitadivergenceatT=TK.CarryingtherescalingouttothirdorderinJproduces 2(0J)3+O((0J)4);(1{26) resultinginamoreaccurateexpressionfortheKondotemperature: 21

PAGE 22

1 2ln1 TK:(1{28) ApplyingEq.( 1{28 )tothemagneticsusceptibilitygives =(gB)2 ln(T=TK)ln[(ln(T=TK))] 2(ln[(T=TK)])2+O1 [ln(T=TK)]2: Equation( 1{30 )stillcontainsalogarithmicdivergenceatT=TK.Thus,evenafterAnderson'sscalingmethods,thelow-temperaturebehaviorremainedamystery. Thepoorman'sscalingmethodprovidedvastprogresstowardsolvingtheKondoandAndersonproblems,butstillbrokedownduetoitsperturbativeapproach.Andersonhadshownthatitwasreasonablefor~J!1asT!0andforimp(T!0)tobenite,thoughhewasunabletoarriveatawell-denedlow-temperatureexpressionforimp.However,theconceptofscalingwouldprovetobeessentialtothesuccessfulnon-perturbativesolutionfoundinthenumericalrenormalizationgroup. 1.2.1TheRenormalizationGroupConcept or,morecompactly, 22

PAGE 23

PAGE 24

PAGE 25

Impuritycoupledtoachainofelectronstates.InWilson'sdiscretization,thecouplingnn=2. Afterdroppingthestateswithq6=0,aLanczostransformationisappliedtothesetofoperatorsfaym;q=0;;bym;q=0;g,resultinginasetffyn;g,denedaslinearcombinationsofaym;q=0;andbym;q=0;. Withthenewfyn;operators,theKondoHamiltonianbecomes ThisHamiltoniancanbepicturedasformingalinearchain(seeFigure 1-3 ),inwhichtheimpurityisconnectedonlytosite0(whosespinoperatoriswrittenas~s0;;0=1 2fy0;~f0;0)andeachsiteiscoupledonlytoitsnearestneighbors. Workingwithadensityofstates()=1=(2D);D<
PAGE 26

2NXn=0n=2nfyn;fn+1;+fyn+1;fn;+2J~S~s0;;0;(1{41) suchthatthesequenceofHamiltoniansH0;H1;H2;:::obeystherecursionrelation Thisrelationshipenablesaniterativesolutionofthechain.First,thesimpleHamiltonianH0=2J~S~s0;;0isdiagonalized,obtainingeigenstatesjN;Q;S;Sz;milabeledbythequantumnumbersQ(chargemeasuredfromhalf-lling),S(totalspin),Sz(spinz-component),andm(whichdistinguishesstateswithallotherquantumnumbersthesame).Theseeigenstatesarethenmultipliedbythestatesofsite1(j0i;j"i;j#i;andj"#i)toconstructabasisofstatesinwhichH1iswritten.ThenewHamiltonianH1isthendiagonalized,obtainingeigenstatesthataremultipliedbythestatesofsite2towriteH2,etc.ThisiterativeprocedureprovidesthebasisoftheRGtransformation. Oneoftheprimarygoalsofthisprocedureistocomparethesetofenergyeigenvalues(alsoreferredtoastheenergyspectrum)ofHNtothatofHN+1tondaxedpointoftheRGtransformation.AsthecharacteristicenergyscaleofeachHNdecreasesasDN=2,inordertomakethecomparisonbetweensuccessivespectra,adimensionlessHamiltonianHNisused: HN2(N1)=2 26

PAGE 27

1{42 )inthismannerleadstotheRGtransformation HN+1=R(HN) (1{44) =1=2HN+N(fyN;fN+1;+h:c:): DiagonalizingHNyieldsthesetofenergyeigenvaluesEN(~m)andtheireigenvectorsjN;~mi(wherejN;~miisashorthandforjN;Q;S;Sz;mi).Computationally,theRGtransformationcanbecarriedoutbycalculating HN+1=X~mEN(~m)jN;~mihN;~mj +NX~m;~m0hhN;~mjfyN;jN;~m0ijN;~mihN;~m0jfN+1;+h:c:i: 1{44 )canbethoughtofasactingonthesetofeigenvaluesofHNandmatrixelementsoffN;: Theiterativeprocedureiscarriedoutonacomputer|hencethename\NumericalRenormalizationGroup."EachHamiltonianHNisblockdiagonalintheconservedquantumnumbersofthemodel.Forexample,itcanbeshownthat[HK;~S]=0,andsotherewillbenonon-zeromatrixelementsbetweenstatesofdierenttotalspinordierentspinz-component.Thisblock-diagonalnaturepermitsafastereigensolution.AsdescribedinChapter 2 ,itisstraightforwardtoobtainthermodynamicandspectralquantitiesasfunctionsoftemperaturebykeepingtrackoftheenergyeigenvaluesandoperatormatrixelementsateachiteration. TheNRGmethod'smostsignicantdetrimentisthatthedimensionoftheHamiltonianHNis22N+3.Sincethetimetodiagonalizeamatrixgrowsasthecubeofitsdimension,thecomputationaleorttoobtainacompletesolutionofHNbecomesprohibitivebeyondN5.Thus,inpractice,thesummationsover~mand~m0arerestrictedtotheNkeepstates 27

PAGE 28

Wall-clocktimefortheNRGprocedureandthehighestenergylevelretainedEkeepvs.Nkeep.Thewall-clocktimeandenergylevelsarescaledbytheirvaluesforNkeep=10.Notehowtheruntime(thecostofthecalculation)growsmuchmorequicklythantheenergylevelretention(thebenet). oflowestenergy,ortoallstateswithenergylowerthansomelevelEkeep,thusreducingthesubsequentHamiltoniantoamanageablesize. Toillustratethestatetruncationapproximation,Figure 1-4 showstheruntimerequiredfortheNRGproceduretobecarriedoutona2.2-GHzAMDOpteronprocessorforvariousvaluesofNkeepandtheresultingtruncatedvalueofEkeep.TheruntimeandEkeepresultsarescaledbytheirvaluesforNkeep=10,whichare0:3secondsand1:68,respectively.Proportionately,theruntimegrowsfasterthanEkeep.(TheseresultsareactuallyfortheNRGstudyoftheAndersonmodel,asdescribedinSection 2.1.1 ,buttheyillustratethepointwell.) Thisstatetruncationisanacceptableapproximation,sincethelow-temperaturebehaviorislargelydeterminedbythelower-lyingenergyeigenstates(EN(~m).10),anda 28

PAGE 29

1-5 ).ItturnsoutthatthetransformationRasgiveninEq.( 1{47 )hasnoxedpoints,sincetheenergyspectraforoddandevenNaremanifestlydierentinform[ 30 ].Thus,itisthetransformationR2thatyieldsxedpoints,andcomparisonsaremadebetweenEN(~m)andEN+2(~m). WilsondemonstratedthattheferromagneticKondomodelhasastablexedpointHJ=0(conrmingtheresultsfrompoorman'sscaling),andthattheantiferromagneticKondomodelhasanunstablexedpointHJ=0andastablexedpointHJ=1.(Technically,eachofthese\xedpoints"isapairofxedpoints|oneforevenNandoneforoddN|butthethermodynamicpropertiesarethesame.) WilsonconstructedHJ=1tosuccessfullyarriveatanitelow-temperatureexpressionforimpintheantiferromagneticcase[ 23 ]: 4kBTK;TTK:(1{48) ThisconstantresultshoweddenitivelythattheimpuritylocalmomentdisappearsasT!0duetotheimpurityspinbeingfullycompensated.Thenumber0:4128,dubbedtheWilsonnumberw,hassincebeenobtainedexactlyfrommethodsbasedontheBetheAnsatz[ 28 29 ],inwhichtheKondomodelisreducedtoasubsidiaryspinproblemassolvedbyYang[ 15 ],producingtheexactresultw=eC+1=4=p 29

PAGE 30

EnergylevelsE(N)vs.iterationnumberN(even)fortheNRGprocedureappliedtotheAndersonmodel.Theenergyeigenstatesarelabeledbyspinandcharge.(Duetospinandparticleholesymmetries,statesofoppositespinoroppositechargearedegenerate.)Theenergylevelsaremeasuredfromthegroundstateenergy,suchthatthelowestenergylevelateachiterationisalways0.Plateausinthespectrumindicatethepresenceofaxedpointoftherenormalizationgrouptransformation.Atiteration6,anunstablexedpointisapproached,butthelevelsowawayfromitbyiteration96.Astablexedpointisreachedatiteration110,signifyingtheonsetoftheKondoeect. Wilsonalsoreproducedthehigh-temperature(TTK)resultsfrompoorman'sscaling[Eq.( 1{30 )],andobtainedanexpressionforthecrossoverregime[ 23 ] whichisidentiedasaCurie-Weisslawforareduced(orpartiallyscreened)localmoment. TheNumericalRenormailzationGroupissimilarlyappliedtotheAndersonmodel;IwilldeferthatdescriptionuntilChapter 2 30

PAGE 31

PAGE 32

PAGE 33

STMdierentialconductancedI=dVforatiplocatedoverbareAu(111)andoverasingleCoatom.TheCotraceisttedwithaFanolineshapeofwidth2kBTK[ 57 ].FromMadhavanetal.,Science24April1998280:567-569.ReprintedwithpermissionfromAAAS. paths[ 73 ],suchthatq=0indicatesnotunnelingtotheimpurity,q=indicatesnotunnelingdirectlytothesurface,andq=1indicatesequaltunnelingthroughbothchannels. Tocomparewiththeexperimentalresults,Eq.( 1{50 )canbeappliedtotheU6=0Andersonmodelwiththemodication Im();(1{52) 33

PAGE 34

STMdierentialconductancedI=dVforatiplocatedoverasingleCeatomonanAg(111)surface.ThedataarettedwithaFanolineshapeofwidth2kBTK[ 59 ].Reprintedwithpermissionfrom J.Li,W.-D.Schneider,R.Berndt,andB.Delley,Phys.Rev.Lett.80,2893(1998) .Copyright1998bytheAmericanPhysicalSociety. where()istheimpurityelectron'sself-energy.ForenergiesneartheKondoresonanceandforT
PAGE 35

1{50 )]stillacquiresanon-uniformshapeduetothepresenceoftheimpurityelectron'sself-energy.Physically,thispropertycanbeunderstoodbyconsideringsecond-orderprocesses,inwhichelectronsthathavetunneledfromthetiptothesurfacethentunnelfromthesurfacetotheimpurityandbackagaintothesurfacebeforereturningtotheSTMapparatus.Theextratunnelingcausestheelectronstopickupaphaseshiftrelativetotheelectronsthatdonottunnel,anddestructiveinterferencetakesplace. TheFanoresonancehasbeenobservedinthedierentialconductanceinmanyotherSTMexperiments[ 62 89 90 ],includingmagneticimpuritiesadsorbedontocarbonnanotubes,whichexhibitaone-dimensionaldensityofstates,furtherdemonstratingtherobustnessoftheFano-resonancebehavior.WewillobservethattheNRGconrmstheFano-resonancebehaviorinChapter 4 STMstudiesalsopermitanexaminationofthespatialextentlKofthescreeningcloudthatcausestheKondoeect.SuchaquantityisnotdirectlycalculatedinthestandardNRGprocedure,butcanbeestimatedexperimentallybymovingtheSTMtiplaterallyawayfromtheimpurity.Figures 1-8 and 1-9 showG(V)forvariouslateraldistancesbetweenthetipandimpurity.ThemagnitudeandasymmetryoftheFano-shapefeaturedecreaseasthelateraldistanceincreases,indicatingareductionofqcausedbyadropintherateofelectronstunnelingtotheimpurity.Sucientlyfarfromtheimpurity,thefeaturelessconductanceisrecovered.Section 6.1 describesanovelapproachtocalculatelKwithanextensionoftheNRG[ 96 ]. 35

PAGE 36

PAGE 37

STMdierentialconductancedI=dVfortiplocatedatvariouslateraldistancesfromasingleCeatomonAg(111)[ 59 ].ThefeatureatV=70mVisattributedtotheAgsurfacestate,andismodiedbytheproximityoftheimpurity.Reprintedwithpermissionfrom J.Li,W.-D.Schneider,R.Berndt,andB.Delley,Phys.Rev.Lett.80,2893(1998) .Copyright1998bytheAmericanPhysicalSociety. 37

PAGE 38

Tuningdpermitscontroloverthenumberofelectronsinthedot.WhenaQDistunedtohaveanoddnumberofelectrons(i.e.,itmeetstheconditionsforlocal-momentformationintheAndersonmodelasdescribedinSection 1.1.2 ),itmusthaveanon-zerototalspin,andthereforecandisplaytheKondoeect.Becauseofthetwo-dimensionalgeometryofmostQDs(suchassemiconductorQDsformedattheinterfacebetweenGaAsandAlGaAs),theKondoeectcausesanincreaseintheconductanceG(V)throughthedot[ 55 64 ]asTdecreases. ThesepredictionsareconrmedintheresultsofaQDexperiment[ 64 ]showninFigure 1-10 .Part(a)ofthegureshowsG(V)vs.thegatevoltageVcorrespondingtod,fortemperaturesrangingfrom25mK(lightbluecurve)to1K(orangecurve).Asthegatevoltageisincreasedovertherangeshowninthegure,theoccupancyofthedotincreasesfromNtoN+4(whereNiseven).TheseregionsformaCoulombblockadepattern,indicatedbypeaksandvalleysintheconductance.Forevenoccupancy,theconductancedecreaseswithtemperature,indicatingthatnoKondoeecttakesplace;foroddoccupancy,theconductanceincreaseswithtemperature,indicatingtheoccurrenceoftheKondoeectandthusthepresenceofalocalmomentonthedot.Similarresultswerefoundin[ 55 ],wheretheenhancedconductancewasshowntodisappearfortemperaturesabovetheKondotemperature. Figure 1-10 (b)showsG(V)vs.TforthreevaluesofVthatproduceoddoccupancyN+1.AsV(i.e.,d)changes,sodoesTK.IfthedataareinsteadplottedasG(V)vs.T=TK,asinpart(c),theresultslieontopofeachother,indicatingthatG(V)isgivenbyauniversalfunctionF(T=TK)asseeninEq.( 1{25 ). 38

PAGE 39

PAGE 40

PAGE 41

1{4 )]isextendedintoitstwo-impurityversion: Here,thenewindexi=1;2labelstheimpuritiesand~riisthepositionvectorofimpurityi.PotentialscatteringtermshavebeenomittedfromEq.( 1{54 )forsimplicity. Aswiththesingle-impurityKondomodel,thetwo-impurityKondomodelcanbethoughtofasaspecialcaseofthetwo-impurityAndersonmodel,wheneachimpurityfavorstheformationofalocalmoment.Thetwo-impurityAndersonmodelwillbedescribedingreaterdetailinChapter 2 Varioustheoreticalandnumericalmethodshavebeenemployedtostudythetwo-impurityKondoandAndersonmodels,includingperturbativescaling[ 32 ],theNRG[ 34 { 36 41 42 49 74 80 ],quantumMonteCarlomethods[ 33 37 ],variationalmethods[ 40 ],andconformaleldtheory[ 43 45 ].Thesestudieshavehighlightedtheimportanceofanumberfeaturesthatarenotpresentinthesingle-impurityKondomodel:thepresenceoftwoeectiveconduction-bandchannels,agreatercomplexityoftheelectrondensityofstates,andacompetitionbetweentheKondoeectandmagneticorderingeects. Therstnewfeature,thepresenceoftwoeectiveconduction-bandchannelsarisingfromthecouplingtotheimpuritiesatlocations~r1and~r2,willbeexploredinmoredetailinSection 2.2.1 .Essentially,itiscomputationallymoreconvenienttodeneconductionelectronstatesthataresymmetricandantisymmetricaboutthemidpointbetweenthetwoimpurities,calledthe\even"channeland\odd"channel,respectively.Whilethisdivisionoftheconductionbandisatheoreticalconstruct,itdoeshavephysicalimplications,suchasthetwo-stageKondoscreeningprocessdescribedbelowandfoundin[ 27 ]. Thesecondnewfeature,agreatercomplexityoftheelectrondensityofstates,willalsobeillustratedinmoredetailinSection 2.2.1 .ThevariableimpurityseparationR=j~r1~r2jmeansthatonecannotsimplyignorethe~k-dependencearisingfromthe 41

PAGE 42

PAGE 43

32 35 37 49 ].Thisspin-1isthenscreenedinatwo-stageKondoeect.Inthistwo-stageKondoeect,thespin-1reducestoaspin-1 2,whichthenreducestoaspin-0.(Forexamplesofthisbehavior,seeFigures 1-17 and 4-10 .)Thesetwostagescanbeexplainedbythepresenceoftheevenandoddconduction-bandchannels(seeabovediscussionand[ 27 ]),eachofwhichexhibitsitsownKondotemperature.Chapter 4 willexplorehowthistwo-stagescreeningprocessmaybeobservedinanSTMexperiment. AnotherclassofbehaviorariseswhenjIRKKYjT1impK.Forthisregime,theRKKYinteractionisnotstrongenoughtocombinetheimpuritymomentsintoaspin-1beforetheKondoeectscreenseachmomentindividually.Thus,thetwoimpuritiesareeectivelyisolatedfromeachother[ 32 49 ].ThisbehaviorwillalsobeobservedinthecontextofSTMexperimentsinChapter 4 AverydierentbehaviorariseswhenIRKKYT1impK.Forthisregime,theRKKYinteractioncausestheimpurityspinstoantiferromagneticallyalignintoaspinsingletstate,suchthatnoKondoeectcantakeplace[ 32 35 37 49 ].ThisbehaviorwillalsobeobservedinthecontextofSTMexperimentsinChapter 4 Onelastfeatureofthemodelthathasbeenexploredingreatdetailisthetransitionbetweentheindependent-impurityregime(jIRKKYjT1impK)andtheimpurity-singletregime(IRKKYT1impK).Theoriginalpoorman'sscalinganalysis[ 32 ]predictedasmoothtransitionbetweenthetworegimes.ThissmoothtransitionwaslatersupportedbyquantumMonteCarlomethods[ 33 37 ].However,anNRGstudy[ 35 ]predictedthepresenceofanunstablecriticalpointseparatingthetworegimes.Atrst,thediscussionofthediscrepancycenteredaroundthedierencesinthemethodsused(forexample,theNRGworkhadnotused~k-dependentcouplingsbetweentheimpuritiesandtheconduction-bandchannels),butnallyaconnectionbetweenthetworesultswasveriedbyvariationalmethods[ 40 ],conformaleldtheory[ 43 45 ],andamodiedversionof 43

PAGE 44

PAGE 45

PAGE 46

PAGE 47

PAGE 48

Setupforcoupleddoublequantumdotexperiment[ 76 ].DotsLandRarecoupledbytheconductingregionC.Gatevoltagesdetermineenergylevels,occupancies,andcouplingofthedots;leadsmeasuretheconductanceofeachdot.FromCraigetal.,Science23April2004304:565-567.ReprintedwithpermissionfromAAAS. totheexchangeofelectronsacrossthecentralregion,leadingtoeitheraspin-0state(antiferromagneticalignment)betweenthetwodotswithnoKondoeect,ortoaspin-1state(ferromagneticalignment)withaweakerKondoeecthavingaTKsmallerthanthetemperatureoftheexperiment.Whicheverbehavioristhecause,themagneticinteractionbetweenthedotssignicantlymodiestheKondoeect. ThismodicationcanbeobservedtodevelopinFigure 1-16 ,whichplotstheconductanceoftheleftdotGL(VL)forvariousstrengthsofthecouplingbetweendotRandtheconductingregion.WhendotLanddotRbothhaveanoddnumberofelectrons(partA),increasingthecoupling(andthereforeincreasingthemagneticorderingeect)causestheKondoresonancetobecomesuppressedandtosplitintotwopeaks.Thissuppressionandsplittingbehavioragreeswiththeoreticalpredictions[ 71 82 88 ]andhasalsobeenobservedinparallel-coupleddoublequantumdotexperiments[ 77 ].WhendotLhasanoddnumberofelectronsanddotRanevennumber(partB),increasingthecouplingcausesslightquantitativemodicationtotheKondoresonance,butthequalitativefeaturesremainthesame. 48

PAGE 49

PAGE 50

DierentialconductanceofdotLfortheDQDsetupshowninFigure 1-14 ,forvariousstrengthsofthecouplingbetweendotRandtheconductingregion,(A)whendotsLandRbothhaveanoddnumberofelectronsand(B)whendotLhasanoddnumberofelectronsanddotRanevennumber[ 76 ].FromCraigetal.,Science23April2004304:565-567.ReprintedwithpermissionfromAAAS. ByconstructingDQDdevicesofdierentgeometriesandne-tuningthedeviceparameterswithgatevoltages,onemaystudyagreatvarietyofphysicalbehaviorspreviouslyinaccessibleinexperiments.Thisnewbreadthofpossibilitieshasstimulatedalargenumberoftheoreticalandnumericalstudies. Forexample,astudyutilizingtheNRGandslave-bosonmean-eldtheory[ 85 ]workingwithaside-coupledDQD(forwhichacentraldotisconnectedtotheleadsandtoasidedot)predictsatwo-stageKondoscreeningeect(cf.Section 1.4.1 ),asseeninFigures 1-17 and 1-18 .TheupperpanelofFigure 1-17 showsTimpvs.T=TK,indicatingthescreeningoftheDQDspinintwostages.Thelowerpanelshowshowthistwo-stageprocessisevidencedintheconductance(giveninunitsofG0=2e2=h).ThescreeningofthecentraldotatTTKenhancestheconductance,whichrisesnearlytoG0.ThescreeningofthesidedotatamuchlowerT104TKleadstoasuppressionofthe 50

PAGE 51

Figure1-17. Side-coupledDQDdisplayingtwo-stageKondoscreeningbehavior[ 85 ].Modelparametersare1=2=0:25,U1=U2=0:5,1=0:035,andinter-dottunnelingamplitude=0:003.Theupperpanelshowsthesquareoftheimpuritymagneticmoment2[kBTimp=(gB)2inthelanguageofthepresentwork]vs.T=TK.Thelowerpanelshowshowthetwo-stageprocessisevidencedintheconductance(giveninunitsofG0=2e2=h).Reprintedwithpermissionfrom P.S.CornagliaandD.R.Grempel,Phys.Rev.B71,075305(2005) .Copyright2005bytheAmericanPhysicalSociety. Figure 1-18 [ 85 ]showstheconductanceforthisside-coupleddeviceas=1=2isvariedatvarioustemperatures.ForT=0(thebottompanel),attheparticle-holesymmetricpoint=U=2wherethetotaldeviceoccupancyiseven,theconductancevanishes,whereasthereisperfectconductanceat0andU,wherethetotaloccupancyofthedeviceisodd.Thus,hereweseethebehaviorofaCoulombblockadevalley,asobservedinexperimentalstudies. 51

PAGE 52

Conductancevs.=1=2(controlledbygatevoltage)foraside-coupledDQD[ 85 ]atvarioustemperatures.ModelparametersareU1=U2=0:25,1=0:125,andinter-dottunnelingamplitude=0:025.NotethattheconductancefeaturesvanishwhenT&TK.Reprintedwithpermissionfrom P.S.CornagliaandD.R.Grempel,Phys.Rev.B71,075305(2005) .Copyright2005bytheAmericanPhysicalSociety. InthemiddlepanelsofFigure 1-18 ,thesecondstageoftheKondoscreeningeectisnottakingplace,andsotheconductanceat=U=2isenhanced.Inthetoppanel,thetemperatureisontheorderofTK,andsothefeaturesoftheconductancespectrumdisappear.(Thisagreeswiththeexperimentin[ 55 ],inwhichtheconductanceenhancementdisappearswhenTexceedsTK.) OtherexperimentalandtheoreticalstudieshaveusedDQDdevicestoexamineCoulomb-blockadebehavior[ 72 ],observeinterferencebetweenelectronpaths[ 75 ], 52

PAGE 53

58 68 69 87 ],observethetwo-channelKondoeect[ 95 ],constructtwo-levelQDdevices[ 75 ],probethelengthoftheKondoscreeningcloud[ 91 ],andobservetheKondoeectinaQDside-coupledtoaquantumwire[ 66 91 ]. Whiletherehasbeenmuchtheoreticalworkdone,themajorityofthesestudieshasfocusedondevicescomposedofidenticalquantumdotsordevicesforwhichbothdotsareintheKondoregime.Recently,variouseorts[ 88 91 94 98 ]havebeenmadetostudyaclassofDQDdevicesinwhichonedotisintheKondoregimeandtheotherisweaklyinteracting(i.e.,hasasmallCoulombenergyU2,andthereforenolocalmoment).AttentionhasbeenpaidtothespeciallimitofU2=0which,thoughdiculttoachieveexperimentally,canbemodeledasasingleAndersonimpurity(Dot1)connectedtoaconductionbandviaanonconstanthybridization[ 88 ] ()X~kjV~kdj2(~k) (1{56) =p whereistheinter-dottunnelingamplitudeandi=V2iistheeectivehybridizationwidthofdotiwiththeleads. Theeectsofthisnonconstanthybridizationhavebeenstudied[ 88 94 98 ]inthespecicgeometriesoftheside-coupledDQD(inwhichcasetheKondo-likeDot1couplestotheleadsonlythroughthenon-interactingDot2)andtheparallelDQD(inwhichcasebothdotsconnectdirectlytotheleads,butnottoeachother).Intheside-coupledconguration,forsucientlystronginter-impuritytunneling,theKondoresonanceinDot1splits,producingnoticeablechangesinthedeviceconductance.Intheparallelconguration,therearisesapairofquantumphasetransitionsseparatingKondo-screenedphasesandalocal-momentphaseinwhichtheimpurityremainsunscreeneddowntozerotemperature.SincetheconditionU2=0isexperimentallyunreasonabletoobtain,itis 53

PAGE 54

5 ,Iwillemploythetwo-impurityAndersonmodel(sincetheeectiveone-impuritymodelislimitedtoU2=0)tostudyhowthesebehaviorsaremodiedforU2&0andeventuallydestroyedforsucientlylargeU2. 1.4 showthatsystemsofcoupledmagneticimpuritiesandtheirDQDanalogsdisplayarichvarietyofbehaviorswarrantingdeeperinvestigation.AsystemoftwocoupledKondo-likeimpuritieshasbeenshowntoexhibitinterestingproperties,butitisdiculttoobservethosepropertiesexperimentallyinbulkhosts.Studyingatwo-impuritysystemonasurfacewithSTMtechniquesshouldallowonetoprobeeectssuchastwo-stageKondoscreeningandimpuritysingletformation.Also,whilesystemsoftwoidentical(oratleasttwoKondo-like)impuritieshavebeenexploredindepth,systemsofhighlyinequivalentimpuritieshavereceivedlittleattention.ThisbroaderrangeofimpuritycongurationsisachievableinsystemsofDQDs.Theremainderofthisstudywillfollowalongthesetwopathsofinterest:Themanifestationofknowntwo-impuritybehaviorinSTMstudiesandthepropertiesofsystemsoftwohighlyinequivalentimpuritiesintheformofDQDs. Beforeembarkinguponthesepaths,inChapter 2 ,Iwillreviewthetransformationofthesingle-impurityandtwo-impurityAndersonHamiltoniansintoformsthatcanbestudiedusingNRGtechniques.IwilldescribetheNRGproceduresusedtodiagonalizetheseHamiltoniansandtocalculatethethermodynamicandspectralpropertiesofinterest.Inthecaseofthetwo-impurityAndersonHamiltonian,Iwillfocusonthespeciallimitsofidenticalimpuritiesandzeroimpurityseparation,asthesecomputationallysimpliedcaseswillbeofuseinChapters 4 and 5 InChapter 3 ,IwillgiveanoverviewofmyeortstoimprovethecomputationaleciencyoftheNRGiterativeprocedurebyapplyingparallelprocessingtechniquestothediagonalizationoftheiterativeHamiltoniansHNandthecalculationofoperator 54

PAGE 55

101 ]routines;inthesecondmethod,smallermatrixblocksarediagonalizedbyindividualprocessorsusingLAPACK[ 100 ]routines.ForthesamplecalculationshowninChapter 3 ,thetwomethodsresultinaminimumrelativewall-clocktimeof65%for4processors,reectingamediocrescalabilityofwall-clocktimewithincreasingthenumberofprocessors.Toparallelizethecalculationofoperatormatrixelements,Ihaveemployedamethodinwhicheachprocessorcalculatesa\chunk"ofmatrixelementsatatime,againtakingadvantageoftheblock-diagonalnatureofHN.ForthesamplecalculationsshowninChapter 3 ,thismethodresultsinamuchbetterscalabilityofwall-clocktimewithincreasingnumberofprocessors. InChapter 4 ,IwillapplytheNRGmethodsofChapter 2 tomodelSTMstudiesofsurfaceimpurities.Iwillbeginbyrecapitulatingthesetupandtheresultsforasinglesurfaceimpurity[ 73 ],andthenpresentnewresultsforapairofidenticalsurfaceimpuritieswithoutadirectinteraction.Byvaryingtheimpurityseparation,IwilldemonstratehowmanyofthebehaviorsdescribedinSection 1.4.1 aremanifestedintheSTMconductancespectrumasthemagnitudeandsignoftheRKKYinteractionchange.Inparticular,wewillseethatimpuritycongurationsthatdisplayatwo-stageKondoeect(IRKKYTK)produceaconductancespectrumgivenbyasumoftwoFanolineshapes(eachhavingitsownqandTK);eectivelyisolatedimpurities(jIRKKYjTK)produceaconductancespectrumgivenbyasingleFanolineshape;andimpuritieslockedintoaspinsingletstate(IRKKYTK)produceagenerallyfeaturelessconductancespectrum. InChapter 5 ,IwillapplytheNRGmethodsofChapter 2 tomodeltheaforementionedclassofhighly-asymmetricDQDdevices,beginningwiththeU2=0specialcasedescribedaboveandexpandingtoincludeU2&0.Thisnewrangeofdeviceparameterswillrequiretheusethetwo-impuritymodelofChapter 2 ,insteadoftheeectiveone-impurity 55

PAGE 56

88 ]discussedabove.Iwillfocusontheside-dotandparallelDQDcongurationsdescribedabove.Intheside-coupledcase,IwillexploretheeectsofpositiveU2onthesplittingoftheKondoresonanceandtheresultingchangesintheconductanceminimum[ 98 ].Intheparallel-dotcase,IwillexplainhowthepairofQPTsevolveandeventuallydisappearasU2isincreasedfromzero. InChapter 6 ,IwillsummarizethebehaviorsobservedinChapters 4 and 5 ,andpresentideasforfuturework. 56

PAGE 57

30 31 ].IwillthenoutlinetheprocedureforusingtheseNRGtechniquestocalculatethermodynamicandspectralpropertiessuchastheimpuritymagneticsusceptibilityandoperatorspectralfunctions.IwillthendescribethexedpointsthatariseintheNRGtreatmentandhowtheyaremanifestedinthethermodynamicandspectralproperties.Lastly,IwilldescribetheextensionoftheNRGtechniquestothetwo-impurityAndersonmodel. 7 ]consistsof(asorderedbelow)aconduction-bandterm,animpurityterm,andahybridizationterm: AllenergiesaremeasuredwithrespecttotheFermilevel("=F,j"jD)andrepeatedspinindices(,inthiscase)areimplicitlysummedover.Thecy~k(c~k)operatorsarestandardfermionoperatorsthatcreate(annihilate)conductionbandelectronsofmomentum~kandspin,andthedy(d)operatorscreate(annihilate)impurityelectronsofenergyd,spin,andon-siteCoulombrepulsionU. ItisstandardtoassumeforsimplicitythatthehybridizationparameterV~kdisaconstant,Vd,andthattheconductionbanddensityofstatesisaconstantoverabandwidthof2D:(")=0forj"j
PAGE 58

30 ]employsthesamediscretizationprocedureaswasdescribedfortheKondomodelinSection 1.2.2 and[ 23 ].TheAndersonHamiltonianisthustransformedintoalinearchainformsimilartoEq.( 1{37 ): +ddyd+U(dy"d")(dy#d#)+p 1{38 )and( 1{39 ). AswiththeNRGtreatmentoftheKondoproblem,thechainHamiltonian( 2{2 )issolvedviaaniterativediagonalizationprocedure.TheiterativedimensionlessHamiltonianis HN=(N1)=2N1Xn=0n=2nfyn;fn+1;+h:c: +Ndyd+UNdyd12+1=2Ndyf0;+h:c:; 1{40 ),and N=2(N1)=2 UN=2(N1)=2 N=2(N1)=2 TheeigenstatesofHN,againabbreviatedjN;~mi,arelabeledbychargeQ,spinS,andspinz-compomentSz,asHNcommuteswiththechargeandspinoperators,denedas[ 30 ] 2NXn=0fyn;~;0fn;0+1 2dy~;0d0: 58

PAGE 59

1.2.3 ,theNRGprocedurebeginsbydiagonalizingtheinitialHamiltonianH0: H0=0dyd+U0dyd12+1=20dyf0;+h:c::(2{9) UsingtheeigenvaluesE0(~m)andeigenstatesj0;~mi,theRGtransformation HN+1=R(HN) (2{10) =1=2HN+N(fyN;fN+1;+h:c:) (2{11) isemployedtoconstructH1,whichisthendiagonalized,etc. Forthecalculationofimp(T),thereisanexponentialtemperaturescalegivenby[ 30 ] whereisachosenparameter.(Wewillseethecriteriaforchoosingbelow.)IfduringarunoftheNRGalgorithmoneutilizestwovaluesofsuchthatTN(1)=TN+1(2)=T,onecancomparethetworesultingvaluesofimp(T)asameasureoftheNRGalgorithm'saccuracy(andtherebyevaluatethechoicesofandthenumberofstateskept). Thecalculationofimp(TN)isgivenby[ 30 ] Trexp(HN)Tr(S0Nz)2exp(H0N) Trexp(H0N);(2{13) 59

PAGE 60

2{8 ),andthesuperscript\0"referstoanoperatorforanimpurity-freesystem,i.e., H0N=(N1)=2"N1Xn=0n=2nfyn;fn+1;+fyn+1;fn;#; 2NXn=0fyn;~;0fn;0: ToevaluatethetracesinEq.( 2{13 ),oneutilizesthefactthattheenergyeigenvaluesofHN(H0N)donotdependonSz.Thus,thetracesreducetosumsoverkandS,wherekrepresentsallquantumnumbersotherthanSandSz.CarryingoutthesumsoverSzproducestheresults Trexp(HN)=Xk;S(2S+1)expEN(k;S); TrS2Nzexp(HN)=Xk;S1 12(2S+1)(2S+1)21expEN(k;S); Trexp(H0N)=Xk;S(2S+1)expE0N(k;S); Tr(S0Nz)2exp(H0N)=Xk;S1 12(2S+1)(2S+1)21expE0N(k;S): Thus,attheendofeachiterationN,TNimp(TN)maybeevaluatedusingonlytheenergyeigenvaluesofHNandH0N. Hereiswhereatrade-oarisesintheNRGprocedure.UsingEq.( 2{13 )incursanerrorontheorderof=(duetothecontributionfromchainsiteswithindexn>Nthatarenotincorporateduntilsubsequentiterations).Thus,choosingasmallvalueofwouldseemtobepreferred.However,inorderforEqs.( 2{16 )through( 2{19 )tobeevaluatedaccurately,allenergyeigenvaluesuptoseveraltimes1=mustbeincluded(sothatexpEN1fortheomittedstates).BecausethehigherenergyeigenstatesatiterationN1aretruncated(seeSection 1.2.3 ),theselargereigenvaluesatiterationNmaynotbecalculatedaccurately.Thus,ischosentokeeptheO( 60

PAGE 61

PAGE 62

2.1.2 ),givenby whereq1.Fortheresultsinthisstudy,Ihaveusedq=1:25. Inpractice,theDiracdelta-functionsinEq.( 2{20 )arereplacedwithlogarithmicGaussianfunctions[ 93 ]: wherebiswidthparameter,typicallychosenbetween0:3and0:7.(Thecalculationsinthisworkhaveusedb=0:5.)Thesetofexponentially-spacedpointsAd(!N)isthenconnectedbyasplinecurve[ 51 ]:apiecewisepolynomialfunctionwithacontinuoussecondderivative.AllofthespectralfunctionresultsinthisdissertationareforT=0,forwhichcasethesummationinEq.( 2{20 )isrestrictedtotermsinwhich~mand/or~m0are/isagroundstateofHN,andZN()reducestothenumberofgroundstates. 2{10 )]fortheAndersonmodelhasmorexedpointsthanfortheKondomodel.EachofthesexedpointsisdescribedbyaneectiveHamiltonianHNobtainedbyinsertingspecialvalues(0or1)ford,U,andintothegeneralHA[Eq.( 2{2 )].Ateachxedpoint,theexcitationspectrumcanberelatedbysuitablephaseshiftstothatatthefree-electronxedpointH0,whoseiterativeformH0NisgiveninEq.( 2{14 ).(AsmentionedinSection 1.2.4 ,eachofthesexedpointsistechnicallyapairofxedpoints,oneforevenNandoneforoddN.) Thefree-orbitalxedpointHFOcanbeobtainedbysettingd=U==0inEq.( 2{2 ).Theimpurityisthusdecoupledfromthemetal,andhasasetoffourdegeneratestates:j0i;j"i;j#i;j"#i.TheiterativeHamiltonianHFO;N=H0N,andthexedpointenergyspectrumisthatofH0Nwithanadditionalfour-folddegeneracyarisingfrom 62

PAGE 63

30 31 ].[Henceforth,IsetkB=(gB)2=1.] Themixed-valencexedpointHMVcanbeobtainedbysettingd==0andU=1(noticethatthisxedpointonlyoccursforparticle-holeasymmetriccases).Thedoubly-occupiedstatej"#iisthusinaccessible,andthestatesj0i;j"i;j#iaredegenerateanddecoupledfromtheconductionband.TheiterativeHamiltonianHMV;N=H0N,andthexedpointenergyspectrumisthatofH0Nwithanadditionalthree-folddegeneracyarisingfromthethreepossiblestatesofthedecoupledimpurity.ThisxedpointischaracterizedbyTimp=1=6[ 30 31 ]. Thelocal-momentxedpointHLMcanbeobtainedbysetting=0andd=U=1(cf.conditionsforlocal-momentformationinSection 1.1.2 ).Thusonlythesingly-occupiedstatesj"i;j#iareaccessible,andtheimpuritybehavesasafreelocalmoment.TheiterativeHamiltonianHLM;N=H0NandthexedpointenergyspectrumisthatofH0Nwithanadditionaltwo-folddegeneracyarisingfromthetwopossiblestatesofthedecoupledimpurity.ThisxedpointisthusanalogoustotheJ=0xedpointfromtheRGsolutionoftheKondoproblemandischaracterizedbyTimp=1=4[ 30 31 ]. Thestrong-couplingxedpointHSCcanbeobtainedbykeepingdandUxedandniteandsetting=1.TheiterativeHamiltonianHSC;NisgivenbyamodiedversionofH0Ninwhichthesumonnistakenfrom1toN1(insteadoffrom0toN1).Thus,theenergyspectrumofHSC;NisgivenbyH0N1.Inthiscase,theimpurityandthefyn=0;statesarelockedintoaspinsingletanddecouplefromtherestoftheconductionband.ThisxedpointisthusanalogoustotheJ=1xedpointfromtheRGsolutionoftheKondoproblemandischaracterizedbyTimp=0[ 30 31 ].Physically,switchingfromtheexcitationspectrumofH0NtothatofH0N1amountstoa=2phaseshiftofthelow-energyconductionelectrons.Thisphaseshiftreectsthefactthatsuchconductionelectronscannothoponorosite0oftheNRGchainbecausethatsiteislockedintoasingletwiththeimpurityspin.Itisthismaximal(unitary)scatteringthatisresponsible 63

PAGE 64

Lastly,thereisapairofsimilarxedpoints:theempty-impurity(orfull-impurity)xedpointcanbeobtainedbysetting=U=0andd=1(ord=),inwhichcaseonlythestatej0i(orj"#i)isallowed.TheiterativeHamiltonianHEI;N(orHFI;N)isequaltoH0N,andthexedpointspectrumisthatofH0Nwithashiftof1(or+1)inthechargequantumnumberQofeachstate.BothofthesexedpointsarecharacterizedbyTimp=0andarerelatedtothestrong-couplingxedpoint[ 30 31 ]. AstheNRGprocedureiscarriedout,theHamiltoniansHN(N=0;1;2;:::)willfollowatrajectorygovernedbytheirproximitytothesexedpoints.ThistrajectorypictureisusefulinunderstandingthethermodynamicresultsoftheNRGmethodbycomparingthetemperatureTNtojdj,U,andTK.Whenever,asTNdecreases(withincreasingN),Timppassesnearoneofitsspecialxed-pointvalues,theiterativeHamiltonianHNisnearthecorrespondingxedpoint. Forexample,Figure 2-1 plotsthemagneticsusceptibilityforasetofNRGcalculationsforthreevaluesofd,U,and.Thesecalculationswereperformedusing=3:0andkeepingNkeep=500statesattheendofeachiteration.Followingconvention[ 30 ],IwillidentifytheKondotemperaturebythecondition whichagreeswithEq.( 1{27 ). CurveAdemonstratesacaseinwhichUd,displayinganumberoftheRGxedpointsovervarioustemperatureregimes.AthighertemperaturesT>U,thereisafree-orbitalregimeinwhichTimp1=8.Astemperaturedecreasestojdj.T.U,thereisamixed-valenceregimeinwhichTimp1=6.WhenTdecreasesfurthertoTK
PAGE 65

zero,thisregimewouldpersistdowntozerotemperature.)Lastly,whenTnearsTK(kBT=D2:61010),thesystementersthestrong-couplingregimeandTimp!0. CurveBdemonstratesacaseinwhichparticle-holesymmetry(d=U=2)isobeyed,andU.Again,thereisafree-orbitalregimeatT>U,alocal-momentregimeatTK.T.jdj,andastrong-couplingregimeatT.TK(kBT=D1:0105),butthemixed-valenceregimehasdisappearedduetotheparticle-holesymmetry. CurveCdemonstratesanotherparticle-hole-symmetriccase,butwith=U.Here,thefree-orbitalregimeisfolloweddirectlybythestrong-couplingregime,withnolocal-momentbehaviorinbetween. Figure 2-2 plotstheT=0impurityspectralfunctioncorrespondingtotheparametersforcurveBinFigure 2-1 .ThemainfeaturesaretheHubbardbandsat~!=dand 65

PAGE 66

PAGE 67

PAGE 68

giving Thequantityinparenthesesisidentiedastheeldoperatorwhichannihilatesaconductionelectronatsite~r,andwhoseHermitianconjugatecreatesaconductionelectronatsite~r: Thus,thehybridizationtermisgivensimplyby InordertorewriteHhybinaone-dimensionalform,wedeneeven(e)andodd(o)parityoperators Replacingthe(~ri)operatorswiththep;operators(p=e;o),Hhybbecomes Jones[ 34 ]showedthat,ifoneconsidersanisotropicconductionband~k=j~kjandtransformsthebasisover~kintoabasisoverthedimensionlessenergy",onlyfourlinearcombinationsofthec~k;operatorscoupletotheimpuritiesateachenergy".Theselinearcombinationsaredenotedc";p;,andobey 68

PAGE 69

where k"R; k"R; and d(2{43) isthedensityofstatesperunitcellperspindirection. Thus,theconductionbandispicturedasconsistingoftwoeectivechannelsofelectrons,labeledbyparityp.Tocompletethepicture,evenandoddimpurityoperatorsaredened: Usingthec";p;anddp;operators,Hhybarrivesatitsnalone-dimensionalform Usingthenewsetofparityoperators,theconduction-bandtermissimilarlyinto 69

PAGE 70

21+1 2U1+2+1 2U2dye;de;+dyo;do;+1 21+1 2U121 2U2dye;do;+dyo;de;+1 2U1dye;de;+dyo;do;+dye;do;+dyo;de;12+1 2U2dye;de;+dyo;do;dye;do;dyo;de;12+dye;de;dyo;do;1 2(U1+U2): 2{26 )through( 2{29 )]intoadiscretizedform,itisnecessarytodeneanewbasisofconductionelectronwavefunctionsa;p;m;q(")(associatedwithenergies">0)andb;p;m;q(")(associatedwithenergies"<0).Thesestates,whicharecenteredaroundtheimpuritysites,aregeneralizationsofthestatesm;q(")denedinEq.( 1{36 ).Again,mistheindexofanenergybin,asdepictedinFigure 1-2 andqistheFourierharmonicindex.Asinthesingle-impuritycase,stateswithq6=0donotcoupletotheimpurity,andcanbeignoredasareasonableapproximationaslongasthediscretizationparameterisnottoomuchgreaterthan1. Thestateswithq=0aregivenby where (2{51) 70

PAGE 71

where 2Z+11d"w2p(R;"):(2{54) Usingthefn;p;operators,HA2isthustransformedintoalinearchainform,withtheconductionbandandhybridizationtermsgivenby andHimpsgivenbyEq.( 2{48 ).Here,thecouplingsp;nandp;naremorecomplicatedthantheirsingle-impurityformsinEqs.( 1{38 )and( 1{39 ),anddependonRthroughwp(R;"). Finally,aniterativedimensionlessHamiltoniananalogoustoEq.( 2{9 )isdened: HN=Hc;N+Himps;N+Hhyb;N; 71

PAGE 72

Hc;N=(N1)=2 Himps;N=1 2sumNdye;de;+dyo;do;+1 2diNdye;do;+dyo;de;+1 2UsumN[dye;"de;"dye;#de;#+dyo;"do;"dyo;#do;#+dye;"de;"dyo;#do;#+dye;"do;"dye;#do;#+dye;"do;"dyo;#de;#+dyo;"de;"dye;#do;#+dyo;"de;"dyo;#de;#+dyo;"do;"dye;#de;#dye;"de;"dye;#de;#dyo;"do;"dyo;#do;#]+1 2UdiN[dye;"do;"dyo;#do;#+dyo;"de;"dyo;#do;#+dyo;"do;"dye;#do;#+dyo;"do;"dyo;#de;#+dyo;"de;"dye;#de;#+dye;"do;"dye;#de;#+dye;"de;"dyo;#de;#+dye;"de;"dye;#do;#dye;"do;"dye;#do;#dyo;"de;"dyo;#de;#]+Ndye;de;dyo;do;; Hhyb;N=XpFp"sumN where sumN=2(N1)=2 2U1+2+1 2U2 diN=2(N1)=2 2U121 2U2 UsumN=(N1)=2 UdiN=(N1)=2 sumN=2(N1)=2 diN=2(N1)=2 72

PAGE 73

2{10 ),thetwo-impurityAndersonmodelmaybesolvediteratively. Thepresenceofthetwoelectronchannelsillustratesthegreatercomputationalcomplexityofthetwo-impurityAndersonmodeloverthesingle-impurityAndersonmodel.ThebasesoftheiterativeHamiltoniansHNsquareinsizecomparedtothesingle-impuritymodel,requiringmorestatestoberetainedaftereachNRGiterationinordertoachievethesamedegreeofaccuracy.Anothercomplicationisthemodicationoftheconduction-electrondensityofstatesbytheadditionofthefactorsin(k"R)=(k"R)inEqs.( 2{41 )and( 2{42 ).Duetothisfactor,thedensitiesofstatesofeven-andodd-paritystatesareinequivalent,andthetwo-impuritymodelbecomesparticle-holeasymmetricevenif()=().Inadditiontothemorecomplexnatureofthecalculations,therearenoweightparameterstoinvestigate(i;Ui;Vi;;andR).However,thesecomplicationscanbesimpliedbyconsideringoneoftwospecialcases:identicalimpuritiesorR=0. Inthespecialcaseofidenticalimpurities(1=2;U1=U2;V1=V2),Eq.( 2{26 )exhibitsparitysymmetry|i.e.,itremainsunchangedunderthetransformationsd1$d2(orde!de;do!doinparity-operatorlanguage)and~r1$~r2.Assuch,parityisaconservedquantumnumber,astherearenomatrixelementsofHNthatconnectstatesofoppositeparity(cf.the\sum"and\di"termsinHN).ThispropertycausesHNtobebrokenupintosmallermatrixblocks.Inadditiontothissimplication,therearethreefewerparameterstoinvestigate.Becauseoneoftenconsidersimpuritiesofthesameelement,thisspecialcaseisveryrelevant.Forexample,IwillutilizeparitysymmetryinthecontextofsurfaceimpuritiesinChapter 4 73

PAGE 74

Eventhoughthereisnoweectivelyonlyoneconductionband,themodelisstillmuchricherthantheone-impurityAndersonmodel:TheimpuritypartoftheHamiltonianHimp;NinEq.( 2{59 )containsfour(ratherthantwo)distinctimpurityoperatorsandseven(ratherthanthree)couplingparameters.WhilethelimitofR=0mayseemphysicallyunreasonableformagneticimpuritiesinametal,itisagoodapproximationwheneverthetwoimpuritiesareseparatedbyadistanceRsatisfyingkFR1,wherekFistheFermiwavevector.ThisconditionisassumedtoapplyinthestudiesofdoublequantumdotsinChapter 5 2.1.2 and 2.1.3 canalsobeappliedtothetwo-impuritymodel.ApplyingEq.( 2{13 )tothetwo-impuritymodelwillcalculatethecontributionofbothimpuritiestothemagneticsusceptibility. TocalculatespectralfunctionsusingaformulaanalogoustoEq.( 2{20 ),thematrixelementsofdyarereplacedbysomecombinationofthematrixelementsofdy1;anddy2;,permittingthecalculationofavarietyofspectralfunctions.Forexample,ifonewishestocalculatethelinearconductancethroughadoublequantumdotdevice,itisnecessaryto 74

PAGE 75

6 18 ] whereG0=2e2=histheconductancequantum,f(!;T)=[exp(!=T)+1]1istheFermi-Diracfunction,andT(!)isthetransmissionmatrix,whoseimaginarypartisgivenby ThespectralfunctionsofinterestmaybecalculatedusingtheNRGformula 75

PAGE 76

AsdescribedinSection 2.2.3 ,oneofthedrawbacksofusingtheNRGtostudytherichphysicsofthetwo-impurityAndersonmodelisthescaleofthecomputationalrequirements.ThebasesoftheiterativeHamiltoniansHNaresquaredinsizecomparedtothesingle-impuritymodel.Becauseofthegreatercomplexityofthesystem(suchasthepresenceoftwoinequivalentchannels),alargernumberofstatesmustberetainedattheendofeachiterationinordertoachieveanacceptablelevelofaccuracy.Thecomputationalrigorbecomesevengreaterwhenstudyingthree-impuritysystems[ 61 84 ]. Inordertoimproveonthecomputationaleciencyoftwo-impuritystudies(andmorecomplicatedsystems),IhavesoughttoadapttheiterativeNRGalgorithmtoutilizeparallelprocessors,implementedusingtheMessagePassingInterface(MPI,[ 99 ]).Thetwomosttime-consumingaspectsofthealgorithmarethediagonalizationofthematricesHNandtherecursivecalculationofoperatormatrixelements[seeEq.( 2{22 )].Asmentionedabove,thematrixelementcalculationsrequiremuchmorecomputertimethantheHamiltoniandiagonalization.Itisshownbelowthatthecalculationofoperatormatrixelementsalsoreapsagreaterbenetfromtheuseofparallelization. 100 ].Thismethodemploysamaster-slavearrangementoftheprocessors,inwhichonenode(themasternode)assignsthematrixblockstotheothernodes(theslavenodes)andmanagestheresults;sincethemasternodedoesnotperformanycalculationsitself,itsitsidleformostoftherun.Becauseofthestraightforwarddivisionoftheindependentdiagonalizationtasks,thisprocedurefallsunderthecategoryof\embarrassinglyparallel"algorithms,meaningthataslongasthere 76

PAGE 77

PAGE 78

101 ]. Ingeneral,itturnsoutthattheiterativeeigensolutionprocessdoesnotbenetsignicantlyfromtheapplicationofparallelprocessingtechniques,asevidencedbytheminimumrelativewall-clocktimeof65%inFigure 3-1 .WhilethereisimprovedperformanceforlargervaluesofNkeep,itisrarelynecessarytokeepmorethan3000states(incurrentNRGendeavors).Also,thebenetaccruedbyusingmultipleprocessorsisusuallyoutweighedbytheamountoftimespentwaitingformanyprocessorstoallbecomeavailableatonceinatypicalhigh-performanceclusterenvironmentinwhichmanyusers'jobscompeteforcomputationalresources.However,futurestudiesutilizingtheNRGprocedurethatrequireahighernumberofstatesmaybenetfromtheprecedingparallelizationtechniques. 2{22 )]isextremelycomputationallydemanding.ThenecessarycomputertimegrowsasthecubeoftheHamiltoniandimension,andisevenlargerthanthetimerequiredfortheeigensolutionprocess.However,Ihavesucceededingreatlyreducingthewall-clocktimeforthematrixelementcalculationby,again,adaptingthealgorithmtorunonparallelprocessors.Inthisimplementation,eachprocessorworksona\chunk"ofmatrixelementsindependently 78

PAGE 79

Wall-clocktimevs.numberofprocessorsNPforeigensolutionofHN=5.Matrixblockswithdimensionlessthan100werediagonalizedbyindividualprocessorsusingLAPACKroutines;matrixblockswithdimensiongreaterthan100werediagonalizedbyallprocessorsusingScaLAPACKroutines.Theverticalaxisisscaledbythewall-clocktimeforacalculationutilizingasingleprocessor. (madepossiblebytheblock-diagonalnatureofHNandofthematrixelementarrays),whichcausesthewall-clocktimetodecreaseratherquicklyasthenumberofprocessorsincreases.Thismethodemploysamaster-slavearrangementoftheprocessors,inwhichonenode(themasternode)assignsthechunkstotheothernodes(theslavenodes)andmanagestheresults.Sincethemasternodedoesnotperformanycalculationsitself,itsitsidleformostoftherun. Figure 3-2 showsthebenetsofapplyingtheaboveproceduretothecalculationofthematrixelementsofdyattheendofasingleiteration.Theverticalaxisisthewall-clocktimescaledbythewall-clocktimeforasingleprocessorrun.Thehorizontal 79

PAGE 80

ForNkeep=2000,thewall-clocktimescalesas1=NP,with1:5.ForNkeep=4000,thewall-clocktimedoesnotscaleassmoothlyasintheNkeep=2000case,butthewall-clocktimestillfollowsthegeneraltrendof1=Nkeep,with1:2.Thesevaluesofwouldseemtoexceedtheoptimumvalueof=1thatonecanexpectfroman\embarrassinglyparallel"algorithm.However,thevariationinthewall-clockdatafortheNkeep=4000runsindicatetheamountoferrorassociatedwithanalyzingthisperformancebehavior.Forexample,arunthatutilizesprocessorslocatedonthesamenodewillhavemuchfasterinter-processorcommunicationthanarunthatutilizesprocessorslocatedondierentnodes.Anotherpossiblefactoristhedicultyinguaranteeingthataparallelcalculationhasaccessto100%ofthecomputationalpowerofallNPnodes;ifotherjobsutilizethesameprocessor,thecomputationalperformancewilldecreaseandwall-clocktimewillincrease.Thissituationistypicallyguardedagainstbyjobschedulingsoftware,butisnotalwaysavoided.Thisseconderrorismorelikelytooccurduringarunthattakesalongeramountoftime,whichmayexplainwhytheuctuationsinthewall-clocktimearesomuchgreaterforthelargervalueofNkeep.Itwouldthereforebeaworthwhileendeavortoseehowthisalgorithmscalesinamorecontrolledparallelprocessingenvironment. Inspiteofsucherrorsanddiculties,theseresultsrepresentamuchlargerpayothanwasseenfortheeigensolutionphaseoftheNRGprocedure.Thissuperiorimprovementisattributedtothefactthattheparallelizedmatrix-elementalgorithmrequiresmuchlessinter-processorcommunicationthantheparallelizedeigensolutiondoes.Notwithstandingtheseimprovements,thebenetsachievedbyutilizingmanyprocessorsareoftenoutweighedbytheamountoftimespentwaitingforthoseprocessors 80

PAGE 81

PAGE 82

PAGE 83

SchematicSTMsetupindicatingpossibletunnelingpathsfromtheSTMtiptothemetalsurface. Following[ 73 ],Iassumethatthesurface-stateoperatory(~rimp)isequivalenttotheinnermostWilson-shelloperatorfy0;.Whilethisequivalencyisasignicantassumption,relaxingitshouldonlychangethequantitativedetailsoftheresults,andnotthequalitativefeatures.UsingtheNRGprocedure,wecancalculatethematrixelementsoftheoperatorsdyandfy0inthebasisoftheeigenstatesofeachHNandevaluateAaat!=!N(seeSection 2.1.3 fordetails)usingaformulaanalogoustoEq.( 2{20 ): 1.3.1 and 1.4.2 ).AsmentionedinSection 2.1.3 ,alloftheresultsinthischapterareforT=0,inwhichcasethesummationinEq.( 4{3 )isrestrictedtotermsinwhich~mand/or~m0isagroundstateofHN,andZN()reducestothenumberofgroundstates. 83

PAGE 84

Figure4-2. Conductance(arbitraryunits)vs.biasvoltageVforanSTMtiplocateddirectlyoverasinglemagneticimpurity.Modelparametersare(inunitsofD)d=U=2=0:5;=0:051,withtd=tcaslabeledinthelegend.TheNRGcalculationswereperformedfor=3,retainingNkeep=500statesaftereachiteration.ThelineshapesaresimilartothoseinFigures 1-6 and 1-7 Figure 4-2 plotstheSTMdierentialconductance(inarbitraryunits)forthissystemforvariousvaluesoftd=tc(theabsolutevaluesoftd,tcandtonlychangetheverticalscale),essentiallyreproducingtheresultsof[ 73 ].(TheNRGdatapointshavebeen 84

PAGE 85

Conductance(arbitraryunits)vs.viasvoltageVforanSTMtiplocateddirectlyoverasinglemagneticimpurity.ModelparametersareasforFigure 4-2 ,excepttc=0.Inthiscase,theconductanceisdirectlyproportionaltotheimpurityspectralfunctionAd(!)(cf.Figure 2-2 ). ttedwithsplinecurves.)Astd=tcincreases,andthefractionofelectronsthattunnelviatheimpuritybecomeslarger,theconductancelineshapebecomesasymmetricduetointerferencebetweenthetwotunnelingpaths.Figure 4-3 showstheconductancefortc=0,inwhichcasethereisnodirecttunnelingtothesurfaceandG(V)hasthesameformasAd(!)(cf.Figure 2-2 ). ItisimportanttonotethatalloftheinterestingfeaturesofthelineshapeoccuronasmallenergyscaledeterminedbykBTK,whichinthiscaseis1:2105D.ThissmallenergyscaleisevenmorevisibleinFigures 4-4 and 4-5 ,whichshowthesameconductancespectraonalogarithmicvoltagescale. 85

PAGE 86

Positive-biasdatafromFigure 4-2 ,replottedonalogarithmicvoltagescale. Asexamples,theNRGresultsfortd=tc=0:0and0:3aresuccessfullyttedwithaFanolineshapeinFigures 4-6 and 4-7 ,respectively.Thetswereobtainedusing[ 73 ] whereg0isabackgroundconductance,aisaconstant,x=eV=kBTK,andqisFano'sasymmetryparameter[cf.Eq.( 1{50 )].(NotethattheNRGresultsshowabroaderresonancethanthets;thisistypicalofNRGspectralresults,whichtendtobelessaccurateathigherenergyscales.)Thettedvaluesofqfortd=tc=0:0and0:3are0:0and0:84,respectively.ThesevaluesaresimilartothosefoundfortheSTMstudiesofsingleCeandCoatomsinSection 1.3.1 ,whichshowedsimilarlineshapes. 86

PAGE 87

Negative-biasdatafromFigure 4-2 ,replottedonalogarithmicvoltagescale. 4.2.1Two-ImpuritySet-up 4-8 showsthesetupforanSTMstudyoftwoimpuritiesseparatedbyadistanceR.Itaketheseimpuritiestobeidentical(i.e.,theyhavethesamevaluesofi=d,Ui=U,andVi=V).IalsoassumethatthesevaluesdonotchangewithR(asddidintheexperimentsdescribedinSection 1.4.2 ),andthatthereisnodirecttunnelingfromoneimpuritytotheother[i.e.,=0inEq.( 2{28 )].Thus,theimpuritiesonlyexchangeelectronsviatheconductionband.IalsoassumethattheSTMtipissituateddirectlyaboveoneoftheimpurities(impurity1inthediagram),suchthatthetip'selectronstunnelonlyintoimpurity1andthesurfacesurroundingit,andnotintoimpurity2.ThisassumptionmeansthatIcanusethesametunnelingparameterstdand 87

PAGE 88

PAGE 89

NRGresultsforthesingle-impuritySTMconductanceG(V)fortheparametersshowninFig. 4-2 withtd=tc=0:3(squares)ttedtoaFanolineshape(line).Thebest-tvalueoftheFanoparameterqinEq.( 4{4 )isq=0:84. Figure4-8. SchematicSTMsetupforastudyoftwoimpurities. 89

PAGE 90

Plotoff(kFR)enteringEq.( 4{6 ),calculatedforthetop-hatdensityofstatesandbanddispersion~kk3employedinthischapter.RegionsofferromagneticandantiferromagneticinteractionarelabeledFMandAFM,respectively. dispersion~kk3,thestrengthoftheRKKYinteractionisgivenby[ 52 ] where0J=8=U1isthespin-exchangecouplingenteringtheKondomodelandf(kFR)isanoscillatingfunctionoftheseparationRmultipliedbytheFermiwavevectorkF,plottedinFigure 4-9 WhenIRKKY>0theimpuritiesexhibitaferromagneticinteraction(favoringtheformationofanimpurityspintriplet),andwhenIRKKY<0theimpuritiesexhibitanantiferromagneticinteraction(favoringtheformationofanimpurityspinsinglet).InanSTMexperiment,changesinthemagnitudeandsignoftheRKKYinteractionmaybeobservedbymovingtheimpuritiesusingtheSTMtip.Weshallseethatthechangebetweenferromagnetic(FM)andantiferromagnetic(AFM)RKKYinteractionsexertsakeyinuenceontheshapeoftheSTMconductancespectrum. 90

PAGE 91

32 35 ]andcompareIRKKYwiththecharacteristicenergyscaleoftheKondoeect,givenbykBtimesthe(R-independent)one-impurityKondotemperature[ 54 ]: ToplottheSTMconductance,IuseanextensionofEq.( 4{1 ) dV=4e2 whereAa1(!)isthespectralfunctionoftheimpurity-1tunnelingoperator AsinSection 4.1.1 ,whereItooky(~rimp)/fy0;,here,Itaketheoperatory(~r1)tobe(uptoaprefactor)f0;1;p 2{54 ).Again,thisassumptionwillonlychangethequantitativedetailsoftheconductanceresults,butnotthequalitativefeatures. 4-10 and 4-11 plotthesusceptibilityandthedy1;-spectralfunctionAd1forseveralvaluesofkFR,alongwiththeone-impurityresultsforcomparison.(Notethattheone-impuritysusceptibilityresultshavebeenmultipliedby2torepresenttwoimpuritiesatinniteseparation.)AsinSection 4.1.2 ,theseresultsarefor(measuredinunitsofD)d=U=2=0:5and=0:051,withatop-hatdensityofstates()=0forjj
PAGE 92

ImpuritysusceptibilityTimpvs.temperaturefortwoimpurities,eachhavingU=2d=Dand=0:051D,andvedierentimpurityseparationsRspeciedinthelegend. thesecalculations,Ikept2000statesattheendofeachiteration(sincethetwo-impurityproblem'sHamiltonianshaveamuchlargerbasis). ForaseparationkFR=1:5,theRKKYinteractionisstronglyferromagnetic,asshownbytheratioIRKKY=T1impK=400.Atrelativelyhightemperatures(kBT=D104),theferromagneticRKKYinteractioncausesthetwoimpurityspinstopointinthesamedirection,creatingaspin-1singletstateforwhichkBTimp=(gB)2=1 2,asseeninFigure 4-10 .Thiscombinedspin-1isthenscreenedoutbytheKondoeectintwostages,seenasthetwoseparatefallsinTimp(occurringatkBT=D105andkBT=D1012)separatedbyarelativelyatregion.Thistwo-stagescreeningisalsomanifestedinFigure 4-11 ,wheretherearetworisesinthespectralfunction,occurringoverthesameenergyrangesatwhichthedropsinTimpoccurred,againindicatingatwo-stageKondoeect. 92

PAGE 93

PAGE 94

2momentsareKondo-screenedindependently.Thisindependent-screeningeectisalsoevidencedinthespectralfunction,whichmatchesupverycloselywiththeone-impurityresults.(Thisindependent-impuritybehaviorissurprisinggiventheestimatedratioofIRKKY=T1impK=50;seebelowforpossibleexplanations.) Finally,thetwoantiferromagneticcases(kFR=2:84and3:3)showverydierentbehaviorthantheferromagneticcasesdid.ThesusceptibilityplotshowsthattheantiferromagneticRKKYinteractioncausesthetwoimpurityspinstopointinoppositedirections,creatingaspin-0stateforwhichTimpfallswithdecreasingtemperatureevenfasterthanitdoesintheone-impurityKondoeect.Becauseofthisdominantdevelopmentofaspin-0system,theKondoresonanceinthespectralfunctionisverystronglysuppressedforthestronglyantiferromagneticcases.Asweshallseeinthenextsection,thissuppressionentersintotheSTMconductancespectrum,aswell. Asmentionedearlier,theindependent-impuritybehaviorfoundforkFR=2:5seemsinconsistentwiththeestimatedratioofIRKKY=T1impK=50.AsimilarinconsistencyoccursinthecaseforkFR=2:84,whichhasapredictedratioofIRKKY=T1impK=0:23,yetshowsbehaviorinwhichthetheantiferromagneticRKKYinteractiondominatesovertheKondoeect.Itismostlikelythattheseinconsistencies(whichoccurasoneapproachesIRKKY=0)indicate(1)aninaccuracyintheperturbativeexpressionsforIRKKYandT1impKinEqs.( 4{6 )and( 4{7 ),and/or(2)errorsarisingfromtheNRGdiscretization.Also,sucherrorsmaybeexacerbatedbytheproximityoftheunstablequantumcriticalpointdescribedinSection 1.4.1 .Whateverthereason,thevecasesstudiedinthissectiondoeectivelydisplaytherangeofqualitativebehaviorsexpectedinthetwo-impurityAndersonmodel. 4-12 through 4-15 plottheSTMdierentialconductanceforeachofthevevaluesofkFRconsideredintheprevioussection,alongwiththeresultsforasingle 94

PAGE 95

4-12 and 4-13 showresultsfortd=tc=0:1,whichcorrespondstoapproximately99%oftheelectronsthatleavethetiptunnelingdirectlyintothesurfaceand1%tunnelingintotheimpurity;Figures 4-14 and 4-15 showtheresultsfortd=tc=0:4,whichcorrespondstoapproximately86%oftheelectronsthatleavethetiptunnelingdirectlyintothesurfaceand14%tunnelingintotheimpurity.Fortheferromagneticcases,increasingtd=tccausesanincreaseinthezero-biasvalueofG,aswasseeninthecaseofaone-impuritystudy(cf.Figures 4-4 and 4-5 ).Fortheantiferromagneticcases,increasingtd=tccausesalmostnochangeintheconductancespectrum,indicatingtheisolatednatureofthespinsingletstateofthetwoimpurities. ForthecaseswithkFR=1:5,G(V)exhibitssimilarfeaturestotheone-impuritycase,but,aswasseeninthesusceptibilityandspectralfunction,thechangesinG(V)occurovertwoeasilydistinguishedenergyscales.ForthecasesofkFR=2:0;G(V)showslineshapessimilartothoseforkFR=1:5,butthetwoenergyscalesaremorediculttodistinguish.Asseeninthesusceptibilityandspectralfunction,theresultsforkFR=2:5closelyresemblethosefortheone-impuritycase.Forthetwostronglyantiferromagneticcases,theconductancespectrumisessentiallyfeaturelessforjeV=Dj<105duetothesuppressionoftheKondoresonanceseeninthespectralfunction. TheseresultsindicatethatthecompetitionbetweenKondoscreeningandtheRKKYinteractionisclearlyrevealedintheconductancespectrum.Thus,itispossibletostudythiscompetitioninSTMexperimentssimplybymovingthetwoimpurities.Wealsoseethatitispossibletostudyone-andtwo-stageKondoscreeningeectsinsuchexperiments.ForarelativelystrongferromagneticRKKYcoupling,theeectiveKondotemperature[asdenedinEq.( 2{25 )]dropswithseparation,and(asaresult)thelowestenergyscaleoftheconductancelineshapealsodecreaseswithseparation.ForarelativelystrongantiferromagneticRKKYcoupling,thesuppressionoftheKondoresonancecausesthelineshapetoremainfeaturelessforsmallerenergies. 95

PAGE 96

STMdierentialconductancevs.positivebiasvoltageforthesametwo-impuritycasesasshowninFigure 4-10 ,withtunnelingintoimpurity1describedbytd=tc=0:1. Thetwo-impurityconductanceresultsmayalsobettedwithFanolineshapes,thoughthetwo-stageKondoscreeningeectrequiresasumoftwoFanolineshapestobeused.Forexample,Figures 4-16 and 4-17 showtheNRGresultsfortd=tc=0:4,kFR=1:5ttedwiththeformula[cf.Eq.( 4{4 )] wherexi=eV=kBTK;i,withTK;1andTK;2beingthetemperaturesatwhichthetwostagesofthescreeningprocessoccur.ThebesttstothedatainFigures 4-16 and 4-17 areachievedwithkBTK;1=D=3:75105,kBTK;2=D=3:401012,q1=1:0,andq2=0:2.(Aswiththesingle-impurityresults,theNRGproducesresonancesslightlybroaderthanthetdoes.) 96

PAGE 97

STMdierentialconductancevs.negativebiasvoltageforthesametwo-impuritycasesasshowninFigure 4-10 ,withtunnelingintoimpurity1describedbytd=tc=0:1. 4{7 )and( 4{6 )].ThisbehaviorisillustratedinFigure 4-18 ,whichplotsthedierentialconductanceforatwo-impuritysystemwithgiveninthelegend.For=0:05and0:1,IRKKYT1impKandtheconductancehastheformofasumoftwoFanolineshapes,indicatingatwo-stageKondoscreeningeect.For=0:15and0:2,IRKKYT1impKandtheconductancehastheformofasingleFanolineshape,indicatingindependent-impurityscreening.(NotethatvaryingcannotchangethesignofIRKKY.) 97

PAGE 98

STMdierentialconductancevs.positivebiasvoltageforthesametwo-impuritycasesasshowninFigure 4-10 ,withtunnelingintoimpurity1describedbytd=tc=0:4. SimilarchangesmaybefoundbydecreasingU,sinceT1impKandIRKKYdependon0J=8=U. Itisalsoimportanttoconsidertheeectsofallowingelectronstotunneldirectlybetweenthetwoimpurities.For>0,impuritystatescreatedbydye;anddyo;haveenergiesd+andd,respectively.TheresultingchangestotheenergyspectrumofHNleadtoamodicationoftheRKKYinteractionIRKKY!IRKKY42=U[ 92 ].Thus,inthecasesinwhichtheKondoeectisobserved(IRKKY>0orIRKKYT1impK),aofsucientlylargemagnitude(forexample,jjinthecaseofUDandR=0[ 92 ])causesimpurity-singletformationanddestroystheKondoeect.Thus,theconductancesignaturesinFigures 4-12 through 4-15 shouldstillbeobservedinthegeneralcase6=0, 98

PAGE 99

STMdierentialconductancevs.negativebiasvoltageforthesametwo-impuritycasesasshowninFigure 4-10 ,withtunnelingintoimpurity1describedbytd=tc=0:4. withthemodicationthatthecomparisonofIRKKY42=UwithT1impKdetermineswhichofthethreetypesofbehaviorisobserved. TheresultsinthischapterdemonstratethatSTMdevicesmaybeusedtostudythecompetitionbetweenmagneticordering(induceddirectlyorindirectly)withtheKondoeectinasimpletwo-impuritysystem.Thedierentbehaviorsoftwo-stageKondoscreening,independent-impurityscreening,andimpurity-singletformationyielddierentsignaturesintheSTMconductancespectrum.Suchsignaturespermittheexperimentalobservationofthesebehaviorsthathavebeenstudiedindepthwithvarioustheoreticalmethods[ 32 { 37 40 { 43 45 49 74 80 ]. 99

PAGE 100

PAGE 101

PAGE 102

STMdierentialconductancevs.positivebiasvoltageforatwo-impuritysystemwithU=2d=D,kFR=1:5,tunnelingintoimpurity1describedbytd=tc=0:1,andvaluesof(inunitsofD)giveninthelegend. 102

PAGE 103

AsseeninSectionsand 1.3.2 and 1.4.3 ,quantumdot(QD)devicespermitthestudyofarichvarietyofphysicalbehaviorsthankstotheexperimentalcontroloftheirgeometryandenergylevelconguration.Duetotheirdiscreteenergylevels,QDdeviceshaverightlybeendescribedasarticialatoms(orarticialmoleculesinthecaseofcoupledQDs).Inkeepingwiththiscomparison,whenaQD'senergylevelsaretunedsuchthatitisenergeticallyfavorableforthedottohaveanoddnumberofelectrons(seeSection 1.3.2 ),thedotinteractswiththeleadelectronsinmuchthesamewaythatamagneticimpurityinteractswithconductionelectrons,thusdisplayingtheKondoeect[ 55 64 76 ].WhenmultipleKondo-regimedotsareconnectedtoeachother,anarticialmultiple-impuritysystemiscreated,permittingthecarefulstudyofmanybehaviorsthatwerepreviouslydiculttoaccessexperimentally. AsdescribedinSection 1.4.3 ,mostoftheprevioustheoreticalworkonDQDshasfocusedondevicesforwhichbothdotsareintheKondoregime.AdierentclassofhighlyasymmetricDQDdeviceshasbeguntobeexplored[ 88 91 94 98 ],revealingnovelproperties,includingasplittingoftheKondoresonanceandapairofquantumphasetransitions.Inthisnewclassofdevices,oneofthedots(hereafterreferredtoasDot1)istunedtobeintheKondoregime,whiletheother(hereafterreferredtoasDot2)iseectivelynon-interactingandistunednearaCoulombblockadepeak[i.e.,atamaximuminG(V)suchasinFigure 1-10 ],suchthatitdoesnotdisplayaKondoresonance.(Thus,thisdeviceexhibitspropertiesverydierentfromthetwo-impuritybehaviorsdiscussedinSection 1.4.1 .) Intermsofthetwo-impurityAndersonmodel,thisdescriptionmeansthatDot1willmeettheconditionsforlocal-momentformation(11,1+U11),whileDot2mustbetunedsuchthat20(tobenearresonancewiththeleads)andconstructedsuchthatU2=0(tobenon-interactingandnotdevelopalocalmoment).Thislast 103

PAGE 104

95 ]. Asweshallseebelow,theconditionU2=0alsoreducesthecomputationalrigorofstudyingthisasymmetricDQDnumerically,permittingtheuseofaneectivesingle-impuritymodel[ 88 94 98 ].However,asitisimpossibletorealizeaperfectlynon-interactingdotinexperiments,itisimportanttoconsidertheeectsofasmallbutnon-zerointeractiononDot2.ThischapterwillfocusonthelimitinwhichU2remainssmallcomparedtootherbareparametersofthetwo-impurityAndersonmodel,inwhichcasenosignicantlocalmomentisexpectedtoformonDot2. Inthischapter,IwillrstdescribetheapplicationoftheAndersonmodeltotheDQDdeviceofinterest.Iwillthenfocusontwodevicecongurations|the\side-dot"and\parallel"geometries|thatexhibitphysicalpropertiesofparticularinterest.Foreachconguration,IwillsummarizethemainfeaturesoftheU2=0specialcase,andthendemonstratehowthesefeaturesaremodiedbyanon-zeroU2. 5.1.1ModelandSimplications 5-1 andisdescribedbytheHamiltonian[ 88 ] Theleftandrightleadsarelabeledby=L;R,respectively,andthedotsarelabeledbyi=1;2.IassumethattheimpurityseparationR=j~r1~r2jsatiseskFR1(sincekFtendstoberelativelysmallinsemiconductorheterostructures),suchthatthedotsmay 104

PAGE 105

PAGE 106

6 18 ],giveninEq.( 2{68 ).ThisformularequiresfourimpurityspectralfunctionsA11(!),A22(!),A12(!),andA21(!),whicharecalculatedusingtheNRGformulainEq.( 2{70 ).Allcalculationsbelowwereperformedfor=3:0,retainingNkeep=500statesaftereachiteration. Evenwiththeabovesimplications,theprecedingHamiltonianisquitediculttostudy.Asafurthersimplication,previouswork[ 88 94 98 ]hasbegunbyconsideringthespecialcaseofU2=0.Ithasbeenshown[ 88 ]thatforU2=0,theabovetwo-impurityAndersonmodelcanbemappedontothatofasingleimpurity(inthiscase,Dot1)describedbyEq.( 1{13 )withanonconstanthybridizationfunction[ 88 ] ()X~kjV~kdj2(~k) (5{2) =p wherei0V2i.Becausethisspeciallimitmapsontoaneectiveone-impurityproblem,itmaybestudiedwithrelativecomputationalease. Thisnonconstanthybridizationgivesrisetoanumberofnoveleects,includingzero-eldsplittingoftheKondoresonanceandapairofquantumphasetransitions.Asweshallsee,bothoftheseeectsareobservableinthelinearconductanceoftheDQDdevice.Theseeectsaremoststraightforwardlyobservedintwospecialcases:theside-coupledDQD(1=0)andtheparallelDQD(=0).Ishallnowexaminethesetwospecialcases,beginningeachbyconsideringthelimitofU2=0andthenexpandingtodiscusstheeectsofsmallbutniteU2. 106

PAGE 107

5.2.1SpecialCase:U2=0 Inthespecialcase1=0and>0,Dot1isconnectedtotheleadsonlythroughDot2.IntheU2=0limit,Eq.( 5{3 )producesaLorentzianhybridizationfunctionofwidth2centeredon!=2.Figure 5-2 showsthat,when2=0(i.e.,Dot2ispreciselyonresonance),forsucientlysmallvaluesof,theDot-1spectralfunctionA11(!)displaysashapesimilartothatseeninthecaseofaconstanthybridization:theHubbardbandsat!1and1+U1,andtheKondoresonanceofwidth2kBTK(cf.Figure 2-2 ).Ifisincreased,however,theKondoresonancesplits.Thissplittingisattributedtothefactthattherelativelylowvalueof(!)forj!j&kBTKcausestheKondoresonancetorise,buttheenhancedvalueof(!=0)causesA11tofallas!!0tosatisfytheFermi-liquidrelationA11(!=0)1=[(0)][ 88 ].Forstillhighervaluesof,thesplittingincreases,indicatingthatTKgrowsasisincreased. SuchasplittingoftheKondoresonanceissimilartothatobservedforaKondoimpurityundertheinuenceofamagneticeld,whichcausesasuppressionoftheKondoeectduetoadestructionoftheKondogroundstateatlowertemperatures.Inthecaseathand,however,onendsuponexaminingtheenergyspectracalculatedbytheNRGprocedure[ 88 ]thattheKondogroundstateispreserved;therefore,theKondoeectisstilltakingplace,ataTKthatincreaseswith.Asimilarsplittingeecthasalsobeenpredictedtooccurinaside-coupledDQDdevicewithidenticalimpurities[ 85 ]. Thezero-eldsplittingoftheKondoresonanceisalsoevidencedinthelinearconductanceoftheside-coupledDQDdevice,asshowninFigure 5-3 .AsalsoseeninFigure 1-18 ,thezero-temperatureconductancevanishesattheparticle-holesymmetricpoint1=U1=2,regardlessofthevalueof;thisfeaturecanbeexplainedbyconsideringthefactthatG(T=0)/sin(22),where22isthephaseshiftofelectronsscatteringfromDot2[ 98 ].Attheparticle-holesymmetricpoint,22makesadiscontinuousjumpfromto0(regardlessofthevalueof),suchthatG(T=0)=0.Awayfromtheparticle-hole 107

PAGE 108

Dot1spectralfunctionforaside-coupledDQDdevicewith1=U1=2=0:25,2=U2=0,2=0:02,andaslabeledinthelegend.Forsucientlylarge,weobserveanovelsplittingoftheKondoresonancethatdoesnotdestroytheKondogroundstate. symmetricpoint,Dot1nolongerhasaKondoresonance,andGapproachesitsmaximumvalueG0=2e2=h.IncreasingcausestheKondoeecttostrengthen(indicatedbyanincreaseinTKaspredictedbythezero-eldsplittingoftheKondoresonance),pushingtheupturnsinGawayfromtheparticle-holesymmetricpoint[ 98 ]. IfwenowconsiderU2>0whileholding2=0,weseeinFigure 5-4 thatforsucientlysmallU2,thereisstillavisiblesplittingoftheKondoresonance,withanewparticle-holeasymmetryintroducedbythefactthattheDQDnolongersatises2=U2=2.AsU2increases,thisasymmetrybecomesmorepronounced,shiftingspectralweightfromthepositive-!sidetothenegative-!side.OnceU2becomesstrongenough 108

PAGE 109

Zero-temperatureconductancevs.Dot1energylevelforaside-coupledDQDwithU1=0:5,2=U2=0,2=0:02,andaslabeledinthelegend. (U20:1inthecaseathand),thepeakonthepositive-!sidedisappears,leavingasinglepeakthatmovestoward!=0. TheasymmetryintheDot-1spectralfunctionmanifestsitselfintheconductancetracesinFigure 5-5 .ForU2.104(notshown),thereisnonoticeabledeviationfromtheU2=0behavior,indicatingtherobustnessoftheeectiveone-impuritymodel.ForU2=0:01,wheretheDot-2interactionhasbecomecomparabletootherenergyscalesofthedevice(suchasU1;1,and2),anoticeableasymmetrydevelopsinGvs.1.UponfurtherincreaseinU2,theconductancedevelopsapeakontheorderofG0ontheright-handside.Thisiscomparabletothepeaksseenintheidentical-impurityside-coupleddeviceinFigure 1-18 .Theabsenceofaleft-handpeakismostlikelyduetotheinequivalenceofthetwodotsinthisdevice.NotethatforallcaseswithU2>0showninFigure 5-5 ,theconductancedropsallthewaytozeroatasinglevalueof1.Justasfor 109

PAGE 110

Zero-temperatureDot-1spectralfunctionforaside-coupledDQDwith1=U1=2=0:25,1=0,2=0,2=0:02,=0:067,andthevaluesofU2speciedinthelegend.TheinteractiononDot2introducesaparticle-holeasymmetrythatbecomesmorepronouncedasU2increases.ForsucientlylargeU2,theright-handpeakinA11(!)disappears. AnimportantquestiontoconsideriswhetherthedeviationsfromtheU2=0behaviorareuniquetodeviceswithU2>0oraremerelytheresultofbrokenparticle-holesymmetryonDot2.ThisissuecanbeinvestigatedbycomparingFigure 5-5 withFigure 5-6 ,whichplotstheconductanceforaside-coupledDQDwithU2=0and26=0(forwhichDot2breaksparticle-holesymmetryeventhoughitisstillnoninteracting).Forrelativelysmalldeviationsfromparticle-holesymmetry(i.e.,U2andj2jsmallerthantheotherdeviceparameters),thereisnoqualitativedierencebetweenthedevicewithU2>0and2=0andthedevicewithU2=0and26=0.Formoresignicantdeviations 110

PAGE 111

Zero-temperatureconductancevs.Dot1energylevelforaside-coupledDQDwithU1=0:5,2=0,2=0:02,=0:063,andthevaluesofU2speciedinthelegend. fromparticle-holesymmetry(i.e.,whenU2and2arecomparabletotheotherdeviceparameters),thereisstillqualitativeagreementonsomefeatures(thepeakofGG0andtheminimumofG=0)buttheconductanceforthedevicewithU2=0and26=0showsaplateauthatisnotpresentinthedevicewithU2>0and2=0.Thus,whilethereissomesimilarityinthebehaviorofthetwodevices,theydoproducenoticeablydierentconductancespectra,signifyingthattheU2>0side-coupledDQDdeviceoersanewrealmofbehaviortoexplore. 4.2.1 ,thedotsstillinteractwitheachotherviatheconductionelectrons.Theprimaryfeatureofthisspecialcaseisapairofquantumphase 111

PAGE 112

Zero-temperatureconductancevs.Dot1energylevelforaside-coupledDQDwithU1=0:5,U2=0,2=0:02,=0:063,andthevaluesof2speciedinthelegend. transitions(QPTs)thatcanbeobservedbytuning1and2.Asintheside-coupledcase,rstthebehaviorofthedotsintheU2=0limitwillbereviewed,andthenthegeneralcaseofU2>0willbeexplored. InthespeciallimitofU2=0,theeectivehybridizationforDot1[Eq.( 5{3 )]vanishesat=2withapower-law()/(2)2.Thus,whenDot2isinresonancewiththeleads(2=0),thehybridizationvanishesattheFermienergy,=0.Thepresenceofsuchapower-lawvanishingofthehybridization(orpseudogap)intheAndersonimpuritymodelisknowntointroduceapairofquantumphasetransitionsbetweenKondo-screenedphasesandalocal-momentphaseinwhichtheimpuritydegreeoffreedomremainsunquenchedatabsolutezero[ 47 48 53 56 67 81 ].IntheDQDdevice 112

PAGE 113

5-7 Figure5-7. Schematiczero-temperaturephasediagramfortheU2=0parallel-dotsystemasafunctionof1. Ifonebeginswithalarge,positive1andmovesthrough1=0tonegativevalues,thebehavioroftheDQDdeviceexhibitsanevolutionfromanempty-impurityregime(EI)toamixed-valenceregime(MV)andthentoaKondoregime(K).Eachoftheseregimes(togethertermedthe\strong-couplingregime")possessesthepropertythattheimpuritymagneticsusceptibilityimpobeysTimp!0asT!0.[However,incontrasttotheconventionalAndersonmodel,forwhichimp(T=0)generallyremainsnite|approaching0:103=TKintheKondolimit[ 30 ]|thepower-lawformoftheeectivehybridizationresultsinastrong-couplingvalueimp(T=0)=0[ 56 ].Thisaspectofthestrong-couplingregimewillbeimportantbelowwhenU2>0isconsidered.]AscanbeseeninFigures 5-8 and 5-9 ,theKondotemperatureTKdenedviatheconditionTKimp(TK)=0:0701islinearly 113

PAGE 114

Figure5-8. If1isdecreasedbelow+1c,thedeviceentersalocal-momentregime(LM)inwhichTimpapproaches1=4asT!0,indicatingthepresenceofanunscreenedlocalmomentonDot1.InthisLMregime,wecandeneacrossovertemperatureT(analogoustoTKinthestrong-couplingregime)viathe(somewhatarbitrary)criterionTimp(T)=1=5.InthevicinityoftheQPT,itisfoundthatT/+1.(ThesamelinearityisobservedforotherchoicesofthecrossovervalueofTimp.) Once1isdecreasedpasttheparticle-holesymmetricpoint1=U1=2,however,Tbeginstodecrease,untilitbecomeszeroatasecondcriticalpoint,1=1c.Similartothebehaviorintheupperhalfofthelocal-momentregime,inthevicinityoftheQPT,Tvarieslinearlywith111c.If1isdecreasedpastthissecondcriticalpoint, 114

PAGE 115

KondotemperatureTKvs.+11+1cforaU2=0parallel-coupledDQD.OtherdotparametersareU1=0:5;1=0:05;2=0;2=0:02;=0. Kondoscreeningagaintakesplace,andthedeviceentersasecondKondoregime(K).Furtherdecreaseof1willbringDot1intoamixedvalenceregime(MV)andnallyafull-impurityregime(FI),allofwhichpossessthepropertythatTimp!0asT!0.Thislowerstrong-couplingregimeissimilartotheupperone,withTK/1. TheQPTat1=1c,whichwasnotdiscussedin[ 88 ],isrelatedtotheQPTat1=+1cbyparticle-holeinversion,asevidencedbythefactthatj+1c+U1=2j=j1c+U1=2jwhen2=U2=2.(ThisruleappliestotheU2>0case,aswell.)Becausethesetwoquantumcriticalpointsarerelatedviasuchasimplesymmetry,theyexhibitmanyofthesamefeatures.Thus,Ishallfocusmydiscussiononthebehavioraround+1c. Figures 5-10 through 5-13 showthespectralfunctionA11(!)ofDot1forvariousvaluesof1intheupperstrong-couplingregimeandtheupperhalfofthelocal-momentregime.Inthestrong-couplingregime(Figures 5-10 and 5-11 ),themainfeatureofthe 115

PAGE 116

Zero-temperatureDot1spectralfunctionA11(!)vs.!>0foraparallel-coupledDQDwithU1=0:5;1=0:05;2=U2=0;2=0:02;=0,andvariousvaluesof1locatedinthestrong-couplingregime,speciedinthelegendthrough+1=1+1c. spectralfunctionisaresonanceorquasiparticlepeakcenteredat!'kBTK.Thispeak,whichisthepseudogapgeneralizationoftheKondoresonance,hasawidthproportionaltoTK,aheightinverselyproportionaltoTK,andthereforeanintegratedareathatremainsconstantasTKvanishesattheQPT.Forj!jTK,A11(!)vanishesas!2.Thispower-lawbehaviorisattributedtothepseudogapnatureofthehybridization,whichalsovanishesas!2.TherearesimilarfeaturesinA11inthelocal-momentregime(Figures 5-12 and 5-13 ),exceptthatnowtheresonanceoccursonthenegativefrequencyside,centeredat!'kBT. 116

PAGE 117

Negative-frequencyspectralfunctionscorrespondingtothedatashowninFigure 5-10 moregeneralsystemcontainsaricherphasediagram,giventhelargerparameterspacetoexplore.WewillbeginbyconsideringtheschematicphasediagraminFigure 5-14 ,whichfocusesonvarying1andU2. Thersteectofnon-zeroU2thatshouldbenotedisachangeinthelocations+1cand1cofthecriticalpoints.Figure 5-14 showsthatthecriticalpoints(atwhichTK!0)shifttowardeachotherasU2increases.Thespecialpropertyj+1c+U1=2j=j1c+U1=2jdoesnotapply,duetothefactthattheDQDdevicebreakstheparticle-holesymmetrycondition2=U2=2. ForU2U1;1;2,thephaseboundariesfollowtheform whereA+()(U1;1;2;2)isnegative(positive).ThisrelationshipisdemonstratedinFigures 5-15 and 5-16 ,whichshowtherelativechangein+1cwithincreasingU2.These 117

PAGE 118

Zero-temperatureDot1spectralfunctionA11(!)vs.!>0foraparallel-coupledDQDwithU1=0:5;1=0:05;2=U2=0;2=0:02;=0,andvariousvaluesof1locatedinthelocal-momentregime,speciedinthelegendthrough+1=1+1c. plotsshowthatA+isproportionalto(2)1butonlyweaklydependenton1.SuchbehaviorisalsofoundforA,conrmingthesimilarityofthetwoQPTs. AsobservedinFigure 5-14 ,whenU2becomescomparabletotheotherenergyscales(U1;1;and2),thecriticalpointsapproacheachothermorequicklythanseeninthelinearbehaviorforsmallerU2.AtacertainvalueofU2(approximately0:046,forthecaseillustratedinFigure 5-14 ),thecriticalpointsmerge,andthelocalmomentregimedisappearsasU2isincreasedfurther.Inthediscussiontofollow,IwillrstfocusonvaluesofU2thataresmallenoughtopreservethelocal-momentregime,andthendiscussthebehaviorofthesystemwhenthelocal-momentregimehasdisappeared. ConsiderableinsightcanbegainedintothebehavioroftheU2>0parallelDQDdevicenearitsquantumcriticalpointsbyexaminingthetemperaturevariationofTimp.Figure 5-17 showsTimpvs.Tforasequenceof1valuesandaxedU2=103.(This 118

PAGE 119

Negative-frequencyspectralfunctionscorrespondingtothedatashowninFigure 5-12 valueofU2hasbeenchosensinceitgivesrisetoverywell-separatedenergyscalesthatwillprovideinsightintobehaviorforlargerU2.) First,as1approaches+1cfromtheupperstrong-couplingregime,Timpremainsnear1=6(asintheU2=0case)untilDot1becomesKondo-screenedandTimpfallstozeroatTK,withTK/+1asbefore.AsinthecaseofU2=0,imp!0asT!0.Aroundacrossovervalue+1>+1c,aregionofnewbehaviorunfolds.ThisnewregionischaracterizedbyariseinTimpfrom1=6toward1=4.However,aslongas1>+1c,Timpneverreaches1=4,andiseventuallyscreeneddowntozero. Anothernewdevelopmentinthisregionisthatimp!0:103=TK(insteadof0)asT!0,mimickingtheconventionalsingle-impurityKondoeect[ 30 ].Thisshiftinimp(T=0)isshowninFigure 5-18 andisalsoevidencedbyhowtheredtracesinFigure 5-17 falltozerolesssteeplythantheblacktraces.Itisimportanttonotethatthechangeinimp(T=0),whileveryrapid,isacontinuousevolutionas1passesbelow+1.Thus, 119

PAGE 120

ApproximatephasediagramontheU2-1planeforaparallel-coupledDQDdevicewithU1=0:5;1=0:05;2=0,and2=0:02.Solidlinesindicatethequantumphasetransitionsseparatingthestrong-couplingandlocal-momentphases.Notetheshiftofthecriticalenergies+1cand1casU2increases,resultinginthedisappearanceofthelocal-momentregimeforU2&0:046.Dashedlines(labeled+1and1)representcrossoversfromregionsofbehavioressentiallyidenticaltothatforU2=0(1>+1and1<1)toregionsofnovelbehaviorlocatedbetweenthesolidanddashedlines.ArrowsindicatevaluesofU2forwhichTKandGareplottedagainst1inlatergures. +1istobeconsideredacrossover,andnotanewcriticalpoint.(Thereissomedegreeofarbitrarinessinthedenitionof+1.)Figure 5-19 showsthatthisnewregion+1>1>+1cexhibitsaveryrapiddropofTKwith+1,departingfromtheuniversalbehavioroftheU2=0limit. IntroductionofaninteractiononDot2notonlychangesthebehaviorinthestrong-couplingphasenearthetransition,butalsochangesthebehavioratthetransition 120

PAGE 121

Criticalvalue+1cvs.U2foraparallel-coupledDQDwithU1=0:5,2=0,=0;and1and2asspeciedinthelegend.Notethelinearvariationof+1cwithU2,andtheweakdependenceon1. 5-17 ,insteadofremainingat1=6downtozerotemperature(asitdidintheU2=0case),Timprisesupto1=4andremainsthereasT!0. As1isloweredbelow+1c,Dot1exhibitslocal-momentbehaviorinwhichTimp=1=4downtozerotemperature.IncontrasttothecaseU2=0,forU2>0,itisnotpossibletodeneacrossoverscaleTthatvanishescontinuouslyas1!+1c.Becauseofthislackofscale,thephasetransitionforU2>0isidentiedasaKosterlitz-Thoulesstypetransition[ 22 ].ThisidenticationissupportedbythefactthatthereissuchastrongdropinTKonthestrong-couplingsideofthetransition,butnosuchfeatureonthelocal-momentside.SuchahighlyasymmetricdivergenceischaracteristicofaKosterlitz-Thoulesstypetransition. 121

PAGE 122

Criticalvalue+1cvs.U2=2foraparallel-coupledDQDdevicewithU1=0:5;2=0,=0,and1=0:05. Approaching1cfromthelocal-momentphase,thereisstillnocharacteristictemperaturescaleatwhichTimp!1=4,indicatingthatthelowerquantumphasetransitionisalsoaKosterlitz-Thoulesstype.For1c>1>1,thereisagainaregionofnewbehaviorinwhichTimpinitiallyriseswithdecreasingTfrom1=6towards1=4,butiseventuallyKondo-screenedtoTimp=0.ThisscreeningsetsinaroundaKondotemperatureTKthatdropstozeroveryrapidlyas1!1cfrombelowandexhibitstheconventionalKondobehaviorofimp(T=0)=0:103=TK.For1<1,theuniversalU2=0behaviorisrecovered,withTK/1=1c1andimp(T=0)=0.(Aswithcrossing+1,thischangeiscontinuous,andso1istoberegardedasacrossoverandnotacriticalpoint.) AsillustratedinFigure 5-14 ,atacertainvalueofU2(approximately0:046,inthecaseathand),thetwocriticalpointsmergeandthelocal-momentphasedisappears. 122

PAGE 123

5-14 )fromtheU2=0pseudogapbehavior(blacktracesintheregionlabeled\1>+1")tonovelbehavior(redtracesintheregionlabeled\+1c<1<+1"). ThisdisappearanceisdemonstratedinFigure 5-20 ,whichshowsthatTKhasanon-zerominimumthatincreaseswithU2.(TheminimumvalueofTKforU2=0:05isontheorderof1090,andnotvisibleintheplot.)ItisnotedthatthewidthofthedipinTKvs.1remainsroughlyconstantwithincreasingU2beyondthedisappearanceofthelocal-momentregime.Indeed,Figure 5-14 showsthatthecrossoverenergies1|denedasthelocationsofthedownturnsinTKvs.1|seemtobecomeindependentofU2forlargeU2.Thus,eventhoughthequantumphasetransitionshavedisappearedforsucientlylargeU2,thereremainsasignatureoftheirproximityinthedipinTKvs1.Itshould 123

PAGE 124

5-19 asagreaternumberofenergyeigenstateswerekeptattheendofeachNRGiterationtoobtaintheseresults.) thereforebefeasibletoprobethequantumphasetransitionsinanexperiment,evenwhenU20isnotsatised. 5-21 showsthespectralfunctionA11(!)foraparallelDQDdevicewithU2=103.(AsinFigure 5-17 ,thisvalueofU2waschosenbecauseitcreateswell-separatedenergyscales.)For1>+1(seenintheblacktraces),themainfeatureofthespectralfunctionisaquasiparticlepeakcenteredat!'kBTK,justasitwasinthecaseoftheU2=0device.However,intheU2>0device, 124

PAGE 125

KondotemperatureTKvs.+1=1+1cforU1=0:5,1=0:05,2=0:02,2==0.ForU2=0,TKdecreaseslinearlywith+1.Thedownwarddeparturefromlinearity,whichsetsinaround1=+1,signalstheonsetofnovelU2>0behavior. ItwillnowbeseenthattheasymmetricnatureofthequantumcriticalpointsforU2>0willhaveinterestingeectsontheconductance.EvenwhenU2issucientlylargetoremovethelocal-momentphase,signaturesofthecriticalpointsarevisible.Figure 5-22 125

PAGE 126

PAGE 127

PAGE 128

PAGE 129

Zero-temperatureconductanceG=G0vs.j+1j=j1+1cjforaparallelDQDdevicewithU2asspeciedinthelegend.Tracesmarkedby1>+1careinthestrong-couplingregimeandmaybecomparedwithtracesinFigure 5-19 .OtherdeviceparametersareasinFigure 5-22 adip(peak)at1=(+)1,withaplateauregioncorrespondingtothedipinTKvs.1.(Theroughregionaround1=0:35intheU2=0:05traceistheresultofanumericalinstabilityassociatedwithbeingnearaKosterlitz-Thoulesstransition.)ItisinterestingtonotethattheshapesoftheconductancetracesforU2>1;2resemblethosefortheside-coupledcasewithU2>2;inFigure 5-5 Finally,asmentionedearlier,itisalsopossibletoobservethesequantumphasetransitionsbyvarying2.Whileacompletephasediagramof1,2,andU2isbeyondthescopeofthepresentstudy,itisinsightfultoexamineacoupleofspeciccases. Figure 5-25 showsthezero-temperatureconductanceas2issweptinaparallelDQDdevicewithaparticle-holesymmetricDot1(1=U1=2).Becauseofthisparticle-holesymmetryonDot1,thetracesinFigure 5-25 aresymmetricabout2=U2=2.For 129

PAGE 130

Zero-temperatureconductanceG=G0vs.1foraparallelDQDdevicewithU2asspeciedinthelegend.ThetracesmaybecomparedwiththoseinFigure 5-20 .OtherdeviceparametersareU1=0:5;1=0:05;2=0;2=0:02,and=0.TheroughfeatureinthesolidredtraceisduetoanumericalinstabilityassociatedwiththeKosterlitz-Thoulesstransition. SimilarbehaviorisalsofoundwhenDot1isparticle-holeasymmetric.Figure 5-26 showsGvs.2with1+1cforeachvalueofU2.Thereisaverypronouncedasymmetryintheseresults,butthesamegeneralfeaturesapplyasintheparticle-holesymmetric 130

PAGE 131

PAGE 132

PAGE 133

ItisclearthatsystemsofmultiplemagneticimpuritiesandtheiranalogousmanifestationsinQDsystemsarenotunderstoodexhaustively.Whilemuchtheoreticalworkhasbeenperformedandnovelexperimentshavebeendesignedtoprobetwo-impuritysystems,therearestillmanymorepossibilitiestoconsider.Inthiswork,Ihavedemonstratedthatanumberofinterestingbehaviors|suchasthecompetitionbetweenmagneticorderingandtheKondoeect,asplittingoftheKondoresonance,andapairofquantumphasetransitions|shouldbeexperimentallyobservableinSTMconductancestudiesofapairofmagneticadatoms,andinelectricaltransportthroughhighlyasymmetricDQDs. Forexample,theresultsofSection 4.2.3 consideredthelimitingcaseofanSTMtippositioneddirectlyoveroneoftheimpurities.ToconsideranSTMstudyperformedwiththetipatavariableposition,itshouldbepossibletoutilizeanextensionofthemethoddevelopedbyBorda[ 96 ]tostudyspatialcorrelationsinthesingle-impurity 133

PAGE 134

2.2 )withthesecondimpuritysite\empty"(i.e.,2;U2;2=0),permittingthecomputationofpropertiesatanydistancefromtheimpurityofstudy(byvaryingtheimpurityseparationR),therebyenablingatheoreticalpredictionoftheKondoscreeninglengthlK.Theextensionofthismethodtoatwo-impuritySTMstudywithageneralizedtippositionwouldthusinvolvesettingupathree-impuritycalculationwiththethirdsite\empty"andcalculatingthematrixelementsoftheimpurity-3operatordy3atsuccessiveiterations.Giventhegreatcomplexityofthree-impuritymodels[ 61 84 ],suchNRGcalculationswouldbenetfromtheparallel-processingtechniquesdescribedinChapter 3 .Studiesofsingleimpurities|suchasdepictedinFigures 1-8 and 1-9 |indicatethattheFanolineshapeintheconductanceonlychangesquantitativelyastheSTMtipismoved.Incontrast,inageneralizedtwo-impuritySTMstudy,itisnecessarytoconsiderthreetunnelingpaths,whichmaygiverisetonovelinterferenceeects,producingamorecomplicatedversionoftheFanolineshape. InChapter 4 ,theimpuritiesunderconsiderationalsoexhibitednodissipativeeects,aswouldbecausedbythepresenceofphononsornoisewhichwouldintroducedecoherencetotheelectrontunnelingpaths,therebymodifyingtheFanolineshape.Thesescenariosmaybemodeledbycouplingtheimpuritiestoabosonicbath.Suchcouplinghasbeenstudiedinsingle-impuritysystemsusingtheNRG[ 50 78 97 ],andisknowntoproduceanumberofnoveleects,includingaquantumphasetransitionaccessiblebytuningthecouplingbetweentheimpurityandthebosonicbath.Thestudyofatwo-impuritysystemcoupledtoabosonicbathhasnotbeenapproachedwiththeNRGduetothecomputationalcomplexity;thus,suchNRGstudieswouldbenetfromtheparallelizationtechniquesdescribedinChapter 3 134

PAGE 135

PAGE 136

FurtherstudyoftheparallelDQDcongurationiscalledfor,includingthedevelopmentofphasediagramsthatillustratestheeectsofchangesinU1,1,and2,whichwereleftxedinChapter 5 .Itwillbeusefultoseetheeectsthatsuchchangesproduceonthebehaviorofthecriticalpoints1cand2c,andalsotheeectsthattheyproduceonthevalueofU2forwhichthelocal-momentphasedisappears. Furtherworkisalsoneededtounderstandthephysicsbehindthesumruleexhibitedbythediscontinuitiesintheconductanceatthecriticalpoints.Todevelopthisunderstanding,itwillbenecessarytoexamineotherdeviceproperties,suchasthephaseshiftsij(i;j=1;2)ofelectronsthattunnelthroughtheDQDdevice.Itwillalsobenecessarytounderstandtheconditionsunderwhichthissumruleisobeyed. Tomakepredictionsthatareexperimentallyrelevant,itwillalsobenecessarytoexploretheeectsofnon-zerotemperatureontheconductancesignaturesreportedforthetwoDQDcongurationsinChapter 5 .CalculatingthespectralfunctionsintheLandauerformulaatnitetemperatureisnotastraightforwardtask,butcanbehandledtoreasonableaccuracy[ 44 93 ].Ingeneral,ithasbeenfoundthattheconductancespectrumremainsqualitativelythesamefor0T.TKandissignicantlydierentforT>TK[ 85 98 ].Intheparallelconguration,itwillbeparticularlyinterestingtoseetheeectsofnitetemperatureonthediscontinuitiesobservedinthezero-temperatureconductanceG. 136

PAGE 137

137

PAGE 138

[1] W.J.deHaas,J.deBoer,andG.J.vanDenBerg,Physica1,1115(1933). [2] C.Zener,Phys.Rev.81,440(1951). [3] A.H.Wilson,TheTheoryofMetals(CambridgeUniversityPress,Cambridge,England,1932). [4] M.A.RudermanandC.Kittel,Phys.Rev.96,99(1954). [5] K.Yosida,Phys.Rev.106,5(1957). [6] R.Landauer,IBMJ.Res.Develop.1,233(1957). [7] P.W.Anderson,Phys.Rev.124,41(1961). [8] U.Fano,Phys.Rev.124,1866(1961). [9] M.Sarachik,E.Corenzwit,andL.D.Longinotti,Phys,Rev.A135,1041(1964). [10] J.Kondo,Prog.Theor.Phys.32,1(1964). [11] A.A.Abrikosov,Physics2,5(1965). [12] J.R.SchrieerandP.A.Wol,Phys.Rev.149,491(1966). [13] K.Yosida,Phys.Rev.147,223(1966). [14] H.IshiiandK.Yosida,Prog.Theor.Phys.38,61(1967). [15] C.N.Yang,Prog.Theor.Phys.55,67(1967). [16] P.W.Anderson,Phys.Rev.164,352(1967). [17] P.W.AndersonandG.Yuval,Phys.Rev.Lett.23,89(1969). [18] R.Landauer,Phil.Mag.21,863(1970). [19] P.W.Anderson,G.Yuval,andD.R.Hamann,Phys.Rev.B1,4464(1970). [20] G.YuvalandP.W.Anderson,Phys.Rev.B1,1522(1970). [21] P.W.Anderson,J.Phys.C3,2439(1970). [22] J.M.KosterlitzandD.J.Thouless,J.Phys.C6,1181(1973). [23] K.G.Wilson,Rev.Mod.Phys.47,773(1975). [24] N.W.AshcroftandN.D.Mermin,SolidStatePhysics(SaundersCollege,Philadelphia,1976). [25] J.H.Jeerson,J.Phys.C10,3589(1977). 138

PAGE 139

F.D.M.Haldane,Phys.Rev.Lett.40,416(1978). [27] P.NozieresandA.Blandin,J.Phys.(Paris)41,193(1980). [28] N.Andrei,Phys.Rev.Lett.45,379(1980). [29] P.B.Wiegmann,Sov.Phys.JETPLett.31,392(1980). [30] H.R.Krishna-murthy,J.W.Wilkins,andK.G.Wilson,Phys.Rev.B21,1003(1980). [31] H.R.Krishna-murthy,J.W.Wilkins,andK.G.Wilson,Phys.Rev.B21,1044(1980). [32] C.Jayaprakash,H.R.Krishna-murthy,andJ.W.Wilkins,Phys.Rev.Lett.47,737(1981). [33] J.E.HirschandR.M.Fye,Phys.Rev.Lett.56,2521(1986). [34] B.A.Jones,Ph.D.thesis,CornellUniversity,1987(unpublished). [35] B.A.JonesandC.M.Varma,Phys.Rev.Lett.58,843(1987). [36] B.A.Jones,C.M.Varma,andJ.W.Wilkins,Phys.Rev.Lett.61,125(1988). [37] R.M.FyeandJ.E.Hirsch,Phys.Rev.B40,4780(1989). [38] A.J.Millis,B.G.Kotliar,andB.A.Jones,inFieldTheoriesinCondensedMatterPhysics,editedbyZ.Tesanovic(Addison-Wesley,RedwoodCity,CA,1990),pp.159-166. [39] O.Sakai,Y.Shimizu,andT.Kasuya,SolidStateCommun.75,81(1990). [40] T.Saso,Phys.Rev.B44,450(1991). [41] O.SakaiandY.Shimizu,J.Phys.Soc.Jap.61,2333(1992). [42] O.SakaiandY.Shimizu,J.Phys.Soc.Jap.61,2348(1992). [43] I.AeckandA.W.W.Ludwig,Phys.Rev.Lett.68,1046(1992). [44] T.A.Costi,A.C.Hewson,andV.Zlatic,J.Phys.:Condens.Matter6,2519(1994). [45] I.Aeck,A.W.W.Ludwig,andB.A.Jones,Phys.Rev.B52,9528(1995). [46] C.J.Chen,IntroductiontoScanningTunnelingMicroscopy(Springer,Berlin,1996). [47] C.Gonzalez-BuxtonandK.Ingersent,Phys.Rev.B54,R15614(1996). [48] K.Ingersent,Phys.Rev.B57,11936(1996). 139

PAGE 140

J.B.Silva,W.L.C.Lima,W.C.Oliveira,J.L.N.Mello,L.N.Oliveira,andJ.W.Wilkins,Phys.Rev.Lett.76,275(1996). [50] T.A.CostiandC.Kieer,Phys.Rev.Lett.76,1683(1996). [51] W.H.Press,S.A.Teukolsky,W.T.Vetterling,B.P.Flannery,NumericalRecipesinFortran77(CambridgeUniversity,Cambridge,England,1996). [52] K.Yoshida,TheoryofMagnetism(Springer,NewYork,1996). [53] R.Bulla,Th.Pruschke,andA.C.Hewson,J.Phys.Condens.Matter9,10463(1997). [54] A.C.Hewson,TheKondoProblemtoHeavyFermions(CambridgeUniversityPress,Cambridge,England,1997). [55] D.Goldhaber-Gordon,H.Shtrikman,D.Mahalu,D.Abusch-Magder,U.Meirav,andM.A.Kastner,Nature391,156(1998). [56] C.Gonzalez-BuxtonandK.Ingersent,Phys.Rev.B57,14254(1998). [57] V.Madhavan,W.Chen,T.Jamneala,M.F.Crommie,andN.S.Wingreen,Science280,567(1998). [58] J.Wu,B.Gu,H.Chen,W.Duan,andY.Kawazoe,Phys.Rev.Lett.80,1952(1998). [59] J.Li,W.-D.Schneider,R.Berndt,andB.Delley,Phys.Rev.Lett.80,2893(1998). [60] W.Chen,T.Jamneala,V.Madhavan,andM.F.Crommie,Phys.Rev.B60,R8529(1999). [61] B.C.Paul,Ph.D.Dissertation,UniversityofFlorida,2000(unpublished). [62] J.Nygard,D.H.Cobden,andP.E.Lindelof,Nature408,342(2000). [63] S.Sasaki,S.deFranceschi,J.M.Elzerman,W.G.vanderWiel,L.P.Kouwenhoven,M.Eto,andS.Tarucha,Nature405,6788(2000). [64] L.KouwenhovenandL.Glazman,Phys.World14,33(2001). [65] H.Jeong,A.M.Chang,andM.R.Melloch,Science293,2221(2001). [66] K.Kang,S.Y.Cho,J.-J.Kim,andS.C.Shin,Phys.Rev.B63,113304(2001). [67] M.VojtaandR.Bulla,Phys.Rev.B65,014511(2001). [68] W.Hofstetter,J.Knig,andH.Schoeller,Phys.Rev.Lett.87,156803(2001). [69] B.KubalaandJ.Knig,Phys.Rev.B65,245301(2002). 140

PAGE 141

V.Madhavan,T.Jamneala,K.Nagaoka,W.Chen,J.L.Li,S.G.Louie,andM.F.Crommie,Phys.Rev.B66,212411(2002). [71] R.AguadoandD.C.Langreth,Phys.Rev.B67,245307(2003). [72] V.M.Apel,M.A.Davidovich,E.Anda,C.A.Busser,andG.Chiappe,MicroelectronicsJournal34,729(2003). [73] P.S.CornagliaandC.A.Balseiro,Phys.Rev.B67,205420(2003). [74] M.Garst,S.Kehrein,T.Pruschke,A.Rosch,andM.Vojta,Phys.Rev.B69,214413(2004). [75] C.A.Busser,G.B.Martins,K.A.Al-Hassanieh,A.Moreo,andE.Dagotto,Phys.Rev.B70,245303(2004). [76] N.J.Craig,J.M.Taylor,E.A.Lester,C.M.Marcus,M.P.Hanson,andA.C.Gossard,Science304,565(2004). [77] J.C.Chen,A.M.Chang,andM.R.Melloch,Phys.Rev.Lett.92,176801(2004). [78] P.S.Cornaglia,H.Ness,andD.R.Grempel,Phys.Rev.Lett.93,147201(2004). [79] A.Fuhrer,T.Ihn,K.Ensslin,W.Wegscheider,andM.Bichler,Phys.Rev.Lett.93,176803(2004). [80] V.L.CampoandL.N.Oliveira,Phys.Rev.B70,153401(2004). [81] L.FritzandM.Vojta,Phys.Rev.B70,214427(2004). [82] P.Simon,R.Lopez,andY.Oreg,Phys.Rev.Lett.94,086602(2005). [83] R.Leturcq,L.Schmid,K.Ensslin,Y.Meir,D.C.Driscoll,andA.C.Gossard,Phys.Rev.Lett.95,126603(2005). [84] K.Ingersent,A.W.W.Ludwig,I.Aeck,Phys.Rev.Lett.95,257204(2005). [85] P.S.CornagliaandD.R.Grempel,Phys.Rev.B71,075305(2005). [86] G.H.Ding,C.K.Kim,andK.Nahm,Phys.Rev.B71,205313(2005). [87] L.G.G.V.DiasdaSilva,S.E.Ulloa,andT.V.Shahbazyan,Phys.Rev.B72,125327(2005). [88] L.G.G.V.DiasdaSilva,N.P.Sandler,K.Ingersent,andS.E.Ulloa,Phys.Rev.Lett.97,096603(2006). [89] A.J.Heinrich,J.A.Gupta,C.P.Lutz,andD.M.Eigler,Science298,1381(2006). [90] A.J.Heinrich,C.P.Lutz,andC.F.Hirjibehedin,e-J.Surf.Sci.Nanotech.4,384(2006). 141

PAGE 142

P.Simon,J.Salomez,andD.Feinberg,Phys.Rev.B73,205325(2006). [92] R.ZitkoandJ.Bonca,Phys.Rev.B74,045312(2006). [93] R.Bulla,T.Costi,andT.Pruschke,arXiv:cond-mat/0701105v1(2007). [94] L.G.G.V.DiasdaSilva,N.P.Sandler,K.Ingersent,andS.E.Ulloa,Phys.Rev.Lett.99,209702(2007). [95] R.M.Potok,I.G.Rau,H.Shtrikman,Y.Oreg,andD.Goldhaber-Gordon,Nature446,167(2007). [96] L.Borda,Phys.Rev.B75,041307(R)(2007). [97] M.T.GlossopandK.Ingersent,Phys.Rev.B75,104410(2007). [98] L.G.G.V.DiasdaSilva,K.Ingersent,N.P.Sandler,andS.E.Ulloa,arXiv:cond-mat/0804.0805(2008). [99] http://www-unix.mcs.anl.gov/mpi/index.htm [100] http://www.netlib.org/lapack/index.html [101] http://www.netlib.org/scalapack/index.html 142

PAGE 143