
Citation 
 Permanent Link:
 http://ufdc.ufl.edu/AA00018872/00001
Material Information
 Title:
 A classification of trees and applications of topology and graph theory to neurosurgery
 Creator:
 Yi, TaeIl, 1962
 Publication Date:
 2000
 Language:
 English
 Physical Description:
 viii, 105 leaves : ill. ; 29 cm.
Subjects
 Subjects / Keywords:
 Algorithms ( jstor )
Brain neoplasms ( jstor ) Dosage ( jstor ) Geometric centers ( jstor ) Integers ( jstor ) Mathematical sequences ( jstor ) Mathematics ( jstor ) Radiosurgery ( jstor ) Tumors ( jstor ) Vertices ( jstor ) Dissertations, Academic  Mathematics  UF ( lcsh ) Mathematics thesis, Ph.D ( lcsh )
 Genre:
 bibliography ( marcgt )
theses ( marcgt ) nonfiction ( marcgt )
Notes
 Thesis:
 Thesis (Ph.D.)University of Florida, 2000.
 Bibliography:
 Includes bibliographical references.
 General Note:
 Printout.
 General Note:
 Vita.
 Statement of Responsibility:
 by TaeIl Yi.
Record Information
 Source Institution:
 University of Florida
 Holding Location:
 University of Florida
 Rights Management:
 Copyright [name of dissertation author]. Permission granted to the University of Florida to digitize, archive and distribute this item for nonprofit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
 Resource Identifier:
 024914146 ( ALEPH )
45805641 ( OCLC )

Downloads 
This item has the following downloads:

Full Text 
A CLASSIFICATION OF TREES AND
APPLICATIONS OF TOPOLOGY AND GRAPH THEORY TO
NEUROSURGERY
By
TAEIL YI
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
2000
Copyright 2000
by
TAEIL YI
TO
My parents, BeomYoung & BongWol,
who show me how to love,
My wife, MoonSil,
who shows me what is love,
and
My children, David & Peter,
who are the reasons why I love
ACKNOWLEDGEMENTS
I would like to express my sincere appreciation to the members of my supervi
sory committee, Dr. Beverly Brechner, Dr. Philip Boyland, Dr. James Keesling, Dr.
LiChien Shen, and Dr. Mary Kantowski, for their patience, knowledge, concern, and
especially for showing me how to study and teach. I would especially like to thank
my advisor, Dr. Beverly Brechner, who accepted me and gave me a chance to work
with her. I also want to express my appreciation for all her effort and all that she has
done for me, not only in my major field, but also in my life. I would like to thank
Matthew Harvey who brought me some wonderful ideas and wrote the Mathematica
programs. I would like to thank Dr. Francis Bova who brought me these questions
which were 'new world' to me. I would like to thank Dr. Meeks for helpful conver
sations, and Dr. Thomas Wagner who gave me the information about neurosurgery
that I did not know before. I would like to thank Dr. Yunmei Chen who gave me
inspiration. I would like to thank Dr. Miklos Bona who showed me some insights of
combinatorics. Finally, I would like to thank the Department of Mathematics that
allowed me to be here and meet all these wonderful people.
TABLE OF CONTENTS
ACKNOWLEDGEMENTS ............................ iv
ABSTRACT ... .. .. .. ... .. .. .. .. .. .. .. .. ... vii
CHAPTERS
1 INTRODUCTION .............................. 1
2 AN AUTOMATED SPHERE PACKING PLAN FOR BRAIN TL'UMORS 5
2.1 Introduction . .. . . 5
2.2 Linac . . . . .. 5
2.3 Procedure of Radiosurgery ........................ 6
2.4 Making a Sphere by Arcs of Beams .................... 10
2.5 Sphere Packing Plan ......................... 10
2.6 Shelling Algorithm .......................... 15
3 A CLASSIFICATION OF UNLABELED TREES ............ 25
3.1 Introduction .. . . . .. 25
3.2 Definitions .. .. .. .. .. .. .. ... .. .. .. .. ... 26
3.3 Vertex Sequence ........................... 28
3.4 Degree Sequence ........................... 36
3.5 Perfect Sequence and Main Theorem .................. 44
3.6 Obtaining Perfect Sequences . . ... .. 45
3.7 Extended Partitions and the Algorithm . .... ..53
4 TREES AS INVARIANTS FOR BRAIN TUMOR SHAPES ...... 59
4.1 Introduction . . . ... .. 59
4.2 Definitions and Some Properties . ... .. 60
4.3 Graphs for Sphere Packing Plans . ... .. 64
4.4 Order in the Vertex Set . . ... .. 65
4.5 Maximal Tree from a Graph . . ... .. 71
4.6 Graph from a Maximal Tree . . ... .. 81
4.7 Algorithm for Our Maximal Tree . ... .. 81
APPENDICES
A SHELLING PROGRAM FOR SPHERE PACKING .......... 82
A.1 Program for Data Reading ....................... 82
A.2 Program for Center Searching ...................... 83
B PROGRAM FOR THE MAXIMAL TREE OF A GRAPH ...... 91
C PROGRAM FOR CLASSIFYING TREES BY SEQUENCES ..... 93
C.1 Program .. .. .. ... .. .. .. ... .. .. .... 93
C.2 Some Program Results ....................... 97
D LABELED GRAPHS AND TREES ...................... 102
D.1 Number of Labeled Graphs and Labeled Trees ......... 102
D.2 Sketch of the Construction of a Priifercode ............. 102
D.3 Reconstruction of a Tree from a Priifercode ........... 103
REFERENCES ................................... 105
BIOGRAPHICAL SKETCH ............................ 105
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
A CLASSIFICATION OF TREES AND
APPLICATIONS OF TOPOLOGY AND GRAPH THEORY TO
NEUROSURGERY
By
TaeIl Yi
August 2000
Chairman: Dr. Beverly L. Brechner
Major Department: Mathematics
The work of this dissertation was motivated by our work with a medical group
at the Brain Institute at the University of Florida. This group was led by Dr. Francis
Bova, who raised the following questions:
(1) Can you automate the timeconsuming sphere packing process for our "multiple
isocenters method of stereographic radiosurgery" for brain tumors? (This surgery is
noninvasive, radiation surgery to destroy brain tumors.)
(2) Can you classify the shapes of brain tumors?
In this dissertation, we construct some algorithms and programs to (partially)
answer these questions. Our major (joint) results are described below.
Our first result is an algorithm and a program that produce a unique sphere
packing plan for any given brain tumor shape, filling the shape with different size
spheres in a unique way. This process automates the 'sphere packing' part of the
surgery plan. Additional code by the medical group was necessary to allow for physi
cian input in certain critical cases. It was then possible for the medical group to
incorporate this into their surgery plan, semiautomating it, and significantly reduc
ing the surgeons' surgery planning time.
Our second result is a complete classification of unlabeled ntrees. We assign
to each ntree, a sequence of n nonnegative integers satisfying certain properties; and
from each such sequence, we can recover the ntree assigned to it. In fact, we present
both an algorithm and code to produce not only the sequences, but also pictures of
the corresponding trees.
Our final result is a first attempt at a topological classification of tumor shapes.
We first use our sphere packing plan to construct a graph representing the tumor.
We then have an algorithm and program which uniquely select a maximal tree in the
graph. Thus we can assign a unique tree to each tumor shape. However, different
graphs may produce the same tree. Our tree classification allows us to compare
two tumor shapes very easily, by just looking at their corresponding sequences. In
particular, different sequences come from different sphere packing plans, but the
converse does not hold.
CHAPTER 1
INTRODUCTION
The work of this dissertation was motivated by two questions raised by Dr.
Francis Bova, of the Brain Institute, Shands Hospital, and the Department of Neuro
surgery, all at the University of Florida. It turns out the differentsounding questions
are very much related and are tied together naturally in this thesis. In fact, our
results for the second question depend upon our results for the first question. Dr.
Bova led a medical group, which joined with our mathematics group, for a number
of discussions over a two year period. This resulted, for us, in both a joint medical
paper [11] and in this dissertation. We also want to reference the dissertation [10] of
Thomas Wagner, Dr. Bova's medical physics student, who worked with us.
We expect to have three mathematics publications from this dissertationfrom
Chapters 2, 3, and 4, respectively. The paper resulting from Chapter 2 will be joint
with Professor Beverly Brechner, Professor Yunmei Chen, and Dr. Thomas Wagner.
The papers resulting from Chapters 3 and 4 will be joint with Matthew Harvey, an
undergraduate student working with Professor James Keesling on his honors thesis.
Further, we anticipate an additional paper, to be joint with Yunmei Chen and her
student, Dr. Stacey Chastain. This anticipated work will be an analysis of the tumor
shapes of approximately 1000 interesting tumor radiation surgeries done by Bova and
his colleagues at the University of Florida.
Dr. Bova's questions were the following:
1. Is it possible to automate the medical group's "multiple isocenters method of
stereographic radiosurgery" to destroy brain tumors?
2. Is it possible to classify brain tumor shapes?
In this dissertation, we develop some algorithms and programs related to these
questions. In stereographic radiosurgery, which is noninvasive surgery, radiation is de
livered by means of collimatorscylindrical metal tubes, with cylindrical holes through
them, through which the radiation flows. The goal of stereotactic radiosurgery for a
brain tumor is to deliver the desired dosage to the target, and only the target. This
is not possible in reality. So they do the next best thing, which is to deliver enough
dosage to the target, to avoid as much normal tissue as possible, and to deliver as
little radiation as possible to whatever normal tissue must be affected. There are two
additional important criteriadose homogeneity and dose conformality. That is, we
do not want 'hot spots,' which have been experimentally determined to cause com
plications; and we do want rapid falloff of dose levels outside the actual tumor. One
of several such radiation surgery methods is called the 'Multiple Isocenter Method.'
This involves filling the tumor image with spheres of different sizes, until the image is
best filled up. In this dissertation, we call such a plan, a 'Sphere Packing Treatment
Plan.'
In Chapter 2, we develop both an algorithm, which is called a 'shelling process'
(or grassfiree process' in medical terminology), and a CH+ program to identify the
centers and sizes of spheres for this sphere packing treatment plan. For any given
brain tumor, our program gives us a unique sphere packing plan. This sphere packing
plan can be represented numerically as a list of parameters of the spheres, namely
the centers and radii of the spheres. Based on this algorithm, the medical group
developed a proprietary program for an automated sphere packing treatment plan,
including dosage distribution, and other things (see Wagner et al.[ll]). Since this
program always produces the same plan for the same tumor, it can be used as a
'benchmark' against which to compare other treatment plans.
In Chapter 3, we obtain a code to denote an unlabeled tree by a unique,
ordered, nonnegative integer sequence satisfying certain properties. We call such
sequences, perfect sequences. That is, we obtain a complete classification of unlabeled
ntrees by means of the perfect sequences of n nonnegative integers. Thus we have our
main theorem of this chapter, the Tree Classification Theorem, (Theorem 3.5.5): For
any positive integer n, let T(n) be the set of unlabeled ntrees and P(n) the set of
perfect sequences of ntrees. Then there is a onetoone correspondence between T(n)
and P(n). We show how to assign a perfect sequence to a given tree, and conversely,
how to reconstruct a tree from a given perfect sequence. We also include an algorithm
and a Mathematica program to produce not only all the perfect nsequences, for any
given positive integer n > 3, but also pictures of the corresponding trees. By means of
this program, we get our classification of unlabeled ntrees and also the total number
of ntrees.
In Chapter 4, we make a first attempt at a topological shape classification of
tumorsactually of the sphere packing plan associated with the tumor shape. We first
assign to each tumor a unique graph, and from that, we determine a unique tree. We
obtain the graph by assigning a vertex for each sphere, and an edge between each pair
of vertices representing adjacent spheres. This produces a unique connected graph
corresponding to each sphere packing plan. Then, by using the notion of cutvertex, we
give an order to the vertex set. We then use this order to build a unique maximal tree
contained in the graph. Next, we develop an algorithm and a Mathematica program
to assign a unique tree to each sphere packing plan. This program produces a list of
edges of the tree, making it clear which edges are incident.
Using our results of Chapter 3, we can now assign exactly one perfect sequence
to each tumor. However, it may happen that more than one tumor shape (or graph)
gives rise to the same tree, and therefore to the same perfect sequence. Thus, we do
not yet have a real classification of tumor shapes. However, if two perfect sequences
are distinct, then their corresponding trees, and therefore their respective correspond
ing graphs and sphere packing plans, are also distinct. Therefore we may consider
4
their shapes to be distinct. Thus, we have our main result of Chapter 4, Theorem
4.5.6: The perfect sequences are invariants of the shapes of arbitrary brain tumors.
That is, if two brain tumors are represented by distinct perfect sequences, then their
corresponding trees are not isomorphic, and therefore their respective graphs and
sphere packing plans are not isomorphic. Thus, we may consider their shapes to be
distinct.
We refer the reader to Douglas[l], Gross[3], and Harary[5] for graph theory.
We also refer the reader to Hall[4], Riordan[7] and Stanley[9] for combinatorics.
CHAPTER 2
AN AUTOMATED SPHERE PACKING PLAN FOR BRAIN TUMORS
2.1 Introduction
The goal of stereotactic radiosurgery for a brain tumor is to deliver the desired
dosage to the target, and only the target. This is not possible in reality. So they do
the next best thing, which is to deliver enough dosage to the target, to avoid as much
normal tissue as possible, and to deliver as little radiation as possible to whatever
normal tissue must be affected. There are two additional important criteriadose
homogeneity and dose conformality. That is, we do not want 'hot spots,' which
have been experimentally determined to cause complications; and we do want rapid
falloff of dose levels outside the actual tumor. One of several such radiation surgery
methods is called the 'Multiple Isocenter Method.' This involves filling the tumor
image with spheres of different sizes, until the image is best filled up. This noninvasive
method of surgery, namely by using radiation, relies on a piece of equipment called the
Linear Accelerator (or simply, Linac). In this chapter, we show how to automate the
construction of a sphere packing plan for the tumor image. Most of the information
in this chapter about treatment of a brain tumor is taken from Friedman et al.[2].
See [2] for further reference.
2.2 Linac
According to Friedman et al.[2], the linear accelerator is a complex machine
capable of producing Xrays. A large amount of energy is generated by the power
supply, which then powers the filament shown. This causes electrons to be emitted
by the filament, which are in turn accelerated to higher energies using a (microwave)
wave guide. The electrons are then changed in direction by the magnet so that they
impact on a heavy metal alloy target. This results in Xray production that can then
be collimated or shaped by both primary and secondary collimators within the linear
accelerator head. This beam is further collimated for radiosurgery by the tertiary
radiosurgery collimator. (See Figure 21 and Figure 22.)
The Linac is mounted on a rotating gantry such that the beam has a center of
rotation about 1.5m above the floor. Usually, the isocenter accuracy is defined within
a 2mm sphere. Because stereotactic radiosurgery depends on optimized accuracy, an
improved system was designed at the University of Florida, by adding a set of bearings
to the stereotactic collimator system and under the patient table. As a result, this
new system achieves mechanical accuracy within 0.2mm0.1mm for defining the
treatment isocenter of beam delivery.
The tertiary collimators are generally circular and allow improved centering
of the treatment beam. The sizes of these collimators are from 5 to 40mm in 2 to
5mm increments.
2.3 Procedure of Radiosurgery
The procedure of radiosurgery is separated into 5 steps as follows.
(1) Head ring attachment
(2) CT scan
(3) Image Processing
(4) Planning
(5) Treatment delivery
For more detailed information, refer to Friedman et al.[2].
(1) Head ring attachment
The stereotactic radiosurgery treatment requires attachment of a stereotactic
head ring. This ring provides accurate information about the position and the shape
Heavy alloy target
Primary (Internal)
collimators
Flattening filter
Secondary (Adjustable)
colimators
Tertiary (Radiosurgery)
collimators
Photon beam for
radiourgery
Floor stand with treatment arm above
Figure 21. System of Linac (Friedman et al.[2])
Figure 22. Operation of Linac (Friedman et al.[2])
of brain tumors from the computer tomography (CT). This ring also helps to insure
that the treatment target is accurately placed at the precise isocenter of the Linac.
(2) CT scan
After attaching the head ring, a series of cut filming is taken. Eighty to 150
CT slices are taken per patient. For radiosurgical treatment planning, getting the
accurate size and shape of the target is very important. Underestimation of the target
size may result in treatment failure. Overestimation of size results in the inclusion of
normal brain tissue within the treatment volume. Misrepresentation of an irregular
target shape may lead to radiation damage of normal brain tissue. The accuracy
of CT scanning depends on the size of the pixels that make up the scan image.
The lateral and AP accuracy of scanning is approximately 0.6mm. But the axial
dimension is determined by slice thickness of the CT scan, which is 1mm in general.
The physicians use the socalled BrownRobertsWells coordinatesor, simply BRW
coordinateswhich are called lateral (Lat), anteriorposterior (AP), and axial (Ax)
coordinates. These are the x, y, and z coordinates, respectively. The direction of
the lateral coordinate is from the left hand side ear to the right hand side ear of the
patient. So the lateral view (or sagittal) is similar to the side view of the head, that
is, views of slices along the lateral axis. This is the same as yzplane views. The
direction of the AP coordinate is from the back side of the head to the face. So the
AP view (or coronal) is similar with the front view of the head, that is, views of slices
along the AP axis. This is the same as xzplane views. The direction of the axial (or
vertical) coordinate is from the neck to the top of the head. So the axial view (or
transaxial) is similar to the sky view of the head, that is, views of slices along the
axial axis. This is the same as xyplane views. (See Figure 24.)
(3) Image Processing
An MRI scan has been done the day before surgery. Using a mouselike device,
the CT and MRI images are transferred to the computer screen and fused. Then,
using geometric equations, the computer determines the lateral, AP, and vertical
stereotactic coordinates of each point in each CT slice.
(4) Planning
Physicians use a mouse and their own judgment to 'eyeball' the location and
the size of spheres for the treatment plan.
(5) Treatment delivery
The geometric sphere packing plan, developed as in (4), is combined with a
proprietary dose planning program, and the resulting treatment plan is delivered by
the Linac.
2.4 Making a Sphere by Arcs of Beams
By varying the angle of the gantry and the angle of the table, one can deliver
a radiation beam to the target from any angle within the range of the rotation. The
shape of the common intersection of an arc of beams passing through one isocenter is
a sphere. The neurosurgeons deliver a series of arcs (usually 5 or 9 arcs) to produce
a single isocentered sphere shape. For an ellipsoidal target, they use fewer numbers
of arcs to make a single isocentered ellipsoid shape. (See Figure 23, 24 and 25.)
So, if the target shape is very close to a sphere or an ellipsoid, then the
treatment plan is relatively easy compared to an irregularly shaped target. In that
case, we need to create a geometric treatment plan.
2.5 Sphere Packing Plan
As seen in the previous section, the physicians know how to irradiateto
(eventually)destroy tumors which are shaped like spheres or ellipsoids. For a non
spherical shape of tumor, they try to fill the target with several spheres of different
sizes. This is called the 'sphere packing' treatment plan.
After finding a sphere packing plan, they treat each sphere separately as de
scribed in the previous section. So, multiple isocenter radiosurgery planning includes
Figure 23. One isocenter plan with five arcs of radiation (Wagner et al.[ll])
(270')
(310*)
(330")
(350) I
(501
(30')
A (lo)
I H G FED C BA
Figure 24. One isocenter plan with nine arcs of radiation (Friedman et al.[2])
13
A F
,,.. ,,"" : ,_ _._ .... **\ '*,f
%1% Ii% l II I
.' ./ .,,,, i, ,,,
> /" "l '. *./,
.,TARGET : _
I,
B
No
TARGET .,
Figure 25. Plan for a sphere shape and an ellipsoidal shape (Friedman et al.[2])
"' i \ ': "' "* "" "* "',' 
..., 2::_'V :2 *
/ ..' : "*
Figure 25. Plan for a sphere shape and an ellipsoidal shape (Friedman et al.[2])
the problem of determining the best sphere packing arrangement with which to fill
the target volume. General methods for this treatment plan are iteratively based,
dosimetrically driven algorithms. But these methods require many computations in
order to compute a radiosurgical plan dose distribution, and then to evaluate the
quality of the dose distribution. So geometrically based radiosurgery optimization
has been suggested as a possible alternative means.
However the method the physicians choose relies on human decisions and
experience. Thus, for the same target, different surgeons may produce different plans.
Even the same surgeon, doing the plan twice for the same target, may produce
different plans. And the planning takes a long time, especially for a complicated
target which needs more than 10 spheres. It might take as much as two hours of
planning for a difficult case which needs about 20 spheres. During that time, the
patient has to wait with the head ring attached to his or her head. And most
importantly, even after spending the time to make a plan, many physicians without
sufficient experience, are not sure if the plan is a 'good' one.
Therefore, we provide an automated sphere packing method for the treatment
plan (see Wagner et al.[11]). This method shows potential to significantly aid the
planning of difficult multiple isocenter cases. Based on tests with irregularly shaped
phantom targets and with a representative sampling of clinical example cases, the
method demonstrates the ability to generate radiosurgery plans comparable to, or
of better quality than, multiple isocenter Linac radiosurgery plans found in other
literature. At the same time, this program always produces the same treatment plan
for the same tumor shape. So it can be used as a 'benchmark' to compare with other
plans for the given tumor shape. Moreover, this program provides the treatment
plan in a relatively short time. For a very difficult case which needed more than 18
spheres, this program took less than 3 minutes instead of more than the 1.5 hours
which were needed when the physicians created the plan using traditional methods.
The algorithm of this program is shown in Figure 26.
In the following section, we explain the 'shelling algorithm' which we used in
this program to get the centers and sizes of spheres for the sphere packing plan.
2.6 Shelling Algorithm
The shelling procedure is best illustrated in Figure 27 to 216. The major
steps of this shelling procedure are as follows.
(1) Transferring the data of the boundary of the target
From the CT scan, each point which is identified from the target contour data
file can be mapped into a three dimensional array, since each CT scan shows the
transaxial view for a specific axial coordinate value. In the three dimensional array,
we assign the lateral, AP and axialcoordinates to be x, y and zaxes, respectively.
For each individual case, there are minimum and maximum values for x, y and z
axes. We assume the minimum value is always one. We get this information from
the Program for Data Reading. (See Appendix A.1.)
Let the target be minimally enclosed in an xrange beginning with one and
ending with M,, a yrange beginning with one and ending with My, and a zrange
beginning with one and ending with M,. Now we put 'safe bumper' layers sur
rounding the target. That is, we enlarge the box containing the target to ET =
[0, M, + 1] x [0, My + 1] x [0, M, + 1]. ET stands for 'enlarged target box.' We assume
that the dimension of each voxel is 1 x 1 x 1. Then we will number the voxels in a lexi
cographic manner, beginning with 0 and ending with (M.++l)x(My+l)x(M2+1)1,
giving us a total of (M, + 1) x (My + 1) x (M, + 1) voxels with which we will work.
Now we want to identify the voxels in ET in two different ways: first, by a set
of coordinates (a, b, c), and second, by an integer representing a counting procedure.
Here, a is the left hand endpoint of the xvalue of that voxel, b is the left hand
endpoint of the yvalue of that voxel, and c is the left hand endpoint of the z
value of that voxel. The second notation for the voxel (a, b, c) is by some integer
n. The voxels are counted, using a lexicographic ordering beginning with 0 and
ending with (M, + 1) x (My + 1) x (M, + 1) 1 which is denoted by nofdata in
the program (see Appendix A.2). Then the voxel denoted by (a, b, c) is numbered
as c x (M. + 1) x (My + 1) + a x (M. + 1) + b. And it is clear that, for every
0 < i < nofdata, there are unique nonnegative integers, a,b,c, such that i =
c x (M, + 1) x (M. + 1) + a x (My + 1) + b. That is, for every 0 < i < nofdata,
one and only one voxel is assigned. Note that Figure 27 to Figure 214 show the
same transaxial view at different stages of the algorithm. Thus these figures show
the data of a fixed axialcoordinate value.
(2) Assign a status value for each voxel
For every 0 < i < nofdata, the status value for the ith voxel is denoted
by cc[i]. That is, we denote the status value for the voxel with coordinates (a, b, c),
by cc[i] where i = c x (M, + 1) x (M. + 1) + a x (M. + 1) + b. The following is
the procedure to assign status values. First of all, let cc[i] = 3 for every 0 <
i < nofdata. Then we assign the number, 1, as the status value for each voxel
corresponding to any of the target contour data. Next, for every 0 < j < M,, we
assign cc[j x (M, + 1) x (My + 1)] = 0, because the j x (M, + 1) x (My + 1)th voxel
is the voxel at the left lower corner for every jth transaxial view; that is, the voxel
belongs to the healthy tissue. (See Figure 28.)
For each voxel with cc[i] = 3, we define cc[i] = 0 if any of cc[i 1], cc[i +
1], cc[i (My + 1)], cc[i + (My + 1)] is 0. Note that this method produces zeros for
every voxel in the safe bumper layers. (See Figure 29.)
(3) Shelling the target voxels
We proceed by mathematical induction on j, for
0 < j < x min{Mn My, Mz} .
12
For i = 0 to (M, + 1) x (My + 1) x (Mz + 1) 1, if the ith voxel is assigned 1
or 3 for its status value, and if any of cc[i 1], cc[i + 1], cc[i (M, + 1)], cc[i +
(My + 1)], cc[i (M, + 1) x (My + 1)], cc[i + (M. + 1) x (My + 1)] is equal to j,
then we assign a new status value for the ith voxel as c[i] = j + 1. Note that Figure
27 to Figure 214 show only a twodimensional figure. But this procedure is held
in a threedimensional array. This process is repeated until the deepest lying voxels
have been identified. The deepest lying voxel in the entire target volume would be
the best location for an isocenter. And at the same time, the status value assigned to
the deepest voxel indicates the radius of the sphere, which is one less than the status
value of the voxel. This completes our induction. (See Figure 210.)
(4) Removing the largest sphere
At this moment, we might have several different voxels which have the same
largest status value. So we choose the smallest numbered voxel as the center of the
first largest sphere to be removed from the target. But in the program in Wagner et
al.[11], we choose the most effective voxel to be the center of the first largest sphere,
by using the 'score function' which was developed by Tom Wagner. We think of
the resulting sphere in the tumor as 'removed,' and therefore we can assign the new
status value zero for these voxels in the sphere. To determine which voxels form the
sphere that is removed, we do the following: Let R = voxel at the center of sphere
to be removed. Let r = status value of R. Then we remove all voxels S with the
property that distance(center of R, center of S) < r 1. (See Figures 211 and 212.)
Now, we proceed by mathematical induction on j, for
0 < j K x min{Mn My, Mz} .
For i = 0 to (Mr + 1) x (My + 1) x (M, + 1) 1, if the ith voxel has status value
> j, we reassign a new status value as follows: if any of cc[i 1], cc[i + 1], cc[i 
(My+ 1)], cc[i+(My+l)], cc[i(M+l)x(My+l 1)], cc[i+(M.+1)x(My+l)] is
equal to j, then we assign a new status value for the ith voxel as c[i] = j + 1. This
process is repeated until the deepest lying voxels have been identified. (See Figures
213 to 215.)
And we repeat step (4) until we get either 0 or 1 for the status value of each
voxel in ET, the enlarged target box. Because the smallest size of the diameter of
collimators is 5mm, we finish our program when the status values of all the voxels are
zeros and/or ones. (See Figure 216.) The remaining part will be taken care of by
the 'score function' (see Wagner et al.[11]). After finishing this procedure, we have
the spheres filling the target. In Figure 217, we see the sphere packing plan within
a target by this method.
For further detailed information about the score function and the automated
sphere packing plan, we refer the reader to Wagner et al.[11] and Wagner[10].
STOP
Figure 26. Sphere packing algorithm block diagram (Wagner et al.[ll])
0 *0 0
00
.0
0 0
0
0
0 5 10 15 20 
Figure 27. Transferring the data of the boundary of the target
3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3
3 3 3 33 3 3 1 1 1 1 1 1 1 3 3 3 313 3 3 3 3
3 3 3 1 1 1 1 1 3 3 33 3 1 11 1 1 1 33 3 3 3
3331,3 3 33 3 3 3333 31 1 3 11 3 3 3 3.
3 1 1 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 1 3 3 3
3 1 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 1 1 3 3
3 1 3 3 33 3 3 3 3 3 3 3 3 33 3 33 33 1 1 3
3 1 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 1 1 3
3 1 1 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 1 3 3
3 3 1 3 3 33 3 3333 3 3 3 3 3 3 1 1 1 3 3
3 3 1 1 1 1 3 3 3 1 1 3 3 3 3 3 3 1 1 3 3 3 3 3
3 3 33 3 3 1 11 1 3 11 1 1 1 1 3 3 3 3 3 3 3
3 3 3 3 33 3 3 3 3 3 3 3 1 1 3 3 3 3 3 3 3 3 3
033 31313 3 3 3 3333 3 3 3 33 3 3 3 3
Figure 28. Status values for the boundary and the initial voxel
0
0
0 1
0
0
0
0
0
0v
0 1
0
0
0 1
0 o
o o o o o olololo/o/o/olo/o olo o o o o o o o
)1o o o o o10ai o o o 01o o o o o o
001 0 0., 000000000
00 0 0 0 0
0 00 0 0
JO 0000 o tooolololololo ofolojololo
010
0 0
Figure 29. Status values for the normal voxels
Figure 210. Shelling the target voxels
I
0 0 0 0 0 0 0 0 0 0 0 0 0 1 000 0 0 0 0 0
0 0 U 0 0 0 0 1 1 1 1 1 1] 1 0 0 J 0 0 0 l0 u
S0 1 2 2 2 3 3 3 3 3 11 1 1 1 0 0 0 0
1i 3 3 3 13 4 4 4 4 4
.1 4 4 4 4 5 5 5 5 4
0i1 4 5 5 5 4 4 5 5 A 1 0.
0 1 1 'l33 4 4 4 3 3 4 4 '1 0 t
0 0 1 1. 2 2 3 3 3 2 2 3 31 1 0
) I,.) 1 2 21 2 1 1 2 0 0 () oo )
0 0 0 0 0 11 11 1 0 1 1 1 1 1 1 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0 0I 0
o0 0 0 0 0 0 0 0 0 0 0 0 00 0 0 0 0 0 0 0 0 0
Figure 211. Choosing the largest sphere
0 0 0 0 0 0 0 0 0 0 ) 0 000
0 0 0 0o III1 0 0 0 0
o I 1 0 1) 0 1) 0 0 0 0 0 1 I 0 'i
O 0 1 .2 l0 0 0 0 0 0 0 0
"2..,0 0 () 0 0 0 0 ..
0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 oi
I .. 1 0 o )u0 0,, 0 o1o
c) i"... 0 01 0 o 0) 0 0 0 0 0 1 ,i 1 1 o
0_ 1 1 0 0 0 0 0 0 02 1 0 0 .
() 0 0 (.2. 0 0 111 0 10 t) 0 1 1 1 11 0 ) 0
0 1) 0 0 0 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0 0 () 1
0 00 00 000 000 0 0 0 0 0 0 0 0
Figure 212. Removing the largest sphere
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 u 0 0 1 0 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0
0 0 0 1 1 0 0 0 0 0 0 1 1 1 11 1 0 0 0 0 j
r0 0 1 1 0 the1 0 r T I I
0 11 0 0 0 u 0 U1) 00 1 1 0 10
U .. .~ I0U~U lT
0 1 1 0 0 0 0 0 0 0 0 0 0 0 1
) To o 1 o o o ( o o o o I T 11
0 1 LT o 1 0 0 0 o .) 0 0 0 0 1 O 2 .. 42 1 0 0)
0 1~ 11O0O O O U 1 1 1) 1 1 1 (
U 0 1 1 1 0 o o o o 01 10 00 0 2 41 1OU 1 U 0 1 00o
0 0 1 1 0 0 01 0 0 0 0 0 1 2 1 0000 000
0 0 1 1 01 0 10 0 0 0 0 0 1 0 lo 0 01, o 0 0 0
0 0 0 000 0 0 0 0 01010 0 0 01 0 0 0 0 0 0 0
Figure 213. Shelling the remaining target voxels
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 1 1 0 0 0 0 0 1 1 1 1 1 1 1 0 0 0 0 0
0 1.&.1(J 00000_0 U1 '.)11 0 1 23 3 0.Z 1 0 0
0 0 0 0 0 ) 0 1 1 0 1 1 1 0 0 0 0
oI 1ooooooo ,0 o1O0ooO
Fi ur o o o o o o o o o s ofo t r emini o a
o I 1 0 0 0 0 0 0 0 0 o 0 0 142 3arge 1 1t
Q I : 1 0 0 0 0 0 0 0(' ) 0 1l 4 A A 1 1 1,0
I 1 0 0 0 0 0 0 u 1.) 0 1l 3 2 1. 0 0
0 0 1 1 0 0 0 10 0 0 0 0 0 1 2 3 3 212 1 1 1 0 0
0 0 1 1 1 0 0 10 0 0 0 0 1 2. 2 1 0 10 0 0 0
0 00 0 0 0 0 0 0 0 0 0 0 1 1 0'L 0] 10 i 0 ) 0 10 0
o 0 0 0 0 0 0 0 0 0 0 0 0 0 0~ 0 0 0 0 00 0
Figure 214. Choosing the largest sphere of the remaining target
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 00000 0 0 0
0( 0 0 0 0 0 0 1 0 1 1 1 1 0 0 0 0 0 0 o 0 0 (
0 0 0 1 1 0 0 0 0 0 o I I IJI I 1 JJ01 0 0 0 (1 0
0 0 ) 1 1 0 C0 0 0 0 0 0 1 1 1 1 1 1 1 0 0 0 '
0 1 1 1 0 0 0 0 0 0 0 0 0 1 1 1 1 0 0 0
0 1 1 0 0 0 0 00 0 O l0 o1 1 01 o
0 1 1 0 0 0 0 0 0 0 0) 0 0 0 0 00 0 01 1 1 0
0 1 1: 0 0 0 0 0 ( 0 0 1 0 0 0 0 1 2 1 1 0
I 1 .2 u 0 0 o 0 0 0 C ) 0 0 0 0 0 0 0 0 0 1 1 ) 0
0 0 1 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 1 1 1 0 0
0 0 1 1 1 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0o 0 0 o0 0
0 0 0 0 0 0 1 1 0 1 0 1 1 1 1 0 0 0 0 0 0 o 0
0 0 0 0 0 0 0 01 0 1 l 0 0 o 00 1) 0 iJ)
0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
Figure 215. Removing the second sphere
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 00 0 0
0 0 0 0 0 0 0 1 0 1 1 1 1 1 0 0 0 0 0 0 0 0 0 0
0 0 0 1 1 0 0 0 0 0 0 0 1 1 1 1 1 1 1 0 0 0 0 0
0 0 1 1 0 0 0 0 0 0 0 (.) 0 1 0 1 1 0 1 1 0 0 0)()
0 1 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0 0 0 0 0 1 0 0
0 1 0 0 0 0 0 0 0 0 0 0 0 0 010 0 0 0 0 0 1 1 0
0 1 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0
0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 1 1 1 0 0
0 0 1 1 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 1 1 0 1 0 1 0 0 0 1 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 00 0 0 0 00 1 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 000 0 0 0 0 0
Figure 216. Removing all the spheres
patB.wol
.35
40.
i ' c
1"" ,.^ ' .... ," ...,,
20 2
100
20 Lot
AP
Patient A. 12 8cc mengloma
.3)5
306
i 0i ^ .....
.40 "
.30
2 20
10 0 10
.1
2D Lat
AP
Figure 217. A sphere packing plan for a brain tumor (Wagner et al.[ll])
CHAPTER 3
A CLASSIFICATION OF UNLABELED TREES
3.1 Introduction
In this chapter, we obtain a code to denote an unlabeled tree. By means
of this code, we classify unlabeled ntrees. One may take an unlabeled ntree to
be the geometric realization of a connected, finite, simplicial complex without any
simple closed curves. In particular, we call a tree an ntree if and only if it is a tree
with n vertices (that is, (n 1) edges). Our code assigns a unique, ordered, 'perfect
sequence', pf(T) = (di, d2,. ., dn), to each unlabeled ntree, T. And, conversely,
given an ordered sequence of n integers satisfying certain properties, it is the perfect
sequence of exactly one unlabeled ntree. Thus, we have our main theorem, the Tree
Classification Theorem (Theorem 3.5.5). We then show how to reconstruct the tree
shape from a given perfect sequence, by hand.
Our work includes an algorithm and a Mathematica program (see Appendix
C.1) that produce a list of all the perfect sequences for all possible ntrees, thus also
producing the number of ntrees, for any given n. For some examples, see Appendix
C.2. Since our algorithm and program are based upon our main theorem, we know
that all possible unlabeled ntrees are produced by the program. Thus the program
gives us, for each positive integer n, a complete list of all unlabeled ntrees.
Given n > 0, with sufficient computer power and time, our program will
list all the unlabeled ntrees, thus telling us how many unlabeled ntrees there are.
However, we do not have a simple formula that tells us how many unlabeled ntrees
there are. This remains an open problem.
The Literature contains work regarding the classification of labeled graphs
and labeled trees. In Lovasz[6], the Priifercode is used to determine the number of
labeled ntrees (see the Appendix D.2 and D.3). In Sloane[8], Sloane shows a table
for n < 26. However, Sloane references Harary for the method he uses. In Harary[5],
there is a simple formula for the total number of labeled graphs with n vertices (see the
Appendix D.1). There is also a formula for the total number of unlabeled ntrees, but
it requires knowing the number of rooted ntrees, which is calculated inductively. And,
even though Harary[5] contains ntree shapes up to n = 10, there is no explanation
of how to get the shapes of these trees, while our work does show how to get these
tree shapes from our code.
This chapter is organized as follows:
Section 1: Introduction
Section 2: Definitions
Section 3: Vertex Sequence
Section 4: Degree Sequence
Section 5: Perfect Sequence and Main Theorem
Section 6: Obtaining Perfect Sequences
Section 7: Extended Partition and the Algorithm
3.2 Definitions
Most of the necessary definitions in this section are from Harary[5]. We refer
the reader to Harary[5] for easy reference.
Definition 3.2.1 A graph G = (V(G),E(G)) consists of a vertex set V(G) =
{vi,...,vn} and an edge set E(G) = {el,..., em} C {{fu,v}u, v E V(G)} in which
each edge is an unordered pair of two distinct vertices u and v, and the edge {u, v}
is said to join u and v. For an edge {u,v} we say that u and v are adjacent
vertices and the vertex u and the edge {u, v} are incident with each other, as are
v and {u,v}. From now on, V(G)I denotes the number of vertices and IE(G)I
denotes the number of edges. For a vertex u E V(G), the vertex set
A(u) = {v E GIv is an adjacent vertex of u}
is called the adjacent set of u. Note that u A(u).
Definition 3.2.2 A walk is an alternating sequence (vo0, ei, v1, e2,... ,ek, Vk) of
vertices and edges s.t. vi,1 vi for i = 1,2,...,k, and ei = {vii,vi} for i =
1,2,..., k. If there is no confusion, a walk can be denoted by a sequence of vertices
only in the path like (Vi, V2,..., Vk) A cycle is a walk (vo, el, vi, e2,...,ek, Vk) with
k > 3 in which v0 = vk is the only vertex repeated. A path is a walk with no
repeated vertex. A (u, v)path is a path with the first vertex u and the last vertex
v. The length of the (u, v)path, l(u,v), in a graph is the number of edges in the
(u, v)path. The distance d(u,v) between two vertices u and v is the length of a
shortest path joining them, if any; otherwise d(u, v) = oo. Note that d(u, u) = 0. A
graph G is connected iff every pair of vertices are joined by a path. A graph having
no cycle is acyclic. Note that an acyclic graph can be disconnected. A forest is an
acyclic graph. A tree is a connected acyclic graph. An ntree is a tree with n vertices.
Definition 3.2.3 We say that a graph G is isomorphic to a graph H if and only
if there exists a bijection f : V(G) + V(H) such that {u, v} E E(G) if and only if
{f(u),f(v)} E E(H).
Definition 3.2.4 An unlabeled tree is an isomorphism class of trees.
We are interested in the total number of unlabeled ntrees and their shapes.
For example, T1 = ({1,2,3},{{1,2},{2,3}}) and T2 = ({1,2,3},{{2,1},{1,3}})
are different trees, but there is an isomorphism f : V(T1) 4 V(T2) by f(1) =
2, f(2) = 1, f(3) = 3. Thus they are the same unlabeled tree, so we want to count
them as one unlabeled 3tree. From now on, a tree implies an unlabeled tree unless
otherwise specified.
Definition 3.2.5 The degree of a vertex v, defined by d(v), is the sum of the
number of edges incident with v. A vertex v is called an endvertex if d(v) = 1.
We will use the geometric, that is, topological realization of a graph inter
changeably with 'graph' as defined above.
Proposition 3.2.6 [5] The following statements are equivalent for a graph T:
(1) T is a tree.
(2) Every two vertices of T are joined by a unique path.
(3) T is connected and IV(T)I = IE(T)I + 1.
(4) T is acyclic and IV(T)I = \E(T)I + 1.
Note that there are exactly n vertices and n1 edges in an ntree.
Proposition 3.2.7 (Degree sum formula) [5] If G is a graph with the vertex
degrees di,...,dn,, then ZvEv(G) di = 21E(G)I.
Corollary 3.2.8 For an nTree T we have
i=1 d(v,) = 2(n 1) = 2E(T)\ = 2(IV(T)I 1)
where vi is a vertex of T for i = 1,2,..., n.
Because a nontrivial tree satisfies EEV(T)d(v) = 21E(T)I = 2(IV(T)I 1),
there are at least two vertices with degree 1. Thus we have the following theorem as
a corollary.
Theorem 3.2.9 [5] Every nontrivial tree has at least two endvertices.
3.3 Vertex Sequence
In this section, we define the vertex sequence of a given ntree.
Definition 3.3.1 Let T = (V,E) be an ntree. If d(v) = m, then there are m
components of T {v}. A branch of v (or simply a vbranch) is a subtree which
is the closure of a component of T {v}. We call the set of vertices in a vbranch
a vbranch vertex set. If u E A(v), then the vbranch vertex set containing u is
called the (v,u)branch vertex set and denoted by B(v,u) (or B(u)). We call the
set B'(v,u) = B(v,u) {v} the (v,u)subbranch vertex set, which is denoted by
B'(u) if there is no confusion. The subtree consisting of B'(v, u) as the vertex set
and {{s,t} E B'(v,u) x B'(v,u) {s,t} E E(T)} as the edge set is called the (v,u)
subbranch.
Example 3.3.2 Let T be the 10tree shown below.
s t u
p q r v w x
Y
'y
Figure 31
Since d(v) = 4, there are 4 components of T{v}. Since A(v)= {r, t,w,y},
we have that
B(v,r) = {v,r,s,q,p}, B'(v,r) = {r,s,q,p}
B(v,t) = {v,t}, B'(v,t) = {t}
B(v,w)= {v,w,u,x}, B'(v,w)= {Iw,u,x}
B(v,y) {v,y}, B(v,y) {y}.
On the other hand, we have that
B(r,v) = {r,v,t,u,w,x,y}.
That is, B(v, r) $ B(r, v). And this is true for any tree except the 2tree. In fact, for
every ntree with n > 2, there are at least 3 vertices, namely u, v, and w. Without
loss of generality, let u and v be vertices adjacent to each other, and let the (w, v)
path contain the vertex u. Then w E B(v,u) and w B(u,v).
Definition 3.3.3 A section of a sequence is a subsequence consisting of consecutive
terms of the sequence.
Definition 3.3.4 Let T = (V, E) be an ntree with V = {x, x2, ... x}. We define
the set, S, of sequences of the form s = (vI, v2,... v, n), where s is a permutation
of the xi's in V(T), to be those sequences satisfying:
(*) forany vj with 1
where k = max{i {1,2,... ,j3} A(v) {v1, v2, ,v} # 0
Then S is called the vertex sequence set of T and each element, s, in S is called a
vertex sequence of T.
Definition 3.3.5 For any v E V, we call S, = { E Svi = v} C S the vvertex
sequence set.
Definition 3.3.4 requires that the vertices in a sequence be listed in a particular
way. The condition (*) ensures that all vertices are listed and that each is listed
exactly once. We can choose any vertex for the first term, vi. Then the condition
(*) says that, for the (i+l)th term, we have to choose an unlisted adjacent vertex
of vi, if possible. If every adjacent vertex of vi is already used, then look backward
along the sequence until we come to the first vertex which has an unused adjacent
vertex. Then, for the (i+l)th term, we assign one of the unused adjacent vertices of
that vertex. To get a vertex sequence, we continue this procedure inductively.
Note that every vertex sequence of an ntree has n entries.
Example 3.3.6 We show how to construct a vertex sequence of a given tree. Let
T be the 7tree shown below.
f g
a b c d e
Figure 32
Since we have freedom to choose the first term, let v, = b. Then v2 E A(vi)
{vl} = A(b){b} = {a,c,f}{b} = {a,c,f} = A(b) = A(vi). So v2 must beoneof
the elements of A(vi) = A(b) = {a, c, f}. Let v2 = c. Since v3 E A(v2) {v1, v2} =
A(c) {b,c} = {b,d,g} {b,c} = {fd,g}, v3 must be either d or g. Let v3 = g.
Then v4 A(v3) {v,v2, v3}, since A(v3) {v1,v2, v3} = {c} {b,c,g} = 0. On
the other hand, we have that A(v2) {v1,v2, V3} = {b,d,g} {b,c,g} = {d} 5 0.
Therefore we have to choose d as v4. Then v5 = e since A(v4) {v, v2, v3,v4} =
{c, e} {b, c,g,d} = {e}. Since we have
A(v5) {Vi, V2, V3, V4, V5} = {d} {b,c,g,d,e} = 0,
A(v4) {v1,v2,v3, v4, v5} = {c,e}{ b,c,g,d,e} =0,
A(v3) {vi, v2, V3, v4, v5} = {c} {b,c,g,d,e} = 0,
A(v2) {v1,v2,v3,v4,v5} = {b,g,d} {b,c,g,d,e} = 0, and
A(vi) {v,,v2, v3, v4, v5} = {a,c,f} {b,c,g,d,e} = {a,f },
v6 must be either a or f. Let v6 = f. Then we have that
A(v6) {vi,v2,v3,v4,V5,V6} = {b} {b,c,g,d,e,f} =0,
A(v5){v1,v2, v3,v4, v5,v6} = {d} {b,c,g,d,e,f} = 0,
A(v4) {vi,v2,v3,v4,v5,v6} = {c,e} {b,c,g,d,e,f} =0,
A(v3) {v1,v2,v3,v4,v5,v6} = {c} {b,c,g,d,e,f} = 0,
A(v2){ fV1,v2, v3,v4, v5, v6} = {b,gd} {b,c,g,d,e,f} = 0, and
A(vi) {v1,v2, v3, v4,v5, v6} = {a,c,f}{b,c,g,d,e,f} = {a}.
Therefore v7 must be a; that is, we construct a vertex sequence
(V1, V2,...,V7) = (b,c,g,d,e, f,a).
For a given tree we will get different vertex sequences if we choose different
vertices whenever we have two or more choices. But the sequence t = (b, c, d, e, f, a, g)
cannot be a vertex sequence of T above. In fact, we have v5 = f. But we have that
A(v4) {vi,v2, v3, v4} = {d} {b,c,d,e} = 0,
A(v3) {Vi, V2, V3, V4} = {c, e} {b, c, d, e} =0,
A(v2) {Vi,V2, V3, V4} = {b,d,g} {b,c,d,e} = {g}, and
A(vi){V1,V2,V3,V4} = {a,c,f}{ b,c,d,e} = {a,f}.
That is, the sequence t has f, one of the elements of A(vi) {v1, v2, v3, v4}, for v5
instead of g which is an element of A(v2) {v1, v2, v3, v4}. It is a contradiction to
the maximality of k in the definition of vertex sequence.
Proposition 3.3.7 Let (v1, V2,..., V,) be a vertex sequence of an ntree T. For any
vj with j > 1 there is unique 1 < k < j such that vj E A(vk) {v1,v2,... vj}.
Proof First of all, for any j > 1, we have vj 5 vi for i = 1,2,... ,j 1 from the
Definition 3.3.4. Assume that there is j > 1 such that
{vj, vj+I,...,vn} n ((Ujl'A(vi)) u {Vl, v2,..., vj1}) = 0.
Then, for any u E {vj,,vj+i,.. .,vn} and any w E (U=IA(v,)) U {v1,v2, ..,vj},
there is no (u,w)path in T. That is, T is not connected. It is a contradiction
to the connectedness of a tree. Therefore there is 1 < k < j such that vj E
A(vk) {Vi,V2,...,vji}. If there are 1 < ki < k2 < j such that vj E A(vk,) 
{v1i,v2, ...,vjl} and vj E A(vk2) {Vi, V2,.. .,vji}, then the two edges, {vj, vk}
and {vj, vk2}, are elements of E(T). There is a path between vk, and vk2 in the
subtree {{vl, v2,... vj}, E'} where E' = {{u,v} E E(T)j u,v v {V1, V2, .. ,Vj}}.
The tree T contains a cycle consisting of {vj, Vk,}, {vj, Vk2} and the (Vkl, Vk2)path.
This contradicts the fact that a tree is acyclic. Therefore there is only one 1 < k < j
such that vj E A(vk) {vi,V2,... ,vj_}. I
From the Definition 3.3.4 and the previous lemma we get the following corol
lary immediately.
Corollary 3.3.8 Let (v, v2,... v,n) be a vertex sequence of an ntree. Then
(1) v2 E A(vi)
(2) If d(vj) > 2 for any j = 2,3,..., n 1, then Vj+l E A(vj).
(3) For any j = 2,3,... ,n 1 if d(vj) = 1 and there is i < j such that d(vi) >_ 3
and d(vk) = 2 for every k = i + 1, i + 2,... ,j 1, then vj+l E A(vi) and Vj+l .
A(vp), for every p= 1,2,...,i 1,i + 1,...,j.
Proposition 3.3.9 Let (Vi, v2,... Vn) be a vertex sequence of an ntree and A(vi) =
{vli,vI,12...,vli} with 1 <11 < 12 <...< lm. Then
{V, V2, Vn} = UhlB(vi, vi,).
Proposition 3.3.10 Let = (1, v2,..., v,.n) be a vertex sequence of a given ntree
T. If d(vj) = m for some j > 1, then there exist m vertices, vi, vj,,vj2,...,viji
such that A(vj) = {vi, vj,,vj2,...,vimi} with i
Proof For any j > 1 there exists unique i < j such that {vi,vj} E E(T)
by Proposition 3.3.7; that is, vi e A(vj). So, for any vk A(vj) {vi}, j < k.
Since d(vj) = m, there are m 1 vertices, Vj Vj2... ,vj_,, such that A(vj) =
{vi, Vjl, Vj2 ,..., vj.I} with < j < j < j2< ... < jm. J
Proposition 3.3.11 Let = (V1, v2,..., v,,} be a vertex sequence of a given ntree
T. Let d(vi) = m and A(vi) = {V11,V12,. ,Vi} be the adjacent set of v1, named
so that 1 <11 < 12< ... < 1m. Then, for any i = 1,2,... ,m, the set of vertices in
the subbranch, B'(vi,vii), forms a section in s starting with vii,.
Proof Clearly, the section containing vi, starts with v1, = v2. Let B'(vi,v11) =
{VI,V11,+i,. .. ,Vl+1m}. Assume that there is a vertex u E V(T) B'(vi,vi,) such
that (v11,v11+1,...,v11+j,u,...) is a sectionof 5. Since u V B'(vi,vi,), A(vll+k)
B'(vi,vl,) = 0 for any k = 0,1,2,... ,j from the definition. Thus U=0A(vl1+k) 
B'(vi, vl) = 0. That is, Uk=oA(vl,+k) D B'(vi,vi,). On theother hand, A(vll+k) C
B'(vi,vl,) forever k = 0,1,2,... ,j. Thus wehavethat U=o=A(vl+k) C B'(vi,vi).
Therefore U =A(vi+k) = B'(vi,vl,). That is, j = mi. Hence B'(vl,vl) forms
a section in s. Since B'(vi,vi) is a subbranch of vI, Vli+mi+l V A(u) for any
u E B'(vi,vl). Thus vll+mi+l E A(vi). That is, Vll+mi+1 = v12. For any i =
1,2,... ,m 1, let B'(vi, vi,) form a section in s and vl,+m+i = v1i+1. By a simi
lar method, B'(vi,vli+,) forms a section in s and Vi;+,i+m +l1 = Vii+2. I
The section formed by a subbranch B'(vi,vi) is denoted by (B'(vi,vl,)).
From the previous proposition, we have the following corollaries.
Corollary 3.3.12 Let A = (v1, V2, ... Vn) be a vertex sequence of a given ntree
T. Let d(vi) = mn and A(vi) = {vi,v12,...,vin} be the adjacent set of vi, named
so that 1 < 11 < 12 < ... < 1m. The section (B'(vi,vi,)) precedes the section
(B'(vi,vi,+,)) in A for i = 1,2,...,m 1, and there is no other vertex between
these two sections.
Corollary 3.3.13 Let (vi,V2,... V) be a vertex sequence of a given ntree T.
For any j > 1 with d(vj) = m > 2 let A(vj) = {vi,v,,v12,..., ,,_i} where i <
j < Ji < J2 < ... < jm1. Then (B'(vj,vj,)) = Vi,vji+ ...,vj(i,+l)1) for any
i = 1,2,..., m 1. In particular, we have that ji = j + 1.
Note that B(vj, vjm ,) = {vj, Vjm_, Vj.,i+i,..., Vj.i } is the (vj, vj,,_ )branch
vertex set for some jm > jm1.
Corollary 3.3.14 Let (vl,v2, ... vn) be a vertex sequence of a given ntree. If
i < j with vj A(vi) and vj 3 A(vk) for every k = i + 1,i + 2,...,j 1, then
vp A(vk) for every p > j and k = i+ 1,i + 2,...,j 1.
For any given vertex sequence s = (Vi, v2,..., vn) and any vertex vj 7$ vi, Vi
belongs to a branch of vj. We want to give a special name to this branch.
Definition 3.3.15 Let = (vi, V2, vn) be a vertex sequence of a given ntree
T. For any j > 1 the root branch of vj in is the branch of vj containing vU.
Note that there is no root branch for vj.
Proposition 3.3.16 Let (vi, v2,. .. v,) E S be the vertex sequence set of a given
ntree T. For any 1 < j < n, if vj+i A(vj), then vj is an end vertex.
Proof If vj+l V A(vj) for some 1 < j < n, then vj has a unique edge with a
vertex, vi, with i < j by Proposition 3.3.7. That is, d(vj) = 1. Therefore vj is an
end vertex. L
Proposition 3.3.17 Let (vi,v2,. ,vn) S be a vertex sequence set of a given
ntree T. For any 1 < j < n there is an integer jm such that
{Vl,U2, ...,vj} U {Vjm,vjm+i,...,vn}
is the vertex set of the root branch of vj.
Proof Let 1 < j < n and d(vj) = m. Then, by Proposition 3.3.10, there exist
m different vertices, vi,vj,,vj2,... ,vjm_ such that A(vj) = {vi,vjl,vj2,.... vii}
with i < < < j
(u=l B(vj, vj,)) = (,vjl ,..., ).
That is, the union of all the subbranches of vj except the root branch is a section
from vj, to Vji. Therefore the set of vertices in the root branch is {v1, v2,... vj}U
{VJm,Vj.m+l, ..,vn}. IL
Note that, if jm > n, then the root branch is {V1, v,..., vj}.
Corollary 3.3.18 Let = (V1, v2,. vn) E S be the vertex sequence set of a given
ntree T. Let j 54 1. Then d(vj) = 1, that is, vj is an end vertex of T, if and only
if the set of vertices of the root branch of vj is V(T) = {v1, v2, ... Vn}
Proposition 3.3.19 Let = (v1, v2,... ,vn) be a vertex sequence of a given n
tree T. For any vertex vj with j > 1 the (vi,vj)path in T is a subsequence of
(Vi, V2, .. vj) .
Proof For any vertex vj with j > 1 there is a unique vertex Vkp with kp < j
such that vj E A(vkp). For the vertex vkp there is another unique vertex vkp_,
with kp1 < kp such that Vkp C A(vkpi). Then we will have that v, is the unique
adjacent vertex of a previous vertex Vk,. Therefore the (vi, vj)path is a subsequence
(vi,vkI,...,vk_,vkfp,vj) of (vl,v2,...,Vj). L
Proposition 3.3.20 For any two vertex sequences S, and s2 in the vvertex se
quence set S, = {s E SIV1 = v} the root branch vertex sets of a vertex x :A v, = v
in s1 and in s2 are same.
Proof Let S1 = (vi, v2, ,v,n) and g2 = (ui,u2,..., Un) be two different vertex
sequences of a given tree T with v, = ul. Let x be an arbitrary vertex in T
with x =A v1. Then there are two integers k and I such that Vk = x = ul. Let
A(x) = {xi,x2,... ,xm1,y}. Since T is a tree, there is a unique (vi,x)path in T.
Then one of the elements of A(x) is in the (vi, x)path. Without loss of generality,
let y be in the (vi, x)path. From the proof of Proposition 3.3.17 we have that
(U j'=B'(x,x,)) = (u 'B'(vk,xp))
= (Vk+l, Vk+2,. ,Vk+r} and
(U=1 B'(x,xp)) = (Um='B'(ul,xp))
= (U+i,UL+2,. ,UI+r)
for some r. Therefore we have that
{Vl,V 2,...,Vk} U {Vk+r+l,..., Vn} = {u1,U2,..., U} U {Iu+r+l,..., n}.
That is, the root branch vertex sets of x in g, and in 2 are same. _
Example 3.3.21 Let T be an 8tree as shown below.
v w
p q r s t U
Figure 33
Clearly, s = (q,r,s,t,u,w,v,p), S2 = (q,p,r,s,w,t,u,v) 5r = {f E
Siv1 = q}. Let
(vi,v2,...,v8) = ).
Then the root branch vertex set of s is {p,q,r,v}. That is, {p, q,r, v} = {q,r} U
{v,p} = {vi,v2} U {v7,v8} and {p,q,r,v} = {q,p,r} U {v} = {u1,u2,u3} U {us}.
Therefore the root branch vertex sets are same but the position might be different.
On the other hand, d(w) = 1 and the unique adjacent vertex of w = V6
is s = v3. Thus s = V3 is in the root branch of w = v6. Therefore the root
branch vertex set of w = v6 is {vi,v2,v3,v4,V5, v6} U {v, v8} = {vi, v2,..., V8s} =
{q,r,s,t,u,w,v,p}. That is, the root branch vertex set of an end vertex is V(T)
itself.
3.4 Degree Sequence
In this section, we define the degree sequence of a vertex sequence.
Definition 3.4.1 Let S be the vertex sequence set of an ntree T. For any ver
tex sequence = (V1,V2,. .,V n) E S, we call Di = (dl,d2,.. .,dn) the degree se
quence of where d, = d(vi) and di = d(vi) 1 for i > 2. We call the set
D = {Djl. E S} the degree sequence set of S. And, for any vertex v E V, we
call DV = {DjsA E S,} C D the vdegree sequence set of S,. For any section
(B(vj, vj,)) = (vj, vj,,vj,,+,...,vj,+_i) of A, formed by the (vj,vj,)branch vertex
set B(vj,vj,), we define Bd(vj,vj,) = {dj,djdj,+i,...,dj,+1_i} as the (vj,vj,)
branch degree setwhere dk = d(vk) 1 for k = j, ji, j +1, .. ,ji+ 1. And we define
the (vj, vj.)branch degree section by (Bd(vj, vj,)) = (dj, dj,, dj,+1, dj,+ I) which
is denoted by (Bd(vj,)) if there is no confusion. In the same way as in Definition
3.3.1, we define (Bd'(vj,)) = (Bd'(vj,vj,)) = (djdj,+...,dj+i1)
Example 3.4.2 Occasionally, it may happen that there are two or more vertex se
quences of a given tree with the same degree sequence. In the following example, there
are several pairs of vertex sequences with the same degree sequences, respectively.
q
p r .5
Figure 34
Since there are 5 vertices in V(T), there are exactly 5! different sequences of vertices
of T having length 5 and using each vertex only once. But only 16 of them satisfy
the condition (*) in Definition 3.3.4. Therefore the vertex sequence set, S, of the
given tree, T, consists of the following sequences
i = (p,r,q,s,t), s = (r,p,q,s,t), 813 = (s,r,p,q,t) ,
2 = (p,r,s,t,q), s8 = (r,p,s,t,q), s4 = (s,r,q,p,t),
S3= (q,r,p,s,t), 9 = (r,q,p,s,t), s15 = (s,t,r,p,q),
S4 = (q,r,s,t,p), Sio= (r,q,s,t,p), s6 = (s,t,r,q,p),
i5 = (t,s,r,p,q), S3i = (r,s,t,p,q),
46 = (t,s,r,q,p), s32 = (r,s,t,q,p).
Then the degree sequences of these vertex sequences are
D, = (1,2,0,1,0), D, = (3,0,0,1,0), D.i3 = (2,2,0,0,0),
Dj = (1,2,1,0,0), D, = (3,0,1,0,0), Dsj, = (2,2,0,0,0),
D3 =(1,2,0,1,0), D9 =(3,0,0,1,0), Dsj5 =(2,0,2,0,0),
= (1,2,1,0,0), D,,0 = (3,0,1,0,0), D,6 = (2,0,2,0,0),
S=(1,1,2,0,0), Dh =(3,1,0,0,0),
D =(1,1,2,0,0), D,2 =(3,1,0,0,0).
Now we have that
Dj = D3, D, = Di4, D5, = D4, D,, = Ds,
Di, = D8l0, Dqs = D82, Dsi3 = Di4, Ds15 = Dai6.
That is, for some different vertex sequences, the degree sequences might be same.
And, if the degree sequences are same, then the trees denoted by these degree se
quences are isomorphic.
Remark: The sum of the terms in any degree sequence of a given ntree is the
same, namely, n 1. It is 4 in the above example, and it is same as the number of
edges in the tree. See Theorem 3.4.6 below. Since d, = d(vi) and d, = d(vi) 1 for
i = 2, 3,..., n, we have the following proposition.
Proposition 3.4.3 Let Di = (d,d2,. ,dn) be a degree sequence of a vertex se
quence s of an ntree T. Then 0 < di < n1, and 0 < di < n1 for i= 2,..., n.
Proposition 3.4.4 If dj 5 0 for any 1 < j < n, then vj+l is an adjacent vertex
of vj.
Proposition 3.4.5 Let = (v1,v2,..., Vn) be a vertex sequence of an ntree T and
Ds = (di,d2,. ..,dn) the degree sequence of A. For any j E {1,2,...,n 1} with
dj = m > 1, if the set {vj+1, Vj+2,..., Vk} is the collection of all the vjsubbranches
except the root subbranch, then k > j + m.
Proof Since dj = m = d(vj) 1, from Proposition 3.3.10, the vertex vj has m
subbranches besides the root subbranch. Thus there are at least mrn endvertices in
the collection of those subbranches other than the root branch. Therefore at least m
terms in (dj,dj+1,...,dk) are 0. Hence k > j + m. .
Theorem 3.4.6 (First Reduced Degree Sum Formula) For any degree sequence
Di = (dl,d2,..., dn) of a vertex sequence A E S of a given ntree T we have
Z1 di = n l = \E
and
d + E2(d 1)= 0.
Proof d, is the degree of a vertex, and the remaining n 1 terms are each one
less than the degree of a vertex. By the degree sum formula (Proposition 3.2.7), it is
n
true that Ei=1 d(vi) = 2\E\. So
nn
j=1 di = d + Ei=2 di
= d(vi) i Z,=2 (d(vi) 1)
^+ En En
= d(vi) + Ei=2 d(v) Ei=2 1
=(d(vi) + =2 d(v)) ( 1)
= d(vi) (n )
= 21E (n 1)
=2(n 1) (n 1)
=n 1= E.
Then we have that
d, + En2(d 1) d, + En2 E=21
iE=(l =d+ i=2d i=21
=Eni=1 di (n 1)
=(ni)(ni)
0.
Corollary 3.4.7 For any degree sequence, Dj = (d1d2, ...,dn) of a vertex sequence
A E S of a given ntree T, we have
d + ,=2, dO(di 1) = E 2, d,=0 1.
Proof From the first reduced degree sum formula we have that d1+Ei=2(d1) = 0.
Then
0 = d + E=2(d 1)
= d1 + Ei=2, do(di 1) + E=2, d,=0(d 1)
= d + Ei=2, do(di 1) + Ei=2, d,=O(1)
= d + Ei=2, d,0(di 1) Ei=2, d=o 1.
Thus we get the result. I
Note that di > 0 for any i = 1,2,...,n. So di 1 < 0 if and only if di = 0.
Corollary 3.4.7 shows that, in a degree sequence D (di, d2,..., dn), the total
number of 0 terms is the same as di + S"=2 (di 1). Therefore there are exactly
di:O
di + En=2 d,0(di 1) endvertices in the given tree. Because of this, we have the
following proposition.
Proposition 3.4.8 For any degree sequence Ds = (dl,d2,...,dn) of a vertex se
quence A E S of a given ntree T, we have
dj + Ez=2(d 1) # o
for every 1 < q < n.
Proof Assume that there is q < n such that di + EZ 2(d 1) = 0. Since D =
(dl,d2, ...,dn) is a degree sequence of a vertex sequence s E S of the given ntree
T, we have that vq+1 A(vp) for some p < q. Thus at least one of the endvertices
counted by Vp is one of Vq+l,Vq+2,... ,Vn. That is, in the section (d1, d2,. dq),
q (i 1), that is,
we have at least one less 0 terms than di + Ei=2, dO(di ) a is
d + q (i_1) _ =2 1 q1
d + Ei=2, dO(dj i) 2 d,=O 
Then
0 =d+E?=2(d 1)
d q+Z (di )+Z2 =O(d? 1)
= d + EL:2, d0i ( d 1) + E?2, d,=o( 1)
= d + EL2, d4o(di 1) + E d=o()
i=2, di i=2, d,=O('
>1.
This is a contradiction. Therefore we have di + E2(di 1) 5 0 for every 1 < q <
n. 
Theorem 3.4.9 (Second Reduced Degree Sum Formula) Let T = (V,E)
be an ntree. Let Ds = (d1,d2,...,dn) be the degree sequence of a vertex sequence
A E S of the given ntree T. For any jk > 1, let B'(vj,) = {Vj,Vjk+I,...,Vjk+11}
be a (vj, vjk)subbranch vertex set. Then we have
dj + (di 1)=0
dk+..i=jk+l 
and
djk + EZ=+l(di 1) 5 0
for every jk < q < jk+I.
Proof Let E(B'(vjk)) = {{u,v} E E(T)ju, v E B'(v,k)}. Then
T(vj,) = (B'(vjk),E(B'(vjk))) c T
is also a tree, and (B'(vjk)) = (VjkVjk+l,... ,Vj+,l_) is a vertex sequence of the
tree T(vjk). Note that d(vjk) = dj +1 in T. Since vj B'(vjk), the edge {V, Vjk}
is not in the edge set E(B'(vjk)). Thus we have that d(vjk) = dj, in the subtree
T(vJk). Then the result comes from the first reduced degree sum formula (Theorem
3.4.6) and Proposition 3.4.8. 1l
Corollary 3.4.10 Let A = (V1,V2,...,v,,) be a vertex sequence of an ntree T and
Ds = (d,d2,. ,d,d) the degree sequence of s. For any dj > 0, if the subsequence
(vj, v+l,... vp) is the collection of all the vjbranches except the root subbranch of
vj, then
dj + E +(d 1)= 0
and
d3 +E=j,+l(di 1) 0
for every j < q < p.
Proof For j = 1, they are true by the first reduced degree sum formula (Theorem
3.4.6) and Proposition 3.4.8. Let j > 1 and vj E A(vt) for some t
(i) If dj = 0, then d(vj) = 1, that is, vj is an end vertex. Since j > 1, vj 5 vi.
Thus the root branch is the whole tree by Corollary 3.3.18. Therefore (vj) is the
collection of all the vjbranches except the root subbranch and dj + EZ=j+ (d 1) =
dj = 0.
(ii) If dj > 0, then the subsequence (vj, vj+l,..., vp) which is the collection of all
the vjbranches except the root subbranch, is the (vt, vj)subbranch, since vj E A(vt).
By the previous theorem, we get dj + E=j+l (di 1) = 0. Thus the collection of all
the vjbranches except the root subbranch of vj is the (vt, vj)subbranch. Therefore,
by the second reduced degree sum formula, we get dj + Eq+( 1) 0 or every
+~ +1 d 1) jf 0 for every
j < q < p.
Not all of the nonnegative integer sequences with n entries can be a vertex
sequence of an ntree. There is a condition to being a degree sequence of an ntree.
Definition 3.4.11 Let D = (dl,d2,..,d, ) be a nonnegative integer sequence.
We say that D satisfies the degree sequence condition, briefly DSC, if and only if it
satisfies all of the following:
(a) di > 0.
(b) E', di = n 1.
(c) For any j < n with dj = m > 0, there are exactly m + 1 integers, j < ji =
J+l
djk + Ejk+''(di 1) = 0 for k = 1,2,...,m.
(d) For any jk with dj = I > 0, there are exactly + 1 integers, jk < JkA =
ik +1< k2 < ... < Jk, < Jk(,+1) = Jk+1 such that
d ,P+ Ej+)l(d 1) 0 for p = 1,2,..., 1.
Theorem 3.4.12 Let D = (,dl ,..., dn) be a nonnegative integer sequence. Then
D is a degree sequence of a vertex sequence of an ntree if and only if D satisfies
DSC.
Proof [=] This follows from Proposition 3.4.3, Theorem 3.4.6 and Theorem 3.4.9.
[=] Let V = {v1,v2,... ,v,} be a set of n different vertices. Let D =
(di, d2,.. .,dn) be a nonnegative integer sequence satisfying the above conditions.
Since d, > 0 and di > 0 for any i = 2,3,..., n, we can assign the degree of each
vertex vi E V by d(vi) = d, and d(vu) = di + 1 for i = 2,3,...,n. Let j < n and
dj = m > 0. From (c), there are exactly m + 1 integers, j < j1 < J2 < ... < jm <
jm+i, such that ji = j + 1, and dj, + EZij+(di 1) = 0 for k = 1,2,...,m. Now
we assign an edge between vj and vj, for k = 1,2,...,m. That is, vjk E A(vj)
for i = 1,2, ...,m.
(i) If djk = > 0, we assign an edge between vjk and vjk for p = 1,2,...,1
by (d). We do a similar procedure for each vertex, vj, assigned to dj > 0. Therefore
vj, C A(vj) and vjP EA(vj,) for k= 1,2,...,m, j= 1,2,...,l.
(ii) If dj, = 0, then we do not assign any new edge for the vertex vi,. But it is
already joined to the vertex vj by an edge.
Let s = (vi, v2,..., vn). Then all the vertices in s are connected. Since each vertex
vi is assigned di new edges and >F>' di = n 1 from (b), we have n vertices and
n 1 edges. Then the set of all the edges assigned above, together with the vertex
set V, forms an ntree by Proposition 3.2.6. 1
Remark: The proof of Theorem 3.4.12 shows how to construct a tree that corre
spond to the given degree sequence satisfying DSC.
Example 3.4.13 Let D = (3,2,3,1,0,0,0,1,0,2,0,0, 0) be a nonnegative integer
sequence having 13 entries. Let V = {V1, v2,..., vn,} be a set of n different vertices.
V9
Y12 V13 V8 V7
VlO V1 V2 V3 V4 V5
Vl1 V6
Figure 35
(a) Clearly, Ei31 di = 12 and d, = 3 > 0. So we expect a 13tree.
(b) Since d(vi) = 3, there are 4 different integers, 1 < 11 = 2 < 12 = 10 < 13 =
13 < 14 = 14, such that d2+Ei=3(j1) = 2+(2+0111+01) =
0, d0io + E =11(di 1) = 2 + (1 1) = 0, and d13 = 0. We assign edges between
v, and each of v2, V10, v13.
(c) (i) Since v2 = 2, there 3 different integers, 2 < 3 < 8 < 10, such that d3 +
Si=4(d i1) = 3+(01 1 1)=0, and d8+(d91) = 11 =0. Weassign
edges between v2 and each of v3 and vs.
(ii) Since v0io = 2, there are 3 different integers, 10 < 11 < 12 < 13, such that
dll = 0 and d12 = 0. We assign edges between vl0 and each of vll and v12.
(iii) Since v13 = 0, we do not assign any new edge for v13. But the vertex v13
is connected to vi by an edge.
In a similar way, we get {v3, v4}, {v3, v6}, {v3, V7}, {Vs, vg}, and {v4, v5}. Let E be
the collection of all the edges above, that is,
E = {{vi, V2}, {Vi, V0io}, {Vi, V13}, {V2, V3}, {vJ2, V8}, {Vio0, V1},
{v10, V12}, {v3, v4}, {v3, v6}, {v3, v7}, {v8, v9}, {v4, v5}}.
Since the edge set E contains 12 elements, the set T = {V, E} is a 13tree.
3.5 Perfect Sequence and Main Theorem
In the previous section, we showed that, for a given ntree T, there exist many
vertex sequences depending on the starting vertex and choice among the adjacent
vertices for each successive vertex. Therefore the degree sequence set D for the
given tree T contains many different degree sequences which denote the same tree
T. We need a way to choose one degree sequence representing the tree T. First of
all we want to define an order on the set of all finite, nonnegative, integer sequences.
Definition 3.5.1 For any two finite nonnegative integer sequences D = (Pl, P2, ..., IPk)
and E = (ql, q2,..., qi), D < E if and only if either there is m < min{k, l} such
that pm
i = 1,2,...,k. We write D < E if D < E or D = E.
Proposition 3.5.2 Let D be the set of all degree sequences of an ntree. The degree
sequence set, (D, <), is a totally ordered finite set.
Proof For any two different degree sequences the order defined above is well defined.
Since each term di of a degree sequence is nonnegative and less than n and the
length of each degree sequence is n, the degree sequence set D is a finite set. Thus
(D, <) is a totally ordered finite set. I
Definition 3.5.3 For a given ntree T, let S be the vertex sequence set of T and
D the degree sequence set of T. Then the maximum degree sequence in D is called
the perfect sequence of the given tree T, and it is denoted by pf(T). For the perfect
sequence, pf(T), of a given tree, T, there is a vertex sequence s = (V1i, v2, ..., Vn) E
S such that the degree sequence of A is pf(T). We call A a perfect vertex sequence
of the tree and v, a perfect vertex of the tree.
In the Example 3.4.2, the perfect sequence is (3,1,0,0, 0) and there are two
different perfect vertex sequences, s\ = (r, s, t, p, q) and 412 = (r,s,t,q,p) for it.
But the trees represented by snl and s12 are isomorphic.
For any given tree the perfect sequence is unique by Proposition 3.5.2. But
there may exist several different perfect vertex sequences. If the given tree is a
labeled tree, then these perfect vertex sequences denote different labeled trees. But
these labeled trees are isomorphic to each other; that is, there is a unique unlabeled
tree. Therefore, for a given perfect sequence, there exists a unique unlabeled tree.
Theorem 3.5.4 Let dj be the perfect sequence of a tree T. If T' is any other tree
whose perfect sequence is also dj, then T' is isomorphic to T.
Proof If T has more than one vertex sequence with the same degree sequence, then
these vertex sequences all produce the isomorphic trees, by using the construction in
the proof of Theorem 3.4.12. 1
Because of this theorem, we have our main theorem below.
Theorem 3.5.5 (Tree Classification Theorem) For any positive integer n, let
T(n) be the set of unlabeled ntrees and P(n) the set of perfect sequences of ntrees.
Then there is a onetoone correspondence between T(n) and P(n).
3.6 Obtaining Perfect Sequences
By Theorem 3.5.5, if we can construct the set of perfect sequences of ntrees,
P(n), without being given any tree shape, then we can classify all the ntrees by this
set. And, in the same way as in Example 3.4.13, we can get the different tree shapes
from the elements of P(n).
Proposition 3.6.1 For the perfect vertex vl of a perfect vertex sequence A =
(vlV2,...,vn), we have d(vi) > d(vi) for i = 2,3,...,n.
Proof If there is some k {2,3,...,n} such that d(vk) > d(vi), then D =
(d(vi),...) < D, = (d(vk),...) for any vertex sequence = (Vk,...). This contra
dicts the maximality of Dj. _
Proposition 3.6.2 If D = (dj, d2, ..., d) is the perfect sequence of a given ntree
T, then n 1 > dl > di > dn = O forany i=2,3,...,n1.
Proof Since T is an ntree and Ds is the perfect sequence, we have n 1 > dl. For
any vertex u E V(T) we have d(vi) > d(u) by Proposition 3.6.1. Since di = d(vi)
and di = d(vi) 1 for every 1 < i < n, we have d, > cdi for 1 < i
must denote one of the end vertices, each of which has degree 1, we have dn = 0.
Therefore we have di > dn = 0 for every 1 < i < n 1. l
Definition 3.6.3 Let D = (di, d2,..., d) be a degree sequence. Then we say that
D satisfies the ordered subbranch condition, denoted OSbC, if and only if D satisfies
the following condition for all j> 1 :
If d(vj) = m + 1 > 2, and B'(vj,), B'(vj2), ..., B'(vj) with j < ji < j2 <
S m< JM are the subbranches of the vertex vj other than the root subbranch, then
(Bd'(vjk,,)) > (Bd'(vjk+l)) for every k = 1,2,...,m 1.
Recall that (Bd'(vj,)) is a section of the degree sequence D. (See Definition
3.3.3 and 3.4.1.)
Remark : If j = 1, then there is no root branch of vi; instead, there is one more
subbranch, Bd'(Vjm+,), such that jm < jm+1 and (B'(vjm)) > (Bd'(vj,,)).
Theorem 3.6.4 Let D = (dj, d2,... dn) be a nonnegative integer sequence. If D
is the perfect sequence of an ntree, then it satisfies the DSC and the OSbC.
Proof It is clear from Theorem 3.4.12 and the maximality of D. I
Proposition 3.6.5 Let T be a tree and v a vertex with the largest degree in T.
Then there is exactly one degree sequence D, for T starting at v and satisfying
both DSC and the OSbC.
By the above proposition, the converse of Theorem 3.6.4 is not true. The
following example is a counterexample.
Example 3.6.6 Let A = (dj,d2,...,d12) = (3,2,2,1,0,0,2,0,0,1,0,0). Then
Z'i di = 11 and d, = 3 > 0. And the sequence A can be separated into 4
sections like (3), (2,2,1,0,0,2,0,0), (1,0), (0). The second section can be sep
arated into 3 sections, (2), (2,1,0,0), (2,0,0), and the second and third sec
tion of these can be separated by (2), (1,0), (0) and (2), (0), (0), respec
tively. All of these subsections satisfy the second and third conditions of DSC.
Thus the sequence A satisfies the DSC. That is, the sequence A is a degree se
quence of a vertex sequence (Vi, v2, ... v12) of a 12tree by Theorem 3.4.12. On the
other hand, the sequence A satisfies the OSbC, since we have (2,2,1,0,0,2,0,0) >
(1,0) > (0). But A is not the perfect sequence of the 12tree. In fact, we rearrange
the sequence A to get B = (dr + 1,d2, d 1, dio, dn,d12,d3, d4, d5,d6, ds, d9) =
(3,2,2,1,0,0,2,1,0,0,0,0). Note that d(vi) = d, and d(v7) = d7 + 1. Clearly the
sequence B also satisfies the DSC and the OSbC. So B is also a degree sequence
of another vertex sequence (V7, v2, v1, V10, v11, V12, v3, v4, v5, v6, V8, vg) of the 12tree.
That is, A and B are degree sequences of the same 12tree (see Figure 36). And
we have
B= (3,2,2,1,0,0,2, 1,0,0,0,0) > (3,2,2, 1,0,0,2,0,0,1,0,0) = A.
Thus A is not the perfect sequence. Furthermore A cannot be the perfect sequence
of any tree, because A and B denote isomorphic trees, and B > A.
V11 V10 V1 V12
V9
V2
V7
V8
V5 V4 V3 V6
Figure 36
In this example we have that d(vi) = 3 = d(v7) which is the largest degree
in the tree. That is, the perfect sequence must start with the largest degree. We will
see in Example 3.6.8 that the degree sequence B is the perfect sequence of this tree.
Note that (1, 0) is the only possible degree sequence when d, = 1. Therefore
we assume d, > 1 from now on.
Now we show how to construct alternate degree sequences from A, similar
to B. Assume that we have a nonnegative integer sequence A = (di,d2,... ,d.)
satisfying the DSC and the OSbC. If there is 1 < i < n such that d, < di, then
A cannot be the perfect sequence of any tree by Proposition 3.6.2. So we assume
that d, > di for i = 2,3,... ,n, and there is at least one integer 1 < j < n such
that dj = cd 1. That is, there is an ntree T = (V, E) such that A is a degree
sequence of a vertex sequence A = (V1, V2,... V,v) of T, and T has at least two
vertices of maximum degree dl. Clearly, dj > 0. Thus vj is not an end vertex in
T. By Corollary 3.3.18, the set of the vertices of the root branch of vj is not the
set of vertices in A itself. So, from Proposition 3.3.17 there is j < m < n such
that RB(vj) = {Vi,V2,...,vj} U {Vm+i,Vm+2,. ..,Vn} is the set of the vertices of
the root branch of vj. Clearly, (Vj+i,Vj+2,... ,Vm) is the section containing all of
the subbranches of vj except the root branch. Therefore, if dj = p, then there are
p subsections of A like (Bd'(vj,)) > (Bd'(vj,)) > ... > (Bd'(vj,)) with j < ji <
j.2 <... < jp < m+1, since A satisfies OSbC. On the other hand, let E(RB'(vj)) =
{{u,v} E E(T)ju,v E RB'(vj)} be the set of edges in the root subbranch of vj. Then
T' = {RB'(vj),E(RB'(vj))} is a subtree of T. Let S(T') be the vertex sequence
set of T'. By Proposition 3.3.7, there is a unique integer 1 < k < j such that
Vk E A(v,). Clearly, vk E RB'(vj). Let Svk(T') be the vkvertex sequence set of
S(T'), in which the first term of each vertex sequence is vk. Let Dvk(T') be the
vkdegree sequence set of Svk(T') (see Definition 3.4.1). Since Dv,(T') is a finite
set, there is a maximum sequence in Dk (T'). We denote the maximum sequence in
Dvk (T') by mdR(vk). Note that d, must be changed to d, 1 in mdR(vk), since
it is not going to be used for the first term anymore in the new degree sequence.
Now we have that either mdR(vk) > (Bd'(vj,)), or there is an integer 1 < q < p
such that (Bd'(vjq)) > mdR(vk) > (Bd'(vjq,+)). Note that mdR(vk) may be the
smallest section, then vjq1 would not exist.
We rearrange the given degree sequence A to a new sequence in which dj + 1
is the first term and the subbranch degree sequences of vj,
mdR(vk), (Bd'(vj,)) (Bd'(vj2)) ..., (Bd'(vj)),
are listed in descending order. Denote this new degree sequence by MA.
Since the subbranch degree sequences of vj are listed in descending order in
MA the new degree sequence MA satisfies the OSbC. Since MA is a rearrangement
of A with di =d(vj) =dj+l1 >0, d, = d(vi) 1 = di 1 for some i> 1, and
A satisfies the DSC, the new degree sequence MA satisfies the conditions (a) and
(b) of the definition of the DSC. Since mdR(vk) is the perfect sequence of T', it
satisfies the conditions (c) and (d) of the definition of the DSC. Clearly, each section,
(Bd'(vj,)), satisfies the conditions (c) and (d) of the definition of the DSC. Therefore
the new degree sequence MA satisfies the DSC. Thus the new degree sequence MA
V) Vi
satisfies both the DSC and the OSbC.
For the given nonnegative integer sequence A = (dj, d2, ... ,dn), the set
{M\Id(vj) = d(vi)} is a nonempty finite set, since A E {MAId(vj) = d(vl)}.
Therefore there is a maximum element in this set. Let PA = max{ MA d(vj) = d(vi)}.
Now A determines a unique tree T. Suppose B is another vertex sequence for T
satisfying the DSC and the OSbC. Then {.1'Id(vq) = d(vi)} = {M d(v) = d(vI),
where Vq is an analogue of vj, by Proposition 3.6.5. Therefore PA is the perfect
sequence for T.
By the above explanation together with Proposition 3.6.2 and Theorem 3.4.12,
we have the following theorem.
Theorem 3.6.7 Let A = (di, d2,. ., d,) be a nonnegative integer sequence satisfy
ing the DSC and the OSbC with di > d, for i = 2,3,...,n. Then A = PA if and
only if A is the perfect sequence of an ntree.
In the following example, we show the degree sequence B in Example 3.6.6
is the perfect sequence of the tree.
Example 3.6.8 In Example 3.6.6,
A= (dl,d2,...,d12) =(3,2,2,1,0,0,2,0,0,1,0,0)
satisfies the DSC and the OSbC and d2 = d3 = d7 = di 1. That is, d(v2) = d(v3) =
d(v7) = d(vi). Thus max{MAIj = 2,3, 7} is the perfect sequence of the tree. For
v7 we have that RB(v7) = {v1, v2, .. ,v7} U {vo10, V11, V12} and (Bd'(vs)) = (0) =
(Bd'(v9g)). Thus T' = {RB'(v7),E(RB'(vr)} is the subtree produced by the root
subbranch of Vy. (See Figure 37.)
Since v7 C A(v2), we get the vrvertex sequence set
Sv,(T') = { (v2,Vi,v1OV11,V12,v3,v4,v5, v6)
(v2, Vl, v10, v11, v12, v3, v6, v4, v5)
(v2, Vi, v12, Vlo, 11, v3, v4, V5s, v6)
(v2, Vl, V12, Vlo, vii, V3, v6, v4, V5)
(v2, v3, v4, V5, v6, V1, Vio, vii, v12)
(v2, V3, V4, v5, v6, V1, V12, v10, vii)
(v2, v3, V6, v4, v5, V1, vio, Vll, v12),
(V2, v3, v6, V4, V5, Vl, V12, vo, V11) }.
Then the v7degree sequence set of S,,(T') is
D,7(T') ={ (2,2,1,0,0,2,1,0,0),
(2,2,1,0,0,2,0,1,0),
(2,2,0,1,0,2,1,0,0),
(2,2,0,1,0,2,0,1,0) ,
(2,2,1,0,0,2,1,0,0),
(2,2,1,0,0,2,0,1,0),
(2,2,0,1,0,2,1,0,0),
(2,2,0,1,0,2,0,1,0) }
= { (2,2,1,0,0,2,1,0,0),
(2,2,1,0,0,2,0,1,0),
(2,2,0,1,0,2,1,0,0),
(2,2,0,1,0,2,0,1,0) }.
Therefore we have the maximum sequence mdR(v7) in D, (T') as mdR(v7) =
(2,2,1,0,0,2,1,0,0). Since mdR(vr) = (2,2,1,0,0,2,1,0,0) > (0) = (Bd'(vs)) =
(Bd'(v9)), we have MA = (3,2,2,1,0,0,2,1,0,0,0,0).
In the same way, we get
MA = (3,2,1,0,0,2,1,0,0,2,0,0)
MV3 =(3,2,2,1,0,0,2,0,0,1,0,0).
Therefore max{Mt,,j = 2,3,7} = M17. Since
MA =(3,2,2,1,0,0,2,1,0,0,0,0)
> (3,2,2,1,0,0,2,00,0,1,0,0)
= A,
M4, which is B in the Example 3.6.6, is the perfect sequence of the tree.
VlI VIO V1 V12
V2
V5 V4 V3 V6
Figure 37
By the Tree Classification Theorem (Theorem 3.5.5) and the previous example,
we have the following important corollary.
Corollary 3.6.9 Let A = (d, d2,. .., dn) be a nonnegative integer sequence satis
fying the DSC and the OSbC. If d, 1 > di for every i = 2,3,..., n, then A is the
perfect sequence for an ntree.
We will see in the following example how we construct a tree shape from a
given maximal degree sequence.
Example 3.6.10 Let D = (di, d2,..., d0io) = (3,2,1,2,0,0,1,0,0,0) be a maximal
degree sequence. Then we expect a 10tree, T. Let s = (V1, v2,... vio) be the vertex
sequence having D as the degree sequence. Then d(vi) = d, = 3, so the vertex v,
has 3 edges. (See Figure 38.) Though there is no specific order for these edges, we
label them as eC, e2, e3 for the convenience of explanation.
e2
e3 V1 el
Figure 38
From the definition of vertex sequence (Definition 3.3.4) v2 is one of the
elements of the adjacent set of v1, A(vl). Without loss of generality, let el =
{vI,v2}. Then v2 has 2 'new' edges, e4 and e5, since d2 = 2. Let e4 = {v2,v3}.
By the same reasoning, v3 has 1 'new' edge, named e6. So e6 = {v3, v4}. Since
d4 = 2, V4 has 2 'new' edges, named e7 and e8. (See Figure 39.)
e2 e5 e8
e3 V1 el V2 e4 V3 e6 v4 e7
Figure 39
Since d5 = 0 and d6 = 0, neither v5 nor v6 has a 'new' edge; that is, V5
and V6 are end vertices. On the other hand, v5 and v6 are the adjacent vertices of
v4, since d4 = 2. Thus V7 is an adjacent vertex of one of {vi,v2,v3}. Since d3 = 1
and v4 is the adjacent vertex of v3, v3 is not an adjacent vertex of v7. Since
d2= 2 and we have only one adjacent vertex of v2, namely 1V3, we add an adjacent
vertex, v7. (See Figure 310.)
V7 V6
C2 C5 e8
e3 v, e1 V2 e4 v3 e6 V4 e7 V5
Figure 310
Since d7 = 1, v7 has 1 'new' edge which connects with v8. Since d8 = 0,
there is no 'new' edge from V8. Since we covered up to d2, the remaining two vertices,
v9 and v0io, are adjacent vertices of v, which has two 'unused' edges, e2 and e3.
Let e2 = {v1, v9}, and e3 = {vi, vIo}. Since we are thinking about an unlabeled tree
and we put the edge names for our convenience, there are no specific names for the
vertices and edges. Thus we get the following 10tree. (See Figure 311.)
(V8)
(V9) ()(V6)
(Vio) (V1) (V2) (V3) (V4) (V5)
Figure 311
3.7 Extended Partitions and the Algorithm
In Definition 3.5.3 and the following explanation, it was shown that any tree
can be represented by a unique perfect sequence. So, in this section, we develop an
algorithm to produce all the perfect sequences for the ntrees for any positive integer
n. From these perfect sequences, we get the shape of each ntree as shown in the
previous example, as well as the total number of ntrees.
From Definition 3.4.1 and the First Reduced Degree Sum Formula (Theorem
3.4.6), we see that a degree sequence for an (n + 1)tree always has n + 1 nonnegative
integers as entries and these integers always add up to n which is the number of
the edges of the given (n + 1)tree. Thus, our first step in listing all possible degree
sequences is to form ordered partitions of the integer n as sums of smaller positive
integers.
Definition 3.7.1 An ordered partition of the positive integer n is a sequence of
positive integers, (pi,P2,... ,Pm) such that E=1 pi = n.
Note that 1 < m < n in the definition above.
Definition 3.7.2 An extended partition of an ordered partition (p1,p2,. ,Pm) of
the positive integer n is a sequence of n + 1 nonnegative integers, (di, d2,..., ,n+)
such that
(i) di = pi, E+1 di = n, and
(ii) there is a subsequence (dj, dj2,..., dim) such that djk = Pk for k = 1,...,m.
Note that any entry di in the extended partition must be 0 unless it is a
member of the ordered partition. It is important that different extended partitions
are produced not only by the different ordered partitions but can also be produced
by the same ordered partition.
For example, let (di,d2,... dn+,i) and (di,d,.... ,dn+i) be two extended
partitions of an ordered partition (pi,p2,. ..,Pm), and let (djl, dj2,... ,dj,,) and
dl, d'2,... ,d'm) be subsequences, respectively, such that djl = d = =i = d\t and
djk =Pk = dt, for k = 2,3,... ,m. Assume that j, $ ti for at least one i. Then
there are sections of zeros with different lengths between d(kl) and dj, and d'
and d' respectively. So the extended partitions are different. Thus the order is
very important in an extended partition.
Also note that the length of the extended partition of n must be n + 1. So,
for any given ordered partition, (p, P2,... ,Pm), we need to add n m + 1 zeros to
make an extended partition.
Proposition 3.7.3 Let (dj, d2,... ,dn+,) be an extended partition of an ordered
partition P = (djl,dj2,... ,djm) of the integer n. Then this extended partition sat
isfies the First Reduced Degree Sum Formula (Theorem 3.4.6).
Proof From the given extended partition (di, d2,... dn+i) of an ordered partition
Ei"+l di dilJ, =n and
(dj dj2,..., d.,) of the integer n, we have that Z 1i = = n and
i + d ( n+1
~ i+ l2(d 1) d + Ek=2(dik 1) + ELi=, di.P(di 1) since d1 5 0
d= d+ Ek=2(dJ 1) + E 1, dp(d, 1)
+ n+1
= EiM_ d3k + E + E+ 2+=1, d p(1)
= n + (m 1)(l) + (n m + 1)(l),
=nm+ln+m 1=0.I
Even though we have a very large number of extended partitions of n, some of
them can not be a maximal degree sequence for an (n + 1)tree. So we first eliminate
some ordered partitions that could not possibly result in maximal degree sequences.
To do so, we need to check some properties of perfect sequences.
Proposition 3.7.4 Let Ds = (d1,d2,... ,dn+) be the perfect sequence of a given
(n + 1)tree T. If di, 0 and di+1 = 0, then 0 = di+1 = di+2 = ... = di+d,.
Proof Since di, 0, the vertex vi has di 'new' edges; that is, there are di adjacent
vertices of vi and they are labeled by d, consecutive integers after i. Since Di is
a perfect sequence, it satisfies the OSbC. That is, the vertex vi+1 is a vertex of the
largest degree among the adjacent vertices of vi. But di+1 = 0. Therefore we have
dj =0 for j = i + 2, i + 3,..., i + di. .
This proposition gives us a method to make an extended partition from an
ordered partition by adding zeros. That is, we know certain extended partitions
cannot be perfect sequences just by checking this property.
Proposition 3.7.5 Let Di = (dl, d2,. ., dn+) be the perfect sequence of a given
(n + 1)tree T. If n > 2, then d > 2.
Proof If d, = 1, then di = 0 for every i = 2,3,... ,n + 1 by Proposition 3.6.2.
Since n > 2, there are at least 3 vertices in T. Thus we have that di+1.=(di1) =
1_+21(1) < 1+(11) = 1 0. This is a contradiction to the First Reduced
Degree Sum Formula. Therefore d, > 2. 1_
According to the above proposition, the only trees whose perfect sequences
start with 0 or 1 are the tree with zero edge and the tree with one edge. The only
degree sequence representation of a tree with zero edge is (0), and the only degree
sequence representation of a tree with one edge is (1,0). These are easy enough to
visualize and count, so for the algorithm we work on the assumption that we are
listing trees with at least two edges. Then the first term of the perfect sequence will
be at least 2.
The following proposition shows that we have an arc shape tree if the first
term of the perfect sequence is 2.
Proposition 3.7.6 If Di = (dia, d2,..., dn+) is the perfect sequence of an (n+ l1)tree
T and d, =2, then di = 1 for i = 2,3,...,n 1 and d4 = dn+, =0.
Proof By Proposition 3.6.2 we have that di, E {0, 1} for any i = 2,3,... ,n + 1,
since di = 2. Then the possible highest degree in T is degree 2. So T is just a
linear tree with n edges. That is, there are exactly two end vertices which cannot be
v1. Therefore there are exactly two zeros in the sequence. Write a degree sequence
for T, in the order of the vertices in the following figure.
   
Vn+l Vi V2 V3 Vn1 Vn
Figure 312
That is, start at a vertex adjacent to an end vertex and move in the longer
direction first. Clearly the degree sequence resulting from this method contains
the two zeros for vn and Vn+l. Then we will get a degree sequence of the form
D = (2, 1,1,..., 1,0,0). For the same tree, the numbers in a degree sequence cannot
change, but only the order in which they occur can change by changing the position
of zeros. And, clearly, the new ordered sequence < D. So the sequence D must be
the perfect sequence of T. That is, Di = D. I
Most of the extended partitions of n start with a 1 or a 2. Since the last two
propositions classify trees whose degree sequences start with these values, we do not
need to consider ordered partitions that start with 1 or 2.
Proposition 3.7.6 shows that for any given number n of edges there is only one
tree with n edges whose perfect sequence starts with a 2. In particular, it is a linear
tree with n edges, since d, = 1 for i = 2, 3,..., n 1. We add this sequence at the
end of the algorithm.
By Proposition 3.6.2, we do not need the ordered partitions in which the largest
summand appears more than once or is not the first term in the ordered partition.
If an ordered partition has either one or two entries only, then we can produce
only one extended partition for each case. See the following Proposition 3.7.7 and
Corollary 3.7.8.
Proposition 3.7.7 Let D = (dl, d2,..., dn+) be the perfect sequence of an (n+l)
tree T with d2 = 0. Then d, =n and d, = 0 for all i > 1.
Proof: Let A be a vertex sequence yielding a perfect sequence Ds of an (n + 1)
tree T. Since Ds is a perfect sequence and 0 = d2 = d(v2) 1, all of the vertices in
the tree other than vi have degree 1, so all the terms in D, other than d, must be
zero. By the First Reduced Degree Sum Formula (Theorem 3.4.6), we have d, = n. 1_
Corollary 3.7.8 If an ordered partition P = (pl,p2) of n has only two entries,
then the produced extended partition, (di, d2,... ,dn+1), from P is such that d, =
p1, d2 =p2 and di = 0 for i = 3,4,...,n + 1.
We demonstrated several necessary conditions which apply to a perfect se
quence of a tree. Unfortunately, while these conditions are sufficient in almost all
cases, there are certain sequences which, although they satisfy all the conditions, are
not perfect sequences of the trees they represent. The problems usually arise when
the tree has two or more vertices of the largest degree. We saw a counterexample in
Example 3.6.6.
Note that in the Example 3.6.6 the two degree sequences A and B have the
same ordered partition, P = (3,2,2,1,2,1). That is, our algorithm produces the
degree sequence B as one of the extended partitions of the ordered partition P.
Therefore all we need to do is just drop the degree sequence A from the result of
our algorithm.
Each step described so far has involved checking each member of a list of
possible degree sequences and removing those which could not be the perfect sequence
of a tree. The last step is no exception. Given a degree sequence, we check if there is
more than one vertex of maximal degree. If there are more than one (in Example 3.6.6,
both vertices v, and v7 = u1 have the same degree, 3), we write degree sequences
for the tree starting at each such vertex. If the largest of these with respect to
the ordering < is the sequence we are checking, then it is a perfect sequence. If
some other vertex gives a larger degree sequence, it is false that the sequence we are
checking is a perfect sequence. This is a surefire approach in that, given a degree
sequence, it checks for a perfect sequence by considering every possible alternative
for the maximum and seeing if the sequence in hand is the largest.
CHAPTER 4
TREES AS INVARIANTS FOR BRAIN TUMOR SHAPES
4.1 Introduction
A sphere packing treatment plan for a given brain tumor is produced in the
first chapter. This plan can be represented numerically as a list of parameters of
the dose spheres. In particular, the list consists of the coordinates of the dose center
in millimeters along the lateral (Lat), anteriorposterior (AP), and axial (Ax) axes
together with the radius of the dose sphere in millimeters. From these data, we
can assign a unique corresponding graph by matching a sphere with a vertex and
matching the adjacency of two spheres with an edge. Then, by using the notion of
cutvertex, we give an order to the vertex set. We use this order to decide which edges
to choose in order to obtain a unique maximal tree contained in the graph.
Our main theorem of this section is Theorem 4.5.6: The perfect sequences are
invariants of the shapes of arbitrary brain tumors. We then develop a Mathematica
program (see Appendix B) to get a maximal tree of the graph for a sphere packing
plan for a given brain tumor. In the near future, we plan to use this program to
classify all the treatment plans which have been done at the Brain Institute of the
University of Florida. We hope that the classification gives some insight about brain
tumor shapes.
In this chapter, we assume that a brain tumor is connected, so the graph
representation for any brain tumor is a connected graph.
4.2 Definitions and Some Properties
Definition 4.2.1 A graph G = (V(G),E(G)) consists of a vertex set V(G) =
{Vi,.. .,vn} and an edge set E(G) = {ei,...,em} C {{u,v}lu,v E V(G)} in which
each edge is an unordered pair of two distinct vertices u and v, and the edge {u, v}
is said to join u and v. For an edge {u, v} we say that u and v are adjacent
vertices and the vertex u and the edge {u, v} are incident with each other, as are
v and {u,v}. From now on, IV(G)I denotes the number of vertices and jE(G)j
denotes the number of edges. For a vertex u V(G), the vertex set
AG(u) = {v E Gjv is an adjacent vertex of u}
is called the adjacent set of u in G.
Since there is no loop represented by a single edge, in a graph in this article,
no vertex is adjacent to itself.
Definition 4.2.2 We say that a graph G is isomorphic to a graph H if and only
if there exists a bijection f: V(G) * V(H) such that {u, v} C E(G) if and only if
{f(u),f(v)} I (H).
Definition 4.2.3 A walk is an alternating sequence (vo, el, vi, e2,..., ek, Vk) of ver
tices and edges s.t. vi1 J vi for i = 1,2,... ,k, and ei = {vii,vi} for every i. A
cycle is a walk (vo, el,v1, e2,..., ek, Vk) with k > 3 in which v0 = vk is the only
vertex repeated. A path is a walk with no repeated vertex. A (u,v)path is a path
with the first vertex u and the last vertex v. A (u,u)path is a vertex u. The length
of the (u,v)path, l(u, v), is the number of edges in the (u,v)path. So the length of
(u,u)path is 0. The distance d(u,v) between two vertices u and v is the length of
a shortest path joining them, if any; otherwise d(u, v) = oo.
Note that a path of length > 1 is of linear shape.
Definition 4.2.4 Let G = (V, E) be a graph where V = {vi,v2,...,v"n}. Then,
for any vertex v E V, we say that PG(v) = j if and only if
j = min{i {1,2,..., n}I there is a (v, vi)path in G}.
We call Pc(v) the pathvalue, or simply the Pvalue of the vertex v in G.
Definition 4.2.5 Let G = (V, E) be a graph. For any nonempth subset V' C V,
let F = { {u,v} E E I u,v E V } C E. For any subset F' of F, the graph
G' = (V', F') is called a subgraph of G. A graph G is connected if and only if every
pair of vertices are joined by a path. A graph having no cycle is acyclic. Note that an
acyclic graph can be disconnected. A forest is an acyclic graph. A tree is a connected
acyclic graph. An ntree is a tree with n vertices.
Definition 4.2.6 The degree of a vertex v, denoted by d(v), is the sum of the
number of edges incident with v.
Definition 4.2.7 A cutvertex of a graph is a vertex of degree greater than 2, whose
removal increases the number of components of the graph.
Definition 4.2.8 A nonseparable graph is connected, nontrivial, and has no cutver
tex. A block of a graph is a maximal nonseparable subgraph.
Thus if v is a cutvertex of a connected graph G, then G v is disconnected.
Note that the traditional definition of cutvertex (usually called cutpoint in graph
theory) does not have the degree condition. But we do not want to separate any
linear shape subgraph into several blocks because there is no difficulty in handling
linear shape subgraphs in our algorithm. Thus we do not call a vertex with degree
2, a cutvertex in this chapter, even though it separates the graph.
And if H is a block, then none of the vertices in H is a cutvertex of H.
Note that if there is no cutvertex in G, then the graph G itself is a block. This
definition differs from the definition of block in Harary [5], because we have a different
definition of cutvertex. Because of this, we have Proposition 4.2.10 below.
Proposition 4.2.9 If a graph G is of linear shape, then the maximal tree of G is
G itself.
Proposition 4.2.10 [5] Let v be a vertex of a connected graph G. The following
statements are equivalent:
(1) v separates G and d(v) > 2.
(2) There exist vertices u and w distinct from v such that v is on every (u,w)
path.
(3) There exists a partition of the set of vertices V(G) {v} into subsets U and W
such that for any vertices u E U and w W, the vertex v is on every (u, w)path.
Definition 4.2.11 Let G = {V, E} be a graph. For any vertex v E V we define
the edge subset IG(v) C E(G) by IaG(v) = {{v,u} E E(G)I u E AG(v)}. We call
IG(v) the incident edge set of v in G.
Clearly, Ia(v) contains all the incident edges of the vertex v in G.
Definition 4.2.12 Let G = {V, E} be a graph. For any subset F of E we call the
set V(F) = {v E V(G)I there is an edge e E F such that v and e are incident}
the F vertex set.
Proposition 4.2.13 [5] The following statements are equivalent for a graph T:
(1) T is a tree.
(2) Every two vertices of T are joined by a unique path.
(3) T is connected and IV(T)I = IE(T)I + 1.
(4) T is acyclic and JV(T)J = JE(T) + 1.
Note that if a tree has n vertices, then there are exactly n 1 edges.
Definition 4.2.14 Let G = {V, E} be a connected graph and E' a subset of E.
A connected subgraph T = {V, E'} is called a maximal tree of G if and only if
T'= {V, E'U {e}} contains a cycle for any edge e E E E'.
Our work in this chapter produces a particular maximal tree for any given
graph. So we have the following (wellknown) theorem.
Theorem 4.2.15 Every connected graph has a maximal tree.
63
Example 4.2.16 Let G be a graph with 14 vertices and 18 edges as in the figure
41 below.
a b c d e f
Figure 41
There are 3 cutvertices, b, c, e. So the given graph G can be separated into
6 blocks as follows:
G1 ={Vi,Ei}={{a,b }, {{a,b}}},
G2 = {V2, E2} = { { b,c, k, ,m,n },
{ {b,c}, {c, k}, {k, {, m}, {mn}, {rn, b}, {b, k}, {k, m} } },
G3 ={V3,E3}={{c,d,e}, {{c,d},{d,e}}},
G4 = {V4, E4} = { { e,f,g }, { {e,f},{f,g},{g,e} } },
G5 = {V5,E5} = { { e,h,i }, { {e,h},{h,i} } }, and
G6 ={V6, E6}={{e,j }, {{e,j}}}.
a b
b
c d e
G3
9
e f
G4
Figure 42
4.3 Graphs for Sphere Packing Plans
The sphere packing treatment plan S having n spheres for a given brain tu
mor, from Chapter 1, consists of the centers, c1, c2,..., cn, and the radii, ri, r2,..., r.,
of the spheres with ci = (xi, yi, zi) for every i, where x, is for the AP coordinate,
y, is for the Lat coordinate and zi is for the Ax coordinate of the center, ci. (Note
that the physicians use diameters, rather than radii, in their data sets.)
We consider the spheres to be the vertices of a graph, and we assign the vertex,
vi, for the sphere centered at ci with radius ri. Now we can assign an edge between
two vertices if the spheres assigned by these two vertices meet. In Chapter 1, we
noted that the collimators do not deliver clear cut boundaries for spheresthere is
some spillover. The neurosurgeons currently assume that the portion between two
spheres gets enough dosage of radiation if the distance between the centers of the
two spheres is less than or equal to eleven tenths of the sum of the radii of the two
spheres. Thus for any two vertices assigned for two spheres satisfying this condition
we assign an edge between them. Then we have the following proposition.
Proposition 4.3.1 Let G = {V,E} be a graph for the sphere packing plan of a
brain tumor. For any vertices vi and vj in V, d(ci,cj) < (ri + r) if and only if
{vi, vj} E E. Thus we have the following edge set;
E = {{i,j}Id(ci, cj) < (ll/10)(ri + rj) where i < j, i = 1,2,...,n 1}.
Since any brain tumor is connected in this article, there is at least one (u, v)
path in G for every two vertices, u and v in G. Thus we have the following propo
sition.
Proposition 4.3.2 The graph G = {V, E} produced by the above method for a
given tumor is a connected graph.
Since we are dealing with unlabeled graphs in this article, we have the following
proposition.
Proposition 4.3.3 For any sphere packing treatment plan for a given brain tumor,
there exists a unique graph that we assign to the treatment plan.
By means of the above method, we get a unique connected graph G = {V, E}
for the sphere packing treatment plan for the tumor. This graph does not have the
information about the positions of the dose spheres as an aspect of the structure and
looks only at how the spheres are attached. It is clear that the graph G does not
have a loop or multiple edges between two vertices, since every edge is assigned for
two different vertices which are assigned for two attached spheres.
4.4 Order in the Vertex Set
To get a maximal tree from a graph we may need to delete some edges in the
graph. Thus we need to label each vertex in order to choose certain edges to delete,
even though we are dealing with an unlabeled graph in this chapter. So there is no
specific meaning for this labeling except that it is only used to choose a maximal tree
for the graph.
Example 4.4.1 Let P be the sphere packing treatment plan (see Figure 43) for a
given brain tumor.
Figure 43
Then, by Proposition 4.3.3, there is a unique graph G for the sphere packing plan.
The graph is given below in Figure 44.
p  s
Figure 44
There are two different isomorphism classes, namely T1 and T2, of maximal trees
for the graph G. That is, we could choose either of these for a maximal tree to assign
the graph G. Since, in our algorithm, we want to get the same maximal tree for a
given graph every time, we need certain rules to get the same tree. (See Figure 45.)
T, T
Figure 45
In this chapter, the graph comes from a sphere packing treatment plan. That is,
the main shape of the brain tumor depends on the size of spheres and the connection
between spheres. Even though we cannot keep the information about the size of
spheres in the graph, it seems to us that the bigger the sphere is, the more it has a
chance to get attached to more spheres. So we assume that the larger degree vertices
play a more important role in classifying the tumor shapes into trees than the smaller
degree vertices. That is, by choosing a largest degree vertex first, the shape of the
tumor is apparently most closely preserved. Thus, in the previous example, we choose
the vertex q or s as the starting vertex; then we pick all the incident edges. Then
we get the tree T1. Therefore we want to choose the tree T1 for the maximal tree of
G. Note that the tree T2 produces a linear graph which does not show the shape of
the tumor as closely as T1.
But there are some vertices which are more important than the bigger degree
vertices. Recall the graph G in Example 4.2.16. Then, in the block G2, it is clear
that the vertex k has the largest degree, 5. But the cutvertices, b and c, play a
critical role in obtaining a maximal tree of G. So, to get a maximal tree for a graph
G, we want to start at the cutvertices of G first, instead of the vertices with the
largest degree. So we want to label the cutvertices first. On the other hand, there
67
are different kinds of 'cutvertices' in some graphs. The following example shows such
a case.
Example 4.4.2 Let G = {V, E} be the graph in Figure 46.
U3
Figure 46
Then there are two cutvertices of the graph G, namely u2 and uj0. Us
ing these two cutvertices, the graph G is separated into four subgraphs, namely
G1, G2, G3 and G4. (See Figure 47.)
1 G2
"II*10
U3 G3
U1
U2 U2
U9
U10 UIO/ \ U
U2
Figure 47
Then the subgraphs G1, G2, and G3 are blocks, but the subgraph G4 con
tains its own cutvertices, u4 and u8, which are not cutvertices of the graph G. We
separate the subgraph G4 into four subgraphs by using its cutvertices u4 and us.
(See Figure 48.) Note that there is no specific order in labeling the subgraphs. So
at this moment, we relabel the subgraphs of G by H1, H2, ..., H7.
U9
H2 U10 U10 / \
UU0 U8
u3 //H 5
H1 / /
H4
U2 U2 U2 U4
U8 U5 ___ u
U7
H6 u6 H7
 U5
U4 U4
Figure 48
Then, again, the subgraphs H1, H2, ..., H6 are blocks, but the subgraph H7
contains its own cutvertex, u6, which is neither a cutvertex of the graph G nor a
cutvertex of the subgraph G4. We separate this subgraph H7 into two subgraphs by
using U6. (See Figure 49.) And we relabel the subgraphs of G by B1, B2, ..., B8.
Therefore {u2, u10} is the set of cutvertices of the graph G. But {u4, us} is
the set of the cutvertices of a subgraph which is produced after the first separation
using the cutvertices of the graph, and u6 is the cutvertex of a subgraph which is
produced after the second separation.
For the purpose of labeling the vertices in a graph, we separate these cutvertex
sets for different levels, from each other.
Definition 4.4.3 Let G be the graph assigned for a given sphere packing plan
P. Then the vertex set V(G) of the graph G is the union of two disjoint subsets,
namely CI(G) and Cc(G), where Ci(G) = {v V(G)I v is a cutvertex of G}
69
and Cc(G) = V(G) CI(G). We call C,(G) the first step cutvertex set of G. For
every i = 2,3,..., IV(G)I, we define a subset Ci(G) of CcI(G) as the collection
of the vertices of Cc I(G), which are cutvertices of a subgraph produced after the
(i 1)th separation using the elements of Cii(G). Then we call C1(G) the ith
step cutvertex set of G, and let C.(G) = Cic(G) Ci(G). We assume that there
are finitely many vertices in a given graph. If there exists at least one cutvertex,
then there exists an integer 1 < k < n such that Ci(G) 54 0 for every i < k, and
Ci(G) = 0 for every i > k + 1. Then we call k the separation step constant of the
graph G, and let Cc(G) = (Ck)c(G).
If there is no cutvertex of G
separation step constant of G is 0.
B2 u10
U11
U3
U1
U2
U8
B6
then the graph is a block, and we say that the
tU4
Figure 49.
U9
U10 U8
B5
B4
U2 U4
U7
U6
 U5
Proposition 4.4.4 Let JV(G)j = n and let the separation step constant of G be
k > 0. Then there is a sequence of integers {ml,m2,...,mk} with 0 < mn1 < m2 <
... < rnk such that 1CI(G)I = mi, and \Ci(G)j = mi, min1, for i = 2,3,...,k.
^,
Proof Since k > 0, the first step cutvertex set Ci(G) is not empty. Thus there is
a positive integer, namely mi, such that ICi(G) = mi. For every i = 2,3,...,k,
let mi = mi1 + C1(G). For every i = 2,3,... k, the ith step cutvertex set Ci(G)
is not empty. So we have that mi > mi1, and ICi(G) = mi mi,. I
Now we label the vertices as follows;
(1) the vertices in Ci(G) are labeled by {1,2,...,mI},
(2) for every i = 2,3,..., k, the vertices in Ci(G) are labeled by {mii + 1, mii +
2,...,mi 1,mi}, and
(3) the vertices in CC(G) are labeled by {mk + 1, mk + 2,..., n}.
We arbitrarily name the vertices of Ci(G) and CC(G), and denote that CI(G)
= {Ui,U2,...,Uml}, Ci(G) = {Um,_i+I,Um,_i+2,...,Um,} for i = 2,3,...,k, and
CC(G) = {Umk+1,Umk+2, ... ,Un}. It is clear that for every vertex up E Ci(G) and
every vertex Uq E Cj(G) with i $ j, we have that p < q if and only if i < j.
And also, for every vertex up E Ci(G) for every i = 1,2,... k, and every vertex
Uq E Cc(G), we have that p < q. Assume that, for every vertex ui, the center and
the radius of the sphere which is assigned to that vertex ui, are ci = {xi, yi, zi} and
ri, respectively.
We now relabel the elements for each of the sets, C1(G), C2(G),..., Ck(G)
and CC(G), in the following way:
for every pair of vertices, ui, uj, in one of CI(G),C2(G),... ,Ck(G) and Cc(G),
i < j if and only if
(i) d(ui) > d(uj); or
(ii) If d(ui) = d(uj), then r, > rj; or
(ii) If d(ui) = d(uj) and r, = rj, then xi < xj; or
(iii) If d(ui) = d(uj), ri = rj, and x, = xj, then yi < yj; or
(iv) If d(ui) = d(uj), r = rj, x, = xj, and y = yj, then zi
Note that d(ui) represents degree of the vertex u, in G.
It is clear that for any two different spheres, at least one of the coordinates is
different. Thus any two distinct spheres can always be compared by the above order.
Therefore the above order for the vertices of the graph is well defined.
4.5 Maximal Tree from a Graph
In this section, we develop an algorithm that will uniquely select a maximal
tree from the graph which is produced for a given brain tumor. The choice is made
so that the tree will most closely resemble, in our opinion, the shape of the tumor 
that is, of the graph obtained for the tumor.
Let G = (V, E) be the graph for a given sphere packing plan. Let V(G) =
{u1, u2,..., uI. for some positive integer n, where all the vertices in V are labeled
by the method in the previous section.
Define a subgraph To = (V(G),Fo) with V(To) = V = V(G) and F0 =
0; that is, there is no edge in To. Then, for every i = 1, 2,... ,n, we have that
PTo(ui) = i. (See Definition 4.2.4.)
(1) For the vertex ul, let F1 = F0 U IG(ul). (See Definition 4.2.11.) Let T1 =
(V(G), F1). The subgraph (V(F1), F1) C T1 is a tree. Thus, if V(F1) = V(G), then
T, = (V(G), F1) = (V(F1), F1) is our maximal tree of G. Note that for every vertex
v E V(F1) we have that PTi (v) = 1, and for every vertex u, V V(F1), we have that
PT, (Uj) = J.
(2) Assume that V(F1) 5 V(G). For the vertex u2, there are two cases, that is,
PTi (u2) = 1 and PT1(U2) = 2. Now define the edge subset I'(u2) by I(u2) =
IG(U2) {{u2,v} V E\ PT1(v) = 1)}.
(a) If PT (u2) = 1, that is, there is a (ui,u2)path in the subgraph T1, then let
F2 = FiUIG(u2). Each edge in these {{u2, v} El PT, (v) 1}= {{u2, v} E Ely E
V(F1)} makes a cycle with {u1,u2} and {ui, V}, since PT, (U2) = 1. Thus there is
no cycle in the subgraph (V(F2),F2), so it is a tree. Let T2 = (V(G),F2) C G.
If V(F2) = V(G), then T2 = (V(G),F2) = (V(F2),F2) is our maximal tree of G.
Note that for every vertex v E V(F2) we have that PT2(V) = 1, and for every vertex
uj 3 V(F2) we have that PT2(Uj)= j.
(b) If PTi(u2) = 2, then there is no (Ul,U2)path in the subgraph T1.
(i) If there is a nonempty subset {u21,, U22,... U2} C V(IG(u2)) where PT1(U2,)
=1 for i = 1,2,...,t and 2i <2j for l
{{u2,u21}}. Then the subgraph (V(F2),F2) C G is a tree. Let T2 = (V(G),F2) C
G. If V(F2) = V(G), then T2 = (V(G),F2) = (V(F2),F2) is our maximal tree of
G. Note that for every vertex v E V(F2) we have that PT2(v) = 1, and for every
vertex uj g V(F2) we have that PT2(Uj) = j.
(ii) If none of the vertices in V(IG(u2)) has pvalue 1, that is, for every vertex
v E V(IaG(u2)) there is no (ul,v)path, then let F2 = F1 U IG(u2). Since PT,(w) = 1
for every vertex w E F1 and PT, (v) = 2 for every vertex v E IG(U2), the subgraph
(V(F2), F2) is a forest, but not a tree. Let T2 = (V(G),F2) C G. It is clear that
the subgraph T2 cannot be our maximal tree, even if we have that V(F2) = V(G),
because the subgraph T2 is disconnected. Note that for every vertex v E V(F2) =
V(F1 U IG(U2)) = V(F1) U V(Io(u2)) we have that PT2(v) = 1 if v E V(F1), and
PT2(v) = 2 if v E V(IG(u2)). And, for every vertex uj V(F2), we have that
PT2(UJ) j.
(3) By induction, at the jth step, the subgraph Tj_j = (V(G),Fji) has been
defined. We want to define the subgraph Tj C G. For the vertex uj, there are two
cases, that is, PT,_(uj) = j and PT,_I(uj) = q for some 1 < q < j. For every
i = 1,2,..., n, define
mj_(i) = min{l {1,2,... ,n}PT_,(ul) = i and ul E AG(u) }.
If the set {l E {1,2,...,n}juj E V(IG(uj)) and PT,_(ui) = i } is empty, then the
vertex u,_,,_(i) is not defined. So there is no edge {uj,Um,_(i)}.
(a) Let PT, (uj) = ; that is, for every i = 1,2,... ,j 1,j + 1,...,n, there is
no (uj,,ui)path in the subgraph Tji = (V(G), Fji). Let
Fj = Fj, U (U={ {uj,um ,(i)} })
Then the subgraph (V(F,),Fj) is a forest. Let Tj = (V(G), F,) C G. If V(F,) =
V(G) and IV(Fj)l = iFj + 1, then, by Proposition 4.2.13, the subgraph Tj =
(V(G), F,) = (V(F,), F,) is our maximal tree of G.
(b) Let PT,_(uj) = q for some 1 < q < j, that is, there is a (uq,Uj)path in the
subgraph Tj, = (V(G), Fji). Now let
Fj = Fj1 U (U'1, ,iq{ {U,,Uj _,(i)}})
Then the subgraph (V(Fj), F) is a forest. Let T, = (V(G),Fj) C G. If V(Fj) =
V(G) and IV(Fj)t = IFj + 1, then by Proposition 4.2.13, the subgraph Tj =
(V(G), Fj) = (V(F,), Fj) is our maximal tree of G.
Note that if u is an ith step cutvertex, then d(u) in G is same as d(u) in
Ti. Also note that, for every i = 1,2,...,n, {uiu2,...,ui} C V(Fi), that is,
V(G) = V(Fn). We will see in Theorem 4.5.3 that Tn = (V(G), Fn) = (V(Fn),F")
is our maximal tree of G. By the Tree Classification Theorem (Theorem 3.6.8), this
tree Tn can be denoted by a unique perfect sequence.
Example 4.5.1 Let G = (V,E) be a graph where V = {u,u2,.. .,u8}, which
are ordered as in the previous subsection, and E = {eli e2, ,..., e13} as in the figure
below. Clearly, the graph G does not have a cutvertex, so the order is decided by
only the sizes of the spheres and the coordinates of the centers of the spheres.
us e13 U?7 e2 "U1
U2 C5 U6
Figure 410
Let To = (V(G),Fo) be a subgraph of G = (V,E) where Fo = 0, as shown in
Figure 411.
U8 U7 U1
0 U
U4
U3. **K
*
Z2 Z6
Figure 411
Since Ia(ul) = {e1,e2}, let F = FoUIG(ul) = {e1,e2}. Then Y(F) = {u1,u5,u7}
Since V(Fi) = {u,u5,u7} 5 V(G), the tree (V(F1),F1) is not a maximal tree of
G. Let T1 = (V(G), F1) which is shown in Figure 412.
Us U7 e2 U1
e6l
U30 U4
UU5
Z2 U6
Figure 412
We have that mi(i) = i for i = 1,2,3,4,6, and 8, but mi(5) and m1(7) do not
exist. Since PT,(U2)= 2 and Aa(u2) = {u3,u4, u6}, let
F2 = F U (U,=3,4,6{{u2,Ui}})
= F1 U IG(U2)
= {e, e2} U {e3, e4, e5} = {fe1,e2, e3,e4,e5}.
Then V(F2) = {ui,u5,U7,U2,uu 4,U6U} $ V(G) and the subgraph (V(F2),F2) is a
forest, not a tree. Thus it cannot be a maximal tree of G. Let T2 = (V(G), F2).
(See Figure 413.)
U8 U7 62 U1
U4 /e
U3\4 U5
e3 6e4
U2 65 U6
Figure 413
Note that PT(Ul) = PT2(U) = PT2(U)= 1, PT2(U2) = PT2(U3) = PT2(U4) =
PT2(U6) = 2, and PT2(us) =8. Thus we have that m2(1) = 1,m2(2) = 2, m2(8) = 8.
Note that m2(i) does not exist for i = 3,4,5,6, 7. Since ul 0 A(u3) and PT2(U3) =
2, weget F3 = F2U{{u3, us8}} = {el,e2,6 e3,6 e4, e5, e7}. Since the subgraph (V(F3),F3)
is a forest, but not a tree, it is not a maximal tree of G. Define T3 = (V(G), F3).
(See Figure 414.)
U8 U7 2 U1
67 61
U3
63 64
U2 6e5 U6
Figure 414
Note that PT3(UI) PT3(US) = PT3(U7) = 1 and PT3(U2) = PT3(U3) = PT3(U4) =
PT3(U6) = PT3(Us) = 2. Thus we have that m3(1) = 5 and m3(2) = 4. Note that
mr3(i) does not exist for i = 3,4,..., 8. Since PT3(U4) = 2, we define F4 = F3 U
{{U4, U5}} = {61,6e2,6 e3,6 e4,6 e5,6 e7, e68}. Since V(F4) = {v1,v2,...,vs} = V(G) and
IV(F4) = 8 = IF41 + 1, the subgraph T = (V(F4), F4) is a maximal tree of G. (See
Figure 415.)
US U7 e2 U1
e7 el
U4
U3 5
U2 e5 U6
Figure 415
At this moment, we have that PT(Ul) = PT( 2) = ...= PT(us) = 1. Then,
for every ut V(G) with t > 4, we have that Ft = F4. Thus, for every t > 4, the
subgraph (V(Ft), Ft) is same as T = (V(F4), F4).
The next example shows how we get the maximal tree for a given graph
containing some cutvertices.
Example 4.5.2 Let G be the graph in Example 4.2.16. (See Figure 41.) Then
CI(G) = {b,c,e} and Cc (G) = {a,d,f,g,h,i,j,k,l,m,n}. In Ci(G), we have that
d(e) = 5, d(b) = 4 and d(c) = 3. On the other hand, in Cc(G), we have that
d(k) = 5, d(m) = d(n) = 3, d(d) = d(f) = d(g) = d(h) = d(l) = 2, d(a) = d(i) =
d(j) = 1. And C2 = 0. Thus we order the vertices as follows; vI = e, v2 = b, v3 = c
and v4 = k, {v5s,v6} = {m,n}, {v7,v8,v9,Vo,vll} = {d,f,g,h,l}, {v12,v13, v14} =
{a,i,j}. Assume that vs = m, v6 = n, v7 = d, V8 = f, v9 = g, vio = h, vi =
1, v12 = a, vi13 = i, v14 = j, which are decided by the sizes and centers of the
corresponding spheres. (See Figure 416.)
V5 V11 V13
V6" V4 Vl0 / V9
V12 V2 V3 V7 V1 V8
Figure 416
Since IG(v) = { {v= v}, {v1, vs}, {v1, v9}, {v1, vio}, {v1, v14} }, let F1 = IG(v1).
Then V(F1) = {vI,v7, v8, V9, Vio, V14}. Since V(F1) V(G), the tree T1 = (V(G),F1)
is not a maximal tree of G. Thus, for the cutvertex vi, we have the following sub
graph which is T1 = (V(G),F1). (See Figure 417.)
* V4l
" V4
*12 V2 3
Vl2 V2> V3
V13
V7
V9
V1 V8
V14
Figure 417
Since PT,(v2) = 2, and, for every vertex v E AG(v2) = {v3, v4, V6, v12}, PTi(V) 7 1,
let F2 = F1 U Ia(v2). The subgraph (V(F2), F2) is a forest, but not a tree. Thus it
cannot be a maximal tree of G. Now, for the cutvertex v2, we have the following
subgraph which is T2 = (V(G), F2). (See Figure 418.)
* VUll
2 V6 V3
V12 V2 V3
V10
V7
Figure 418
V9
V1 V8
V14
Since AG(v3) = {v2, v4, v7}, PT (V2) = PT2(V4) = 2 = PT2(V3), and PT2(v7) = 1, let
F3 = F2 U {{v3, V7}}. Then, for the cutvertex v3, we have the following subgraph
which is (V(F3),F3). Then the subgraph (V(F3), F3) is a tree. But V(F3) 5 V(G).
Thus it cannot be a maximal tree of G. We have the following subgraph which is
T3= (V(G),F3). (See Figure 419.)
. Vll
V12 V2 V3 V7 Vi V8
Figure 419
Since AaG(v4) {v V2, v3,V5, v6, V11}, and
PT3(V2) = PT,(V3) = PT 3(V6) 1 = PT3(V4),PT3(V) =5,PT3(vn) 11,
we define F4 = F3 U {{v4, V5}} U {{V4, v11}}. Then the subgraph (V(F4), F4) is a
tree. But V(F4) 5 V(G). Thus it cannot be a maximal tree of G. Then, for the
cutvertex v4, we have that the following subgraph which is T4 = (V(G), F4). (See
Figure 420.)
V5 V11 V13.
V6 V4 V10 V9
V12 V2 V3 V7 V1 V8
V14
Figure 420
For the vertices vs, v6, v7, vs, v9, there is no 'new' edge. Thus we get the same sub
graph as above, that is, T4 = T5 = T6 = T7 = T = Tg. But, for the cutver
tex vl0, we have that v13 e V(IG(vlo)) and PT,(V13) = 13 $ 1 = PT9(vio). Thus
F10o = F9 U { {vIo,v 13 } and V(Flo) = V(G). Therefore we have the following sub
graph which is T = (V(G), Flo) = (V(Flo), Flo). (See Figure 421.)
V5 V11 V13
V6' V4 Vo1 V9
V12 V2 V3 V7 V1l V8
SV14
Figure 421
Clearly, the subgraph T = (V(F1o), Fo10) does not include a cycle. Thus it is a max
imal tree of G. That is, for every t = 11, 12,13,14, we have that (V(Ft), Ft) =
(V(Flo), Flo). Therefore, the subgraph T = (V(F1o),Flo) = (V(G), Fio) is the max
imal tree of the graph G.
For this resulting tree T = (V(F1o),Flo), choose a vertex sequence s by
s = (V,V7,V3,v2,V4,V5, Vi,V6,v12,Vlo,V13,Vs,V9,V14). Then the sequence
P = (5,1,1,3,2,0,0,0,0,1,0,0,0,0)
is the degree sequence of A. By Theorem 3.4.12, P satisfies DSC. It is clear that
P also satisfies DSbC. By Corollary 3.6.10, P is the perfect sequence for the tree
T = (V(F1o), Fo).
Theorem 4.5.3 Let G = {V, E} be a graph for a sphere packing plan where
V = {u1, u2,..., un} with the order described in Section 4.4. Let the subgraph
(V(Fn),F,) of G come from the above procedure. Then V(F,) = V(G). There
fore (V(Fn), Fn) is a maximal tree of G and there is a unique perfect nsequence for
the tree (V(F,), F,).
Proof The subgraph (V(Fn), Fn) is connected. In fact, assume that the sub
graph (V(Fn),Fn) contains two disjoint subtrees, namely T, and T2. Then the
pathvalue for every vertex in T1 is different from the pathvalue for every ver
tex in T2. Since the graph G is connected, there is at least one edge e E E(G)
which connects the two subgraphs T1 and T2 and e V E(T1 U T2) = Fn. Let
{ {VJl, Vk }, {vj2, Vk2 },. .. {Vjq, Vkq} } be the collection of such edges where vj, E T1
and vkj C T2 for i = 1,2,. ., q. Without loss of generality, let j, < ji for every
i = 2,3,..., q. Then it is clear that {vjj, Vk} I Fj1 by the procedure of getting the
subgraph (V(Fn), F.). Then {vj,, vk } C F,. This is a contradiction. Therefore the
subgraph (V(F,), Fn) is connected. The pathvalue in the procedure of getting the
subgraph (V(Fn),Fn) guarantees that the subgraph (V(F,,),Fn) is acyclic. There
fore (V(F,), Fn) is a tree. It is clear that, for every i = 1,2,... n, the vertex set
V(Fi) contains the vertices u1, u2,...,i u. Thus V(Fn) contains all the vertices of
G, since V(Fn) D {u,u2,...,un} =, V(G). That is, V(Fn) = V = V(G). Hence
the subgraph (V(F,), F,) is a maximal tree of G. By the Tree Classification The
orem (Theorem 3.6.8), it is clear that there is a perfect nsequence for the tree
(V(F,), F). I
Definition 4.5.4 The tree shape of a brain tumor is the tree corresponding to that
tumor, or equivalently, its unique corresponding perfect sequence.
Definition 4.5.5 Two tumors have the same shape if and only if their corresponding
graphs obtained from their respective sphere packing plans are the same.
As in Chapter 2, for any brain tumor, we have a unique sphere packing plan
which can be denoted by a graph. From the previous theorem, we have a unique
perfect sequence for the graph. That is, we can assign exactly one perfect sequence to
each brain tumor. So, if two perfect sequences are distinct, then their corresponding
trees, and therefore their respective corresponding graphs and sphere packing, are
also distinct. That is, if two tumors are represented by distinct perfect sequences,
then their corresponding trees are not isomorphic. And their respective graphs and
sphere packing plans are not isomorphic. Thus, we may consider their shapes to be
distinct. Therefore we have the following as our main theorem of this section.
Theorem 4.5.6 The perfect sequences are invariants of the shapes of arbitrary brain
tumors.
4.6 Graph from a Maximal Tree
Let T = {V, E'} be a maximal tree for a graph G = {V, E} which is obtained
from the sphere packing plan of a given tumor or solid shape. We observe that we can
recover the original graph, G, from T. Since V(T) = V = V(G) and E(T) = E' C
E = E(G), we need to add the remaining edges to the edge set of the maximal tree to
get the graph G. Since the graph G comes from the sphere packing treatment plan,
each vertex v in G denotes a sphere of the sphere packing plan. Thus, for every
vertex vi E G, let c, = (xi, y1i, zi) and r, be the center and the radius of the sphere.
By Proposition 4.3.1, for any two vertices vi and vj in V if d(ci, cj) < (r, + r),
then the edge {vi, vj} is in the edge set, E(G), of the graph G. Thus
E'U {{vi,vs} E V x VI d(c2,cj) < "(ri +rj)} = E.
Therefore we get the graph G = {V, E}.
4.7 Algorithm for Our Maximal Tree
Let G = (V, E) be the graph for a sphere packing plan with IV[I = n.
STEP A Find all the ith step cutvertices of the graph G, for 1 < i < n.
STEP B Label the vertices in the vertex set V by the order explained in Section
4.4.
STEP C By induction, for each i = 1,2,...,n, get the subgraph (V(Fi),Fi) of
G. Then T = (V(Fn),Fn) is our maximal tree.
APPENDIX A
SHELLING PROGRAM FOR SPHERE PACKING
In this appendix, we provide the C++ program we developed to get the sphere
packing plan for a given tumor shape. The first section shows the program translating
the data from CT scan to the data of appropriate form for the main program, which
is in the second section. Note that this program does not include the score function
which was developed by Thomas Wagner. We refer the reader Appendix E for the
complete article on the sphere packing treatment plan.
A.1 Program for Data Reading
// rewrite data file into x,y,z format
// & produce the total # of data, x,y,z min & max
#include
#include
#include
#include
#include
main()
{ const int nofdata=3370; // comes from dataOOS.txt
int xposition[nofdata], yposition[nofdata], zposition[nofdata];
int q;
float a; float b; float c;
int k=0;
{ ifstream inClientFile("data003.txt", ios::in);
while (inClientFile >> a >> b >> c)
{ xposition[k] = int(floor(a)); // check the number
yposition[k] = int(floor(b)); // check the number
zposition[k] = int(floor(c)); // check the number
k++;
}
}
int xmin=xposition[O];
int xmax=xposition[O];
int ymin=yposition[O];
int ymax=yposition[O];
int zmin=zposition[O];
int zmax=zposition[O];
for (int i=O; i
{ if ( xposition[i] < xmin )
xmin = xposition[i];
83
else if ( xposition[i] > xmax )
xmax = xposition[i];
else if ( yposition[i] < ymin )
ymin = yposition[i];
else if ( yposition[i] > ymax )
ymax = yposition[i];
else if ( zposition[i] < zmin )
zmin = zposition[i];
else if ( zposition[i] > zmax )
zmax = zposition[i];
}
// output the result of centers
{ ofstream outClientFile("mM003.txt", ios::app);
{ outClientFile << nofdata << setw(10) << xmin << setw(5)
<< xmax << setw(5) << ymin << setw(5) << ymax
<< setw(5) << zmin << setw(5) << zmax << endl;
}
}
cout << nofdata << setw(5) << xmin << setw(5) << xmax
<< setw(5) << ymin << setw(5) << ymax << setw(5)
<< zmin << setw(5) << zmax;
cin >> q;
}
A.2 Program for Center Searching
// searching the centers for pat 2
#include
#include
#include
#include
#include
int square(int); // function prototype
int MAX(int,int,int); // function prototype
main()
{ const int nofdata=3370; // check the number from mM003.txt
const int xmin = 32; // check the number
const int xmax = 4; // check the number
const int ymin = 29; // check the number
const int ymax = 4; // check the number
const int zmin = 57; // check the number
const int zmax = 37; // check the number
const int xfix = 1xmin;
const int yfix = 1ymin;
const int zfix = 1zmin;
int xposition[nofdata], yposition[nofdata], zposition[nofdata];
84
int q;
float a; float b; float c;
int k=0;
{ ifstream inClientFile("dataOO3.txt", los::in);
while (inClientFile >> a >> b >> c)
{ xposition[k] = int(floor(a))+xfix;
yposition[k] = int(floor(b))+yfix;
zposition[k] = int(floor(c))+zfix;
k++;
}
const int x = max xmin + 3;
const int y = ymax ymin + 3;
const int z = zmax zmin + 3;
const int nc = x*y*z; // number of cells
int cc[nc]; // status value
for (int i=O; i
cc[i] = 3;
int xc[nc]; // xcoord. of c[nc]
int yc[nc]; // ycoord. of c[nc]
int zc[nc]; // zcoord. of c[nc]
// setting the x,y,zcoord. for each cell
for (int i=O; i
{ zc[i] = floor(i/(y*x));
xc[i] = floor((izc[i]*y*x)/x);
yc[i] = izc[i]*y*xxc[i]*x;
}
// setting the boundary of tumor by 1
for (int i=O; i
{ cc[zposition[i]*x*y + xposition[i]*y + yposition[i]] = 1;
}
// setting the initial of normal by 0
cc[0] = 0;
// setting the normal cell by 0
int h=0;
for (int j=O; j
{ for (int i=O; i
{ if ( cc[i] == 3 )
{ if (cc[i1]==Ol Icc[i+l]==Ol Icc[iy]==Ol Icc[i+y]==0)
{ cc[i] = 0;
h++;
}
else;
}
else;
if (h==0)
break;
else
h=0;
// setting the remaining tumor cell by 2
for (int i=0; i
{ if ( cc[i] == 3 ) // checking
cc[i] = 2; // checking
else; // checking
} // checking
// setting the boundary cells
int maxrad;
int nofmaxrad;
for (int j=O; j<=MAX(x,y,z)/2+l; j++)
{ int k=0;
for (int i=O; i
{ if (cc[i] == 1 II cc[i] == 2)
{ k++:
if ( cc[i1] == j I
I I cc[i+y]==j
{ cc[i] = j+1;
nofmaxrad = k;
I cc[i+l] ==j 1i cc[iy]==j
II cc[iy*x]==j II cc[i+y*x]==j )
else;
else;
maxrad = j;
if (k==0)
break;
else;
Checking the no. of nofmaxrad
Sif nofmaxrad > 100, then you have change the number
S(nofmaxrad >100)
cout << endl << endl << endl << endl << endl <<
<< endl << endl << endl << endl;
cin >> q;
else;
// pick the boxes with maxrad
int suprad[100];
int TotalNofBox4[100];
int TotalNofMax=0;
for (int i=O; i
{ if (cc[i]==maxrad)
{ suprad[TotalNofMax] =i;
TotalNofMax++;
}
else;
}
cout << TotalNofMax << endl;
cin >> q;
// Count for the number of centers of each case
for (int k=O; k
{ // reset TotalNofBox4[ I to 0
for (int i=O; i<100; i++)
{ TotalNofBox4[i] = 0;
}
TotalNofBox4[maxrad] ++;
int tcc[nc];
for (int i = 0; i
tcc[i] = cc[i];
// remove th<
for (int p=O:
{ int dist
if ( dist
tcc[p]
else;
// temporary cell notion
++i)
// reassign tcc[i] for 'for' loop
a biggest sphere
p
Sfloor(sqrt( square(yc[suprad[k]] yc[p])
+ square (xc [suprad [k] ]xc [p] )
+ square(zc[suprad[k]]zc[p]) ));
< maxrad )
= 0;
// setting the remaining tumor cell by 2
for (int i=O; i
{ if ( tcc[i] >= 1 )
tcc[i] = 2;
else;
// setting the boundary cells
for (int j=O; j<=MAX(x,y,z)/2+l; j++)
{ int 1=0;
for (int i=O; i
{ if (tcc[i] == 2)
{ if (tcc[i1] == j I I tcc[i+l] ==j II tcc[iy]==j
II tcc[i+y]==j II tcc[iy*x]==j II tcc[i+y*x]==j )
{ tcc[i] = j+1;
++I;
}
else;
}
else;
(1==0)
break;
else;
// check if there is more maxrad pixcel
for (int i=0; i
{ if ( tcc[i] == maxrad )
{ cout << endl << endl << endl <<
WARNING!!! WE HAVE MORE 'MAXRAD' PIXCEL AT ("
<< xc[i] << "," << yc[i] "," << zc[i] << ")"
<< endl;
cin >> q;
break;
else;
//remove the smaller or = sphere
for (int j=maxradl; j>1; j)
{ for (int i=0; i
{ if ( tcc[i] ==j )
{ TotalNofBox4[j]++;
// remove the sphere
for (int 1=0; l
{ int dist = floor(sqrt( square(yc[i]yc[l])
+ square(xc[i] xc[1])
+ square(zc[i]zc[l]) ));
if ( dist < j )
tcc[l] = 0;
else;
}
// setting the remaining tumor cell by 2
for (int i=0; i
{ if ( tcc[i] >= 1 )
tcc[i] = 2;
else;
}
// setting the boundary cells
for (int j=0; j<=MAX(x,y,z)/2+l; j++)
{ int 1=0;
for (int i=0; i
{ if (tcc[i] == 2)
{ if ( tcc[i1] == j II tcc[i+1] ==j 
tcc[iy]==j II tcc[i+y]==j II
tcc[iy*x]==j II tcc[i+y*x]==j )
{ tcc[i] = j+1;
++i;
}
else;
}
else;
(1==0)
break;
else;
else;
output the result of centers
ofstream outClientFile("patOO3NofCenters.txt", ios::app);
{ outClientFile << endl << endl << endl << endl << endl <<
(k+1) << setw(5) << "(" << (xc[suprad[k]]xfix) << ","
<< (yc[suprad[k]]yfix) << "," << (zc[suprad[k]]zfix)
<< ") with radius = << maxrad << endl;
for (int j=maxrad; j>l; j)
{ ofstream outClientFile("patOO3NofCenters.txt", ios::app);
{ outClientFile << setw(4) << j << setw(4)
<< TotalNofBox4[j] << endl;
// Search for centers
for (int k=O; k
{ int tcc[nc]; // temporary cell notion
for (int i = 0; i
tcc[i] = cc[i]; // reassign tcc[i] for 'for' loop
// output the result of centers
{ ofstream outClientFile("patOO3centers.txt", ios::app);
{ outClientFile << (k+1) << endl << setw(5) << "x"
<< setw(5) << "y" << setw(5) << "z" << setw(10)
<< "radius" << endl << setw(5) << (xc[suprad[k]]xfix)
<< setw(5) << (yc[suprad[k]]yfix) << setw(5)
<< (zc[suprad[k]]zfix) << setw(10) << maxrad << endl;
}
}
// remove the biggest sphere
for (int p=O; p
{ int dist = floor(sqrt( square(yc[suprad[k]]yc[p])
+ square(xc[suprad[k]]xc[p])
+ square(zc[suprad[k]]zc[p]) ));
if ( dist < maxrad )
tcc[p] = 0;
else;
// setting the remaining tumor cell by 2
for (int i=0; i
{ if ( tcc[i] >= 1 )
tcc[i] = 2;
else;
// setting the boundary cells
for (int j=O; j<=MAX(x,y,z)/2+l; j++)
{ int 1=0;
for (int i=O; i
{ if (tcc[i] == 2)
{ if ( tcc[i1] == j II tcc[i+l] ==j I
II tcc[i+y]==j II tcc[iy*x]==j II
{ tcc[i] = j+1;
++i;
I tcc[iy]==j
tcc[i+y*x]==j )
else;
else;
}
if (1==0)
break;
else;
// check if there is more maxrad pixcel
for (int i=O; i
{ if ( tcc[i] == maxrad )
{ cout << endl << endl << endl <<
WARNING!!! WE HAVE MORE 'MAXRAD' PIXCEL AT ("
<< xc[i] "," << yc[i] << "," << zc[i] << ")" << endl;
cin >> q;
break;
}
else;
}
//remove the smaller or = sphere
for (int j=maxradl; j>1; j)
{ for (int i=0; i
{ if ( tcc[i] ==j )
{ TotalNofBox4[j]++;
// output the result of centers
{ ofstream outClientFile("patOO3centers.txt",
ios::app);
{ outClientFile << setw(5) << (xc[i]xfix)
<< setw(5) <<(yc[i]yfix) << setw(5)
<< (zc[i]zfix) << setw(10) << j << endl;
}
// remove the sphere
for (int 1=0; l
{ int dist = floor(sqrt( square(yc[i]yc[l])
+ square (xc [i] xc []l)
+ square (zc[i]zc[l]) ));
if ( dist < j )
tcc[l] = 0;
else;
}
// setting the remaining tumor cell by 2
for (int i=O; i
{ if ( tcc[i] >= 1 )
tcc[i] = 2;
else;
}
// setting the boundary cells
for (int j=O; j<=MAX(x,y,z)/2+l; j++)
{ int 1=0;
for (int i=O; i
{ if (tcc[i] == 2)
{ if ( tcc[i1] == j I I tcc[i+1] ==j I I
tcc[iy]==j II tcc[i+y]==j II
tcc[iy*x]==j I I tcc[i+y*x]==j )
{ tcc[i] = j+1;
++i;
else;
else;
else;
if (1==o)
break;
else;
}
else;
}
ofstream outClientFile("patOO3centers.txt", ios::app);
{ outClientFile << endl;
int square(int a)
{ return a*a;
}
int MAX(int a, int b, int c)
{ int p=0;
p = a;
if (a<=b)
p = b;
else if (a<=c)
p = c;
else;
return p;
APPENDIX B
PROGRAM FOR THE MAXIMAL TREE OF A GRAPH
In this appendix, we provide the Mathematica program we developed in joint
work with Matthew Harvey. The program below was written by Matthew Harvey to
get the maximal tree for the graph which we assigned for a brain tumor. Note that
this program does not use the method of ordering for the vertices as in Chapter 3.
Instead this program uses some commands which are in the function package in the
Mathematica program library. If anyone wants to develop a program like this using a
different program language, we recommend the vertex ordering algorithm in Chapter
3.
<< DiscreteMath'Combinatorica'
Off[General: :spell]
LexPrecedes[a_, b_] := Module[{i},
For[i = 1, i <= Min[Length[a], Length[b]], i++,
If[ a[[i]] != b[[i]],
If[ a[[i]] > b[[i]], Return[True], Return[False] ] ]
1;
If[ Length[b] > Length[a], Return[False] 1;
Return [True]]
DegreePoint[a_, k_] := Count[Flatten[ToUnorderedPairs[a]], k];
IndSubgraph[Graph[g_, v_], s._List] := Module[{i, j, newmat, sc},
newmat = g;
sc = Complement[Range[Length[v]], s] ;
For[i = 1, i <= Length[g], i++,
For[j = I, j <= Length[g], j++,
If[MemberQ[sc, i] II MemberQ[sc, j], newmat[[i, j]] = 0]]];
Graph[newmat, v]];
CutPoints[gr_] :=
Select[ArticulationVertices[gr], DegreePoint[gr, #] > 2 &]
NonCutPartitions[gr_] := Module[{partitions, cpts, bccs, skeleton},
partitions = {};
bccs = Select[BiconnectedComponents[gr], Length[#] > 2 &] ;
partitions = Union[partitions, bccs];
skeleton =
IndSubgraph[gr,
Union[
Complement[ Range[V[gr]], Flatten[bccs] ],
CutPoints [gr]
]
];
partitions =
Union [partitions,
Select[ConnectedComponents [
FromUnorderedPairs[
Select[ToUnorderedPairs[
skeleton], # !=
Intersection[#,
Flatten[
Select[BiconnectedComponents[skeleton],
Length[#] > 2 &]
]
&]
]
], Length[#] > 1
&I
] ;
Table[
Select[
ToUnorderedPairs[gr],
# == Intersection[ #, partitions[[i]] ]
&], {i, Length[partitions]}]]

Full Text 
PAGE 1
$ &/$66,),&$7,21 2) 75((6 $1' $33/,&$7,216 2) 7232/2*< $1' *5$3+ 7+(25< 72 1(852685*(5< %\ 7$(,/ <, $ ',66(57$7,21 35(6(17(' 72 7+( *5$'8$7( 6&+22/ 2) 7+( 81,9(56,7< 2) )/25,'$ ,1 3$57,$/ )8/),//0(17 2) 7+( 5(48,5(0(176 )25 7+( '(*5(( 2) '2&725 2) 3+,/2623+< 81,9(56,7< 2) )/25,'$
PAGE 2
&RS\ULJKW E\ 7$(,/ <,
PAGE 3
72 0\ SDUHQWV %HRP
PAGE 4
$&.12:/('*(0(176 ZRXOG OLNH WR H[SUHVV P\ VLQFHUH DSSUHFLDWLRQ WR WKH PHPEHUV RI P\ VXSHUYLn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fQHZ ZRUOGf WR PH ZRXOG OLNH WR WKDQN 'U 0HHNV IRU KHOSIXO FRQYHUn VDWLRQV DQG 'U 7KRPDV :DJQHU ZKR JDYH PH WKH LQIRUPDWLRQ DERXW QHXURVXUJHU\ WKDW GLG QRW NQRZ EHIRUH ZRXOG OLNH WR WKDQN 'U
PAGE 5
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
PAGE 6
$33(1',&(6 $ 6+(//,1* 352*5$0 )25 63+(5( 3$&.,1* $O 3URJUDP IRU 'DWD 5HDGLQJ $ 3URJUDP IRU &HQWHU 6HDUFKLQJ % 352*5$0 )25 7+( 0$;,0$/ 75(( 2) $ *5$3+ & 352*5$0 )25 &/$66,)<,1* 75((6 %< 6(48(1&(6 &O 3URJUDP & 6RPH 3URJUDP 5HVXOWV /$%(/(' *5$3+6 $1' 75((6 'O 1XPEHU RI /DEHOHG *UDSKV DQG /DEHOHG 7UHHV 6NHWFK RI WKH &RQVWUXFWLRQ RI D 3ULLIHUFRGH 5HFRQVWUXFWLRQ RI D 7UHH IURP D 3ULLIHUFRGH 5()(5(1&(6 %,2*5$3+,&$/ 6.(7&+ YL
PAGE 7
$EVWUDFW RI 'LVVHUWDWLRQ 3UHVHQWHG WR WKH *UDGXDWH 6FKRRO RI WKH 8QLYHUVLW\ RI )ORULGD LQ 3DUWLDO )XOILOOPHQW RI WKH 5HTXLUHPHQWV IRU WKH 'HJUHH RI 'RFWRU RI 3KLORVRSK\ $ &/$66,),&$7,21 2) 75((6 $1' $33/,&$7,216 2) 7232/2*< $1' *5$3+ 7+(25< 72 1(852685*(5< %\ 7DH,O
PAGE 8
VXUJHU\ SODQ $GGLWLRQDO FRGH E\ WKH PHGLFDO JURXS ZDV QHFHVVDU\ WR DOORZ IRU SK\VLn FLDQ LQSXW LQ FHUWDLQ FULWLFDO FDVHV ,W ZDV WKHQ SRVVLEOH IRU WKH PHGLFDO JURXS WR LQFRUSRUDWH WKLV LQWR WKHLU VXUJHU\ SODQ VHPLDXWRPDWLQJ LW DQG VLJQLILFDQWO\ UHGXFn LQJ WKH VXUJHRQVf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
PAGE 9
&+$37(5 ,1752'8&7,21 7KH ZRUN RI WKLV GLVVHUWDWLRQ ZDV PRWLYDWHG E\ WZR TXHVWLRQV UDLVHG E\ 'U )UDQFLV %RYD RI WKH %UDLQ ,QVWLWXWH 6KDQGV +RVSLWDO DQG WKH 'HSDUWPHQW RI 1HXURn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fV PHGLFDO SK\VLFV VWXGHQW ZKR ZRUNHG ZLWK XV :H H[SHFW WR KDYH WKUHH PDWKHPDWLFV SXEOLFDWLRQV IURP WKLV GLVVHUWDWLRQIURP &KDSWHUV DQG UHVSHFWLYHO\ 7KH SDSHU UHVXOWLQJ IURP &KDSWHU ZLOO EH MRLQW ZLWK 3URIHVVRU %HYHUO\ %UHFKQHU 3URIHVVRU
PAGE 10
,Q WKLV GLVVHUWDWLRQ ZH GHYHORS VRPH DOJRULWKPV DQG SURJUDPV UHODWHG WR WKHVH TXHVWLRQV ,Q VWHUHRJUDSKLF UDGLRVXUJHU\ ZKLFK LV QRQLQYDVLYH VXUJHU\ UDGLDWLRQ LV GHn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fKRW VSRWVf ZKLFK KDYH EHHQ H[SHULPHQWDOO\ GHWHUPLQHG WR FDXVH FRPn SOLFDWLRQV DQG ZH GR ZDQW UDSLG IDOORII RI GRVH OHYHOV RXWVLGH WKH DFWXDO WXPRU 2QH RI VHYHUDO VXFK UDGLDWLRQ VXUJHU\ PHWKRGV LV FDOOHG WKH f0XOWLSOH ,VRFHQWHU 0HWKRGf 7KLV LQYROYHV ILOOLQJ WKH WXPRU LPDJH ZLWK VSKHUHV RI GLIIHUHQW VL]HV XQWLO WKH LPDJH LV EHVW ILOOHG XS ,Q WKLV GLVVHUWDWLRQ ZH FDOO VXFK D SODQ D f6SKHUH 3DFNLQJ 7UHDWPHQW 3ODQf ,Q &KDSWHU ZH GHYHORS ERWK DQ DOJRULWKP ZKLFK LV FDOOHG D fVKHOOLQJ SURFHVVf RU fJUDVVILUH SURFHVVf LQ PHGLFDO WHUPLQRORJ\f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f 6LQFH WKLV SURJUDP DOZD\V SURGXFHV WKH VDPH SODQ IRU WKH VDPH WXPRU LW FDQ EH XVHG DV D fEHQFKPDUNf DJDLQVW ZKLFK WR FRPSDUH RWKHU WUHDWPHQW SODQV ,Q &KDSWHU ZH REWDLQ D FRGH WR GHQRWH DQ XQODEHOHG WUHH E\ D XQLTXH RUGHUHG QRQQHJDWLYH LQWHJHU VHTXHQFH VDWLVI\LQJ FHUWDLQ SURSHUWLHV :H FDOO VXFK
PAGE 11
VHTXHQFHV SHUIHFW VHTXHQFHV 7KDW LV ZH REWDLQ D FRPSOHWH FODVVLILFDWLRQ RI XQODEHOHG QWUHHV E\ PHDQV RI WKH SHUIHFW VHTXHQFHV RI Q QRQQHJDWLYH LQWHJHUV 7KXV ZH KDYH RXU PDLQ WKHRUHP RI WKLV FKDSWHU WKH 7UHH &ODVVLILFDWLRQ 7KHRUHP 7KHRUHP f )RU DQ\ SRVLWLYH LQWHJHU Q OHW 7Qf EH WKH VHW RI XQODEHOHG QWUHHV DQG n3Qf WKH VHW RI SHUIHFW VHTXHQFHV RI QWUHHV 7KHQ WKHUH LV D RQHWRRQH FRUUHVSRQGHQFH EHWZHHQ 7Qf DQG 9^Qf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f JLYHV ULVH WR WKH VDPH WUHH DQG WKHUHIRUH WR WKH VDPH SHUIHFW VHTXHQFH 7KXV ZH GR QRW \HW KDYH D UHDO FODVVLILFDWLRQ RI WXPRU VKDSHV +RZHYHU LI WZR SHUIHFW VHTXHQFHV DUH GLVWLQFW WKHQ WKHLU FRUUHVSRQGLQJ WUHHV DQG WKHUHIRUH WKHLU UHVSHFWLYH FRUUHVSRQGn LQJ JUDSKV DQG VSKHUH SDFNLQJ SODQV DUH DOVR GLVWLQFW 7KHUHIRUH ZH PD\ FRQVLGHU
PAGE 12
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
PAGE 13
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fKRW VSRWVf ZKLFK KDYH EHHQ H[SHULPHQWDOO\ GHWHUPLQHG WR FDXVH FRPSOLFDWLRQV DQG ZH GR ZDQW UDSLG IDOORII RI GRVH OHYHOV RXWVLGH WKH DFWXDO WXPRU 2QH RI VHYHUDO VXFK UDGLDWLRQ VXUJHU\ PHWKRGV LV FDOOHG WKH f0XOWLSOH ,VRFHQWHU 0HWKRGf 7KLV LQYROYHV ILOOLQJ WKH WXPRU LPDJH ZLWK VSKHUHV RI GLIIHUHQW VL]HV XQWLO WKH LPDJH LV EHVW ILOOHG XS 7KLV QRQLQYDVLYH PHWKRG RI VXUJHU\ QDPHO\ E\ XVLQJ UDGLDWLRQ UHOLHV RQ D SLHFH RI HTXLSPHQW FDOOHG WKH /LQHDU $FFHOHUDWRU RU VLPSO\ /LQDFf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f
PAGE 14
ZDYH JXLGH 7KH HOHFWURQV DUH WKHQ FKDQJHG LQ GLUHFWLRQ E\ WKH PDJQHW VR WKDW WKH\ LPSDFW RQ D KHDY\ PHWDO DOOR\ WDUJHW 7KLV UHVXOWV LQ ;UD\ SURGXFWLRQ WKDW FDQ WKHQ EH FROOLPDWHG RU VKDSHG E\ ERWK SULPDU\ DQG VHFRQGDU\ FROOLPDWRUV ZLWKLQ WKH OLQHDU DFFHOHUDWRU KHDG 7KLV EHDP LV IXUWKHU FROOLPDWHG IRU UDGLRVXUJHU\ E\ WKH WHUWLDU\ UDGLRVXUJHU\ FROOLPDWRU 6HH )LJXUH DQG )LJXUH f 7KH /LQDF LV PRXQWHG RQ D URWDWLQJ JDQWU\ VXFK WKDW WKH EHDP KDV D FHQWHU RI URWDWLRQ DERXW P DERYH WKH IORRU 8VXDOO\ WKH LVRFHQWHU DFFXUDF\ LV GHILQHG ZLWKLQ D PP VSKHUH %HFDXVH VWHUHRWDFWLF UDGLRVXUJHU\ GHSHQGV RQ RSWLPL]HG DFFXUDF\ DQ LPSURYHG V\VWHP ZDV GHVLJQHG DW WKH 8QLYHUVLW\ RI )ORULGD E\ DGGLQJ D VHW RI EHDULQJV WR WKH VWHUHRWDFWLF FROOLPDWRU V\VWHP DQG XQGHU WKH SDWLHQW WDEOH $V D UHVXOW WKLV QHZ V\VWHP DFKLHYHV PHFKDQLFDO DFFXUDF\ ZLWKLQ PPsPP IRU GHILQLQJ WKH WUHDWPHQW LVRFHQWHU RI EHDP GHOLYHU\ 7KH WHUWLDU\ FROOLPDWRUV DUH JHQHUDOO\ FLUFXODU DQG DOORZ LPSURYHG FHQWHULQJ RI WKH WUHDWPHQW EHDP 7KH VL]HV RI WKHVH FROOLPDWRUV DUH IURP WR PP LQ WR PP LQFUHPHQWV 3URFHGXUH RI 5DGLRVXUJHU\ 7KH SURFHGXUH RI UDGLRVXUJHU\ LV VHSDUDWHG LQWR VWHSV DV IROORZV f +HDG ULQJ DWWDFKPHQW f &7 VFDQ f ,PDJH 3URFHVVLQJ f 3ODQQLQJ f 7UHDWPHQW GHOLYHU\ )RU PRUH GHWDLOHG LQIRUPDWLRQ UHIHU WR )ULHGPDQ HW DO>@ f +HDG ULQJ DWWDFKPHQW 7KH VWHUHRWDFWLF UDGLRVXUJHU\ WUHDWPHQW UHTXLUHV DWWDFKPHQW RI D VWHUHRWDFWLF KHDG ULQJ 7KLV ULQJ SURYLGHV DFFXUDWH LQIRUPDWLRQ DERXW WKH SRVLWLRQ DQG WKH VKDSH
PAGE 15
)LODPHQW :DYHJXLGH (OHFWURQ EHDP 3RZHU VXSSO\ ? WL fÂ§0 +,+,+fÂ§ :KU 6HFRQGDU\ $GMXVWDEOHf FROOLPDWRUV 7HUWLDU\ 5DGLRVXUJHU\f FROOLPDWRUV 3KRWRQ EHDP WRU UDGLRVXUJHU\ )ORRU VWDQG ZLWK WUHDWPHQW DUP DERYH )LJXUH 6\VWHP RI /LQDF )ULHGPDQ HW DO>@f
PAGE 16
)LJXUH 2SHUDWLRQ RI /LQDF )ULHGPDQ HW DO>@f
PAGE 17
RI EUDLQ WXPRUV IURP WKH FRPSXWHU WRPRJUDSK\ &7f 7KLV ULQJ DOVR KHOSV WR LQVXUH WKDW WKH WUHDWPHQW WDUJHW LV DFFXUDWHO\ SODFHG DW WKH SUHFLVH LVRFHQWHU RI WKH /LQDF f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f DQWHULRUSRVWHULRU $3f DQG D[LDO $[f FRRUGLQDWHV 7KHVH DUH WKH [ \ DQG ] FRRUGLQDWHV UHVSHFWLYHO\ 7KH GLUHFWLRQ RI WKH ODWHUDO FRRUGLQDWH LV IURP WKH OHIW KDQG VLGH HDU WR WKH ULJKW KDQG VLGH HDU RI WKH SDWLHQW 6R WKH ODWHUDO YLHZ RU VDJLWWDOf LV VLPLODU WR WKH VLGH YLHZ RI WKH KHDG WKDW LV YLHZV RI VOLFHV DORQJ WKH ODWHUDO D[LV 7KLV LV WKH VDPH DV \]SODQH YLHZV 7KH GLUHFWLRQ RI WKH $3 FRRUGLQDWH LV IURP WKH EDFN VLGH RI WKH KHDG WR WKH IDFH 6R WKH $3 YLHZ RU FRURQDOf LV VLPLODU ZLWK WKH IURQW YLHZ RI WKH KHDG WKDW LV YLHZV RI VOLFHV DORQJ WKH $3 D[LV 7KLV LV WKH VDPH DV []SODQH YLHZV 7KH GLUHFWLRQ RI WKH D[LDO RU YHUWLFDOf FRRUGLQDWH LV IURP WKH QHFN WR WKH WRS RI WKH KHDG 6R WKH D[LDO YLHZ RU WUDQVD[LDOf LV VLPLODU WR WKH VN\ YLHZ RI WKH KHDG WKDW LV YLHZV RI VOLFHV DORQJ WKH D[LDO D[LV 7KLV LV WKH VDPH DV [\SODQH YLHZV 6HH )LJXUH f f ,PDJH 3URFHVVLQJ $Q 05, VFDQ KDV EHHQ GRQH WKH GD\ EHIRUH VXUJHU\ 8VLQJ D PRXVHOLNH GHYLFH WKH &7 DQG 05, LPDJHV DUH WUDQVIHUUHG WR WKH FRPSXWHU VFUHHQ DQG IXVHG 7KHQ
PAGE 18
XVLQJ JHRPHWULF HTXDWLRQV WKH FRPSXWHU GHWHUPLQHV WKH ODWHUDO $3 DQG YHUWLFDO VWHUHRWDFWLF FRRUGLQDWHV RI HDFK SRLQW LQ HDFK &7 VOLFH f 3ODQQLQJ 3K\VLFLDQV XVH D PRXVH DQG WKHLU RZQ MXGJPHQW WR fH\HEDOOf WKH ORFDWLRQ DQG WKH VL]H RI VSKHUHV IRU WKH WUHDWPHQW SODQ f 7UHDWPHQW GHOLYHU\ 7KH JHRPHWULF VSKHUH SDFNLQJ SODQ GHYHORSHG DV LQ f LV FRPELQHG ZLWK D SURSULHWDU\ GRVH SODQQLQJ SURJUDP DQG WKH UHVXOWLQJ WUHDWPHQW SODQ LV GHOLYHUHG E\ WKH /LQDF 0DNLQJ D 6SKHUH E\ $UFV RI %HDPV %\ YDU\LQJ WKH DQJOH RI WKH JDQWU\ DQG WKH DQJOH RI WKH WDEOH RQH FDQ GHOLYHU D UDGLDWLRQ EHDP WR WKH WDUJHW IURP DQ\ DQJOH ZLWKLQ WKH UDQJH RI WKH URWDWLRQ 7KH VKDSH RI WKH FRPPRQ LQWHUVHFWLRQ RI DQ DUF RI EHDPV SDVVLQJ WKURXJK RQH LVRFHQWHU LV D VSKHUH 7KH QHXURVXUJHRQV GHOLYHU D VHULHV RI DUFV XVXDOO\ RU DUFVf WR SURGXFH D VLQJOH LVRFHQWHUHG VSKHUH VKDSH )RU DQ HOOLSVRLGDO WDUJHW WKH\ XVH IHZHU QXPEHUV RI DUFV WR PDNH D VLQJOH LVRFHQWHUHG HOOLSVRLG VKDSH 6HH )LJXUH DQG f 6R LI WKH WDUJHW VKDSH LV YHU\ FORVH WR D VSKHUH RU DQ HOOLSVRLG WKHQ WKH WUHDWPHQW SODQ LV UHODWLYHO\ HDV\ FRPSDUHG WR DQ LUUHJXODUO\ VKDSHG WDUJHW ,Q WKDW FDVH ZH QHHG WR FUHDWH D JHRPHWULF WUHDWPHQW SODQ 6SKHUH 3DFNLQJ 3ODQ $V VHHQ LQ WKH SUHYLRXV VHFWLRQ WKH SK\VLFLDQV NQRZ KRZ WR LUUDGLDWHWR HYHQWXDOO\fGHVWUR\ WXPRUV ZKLFK DUH VKDSHG OLNH VSKHUHV RU HOOLSVRLGV )RU D QRQ VSKHULFDO VKDSH RI WXPRU WKH\ WU\ WR ILOO WKH WDUJHW ZLWK VHYHUDO VSKHUHV RI GLIIHUHQW VL]HV 7KLV LV FDOOHG WKH fVSKHUH SDFNLQJf WUHDWPHQW SODQ $IWHU ILQGLQJ D VSKHUH SDFNLQJ SODQ WKH\ WUHDW HDFK VSKHUH VHSDUDWHO\ DV GHn VFULEHG LQ WKH SUHYLRXV VHFWLRQ 6R PXOWLSOH LVRFHQWHU UDGLRVXUJHU\ SODQQLQJ LQFOXGHV
PAGE 19
)LJXUH 2QH LVRFHQWHU SODQ ZLWK ILYH DUFV RI UDGLDWLRQ :DJQHU HW DO>OO@f
PAGE 20
rf )LJXUH 2QH LVRFHQWHU SODQ ZLWK QLQH DUFV RI UDGLDWLRQ )ULHGPDQ HW DO>@f 82 2 P
PAGE 21
)LJXUH 3ODQ IRU D VSKHUH VKDSH DQG DQ HOOLSVRLGDO VKDSH )ULHGPDQ HW DO>@f
PAGE 22
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fJRRGf RQH 7KHUHIRUH ZH SURYLGH DQ DXWRPDWHG VSKHUH SDFNLQJ PHWKRG IRU WKH WUHDWPHQW SODQ VHH :DJQHU HW DO >@f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fEHQFKPDUNf WR FRPSDUH ZLWK RWKHU SODQV IRU WKH JLYHQ WXPRU VKDSH 0RUHRYHU WKLV SURJUDP SURYLGHV WKH WUHDWPHQW SODQ LQ D UHODWLYHO\ VKRUW WLPH )RU D YHU\ GLIILFXOW FDVH ZKLFK QHHGHG PRUH WKDQ VSKHUHV WKLV SURJUDP WRRN OHVV WKDQ PLQXWHV LQVWHDG RI PRUH WKDQ WKH KRXUV
PAGE 23
ZKLFK ZHUH QHHGHG ZKHQ WKH SK\VLFLDQV FUHDWHG WKH SODQ XVLQJ WUDGLWLRQDO PHWKRGV 7KH DOJRULWKP RI WKLV SURJUDP LV VKRZQ LQ )LJXUH ,Q WKH IROORZLQJ VHFWLRQ ZH H[SODLQ WKH fVKHOOLQJ DOJRULWKPf ZKLFK ZH XVHG LQ WKLV SURJUDP WR JHW WKH FHQWHUV DQG VL]HV RI VSKHUHV IRU WKH VSKHUH SDFNLQJ SODQ 6KHOOLQJ $OJRULWKP 7KH VKHOOLQJ SURFHGXUH LV EHVW LOOXVWUDWHG LQ )LJXUH WR 7KH PDMRU VWHSV RI WKLV VKHOOLQJ SURFHGXUH DUH DV IROORZV f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f /HW WKH WDUJHW EH PLQLPDOO\ HQFORVHG LQ DQ [UDQJH EHJLQQLQJ ZLWK RQH DQG HQGLQJ ZLWK 0[ D \UDQJH EHJLQQLQJ ZLWK RQH DQG HQGLQJ ZLWK 0\ DQG D ]UDQJH EHJLQQLQJ ZLWK RQH DQG HQGLQJ ZLWK 0] 1RZ ZH SXW fVDIH EXPSHUf OD\HUV VXUn URXQGLQJ WKH WDUJHW 7KDW LV ZH HQODUJH WKH ER[ FRQWDLQLQJ WKH WDUJHW WR (7 > 0[ @ [ > 0\ @ [ > 0] @ (7 VWDQGV IRU fHQODUJHG WDUJHW ER[f :H DVVXPH WKDW WKH GLPHQVLRQ RI HDFK YR[HO LV O[O[O 7KHQ ZH ZLOO QXPEHU WKH YR[HOV LQ D OH[Ln FRJUDSKLF PDQQHU EHJLQQLQJ ZLWK DQG HQGLQJ ZLWK 0[ f [ 0\ f [0r Of fÂ§ JLYLQJ XV D WRWDO RI 0[ f [ 0\ f [ 0] f YR[HOV ZLWK ZKLFK ZH ZLOO ZRUN 1RZ ZH ZDQW WR LGHQWLI\ WKH YR[HOV LQ (7 LQ WZR GLIIHUHQW ZD\V ILUVW E\ D VHW RI FRRUGLQDWHV D Ff DQG VHFRQG E\ DQ LQWHJHU UHSUHVHQWLQJ D FRXQWLQJ SURFHGXUH +HUH D LV WKH OHIW KDQG HQGSRLQW RI WKH [YDOXH RI WKDW YR[HO E LV WKH OHIW KDQG
PAGE 24
HQGSRLQW RI WKH \YDOXH RI WKDW YR[HO DQG F LV WKH OHIW KDQG HQGSRLQW RI WKH ] YDOXH RI WKDW YR[HO 7KH VHFRQG QRWDWLRQ IRU WKH YR[HO D Ff LV E\ VRPH LQWHJHU Q 7KH YR[HOV DUH FRXQWHG XVLQJ D OH[LFRJUDSKLF RUGHULQJ EHJLQQLQJ ZLWK DQG HQGLQJ ZLWK 0[ f [ 0\ I f [ 0 f fÂ§ ZKLFK LV GHQRWHG E\ QRIGDWD LQ WKH SURJUDP VHH $SSHQGL[ $f 7KHQ WKH YR[HO GHQRWHG E\ D Ff LV QXPEHUHG DV F [ 0[ f [ 0\ f D [ 0\ f E $QG LW LV FOHDU WKDW IRU HYHU\ L QRIGDWD WKHUH DUH XQLTXH QRQQHJDWLYH LQWHJHUV D F VXFK WKDW L fÂ§ F [ 0[ f [ 0\ f D [ 0\ f E 7KDW LV IRU HYHU\ L QRIGDWD RQH DQG RQO\ RQH YR[HO LV DVVLJQHG 1RWH WKDW )LJXUH WR )LJXUH VKRZ WKH VDPH WUDQVD[LDO YLHZ DW GLIIHUHQW VWDJHV RI WKH DOJRULWKP 7KXV WKHVH ILJXUHV VKRZ WKH GDWD RI D IL[HG D[LDOFRRUGLQDWH YDOXH f $VVLJQ D VWDWXV YDOXH IRU HDFK YR[HO )RU HYHU\ L QRIGDWD WKH VWDWXV YDOXH IRU WKH ]WK YR[HO LV GHQRWHG E\ FF>L? 7KDW LV ZH GHQRWH WKH VWDWXV YDOXH IRU WKH YR[HO ZLWK FRRUGLQDWHV D Ff E\ FF>L@ ZKHUH L F [ 0U f [ 0\ f D [ 0\ f E 7KH IROORZLQJ LV WKH SURFHGXUH WR DVVLJQ VWDWXV YDOXHV )LUVW RI DOO OHW FF>L? fÂ§ IRU HYHU\ L QRIGDWD 7KHQ ZH DVVLJQ WKH QXPEHU fÂ§ DV WKH VWDWXV YDOXH IRU HDFK YR[HO FRUUHVSRQGLQJ WR DQ\ RI WKH WDUJHW FRQWRXU GDWD 1H[W IRU HYHU\ M 0] ZH DVVLJQ FF>M [ 0[ f [ 0\ I f@ EHFDXVH WKH M [ 0[ f [ 0\ OfWK YR[HO LV WKH YR[HO DW WKH OHIW ORZHU FRUQHU IRU HYHU\ MWK WUDQVD[LDO YLHZ WKDW LV WKH YR[HO EHORQJV WR WKH KHDOWK\ WLVVXH 6HH )LJXUH f )RU HDFK YR[HO ZLWK FF>L? fÂ§ ZH GHILQH FF>L? LI DQ\ RI FF>L fÂ§ @ FF>L @ FF>L fÂ§ 0\ f@ FF>L 0\ I f@ LV 1RWH WKDW WKLV PHWKRG SURGXFHV ]HURV IRU HYHU\ YR[HO LQ WKH VDIH EXPSHU OD\HUV 6HH )LJXUH f f 6KHOOLQJ WKH WDUJHW YR[HOV :H SURFHHG E\ PDWKHPDWLFDO LQGXFWLRQ RQ M IRU [ PLQ^0[ 0\ 0]`
PAGE 25
)RU L WR 0[ f [ 0\ I f [ 0] f LI WKH LWK YR[HO LV DVVLJQHG RU fÂ§ IRU LWV VWDWXV YDOXH DQG LI DQ\ RI FF>r fÂ§ @ FF>L @ FF>L fÂ§ 0\ f@ FF>L 0\ f@ FF>L 0[ f [ 0\ f@ FF>L 0[ f [ 0\ f@ LV HTXDO WR M WKHQ ZH DVVLJQ D QHZ VWDWXV YDOXH IRU WKH Â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f f 5HPRYLQJ WKH ODUJHVW VSKHUH $W WKLV PRPHQW ZH PLJKW KDYH VHYHUDO GLIIHUHQW YR[HOV ZKLFK KDYH WKH VDPH ODUJHVW VWDWXV YDOXH 6R ZH FKRRVH WKH VPDOOHVW QXPEHUHG YR[HO DV WKH FHQWHU RI WKH ILUVW ODUJHVW VSKHUH WR EH UHPRYHG IURP WKH WDUJHW %XW LQ WKH SURJUDP LQ :DJQHU HW DO>OO@ ZH FKRRVH WKH PRVW HIIHFWLYH YR[HO WR EH WKH FHQWHU RI WKH ILUVW ODUJHVW VSKHUH E\ XVLQJ WKH fVFRUH IXQFWLRQf ZKLFK ZDV GHYHORSHG E\ 7RP :DJQHU :H WKLQN RI WKH UHVXOWLQJ VSKHUH LQ WKH WXPRU DV fUHPRYHGf DQG WKHUHIRUH ZH FDQ DVVLJQ WKH QHZ VWDWXV YDOXH ]HUR IRU WKHVH YR[HOV LQ WKH VSKHUH 7R GHWHUPLQH ZKLFK YR[HOV IRUP WKH VSKHUH WKDW LV UHPRYHG ZH GR WKH IROORZLQJ /HW 5 YR[HO DW WKH FHQWHU RI VSKHUH WR EH UHPRYHG /HW U VWDWXV YDOXH RI 5 7KHQ ZH UHPRYH DOO YR[HOV 6 ZLWK WKH SURSHUW\ WKDW GLVWDQFHFHQWHU RI 5 FHQWHU RI 6f U fÂ§ 6HH )LJXUHV DQG f 1RZ ZH SURFHHG E\ PDWKHPDWLFDO LQGXFWLRQ RQ M IRU M [ PLQ^0[ 0\ 0]` )RU L WR 0[ f [ 0\ f [ 0] f fÂ§ LI WKH rWK YR[HO KDV VWDWXV YDOXH M ZH UHDVVLJQ D QHZ VWDWXV YDOXH DV IROORZV LI DQ\ RI FF>L fÂ§ @ FF>L @ FF>L fÂ§ 0\ f@ FF>L 0\f@ FF>L fÂ§ 0[ Of[0\ Of@ FF>L 0[ Of[0\ Of@ LV HTXDO WR M WKHQ ZH DVVLJQ D QHZ VWDWXV YDOXH IRU WKH LWK YR[HO DV F>L? M 7KLV
PAGE 26
SURFHVV LV UHSHDWHG XQWLO WKH GHHSHVW O\LQJ YR[HOV KDYH EHHQ LGHQWLILHG 6HH )LJXUHV WR f $QG ZH UHSHDW VWHS f XQWLO ZH JHW HLWKHU RU IRU WKH VWDWXV YDOXH RI HDFK YR[HO LQ (7 WKH HQODUJHG WDUJHW ER[ %HFDXVH WKH VPDOOHVW VL]H RI WKH GLDPHWHU RI FROOLPDWRUV LV PP ZH ILQLVK RXU SURJUDP ZKHQ WKH VWDWXV YDOXHV RI DOO WKH YR[HOV DUH ]HURV DQGRU RQHV 6HH )LJXUH f 7KH UHPDLQLQJ SDUW ZLOO EH WDNHQ FDUH RI E\ WKH fVFRUH IXQFWLRQf VHH :DJQHU HW DO> @f $IWHU ILQLVKLQJ WKLV SURFHGXUH ZH KDYH WKH VSKHUHV ILOOLQJ WKH WDUJHW ,Q )LJXUH ZH VHH WKH VSKHUH SDFNLQJ SODQ ZLWKLQ D WDUJHW E\ WKLV PHWKRG )RU IXUWKHU GHWDLOHG LQIRUPDWLRQ DERXW WKH VFRUH IXQFWLRQ DQG WKH DXWRPDWHG VSKHUH SDFNLQJ SODQ ZH UHIHU WKH UHDGHU WR :DJQHU HW DO>@ DQG :DJQHU>@ )LJXUH 6SKHUH SDFNLQJ DOJRULWKP EORFN GLDJUDP :DJQHU HW DO>@f
PAGE 27
f f f f f f m f f }f } m f f m f f f f f f f f f f $ f f f f f f f f f f f f f f f f f f f f f } f f f f f f f f f 2 [ )LJXUH 7UDQVIHUULQJ WKH GDWD RI WKH ERXQGDU\ RI WKH WDUJHW )LJXUH 6WDWXV YDOXHV IRU WKH ERXQGDU\ DQG WKH LQLWLDO YR[HO
PAGE 28
)LJXUH 6WDWXV YDOXHV IRU WKH QRUPDO YR[HOV RO L } L! 6 6 U! _R ,7 )LJXUH 6KHOOLQJ WKH WDUJHW YR[HOV
PAGE 29
)LJXUH &KRRVLQJ WKH ODUJHVW VSKHUH )LJXUH 5HPRYLQJ WKH ODUJHVW VSKHUH
PAGE 30
)LJXUH 6KHOOLQJ WKH UHPDLQLQJ WDUJHW YR[HOV )LJXUH &KRRVLQJ WKH ODUJHVW VSKHUH RI WKH UHPDLQLQJ WDUJHW
PAGE 31
)LJXUH 5HPRYLQJ WKH VHFRQG VSKHUH )LJXUH 5HPRYLQJ DOO WKH VSKHUHV
PAGE 32
)LJXUH $ VSKHUH SDFNLQJ SODQ IRU D EUDLQ WXPRU :DJQHU HW DO>OO@f
PAGE 33
&+$37(5 $ &/$66,),&$7,21 2) 81/$%(/(' 75((6 ,QWURGXFWLRQ ,Q WKLV FKDSWHU ZH REWDLQ D FRGH WR GHQRWH DQ XQODEHOHG WUHH %\ PHDQV RI WKLV FRGH ZH FODVVLI\ XQODEHOHG QWUHHV 2QH PD\ WDNH DQ XQODEHOHG QWUHH WR EH WKH JHRPHWULF UHDOL]DWLRQ RI D FRQQHFWHG ILQLWH VLPSOLFLDO FRPSOH[ ZLWKRXW DQ\ VLPSOH FORVHG FXUYHV ,Q SDUWLFXODU ZH FDOO D WUHH DQ QWUHH LI DQG RQO\ LI LW LV D WUHH ZLWK Q YHUWLFHV WKDW LV Q fÂ§ f HGJHVf 2XU FRGH DVVLJQV D XQLTXH RUGHUHG fSHUIHFW VHTXHQFHf SI7f FIL G GQf WR HDFK XQODEHOHG QWUHH 7 $QG FRQYHUVHO\ JLYHQ DQ RUGHUHG VHTXHQFH RI Q LQWHJHUV VDWLVI\LQJ FHUWDLQ SURSHUWLHV LW LV WKH SHUIHFW VHTXHQFH RI H[DFWO\ RQH XQODEHOHG QWUHH 7KXV ZH KDYH RXU PDLQ WKHRUHP WKH 7UHH &ODVVLILFDWLRQ 7KHRUHP 7KHRUHP f :H WKHQ VKRZ KRZ WR UHFRQVWUXFW WKH WUHH VKDSH IURP D JLYHQ SHUIHFW VHTXHQFH E\ KDQG 2XU ZRUN LQFOXGHV DQ DOJRULWKP DQG D 0DWKHPDWLFD SURJUDP VHH $SSHQGL[ &Of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
PAGE 34
7KH /LWHUDWXUH FRQWDLQV ZRUN UHJDUGLQJ WKH FODVVLILFDWLRQ RI ODEHOHG JUDSKV DQG ODEHOHG WUHHV ,Q /RYÂ£V]>@ WKH 3ULLIHUFRGH LV XVHG WR GHWHUPLQH WKH QXPEHU RI ODEHOHG QWUHHV VHH WKH $SSHQGL[ DQG 'f ,Q 6ORDQH>@ 6ORDQH VKRZV D WDEOH IRU Q +RZHYHU 6ORDQH UHIHUHQFHV +DUDU\ IRU WKH PHWKRG KH XVHV ,Q +DUDU\>@ WKHUH LV D VLPSOH IRUPXOD IRU WKH WRWDO QXPEHU RI ODEHOHG JUDSKV ZLWK Q YHUWLFHV VHH WKH $SSHQGL[ 'Of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f (*ff FRQVLVWV RI D YHUWH[ VHW 9*f ^W!L L!f` DQG DQ HGJH VHW (*f ^HL HP` & ^^X X`_XY 9*f` LQ ZKLFK HDFK HGJH LV DQ XQRUGHUHG SDLU RI WZR GLVWLQFW YHUWLFHV X DQG Y DQG WKH HGJH ^XX` LV VDLG WR MRLQ X DQG Y )RU DQ HGJH ^XX` ZH VD\ WKDW X DQG Y DUH DGMDFHQW YHUWLFHV DQG WKH YHUWH[ X DQG WKH HGJH ^XY` DUH LQFLGHQW ZLWK HDFK RWKHU DV DUH
PAGE 35
Y DQG ^XY` )URP QRZ RQ _9Mf_ GHQRWHV WKH QXPEHU RI YHUWLFHV DQG _(*f? GHQRWHV WKH QXPEHU RI HGJHV )RU D YHUWH[ X f 9*f WKH YHUWH[ VHW $^Xf ^Y *?Y LV DQ DGMDFHQW YHUWH[ RI X` LV FDOOHG WKH DGMDFHQW VHW RI X 1RWH WKDW X $^Xf 'HILQLWLRQ $ ZDON LV DQ DOWHUQDWLQJ VHTXHQFH L! HL 9? H Hr Xrf RI YHUWLFHV DQG HGJHV VW XBL 9^ IRU L fÂ§ DQG H ^XBLQ` IRU L N ,I WKHUH LV QR FRQIXVLRQ D ZDON FDQ EH GHQRWHG E\ D VHTXHQFH RI YHUWLFHV RQO\ LQ WKH SDWK OLNH LT Y Xrf $ F\FOH LV D ZDON Q HL Y? H Hr 9Nf ZLWK N LQ ZKLFK 9R 9N LV WKH RQO\ YHUWH[ UHSHDWHG $ SDWK LV D ZDON ZLWK QR UHSHDWHG YHUWH[ $ XYfSDWK LV D SDWK ZLWK WKH ILUVW YHUWH[ X DQG WKH ODVW YHUWH[ Y 7KH OHQJWK RI WKH XYfSDWK OXYf LQ D JUDSK LV WKH QXPEHU RI HGJHV LQ WKH XL!fSDWK 7KH GLVWDQFH GXYf EHWZHHQ WZR YHUWLFHV X DQG Y LV WKH OHQJWK RI D VKRUWHVW SDWK MRLQLQJ WKHP LI DQ\ RWKHUZLVH GXYf RR 1RWH WKDW GXXf $ JUDSK LV FRQQHFWHG LII HYHU\ SDLU RI YHUWLFHV DUH MRLQHG E\ D SDWK $ JUDSK KDYLQJ QR F\FOH LV DF\FOLF 1RWH WKDW DQ DF\FOLF JUDSK FDQ EH GLVFRQQHFWHG $ IRUHVW LV DQ DF\FOLF JUDSK $ WUHH LV D FRQQHFWHG DF\FOLF JUDSK $Q QWUHH LV D WUHH ZLWK Q YHUWLFHV 'HILQLWLRQ :H VD\ WKDW D JUDSK LV LVRPRUSKLF WR D JUDSK + LI DQG RQO\ LI WKHUH H[LVWV D ELMHFWLRQ 9*f fÂ§! 9+f VXFK WKDW ^XX` f (*f LI DQG RQO\ LI ^rf}f` 6(+f 'HILQLWLRQ $Q XQODEHOHG WUHH LV DQ LVRPRUSKLVP FODVV RI WUHHV :H DUH LQWHUHVWHG LQ WKH WRWDO QXPEHU RI XQODEHOHG QWUHHV DQG WKHLU VKDSHV )RU H[DPSOH ? ^` ^^` ^``f DQG 7 ^` ^^` ^``f DUH GLIIHUHQW WUHHV EXW WKHUH LV DQ LVRPRUSKLVP 9n7Mf 97f E\ f f f 7KXV WKH\ DUH WKH VDPH XQODEHOHG WUHH VR ZH ZDQW WR FRXQW WKHP DV RQH XQODEHOHG WUHH )URP QRZ RQ D WUHH LPSOLHV DQ XQODEHOHG WUHH XQOHVV RWKHUZLVH VSHFLILHG
PAGE 36
'HILQLWLRQ 7KH GHJUHH RI D YHUWH[ Y GHILQHG E\ GYf LV WKH VXP RI WKH QXPEHU RI HGJHV LQFLGHQW ZLWK Y $ YHUWH[ Y LV FDOOHG DQ HQGYHUWH[ LI G^Yf fÂ§ :H ZLOO XVH WKH JHRPHWULF WKDW LV WRSRORJLFDO UHDOL]DWLRQ RI D JUDSK LQWHUn FKDQJHDEO\ ZLWK fJUDSKf DV GHILQHG DERYH 3URSRVLWLRQ >@ 7KH IROORZLQJ VWDWHPHQWV DUH HTXLYDOHQW IRU D JUDSK 7 f 7 LV D WUHH f (YHU\ WZR YHUWLFHV RI 7 DUH MRLQHG E\ D XQLTXH SDWK f 7 LV FRQQHFWHG DQG _97f_ ?(7f? f 7 LV DF\FOLF DQG _97f_ _Âe7f_ 1RWH WKDW WKHUH DUH H[DFWO\ Q YHUWLFHV DQG Q HGJHV LQ DQ QWUHH 3URSRVLWLRQ 'HJUHH VXP IRUPXODf >@ ,I LV D JUDSK ZLWK WKH YHUWH[ GHJUHHV GX WKHQ (f*9*f ?(*f? &RUROODU\ )RU DQ Q7UHH 7 ZH KDYH (7 L rmf Q f ?(7f? _37f_ f ZKHUH LV D YHUWH[ RI 7 IRU L Q %HFDXVH D QRQWULYLDO WUHH VDWLVILHV @ (YHU\ QRQWULYLDO WUHH KDV DW OHDVW WZR HQGYHUWLFHV 9HUWH[ 6HTXHQFH ,Q WKLV VHFWLRQ ZH GHILQH WKH YHUWH[ VHTXHQFH RI D JLYHQ QWUHH 'HILQLWLRQ /HW 7 9 (f EH DQ QWUHH ,I G\f P WKHQ WKHUH DUH P FRPSRQHQWV RI 7 fÂ§ ^X` $ EUDQFK RI Y RU VLPSO\ D YEUDQFKf LV D VXEWUHH ZKLFK LV WKH FORVXUH RI D FRPSRQHQW RI 7 fÂ§ ^X` :H FDOO WKH VHW RI YHUWLFHV LQ D XEUDQFK D YEUDQFK YHUWH[ VHW ,I X $Xf WKHQ WKH ZEUDQFK YHUWH[ VHW FRQWDLQLQJ X LV FDOOHG WKH YXfEUDQFK YHUWH[ VHW DQG GHQRWHG E\ %YXf RU %Xff :H FDOO WKH
PAGE 37
VHW %nYXf %YXf fÂ§ ^R` WKH W!LLfVXEEUDQFK YHUWH[ VHW ZKLFK LV GHQRWHG E\ %?Xf LI WKHUH LV QR FRQIXVLRQ 7KH VXEWUHH FRQVLVWLQJ RI %nYXf DV WKH YHUWH[ VHW DQG ^^VW` H %?YXf [ %nYXf_ ^VW` f (7f` DV WKH HGJH VHW LV FDOOHG WKH W!Xf VXEEUDQFK ([DPSOH /HW 7 EH WKH WUHH VKRZQ EHORZ V W X S U 9 Z ; \ )LJXUH 6LQFH GYf WKHUH DUH FRPSRQHQWV RI 7 fÂ§ ^Y` 6LQFH $Yf ^U W WR \` ZH KDYH WKDW %^YUf ^YUVTS` %YWf ^0` %YZf fÂ§ ^R UR X [` %Y\f ^Y\` %nYUf ^UVTS` %nYWf ^W` %nY Zf ^UR X D` %nY\f ^\` 2Q WKH RWKHU KDQG ZH KDYH WKDW %UYf ^UYWXZ[\` 7KDW LV %Y Uf %U Yf $QG WKLV LV WUXH IRU DQ\ WUHH H[FHSW WKH WUHH ,Q IDFW IRU HYHU\ QWUHH ZLWK Q WKHUH DUH DW OHDVW YHUWLFHV QDPHO\ XY DQG Z :LWKRXW ORVV RI JHQHUDOLW\ OHW X DQG Y EH YHUWLFHV DGMDFHQW WR HDFK RWKHU DQG OHW WKH WR Rf SDWK FRQWDLQ WKH YHUWH[ X 7KHQ Z e %YXf DQG Z A %XYf 'HILQLWLRQ $ VHFWLRQ RI D VHTXHQFH LV D VXEVHTXHQFH FRQVLVWLQJ RI FRQVHFXWLYH WHUPV RI WKH VHTXHQFH 'HILQLWLRQ /HW 7 9 (f EH DQ QWUHH ZLWK 9 ^rL ; [Q` :H GHILQH WKH VHW 6 RI VHTXHQFHV RI WKH IRUP V R 9 YQf ZKHUH V LV D SHUPXWDWLRQ RI WKH [fV LQ 97f WR EH WKRVH VHTXHQFHV VDWLVI\LQJ
PAGE 38
rf IRU DQ\ 9M ZLWK M Q 9M? f $YNf fÂ§ ^QL Y f f f 8M` ZKHUH N PD[^L ^ M`_ $^Y^f fÂ§ ^YLY 9M` ` 7KHQ 6 LV FDOOHG WKH YHUWH[ VHTXHQFH VHW RI 7 DQG HDFK HOHPHQW V LQ 6 LV FDOOHG D YHUWH[ VHTXHQFH RI 7 'HILQLWLRQ )RU DQ\ Y f 9 ZH FDOO 6Y ^ f 6MIL Y` & 6 WKH YYHUWH[ VHTXHQFH VHW 'HILQLWLRQ UHTXLUHV WKDW WKH YHUWLFHV LQ D VHTXHQFH EH OLVWHG LQ D SDUWLFXODU ZD\ 7KH FRQGLWLRQ rf HQVXUHV WKDW DOO YHUWLFHV DUH OLVWHG DQG WKDW HDFK LV OLVWHG H[DFWO\ RQFH :H FDQ FKRRVH DQ\ YHUWH[ IRU WKH ILUVW WHUP LT 7KHQ WKH FRQGLWLRQ rf VD\V WKDW IRU WKH LOfWK WHUP ZH KDYH WR FKRRVH DQ XQOLVWHG DGMDFHQW YHUWH[ RI 9^ LI SRVVLEOH ,I HYHU\ DGMDFHQW YHUWH[ RI LV DOUHDG\ XVHG WKHQ ORRN EDFNZDUG DORQJ WKH VHTXHQFH XQWLO ZH FRPH WR WKH ILUVW YHUWH[ ZKLFK KDV DQ XQXVHG DGMDFHQW YHUWH[ 7KHQ IRU WKH LOfWK WHUP ZH DVVLJQ RQH RI WKH XQXVHG DGMDFHQW YHUWLFHV RI WKDW YHUWH[ 7R JHW D YHUWH[ VHTXHQFH ZH FRQWLQXH WKLV SURFHGXUH LQGXFWLYHO\ 1RWH WKDW HYHU\ YHUWH[ VHTXHQFH RI DQ QWUHH KDV Q HQWULHV ([DPSOH :H VKRZ KRZ WR FRQVWUXFW D YHUWH[ VHTXHQFH RI D JLYHQ WUHH /HW 7 EH WKH WUHH VKRZQ EHORZ I D E F G H )LJXUH 6LQFH ZH KDYH IUHHGRP WR FKRRVH WKH ILUVW WHUP OHW 9? E 7KHQ Y f ALTf fÂ§ ^Y[` fÂ§ $Ef fÂ§^E` ^DF I` fÂ§ ^E` ^DF I` $Ef fÂ§ $YLf 6R Y PXVW EH RQH RI WKH HOHPHQWV RI $XLf $Ef ^DFI` /HW Y F 6LQFH Y f $Yf fÂ§ ^XLX` $^Ff fÂ§ ^E F` ^E G J` fÂ§ ^E F` ^GJ` Y PXVW EH HLWKHU G RU J /HW Y J 7KHQ Y $Yf ^YLYY` VLQFH $Yf ^YXYY` ^F` ^EFJ` 2Q
PAGE 39
WKH RWKHU KDQG ZH KDYH WKDW $Xf fÂ§ ^W!LQ X` ^e! GJ` fÂ§ ^EFJ` fÂ§ ^G` 7KHUHIRUH ZH KDYH WR FKRRVH G DV Y 7KHQ YA fÂ§ H VLQFH $Yf fÂ§ ^LTX` ^F H` fÂ§ ^EFJG` ^H` 6LQFH ZH KDYH $Xf ^XL8YY8` ^G` ^EFJGH` $Xf ^YXYYY Q` ^F H` ^t FS G H` $Qf ^XLXQ8X` ^F` ^t F G H` $Xf ^YL:Y9` ^0! f ^EFJGH` DQG $LTf ^QLQQQQ` ^D F ` ^t F G H@ ^D` Q PXVW EH HLWKHU D RU /HW Q 7KHQ ZH KDYH WKDW $Xf ^WT8X8XL!` ^t` a ^EFJGHI` $Zf ^W!OW!W!W!W!W!` ^G` ^tFGH` $Xf fÂ§ ^XL X X X X X` ^F H` fÂ§ ^ F G H ` $Xf ^UTXYX8X` ^F` ^FIOIGH` $Yf^Y Y W Y X X` ^IOIG` ^tFGH` DQG $LTf ^XLXXXXL` ^D F ` ^t FJ G H ` ^D` 7KHUHIRUH WT PXVW EH D WKDW LV ZH FRQVWUXFW D YHUWH[ VHTXHQFH YXYYf EFJGHIDf )RU D JLYHQ WUHH ZH ZLOO JHW GLIIHUHQW YHUWH[ VHTXHQFHV LI ZH FKRRVH GLIIHUHQW YHUWLFHV ZKHQHYHU ZH KDYH WZR RU PRUH FKRLFHV %XW WKH VHTXHQFH W F G H D f FDQQRW EH D YHUWH[ VHTXHQFH RI 7 DERYH ,Q IDFW ZH KDYH Y I %XW ZH KDYH WKDW $Xf ^YLW 8 9` ^G` ^Â! F G H` $Xf ^LTXXX` ^FH` ^KFGH` $Yf ^UT X W! ` ^ G` ^ F G H` ^` DQG $WTf ^LTXQX` ^D F ` fÂ§ ^ F G H` ^D` 7KDW LV WKH VHTXHQFH W KDV RQH RI WKH HOHPHQWV RI $UTf fÂ§ ^LT Q X ` IRU 8 LQVWHDG RI J ZKLFK LV DQ HOHPHQW RI $Xf fÂ§ ^LT X X X` ,W LV D FRQWUDGLFWLRQ WR WKH PD[LPDOLW\ RI N LQ WKH GHILQLWLRQ RI YHUWH[ VHTXHQFH 3URSRVLWLRQ /HW LT X YQf EH D YHUWH[ VHTXHQFH RI DQ QWUHH 7 )RU DQ\ 9M ZLWK M WKHUH LV XQLTXH N M VXFK WKDW 9M $UTf fÂ§ ^UTW! LTB ` 3URRI )LUVW RI DOO IRU DQ\ M ZH KDYH 9M Q IRU L M fÂ§ IURP WKH 'HILQLWLRQ $VVXPH WKDW WKHUH LV M VXFK WKDW ^YM9MX9Q` 8)A$Aff 8 ^9LWWMBL`f
PAGE 40
7KHQ IRU DQ\ X ^YMYMLYf` DQG DQ\ Z f 8ÂL $QÂff 8 ^U!L Y Â‘ 9M` WKHUH LV QR XLQfSDWK LQ 7 7KDW LV 7 LV QRW FRQQHFWHG ,W LV D FRQWUDGLFWLRQ WR WKH FRQQHFWHGQHVV RI D WUHH 7KHUHIRUH WKHUH LV O N M VXFK WKDW YM f $YNf fÂ§ ^XL Y Â‘ Â‘ Â‘ 9ML` ,I WKHUH DUH N? N M VXFK WKDW 9M f $L\Af fÂ§ ^YLY 9M` DQG 9M f $YNf fÂ§ ^XL Y f f f WKHQ WKH WZR HGJHV ^YMYNO` DQG ^YM9N`L DUH HOHPHQWV RI (7f 7KHUH LV D SDWK EHWZHHQ 9N[ DQG 9N LQ WKH VXEWUHH ^^W!L Y 8ML` (f` ZKHUH (n ^^X X` f (7f?X Y ^ZL Y XÂBL`` 7KH WUHH 7 FRQWDLQV D F\FOH FRQVLVWLQJ RI ^YM9N` DQG WKH YA 8IFfSDWK 7KLV FRQWUDGLFWV WKH IDFW WKDW D WUHH LV DF\FOLF 7KHUHIRUH WKHUH LV RQO\ RQH O N M VXFK WKDW YM f $YNf fÂ§ ^YLY B/ )URP WKH 'HILQLWLRQ DQG WKH SUHYLRXV OHPPD ZH JHW WKH IROORZLQJ FRUROn ODU\ LPPHGLDWHO\ &RUROODU\ /HW XL Y YQf EH D YHUWH[ VHTXHQFH RI DQ QWUHH 7KHQ f YH f ,I GYMf IRU DQ\ M Q fÂ§ WKHQ 9ML f $^YMf f )RU DQ\ M Q fÂ§ LI GYMf DQG WKHUH LV L M VXFK WKDW GYIf DQG GYNf IRU HYHU\ N L L M fÂ§ WKHQ Xf $Xf DQG 9ML e $YSf IRU HYHU\ S L r M 3URSRVLWLRQ /HW XL8 YQf EH D YHUWH[ VHTXHQFH RI DQ QWUHH DQG $XLf XOP` ZLWK OL OUR 7KHQ ^YLYYQ` 8 %YLYLWf 3URSRVLWLRQ /HW V W"[ Y YQf EH D YHUWH[ VHTXHQFH RI D JLYHQ QWUHH 7 ,I GYMf P IRU VRPH M WKHQ WKHUH H[LVW P YHUWLFHV X 9M 9M XPB VXFK WKDW $YMf ^XÂ YMO YK XMPB` ZLWK L M ML M MP f 3URRI )RU DQ\ M WKHUH H[LVWV XQLTXH L M VXFK WKDW ^X8M` f (7f E\ 3URSRVLWLRQ WKDW LV f $YMf 6R IRU DQ\ 9N f $^YMf fÂ§ ^X` M N
PAGE 41
6LQFH GYMf P WKHUH DUH P fÂ§ YHUWLFHV 9M[9M 8MPQ VXFK WKDW $YMf fÂ§ ^XÂ YK YK YMUQB[` ZLWK L M ML ML MPfÂ§O f / 3URSRVLWLRQ /HW V Y? X Â‘ Â‘ Â‘ YQf EH D YHUWH[ VHTXHQFH RI D JLYHQ QWUHH 7 /HW GYLf P DQG $^YLf ^X[Q YO YLP` EH WKH DGMDFHQW VHW RI YX QDPHG VR WKDW OL OP 7KHQ IRU DQ\ L P WKH VHW RI YHUWLFHV LQ WKH VXEEUDQFK %nYLXLMf IRUPV D VHFWLRQ LQ V VWDUWLQJ ZLWK Y?W 3URRI &OHDUO\ WKH VHFWLRQ FRQWDLQLQJ 8[ VWDUWV ZLWK 9L[ X /HW %n>\LQLMf ^XOQ 8LML ILPL `f $VVXPH WKDW WKHUH LV D YHUWH[ X f 97f fÂ§ %nYL 8LVXFK WKDW YL[ XLL YOOM Xf LV D VHFWLRQ RI V 6LQFH X e %nYL 9Lf $YOONfa %nY[YLIf IRU DQ\ N M IURP WKH GHILQLWLRQ 7KXV 8Â $XLDf fÂ§ %nYXYLf 7KDW LV 8A $XLIFf %nYL XMf 2Q WKH RWKHU KDQG $YLLNf & %nY LXLMf IRUHYHU\ N M 7KXV ZH KDYH WKDW 8A $XOOIFf & %nYL YO[f 7KHUHIRUH 8AB$XLIFf %nY7KDW LV M PL +HQFH %n>\LQLf IRUPV D VHFWLRQ LQ V 6LQFH %nYL8LMf LV D VXEEUDQFK RI Y? 8[72O[ $Xf IRU DQ\ X %nYLY;Of 7KXV XOOPL $Y[f 7KDW LV X[PL[ YO )RU DQ\ L fÂ§ OHW %nY LXLÂf IRUP D VHFWLRQ LQ V DQG 8[LPL[ L![L %\ D VLPLn ODU PHWKRG %n\L8LOf IRUPV D VHFWLRQ LQ V DQG 8LLURL AÂ 7 7KH VHFWLRQ IRUPHG E\ D VXEEUDQFK %n\LX[M LV GHQRWHG E\ %?Y[X[Âff )URP WKH SUHYLRXV SURSRVLWLRQ ZH KDYH WKH IROORZLQJ FRUROODULHV &RUROODU\ /HW V X[ X YQf EH D YHUWH[ VHTXHQFH RI D JLYHQ QWUHH 7 /HW GY[f P DQG $^YLf ^XLM L![ 9LP` EH WKH DGMDFHQW VHW RI 8[ QDPHG VR WKDW O[ OP 7KH VHFWLRQ %nYLWfLLff SUHFHGHV WKH VHFWLRQ %?Y [Y[Mff LQ V IRU L fÂ§ P fÂ§ DQG WKHUH LV QR RWKHU YHUWH[ EHWZHHQ WKHVH WZR VHFWLRQV &RUROODU\ /HW X[ X YQf EH D YHUWH[ VHTXHQFH RI D JLYHQ QWUHH 7 )RU DQ\ M ZLWK GYMf P OHW $^YIf fÂ§ ^X 9M[9M 9MPB[` ZKHUH L
PAGE 42
M ML fff MPfÂ§L f 7KHQ ^%nLYMW9MLff YMLW YML XWfLf IRU DQ\ L P fÂ§ ,Q SDUWLFXODU ZH KDYH WKDW M? M 1RWH WKDW ^YM XPBQ XMPBL XPBL` LV WKH ZMXLPBfEUDQFK YHUWH[ VHW IRU VRPH MP MP &RUROODU\ /HW WT 9 YQf EH D YHUWH[ VHTXHQFH RI D JLYHQ QWUHH ,I L M ZLWK YM f ML!f DQG 9M $YNf IRU HYHU\ N L L M fÂ§ WKHQ YS I $^\Nf IRU HYHU\ S M DQG N ]f r M fÂ§ )RU DQ\ JLYHQ YHUWH[ VHTXHQFH V ZLf f f YQf DQG DQ\ YHUWH[ 9M A XO Y? EHORQJV WR D EUDQFK RI 9M :H ZDQW WR JLYH D VSHFLDO QDPH WR WKLV EUDQFK 'HILQLWLRQ /HW V LT Y! f f f XQf EH D YHUWH[ VHTXHQFH RI D JLYHQ QWUHH 7 )RU DQ\ M WKH URRW EUDQFK RI 9M LQ V LV WKH EUDQFK RI 9M FRQWDLQLQJ 9? 1RWH WKDW WKHUH LV QR URRW EUDQFK IRU LT 3URSRVLWLRQ /HW LT X YQf 6 EH WKH YHUWH[ VHTXHQFH VHW RI D JLYHQ QWUHH 7 )RU DQ\ M Q LI LTL A WKHQ YM LV DQ HQG YHUWH[ 3URRI ,I 9M A $>YMf IRU VRPH M Q WKHQ YM KDV D XQLTXH HGJH ZLWK D YHUWH[ XÂ ZLWK L M E\ 3URSRVLWLRQ 7KDW LV GYMf 7KHUHIRUH 9M LV DQ HQG YHUWH[ ; 3URSRVLWLRQ /HW WT X YQf f 6 EH D YHUWH[ VHTXHQFH VHW RI D JLYHQ QWUHH 7 )RU DQ\ M Q WKHUH LV DQ LQWHJHU MP VXFK WKDW ^YLY Â‘ Â‘ Â‘ 9M` 8 ^YMP 9MP8 9Q` LV WKH YHUWH[ VHW RI WKH URRW EUDQFK RI 9M 3URRI /HW M Q DQG GYMf P 7KHQ E\ 3URSRVLWLRQ WKHUH H[LVW P GLIIHUHQW YHUWLFHV WT 9M WT 9MPBO VXFK WKDW $^YIf ^XÂ 9M WT 9MPBO` ZLWK L M ML MY MPfÂ§Lf )URP &RUROODU\ ZH KDYH WKDW ErIF L % YM9MNfA fÂ§ L!MM IMML
PAGE 43
7KDW LV WKH XQLRQ RI DOO WKH VXEEUDQFKHV RI 9M H[FHSW WKH URRW EUDQFK LV D VHFWLRQ IURP 9M^ WR L!MPBL 7KHUHIRUH WKH VHW RI YHUWLFHV LQ WKH URRW EUDQFK LV ^LT X L!M`8 ^9MPO9MP OO f f f 9Q`f / 1RWH WKDW LI MP Q WKHQ WKH URRW EUDQFK LV ^LT X 8M` &RUROODU\ /HW V LT X f f Â‘ YQf f 6 EH WKH YHUWH[ VHTXHQFH VHW RI D JLYHQ QWUHH 7 /HW M 7KHQ GYMf WKDW LV 9M LV DQ HQG YHUWH[ RI 7 LI DQG RQO\ LI WKH VHW RI YHUWLFHV RI WKH URRW EUDQFK RI 9M LV 97f ^LT X Xf` 3URSRVLWLRQ /HW V ^Y? X YQf EH D YHUWH[ VHTXHQFH RI D JLYHQ Q WUHH 7 )RU DQ\ YHUWH[ 9M ZLWK M WKH Y?YMfSDWK LQ 7 LV D VXEVHTXHQFH RI 9L 99Mf 3URRI )RU DQ\ YHUWH[ 9M ZLWK M WKHUH LV D XQLTXH YHUWH[ 9NS ZLWK NY M VXFK WKDW 9M f $YNSf )RU WKH YHUWH[ 9NS WKHUH LV DQRWKHU XQLTXH YHUWH[ 9NSB ZLWK $SBL NS VXFK WKDW 9NS f $YNSBIf 7KHQ ZH ZLOO KDYH WKDW LT LV WKH XQLTXH DGMDFHQW YHUWH[ RI D SUHYLRXV YHUWH[ Â‘ 7KHUHIRUH WKH UT YÂfSDWK LV D VXEVHTXHQFH ^YLYNOYNSBOYNSYMf RI 8LAAf 3URSRVLWLRQ )RU DQ\ WZR YHUWH[ VHTXHQFHV ÂM DQG V LQ WKH YYHUWH[ VHn TXHQFH VHW 6Y ^L f MWT Y` WKH URRW EUDQFK YHUWH[ VHWV RI D YHUWH[ [ A UT fÂ§ Y LQ 6L DQG LQ V DUH VDPH 3URRI /HW VM WT X YQf DQG Â X? X XQf EH WZR GLIIHUHQW YHUWH[ VHTXHQFHV RI D JLYHQ WUHH 7 ZLWK WT XL /HW [ EH DQ DUELWUDU\ YHUWH[ LQ 7 ZLWK [ WT 7KHQ WKHUH DUH WZR LQWHJHUV N DQG VXFK WKDW 9N [ WT /HW $[f ^DT [ ePBL \` 6LQFH 7 LV D WUHH WKHUH LV D XQLTXH WT[fSDWK LQ 7 7KHQ RQH RI WKH HOHPHQWV RI $^[f LV LQ WKH LT [fSDWK :LWKRXW ORVV RI JHQHUDOLW\ OHW \ EH LQ WKH LT UfSDWK )URP WKH SURRI RI 3URSRVLWLRQ ZH KDYH WKDW 8 9 %n[ rff 8f9 %nYN [Mf AN?" f f f 4,OG 8 A%n[[Sff 8ff%nm;3f! fÂ§ 8O 8O Â‘ f f 8OUf
PAGE 44
IRU VRPH U 7KHUHIRUH ZH KDYH WKDW ^LT A" f f f M 9N` 8 ^8IFMUO } f f f M W\L` fÂ§ K^ULMU_O Â‘ f M AQ`f 7KDW LV WKH URRW EUDQFK YHUWH[ VHWV RI [ LQ VM DQG LQ V DUH VDPH B ([DPSOH /HW 7 EH DQ WUHH DV VKRZQ EHORZ Y Z S T U V W X )LJXUH &OHDUO\ VM TUVWXZYSf V TSUVZWXYf H 6U ^V f 6MWT T` /HW YLY Yf TUVWXZYSf DQG LT X Xf ^TSUVZWXYf 7KHQ WKH URRW EUDQFK YHUWH[ VHW RI V LV ^S T U L!` 7KDW LV ^STUY` ^T U` 8 ^YS` ^XLW` 8 ^YX` DQG ^STUY` ^" S U` 8 ^X` ^XLWLX` 8 ^X` 7KHUHIRUH WKH URRW EUDQFK YHUWH[ VHWV DUH VDPH EXW WKH SRVLWLRQ PLJKW EH GLIIHUHQW 2Q WKH RWKHU KDQG GZf DQG WKH XQLTXH DGMDFHQW YHUWH[ RI Z fÂ§ Yt LV V Z 7KXV V X LV LQ WKH URRW EUDQFK RI Z YT 7KHUHIRUH WKH URRW EUDQFK YHUWH[ VHW RI Z Yi LV ^LT Y X 8 X X` 8 ^X X` ^WT Y X` ^TUVWXZYS` 7KDW LV WKH URRW EUDQFK YHUWH[ VHW RI DQ HQG YHUWH[ LV 9f7f LWVHOI 'HJUHH 6HTXHQFH ,Q WKLV VHFWLRQ ZH GHILQH WKH GHJUHH VHTXHQFH RI D YHUWH[ VHTXHQFH 'HILQLWLRQ /HW 6 EH WKH YHUWH[ VHTXHQFH VHW RI DQ QWUHH 7 )RU DQ\ YHUn WH[ VHTXHQFH V WT Y YQf f 6 ZH FDOO 'D RIL G GQf WKH GHJUHH VHn TXHQFH RI V ZKHUH GL fÂ§ GWTf DQG GY^f fÂ§ IRU L :H FDOO WKH VHW ^'V_V f ` WKH GHJUHH VHTXHQFH VHW RI 6 $QG IRU DQ\ YHUWH[ Y e 9 ZH
PAGE 45
FDOO 'Y fÂ§ ^'c?V f 6n` & WKH YGHJUHH VHTXHQFH VHW RI 6Y )RU DQ\ VHFWLRQ %IDYMWff YMYMLYMLXYMLOLf RI V IRUPHG E\ WKH YM L!rfEUDQFK YHUWH[ VHW %YMYWf ZH GHILQH %GYMYMWf ^GM GÂ GMLL GÂLL` DV WKH YM9M^f EUDQFK GHJUHH VHW ZKHUH GA fÂ§ GYNf fÂ§ IRU N M MLML O Â‘ Â‘ Â‘ MLL fÂ§ $QG ZH GHILQH WKH YÂ 9MAEUDQFK GHJUHH VHFWLRQ E\ %GYM 8Mff ^GMGMLGML GLBLf ZKLFK LV GHQRWHG E\ %GYM^ff LI WKHUH LV QR FRQIXVLRQ ,Q WKH VDPH ZD\ DV LQ 'HILQLWLRQ ZH GHILQH ^%GnYMWff ^%G?YMYXff L! f f f}f ([DPSOH 2FFDVLRQDOO\ LW PD\ KDSSHQ WKDW WKHUH DUH WZR RU PRUH YHUWH[ VHn TXHQFHV RI D JLYHQ WUHH ZLWK WKH VDPH GHJUHH VHTXHQFH ,Q WKH IROORZLQJ H[DPSOH WKHUH DUH VHYHUDO SDLUV RI YHUWH[ VHTXHQFHV ZLWK WKH VDPH GHJUHH VHTXHQFHV UHVSHFWLYHO\ 4 S U V L )LJXUH 6LQFH WKHUH DUH YHUWLFHV LQ 97f WKHUH DUH H[DFWO\ GLIIHUHQW VHTQHQFHV RI YHUWLFHV RI 7 KDYLQJ OHQJWK DQG XVLQJ HDFK YHUWH[ RQO\ RQFH %XW RQO\ RI WKHP VDWLVI\ WKH FRQGLWLRQ rf LQ 'HILQLWLRQ 7KHUHIRUH WKH YHUWH[ VHTXHQFH VHW 6 RI WKH JLYHQ WUHH 7 FRQVLVWV RI WKH IROORZLQJ VHTXHQFHV mL ^3UTVWf USLf VK ^VUSTWf m 3UVWTf V USVWTf V? VUTSWf Â TUSVWf UTSVWf Â VWUSTf Â TUVWSf V? UTVWSf c VWUTSf V WVUSTf VcL UVWSTf V W V U T Sf V? U W T Sf 7KHQ WKH GHJUHH VHTXHQFHV RI WKHVH YHUWH[ VHTXHQFHV DUH 'IO f r f 'VK f 'V f $UJ f 'V8 f $V f $ f $Lr f $ f $,R f $L f $V f $K f $ f $K f
PAGE 46
1RZ ZH KDYH WKDW 'V? 'V ' A6J fÂ§ 6 'Vf fÂ§ GÂD 'VQ 'V? 'VV fÂ§ '6 '6? O 6? Q fÂ§ ,f A 6c fÂ§ fnVJ f 'VL 'V^ 7KDW LV IRU VRPH GLIIHUHQW YHUWH[ VHTXHQFHV WKH GHJUHH VHTXHQFHV PLJKW EH VDPH $QG LI WKH GHJUHH VHTXHQFHV DUH VDPH WKHQ WKH WUHHV GHQRWHG E\ WKHVH GHJUHH VHn TXHQFHV DUH LVRPRUSKLF 5HPDUN 7KH VXP RI WKH WHUPV LQ DQ\ GHJUHH VHTXHQFH RI D JLYHQ QWUHH LV WKH VDPH QDPHO\ Q fÂ§ ,W LV LQ WKH DERYH H[DPSOH DQG LW LV VDPH DV WKH QXPEHU RI HGJHV LQ WKH WUHH 6HH 7KHRUHP EHORZ 6LQFH G? GYLf DQG G GXf fÂ§ IRU L Q ZH KDYH WKH IROORZLQJ SURSRVLWLRQ 3URSRVLWLRQ /HW 'V fÂ§ GL Â f f f GQf EH D GHJUHH VHTXHQFH RI D YHUWH[ VHn TXHQFH V RI DQ QWUHH 7 7KHQ G? Q fÂ§ DQG GÂ Q fÂ§ IRU L Q 3URSRVLWLRQ ,I GM A IRU DQ\ M Q WKHQ 9M? LV DQ DGMDFHQW YHUWH[ RI 9M 3URSRVLWLRQ /HW V XM X YQf EH D YHUWH[ VHTXHQFH RI DQ QWUHH 7 DQG 'c GL G Gff WKH GHJUHH VHTXHQFH RI V )RU DQ\ M f ^ Q fÂ§ ` ZLWK GM fÂ§ P LI WKH VHW ^XML XÂ 9N` LV WKH FROOHFWLRQ RI DOO WKH YMVXEEUDQFKHV H[FHSW WKH URRW VXEEUDQFK WKHQ N M P 3URRI 6LQFH GM P GYMf fÂ§ IURP 3URSRVLWLRQ WKH YHUWH[ 9M KDV P VXEEUDQFKHV EHVLGHV WKH URRW VXEEUDQFK 7KXV WKHUH DUH DW OHDVW P HQGYHUWLFHV LQ WKH FROOHFWLRQ RI WKRVH VXEEUDQFKHV RWKHU WKDQ WKH URRW EUDQFK 7KHUHIRUH DW OHDVW P WHUPV LQ GM GML Grf DUH +HQFH N M P B 7KHRUHP )LUVW 5HGXFHG 'HJUHH 6XP )RUPXODf )RU DQ\ GHJUHH VHTXHQFH 'V GL G GQf RI D YHUWH[ VHTXHQFH V ( 6 RI D JLYHQ QWUHH 7 ZH KDYH (/L Gc Q ?(? DQG (/A fÂ§ f
PAGE 47
3URRI G? LV WKH GHJUHH RI D YHUWH[ DQG WKH UHPDLQLQJ Q fÂ§ WHUPV DUH HDFK RQH OHVV WKDQ WKH GHJUHH RI D YHUWH[ %\ WKH GHJUHH VXP IRUPXOD 3URSRVLWLRQ f LW LV WUXH WKDW A= GZÂf ?(? 6R ( L GL GL (/ GL GWILf (,8 0Xmf f rf G>YLf <L UI8mf (f .0 (8 m0f mf VU L PR mf _e_QOf Q fÂ§ f fÂ§ Q fÂ§ f Q fÂ§ ?(? 7KHQ ZH KDYH WKDW GL (,8. a f G? ( GL a (,8 (/LÂÂmf Q f Q f B &RUROODU\ )RU DQ\ GHJUHH VHTXHQFH 'V RK G GQf RI D YHUWH[ VHTXHQFH V e 6 RI D JLYHQ QWUHH 7 ZH KDYH GL (Â GÂRGL fÂ§ f (M GL R 3URRI )URP WKH ILUVW UHGXFHG GHJUHH VXP IRUPXOD ZH KDYH WKDW GL (MA:nf 7KHQ G? (,8. fÂ§ GL (/ GARGL a f (f GL RGL f GL (f GAR. a f ( GL Rf GO (W GLÂRGL fÂ§ f fÂ§ ( G f 7KXV ZH JHW WKH UHVXOW 7 1RWH WKDW GL IRU DQ\ Â Q 6R G fÂ§ LI DQG RQO\ LI GL &RUROODU\ VKRZV WKDW LQ D GHJUHH VHTXHQFH HIL Â GQf WKH WRWDO QXPEHU RI WHUPV LV WKH VDPH DV G? ( GÂ fÂ§ f 7KHUHIRUH WKHUH DUH H[DFWO\ G ÂL H OB GÂRGL fÂ§ f HQGYHUWLFHV LQ WKH JLYHQ WUHH %HFDXVH RI WKLV ZH KDYH WKH IROORZLQJ SURSRVLWLRQ 3URSRVLWLRQ )RU DQ\ GHJUHH VHTXHQFH G" Â f ff GQf RI D YHUWH[ VHn TXHQFH V e 6 RI D JLYHQ QWUHH 7 ZH KDYH
PAGE 48
GL (Â Âmn f f IRU HYHU\ T Q 3URRI $VVXPH WKDW WKHUH LV T Q VXFK WKDW G? (/Af fÂ§ f 6LQFH 'c ÂL ÂGQf LV D GHJUHH VHTXHQFH RI D YHUWH[ VHTXHQFH V e 6 RI WKH JLYHQ QWUHH 7 ZH KDYH WKDW YTL e $XSf IRU VRPH S T 7KXV DW OHDVW RQH RI WKH HQGYHUWLFHV FRXQWHG E\ YS LV RQH RI YTLYT Â‘ YQ 7KDW LV LQ WKH VHFWLRQ GL Âf f f f GTf ZH KDYH DW OHDVW RQH OHVV WHUPV WKDQ G? (O GÂRGL aa f} LV" G? (8 GARGL f (Â GL L GL (/ GL fÂ§ f G? (" GÂRGL f (8 G G? (/ GLÂRGL f (8 G^ GL (/ GLÂRGL f (8 G 7KLV LV D FRQWUDGLFWLRQ 7KHUHIRUH ZH KDYH G? <" GL a f IrU HYHU< T Q ; 7KHRUHP 6HFRQG 5HGXFHG 'HJUHH 6XP )RUPXODf /HW 7 fÂ§ 9 (f EH DQ QWUHH /HW 'c G? Â f ff GQf EH WKH GHJUHH VHTXHQFH RI D YHUWH[ VHTXHQFH V e 6 RI WKH JLYHQ QWUHH 7 )RU DQ\ MN OHW %nYMNf EH D YM9MNfVXEEUDQFK YHUWH[ VHW 7KHQ ZH KDYH etm f R DQG GMN (L MNLGL fÂ§ f A R IRU HYHU\ MN T MN 3URRI /HW (%nYMNff ^^XX` e (7f?XY e %nYMNf` 7KHQ 7YMNf %nYMNf(%nYMNfffF7 LV DOVR D WUHH DQG %nYMNff YMN YMN ZÂLf LV D YHUWH[ VHTXHQFH RI WKH WUHH 7YMNf 1RWH WKDW GYMNf fÂ§ GMN LQ 7 6LQFH 9M e %nYMNf WKH HGJH ^YMYMN` LV QRW LQ WKH HGJH VHW (%fYMNff 7KXV ZH KDYH WKDW GYMNf GMN LQ WKH VXEWUHH
PAGE 49
7YMNf 7KHQ WKH UHVXOW FRPHV IURP WKH ILUVW UHGXFHG GHJUHH VXP IRUPXOD 7KHRUHP f DQG 3URSRVLWLRQ / &RUROODU\ /HW V WT L" Â‘ Â‘ YQf EH D YHUWH[ VHTXHQFH RI DQ QWUHH 7 DQG 'V fÂ§ G?G GQf WKH GHJUHH VHTXHQFH RI V )RU DQ\ GM LI WKH VXEVHTXHQFH YM 9ML YSf LV WKH FROOHFWLRQ RI DOO WKH 9MEUDQFKHV H[FHSW WKH URRW VXEEUDQFK RI 9M WKHQ GM
PAGE 50
Df GL Ef 7cL LGL QO Ff )RU DQ\ M Q ZLWK GM P WKHUH DUH H[DFWO\ P LQWHJHUV M ML M M f f f MUQ MUQ m LI M f VXFK WKDW (L ÂcLG f IRU IF P Gf )RU DQ\ MN ZLWK GMN O WKHUH DUH H[DFWO\ O LQWHJHUV MN MN[ MN O MN f f f MN MN^Of MNL VXFK WKDW 3 (6e7GWOf R IRU S fÂ§ 7KHRUHP /H fÂ§ GM G" f f f AQf D QRQQHJDWLYH LQWHJHU VHTXHQFH 7KHQ LV D GHJUHH VHTXHQFH RI D YHUWH[ VHTXHQFH RI DQ QWUHH LI DQG RQO\ LI VDWLVILHV '6& 3URRI > !f@ 7KLV IROORZV IURP 3URSRVLWLRQ 7KHRUHP DQG 7KHRUHP > @ /HW 9 ^YL 8f f f f Xf` EH D VHW RI Q GLIIHUHQW YHUWLFHV /HW GL G f f f GQf EH D QRQQHJDWLYH LQWHJHU VHTXHQFH VDWLVI\LQJ WKH DERYH FRQGLWLRQV 6LQFH GL DQG GÂ IRU DQ\ L Q ZH FDQ DVVLJQ WKH GHJUHH RI HDFK YHUWH[ f 9 E\ G^\?f G? DQG GY^f G IRU L Q /HW M Q DQG GM P )URP Ff WKHUH DUH H[DFWO\ P LQWHJHUV M M? M" MP MPL VXFK WKDW ML M DQG GMN (Â MrL.n f IRU N P 1RZ ZH DVVLJQ DQ HGJH EHWZHHQ 9M DQG 9MN IRU N P 7KDW LV YMN f $^YMf IRU Â P Lf ,I GMN O ZH DVVLJQ DQ HGJH EHWZHHQ 9MN DQG 9MNS IRU S E\ Gf :H GR D VLPLODU SURFHGXUH IRU HDFK YHUWH[ 9M DVVLJQHG WR GM 7KHUHIRUH YMN H $YMf DQG YMNS ( $YMNf IRU N P M LLf ,I GM^ WKHQ ZH GR QRW DVVLJQ DQ\ QHZ HGJH IRU WKH YHUWH[ 9MW %XW LW LV DOUHDG\ MRLQHG WR WKH YHUWH[ 9M E\ DQ HGJH /HW V YL9 YQf 7KHQ DOO WKH YHUWLFHV LQ V DUH FRQQHFWHG 6LQFH HDFK YHUWH[ 9L LV DVVLJQHG GL QHZ HGJHV DQG ( L GÂ Q fÂ§ IURP Ef ZH KDYH Q YHUWLFHV DQG
PAGE 51
Q fÂ§ HGJHV 7KHQ WKH VHW RI DOO WKH HGJHV DVVLJQHG DERYH WRJHWKHU ZLWK WKH YHUWH[ VHW 9 IRUPV DQ QWUHH E\ 3URSRVLWLRQ B/ 5HPDUN 7KH SURRI RI 7KHRUHP VKRZV KRZ WR FRQVWUXFW D WUHH WKDW FRUUHn VSRQG WR WKH JLYHQ GHJUHH VHTXHQFH VDWLVI\LQJ '6& ([DPSOH /HW f EH D QRQQHJDWLYH LQWHJHU VHTXHQFH KDYLQJ HQWULHV /HW 9 fÂ§ ^ILW! f f LYQ` EH D VHW RI Q GLIIHUHQW YHUWLFHV A IQ 9O 9 Y 9O2 9 9 9 9 8Q Y )LJXUH Df &OHDUO\ (Â L GÂ DQG G? 6R ZH H[SHFW D WUHH Ef 6LQFH G\Lf fÂ§ WKHUH DUH GLIIHUHQW LQWHJHUV OL , VXFK WKDW G Of O O O Of G (Â QGÂ fÂ§ f fÂ§ fÂ§ f DQG Â :H DVVLJQ HGJHV EHWZHHQ 9? DQG HDFK RI XL!L 8 Ff Lf 6LQFH Y WKHUH GLIIHUHQW LQWHJHUV VXFK WKDW Â (/M:n fÂ§ f fÂ§ fÂ§ fÂ§ f DQG G Â fÂ§ f fÂ§ :H DVVLJQ HGJHV EHWZHHQ Y DQG HDFK RI 9 DQG X LLf 6LQFH X WKHUH DUH GLIIHUHQW LQWHJHUV VXFK WKDW GQ DQG Â :H DVVLJQ HGJHV EHWZHHQ L!LR DQG HDFK RI W!Q DQG 9L LLLf 6LQFH YL fÂ§ ZH GR QRW DVVLJQ DQ\ QHZ HGJH IRU 8 %XW WKH YHUWH[ 9 LV FRQQHFWHG WR LT E\ DQ HGJH
PAGE 52
,Q D VLPLODU ZD\ ZH JHW ^X Y` ^L! X` ^X X` ^X8` DQG ^QQ` /HW ( EH WKH FROOHFWLRQ RI DOO WKH HGJHV DERYH WKDW LV ( ^^m X` ^L!L YLR` ^YX 9` ^A X` ^X XV` ^XLR 8Q` ^XLR 8` ^X 8` ^X 8` ^X X` ^X X` ^X X`` 6LQFH WKH HGJH VHW ( FRQWDLQV HOHPHQWV WKH VHW 7 fÂ§ ^9 (`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Â‘ fÂ‘L3Nf DQG ( fÂ§ TL T Tcf ( LI DQG RQO\ LI HLWKHU WKHUH LV P PLQ^$` VXFK WKDW SP TP DQG SL fÂ§ TL IRU DOO L P fÂ§ RU N O DQG SL Tc IRU DOO L N :H ZULWH ( LI ( RU fÂ§ ( 3URSRVLWLRQ /HW EH WKH VHW RI DOO GHJUHH VHTXHQFHV RI DQ QWUHH 7KH GHJUHH VHTXHQFH VHW f LV D WRWDOO\ RUGHUHG ILQLWH VHW 3URRI )RU DQ\ WZR GLIIHUHQW GHJUHH VHTXHQFHV WKH RUGHU GHILQHG DERYH LV ZHOO GHILQHG 6LQFH HDFK WHUP GW RI D GHJUHH VHTXHQFH LV QRQQHJDWLYH DQG OHVV WKDQ Q DQG WKH OHQJWK RI HDFK GHJUHH VHTXHQFH LV Q WKH GHJUHH VHTXHQFH VHW LV D ILQLWH VHW 7KXV f LV D WRWDOO\ RUGHUHG ILQLWH VHW B/ 'HILQLWLRQ )RU D JLYHQ QWUHH 7 OHW 6 EH WKH YHUWH[ VHTXHQFH VHW RI 7 DQG WKH GHJUHH VHTXHQFH VHW RI 7 7KHQ WKH PD[LPXP GHJUHH VHTXHQFH LQ LV FDOOHG WKH SHUIHFW VHTXHQFH RI WKH JLYHQ WUHH 7 DQG LW LV GHQRWHG E\ SI7f )RU WKH SHUIHFW VHTXHQFH SI7f RI D JLYHQ WUHH 7 WKHUH LV D YHUWH[ VHTXHQFH V LT X YQf e
PAGE 53
6 VXFK WKDW WKH GHJUHH VHTXHQFH RI V LV SI^7f :H FDOO V D SHUIHFW YHUWH[ VHTXHQFH RI WKH WUHH DQG 9? D SHUIHFW YHUWH[ RI WKH WUHH ,Q WKH ([DPSOH WKH SHUIHFW VHTXHQFH LV ` DQG WKHUH DUH WZR GLIIHUHQW SHUIHFW YHUWH[ VHTXHQFHV 6Q U VWS Tf DQG Vc UVWTSf IRU LW %XW WKH WUHHV UHSUHVHQWHG E\ 6Q DQG 6 DUH LVRPRUSKLF )RU DQ\ JLYHQ WUHH WKH SHUIHFW VHTXHQFH LV XQLTXH E\ 3URSRVLWLRQ %XW WKHUH PD\ H[LVW VHYHUDO GLIIHUHQW SHUIHFW YHUWH[ VHTXHQFHV ,I WKH JLYHQ WUHH LV D ODEHOHG WUHH WKHQ WKHVH SHUIHFW YHUWH[ VHTXHQFHV GHQRWH GLIIHUHQW ODEHOHG WUHHV %XW WKHVH ODEHOHG WUHHV DUH LVRPRUSKLF WR HDFK RWKHU WKDW LV WKHUH LV D XQLTXH XQODEHOHG WUHH 7KHUHIRUH IRU D JLYHQ SHUIHFW VHTXHQFH WKHUH H[LVWV D XQLTXH XQODEHOHG WUHH 7KHRUHP /HW Gc EH WKH SHUIHFW VHTXHQFH RI D WUHH 7 ,I 7n LV DQ\ RWKHU WUHH ZKRVH SHUIHFW VHTXHQFH LV DOVR G WKHQ 7n LV LVRPRUSKLF WR 7 3URRI ,I 7 KDV PRUH WKDQ RQH YHUWH[ VHTXHQFH ZLWK WKH VDPH GHJUHH VHTXHQFH WKHQ WKHVH YHUWH[ VHTXHQFHV DOO SURGXFH WKH LVRPRUSKLF WUHHV E\ XVLQJ WKH FRQVWUXFWLRQ LQ WKH SURRI RI 7KHRUHP 7 %HFDXVH RI WKLV WKHRUHP ZH KDYH RXU PDLQ WKHRUHP EHORZ 7KHRUHP 7UHH &ODVVLILFDWLRQ 7KHRUHPf )RU DQ\ SRVLWLYH LQWHJHU Q OHW 7Qf EH WKH VHW RI XQODEHOHG QWUHHV DQG 9Qf WKH VHW RI SHUIHFW VHTXHQFHV RI QWUHHV 7KHQ WKHUH LV D RQHWRRQH FRUUHVSRQGHQFH EHWZHHQ 7Qf DQG 9Qf 2EWDLQLQJ 3HUIHFW 6HTXHQFHV %\ 7KHRUHP LI ZH FDQ FRQVWUXFW WKH VHW RI SHUIHFW VHTXHQFHV RI QWUHHV 3Qf ZLWKRXW EHLQJ JLYHQ DQ\ WUHH VKDSH WKHQ ZH FDQ FODVVLI\ DOO WKH QWUHHV E\ WKLV VHW $QG LQ WKH VDPH ZD\ DV LQ ([DPSOH ZH FDQ JHW WKH GLIIHUHQW WUHH VKDSHV IURP WKH HOHPHQWV RI 9Qf 3URSRVLWLRQ )RU WKH SHUIHFW YHUWH[ LT RI D SHUIHFW YHUWH[ VHTXHQFH V WT Y YQf ZH KDYH G^\Lf GYLf IRU Â Q
PAGE 54
3URRI ,I WKHUH LV VRPH N f ^ Q` VXFK WKDW G^Y G^Y Lf WKHQ 'V GYLff 'J fÂ§ GYNf Â‘ Â‘ f IRU DQ\ YHUWH[ VHTXHQFH Vn fÂ§ WTf 7KLV FRQWUDn GLFWV WKH PD[LPDOLW\ RI 'V 7 3URSRVLWLRQ ,I 'V fÂ§ GL G GQf LV WKH SHUIHFW VHTXHQFH RI D JLYHQ QWUHH 7 WKHQ Q fÂ§ GL GÂ GQ IRU DQ\ L Q fÂ§ 3URRI 6LQFH 7 LV DQ QWUHH DQG 'V LV WKH SHUIHFW VHTXHQFH ZH KDYH Q fÂ§ G? )RU DQ\ YHUWH[ X e 97f ZH KDYH GLTf GXf E\ 3URSRVLWLRQ 6LQFH G? G^YLf DQG G^ GXWf fÂ§ IRU HYHU\ L Q ZH KDYH GL GL IRU L Q 6LQFH GQ PXVW GHQRWH RQH RI WKH HQG YHUWLFHV HDFK RI ZKLFK KDV GHJUHH ZH KDYH GQ 7KHUHIRUH ZH KDYH GW GQ IRU HYHU\ L Q fÂ§ 7 'HILQLWLRQ /HW GO G GQf EH D GHJUHH VHTXHQFH 7KHQ ZH VD\ WKDW VDWLVILHV WKH RUGHUHG VXEEUDQFK FRQGLWLRQ GHQRWHG 26E& LI DQG RQO\ LI VDWLVILHV WKH IROORZLQJ FRQGLWLRQ IRU DOO M ,I GXMf P DQG I7XMf %nYMPf ZLWK M ML M MP DUH WKH VXEEUDQFKHV RI WKH YHUWH[ 9M RWKHU WKDQ WKH URRW VXEEUDQFK WKHQ %GfYMNff %GnYMNff IRU HYHU\ N P fÂ§ 5HFDOO WKDW %GnA9MIff LV D VHFWLRQ RI WKH GHJUHH VHTXHQFH 6HH 'HILQLWLRQ DQG f 5HPDUN ,I M WKHQ WKHUH LV QR URRW EUDQFK RI LT LQVWHDG WKHUH LV RQH PRUH VXEEUDQFK %GnYMPOf VXFK WKDW MP MPL DQG %nYMPff %GnYMPff 7KHRUHP /HW GL G GQf EH D QRQQHJDWLYH LQWHJHU VHTXHQFH ,I LV WKH SHUIHFW VHTXHQFH RI DQ QWUHH WKHQ LW VDWLVILHV WKH '6& DQG WKH 26E& 3URRI ,W LV FOHDU IURP 7KHRUHP DQG WKH PD[LPDOLW\ RI B/ 3URSRVLWLRQ /HW 7 EH D WUHH DQG Y D YHUWH[ ZLWK WKH ODUJHVW GHJUHH LQ 7 7KHQ WKHUH LV H[DFWO\ RQH GHJUHH VHTXHQFH 'Y IRU 7 VWDUWLQJ DW Y DQG VDWLVI\LQJ ERWK '6& DQG WKH 26E&
PAGE 55
%\ WKH DERYH SURSRVLWLRQ WKH FRQYHUVH RI 7KHRUHP LV QRW WUXH 7KH IROORZLQJ H[DPSOH LV D FRXQWHUH[DPSOH ([DPSOH /HW $ G? b Âf f 7KHQ ÂL DQG GL $QG WKH VHTXHQFH $ FDQ EH VHSDUDWHG LQWR VHFWLRQV OLNH f f f f 7KH VHFRQG VHFWLRQ FDQ EH VHSn DUDWHG LQWR VHFWLRQV f f f DQG WKH VHFRQG DQG WKLUG VHFn WLRQ RI WKHVH FDQ EH VHSDUDWHG E\ f f f DQG f f f UHVSHFn WLYHO\ $OO RI WKHVH VXEVHFWLRQV VDWLVI\ WKH VHFRQG DQG WKLUG FRQGLWLRQV RI '6& 7KXV WKH VHTXHQFH $ VDWLVILHV WKH '6& 7KDW LV WKH VHTXHQFH $ LV D GHJUHH VHn TXHQFH RI D YHUWH[ VHTXHQFH LT W! 9f RI D WUHH E\ 7KHRUHP 2Q WKH RWKHU KDQG WKH VHTXHQFH $ VDWLVILHV WKH 26E& VLQFH ZH KDYH f f f %XW $ LV QRW WKH SHUIHFW VHTXHQFH RI WKH WUHH ,Q IDFW ZH UHDUUDQJH WKH VHTXHQFH $ WR JHW % G G" GL fÂ§ GMR GQ Â G G G Gt GJ GJf f 1RWH WKDW GY?f fÂ§ G? DQG GYMf fÂ§ Â &OHDUO\ WKH VHTXHQFH % DOVR VDWLVILHV WKH '6& DQG WKH 26E& 6R % LV DOVR D GHJUHH VHTXHQFH RI DQRWKHU YHUWH[ VHTXHQFH Y\ Y\ X Y\\ 8 8 UT X YT Y YTf RI WKH WUHH 7KDW LV $ DQG % DUH GHJUHH VHTXHQFHV RI WKH VDPH WUHH VHH )LJXUH f $QG ZH KDYH % f f $ 7KXV $ LV QRW WKH SHUIHFW VHTXHQFH )XUWKHUPRUH $ FDQQRW EH WKH SHUIHFW VHTXHQFH RI DQ\ WUHH EHFDXVH $ DQG % GHQRWH LVRPRUSKLF WUHHV DQG % $ YQ 9LR 9\ 8
PAGE 56
,Q WKLV H[DPSOH ZH KDYH WKDW GYLf G^Yf ZKLFK LV WKH ODUJHVW GHJUHH LQ WKH WUHH 7KDW LV WKH SHUIHFW VHTXHQFH PXVW VWDUW ZLWK WKH ODUJHVW GHJUHH :H ZLOO VHH LQ ([DPSOH WKDW WKH GHJUHH VHTXHQFH % LV WKH SHUIHFW VHTXHQFH RI WKLV WUHH 1RWH WKDW f LV WKH RQO\ SRVVLEOH GHJUHH VHTXHQFH ZKHQ G? fÂ§ 7KHUHIRUH ZH DVVXPH G? IURP QRZ RQ 1RZ ZH VKRZ KRZ WR FRQVWUXFW DOWHUQDWH GHJUHH VHTXHQFHV IURP $ VLPLODU WR % $VVXPH WKDW ZH KDYH D QRQQHJDWLYH LQWHJHU VHTXHQFH $ FIL G Â‘ Â‘ GQf VDWLVI\LQJ WKH '6& DQG WKH 26E& ,I WKHUH LV L Q VXFK WKDW G? G WKHQ $ FDQQRW EH WKH SHUIHFW VHTXHQFH RI DQ\ WUHH E\ 3URSRVLWLRQ 6R ZH DVVXPH WKDW G? G^ IRU L fÂ§ Q DQG WKHUH LV DW OHDVW RQH LQWHJHU M Q VXFK WKDW GM G? fÂ§ 7KDW LV WKHUH LV DQ QWUHH 7 9 (f VXFK WKDW $ LV D GHJUHH VHTXHQFH RI D YHUWH[ VHTXHQFH V UT 9 YQf RI 7 DQG 7 KDV DW OHDVW WZR YHUWLFHV RI PD[LPXP GHJUHH G? &OHDUO\ GM 7KXV 9M LV QRW DQ HQG YHUWH[ LQ 7 %\ &RUROODU\ WKH VHW RI WKH YHUWLFHV RI WKH URRW EUDQFK RI 9M LV QRW WKH VHW RI YHUWLFHV LQ $ LWVHOI 6R IURP 3URSRVLWLRQ WKHUH LV M P Q VXFK WKDW 5%YMf ^XM Xf f f f 9M` 8 ^X72L XP f f f YQ` LV WKH VHW RI WKH YHUWLFHV RI WKH URRW EUDQFK RI YM &OHDUO\ ZML XÂ YPf LV WKH VHFWLRQ FRQWDLQLQJ DOO RI WKH VXEEUDQFKHV RI 9M H[FHSW WKH URRW EUDQFK 7KHUHIRUH LI GM fÂ§ S WKHQ WKHUH DUH S VXEVHFWLRQV RI $ OLNH %GnA9MAf %GnYMff ! GnXSff ZLWK M ML M MS P O VLQFH $ VDWLVILHV 26E& 2Q WKH RWKHU KDQG OHW (5%nYMff ^^XX` e (7f?XY e 5%nYMf` EH WKH VHW RI HGJHV LQ WKH URRW VXEEUDQFK R 9M 7KHQ 9 ^5%nYMf (5%nYMff` LV D VXEWUHH RI 7 /HW 67nf EH WKH YHUWH[ VHTXHQFH VHW RI 7n %\ 3URSRVLWLRQ WKHUH LV D XQLTXH LQWHJHU $ M VXFK WKDW 9N f $^YMf &OHDUO\ 9N f 5%nYMf /HW 69N7nf EH WKH XAYHUWH[ VHTXHQFH VHW RI 67nf LQ ZKLFK WKH ILUVW WHUP RI HDFK YHUWH[ VHTXHQFH LV Xr /HW '9N7nf EH WKH 9NGHJUHH VHTXHQFH VHW RI 69N7nf VHH 'HILQLWLRQ f 6LQFH '9N7nf LV D ILQLWH VHW WKHUH LV D PD[LPXP VHTXHQFH LQ '9N7nf :H GHQRWH WKH PD[LPXP VHTXHQFH LQ
PAGE 57
'9N7nf E\ PG5YNf 1RWH WKDW PXVW EH FKDQJHG WR G? fÂ§ LQ PG5YNf VLQFH LW LV QRW JRLQJ WR EH XVHG IRU WKH ILUVW WHUP DQ\PRUH LQ WKH QHZ GHJUHH VHTXHQFH 1RZ ZH KDYH WKDW HLWKHU PG5YNf %GnYMff RU WKHUH LV DQ LQWHJHU T S VXFK WKDW %GnYMTff PG5YNf %GnYMT f 1RWH WKDW PG5YNf PD\ EH WKH VPDOOHVW VHFWLRQ WKHQ 9MTO ZRXOG QRW H[LVW :H UHDUUDQJH WKH JLYHQ GHJUHH VHTXHQFH $ WR D QHZ VHTXHQFH LQ ZKLFK GM LV WKH ILUVW WHUP DQG WKH VXEEUDQFK GHJUHH VHTXHQFHV RI 9M PG5YNf %G^YKff %GnYKff %GnYMSff DUH OLVWHG LQ GHVFHQGLQJ RUGHU 'HQRWH WKLV QHZ GHJUHH VHTXHQFH E\ 6LQFH WKH VXEEUDQFK GHJUHH VHTXHQFHV RI YM DUH OLVWHG LQ GHVFHQGLQJ RUGHU LQ 0\ WKH QHZ GHJUHH VHTXHQFH 0A VDWLVILHV WKH 26E& 6LQFH LV D UHDUUDQJHPHQW RI $ ZLWK GL GYMf GM GL GYLf fÂ§ FIL fÂ§ IRU VRPH L DQG $ VDWLVILHV WKH '6& WKH QHZ GHJUHH VHTXHQFH 0A VDWLVILHV WKH FRQGLWLRQV Df DQG Ef RI WKH GHILQLWLRQ RI WKH '6& 6LQFH PG5YNf LV WKH SHUIHFW VHTXHQFH RI 7n LW VDWLVILHV WKH FRQGLWLRQV Ff DQG Gf RI WKH GHILQLWLRQ RI WKH '6& &OHDUO\ HDFK VHFWLRQ %GnYMWff VDWLVILHV WKH FRQGLWLRQV Ff DQG Gf RI WKH GHILQLWLRQ RI WKH '6& 7KHUHIRUH WKH QHZ GHJUHH VHTXHQFH VDWLVILHV WKH '6& 7KXV WKH QHZ GHJUHH VHTXHQFH 0A VDWLVILHV ERWK WKH '6& DQG WKH 26E& )RU WKH JLYHQ QRQQHJDWLYH LQWHJHU VHTXHQFH $ G? f f f GQf WKH VHW ^0\MGYMf fÂ§ GYLf` LV D QRQHPSW\ ILQLWH VHW VLQFH $ ^0\GYMf fÂ§ G^LTf` 7KHUHIRUH WKHUH LV D PD[LPXP HOHPHQW LQ WKLV VHW /HW 3D PD[^0A?GYMf fÂ§ GXLf` 1RZ $ GHWHUPLQHV D XQLTXH WUHH 7 6XSSRVH % LV DQRWKHU YHUWH[ VHTXHQFH IRU 7 VDWLVI\LQJ WKH '6& DQG WKH 26E& 7KHQ ^0\MGYTf GXMf` ^0A_GXÂf ÂLTf` ZKHUH YT LV DQ DQDORJXH RI YM E\ 3URSRVLWLRQ 7KHUHIRUH 3D LV WKH SHUIHFW VHTXHQFH IRU 7 %\ WKH DERYH H[SODQDWLRQ WRJHWKHU ZLWK 3URSRVLWLRQ DQG 7KHRUHP ZH KDYH WKH IROORZLQJ WKHRUHP
PAGE 58
7KHRUHP /HW $ G? G GQf EH D QRQQHJDWLYH LQWHJHU VHTXHQFH VDWLVI\n LQJ WKH '6& DQG WKH 26E& ZLWK G? GÂ IRU L fÂ§ Q 7KHQ $ 3D LI DQG RQO\ LI $ LV WKH SHUIHFW VHTXHQFH RI DQ QWUHH ,Q WKH IROORZLQJ H[DPSOH ZH VKRZ WKH GHJUHH VHTXHQFH % LQ ([DPSOH LV WKH SHUIHFW VHTXHQFH RI WKH WUHH ([DPSOH ,Q ([DPSOH $ fÂ§ G? Â f f f M Af f VDWLVILHV WKH '6& DQG WKH 26E& DQG G G G GL fÂ§ 7KDW LV GYf GYf GYf fÂ§ G\Lf 7KXV PD[^0A_M ` LV WKH SHUIHFW VHTXHQFH RI WKH WUHH )RU Y ZH KDYH WKDW 5%Yf ^XL Y Y` 8 ^XLR XQ 8` DQG %GnYJff f %GnYff 7KXV 7n ^5%?Yf(5%?Yf` LV WKH VXEWUHH SURGXFHG E\ WKH URRW VXEEUDQFK RI Y 6HH )LJXUH f 6LQFH Y e $Yf ZH JHW WKH YYHUWH[ VHTXHQFH VHW 699f ^ Y 9? 8 ,fOO 8 9 9 YHf A 9L 9 8Q 8 Y X Y Yf X 9L 8 8 9 Y Y Y Yf Y 9L9 8 9 Y 9* Y Yf Y 9 9 96 9 99 YQ Yf Y 9 9 9V 9 9L9L LR f : 9 9V 9L9LR 9X 9f m r! 9 9V 9L 9 9 O!OOf `f 7KHQ WKH XGHJUHH VHTXHQFH VHW RI 697nf LV D HQ ^ f f f f f f f f ` ^ f f f f `
PAGE 59
7KHUHIRUH ZH KDYH WKH PD[LPXP VHTXHQFH PG5Yf LQ '97nf DV PG5\Lf fÂ§ f 6LQFH PG5Yf f f %Gn^Yff %GnYff ZH KDYH 0e f ,Q WKH VDPH ZD\ ZH JHW 0r f 0r f 7KHUHIRUH PD[^09M?M ` 09 6LQFH f f $ 0Âf ZKLFK LV % LQ WKH ([DPSOH LV WKH SHUIHFW VHTXHQFH RI WKH WUHH 9OO Y 9L 8 9 9 9 9 9 )LJXUH %\ WKH 7UHH &ODVVLILFDWLRQ 7KHRUHP 7KHRUHP f DQG WKH SUHYLRXV H[DPSOH ZH KDYH WKH IROORZLQJ LPSRUWDQW FRUROODU\ &RUROODU\ /HW $ IL Â f f f GQf EH D QRQQHJDWLYH LQWHJHU VHTXHQFH VDWLVn I\LQJ WKH '6& DQG WKH 26E& ,I G fÂ§ IRU HYHU\ L Q WKHQ $ LV WKH SHUIHFW VHTXHQFH IRU DQ QWUHH :H ZLOO VHH LQ WKH IROORZLQJ H[DPSOH KRZ ZH FRQVWUXFW D WUHH VKDSH IURP D JLYHQ PD[LPDO GHJUHH VHTXHQFH ([DPSOH /HW RIL G f f f GLf f EH D PD[LPDO GHJUHH VHTXHQFH 7KHQ ZH H[SHFW D WUHH 7 /HW V WT Y Xf EH WKH YHUWH[ VHTXHQFH KDYLQJ DV WKH GHJUHH VHTXHQFH 7KHQ GY f G[ VR WKH YHUWH[ Y
PAGE 60
KDV HGJHV 6HH )LJXUH f 7KRXJK WKHUH LV QR VSHFLILF RUGHU IRU WKHVH HGJHV ZH ODEHO WKHP DV HLHH IRU WKH FRQYHQLHQFH RI H[SODQDWLRQ H H YL HL )LJXUH )URP WKH GHILQLWLRQ RI YHUWH[ VHTXHQFH 'HILQLWLRQ f X LV RQH RI WKH HOHPHQWV RI WKH DGMDFHQW VHW RI LT $LTf :LWKRXW ORVV RI JHQHUDOLW\ OHW HM ^XLX` 7KHQ X KDV fQHZf HGJHV H DQG H VLQFH G /HW H ^XX` %\ WKH VDPH UHDVRQLQJ Y KDV fQHZf HGJH QDPHG HA 6R Ht ^XX` 6LQFH G fÂ§ Y KDV fQHZf HGJHV QDPHG H DQG H 6HH )LJXUH f H A H H 9? HO O! H 9 A Y H )LJXUH 6LQFH G DQG G QHLWKHU Y QRU Y KDV D fQHZf HGJH WKDW LV X DQG Y DUH HQG YHUWLFHV 2Q WKH RWKHU KDQG Y DQG YH DUH WKH DGMDFHQW YHUWLFHV RI W! VLQFH G 7KXV Y LV DQ DGMDFHQW YHUWH[ RI RQH RI ^XLXX` 6LQFH G DQG Y LV WKH DGMDFHQW YHUWH[ RI Y Y LV QRW DQ DGMDFHQW YHUWH[ RI Y 6LQFH G DQG ZH KDYH RQO\ RQH DGMDFHQW YHUWH[ RI X QDPHO\ Y ZH DGG DQ DGMDFHQW YHUWH[ Y 6HH )LJXUH f Y YH H H H Y &L 9 H 9 H Y H Y )LJXUH
PAGE 61
6LQFH G Y KDV fQHZf HGJH ZKLFK FRQQHFWV ZLWK YV 6LQFH G WKHUH LV QRfQHZfHGJH IURP Y 6LQFH ZH FRYHUHG XS WR WKH UHPDLQLQJ WZR YHUWLFHV YT DQG L!LR DUH DGMDFHQW YHUWLFHV RI UT ZKLFK KDV WZR fXQXVHGf HGJHV H DQG H /HW H ^XX` DQG H ^XLXLR` 6LQFH ZH DUH WKLQNLQJ DERXW DQ XQODEHOHG WUHH DQG ZH SXW WKH HGJH QDPHV IRU RXU FRQYHQLHQFH WKHUH DUH QR VSHFLILF QDPHV IRU WKH YHUWLFHV DQG HGJHV 7KXV ZH JHW WKH IROORZLQJ WUHH 6HH )LJXUH f mf XLRf W!Lf 0 0 Yf Yf )LJXUH mmf 0 ([WHQGHG 3DUWLWLRQV DQG WKH $OJRULWKP ,Q 'HILQLWLRQ DQG WKH IROORZLQJ H[SODQDWLRQ LW ZDV VKRZQ WKDW DQ\ WUHH FDQ EH UHSUHVHQWHG E\ D XQLTXH SHUIHFW VHTXHQFH 6R LQ WKLV VHFWLRQ ZH GHYHORS DQ DOJRULWKP WR SURGXFH DOO WKH SHUIHFW VHTXHQFHV IRU WKH QWUHHV IRU DQ\ SRVLWLYH LQWHJHU Q )URP WKHVH SHUIHFW VHTXHQFHV ZH JHW WKH VKDSH RI HDFK QWUHH DV VKRZQ LQ WKH SUHYLRXV H[DPSOH DV ZHOO DV WKH WRWDO QXPEHU RI QWUHHV )URP 'HILQLWLRQ DQG WKH )LUVW 5HGXFHG 'HJUHH 6XP )RUPXOD 7KHRUHP f ZH VHH WKDW D GHJUHH VHTXHQFH IRU DQ Q OfWUHH DOZD\V KDV Q QRQQHJDWLYH LQWHJHUV DV HQWULHV DQG WKHVH LQWHJHUV DOZD\V DGG XS WR Q ZKLFK LV WKH QXPEHU RI WKH HGJHV RI WKH JLYHQ Q OfWUHH 7KXV RXU ILUVW VWHS LQ OLVWLQJ DOO SRVVLEOH GHJUHH VHTXHQFHV LV WR IRUP RUGHUHG SDUWLWLRQV RI WKH LQWHJHU Q DV VXPV RI VPDOOHU SRVLWLYH LQWHJHUV
PAGE 62
'HILQLWLRQ $Q RUGHUHG SDUWLWLRQ RI WKH SRVLWLYH LQWHJHU Q LV D VHTXHQFH RI SRVLWLYH LQWHJHUV SLS f f f 3Pf VXFK WKDW
PAGE 63
3URRI )URP WKH JLYHQ H[WHQGHG SDUWLWLRQ GL Â f f f GQf RI DQ RUGHUHG SDUWLWLRQ GML GM GMPf RI WKH LQWHJHU UH ZH KDYH WKDW
PAGE 64
$FFRUGLQJ WR WKH DERYH SURSRVLWLRQ WKH RQO\ WUHHV ZKRVH SHUIHFW VHTXHQFHV VWDUW ZLWK RU DUH WKH WUHH ZLWK ]HUR HGJH DQG WKH WUHH ZLWK RQH HGJH 7KH RQO\ GHJUHH VHTXHQFH UHSUHVHQWDWLRQ RI D WUHH ZLWK ]HUR HGJH LV f DQG WKH RQO\ GHJUHH VHTXHQFH UHSUHVHQWDWLRQ RI D WUHH ZLWK RQH HGJH LV f 7KHVH DUH HDV\ HQRXJK WR YLVXDOL]H DQG FRXQW VR IRU WKH DOJRULWKP ZH ZRUN RQ WKH DVVXPSWLRQ WKDW ZH DUH OLVWLQJ WUHHV ZLWK DW OHDVW WZR HGJHV 7KHQ WKH ILUVW WHUP RI WKH SHUIHFW VHTXHQFH ZLOO EH DW OHDVW 7KH IROORZLQJ SURSRVLWLRQ VKRZV WKDW ZH KDYH DQ DUF VKDSH WUHH LI WKH ILUVW WHUP RI WKH SHUIHFW VHTXHQFH LV 3URSRVLWLRQ ,I 'V HIL Â Â‘ f f GQLf LV WKH SHUIHFW VHTXHQFH RI DQ Q OfWUHH 7 DQG G? fÂ§ WKHQ GL IRU L Q fÂ§ DQG GQ GQ 3URRI %\ 3URSRVLWLRQ ZH KDYH WKDW f ^` IRU DQ\ L Q VLQFH Â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f )RU WKH VDPH WUHH WKH QXPEHUV LQ D GHJUHH VHTXHQFH FDQQRW FKDQJH EXW RQO\ WKH RUGHU LQ ZKLFK WKH\ RFFXU FDQ FKDQJH E\ FKDQJLQJ WKH SRVLWLRQ RI ]HURV $QG FOHDUO\ WKH QHZ RUGHUHG VHTXHQFH 6R WKH VHTXHQFH PXVW EH WKH SHUIHFW VHTXHQFH RI 7 7KDW LV 'c s
PAGE 65
0RVW RI WKH H[WHQGHG SDUWLWLRQV RI Q VWDUW ZLWK D RU D 6LQFH WKH ODVW WZR SURSRVLWLRQV FODVVLI\ WUHHV ZKRVH GHJUHH VHTXHQFHV VWDUW ZLWK WKHVH YDOXHV ZH GR QRW QHHG WR FRQVLGHU RUGHUHG SDUWLWLRQV WKDW VWDUW ZLWK RU 3URSRVLWLRQ VKRZV WKDW IRU DQ\ JLYHQ QXPEHU Q RI HGJHV WKHUH LV RQO\ RQH WUHH ZLWK Q HGJHV ZKRVH SHUIHFW VHTXHQFH VWDUWV ZLWK D ,Q SDUWLFXODU LW LV D OLQHDU WUHH ZLWK Q HGJHV VLQFH GÂ IRU L Q fÂ§ :H DGG WKLV VHTXHQFH DW WKH HQG RI WKH DOJRULWKP %\ 3URSRVWLRQ ZH GR QRW QHHG WKH RUGHUHG SDUWLWLRQV LQ ZKLFK WKH ODUJHVW VXPPDQG DSSHDUV PRUH WKDQ RQFH RU LV QRW WKH ILUVW WHUP LQ WKH RUGHUHG SDUWLWLRQ ,I DQ RUGHUHG SDUWLWLRQ KDV HLWKHU RQH RU WZR HQWULHV RQO\ WKHQ ZH FDQ SURGXFH RQO\ RQH H[WHQGHG SDUWLWLRQ IRU HDFK FDVH 6HH WKH IROORZLQJ 3URSRVLWLRQ DQG &RUROODU\ 3URSRVLWLRQ /HW 'V fÂ§ G[ G f f f GQLf EH WKH SHUIHFW VHTXHQFH RI DQ Q Of WUHH 7 ZLWK G 7KHQ GL fÂ§ Q DQG GL IRU DOO L 3URRI /HW EH D YHUWH[ VHTXHQFH \LHOGLQJ D SHUIHFW VHTXHQFH 'c RI DQ Q f WUHH 7 6LQFH 'V LV D SHUIHFW VHTXHQFH DQG G GYf fÂ§ DOO RI WKH YHUWLFHV LQ WKH WUHH RWKHU WKDQ Y? KDYH GHJUHH VR DOO WKH WHUPV LQ RWKHU WKDQ G? PXVW EH ]HUR %\ WKH )LUVW 5HGXFHG 'HJUHH 6XP )RUPXOD 7KHRUHP f ZH KDYH G? fÂ§ Q B &RUROODU\ ,I DQ RUGHUHG SDUWLWLRQ 3 fÂ§ S?Sf RI Q KDV RQO\ WZR HQWULHV WKHQ WKH SURGXFHG H[WHQGHG SDUWLWLRQ G" G GQLf IURP 3 LV VXFK WKDW G[ 3L G S DQG G IRU L Q :H GHPRQVWUDWHG VHYHUDO QHFHVVDU\ FRQGLWLRQV ZKLFK DSSO\ WR D SHUIHFW VHn TXHQFH RI D WUHH 8QIRUWXQDWHO\ ZKLOH WKHVH FRQGLWLRQV DUH VXIILFLHQW LQ DOPRVW DOO FDVHV WKHUH DUH FHUWDLQ VHTXHQFHV ZKLFK DOWKRXJK WKH\ VDWLVI\ DOO WKH FRQGLWLRQV DUH QRW SHUIHFW VHTXHQFHV RI WKH WUHHV WKH\ UHSUHVHQW 7KH SUREOHPV XVXDOO\ DULVH ZKHQ WKH WUHH KDV WZR RU PRUH YHUWLFHV RI WKH ODUJHVW GHJUHH :H VDZ D FRXQWHUH[DPSOH LQ ([DPSOH
PAGE 66
1RWH WKDW LQ WKH ([DPSOH WKH WZR GHJUHH VHTXHQFHV $ DQG % KDYH WKH VDPH RUGHUHG SDUWLWLRQ 3 f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f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
PAGE 67
&+$37(5 75((6 $6 ,19$5,$176 )25 %5$,1 78025 6+$3(6 ,QWURGXFWLRQ $ VSKHUH SDFNLQJ WUHDWPHQW SODQ IRU D JLYHQ EUDLQ WXPRU LV SURGXFHG LQ WKH ILUVW FKDSWHU 7KLV SODQ FDQ EH UHSUHVHQWHG QXPHULFDOO\ DV D OLVW RI SDUDPHWHUV RI WKH GRVH VSKHUHV ,Q SDUWLFXODU WKH OLVW FRQVLVWV RI WKH FRRUGLQDWHV RI WKH GRVH FHQWHU LQ PLOOLPHWHUV DORQJ WKH ODWHUDO /DWf DQWHULRUSRVWHULRU $3f DQG D[LDO $[f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f WR JHW D PD[LPDO WUHH RI WKH JUDSK IRU D VSKHUH SDFNLQJ SODQ IRU D JLYHQ EUDLQ WXPRU ,Q WKH QHDU IXWXUH ZH SODQ WR XVH WKLV SURJUDP WR FODVVLI\ DOO WKH WUHDWPHQW SODQV ZKLFK KDYH EHHQ GRQH DW WKH %UDLQ ,QVWLWXWH RI WKH 8QLYHUVLW\ RI )ORULGD :H KRSH WKDW WKH FODVVLILFDWLRQ JLYHV VRPH LQVLJKW DERXW EUDLQ WXPRU VKDSHV ,Q WKLV FKDSWHU ZH DVVXPH WKDW D EUDLQ WXPRU LV FRQQHFWHG VR WKH JUDSK UHSUHVHQWDWLRQ IRU DQ\ EUDLQ WXPRU LV D FRQQHFWHG JUDSK
PAGE 68
'HILQLWLRQV DQG 6RPH 3URSHUWLHV 'HILQLWLRQ $ JUDSK fÂ§ 9*f (*ff FRQVLVWV RI D YHUWH[ VHW 9*f ^XL YQ` DQG DQ HGJH VHW (*f ^HL HP` & ^^X Y`_X Y f 9*f` LQ ZKLFK HDFK HGJH LV DQ XQRUGHUHG SDLU RI WZR GLVWLQFW YHUWLFHV X DQG Y DQG WKH HGJH ^X X` LV VDLG WR MRLQ X DQG Y )RU DQ HGJH ^XY` ZH VD\ WKDW X DQG Y DUH DGMDFHQW YHUWLFHV DQG WKH YHUWH[ X DQG WKH HGJH ^XY` DUH LQFLGHQW ZLWK HDFK RWKHU DV DUH Y DQG ^XW!` )URP QRZ RQ _3*nf_ GHQRWHV WKH QXPEHU RI YHUWLFHV DQG _(*f? GHQRWHV WKH QXPEHU RI HGJHV )RU D YHUWH[ X f 3*f WKH YHUWH[ VHW $JXf ^X f *?Y LV DQ DGMDFHQW YHUWH[ RI X` LV FDOOHG WKH DGMDFHQW VHW RI X LQ 6LQFH WKHUH LV QR ORRS UHSUHVHQWHG E\ D VLQJOH HGJH LQ D JUDSK LQ WKLV DUWLFOH QR YHUWH[ LV DGMDFHQW WR LWVHOI 'HILQLWLRQ :H VD\ WKDW D JUDSK LV LVRPRUSKLF WR D JUDSK + LI DQG RQO\ LI WKHUH H[LVWV D ELMHFWLRQ 9^*f fÂ§!Â‘ 9+f VXFK WKDW ^XX` e (*f LI DQG RQO\ LI ^mf IYf`f(+f 'HILQLWLRQ $ ZDON LV DQ DOWHUQDWLQJ VHTXHQFH XT HL WT H Hr 9Nf RI YHUn WLFHV DQG HGJHV VW XB L 9^ IRU L fÂ§ IF DQG H IRU HYHU\ L $ F\FOH LV D ZDON X HL 8L H HÂ 9Nf ZLWK N LQ ZKLFK Y 9N LV WKH RQO\ YHUWH[ UHSHDWHG $ SDWK LV D ZDON ZLWK QR UHSHDWHG YHUWH[ $ XYfSDWK LV D SDWK ZLWK WKH ILUVW YHUWH[ X DQG WKH ODVW YHUWH[ Y $ XXfSDWK LV D YHUWH[ X 7KH OHQJWK RI WKH XYfSDWK OXYf LV WKH QXPEHU RI HGJHV LQ WKH XYfSDWK 6R WKH OHQJWK RI X XfSDWK LV 7KH GLVWDQFH GXYf EHWZHHQ WZR YHUWLFHV X DQG Y LV WKH OHQJWK RI D VKRUWHVW SDWK MRLQLQJ WKHP LI DQ\ RWKHUZLVH GXYf fÂ§ RR 1RWH WKDW D SDWK RI OHQJWK LV RI OLQHDU VKDSH 'HILQLWLRQ /HW fÂ§ 9 (f EH D JUDSK ZKHUH 9 ^XL X YQ` 7KHQ IRU DQ\ YHUWH[ Y f 9 ZH VD\ WKDW 3*Yf M LI DQG RQO\ LI
PAGE 69
M PLQ^I f ^ Q`_ WKHUH LV D Y XfSDWK LQ *` :H FDOO 3JYf WKH SDWKYDOXH RU VLPSO\ WKH 3YDOXH RI WKH YHUWH[ Y LQ 'HILQLWLRQ /HW 9 (f EH D JUDSK )RU DQ\ QRQHPSWK VXEVHW 9 & 9 OHW ) ^ ^XX` f ( X Y f 9 ` & ( )RU DQ\ VXEVHW )n RI ) WKH JUDSK *n 9n )nf LV FDOOHG D VXEJUDSK RI $ JUDSK LV FRQQHFWHG LI DQG RQO\ LI HYHU\ SDLU RI YHUWLFHV DUH MRLQHG E\ D SDWK $ JUDSK KDYLQJ QR F\FOH LV DF\FOLF 1RWH WKDW DQ DF\FOLF JUDSK FDQ EH GLVFRQQHFWHG $ IRUHVW LV DQ DF\FOLF JUDSK $ WUHH LV D FRQQHFWHG DF\FOLF JUDSK $Q QWUHH LV D WUHH ZLWK Q YHUWLFHV 'HILQLWLRQ 7KH GHJUHH RI D YHUWH[ Y GHQRWHG E\ GYf LV WKH VXP RI WKH QXPEHU RI HGJHV LQFLGHQW ZLWK Y 'HILQLWLRQ $ FXWYHUWH[ RI D JUDSK LV D YHUWH[ RI GHJUHH JUHDWHU WKDQ ZKRVH UHPRYDO LQFUHDVHV WKH QXPEHU RI FRPSRQHQWV RI WKH JUDSK 'HILQLWLRQ $ QRQVHSDUDEOH JUDSK LV FRQQHFWHG QRQWULYLDO DQG KDV QR FXWYHUn WH[ $ EORFN RI D JUDSK LV D PD[LPDO QRQVHSDUDEOH VXEJUDSK 7KXV LI Y LV D FXWYHUWH[ RI D FRQQHFWHG JUDSK WKHQ fÂ§ Y LV GLVFRQQHFWHG 1RWH WKDW WKH WUDGLWLRQDO GHILQLWLRQ RI FXWYHUWH[ XVXDOO\ FDOOHG FXWSRLQW LQ JUDSK WKHRU\f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
PAGE 70
3URSRVLWLRQ >@ /HW Y EH D YHUWH[ RI D FRQQHFWHG JUDSK 7KH IROORZLQJ VWDWHPHQWV DUH HTXLYDOHQW f Y VHSDUDWHV DQG GYf f 7KHUH H[LVW YHUWLFHV X DQG Z GLVWLQFW IURP Y VXFK WKDW Y LV RQ HYHU\ XZf SDWK f 7KHUH H[LVWV D SDUWLWLRQ RI WKH VHW RI YHUWLFHV 9*f fÂ§ ^W!` LQWR VXEVHWV 8 DQG : VXFK WKDW IRU DQ\ YHUWLFHV X f 8 DQG LU f 93 WKH YHUWH[ Y LV RQ HYHU\ XZfSDWK 'HILQLWLRQ /HW ^9 (` EH D JUDSK )RU DQ\ YHUWH[ Y f 9 ZH GHILQH WKH HGJH VXEVHW ,JYf & (*f E\ ,J^Yf ^^YX` f (*f? X ( $JXf` :H FDOO ,JYf WKH LQFLGHQW HGJH VHW RI Y LQ &OHDUO\ ,JYf FRQWDLQV DOO WKH LQFLGHQW HGJHV RI WKH YHUWH[ Y LQ 'HILQLWLRQ /HW fÂ§ ^9(` EH D JUDSK )RU DQ\ VXEVHW ) RI ( ZH FDOO WKH VHW 9)f ^X f 9*f_ WKHUH LV DQ HGJH H ) VXFK WKDW Y DQG H DUH LQFLGHQW ` WKH ) YHUWH[ VHW 3URSRVLWLRQ >@ 7KH IROORZLQJ VWDWHPHQWV DUH HTXLYDOHQW IRU D JUDSK 7 f 7 LV D WUHH f (YHU\ WZR YHUWLFHV RI 7 DUH MRLQHG E\ D XQLTXH SDWK f 7 LV FRQQHFWHG DQG f 7 LV DF\FOLF DQG _37f_ ?(7f? 1RWH WKDW LI D WUHH KDV Q YHUWLFHV WKHQ WKHUH DUH H[DFWO\ Q fÂ§ HGJHV 'HILQLWLRQ /HW ^9 (` EH D FRQQHFWHG JUDSK DQG (n D VXEVHW RI ( $ FRQQHFWHG VXEJUDSK 7 ^9 (n` LV FDOOHG D PD[LPDO WUHH RI LI DQG RQO\ LI 7n ^9 (n 8 ^H`` FRQWDLQV D F\FOH IRU DQ\ HGJH H f ( fÂ§ (n 2XU ZRUN LQ WKLV FKDSWHU SURGXFHV D SDUWLFXODU PD[LPDO WUHH IRU DQ\ JLYHQ JUDSK 6R ZH KDYH WKH IROORZLQJ ZHOONQRZQf WKHRUHP 7KHRUHP (YHU\ FRQQHFWHG JUDSK KDV D PD[LPDO WUHH
PAGE 71
([DPSOH /HW EH D JUDSK ZLWK YHUWLFHV DQG HGJHV DV LQ WKH ILJXUH EHORZ 7KHUH DUH FXWYHUWLFHV F H 6R WKH JLYHQ JUDSK FDQ EH VHSDUDWHG LQWR EORFNV DV IROORZV *L ^9(O` ^^DE` ^ ^DE` ` ` ^9(` ^ ^ EFNOPQ ` ^ ^EF`^FN`^NO`^OP`^PQ`^QE`^EN`^NP` ` ` ^I (` ^ ^ F G H ` ^ ^F G` ^G H` ` ` ^9(` ^ ^ HIJ ` ^ ^H ` ^IJ` ^J H` ` ` ^9(` ^ ^ HKL ` ^ ^HK`^KL` ` ` DQG ^9e` ^ ^ HM ` ^ ^HM` ` `
PAGE 72
*UDSKV IRU 6SKHUH 3DFNLQJ 3ODQV 7KH VSKHUH SDFNLQJ WUHDWPHQW SODQ 6 KDYLQJ Q VSKHUHV IRU D JLYHQ EUDLQ WXn PRU IURP &KDSWHU FRQVLVWV RI WKH FHQWHUV FL & &Q DQG WKH UDGLL UM U UQ RI WKH VSKHUHV ZLWK FÂ [Â WÂ f IRU HYHU\ L ZKHUH ef LV IRU WKH $3 FRRUGLQDWH \c LV IRU WKH /DW FRRUGLQDWH DQG LV IRU WKH $[ FRRUGLQDWH RI WKH FHQWHU F 1RWH WKDW WKH SK\VLFLDQV XVH GLDPHWHUV UDWKHU WKDQ UDGLL LQ WKHLU GDWD VHWVf :H FRQVLGHU WKH VSKHUHV WR EH WKH YHUWLFHV RI D JUDSK DQG ZH DVVLJQ WKH YHUWH[ IRU WKH VSKHUH FHQWHUHG DW FÂ ZLWK UDGLXV UÂ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fÂ§ ^9 (` EH D JUDSK IRU WKH VSKHUH SDFNLQJ SODQ RI D EUDLQ WXPRU )RU DQ\ YHUWLFHV 9^ DQG 9M LQ 9 GFLFMf AcU 7 UIf LI DQG RQO\ LI e ( 7KXV ZH KDYH WKH IROORZLQJ HGJH VHW ( ^^] M`?GFL&Mf OOfU UMf ZKHUH L M L Q fÂ§ ` 6LQFH DQ\ EUDLQ WXPRU LV FRQQHFWHG LQ WKLV DUWLFOH WKHUH LV DW OHDVW RQH X Xf SDWK LQ IRU HYHU\ WZR YHUWLFHV X DQG Y LQ 7KXV ZH KDYH WKH IROORZLQJ SURSRn VLWLRQ 3URSRVLWLRQ 7KH JUDSK ^9 (` SURGXFHG E\ WKH DERYH PHWKRG IRU D JLYHQ WXPRU LV D FRQQHFWHG JUDSK 6LQFH ZH DUH GHDOLQJ ZLWK XQODEHOHG JUDSKV LQ WKLV DUWLFOH ZH KDYH WKH IROORZLQJ SURSRVLWLRQ
PAGE 73
3URSRVLWLRQ )RU DQ\ VSKHUH SDFNLQJ WUHDWPHQW SODQ IRU D JLYHQ EUDLQ WXPRU WKHUH H[LVWV D XQLTXH JUDSK WKDW ZH DVVLJQ WR WKH WUHDWPHQW SODQ %\ PHDQV RI WKH DERYH PHWKRG ZH JHW D XQLTXH FRQQHFWHG JUDSK ^9 (`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f IRU D JLYHQ EUDLQ WXPRU )LJXUH 7KHQ E\ 3URSRVLWLRQ WKHUH LV D XQLTXH JUDSK IRU WKH VSKHUH SDFNLQJ SODQ 7KH JUDSK LV JLYHQ EHORZ LQ )LJXUH 3 )LJXUH
PAGE 74
7KHUH DUH WZR GLIIHUHQW LVRPRUSKLVP FODVVHV QDPHO\ 7L DQG 7 RI PD[LPDO WUHHV IRU WKH JUDSK 7KDW LV ZH FRXOG FKRRVH HLWKHU RI WKHVH IRU D PD[LPDO WUHH WR DVVLJQ WKH JUDSK 6LQFH LQ RXU DOJRULWKP ZH ZDQW WR JHW WKH VDPH PD[LPDO WUHH IRU D JLYHQ JUDSK HYHU\ WLPH ZH QHHG FHUWDLQ UXOHV WR JHW WKH VDPH WUHH 6HH )LJXUH f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
PAGE 75
DUH GLIIHUHQW NLQGV RI fFXWYHUWLFHVf LQ VRPH JUDSKV 7KH IROORZLQJ H[DPSOH VKRZV VXFK D FDVH ([DPSOH /HW fÂ§ ^9 (` EH WKH JUDSK LQ )LJXUH X X 8 7KHQ WKHUH DUH WZR FXWYHUWLFHV RI WKH JUDSK QDPHO\ 8 DQG 8LR 8Vn LQJ WKHVH WZR FXWYHUWLFHV WKH JUDSK LV VHSDUDWHG LQWR IRXU VXEJUDSKV QDPHO\ *? * DQG 6HH )LJXUH f 7KHQ WKH VXEJUDSKV *L DQG DUH EORFNV EXW WKH VXEJUDSK FRQn WDLQV LWV RZQ FXWYHUWLFHV DQG X ZKLFK DUH QRW FXWYHUWLFHV RI WKH JUDSK :H VHSDUDWH WKH VXEJUDSK LQWR IRXU VXEJUDSKV E\ XVLQJ LWV FXWYHUWLFHV X DQG X}
PAGE 76
6HH )LJXUH f 1RWH WKDW WKHUH LV QR VSHFLILF RUGHU LQ ODEHOLQJ WKH VXEJUDSKV 6R DW WKLV PRPHQW ZH UHODEHO WKH VXEJUDSKV RI E\ +? + +U X )LJXUH 7KHQ DJDLQ WKH VXEJUDSKV +L + +H DUH EORFNV EXW WKH VXEJUDSK +U FRQWDLQV LWV RZQ FXWYHUWH[ XH ZKLFK LV QHLWKHU D FXWYHUWH[ RI WKH JUDSK QRU D FXW YHUWH[ RI WKH VXEJUDSK :H VHSDUDWH WKLV VXEJUDSK +U LQWR WZR VXEJUDSKV E\ XVLQJ XH 6HH )LJXUH f $QG ZH UHODEHO WKH VXEJUDSKV RI E\ %? % %J 7KHUHIRUH ^mA` LV WKH VHW RI FXWYHUWLFHV RI WKH JUDSK %XW ^XX` LV WKH VHW RI WKH FXWYHUWLFHV RI D VXEJUDSK ZKLFK LV SURGXFHG DIWHU WKH ILUVW VHSDUDWLRQ XVLQJ WKH FXWYHUWLFHV RI WKH JUDSK DQG XJ LV WKH FXWYHUWH[ RI D VXEJUDSK ZKLFK LV SURGXFHG DIWHU WKH VHFRQG VHSDUDWLRQ )RU WKH SXUSRVH RI ODEHOLQJ WKH YHUWLFHV LQ D JUDSK ZH VHSDUDWH WKHVH FXWYHUWH[ VHWV IRU GLIIHUHQW OHYHOV IURP HDFK RWKHU 'HILQLWLRQ /HW EH WKH JUDSK DVVLJQHG IRU D JLYHQ VSKHUH SDFNLQJ SODQ 3 7KHQ WKH YHUWH[ VHW 9*f RI WKH JUDSK LV WKH XQLRQ RI WZR GLVMRLQW VXEVHWV QDPHO\ &L*f DQG &>*f ZKHUH &L*f ^Y e 9*f? Y LV D FXWYHUWH[ RI *`
PAGE 77
DQG &>*f 9*f fÂ§ &?*f :H FDOO &?*f WKH ILUVW VWHS FXWYHUWH[ VHW RI )RU HYHU\ L _9*f_ ZH GHILQH D VXEVHW &IL*f RI &IBIL*f DV WKH FROOHFWLRQ RI WKH YHUWLFHV RI &IB*f ZKLFK DUH FXW YHUWLFHV RI D VXEJUDSK SURGXFHG DIWHU WKH L fÂ§ OfWK VHSDUDWLRQ XVLQJ WKH HOHPHQWV RI &LL*f 7KHQ ZH FDOO &IL*f WKH LWK VWHS FXWYHUWH[ VHW RI DQG OHW &*f &AB;^*f fÂ§ &IL*f :H DVVXPH WKDW WKHUH DUH ILQLWHO\ PDQ\ YHUWLFHV LQ D JLYHQ JUDSK ,I WKHUH H[LVWV DW OHDVW RQH FXWYHUWH[ WKHQ WKHUH H[LVWV DQ LQWHJHU N Q VXFK WKDW &IL*f A IRU HYHU\ L N DQG &L*f IRU HYHU\ L N 7KHQ ZH FDOO N WKH VHSDUDWLRQ VWHS FRQVWDQW RI WKH JUDSK DQG OHW &&*f &NfF*f ,I WKHUH LV QR FXWYHUWH[ RI WKHQ WKH JUDSK LV D EORFN DQG ZH VD\ WKDW WKH VHSDUDWLRQ VWHS FRQVWDQW RI LV X )LJXUH 3URSRVLWLRQ /HW _e"f_ Q DQG OHW WKH VHSDUDWLRQ VWHS FRQVWDQW RI EH N 7KHQ WKHUH LV D VHTXHQFH RI LQWHJHUV ^PL P Pr` ZLWK P[ P PN VXFK WKDW _&nL*f_ PX DQG ?&IL*f? PL PBL IRU L N
PAGE 78
3URRI 6LQFH N WKH ILUVW VWHS FXWYHUWH[ VHW &L*f LV QRW HPSW\ 7KXV WKHUH LV D SRVLWLYH LQWHJHU QDPHO\ PL VXFK WKDW _*L*f_ P? )RU HYHU\ L $ OHW P PBL _&*f_ )RU HYHU\ L N WKH LWK VWHS FXWYHUWH[ VHW *Â*f LV QRW HPSW\ 6R ZH KDYH WKDW P PÂBL DQG _&Â*f_ PÂ fÂ§ PBL 1RZ ZH ODEHO WKH YHUWLFHV DV IROORZV f WKH YHUWLFHV LQ &?^*f DUH ODEHOHG E\ ^ PL` f IRU HYHU\ L N WKH YHUWLFHV LQ **f DUH ODEHOHG E\ ^PBL PÂBL PÂ fÂ§ PÂ` DQG f WKH YHUWLFHV LQ &&*f DUH ODEHOHG E\ ^Pr PW Qf :H DUELWUDULO\ QDPH WKH YHUWLFHV RI **f DQG &&*f DQG GHQRWH WKDW *L*f `XL8` &L*f fÂ§ ^XPLBOABL :M7LLfÂ§L f f f ` IRU L N DQG *&*f ^XPIFL XPMW XQ` ,W LV FOHDU WKDW IRU HYHU\ YHUWH[ XS f **f DQG HYHU\ YHUWH[ XT f *M*f ZLWK L A M ZH KDYH WKDW S T LI DQG RQO\ LI Â M $QG DOVR IRU HYHU\ YHUWH[ XY f *Â*f IRU HYHU\ L ON DQG HYHU\ YHUWH[ XT f &&*f ZH KDYH WKDW S T $VVXPH WKDW IRU HYHU\ YHUWH[ 8^ WKH FHQWHU DQG WKH UDGLXV RI WKH VSKHUH ZKLFK LV DVVLJQHG WR WKDW YHUWH[ X DUH FÂ ^[Â \ ]Âf DQG UL UHVSHFWLYHO\ :H QRZ UHODEHO WKH HOHPHQWV IRU HDFK RI WKH VHWV *L*f **f &N*f DQG &&*f LQ WKH IROORZLQJ ZD\ IRU HYHU\ SDLU RI YHUWLFHV X 8M LQ RQH RI *L*f **f *IF*f DQG &&*f L M LI DQG RQO\ LI Lf GXLf GXMf RU LLf ,I GXLf GXMf WKHQ UL UM RU LLf ,I GXLf GXMf DQG UÂ UÂ WKHQ [Â ;M RU LLLf ,I GXLf G^XMf U^ UM DQG [Â [M WKHQ \\ RU LYf ,I GXLf GXMf UL UM [ ;M DQG \^ \M WKHQ =M 1RWH WKDW GXf UHSUHVHQWV GHJUHH RI WKH YHUWH[ LQ *
PAGE 79
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f EH WKH JUDSK IRU D JLYHQ VSKHUH SDFNLQJ SODQ /HW *f ^XL X XQ` IRU VRPH SRVLWLYH LQWHJHU Q ZKHUH DOO WKH YHUWLFHV LQ 9 DUH ODEHOHG E\ WKH PHWKRG LQ WKH SUHYLRXV VHFWLRQ 'HILQH D VXEJUDSK 7 fÂ§ *f )f ZLWK )f 9 *f DQG ) WKDW LV WKHUH LV QR HGJH LQ 7 7KHQ IRU HYHU\ L Q ZH KDYH WKDW 3WXf L 6HH 'HILQLWLRQ f f )RU WKH YHUWH[ LT OHW ) 8 ,JXLf 6HH 'HILQLWLRQ f /HW *f )Lf 7KH VXEJUDSK ,)f)f & 7? LV D WUHH 7KXV LI 9)Lf *f WKHQ 7? fÂ§ *f )Lf )Lf )?f LV RXU PD[LPDO WUHH RI 1RWH WKDW IRU HYHU\ YHUWH[ Y f )Mf ZH KDYH WKDW 3WAYf fÂ§ DQG IRU HYHU\ YHUWH[ 8M I)Lf ZH KDYH WKDW 37 8Mf M f $VVXPH WKDW )Lf A*f )RU WKH YHUWH[ 8 WKHUH DUH WZR FDVHV WKDW LV )WLXf DQG 3WcXf fÂ§ 1RZ GHILQH WKH HGJH VXEVHW ,n*Xf E\ ÂXf ,J0 ^^XX` H (? )7OXf ` Df ,I 3W[Xf WKDW LV WKHUH LV D XLXfSDWK LQ WKH VXEJUDSK L WKHQ OHW ) )M8ÂXf (DFK HGJH LQ WKH VHW ^^X X` f (? 37O Yf ` ^^X Yf (_ X )Lf` PDNHV D F\FOH ZLWK ^XXX` DQG ^LmLX` VLQFH 37Xf 7KXV WKHUH LV QR F\FOH LQ WKH VXEJUDSK )f)f VR LW LV D WUHH /HW 7 *f )f & *
PAGE 80
,I 9)f 9*f WKHQ 7 ^9*f)f 9)f)f LV RXU PD[LPDO WUHH RI 1RWH WKDW IRU HYHU\ YHUWH[ Y e 9)f ZH KDYH WKDW 3W$ff DQG IRU HYHU\ YHUWH[ XM 9)f ZH KDYH WKDW 3M XMf fÂ§ M Ef ,I 3Wc^Xf WKHQ WKHUH LV QR LWLXfSDWK LQ WKH VXEJUDSK 7? Lf ,I WKHUH LV D QRQHPSW\ VXEVHW ^XLXL XW` & 9,JXff ZKHUH 3WLXf IRU L W DQG M IRU L M W WKHQ OHW ) )? 8 3*Xf 8 ^^XX`` 7KHQ WKH VXEJUDSK 9)f)f & LV D WUHH /HW 7 A*fAf & ,I 9)f 9*f WKHQ 7 A*fAf 9)f )f LV RXU PD[LPDO WUHH RI 1RWH WKDW IRU HYHU\ YHUWH[ Y 9)f ZH KDYH WKDW 3W Yf DQG IRU HYHU\ YHUWH[ 8M e 9)f ZH KDYH WKDW 3W0 LLf ,I QRQH RI WKH YHUWLFHV LQ 9^,J^Xff KDV SYDOXH WKDW LV IRU HYHU\ YHUWH[ Y e 9,JXff WKHUH LV QR XLQfSDWK WKHQ OHW ) )? 8 ,*Xf 6LQFH 3W[^Zf IRU HYHU\ YHUWH[ Z e )? DQG 3W[Yf IRU HYHU\ YHUWH[ Y e ,JXf WKH VXEJUDSK 9)f)f LV D IRUHVW EXW QRW D WUHH /HW 7 9*f)f & ,W LV FOHDU WKDW WKH VXEJUDSK 7 FDQQRW EH RXU PD[LPDO WUHH HYHQ LI ZH KDYH WKDW 9)f fÂ§ 9*f EHFDXVH WKH VXEJUDSK 7 LV GLVFRQQHFWHG 1RWH WKDW IRU HYHU\ YHUWH[ Y e 9)f 9)L 8 JZff 9^)Lf 8 9,J^Xff ZH KDYH WKDW 3WYf LI Y e 9)?f DQG 3W^Yf LI Y e 9,JXff $QG IRU HYHU\ YHUWH[ 8M 9)f ZH KDYH WKDW SW$XMf f %\ LQGXFWLRQ DW WKH MWK VWHS WKH VXEJUDSK 7MBL 9*f)MLf KDV EHHQ GHILQHG :H ZDQW WR GHILQH WKH VXEJUDSK 7M & )RU WKH YHUWH[ LÂ WKHUH DUH WZR FDVHV WKDW LV 3UMBLXMf M DQG 3WB$XMf T IRU VRPH T M )RU HYHU\ L Q GHILQH UQLQ^O e ^ Q`?37@B$XLf r DQG XW e $*^XMf ` ,I WKH VHW ^ e ^ Q`_LLM e 9O*XMff DQG 3WALXLf ] ` LV HPSW\ WKHQ WKH YHUWH[ XPM OA LV QRW GHILQHG 6R WKHUH LV QR HGJH ^XM XPMfÂ§ rf`
PAGE 81
Df /HW 3U!BLmMf M WKDW LV IRU HYHU\ L M fÂ§ OM Q WKHUH LV QR LLM XfSDWK LQ WKH VXEJUDSK 7ML 9*f)Mf /HW )M )ML 8 8W ^ ^XM `f 7KHQ WKH VXEJUDSK 9)Mf)Mf LV D IRUHVW /HW 7M 9*f)Mf & ,I 9)Mf 9*f DQG _9)Âf_ ?)M? WKHQ E\ 3URSRVLWLRQ WKH VXEJUDSK 7M 9*f )Mf 9)Mf )Mf LV RXU PD[LPDO WUHH RI Ef /HW 3WM XMf T IRU VRPH T M WKDW LV WKHUH LV D X WÂfSDWK LQ WKH VXEJUDSK 7MBM 9*f )ÂBLf 1RZ OHW )L )ML 8 X @ ^XM `f 7KHQ WKH VXEJUDSK 9)Mf )Mf LV D IRUHVW /HW 7M 9*f)Mf & ,I 9)Mf 9^*f DQG ?)M? WKHQ E\ 3URSRVLWLRQ WKH VXEJUDSK 7M 9*f )Mf 9^)Mf )Mf LV RXU PD[LPDO WUHH RI 1RWH WKDW LI X LV DQ ÂWK VWHS FXWYHUWH[ WKHQ GXf LQ LV VDPH DV GXf LQ 7LBM $OVR QRWH WKDW IRU HYHU\ L Q ^XL WÂ LW` & 9A)f WKDW LV 9*f 9)Qf :H ZLOO VHH LQ 7KHRUHP WKDW 7Q 9*f)Qf 9)Qf )Qf LV RXU PD[LPDO WUHH RI %\ WKH 7UHH &ODVVLILFDWLRQ 7KHRUHP 7KHRUHP f WKLV WUHH 7Q FDQ EH GHQRWHG E\ D XQLTXH SHUIHFW VHTXHQFH ([DPSOH /HW 9 (f EH D JUDSK ZKHUH 9 ^mL LL" f f f LWV` ZKLFK DUH RUGHUHG DV LQ WKH SUHYLRXV VXEVHFWLRQ DQG ( ^HL H H` DV LQ WKH ILJXUH EHORZ &OHDUO\ WKH JUDSK GRHV QRW KDYH D FXWYHUWH[ VR WKH RUGHU LV GHFLGHG E\ RQO\ WKH VL]HV RI WKH VSKHUHV DQG WKH FRRUGLQDWHV RI WKH FHQWHUV RI WKH VSKHUHV 8V HL ,Â H 8M X H X )LJXUH 8
PAGE 82
/HW 7R 9*f)Rf EH D VXEJUDSK RI 9 (f ZKHUH )R DV VKRZQ LQ )LJXUH X X 8 X X f f X 8 )LJXUH 6LQFH ,*^XLf ^HLH` OHW )[ )8JmLf ^HLH` 7KHQ 9)Lf ^XLXX` 6LQFH 9)Lf ^XLXX` A 9^*f WKH WUHH 9)Lf )Lf LV QRW D PD[LPDO WUHH RI /HW 7? 9*f)Lf ZKLFK LV VKRZQ LQ )LJXUH f f 8 8t )LJXUH :H KDYH WKDW PLÂf L IRU L DQG EXW PLf DQG PLf GR QRW H[LVW 6LQFH 3UXf DQG $J^Xf ^m X X` OHW ) )L 8 8 LL^^X X``f )L 8 ,J^Xf ^HLL H` 8 ^H H HV` ^HL H H H HV` 7KHQ 9)f ^XOXX XXXX` A 9*f DQG WKH VXEJUDSK 9)f)f LVD IRUHVW QRW D WUHH 7KXV LW FDQQRW EH D PD[LPDO WUHH RI /HW 7 9*f )f 6HH )LJXUH f
PAGE 83
X H X )LJXUH 1RWH WKDW 3W^XLf 3WXf 3WXf 3W^Xf 3W^Xf 3WXf 3W^Xf fÂ§ DQG 3WXVf 7KXV ZH KDYH WKDW P f OPf Pf 1RWH WKDW PÂf GRHV QRW H[LVW IRU Â 6LQFH 8L $Xf DQG 3WXffÂ§ ZHJHW ) )8^^XX`` ^HL H H H H H` 6LQFH WKH VXEJUDSK 9)f)f LV D IRUHVW EXW QRW D WUHH LW LV QRW D PD[LPDO WUHH RI 'HILQH 7 9*f)f 6HH )LJXUH f X H X )LJXUH 1RWH WKDW 3WXLf 3WXf 3WXf DQG 3WXf 3WXf 3WXf 3W^XTf 3WXVf fÂ§ 7KXV ZH KDYH WKDW POf DQG Pf 1RWH WKDW PLf GRHV QRW H[LVW IRU L 6LQFH 3WXf ZH GHILQH ) ) 8 ^^XX`` ^HHHHHHH` 6LQFH 9)f ^XM Y X` 9*f DQG _9)f_ _)_ WKH VXEJUDSK 7 9)f )f LV D PD[LPDO WUHH RI 6HH )LJXUH f
PAGE 84
X 8U H 0L )LJXUH $W WKLV PRPHQW ZH KDYH WKDW 3W^X?f 3WPf 3W^Xf 7KHQ IRU HYHU\ XW f 9*f ZLWK W ZH KDYH WKDW )W ) 7KXV IRU HYHU\ W WKH VXEJUDSK 9)Wf)Wf LV VDPH DV 7 9)f )f 7KH QH[W H[DPSOH VKRZV KRZ ZH JHW WKH PD[LPDO WUHH IRU D JLYHQ JUDSK FRQWDLQLQJ VRPH FXWYHUWLFHV ([DPSOH /HW EH WKH JUDSK LQ ([DPSOH 6HH )LJXUH f 7KHQ &L*f ^FH` DQG &^^*f fÂ§ ^DGIJKLMNOPQ` ,Q &L*f ZH KDYH WKDW GHf GEf DQG GFf 2Q WKH RWKHU KDQG LQ &nL*f ZH KDYH WKDW GNf GPf GQf GGf GIf GJf GKf GOf GDf GLf GMf fÂ§ $QG & 7KXV ZH RUGHU WKH YHUWLFHV DV IROORZV Y? H P X F DQG Y $ ^XX` ^PQ` ^X X Y P 0X` ^G IJ K ` ^P0L P` ^D Â M` $VVXPH WKDW X P P Q X G Y X I X K XQ O YX D PL ] X M ZKLFK DUH GHFLGHG E\ WKH VL]HV DQG FHQWHUV RI WKH FRUUHVSRQGLQJ VSKHUHV 6HH )LJXUH f )LJXUH
PAGE 85
6LQFH ,*YLf ^ ` OHW )L ,*YLf 7KHQ 9)Lf ^L; Y Y Y Y Yf 6LQFH 9)Wf A 9*f WKH WUHH 7L 9*f)Lf LV QRW D PD[LPDO WUHH RI 7KXV IRU WKH FXW YHUWH[ L!L ZH KDYH WKH IROORZLQJ VXEn JUDSK ZKLFK LV 7? O&Uf)Lf 6HH )LJXUH f W9 mXQ Â9 f 9 9 9 9 )LJXUH 6LQFH )7OXf DQG IRU HYHU\ YHUWH[ Y H $*Yf ^YYYYL` 37OYf s OHW ) )L 8 ,*Yf 7KH VXEJUDSK 9)f )f LV D IRUHVW EXW QRW D WUHH 7KXV LW FDQQRW EH D PD[LPDO WUHH RI 1RZ IRU WKH FXWYHUWH[ Z ZH KDYH WKH IROORZLQJ VXEJUDSK ZKLFK LV 7 *f)f 6HH )LJXUH f r9 f YQ WY} 6LQFH $*Yf ^XXX` 3W^Yf 3UW!f 3WYf DQG 3W^Yf OHW ) ) 8 ^^XZ`` 7KHQ IRU WKH FXWYHUWH[ X ZH KDYH WKH IROORZLQJ VXEJUDSK ZKLFK LV )f)f 7KHQ WKH VXEJUDSK 9)f)f LV D WUHH %XW 9)f IL 9*f
PAGE 86
7KXV LW FDQQRW EH D PD[LPDO WUHH RI :H KDYH WKH IROORZLQJ VXEJUDSK ZKLFK LV 7 9*f)f 6HH )LJXUH f 9f f 98 Xr 6LQFH $*^\f ^X X X X W!Q` DQG 3W^Yf 3WYf 3WYf 3WYf 3W^Yf 37LYQf ZH GHILQH ) ) 8 ^^X8V`` 8 ^^AA`` 7KHQ WKH VXEJUDSK 9)f )f LV D WUHH %XW 9^)f A 9*f 7KXV LW FDQQRW EH D PD[LPDO WUHH RI 7KHQ IRU WKH FXWYHUWH[ Y ZH KDYH WKDW WKH IROORZLQJ VXEJUDSK ZKLFK LV 7 9*f )f 6HH )LJXUH f )RU WKH YHUWLFHV YYYYY WKHUH LV QR fQHZf HGJH 7KXV ZH JHW WKH VDPH VXEn JUDSK DV DERYH WKDW LV 7 7 7 7 fÂ§ 7 fÂ§ 7 %XW IRU WKH FXWYHUn WH[ 8 ZH KDYH WKDW 8 e 9,JY ff DQG 3W^Y f A 3W^YLRf 7KXV
PAGE 87
) ) 8^ ^AA ` DQG 9):f 9*f 7KHUHIRUH ZH KDYH WKH IROORZLQJ VXEn JUDSK ZKLFK LV 7 9*f)LRf 9)LRf )LRf 6HH )LJXUH f &OHDUO\ WKH VXEJUDSK 7 fÂ§ ))LRf )LRf GRHV QRW LQFOXGH D F\FOH 7KXV LW LV D PD[n LPDO WUHH RI 7KDW LV IRU HYHU\ W fÂ§ ZH KDYH WKDW ))Wf)Wf 9^)?Rf )f 7KHUHIRUH WKH VXEJUDSK 7 9)Lf )Lf 9n*f )f LV WKH PD[n LPDO WUHH RI WKH JUDSK )RU WKLV UHVXOWLQJ WUHH 7 ))LRf )LRf FKRRVH D YHUWH[ VHTXHQFH V E\ V YLYY99Y9Q9YLYLRYL9YYOf 7KHQ WKH VHTXHQFH 3 f LV WKH GHJUHH VHTXHQFH RI V %\ 7KHRUHP 3 VDWLVILHV '6& ,W LV FOHDU WKDW 3 DOVR VDWLVILHV '6E& %\ &RUROODU\ 3 LV WKH SHUIHFW VHTXHQFH IRU WKH WUHH 7 ^9)LRf )f 7KHRUHP /HW fÂ§ ^9 )` EH D JUDSK IRU D VSKHUH SDFNLQJ SODQ ZKHUH 9 ^XM m XQ` ZLWK WKH RUGHU GHVFULEHG LQ 6HFWLRQ II /HW WKH VXEJUDSK 9)Qf)Qf RI FRPH IURP WKH DERYH SURFHGXUH 7KHQ 9)Qf 9*f 7KHUHn IRUH 9)Qf )ff LV D PD[LPDO WUHH RI DQG WKHUH LV D XQLTXH SHUIHFW QVHTXHQFH IRU WKH WUHH 9)Qf)Qf 3URRI 7KH VXEJUDSK 9)Qf )Qf LV FRQQHFWHG ,Q IDFW DVVXPH WKDW WKH VXEn JUDSK 9)Qf)Qf FRQWDLQV WZR GLVMRLQW VXEWUHHV QDPHO\ L DQG 7 7KHQ WKH
PAGE 88
SDWKYDOXH IRU HYHU\ YHUWH[ LQ 7? LV GLIIHUHQW IURP WKH SDWKYDOXH IRU HYHU\ YHUn WH[ LQ 7 6LQFH WKH JUDSK LV FRQQHFWHG WKHUH LV DW OHDVW RQH HGJH H f (*f ZKLFK FRQQHFWV WKH WZR VXEJUDSKV 7? DQG 7 DQG H A (7? 8 7f )Q /HW ^ ^XMLXrL`^XM8MW` EH WKH FROOHFWLRQ RI VXFK HGJHV ZKHUH YMW f 7L DQG 9N f 7 IRU L :LWKRXW ORVV RI JHQHUDOLW\ OHW ML ML IRU HYHU\ L T 7KHQ LW LV FOHDU WKDW ^XM ` f )M E\ WKH SURFHGXUH RI JHWWLQJ WKH VXEJUDSK 9)Qf )Qf 7KHQ )Q 7KLV LV D FRQWUDGLFWLRQ 7KHUHIRUH WKH VXEJUDSK 9)Qf )Qf LV FRQQHFWHG 7KH SDWKYDOXH LQ WKH SURFHGXUH RI JHWWLQJ WKH VXEJUDSK 9)Qf)Qf JXDUDQWHHV WKDW WKH VXEJUDSK 9)Qf)Qf LV DF\FOLF 7KHUHn IRUH 9)Qf )Qf LV D WUHH ,W LV FOHDU WKDW IRU HYHU\ L Q WKH YHUWH[ VHW 9)Lf FRQWDLQV WKH YHUWLFHV X? X XÂ 7KXV 9)Qf FRQWDLQV DOO WKH YHUWLFHV RI VLQFH 9)Qf ^XXX XQ` 9^*f 7KDW LV 9)Qf 9 9*f +HQFH WKH VXEJUDSK 9)Qf )Qf LV D PD[LPDO WUHH RI %\ WKH 7UHH &ODVVLILFDWLRQ 7KHn RUHP 7KHRUHP f LW LV FOHDU WKDW WKHUH LV D SHUIHFW QVHTXHQFH IRU WKH WUHH n9)Qf)Qf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
PAGE 89
VSKHUH SDFNLQJ SODQV DUH QRW LVRPRUSKLF 7KXV ZH PD\ FRQVLGHU WKHLU VKDSHV WR EH GLVWLQFW 7KHUHIRUH ZH KDYH WKH IROORZLQJ DV RXU PDLQ WKHRUHP RI WKLV VHFWLRQ 7KHRUHP 7KH SHUIHFW VHTXHQFHV DUH LQYDULDQWV RI WKH VKDSHV RI DUELWUDU\ EUDLQ WXPRUV *UDSK IURP D 0D[LPDO 7UHH /HW 7 ^9 (n` EH D PD[LPDO WUHH IRU D JUDSK ^9 (` ZKLFK LV REWDLQHG IURP WKH VSKHUH SDFNLQJ SODQ RI D JLYHQ WXPRU RU VROLG VKDSH :H REVHUYH WKDW ZH FDQ UHFRYHU WKH RULJLQDO JUDSK IURP 7 6LQFH 9^7f fÂ§ 9 9*f DQG (7f fÂ§ (n & ( (*f ZH QHHG WR DGG WKH UHPDLQLQJ HGJHV WR WKH HGJH VHW RI WKH PD[LPDO WUHH WR JHW WKH JUDSK 6LQFH WKH JUDSK FRPHV IURP WKH VSKHUH SDFNLQJ WUHDWPHQW SODQ HDFK YHUWH[ Y LQ GHQRWHV D VSKHUH RI WKH VSKHUH SDFNLQJ SODQ 7KXV IRU HYHU\ YHUWH[ f OHW F D \ ]If DQG UÂ EH WKH FHQWHU DQG WKH UDGLXV RI WKH VSKHUH %\ 3URSRVLWLRQ IRU DQ\ WZR YHUWLFHV W! DQG 9M LQ 9 LI GF&Mf MAUf UÂf WKHQ WKH HGJH LV LQ WKH HGJH VHW )*f RI WKH JUDSK 7KXV (n 8 ^^XÂL!M` f 9 [ 9? GFL&Mf ^ÂU Uf` ( 7KHUHIRUH ZH JHW WKH JUDSK ^9 (` $OJRULWKP IRU 2XU 0D[LPDO 7UHH /HW 9 (f EH WKH JUDSK IRU D VSKHUH SDFNLQJ SODQ ZLWK _9_ Q 67(3 $ )LQG DOO WKH ÂWK VWHS FXWYHUWLFHV RI WKH JUDSK IRU L Q 67(3 % /DEHO WKH YHUWLFHV LQ WKH YHUWH[ VHW 9 E\ WKH RUGHU H[SODLQHG LQ 6HFWLRQ 67(3 & %\ LQGXFWLRQ IRU HDFK L Q JHW WKH VXEJUDSK 3)f )Âf RI 7KHQ 7 9)Qf )Qf LV RXU PD[LPDO WUHH
PAGE 90
$33(1',; $ 6+(//,1* 352*5$0 )25 63+(5( 3$&.,1* ,Q WKLV DSSHQGL[ ZH SURYLGH WKH & SURJUDP ZH GHYHORSHG WR JHW WKH VSKHUH SDFNLQJ SODQ IRU D JLYHQ WXPRU VKDSH 7KH ILUVW VHFWLRQ VKRZV WKH SURJUDP WUDQVODWLQJ WKH GDWD IURP &7 VFDQ WR WKH GDWD RI DSSURSULDWH IRUP IRU WKH PDLQ SURJUDP ZKLFK LV LQ WKH VHFRQG VHFWLRQ 1RWH WKDW WKLV SURJUDP GRHV QRW LQFOXGH WKH VFRUH IXQFWLRQ ZKLFK ZDV GHYHORSHG E\ 7KRPDV :DJQHU :H UHIHU WKH UHDGHU $SSHQGL[ ( IRU WKH FRPSOHWH DUWLFOH RQ WKH VSKHUH SDFNLQJ WUHDWPHQW SODQ $O 3URJUDP IRU 'DWD 5HDGLQJ UHZULWH GDWD ILOH LQWR [\] IRUPDW t SURGXFH WKH WRWDO RI GDWD [\] PLQ t PD[ LQFOXGH LRVWUHDPK! LQFOXGH IVWUHDPK! LQFOXGH LRPDQLSK! LQFOXGH VWGOLEK! LQFOXGH PDWKK! PDLQ2 ^ FRQVW LQW QRIGDWD FRPHV IURP GDWDW[W LQW [SRVLWLRQ>QRIGDWD@ \SRVLWLRQIQRIGDWD@ ]SRVLWLRQ>QRIGDWD@ LQW T IORDW D IORDW E IORDW F LQW N ^ LIVWUHDP LQ&OLHQW)LOHGDWDW[W ZKLOH LQ&OLHQW)LOH !! D !! E !! Ff ^ [SRVLWLRQ >N@ LQWIORRUDff \SRVLWLRQIN@ LQWIORRUEff ]SRVLWLRQIN@ LQWIORRUFff N LRVLQf FKHFN WKH QXPEHU FKHFN WKH QXPEHU FKHFN WKH QXPEHU LQW [PLQ [SRVLWLRQ>@ LQW [PD[ [SRVLWLRQ>@ LQW \PLQ \SRVLWLRQ>@ LQW \PD[ \SRVLWLRQ>@ LQW ]PLQ ]SRVLWLRQ>@ LQW ]PD[ ]SRVLWLRQ>@ IRU LQW L LQRIGDWD Lf ^ LI [SRVLWLRQ>L@ [PLQ f [PLQ [SRVLWLRQ>L@
PAGE 91
! ^ HOVH LI [SRVLWLRQ>L@ [PD[ [SRVLWLRQ>L@ HOVH LI \SRVLWLRQ>L@ \PLQ \SRVLWLRQ>L@ HOVH LI \SRVLWLRQ>L@ \PD[ \SRVLWLRQ>L@ HOVH LI ]SRVLWLRQ>L@ ]PLQ ]SRVLWLRQ>L@ HOVH LI ]SRVLWLRQ>L@ ]PD[ ]SRVLWLRQ>L@ [PD[ f \PLQ f \PD[ f ]PLQ f ]PD[ f RXWSXW WKH UHVXOW RI FHQWHUV RIVWUHDP RXW&OLHQW)LOHP0W[W LRVDSSf ^ RXW&OLHQW)LOH QRIGDWD VHWZf [PLQ VHWZf [PD[ VHWZf \PLQ VHWZf \PD[ VHWZf ]PLQ VHWZf ]PD[ HQGO FRXW QRIGDWD VHWZf [PLQ VHWZf [PD[ VHWZf \PLQ VHWZf \PD[ VHWZf ]PLQ VHWZf ]PD[ FLQ !! T $ 3URJUDP IRU &HQWHU 6HDUFKLQJ VHDUFKLQJ WKH FHQWHUV IRU SDW LQFOXGH LRVWUHDPK! LQFOXGH IVWUHDPK! LQFOXGH LRPDQLSK! mLQFOXGH VWGOLEK! LQFOXGH PDWKK! LQW VTXDUHLQWf IXQFWLRQ SURWRW\SH LQW 0$;LQWLQWLQWf IXQFWLRQ SURWRW\SH PDLQf ^ FRQVW LQW QRIGDWD FKHFN WKH QXPEHU IURP P0W[W FRQVW LQW [PLQ FKHFN WKH QXPEHU FRQVW LQW [PD[ FKHFN WKH QXPEHU FRQVW LQW \PLQ FKHFN WKH QXPEHU FRQVW LQW \PD[ FKHFN WKH QXPEHU FRQVW LQW ]PLQ FKHFN WKH QXPEHU FRQVW LQW ]PD[ FKHFN WKH QXPEHU FRQVW LQW [I L[ [PLQ FRQVW LQW \IL[ \PLQ FRQVW LQW ]I L[ ]PLQ LQW [SRVLWLRQ>QRIGDWD@ \SRVLWLRQ >QRIGDWD@ ]SRVLWLRQ >QRIGDWD@
PAGE 92
LQW T IORDW D IORDW E IORDW F LQW N ^ LIVWUHDP LQ&OLHQW)LOHGDWDW[W LRVLQf ZKLOH LQ&OLHQW)LOH !! D !! E !! Ff ^ [SRVLWLRQ>N@ LQWIORRUDff[IL[ \SRVLWLRQ>N@ LQWIORRUEff\IL[ ]SRVLWLRQ>N@ LQWIORRUFff]IL[ N ! FRQVW LQW [ [PD[ [PLQ FRQVW LQW \ \PD[ \PLQ FRQVW LQW ] ]PD[ ]PLQ FRQVW LQW QF [r\r] QXPEHU RI FHOOV LQW FF>QF@ VWDWXV YDOXH IRU LQW L LQF Lf FF>L@ LQW [F>QF@ LQW \F >QF@ LQW ]F>QF@ [FRRUG RI F>QF@ \FRRUG RI F>QF@ ]FRRUG RI F>QF@ VHWWLQJ WKH [\]FRRUG IRU HDFK FHOO IRU LQW L LQF Lf ^ ]F>L@ IORRUL\r[ff [F>L@ IORRUL]F>L@r\r[f[f \F>L@ L]F >L@ r\r[[F>L@ r[ VHWWLQJ WKH ERXQGDU\ RI WXPRU E\ IRU LQW L LQRIGDWD Lf ^ FF>]SRVLWLRQ>L@r[r\ [SRVLWLRQ>L@r\ \SRVLWLRQ>L@@ VHWWLQJ WKH LQLWLDO RI QRUPDO E\ FF>@ VHWWLQJ WKH QRUPDO FHOO E\ LQW K IRU LQW M M[r\ Mf ^ IRU LQW L LQF Lf ^ LI FF>L@ f ^ LI FF>L@ __FF>LO@ ,_FF>L\@ __FF>L\@ f ^ FF>L@ K HOVH HOVH !
PAGE 93
! LI K f EUHDN HOVH K VHWWLQJ WKH UHPDLQLQJ WXPRU FHOO E\ IRU LQW L LQF Lf ^ LI FF>L@ f FF>L@ HOVH ` FKHFNLQJ FKHFNLQJ FKHFNLQJ FKHFNLQJ FKHFNLQJ VHWWLQJ WKH ERXQGDU\ FHOOV LQW PD[UDG LQW QRIPD[UDG IRU LQW M M 0$;[\]fO Mf ^ LQW N IRU LQW L LQF Lf ^ LI FF>L@ ,, FF>L@ f ^ N LI FF>LO@ M , FF>LO@ M __ FF>L\@ M ,, FF>L\@ M ,, FF>L\r[@ M ,, FF>L\r[@ ^ FF>L@ M QRIPD[UDG N ` HOVH HOVH ` PD[UDG L LI N f EUHDN HOVH FKHFNLQJ WKH QR RI QRIPD[UDG LI QRIPD[UDG WKHQ \RX KDYH FKDQJH WKH QXPEHU LI QRIPD[UDG !f ^ FRXW HQGO HQGO HQGO HQGO HQGO HQGO HQGO HQGO HQGO FLQ !! T HOVH SLFN WKH ER[HV ZLWK PD[UDG LQW VXSUDG>O@ LQW 7RWDO1RI%R[>O@ LQW 7RWDO1RI0D[ IRU LQW L LQF Lf ^ LI FF>L@ PD[UDGf
PAGE 94
^ VXSUDG>7RWDO1RI0D[@ L 7RWDO1RI0D[ HOVH FRXW 7RWDO1RI0D[ HQGO FLQ !! T &RXQW IRU WKH QXPEHU RI FHQWHUV RI HDFK FDVH IRU LQW N N7RWDO1RI0D[ Nf ^ UHVHW 7RWDO1RI%R[>@ WR IRU LQW L L Lf ^ 7RWDO1RI%R[>L@ ` 7RWDO1RI%R[>PD[UDG@ LQW WFF>QF@ WHPSRUDU\ FHOO QRWLRQ IRU LQW L LQF Lf WFF>L@ FF>L@ UHDVVLJQ WFF>L@ IRU nIRUn ORRS UHPRYH WKH ELJJHVW VSKHUH IRU LQW S SQF Sf ^ LQW GLVW IORRUVTUW VTXDUH\F>VXSUDG>N@@\F>S@f VTXDUH [F >VXSUDG >N@@[F>S@f VTXDUH]F>VXSUDG>N@@]F>S@f ff LI GLVW PD[UDG f WFF>S@ HOVH ` VHWWLQJ WKH UHPDLQLQJ WXPRU FHOO E\ IRU LQW L LQF Lf ^ LI WFF>L@ f WFF>L@ HOVH VHWWLQJ WKH ERXQGDU\ FHOOV IRU LQW M M 0$;[\]fO Mf ^ LQW IRU LQW L LQF Lf ^ LI WFF>LM f ^ LI WFF>LO@ M __ WFF>LO@ M __ WFF>L\@ M ,, WFF>L\@ M __ WFF>L\r[@ M __ WFF >L\r[@ M ^ WFF>L@ M HOVH HOVH LI f EUHDN
PAGE 95
HOVH FKHFN LI WKHUH LV PRUH PD[UDG SL[FHO IRU LQW L LQF Lf ^ LI WFF>L@ PD[UDG f ^ FRXW HQGO HQGO HQGO :$51,1* :( +$9( 025( n0$;5$'n 3,;&(/ $7 m [F >L@ m m \F>L@ m m ]F >L@ m f HQGO FLQ !! T EUHDN HOVH UHPRYH WKH VPDOOHU RU VSKHUH IRU LQW M PD[UDGO M! fÂ§Mf ^ IRU LQW L LQF Lf ^ LI WFF>L@ M f ^ 7RWDO1RI%R[>M@ UHPRYH WKH VSKHUH IRU LQW OQF f ^ LQW GLVW IORRUVTUW VTXDUH\H>L@\F>@f VTXDUH [F >L@[F >@ f VTXDUH ]F >L@]F >@ f ff LI GLVW M f WFF>O@ HOVH VHWWLQJ WKH UHPDLQLQJ WXPRU FHOO E\ IRU LQW L LQF Lf ^ LI WFF>L@ f WFF>L@ HOVH VHWWLQJ WKH ERXQGDU\ FHOOV IRU LQW M M 0$;[\]fO Mf ^ LQW IRU LQW L LQF Lf ^ LI WFF>LM f ^ LI WFF>LfÂ§@ M __ WFF>LO@ M __ WFF>L\@ M ,, WFF>L\@ M , WFF>L\r[@ M __ WFF>L\r[@ M f ^ WFF>L@ M HOVH HOVH ` LI f EUHDN
PAGE 96
HOVH ! HOVH ! RXWSXW WKH UHVXOW RI FHQWHUV ^ RIVWUHDP RXW&OLHQW)LOHSDW1RI&HQWHUVW[W LRVDSSf ^ RXW&OLHQW)LOH HQGO HQGO HQGO HQGO HQGO Nf VHWZf [F>VXSUDG>N@@[IL[f \F>VXSUDG>N@@\IL[f ]F>VXSUDG>N@@]IL[f f ZLWK UDGLXV PD[UDG HQGO ! IRU LQW M PD[UDG M! M fÂ§f ^ RIVWUHDP RXW&OLHQW)LOHSDW1RI&HQWHUVW[W LRVDSSf ^ RXW&OLHQW)LOH VHWZf M VHWZf 7RWDO1RI%R[>M@ HQGO ` 6HDUFK IRU FHQWHUV IRU LQW N N7RWDO1RI0D[ Nf ^ LQW WFF>QF@ WHPSRUDU\ FHOO QRWLRQ IRU LQW L LQF Lf WFF>L@ FF>L@ UHDVVLJQ WFF>L@ IRU nIRUn ORRS RXWSXW WKH UHVXOW RI FHQWHUV ^ RIVWUHDP RXW&OLHQW)LOHSDWFHQWHUVW[W LRVDSSf ^ RXW&OLHQW)LOH Nf HQGO VHWZf [ VHWZf \ VHWZf ] VHWZf UDGLXV HQGO VHWZf [F>VXSUDG>N@@[IL[f VHWZf \F>VXSUDG>N@@\IL[f VHWZf ]F>VXSUDG>N@@]IL[f VHWZf PD[UDG HQGO ` UHPRYH WKH ELJJHVW VSKHUH IRU LQW S SQF Sf ^ LQW GLVW IORRUVTUW VTXDUH\F>VXSUDG>N@@\F>S@f VTXDUH [F >VXSUDG >N@@[F>S@f VTXDUH ]F >VXSUDG >N@@]F>S@f ff LI GLVW PD[UDG f WFF>S@ HOVH VHWWLQJ WKH UHPDLQLQJ WXPRU FHOO E\ IRU LQW L LQF Lf & LI WFF>L@ f WFF>L@
PAGE 97
HOVH ` VHWWLQJ WKH ERXQGDU\ FHOOV IRU LQW M M 0$;[\]fO Mf ^ LQW IRU LQW L LQF Lf ^ LI WFF>LM f & LI & WFF>LO@ M , WFF>LO@ M ,, WFF>L\@ M ,, WFF>L\@ M ,, WFF>L\r[@ M ,, WFF>L\r[@ M f ^ WFF >L@ M ` HOVH HOVH ` LI fÂ§f EUHDN HOVH ` FKHFN LI WKHUH LV PRUH PD[UDG SL[FHO IRU LQW L LQF Lf ^ LI WFF>L@ PD[UDG f ^ FRXW HQGO HQGO HQGO :$51,1* :( +$9( 025( f0$;5$'f 3,;&(/ $7 [F>L@ \F>L@ ]F>L@ f HQGO FLQ !! T EUHDN ` HOVH UHPRYH WKH VPDOOHU RU VSKHUH IRU LQW M PD[UDGO M! fÂ§Mf ^ IRU LQW L LQF Lf ^ LI WFF>L@ M f ^ 7RWDO1RI%R[>M@ RXWSXW WKH UHVXOW RI FHQWHUV ^ RIVWUHDP RXW&OLHQW)LOHSDWFHQWHUVW[W LRVDSSf ^ RXW&OLHQW)LOH VHWZf [F>L@[IL[f VHWZf \F>L@\IL[f VHWZf ]F>L@]IL[f VHWZf M HQGO ` UHPRYH WKH VSKHUH IRU LQW OQF f ^ LQW GLVW IORRUVTUW VTXDUH\H>L@\F >@f VTXDUH [F>L@[F>O@f VTXDUH]F>L@]F>@f ff LI GLVW M f WFF>O@
PAGE 98
HOVH VHWWLQJ WKH UHPDLQLQJ WXPRU FHOO E\ IRU LQW L LQF Lf ^ LI WFF>L@ f WFF>L@ HOVH VHWWLQJ WKH ERXQGDU\ FHOOV IRU LQW M M 0$;[\]fO Mf ^ LQW IRU LQW L LQF Lf ^ LI WFF>L@ f ^ LI WFF>LO@ M , WFF>LO@ M ,, WFF>L\@ M ,, WFF>L\@ M ,, WFF >L\r[@ M __ WFF>L\r[@ M f ^ WFF >L@ M HOVH ` HOVH ` LI fÂ§f EUHDN HOVH ` ` HOVH ! ^ RIVWUHDP RXW&OLHQW)LOHSDWFHQWHUVW[W LRVDSSf ^ RXW&OLHQW)LOH HQGO ` ` LQW VTXDUHLQW Df ^ UHWXUQ DrD LQW 0$;LQW D LQW E LQW Ff ^ LQW S S D LI D Ef S E HOVH LI D Ff S F HOVH UHWXUQ S !
PAGE 99
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n&RPELQDWRULFDn 2II>*HQHUDOVSHOO@ /H[3UHFHGHV>DB EB@ 0RGXOH>^L` )RU>L L 0LQ>/HQJWK>D@ /HQJWK>E@@ L ,I> D>>L@@ E>>L@@ ,I> D>>L@@ E > >L@@ 5HWXUQ>7UXH@ 5HWXUQ >)DOVH@ @ @ @ ,I> /HQJWK >E@ /HQJWK >D@ 5HWXUQ >)DOVH@ @ 5HWXUQ>7UXH@@ 'HJUHH3RLQW>DB NB@ &RXQW>)ODWWHQ>7R8QRUGHUHG3DLUV>D@@ N@ ,QG6XEJUDSK>*UDSK>JB YB@ VB/LVW@ 0RGXOH>^L M QHZPDW VF` QHZPDW J VF &RPSOHPHQW>5DQJH>/HQJWK>Y@@ V@ )RU>L L /HQJWK>J@ L )RU >M M /HQJWK >J@ M ,I>0HPEHU4>VF L@ __ 0HPEHU4>VF M@ QHZPDW>>L M@@ @@@ *UDSK>QHZPDW Y@@ &XW3RLQWV>JUB@ 6HOHFW>$UWLFXODWLRQ9HUWLFHV>JU@ 'HJUHH3RLQW>JU @ t@ 1RQ&XW3DUWLWLRQV >JUB@ 0RGXOH>^SDUWLWLRQV FSWV EHHV VNHOHWRQ` SDUWLWLRQV ^` EHHV 6HOHFW>%LFRQQHFWHG&RPSRQHQWV>JU@ /HQJWK>@ t@ SDUWLWLRQV 8QLRQ>SDUWLWLRQV EHHV@ VNHOHWRQ ,QG6XEJUDSK>JU 8QLRQ> &RPSOHPHQW> 5DQJH>9>JU@@ )ODWWHQ>EHHV@ @
PAGE 100
&XW3RLQWV>JU@ @ @ SDUWLWLRQV 8QLRQ>SDUWLWLRQV 6HOHFW>&RQQHFWHG&RPSRQHQWV> )URP8QRUGHUHG3DLUV> 6HOHFW>7R8QRUGHUHG3DLUV> VNHOHWRQ@ ,QWHUVHFWLRQ> )ODWWHQ> 6HOHFW>%LFRQQHFWHG&RUDSRQHQWV>VNHOHWRQ@ /HQJWK >@ t@ @ @ t@ @ @ /HQJWK>@ t@ @ 7DEOH> 6HOHFW> 7R8QRUGHUHG3DLUV>JU@ ,QWHUVHFWLRQ> SDUWLWLRQV >>L@@ @ t@ ^L /HQJWK>SDUWLWLRQV@`@@
PAGE 101
$33(1',; & 352*5$0 )25 &/$66,)<,1* 75((6 %< 6(48(1&(6 ,Q WKH ILUVW VHFWLRQ ZH SURYLGH WKH 0DWKHPDWLFD SURJUDP ZH GHYHORSHG LQ MRLQW ZRUN ZLWK 0DWWKHZ +DUYH\ 7KLV SURJUDP ZDV ZULWWHQ E\ 0DWWKHZ +DUYH\ WR JHW WKH PD[LPDO GHJUHH VHTXHQFHV IRU DQ\ JLYHQ LQWHJHU Q $QG ZH VKRZ WKH SURJUDP UHVXOWV XS WR Q LQ WKH VHFRQG VHFWLRQ &O 3URJUDP 2II>*HQHUDOVSHOO@ 'LVFUHWH0DWKf&RPELQDWRULFDr 'R3DUWLWLRQV>QB@ 0RGXOH>^L N WHPSOLVW UHWXUQOLVW` ,I>Q ^^`` UHWXUQOLVW ^` )RU>N Q N NfÂ§ WHPSOLVW 'R3DUWLWLRQV>Q N@ )RU>L L 'LPHQVLRQV>WHPSOLVW@ >>O@@ L UHWXUQOLVW $SSHQG >UHWXUQOLVW )ODWWHQ>^N WHPSOLVW >>L@@`@ @ @ @ UHWXUQOLVW @ @ )LOWHU>WKHOLVWB@ 6HOHFW>WKHOLVW 0HPEHU4>^ ` >>@@@ __ 0D[>O@ )LUVW>@f ,, &RXQW > 0D[>O@@ f f t@ $GG=HURV7R6HT>SDUWB@ 0RGXOH>^Q=HURV Q L M N VXP SRVLWLRQV VHTV ZLWK]HURV` ,I >/HQJWK >SDUW@ 5HWXUQ >^7DEOH >,I >L SDUW>>O@@ @ ^L SDUW >>O@@ !@!@ @ VHTV ^! VXP 6XP>SDUW>>L@@ ^L /HQJWK>SDUW@!@ Q=HURV VXP f /HQJWK>SDUW@ Q 7DEOH>N ^N VXP`@ SRVLWLRQV .6XEVHWV>Q Q=HURV @
PAGE 102
)RUG L /HQJWK >SRVLWLRQV@ L N ZLWK]HURV ^)LUVW>SDUW@` )RU>M M VXP M ,I>0HPEHU4>SRVLWLRQV>>L@@ M@ ZLWK]HURV $SSHQG>ZLWK]HURV @ ZLWK]HURV $SSHQG>ZLWK]HURV SDUW>>N@@@ N @ @ ZLWK]HURV $SSHQG>ZLWK]HURV @ VHTV $SSHQG>VHTV ZLWK]HURV@ @ VHTV @ $GG=HURV>VHTB@ )ODWWHQ>7DEOH> $GG=HURV7R6HT>VHT>>L@@@ ^L /HQJWK>VHT@`@@ 6XE6HTV>GVHTB@ 0RGXOH> ^EUDQFKOLVW EUDQFK ZRUNVHT EUDQFKHV L` EUDQFKHV GVHT>>O@@ ZRUNVHT 'URS>GVHT @ EUDQFKOLVW ^` )RU>L L EUDQFKHV L GVXP ZRUNVHT>>@@ ZRUNVHT 'URS>ZRUNVHT @ EUDQFK ^GVXP` :KLOH>GVXP GVXP ZRUNVHT>>O@@ f EUDQFK $SSHQG> EUDQFK ZRUNVHT>>O@@ @ ZRUNVHT 'URS>ZRUNVHT @ @ f EUDQFKOLVW $SSHQG>EUDQFKOLVW EUDQFK@ @ L EUDQFKOLVW @ /H[3UHFHGHV>DB EB@ 0RGXOH>^L` )RU>L L 0LQ>/HQJWK>D@ /HQJWK>E@@ L ,I>D>>L@@ E>>L@@ ,I > D>>L@@ E>>L@@ 5HWXUQ >7UXH@ 5HWXUQ >)DOVH@ @ @ @ ,I>/HQJWK>E@ /HQJWK>D@ 5HWXUQ >)DOVH@ @ 5HWXUQ>7UXH@ @ :HLJKWV4>GVHTB@ 0RGXOH>^L VXEVHTV IODJ` VXEVHTV 6XE6HTV>GVHT@ IODJ ,I>2UGHUHG4>VXEVHTV /H[3UHFHGHV@
PAGE 103
)RUG L /HQJWK>VXEVHTV@ L ,I& :HLJKWV4>VXEVHTV>>L@@@ IODJ IODJr O @ @ 5HWXUQ>)DOVH@ @ ,I> IODJ 5HWXUQ>7UXH@ 5HWXUQ>)DOVH@ @ @ 'HJUHH6XP4>GVHTB@ 0RGXOH >^GVXP` GVXP GVHT>>O@@ )RU>L L /HQJWK>GVHT@ L GVXP GVHT>>L@@ ,I>GVXP ,I >L /HQJWK>GVHT@5HWXUQ>7UXH@5HWXUQ>)DOVH@ @ @ @ 5HWXUQ>)DOVH@ @ ([WUDFW6XEWUHH>GVHTB@ 0RGXOH >^UWRWDO` UWRWDO GVHT>>O@@ L :KLOH>UWRWDO L UWRWDO GVHT>>L@@ @ 7DNH>GVHT L@ @ 5HZULWH7UHH >GVHTB LB@ 0RGXOH>^EUDQFKHVRII VXEWUHH EUOHQJWK VHTOVW` ,I>L 5HWXUQ>GVHT@@ EUDQFKHVRII L :KLOH> EUDQFKHVRII tt 6XP>GVHT>>N@@ ^N EUDQFKHVRII L `@ f __ GVHT>>EUDQFKHVRII@@ EUDQFKHVRIIfÂ§ @ EUOHQJWK /HQJWK> ([WUDFW6XEWUHH>'URS>GVHT L @@ @ VXEWUHH 'URS>GVHT ^L L EUOHQJWK `@ VXEWUHH>>EUDQFKHVRII@@fÂ§ VHTOVW 6XE6HTV>7DNH>GVHT ^L L EUOHQJWK `@@ VHTOVW $SSHQG>VHTOVW 5HZULWH7UHH>VXEWUHHEUDQFKHVRII@@ VHTOVW 6RUW>VHTOVW /H[3UHFHGHV> @ t@ )ODWWHQ>^GVHT>>L@@ VHTOVW`@ 5LJKW2UGHU4>GVHTB@ 0RGXOH>^SRV VHTV M UHZULWWHQ` SRV )ODWWHQ> 3RVLWLRQ>GVHT GVHT>>O@@ @ @ ,I>SRV ^` 5HWXUQ>7UXH@@ VHTV ^` )RU >M M /HQJWK >SRV@ M UHZULWWHQ 5HZULWH7UHH> GVHT SRV>>M@@ @
PAGE 104
,I>/H[3UHFHGHV>UHZULWWHQ GVHT@ tt UHZULWWHQ GVHT 5HWXUQ>)DOVH@ @ @ 5HWXUQ>7UXH@ /HPPDV4>GVHTB@ 'HJUHH6XP4>GVHT@ tt :HLJKWV4>GVHT@ tt 5LJKW2UGHU4>GVHT@ &KHFN/HPPDV>OVWVHTB@ 0RGXOH>^UHWVHT WZRVHT L M Q S` UHWVHT ^` M 3 Q /HQJWK>OVWVHT@ )RU>L L Q L ,I> M Q 3ULQW >S n@ S M @ M ,I >/HPPDV4 > OVWVHT>>L@@ @ UHWVHT $SSHQG> UHWVHT OVWVHT>>L@@ @ @ @ WZRVHT ^` ,I >/HQJWK>OVWVHT@ )RU>L L /HQJWK> UHWVHT>>O@@ @ L WZRVHT $SSHQG>WZRVHT @ @ @ WZRVHT RLQ>WZRVHT ^ `@ UHWVHT $SSHQG>UHWVHT WZRVHT@ UHWVHT @ '675HFXUVLYH>GVHTB EYHUWB KYHUWB WUSDUDPB @ 0RGXOH>^L Y WUHH WHPSVHT` Y KYHUW WUHH WUSDUDP ,I>GVHT ^` WHPSVHT 'URS>GVHT @ )RU>L L GVHT>>O@@ L Y WUHH $SSHQG>WUHH ^EYHUW Y`@ VXEVHT ([WUDFW6XEWUHH>WHPSVHT@ WHPSVHT 'URS>WHPSVHT /HQJWK>VXEVHT@@ ^WUHH Y` '675HFXUVLYH>VXEVHT Y Y WUHH@ @ ^WUHH Y`
PAGE 105
'HJUHH6HT7R7UHH>GVHTB@ )LUVW> '675HFXUVLYH>GVHT ^`@ @ 7R*UDSKLFV>FNGB@ )RU>W W /HQJWK>FNG@ W 3ULQW>FNG>>W@@@ 6KRZ*UDSK>5DGLDO(PEHGGLQJ>)URP8QRUGHUHG3DLUV > 'HJUHH6HT7R7UHH>FNG>>W@ @ @ @ @ @ @ & 6RPH 3URJUDP 5HVXOWV 7KH QRWDWLRQ IRU D GHJUHH VHTXHQFH LV G? J" GQf %XW WKH 0DWKHPDWLFD SURJUDP SURGXFHV WKH VHTXHQFH QRWDWLRQ ZLWK ^ DQG ` 6R LQ WKLV VHFWLRQ DQ\ VHW QRWDWLRQ LV DFWXDOO\ GHQRWHV WKH GHJUHH VHTXHQFH 1RWH WKDW DQ QWUHH KDV Q YHUWLFHV VR WKHUH DUH Q HGJHV Q WRWDO ^ ` Q WRWDO & ` ^ Q WRWDO ^ ^ ` ^ `
PAGE 106
Q WRWDO ^ ` ^ ` ^ ` ^ ` ^ ` ^ ` Q WRWDO Â R ^ ^ R R ` ^ ^ ^ R ` ^ ^ ` ^ R ` ^ ^ `
PAGE 107
U+ U6 U6 U6 U6 U6 U6 U6 U6 U6 U6 U6 ,2f8&rf:MAAL:O6 SKNN,20KK,2K02 3,fO,2O2.,22322 WrfO0 fÂ‘2LL2!N2Lf2Wf22&! f02222222222 220222222222 RRRRRRRRRRRR 6n6n6f6n6nnUnU6fOU6nOU6n U6U6U6U6U6U6U6U6U6U6U6 &2 &2&2&2&2&2AIAAWA&O2L LfÂ§OfÂ§/ OfÂ§N OfÂ§rÂ‘ OfÂ§N fÂ§N &' OfÂ§N &' &' R! N2O2LN2+A.N22L$ fÂ‘f222222222 2KA22WA2WA2222 RRRRRRRRRRR RRRRRRRRRRR 6! 6r 6r 6 6 6 66! 6 6 6! &2 &2 Q WRWDO
PAGE 108
WRWDO 2 R U6 U6 U6 U6 U6 fÂ§6 U6 Uf6 U6 U6 U6 U6 Uf6 U6 U6 U6 Uf6 U6 U6 U6 U6 U6 fÂ§6 2 R R R R R R R R R R R R R R R R R R 2 R R R 2 R R R R R R R R R R R R R R R R R R R U+ R R R R R R R R 2 nU+ 2 2 2 2 2 2 2 U+ R R R 2 U+ R R 2 2 2 7+ R 2 2 U+ 2 U+ 2 2 W+ W+ 2 2 2 U+ R U+ U+ U+ R R R R R R &1 W+ 2 2 2 2 + 2 2 2 2 U+ R U+ U+ 2 U+ R Q+ &= &= U+ WfÂ§L 2 2 2 U+ R 12U+2&1U++ U+ &1 2 U+ U+ R R 2 }K R Â‘r} 2 2 + U+ 1 7+ Q+ W+ &1 W+ W+ U+ &1 U+ U+ U+ U+ U+ + &1 + + &1 &1 &1 &1 U+ + + + &1 &1 2, &1 U+ U+ U+ U+ U+ 1 LQLQLUXQAnnAA A &2 &2 &2 &2 FR FR FR FR FR 66r 6U6r 6r 6Ur 6Ur 6Un 6Ur 6U6Ur 6r 6r 6U6Ur 6Un 6Un 6Ur 6Ur 6n 6Ur 6r 6n U6 U6 U6 U6 U6 Uf6 U6 U6 U6 U6 U6 Uf6 Ur6 U6 UA} U6 U6 U6 U6 U6 U6 U6 U6 U6 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 R 2 R R R R R 2 2 2 2 2 2 U+ R 2 2 2 2 R R R U+ R R 2 R R R R R 2 2 2 2 2 2 2 2 2 2 L+ R U+ R R 2 U+ U+ 2 R R R R R 2 2 W+ 2 2 2 U+ W+ R 2 2 W+ 2 R R U+ 2 U+ R 2 2 U+ R 7+ 2 2 U+ W+ 2 2 2 U+ 7+ R U+ R R U+ &1 2 U+ U+ 2 R R R R R 2 2 2 2 2 + W+ 2 W+ W+ 2 nU+ U+ U+ U+ U+ U+ U+ R 2 7+ 2 &1 U+ U+ &1 2 7+ &2 &1 2 U+ U+ &1 U+ U+ &1 &1 U+ U+ U+ U+ R &1 U+ &1 + + &2 &1 &1 &1 U+ U+ U+ U+ U+ &1 &1 &1 U+ U+ U+ U+ U+ U+ FR &' &' /2 /2 /2 A A A A A Â‘L Q nW A UR Q Q FR FR FR FR FR &1 6} YYYYY 6} 6r 6r 6} 6} 6} 6} 6} 6} 6} 6! 6r 6} 6} 6} 6r 66UWRWDO U6 U6 U6 U6 U6 U6 U6 U6 U6 U6 6 U6 U6 U6 U6 6 U6 U6 U6 6 U6 Uf6 Uf6 R R 2 R 2 2 2 R R 2 2 2 2 2 R R 2 R 2 2 2 2 2 2 R 2 R 2 R R R R 2 R R R R 2 2 R R R 2 2 U+ R R R 2 R 2 R R R R R R R R 2 R R R R R 2 2 2 R R R R R 2 R R R U+ R R R R U+ R R R R U+ 2 2 U+ U+ R R R R U+ R U+ R R U+ R U+ R 2 U+ R R R 2 R &1 2 U+ R R R R 2 R 2 &1 U+ 2 R 2 R U+ 2 R fÂ§ R R R 2 U+ U+ R U+ R R U+ U+ R R R R U+ R &1 2 U+ R R R R U+ R 2 2 R 2 R U+ 2 U+ R R 2 U+ U+ &1 U+ U+ U+ &1 U+ FR &1 &1 U+ R R U+ W+ FR &1 U+ U+ FR &1 &1 &1 &1 U+ U+ U+ U+ &2 FR &1 &1 &1 &1 &1 &1 FRVW'W'WR2/2LILRXfLRLUfLRLQL2nAAAWAAA I6 U6 U6 U6 Uf6 U6 U6 U6 U6 U6 U6 U6 U6 U6 Uf6 U6 LfÂ§6 U6 Uf6 Uf6 U6 U6 U6 2 R R R R R R R R R R R 2 2 2 2 2 2 2 2 2 2 2 2 R R R R R R R R R R R R R R R R R R R R 2 R R R R R R R R R R R R R R U+ R R R R U+ R R 2 U+ R R R R R R R R R R R R R 2 R R U+ R 2 R R 2 2 R R R R R R R R R U+ R R R U+ U+ R 2 R R U+ &1 2 U+ R R R U+ R U+ R R R U+ U+ R R 2 U+ U+ R R &1 2 U+ 2 U+ R R R R R R R R R R R R U+ U+ 2 U+ R U+ 2 R 2 &1 2 R R U+ R &1 U+ R U+ &1 R U+ FR &1 2 U+ U+ U+ U+ R &1 R U+ R R &1 U+ &1 U+ U+ FR &1 &1 &1 U+ U+ U+ U+ U+ FR FR &1 &1 &1 &1 &1
PAGE 109
HQG RI GDWD Uraa? Urf! r} r} L} UrW U6 r! UrfL UraL Ur? Ura\ Uraa? WaraL UAfÂ§L UAW UAW UraaW Ur} SrfnL U} UrA UraaW UraA Ur! Ur} U} fÂ§rfÂ§} FRFRFRFRFRFRFRFRFRFRFRFRFRFRFRFRFRFR!e!!I}!I}!W!Ae}!I!!IrAe}AWe KN KfÂ§/ KN KN KN KN KN KN KN WR WR WR WR WR WR WR WR WR KN KN KN KN KN KN KN KN KN WR WR WR KN KN KN KN KN KN WR WR WR KN KN KN KN KN KN WR WR WR KN KN KN KN KN WR WR WR &2 KN KN KN KN KN KN R 0 WR KN KN KN KN KN KN R WR WR KN KN WR KN KN R KN WR KN KN R KN KN R R KN KN R KN W2 KN KN KN WR KN R R KN KN R R R R KN R KN WR R KN R R R R KN R KN R KN KN KN R KN R R R KN R R R R KN R R KN KN KN R R R R KN R KN R KN Kr KN KN R R KN R R R KN KN KN KN R R R R R R KN R R KN R KN R R R KN R R KN KN KN R R R KN R R R KN R R KN R KN R R R KN R R R R KN R R R KN 222.n22222222022222222222222222 RRRRRRRRRRRRRRRRRRRRRRRRRRRRRR 9999 :9999999999999 ::: ZYYY _3aL U6 U6 Lr} UrY LfL LrfL U6 LrW rr ? Urr Srr Sr Sr} Sr! Sr} UAa? Ur} SL Urr Sr SW SrfL Sar Srr Srf? W2&2&2&2&2&2&2&2&2&2&2&2&2&2&2&2&2&2&2e!!I n!3!A!enA!en!e!!e!!IrWIr .....333.:,2,2:I2,2O20I203LA+3333+:W2: 3333333W2I2333333,2I20W23333300I2333 33333003333333,23322.3208.2,2332 }002OAWfOfOf2OfWf22W2Wf222Wf2OfOf2222!f22 fWfn20WOf22OfOfn2Of22222}rW2Or222OfWf222Ofn KN2KNLN222LN2222O2WL2LNLN2Or2ON222OL22Or0 OLLO2222Of2ON2K}nON22222222}W222222222 22WnOn22222222022222222022222222 RRRRRRRRRRRRRRRRRRRRRRRRRRRRRR ZZZZZZZZZZZZZZZ R
PAGE 110
$33(1',; /$%(/(' *5$3+6 $1' 75((6 'O 1XPEHU RI /DEHOHG *UDSKV DQG /DEHOHG 7UHHV 7R REWDLQ WKH QXPEHU RI ODEHOHG JUDSKV ZLWK Q YHUWLFHV ZH QHHG RQO\ REVHUYH WKDW HDFK RI WKH 4f SRVVLEOH HGJHV LV HLWKHU SUHVHQW RU DEVHQW 7KXV ZH KDYH WKH IROORZLQJ WKHRUHP 7KHRUHP >@ 7KH QXPEHU RI JUDSKV RQ Q ODEHOHG YHUWLFHV LV ,Q /RYÂ£V] >@ WKH 3ULLIHUFRGH LV XVHG WR GHQRWH D ODEHOHG WUHH $QG IRU DQ\ JLYHQ 3ULLIHUFRGH RQH FDQ FRQVWUXFW WKH WUHH ZKLFK LV GHQRWHG E\ WKDW FRGH 7KHQ XVLQJ WKH SURSHUWLHV RI WKH 3ULLIHUFRGH &D\OH\ SURYHG WKH IROORZLQJ WKHRUHP 7KHRUHP 'O &D\OH\f >@ 7KH QXPEHU RI WUHHV RQ Q ODEHOHG YHUWLFHV LV QQa 6NHWFK RI WKH &RQVWUXFWLRQ RI D 3ULLIHUFRGH $PRQJ WKH HQG YHUWLFHV RI D JLYHQ ODEHOHG WUHH SLFN WKH RQH ZKRVH ODEHO LV PLQLPDO &XW LW RII IURP WKH WUHH DQG ZULWH GRZQ WKH ODEHO RI LWV DGMDFHQW YHUWH[ 5HSHDW WKLV SURFHVV XQWLO RQO\ RQH HGJH UHPDLQ $W WKLV SRLQW ZH KDYH D VHTXHQFH RI Q fÂ§ QXPEHUV ZULWWHQ GRZQ :H FDOO WKLV UHVXOWLQJ VHTXHQFH WKH 3ULLIHUFRGH RI WKH WUHH 7KH IROORZLQJ H[DPSOH VKRZV KRZ ZH FRQVWUXFW WKH 3ULLIHUFRGH RI D JLYHQ WUHH ([DPSOH /HW 7 EH WKH WUHH VKRZQ EHORZ )LJXUH 'O 7KHUH DUH HQG YHUWLFHV :H FXW RII WKH VPDOOHVW ODEHOHG HQG YHUWH[ ZKLFK LV 7KHQ WKH VHTXHQFH LV f DQG WKH UHPDLQLQJ WUHH LV )LJXUH EHORZ 7KHQ ZH FXW RII WKH QH[W VPDOOHVW ODEHOHG HQG YHUWH[ ZKLFK LV DQG DGG WKH ODEHO RI WKH DGMDFHQW YHUWH[ RI WR WKH VHTXHQFH 7KDW LV WKH VHTXHQFH LV f )URP WKH UHPDLQLQJ WUHH ZH FXW RII WKH QH[W VPDOOHVW ODEHOHG HQG YHUWH[ ZKLFK LV 7KHQ WKH VHTXHQFH LV f $QG ZH JHW WKH WUHH VKDSH LQ )LJXUH
PAGE 111
)LJXUH )LJXUH $PRQJ WKH HQG YHUWLFHV LV WKH VPDOOHVW ODEHOHG RQH 6R ZH FXW RII WKH YHUWH[ DQG ZULWH GRZQ WKH ODEHO RI WKH DGMDFHQW YHUWH[ RI 7KH VHTXHQFH WKHQ ZLOO EH f DQG WKH UHPDLQLQJ WUHH LV )LJXUH )LJXUH 7KH UHPDLQLQJ WUHH KDV WZR HQG YHUWLFHV WKDW LV LW LV D OLQHDU VKDSH :H FXW RII WKH VPDOOHU ODEHOHG YHUWH[ $QG WKH VHTXHQFH ZLOO EH f $IWHU ZH FXW RII WKH QH[W VPDOOHU ODEHOHG YHUWH[ ZKLFK LV DQG JHW WKH VHTXHQFH f WKH UHPDLQLQJ WUHH KDV RQO\ WZR YHUWLFHV ,WfV FOHDU ZH KDYH WR FXW RII WKH VPDOOHU ODEHOHG YHUWH[ ZKLFK LV DQG WKH DGMDFHQW YHUWH[ ZLOO EH 7KHUHIRUH ZH GR QRW QHHG WR DGG WKLV QXPEHU WR WKH VHTXHQFH +HQFH WKH ODVW VHTXHQFH f LV WKH 3ULLIHUFRGH RI WKH JLYHQ WUHH 7 ,I ZH DVVLJQ GLIIHUHQW ODEHOV IRU WKH WUHH DERYH WKHQ ZH ZLOO JHW D GLIIHUHQW 3ULLIHUFRGH %XW WKH FRGH PXVW FRQVLVW RI WHUPV ZKLFK LV WZR OHVV WKDQ WKH WRWDO QXPEHU RI YHUWLFHV LQ WKH WUHH $QG HDFK WHUP FDQ EH DQ\ QXPEHU DPRQJ 7KXV ZH KDYH WKH IROORZLQJ WKHRUHP 5HFRQVWUXFWLRQ RI D 7UHH IURP D 3ULLIHUFRGH /HW DL RfBf EH D 3ULLIHUFRGH 7KDW LV ZH H[SHFW DQ QWUHH /HW L f f f QBL EH WKH LQGLFHV RI WKH UHPRYHG SRLQWV 7KHQ IRU DQ\ L E^ LV REYLRXVO\ GLIIHUHQW IURP BLD $OVR A DM LI M L EHFDXVH LV UHPRYHG LQ WKH LWK VWHS VR LW FDQ QRW EH DQ DGMDFHQW YHUWH[ ODWHU &RQYHUVHO\ LI N e ^" B[ D D[ DQB` WKHQ 9N LV DQ HQG YHUn WH[ RI 97f fÂ§ ^XMMM8MM 7KXV
PAGE 112
EL PLQ^$_$ e ^E E L DÂ DL f f f A` /HW WKH VHTXHQFH f EH D 3ULLIHUFRGH 7KDW LV D D OD OD D fÂ§ D 7KHQ PLQ^$_$ A ^DL D D`` PLQ^IF_$ ^`` PLQ^` 7KDW LV WKHUH LV DQ HGJH EHWZHHQ D[ DQG E? 7KHQ t PLQ^$_$ A ^ D D`` PLQ^$_$ A ^`` PLQ^` 7KXV WKH WUHH 7 KDV DQ HGJH EHWZHHQ D DQG E ,Q D VLPLODU PDQQHU ZH KDYH WKH IROORZLQJ PLQ^$_IF A ^LD D`` PLQ^IF_$ A ^ `` PLQ^ ` 7KXV 7 KDV DQ HGJH ^D` ^`f PLQ^IF_IF A ^DDD`` PLQ^$_$ A ^`` PLQ^O` 7KXV 7 KDV DQ HGJH ^D` ^`f E PLQ^$_$ A ^LDD`` PLQ^$_$ A ^ `` PLQ^` 7KXV 7 KDV DQ HGJH ^` ^`f E PP^N?N A ^EXEEEEDH`` fÂ§ PLQ^IF_IF A ^ `` PLQ^` 7KXV 7 KDV DQ HGJH ^D` ^`f $QG IRU WKH YHUWH[ Dt WKH ODUJHVW ODEHOHG YHUWH[ X ZLOO EH WKH UHPDLQLQJ DGMDFHQW YHUWH[ WKDW LV WKHUH LV DQ HGJH ^` 6R WKH WUHH 7 KDV WKH YHUWH[ VHW 9 ^ ` DQG WKH HGJH VHW ( ^^` ^` ^` ^` ^` ^ ` ^`` ZKLFK LV WKH WUHH VKDSH LQ WKH H[DPSOH DERYH 7KDW LV ZH FDQ FRQVWUXFW D ODEHOHG WUHH IURP D 3ULLIHUFRGH &OHDUO\ GLIIHUHQW 3ULLIHUFRGHV SURGXFH GLIIHUHQW ODEHOHG WUHHV
PAGE 113
%,2*5$3+,&$/ 6.(7&+ 7DH,O
PAGE 114
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r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
PAGE 115
HDPDWD
A CLASSIFICATION OF TREES AND
APPLICATIONS OF TOPOLOGY AND GRAPH THEORY TO
NEUROSURGERY
By
TAEIL YI
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
2000
Copyright 2000
by
TAEIL YI
TO
My parents, BeomYoung & BongWol,
who show me how to love,
My wife, MoonSil,
who shows me what is love,
and
My children, David & Peter,
who are the reasons why I love
ACKNOWLEDGEMENTS
I would like to express my sincere appreciation to the members of my superviÂ¬
sory committee, Dr. Beverly Brechner, Dr. Philip Boyland, Dr. James Keesling, Dr.
LiChien Shen, and Dr. Mary Kantowski, for their patience, knowledge, concern, and
especially for showing me how to study and teach. I would especially like to thank
my advisor, Dr. Beverly Brechner, who accepted me and gave me a chance to work
with her. I also want to express my appreciation for all her effort and all that she has
done for me, not only in my major field, but also in my life. I would like to thank
Matthew Harvey who brought me some wonderful ideas and wrote the Mathematica
programs. I would like to thank Dr. Francis Bova who brought me these questions
which were â€˜new worldâ€™ to me. I would like to thank Dr. Meeks for helpful converÂ¬
sations, and Dr. Thomas Wagner who gave me the information about neurosurgery
that I did not know before. I would like to thank Dr. Yunmei Chen who gave me
inspiration. I would like to thank Dr. Miklos Bona who showed me some insights of
combinatorics. Finally, I would like to thank the Department of Mathematics that
allowed me to be here and meet all these wonderful people.
IV
TABLE OF CONTENTS
ACKNOWLEDGEMENTS iv
ABSTRACT vii
CHAPTERS
1 INTRODUCTION 1
2 AN AUTOMATED SPHERE PACKING PLAN FOR BRAIN TUMORS 5
2.1 Introduction 5
2.2 Linac 5
2.3 Procedure of Radiosurgery 6
2.4 Making a Sphere by Arcs of Beams 10
2.5 Sphere Packing Plan 10
2.6 Shelling Algorithm 15
3 A CLASSIFICATION OF UNLABELED TREES 25
3.1 Introduction 25
3.2 Definitions 26
3.3 Vertex Sequence 28
3.4 Degree Sequence 36
3.5 Perfect Sequence and Main Theorem 44
3.6 Obtaining Perfect Sequences 45
3.7 Extended Partitions and the Algorithm 53
4 TREES AS INVARIANTS FOR BRAIN TUMOR SHAPES 59
4.1 Introduction 59
4.2 Definitions and Some Properties 60
4.3 Graphs for Sphere Packing Plans 64
4.4 Order in the Vertex Set 65
4.5 Maximal Tree from a Graph 71
4.6 Graph from a Maximal Tree 81
4.7 Algorithm for Our Maximal Tree 81
v
APPENDICES
A SHELLING PROGRAM FOR SPHERE PACKING 82
A.l Program for Data Reading 82
A.2 Program for Center Searching 83
B PROGRAM FOR THE MAXIMAL TREE OF A GRAPH 91
C PROGRAM FOR CLASSIFYING TREES BY SEQUENCES 93
C.l Program 93
C.2 Some Program Results 97
D LABELED GRAPHS AND TREES 102
D.l Number of Labeled Graphs and Labeled Trees 102
D.2 Sketch of the Construction of a Priifercode 102
D.3 Reconstruction of a Tree from a Priifercode 103
REFERENCES 105
BIOGRAPHICAL SKETCH 105
vi
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
A CLASSIFICATION OF TREES AND
APPLICATIONS OF TOPOLOGY AND GRAPH THEORY TO
NEUROSURGERY
By
TaeIl Yi
August 2000
Chairman: Dr. Beverly L. Brechner
Major Department: Mathematics
The work of this dissertation was motivated by our work with a medical group
at the Brain Institute at the University of Florida. This group was led by Dr. Francis
Bova, who raised the following questions:
(1) Can you automate the timeconsuming sphere packing process for our â€œmultiple
isocenters method of stereographic radiosurgeryâ€ for brain tumors? (This surgery is
noninvasive, radiation surgery to destroy brain tumors.)
(2) Can you classify the shapes of brain tumors?
In this dissertation, we construct some algorithms and programs to (partially)
answer these questions. Our major (joint) results are described below.
Our first result is an algorithm and a program that produce a unique sphere
packing plan for any given brain tumor shape, filling the shape with different size
spheres in a unique way. This process automates the â€˜sphere packingâ€™ part of the
Vll
surgery plan. Additional code by the medical group was necessary to allow for physiÂ¬
cian input in certain critical cases. It was then possible for the medical group to
incorporate this into their surgery plan, semiautomating it, and significantly reducÂ¬
ing the surgeonsâ€™ surgery planning time.
Our second result is a complete classification of unlabeled ntrees. We assign
to each ntree, a sequence of n nonnegative integers satisfying certain properties; and
from each such sequence, we can recover the ntree assigned to it. In fact, we present
both an algorithm and code to produce not only the sequences, but also pictures of
the corresponding trees.
Our final result is a first attempt at a topological classification of tumor shapes.
We first use our sphere packing plan to construct a graph representing the tumor.
We then have an algorithm and program which uniquely select a maximal tree in the
graph. Thus we can assign a unique tree to each tumor shape. However, different
graphs may produce the same tree. Our tree classification allows us to compare
two tumor shapes very easily, by just looking at their corresponding sequences. In
particular, different sequences come from different sphere packing plans, but the
converse does not hold.
Vlll
CHAPTER 1
INTRODUCTION
The work of this dissertation was motivated by two questions raised by Dr.
Francis Bova, of the Brain Institute, Shands Hospital, and the Department of NeuroÂ¬
surgery, all at the University of Florida. It turns out the differentsounding questions
are very much related and are tied together naturally in this thesis. In fact, our
results for the second question depend upon our results for the first question. Dr.
Bova led a medical group, which joined with our mathematics group, for a number
of discussions over a two year period. This resulted, for us, in both a joint medical
paper [11] and in this dissertation. We also want to reference the dissertation [10] of
Thomas Wagner, Dr. Bovaâ€™s medical physics student, who worked with us.
We expect to have three mathematics publications from this dissertationfrom
Chapters 2, 3, and 4, respectively. The paper resulting from Chapter 2 will be joint
with Professor Beverly Brechner, Professor Yunmei Chen, and Dr. Thomas Wagner.
The papers resulting from Chapters 3 and 4 will be joint with Matthew Harvey, an
undergraduate student working with Professor James Keesling on his honors thesis.
Further, we anticipate an additional paper, to be joint with Yunmei Chen and her
student, Dr. Stacey Chastain. This anticipated work will be an analysis of the tumor
shapes of approximately 1000 interesting tumor radiation surgeries done by Bova and
his colleagues at the University of Florida.
Dr. Bovaâ€™s questions were the following:
1. Is it possible to automate the medical groupâ€™s â€œmultiple isocenters method of
stereographic radiosurgeryâ€ to destroy brain tumors?
2. Is it possible to classify brain tumor shapes?
1
2
In this dissertation, we develop some algorithms and programs related to these
questions. In stereographic radiosurgery, which is noninvasive surgery, radiation is deÂ¬
livered by means of collimatorscylindrical metal tubes, with cylindrical holes through
them, through which the radiation flows. The goal of stereotactic radiosurgery for a
brain tumor is to deliver the desired dosage to the target, and only the target. This
is not possible in reality. So they do the next best thing, which is to deliver enough
dosage to the target, to avoid as much normal tissue as possible, and to deliver as
little radiation as possible to whatever normal tissue must be affected. There are two
additional important criteriadose homogeneity and dose conformality. That is, we
do not want â€˜hot spots,â€™ which have been experimentally determined to cause comÂ¬
plications; and we do want rapid falloff of dose levels outside the actual tumor. One
of several such radiation surgery methods is called the â€˜Multiple Isocenter Method.â€™
This involves filling the tumor image with spheres of different sizes, until the image is
best filled up. In this dissertation, we call such a plan, a â€˜Sphere Packing Treatment
Plan.â€™
In Chapter 2, we develop both an algorithm, which is called a â€˜shelling processâ€™
(or â€˜grassfire processâ€™ in medical terminology), and a CH program to identify the
centers and sizes of spheres for this sphere packing treatment plan. For any given
brain tumor, our program gives us a unique sphere packing plan. This sphere packing
plan can be represented numerically as a list of parameters of the spheres, namely
the centers and radii of the spheres. Based on this algorithm, the medical group
developed a proprietary program for an automated sphere packing treatment plan,
including dosage distribution, and other things (see Wagner et al. [11]). Since this
program always produces the same plan for the same tumor, it can be used as a
â€˜benchmarkâ€™ against which to compare other treatment plans.
In Chapter 3, we obtain a code to denote an unlabeled tree by a unique,
ordered, nonnegative integer sequence satisfying certain properties. We call such
3
sequences, perfect sequences. That is, we obtain a complete classification of unlabeled
ntrees by means of the perfect sequences of n nonnegative integers. Thus we have our
main theorem of this chapter, the Tree Classification Theorem, (Theorem 3.5.5): For
any positive integer n, let T(n) be the set of unlabeled ntrees and F(n) the set of
perfect sequences of ntrees. Then there is a onetoone correspondence between T{n)
and V{n). We show how to assign a perfect sequence to a given tree, and conversely,
how to reconstruct a tree from a given perfect sequence. We also include an algorithm
and a Mathematica program to produce not only all the perfect nsequences, for any
given positive integer n > 3, but also pictures of the corresponding trees. By means of
this program, we get our classification of unlabeled ntrees and also the total number
of ntrees.
In Chapter 4, we make a first attempt at a topological shape classification of
tumorsactually of the sphere packing plan associated with the tumor shape. We first
assign to each tumor a unique graph, and from that, we determine a unique tree. We
obtain the graph by assigning a vertex for each sphere, and an edge between each pair
of vertices representing adjacent spheres. This produces a unique connected graph
corresponding to each sphere packing plan. Then, by using the notion of cutvertex, we
give an order to the vertex set. We then use this order to build a unique maximal tree
contained in the graph. Next, we develop an algorithm and a Mathematica program
to assign a unique tree to each sphere packing plan. This program produces a list of
edges of the tree, making it clear which edges are incident.
Using our results of Chapter 3, we can now assign exactly one perfect sequence
to each tumor. However, it may happen that more than one tumor shape (or graph)
gives rise to the same tree, and therefore to the same perfect sequence. Thus, we do
not yet have a real classification of tumor shapes. However, if two perfect sequences
are distinct, then their corresponding trees, and therefore their respective correspondÂ¬
ing graphs and sphere packing plans, are also distinct. Therefore we may consider
4
their shapes to be distinct. Thus, we have our main result of Chapter 4, Theorem
4.5.6: The perfect sequences are invariants of the shapes of arbitrary brain tumors.
That is, if two brain tumors are represented by distinct perfect sequences, then their
corresponding trees are not isomorphic, and therefore their respective graphs and
sphere packing plans are not isomorphic. Thus, we may consider their shapes to be
distinct.
We refer the reader to Douglas[1], Gross[3], and Harary[5] for graph theory.
We also refer the reader to Hall[4], Riordan[7] and Stanley[9] for combinatorics.
CHAPTER 2
AN AUTOMATED SPHERE PACKING PLAN FOR BRAIN TUMORS
2.1 Introduction
The goal of stereotactic radiosurgery for a brain tumor is to deliver the desired
dosage to the target, and only the target. This is not possible in reality. So they do
the next best thing, which is to deliver enough dosage to the target, to avoid as much
normal tissue as possible, and to deliver as little radiation as possible to whatever
normal tissue must be affected. There are two additional important criteriadose
homogeneity and dose conformality. That is, we do not want â€˜hot spots,â€™ which
have been experimentally determined to cause complications; and we do want rapid
falloff of dose levels outside the actual tumor. One of several such radiation surgery
methods is called the â€˜Multiple Isocenter Method.â€™ This involves filling the tumor
image with spheres of different sizes, until the image is best filled up. This noninvasive
method of surgery, namely by using radiation, relies on a piece of equipment called the
Linear Accelerator (or simply, Linac). In this chapter, we show how to automate the
construction of a sphere packing plan for the tumor image. Most of the information
in this chapter about treatment of a brain tumor is taken from Friedman et al. [2].
See [2] for further reference.
2.2 Linac
According to Friedman et al. [2], the linear accelerator is a complex machine
capable of producing Xrays. A large amount of energy is generated by the power
supply, which then powers the filament shown. This causes electrons to be emitted
by the filament, which are in turn accelerated to higher energies using a (microwave)
5
6
wave guide. The electrons are then changed in direction by the magnet so that they
impact on a heavy metal alloy target. This results in Xray production that can then
be collimated or shaped by both primary and secondary collimators within the linear
accelerator head. This beam is further collimated for radiosurgery by the tertiary
radiosurgery collimator. (See Figure 21 and Figure 22.)
The Linac is mounted on a rotating gantry such that the beam has a center of
rotation about 1.5m above the floor. Usually, the isocenter accuracy is defined within
a 2mm sphere. Because stereotactic radiosurgery depends on optimized accuracy, an
improved system was designed at the University of Florida, by adding a set of bearings
to the stereotactic collimator system and under the patient table. As a result, this
new system achieves mechanical accuracy within 0.2mmÂ±0.1mm for defining the
treatment isocenter of beam delivery.
The tertiary collimators are generally circular and allow improved centering
of the treatment beam. The sizes of these collimators are from 5 to 40mm in 2 to
5mm increments.
2.3 Procedure of Radiosurgery
The procedure of radiosurgery is separated into 5 steps as follows.
(1) Head ring attachment
(2) CT scan
(3) Image Processing
(4) Planning
(5) Treatment delivery
For more detailed information, refer to Friedman et al.[2].
(1) Head ring attachment
The stereotactic radiosurgery treatment requires attachment of a stereotactic
head ring. This ring provides accurate information about the position and the shape
Filament Waveguide
Electron beam
Power supply
\.
ti II
â€¢â€”H
Whr
Secondary (Adjustable)
collimators
Tertiary (Radiosurgery)
collimators
Photon beam tor
radiosurgery
Floor stand with treatment arm above
Figure 21. System of Linac (Friedman et al.[2])
8
Figure 22. Operation of Linac (Friedman et al.[2])
9
of brain tumors from the computer tomography (CT). This ring also helps to insure
that the treatment target is accurately placed at the precise isocenter of the Linac.
(2) CT scan
After attaching the head ring, a series of cut filming is taken. Eighty to 150
CT slices are taken per patient. For radiosurgical treatment planning, getting the
accurate size and shape of the target is very important. Underestimation of the target
size may result in treatment failure. Overestimation of size results in the inclusion of
normal brain tissue within the treatment volume. Misrepresentation of an irregular
target shape may lead to radiation damage of normal brain tissue. The accuracy
of CT scanning depends on the size of the pixels that make up the scan image.
The lateral and AP accuracy of scanning is approximately 0.6mm. But the axial
dimension is determined by slice thickness of the CT scan, which is 1mm in general.
The physicians use the socalled BrownRobertsWells coordinatesor, simply BRW
coordinateswhich are called lateral (Lat), anteriorposterior (AP), and axial (Ax)
coordinates. These are the x, y, and z coordinates, respectively. The direction of
the lateral coordinate is from the left hand side ear to the right hand side ear of the
patient. So the lateral view (or sagittal) is similar to the side view of the head, that
is, views of slices along the lateral axis. This is the same as yzplane views. The
direction of the AP coordinate is from the back side of the head to the face. So the
AP view (or coronal) is similar with the front view of the head, that is, views of slices
along the AP axis. This is the same as xzplane views. The direction of the axial (or
vertical) coordinate is from the neck to the top of the head. So the axial view (or
transaxial) is similar to the sky view of the head, that is, views of slices along the
axial axis. This is the same as xyplane views. (See Figure 24.)
(3) Image Processing
An MRI scan has been done the day before surgery. Using a mouselike device,
the CT and MRI images are transferred to the computer screen and fused. Then,
10
using geometric equations, the computer determines the lateral, AP, and vertical
stereotactic coordinates of each point in each CT slice.
(4) Planning
Physicians use a mouse and their own judgment to â€˜eyeballâ€™ the location and
the size of spheres for the treatment plan.
(5) Treatment delivery
The geometric sphere packing plan, developed as in (4), is combined with a
proprietary dose planning program, and the resulting treatment plan is delivered by
the Linac.
2.4 Making a Sphere by Arcs of Beams
By varying the angle of the gantry and the angle of the table, one can deliver
a radiation beam to the target from any angle within the range of the rotation. The
shape of the common intersection of an arc of beams passing through one isocenter is
a sphere. The neurosurgeons deliver a series of arcs (usually 5 or 9 arcs) to produce
a single isocentered sphere shape. For an ellipsoidal target, they use fewer numbers
of arcs to make a single isocentered ellipsoid shape. (See Figure 23, 24 and 25.)
So, if the target shape is very close to a sphere or an ellipsoid, then the
treatment plan is relatively easy compared to an irregularly shaped target. In that
case, we need to create a geometric treatment plan.
2.5 Sphere Packing Plan
As seen in the previous section, the physicians know how to irradiateto
(eventually)destroy tumors which are shaped like spheres or ellipsoids. For a non
spherical shape of tumor, they try to fill the target with several spheres of different
sizes. This is called the â€˜sphere packingâ€™ treatment plan.
After finding a sphere packing plan, they treat each sphere separately as deÂ¬
scribed in the previous section. So, multiple isocenter radiosurgery planning includes
11
Figure 23. One isocenter plan with five arcs of radiation (Wagner et al.[ll])
12
(270*)
Figure 24. One isocenter plan with nine arcs of radiation (Friedman et al.[2])
111 Q U CQ
13
TARGET
Figure 25. Plan for a sphere shape and an ellipsoidal shape (Friedman et al.[2])
14
the problem of determining the best sphere packing arrangement with which to fill
the target volume. General methods for this treatment plan are iteratively based,
dosimetrically driven algorithms. But these methods require many computations in
order to compute a radiosurgical plan dose distribution, and then to evaluate the
quality of the dose distribution. So geometrically based radiosurgery optimization
has been suggested as a possible alternative means.
However the method the physicians choose relies on human decisions and
experience. Thus, for the same target, different surgeons may produce different plans.
Even the same surgeon, doing the plan twice for the same target, may produce
different plans. And the planning takes a long time, especially for a complicated
target which needs more than 10 spheres. It might take as much as two hours of
planning for a difficult case which needs about 20 spheres. During that time, the
patient has to wait with the head ring attached to his or her head. And most
importantly, even after spending the time to make a plan, many physicians without
sufficient experience, are not sure if the plan is a â€˜goodâ€™ one.
Therefore, we provide an automated sphere packing method for the treatment
plan (see Wagner et al.[ll]). This method shows potential to significantly aid the
planning of difficult multiple isocenter cases. Based on tests with irregularly shaped
phantom targets and with a representative sampling of clinical example cases, the
method demonstrates the ability to generate radiosurgery plans comparable to, or
of better quality than, multiple isocenter Linac radiosurgery plans found in other
literature. At the same time, this program always produces the same treatment plan
for the same tumor shape. So it can be used as a â€˜benchmarkâ€™ to compare with other
plans for the given tumor shape. Moreover, this program provides the treatment
plan in a relatively short time. For a very difficult case which needed more than 18
spheres, this program took less than 3 minutes instead of more than the 1.5 hours
15
which were needed when the physicians created the plan using traditional methods.
The algorithm of this program is shown in Figure 26.
In the following section, we explain the â€˜shelling algorithmâ€™ which we used in
this program to get the centers and sizes of spheres for the sphere packing plan.
2.6 Shelling Algorithm
The shelling procedure is best illustrated in Figure 27 to 216. The major
steps of this shelling procedure are as follows.
(1) Transferring the data of the boundary of the target
From the CT scan, each point which is identified from the target contour data
file can be mapped into a three dimensional array, since each CT scan shows the
transaxial view for a specific axial coordinate value. In the three dimensional array,
we assign the lateral, AP and axialcoordinates to be x, y and zaxes, respectively.
For each individual case, there are minimum and maximum values for x, y and z
axes. We assume the minimum value is always one. We get this information from
the Program for Data Reading. (See Appendix A.l.)
Let the target be minimally enclosed in an xrange beginning with one and
ending with Mx, a yrange beginning with one and ending with My, and a zrange
beginning with one and ending with Mz. Now we put â€˜safe bumperâ€™ layers surÂ¬
rounding the target. That is, we enlarge the box containing the target to ET =
[0, Mx +1] x [0, My +1] x [0, Mz + 1]. ET stands for â€˜enlarged target box.â€™ We assume
that the dimension of each voxel is lxlxl. Then we will number the voxels in a lexiÂ¬
cographic manner, beginning with 0 and ending with (Mx +1) x (My + 1) x(M* + l) â€” 1,
giving us a total of (Mx + 1) x (My + 1) x (Mz + 1) voxels with which we will work.
Now we want to identify the voxels in ET in two different ways: first, by a set
of coordinates (a, 6, c), and second, by an integer representing a counting procedure.
Here, a is the left hand endpoint of the xvalue of that voxel, b is the left hand
16
endpoint of the yvalue of that voxel, and c is the left hand endpoint of the z
value of that voxel. The second notation for the voxel (a, 6, c) is by some integer
n. The voxels are counted, using a lexicographic ordering beginning with 0 and
ending with (Mx + 1) x (My f 1) x (Mz + 1) â€” 1 which is denoted by nofdata in
the program (see Appendix A.2). Then the voxel denoted by (a, 6, c) is numbered
as c x (Mx + 1) x (My + 1) f a x (My + 1) + b. And it is clear that, for every
0 < i < nofdata, there are unique nonnegative integers, a, 6, c, such that i â€”
c x (Mx + 1) x (My + 1) + a x (My + 1) + b. That is, for every 0 < i < nofdata,
one and only one voxel is assigned. Note that Figure 27 to Figure 214 show the
same transaxial view at different stages of the algorithm. Thus these figures show
the data of a fixed axialcoordinate value.
(2) Assign a status value for each voxel
For every 0 < i < nofdata, the status value for the zth voxel is denoted
by cc[i\. That is, we denote the status value for the voxel with coordinates (a, 6, c),
by cc[i\ where i = c x (Mr + 1) x (My + 1) + a x (My + 1) + b. The following is
the procedure to assign status values. First of all, let cc[i\ = â€”3 for every 0 <
i < nofdata. Then we assign the number, â€”1, as the status value for each voxel
corresponding to any of the target contour data. Next, for every 0 < j < Mz, we
assign cc[j x (Mx + 1) x (My J1)] = 0, because the j x (Mx + 1) x (My + l)th voxel
is the voxel at the left lower corner for every jth transaxial view; that is, the voxel
belongs to the healthy tissue. (See Figure 28.)
For each voxel with cc[i\ = â€”3, we define cc[i\ â€” 0 if any of cc[i â€” 1], cc[i +
1], cc[i â€” (My + 1)], cc[i + (My + 1)] is 0. Note that this method produces zeros for
every voxel in the safe bumper layers. (See Figure 29.)
(3) Shelling the target voxels
We proceed by mathematical induction on j, for
0
 x min{Mx, My, Mz}
17
For i = 0 to (Mx + 1) x (My f 1) x (Mz + 1)  1, if the ith voxel is assigned 1
or â€”3 for its status value, and if any of cc[i â€” 1], cc[i + 1], cc[i â€” (My + 1)], cc[i +
(My + 1)], cc[i  (Mx + 1) x (My + 1)], cc[t + (Mx + 1) x (My + 1)] is equal to j,
then we assign a new status value for the Â¿th voxel as c\i\ = j + 1. Note that Figure
27 to Figure 214 show only a twodimensional figure. But this procedure is held
in a threedimensional array. This process is repeated until the deepest lying voxels
have been identified. The deepest lying voxel in the entire target volume would be
the best location for an isocenter. And at the same time, the status value assigned to
the deepest voxel indicates the radius of the sphere, which is one less than the status
value of the voxel. This completes our induction. (See Figure 210.)
(4) Removing the largest sphere
At this moment, we might have several different voxels which have the same
largest status value. So we choose the smallest numbered voxel as the center of the
first largest sphere to be removed from the target. But in the program in Wagner et
al. [11], we choose the most effective voxel to be the center of the first largest sphere,
by using the â€˜score functionâ€™ which was developed by Tom Wagner. We think of
the resulting sphere in the tumor as â€˜removed,â€™ and therefore we can assign the new
status value zero for these voxels in the sphere. To determine which voxels form the
sphere that is removed, we do the following: Let R = voxel at the center of sphere
to be removed. Let r = status value of R. Then we remove all voxels S with the
property that distance(center of R, center of S) < r â€” 1. (See Figures 211 and 212.)
Now, we proceed by mathematical induction on j, for
0
 x min{Mx, My, Mz}
For i = 0 to (Mx + 1) x (My + 1) x (Mz + 1) â€” 1, if the Â¿th voxel has status value
> j, we reassign a new status value as follows: if any of cc[i â€” 1], cc[i + 1], cc[i â€”
(My + 1)], cc[i + (My + 1)], cc[i(Mx + l) x (My + 1)], cc[i + (Mx + 1) x (My + 1)] is
equal to j, then we assign a new status value for the Â¿th voxel as c[Â¿] = j + 1. This
18
process is repeated until the deepest lying voxels have been identified. (See Figures
213 to 215.)
And we repeat step (4) until we get either 0 or 1 for the status value of each
voxel in ET, the enlarged target box. Because the smallest size of the diameter of
collimators is 5mm, we finish our program when the status values of all the voxels are
zeros and/or ones. (See Figure 216.) The remaining part will be taken care of by
the â€˜score functionâ€™ (see Wagner et al.[ 11 ]). After finishing this procedure, we have
the spheres filling the target. In Figure 217, we see the sphere packing plan within
a target by this method.
For further detailed information about the score function and the automated
sphere packing plan, we refer the reader to Wagner et al.[11] and Wagner[10].
Figure 26. Sphere packing algorithm block diagram (Wagner et al.[11])
19
1
â€¢
â€¢
â€¢
â€¢
â€¢
â€¢
Â«
â€¢
â€¢ 4
Â»â€¢
4
Â»
Â«
â€¢
â€¢
Â«
â€¢
â€¢
â€¢
â€¢
â€¢
â€¢
â€¢
â€¢
â€¢
â€¢
A
â€¢
â€¢
â€¢
â€¢
â€¢
â€¢
â€¢
â€¢
â€¢
â€¢
â€¢
â€¢
â€¢
â€¢
â€¢
â€¢
â€¢
â€¢
â€¢
â€¢
â€¢
â€¢
Â»
â€¢
â€¢
â€¢
â€¢
â€¢
â€¢
â€¢
â€¢
â€¢
O 5 10 15 20 > x
Figure 27. Transferring the data of the boundary of the target
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
1
1
1
1
1
1
1
3
3
3
3
3
3
3
3
3
3
3
3
3
1
1
1
1
1
3
3
3
3
3
1
1
1
1
1
1
3
3
3
3
3
3
3
1
3
3
3
3
3
3
3
3
3
3
3
3
1
1
3
1
1
3
3
3
3
3
1
1
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
1
3
3
3
3
1
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
1
1
3
3
3
1
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
1
1
3
3
1
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
1
1
3
3
1
1
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
1
3
3
3
3
1
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
1
1
1
3
3
3
3
1
1
1
1
3
3
3
1
1
3
3
3
3
3
3
1
1
3
3
3
3
3
3
3
3
3
3
3
1
1
1
1
3
1
1
1
1
1
1
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
1
1
3
3
3
3
3
3
3
3
3
0
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
Figure 28. Status values for the boundary and the initial voxel
20
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
1
1
1
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
1
1
1
3
3
3
3
3
1
1
1
1
1
1
0
0
0
0
0
0
0
1
3
3
3
3
3
3
3
3
3
3
3
3
1
1
3
1
1
0
0
0
0
0
1
1
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
1
0
0
0
0
1
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
1
1
0
0
0
1
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
1
1
0
0
1
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
1
1
0
0
1
1
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
1
0
0
0
0
1
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
3
1
1
1
0
0
0
0
1
1
1
1
3
3
3
1
1
3
3
3
3
3
3
1
1
0
0
0
0
0
0
0
0
0
0
0
1
1
1
1
0
1
1
1
1
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
Figure 29. Status values for the normal voxels
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
ol
0
0
0
0
i 0
0
0
0
1
1
1
1
1
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
1
1
1
2
2
2
2
2
1
1
1
1
1
1
0
0
0
0
0
0
0
1
2
2
2
2
2
3
3
3
3
3
2
2
1
1
2
1
1
0
0
0
0
0
1
1
2
3
3
3
3
4
4
4
4
4
3
3
2
2
3
2
2
1
0
0
0
0
1
2
3
4
4
4
4
5
5
5
5
5
4
4
3
3
4
3
2
1
1
0
0
0
1
2
3
4
5
h
5
6
5
5
(S
Â©
5
5
4
4
5
4
3
2
1
1
0
0
1
2
3
4
4
5
5
5
4
4
5
5
5
5
5
5
4
4
3
2
1
1
0
0
1
1
2
3
3
4
4
4
3
3
4
4
4
4
4
4
3
3
2
2
1
0
0
0
0
1
2
2
2
3
3
3
2
2
3
3
3
3
3
3
2
2
1
1
1
0
0
0
0
1
1
1
1
2
2
2
1
1
2
2
2
2
2
2
1
1
0
0
0
0
0
0
0
0
0
0
0
1
1
1
1
0
1
1
1
1
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
~cT
0
Figure 210. Shelling the target voxels
21
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
1
1
1
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
1
1
1
2
2
2
2
2
1
1
1
1
1
1
0
0
0
0
0
0
0
1
2
2
2
2
2
3
3
3
3
3
2
2
1
1
2
1
1
0
0
0
0
0
1
1
2
3
3
3
3
4
4
4
4
4
3
3
2
2
3
2
2
1
0
0
0
0
1
2
3
4
4
4
4
5
5
5
5
5
4
4
3
3
4
3
2
1
1
0
0
0
1
2
3
4
5
5
5
6
5
5
6
6
5
5
4
4
5
4
3
2
1
1
0
0
1
2
3
4
4
5
5
5
4
4
5
5
5
5
5
5
4
4
3
2
1
1
0
0
1
1
2
3
3
4
4
4
3
3
4
4
4
4
4
4
3
3
2
2
1
0
0
0
0
1
2
2
2
3
3
3
2
2
3
3
3
3
3
3
2
2
1
1
1
0
0
0
0
1
1
1
1
2
2
2
1
1
2
2
2
2
2
2
1
1
0
0
0
0
0
0
0
0
0
0
0
1
1
1
1
0
1
1
1
1
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
Figure 211. Choosing the largest sphere
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
0
1
1
1
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
0
0
0
0
0
0
0
2
1
1
1
1
1
1
0
0
0
0
0
0
0
1
2
0
0
0
0
0
0
0
0
0
2
2
1
1
2
1
1
0
0
0
0
0
1
1
2
0
0
0
0
0
0
0
0
0
3
3
2
2
3
2
2
1
0
0
0
0
1
2
3
0
0
0
0
0
0
0
0
0
4
4
3
3
4
3
2
1
1
0
0
0
1
2
0
0
0
0
0
0
0
0
0
0
0
5
4
4
5
4
3
2
1
1
0
0
1
2
3
0
0
0
0
0
0
0
0
0
5
5
5
5
4
4
3
2
1
1
0
0
1
1
2
0
0
0
0
0
0
0
0
0
4
4
4
4
3
3
2
2
1
0
0
0
0
1
2
0
0
0
0
0
0
0
0
0
3
3
3
3
2
2
1
1
1
0
0
0
0
1
1
1
0
0
0
0
0
0
0
2
2
2
2
2
1
1
0
0
0
0
0
0
0
0
0
0
0
1
1
0
1
0
1
1
1
1
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
Figure 212. Removing the largest sphere
22
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
0
1
1
1
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
0
0
0
0
0
0
0
1
1
1
1
1
1
1
0
0
0
0
0
0
0
1
1
0
0
0
0
0
0
0
0
0
1
2
1
1
2
1
1
0
0
0
0
0
1
2
1
0
0
0
0
0
0
0
0
0
1
2
2
2
3
2
2
1
0
0
0
0
1
2
1
0
0
0
0
0
0
0
0
0
1
2
3
3
4
3
2
1
1
0
0
0
1
1
0
0
0
0
0
0
0
0
0
0
0
1
2
3
4
4
3
2
1
1
0
0
1
2
1
0
0
0
0
0
0
0
0
0
1
2
3
4
4
4
3
2
1
1
0
0
1
2
1
0
0
0
0
0
0
0
0
0
1
2
3
4
3
3
2
2
1
0
0
0
0
1
1
0
0
0
0
0
0
0
0
0
1
2
3
3
2
2
1
1
1
0
0
0
0
1
1
1
0
0
0
0
0
0
0
1
2
2
2
2
1
1
0
0
0
0
0
0
0
0
0
0
0
1
1
0
1
0
1
1
2
2
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
Figure 213. Shelling the remaining target voxels
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
0
1
1
1
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
0
0
0
0
0
0
0
1
1
1
1
1
1
1
0
0
0
0
0
0
0
1
1
0
0
0
0
0
0
0
0
0
1
2
1
1
2
1
1
0
0
0
0
0
1
2
1
0
0
0
0
0
0
0
0
0
1
2
2
2
3
2
2
1
0
0
0
0
1
2
1
0
0
0
0
0
0
0
0
0
1
2
3
3
4
3
2
1
1
0
0
0
1
1
0
0
0
0
0
0
0
0
0
0
0
1
2
3
4
4
3
2
1
1
0
0
1
2
1
0
0
0
0
0
0
0
0
0
1
2
3
4
4
4
3
2
1
1
0
0
1
2
1
0
0
0
0
0
0
0
0
0
1
2
3
4
3
3
2
2
1
0
0
0
0
1
1
0
0
0
0
0
0
0
0
0
1
2
3
3
2
2
1
1
1
0
0
0
0
1
1
1
0
0
0
0
0
0
0
1
2
2
2
2
1
1
0
0
0
0
0
0
0
0
0
0
0
1
1
0
1
0
1
1
2
2
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
Figure 214. Choosing the largest sphere of the remaining target
23
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
0
1
1
1
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
0
0
0
0
0
0
0
1
1
1
1
1
1
1
0
0
0
0
0
0
0
1
1
0
0
0
0
0
0
0
0
0
1
2
1
1
2
1
1
0
0
0
0
0
1
2
1
0
0
0
0
0
0
0
0
0
1
2
2
1
2
2
2
1
0
0
0
0
1
2
1
0
0
0
0
0
0
0
0
0
1
1
1
0
1
1
2
1
1
0
0
0
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
2
1
1
0
0
1
2
1
0
0
0
0
0
0
0
0
0
1
0
0
0
0
0
1
2
1
1
0
0
1
2
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
0
0
0
0
1
1
0
0
0
0
0
0
0
0
0
1
0
0
0
0
0
1
1
1
0
0
0
0
1
1
1
0
0
0
0
0
0
0
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
0
1
0
1
1
2
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
Figure 215. Removing the second sphere
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
0
1
1
1
1
1
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
0
0
0
0
0
0
0
1
1
1
1
1
1
1
0
0
0
0
0
0
0
1
1
0
0
0
0
0
0
0
0
0
1
0
1
1
0
1
1
0
0
0
0
0
1
0
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
0
1
0
0
0
0
0
1
0
0
0
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
0
0
1
0
1
0
0
0
0
0
0
0
0
0
1
0
0
0
0
0
0
0
0
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
0
0
0
0
0
1
0
0
0
0
0
1
1
1
0
0
0
0
1
1
1
0
0
0
0
0
0
0
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
0
1
0
1
0
0
0
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
Figure 216. Removing all the spheres
24
Figure 217. A sphere packing plan for a brain tumor (Wagner et al.[ll])
CHAPTER 3
A CLASSIFICATION OF UNLABELED TREES
3.1 Introduction
In this chapter, we obtain a code to denote an unlabeled tree. By means
of this code, we classify unlabeled ntrees. One may take an unlabeled ntree to
be the geometric realization of a connected, finite, simplicial complex without any
simple closed curves. In particular, we call a tree an ntree if and only if it is a tree
with n vertices (that is, (n â€” 1) edges). Our code assigns a unique, ordered, â€˜perfect
sequenceâ€™, pf(T) = (cfi, d2,..., dn), to each unlabeled ntree, T. And, conversely,
given an ordered sequence of n integers satisfying certain properties, it is the perfect
sequence of exactly one unlabeled ntree. Thus, we have our main theorem, the Tree
Classification Theorem (Theorem 3.5.5). We then show how to reconstruct the tree
shape from a given perfect sequence, by hand.
Our work includes an algorithm and a Mathematica program (see Appendix
C.l) that produce a list of all the perfect sequences for all possible ntrees, thus also
producing the number of ntrees, for any given n. For some examples, see Appendix
C.2. Since our algorithm and program are based upon our main theorem, we know
that all possible unlabeled ntrees are produced by the program. Thus the program
gives us, for each positive integer n, a complete list of all unlabeled ntrees.
Given n > 0, with sufficient computer power and time, our program will
list all the unlabeled ntrees, thus telling us how many unlabeled ntrees there are.
However, we do not have a simple formula that tells us how many unlabeled ntrees
there are. This remains an open problem.
25
26
The Literature contains work regarding the classification of labeled graphs
and labeled trees. In LovÃ¡sz[6], the Priifercode is used to determine the number of
labeled ntrees (see the Appendix D.2 and D.3). In Sloane[8], Sloane shows a table
for n < 26. However, Sloane references Harary for the method he uses. In Harary[5],
there is a simple formula for the total number of labeled graphs with n vertices (see the
Appendix D.l). There is also a formula for the total number of unlabeled ntrees, but
it requires knowing the number of rooted ntrees, which is calculated inductively. And,
even though Harary[5] contains ntree shapes up to n = 10, there is no explanation
of how to get the shapes of these trees, while our work does show how to get these
tree shapes from our code.
This chapter is organized as follows:
Section 1: Introduction
Section 2: Definitions
Section 3: Vertex Sequence
Section 4: Degree Sequence
Section 5: Perfect Sequence and Main Theorem
Section 6: Obtaining Perfect Sequences
Section 7: Extended Partition and the Algorithm
3.2 Definitions
Most of the necessary definitions in this section are from Harary[5]. We refer
the reader to Harary[5] for easy reference.
Definition 3.2.1 A graph G = (V(G), E(G)) consists of a vertex set V(G) =
{t>i, and an edge set E(G) = {e i,..., em} C {{u, u} u,v G V(G)} in which
each edge is an unordered pair of two distinct vertices u and u, and the edge {u,u}
is said to join u and v. For an edge {u,u} we say that u and v are adjacent
vertices and the vertex u and the edge {u,u} are incident with each other, as are
27
v and {u,v}. From now on, V(G) denotes the number of vertices and E(G)\
denotes the number of edges. For a vertex u â‚¬ F(G), the vertex set
A{u) = {u Â£ G\v is an adjacent vertex of u}
is called the adjacent set of u. Note that u # A(u).
Definition 3.2.2 A walk is an alternating sequence (i>0, ei, v\, e2,..., e*, Vk) of
vertices and edges s.t. u,_i / ut for i â€” 1,2, and e, = {u,_i,n,} for i =
1,2,..., fc. If there is no confusion, a walk can be denoted by a sequence of vertices
only in the path like (iq, v2,..., Vk) . A cycle is a walk (n0, ei, iq, e2,..., e*, n*) with
A: > 3 in which Vo = Vk is the only vertex repeated. A path is a walk with no
repeated vertex. A (u,v)path is a path with the first vertex u and the last vertex
v. The length of the (u,u)path, l(u,v), in a graph is the number of edges in the
(u,i>)path. The distance d(u,v) between two vertices u and v is the length of a
shortest path joining them, if any; otherwise d(u,v) â€” oo. Note that d(u,u) = 0. A
graph G is connected iff every pair of vertices are joined by a path. A graph having
no cycle is acyclic. Note that an acyclic graph can be disconnected. A forest is an
acyclic graph. A tree is a connected acyclic graph. An ntree is a tree with n vertices.
Definition 3.2.3 We say that a graph G is isomorphic to a graph H if and only
if there exists a bijection / : V{G) â€”> V(H) such that {u,i>} Â£ E(G) if and only if
{Hu)J(v)}â‚¬E(H).
Definition 3.2.4 An unlabeled tree is an isomorphism class of trees.
We are interested in the total number of unlabeled ntrees and their shapes.
For example, Tx = ({1,2,3}, {{1,2}, {2,3}}) and T2 = ({1,2,3}, {{2,1}, {1,3}})
are different trees, but there is an isomorphism / : V'(Tj) > V(T2) by /(1) =
2, /(2) = 1, /(3) = 3. Thus they are the same unlabeled tree, so we want to count
them as one unlabeled 3tree. From now on, a tree implies an unlabeled tree unless
otherwise specified.
28
Definition 3.2.5 The degree of a vertex v, defined by d(v), is the sum of the
number of edges incident with v. A vertex v is called an endvertex if d{v) â€” 1.
We will use the geometric, that is, topological realization of a graph interÂ¬
changeably with â€˜graphâ€™ as defined above.
Proposition 3.2.6 [5] The following statements are equivalent for a graph T :
(1) T is a tree.
(2) Every two vertices of T are joined by a unique path.
(3) T is connected and V(T) = \E(T)\ + 1.
(4) T is acyclic and V(T) = Â¿Â£(T) + 1.
Note that there are exactly n vertices and n1 edges in an ntree.
Proposition 3.2.7 (Degree sum formula) [5] If G is a graph with the vertex
degrees du. then Eâ€žGV(G) = 2\E(G)\
Corollary 3.2.8 For an nTree T we have
ET=i d{vi) = 2(n  1) = 2\E(T)\ = 2(\V(T)\  1)
where is a vertex of T for i = 1,2,..., n.
Because a nontrivial tree satisfies Ylvev(T)d(v) â€” 2\E(T)\ â€” 2(\V(T)\ â€” 1),
there are at least two vertices with degree 1. Thus we have the following theorem as
a corollary.
Theorem 3.2.9 [5] Every nontrivial tree has at least two endvertices.
3.3 Vertex Sequence
In this section, we define the vertex sequence of a given ntree.
Definition 3.3.1 Let T = (V, E) be an ntree. If d(y) = m, then there are m
components of T â€” {u}. A branch of v (or simply a vbranch) is a subtree which
is the closure of a component of T â€” {u}. We call the set of vertices in a ubranch
a vbranch vertex set. If u 6 A(u), then the wbranch vertex set containing u is
called the (v,u)branch vertex set and denoted by B(v,u) (or B(u)). We call the
29
set B'(v,u) = B(v,u) â€” {u} the (t>,ii)subbranch vertex set, which is denoted by
B\u) if there is no confusion. The subtree consisting of B'(v,u) as the vertex set
and {{s,t} e B\v,u) x B'(v,u) {s,t} â‚¬ E(T)} as the edge set is called the (i>,u)
subbranch.
Example 3.3.2 Let T be the 10tree shown below.
s
t
u
p
<7
r
V
w
X
y
Figure 31
Since d(v) = 4, there are 4 components of T â€” {v}. Since A(v) = {r, t, to, y},
we have that
B{v,r) = {v,r,s,q,p},
B(v,t) = {M},
B(v,w) â€” {v, io, u, a:},
B{v,y) = {v,y},
B'(v,r) = {r,s,q,p}
B'(v,t) = {Â¿}
B'(v, w) = {w, u,
B'(v,y) = {y}.
On the other hand, we have that
B(r,v) = {r,v,t,u,w,x,y}.
That is, B(v, r) / B(r, v). And this is true for any tree except the 2tree. In fact, for
every ntree with n > 2, there are at least 3 vertices, namely u,v, and w. Without
loss of generality, let u and v be vertices adjacent to each other, and let the (io, v)
path contain the vertex u. Then w Â£ B(v,u) and w ^ B(u,v).
Definition 3.3.3 A section of a sequence is a subsequence consisting of consecutive
terms of the sequence.
Definition 3.3.4 Let T = (V, E) be an ntree with V = {aq, X2,..., xn}. We define
the set, S, of sequences of the form s = (tq, V2,..., vn), where s is a permutation
of the Xi s in V(T), to be those sequences satisfying:
30
(*) for any Vj with 1 < j < n, Vj+\ â‚¬ A(vk) â€” {ni, v2,..., Uj}
where k = max{i â‚¬ {1,2,... ,j} A(v{) â€” {vi,v2,... ,Vj} / 0}.
Then S is called the vertex sequence set of T and each element, s, in S is called a
vertex sequence of T.
Definition 3.3.5 For any v â‚¬ V, we call Sv = â‚¬ 5ui = v} C S the vvertex
sequence set.
Definition 3.3.4 requires that the vertices in a sequence be listed in a particular
way. The condition (*) ensures that all vertices are listed and that each is listed
exactly once. We can choose any vertex for the first term, iq. Then the condition
(*) says that, for the (i+l)th term, we have to choose an unlisted adjacent vertex
of V{, if possible. If every adjacent vertex of u, is already used, then look backward
along the sequence until we come to the first vertex which has an unused adjacent
vertex. Then, for the (i+l)th term, we assign one of the unused adjacent vertices of
that vertex. To get a vertex sequence, we continue this procedure inductively.
Note that every vertex sequence of an ntree has n entries.
Example 3.3.6 We show how to construct a vertex sequence of a given tree. Let
T be the 7tree shown below.
f 9
a b c d e
Figure 32
Since we have freedom to choose the first term, let V\ = b. Then v2 â‚¬ Ã4(Ã¼i) â€”
{ui} = A(b) â€” {6} = {a, c, /} â€” {6} = {a, c, /} = A(b) = A(ui). So v2 must be one of
the elements of A(ui) = A(b) = {a,c, /}. Let v2 = c. Since v3 â‚¬ A(v2) â€” {t>i,u2} =
A{c) â€” {b, c} = {b, d, g} â€” {b, c} = {d,g}, v3 must be either d or g. Let v3 = g.
Then v4 # A(v3)  {vi,v2,v3}, since A(v3)  {vuv2,v3} = {c}  {b,c,g} = 0. On
31
the other hand, we have that A(v2) â€” {t>i,n2, u3} = {&, d,g} â€” {b,c,g} â€” {d} / 0.
Therefore we have to choose d as u4. Then v^ â€” e since A(v4) â€” {iq,v2,u3,14} =
{c, e} â€” {b,c,g,d} = {e}. Since we have
A(u5)  {vi,V2,V3,v4,v5} = {d}  {b,c,g,d,e} =0,
A(n4)  {vi,v2,v3, v4, u5} = {c, e}  {6, c,p, d, e} =0,
A(n3)  {ui,n2,n3,n4,n5} = {c}  {&, c,g, d, e] =0,
A(u2)  {ui,n2,u3,u4,n5} = {M, d}  {&, d, e] =0, and
A(Ã¼i)  {vi,v2,u3,u4,u5} = {a, c, /}  {6, c,g, d, e] ={a,f},
ve must be either a or /. Let ve = /. Then we have that
A{vG)  {vi,v2,v3,v4,v5,ve} = {&}  {b,c,g,d,e,f} =0,
A{v^)  {vuv2,v^vA,vh,vG} = {d}  {b,c,g,d,e,f} =0,
A(v4)  {v1,v2,v3,v4,v5,ve} = {c,e}  {b,c,g,d,e,f} =0,
A(u3) â€” {ui, v2, t73, u4,175, v6} = {c}  {b,c,g,d,e,f} =0,
A(v2){vi,u2,u3,u4,u5,u6} = {6,sr, rf}  {Â¿>,c,Â£,d,e,/} =0, and
A(ni)  {vuv2,v3,v4,v5,v6} = {a, c, /}  {&, c,Â£, d, e, /} = {a}.
Therefore iq must be a; that is, we construct a vertex sequence
(iq,u2,...,iq) = (b,c,g,d,e,f,a).
For a given tree we will get different vertex sequences if we choose different
vertices whenever we have two or more choices. But the sequence t = (6, c, d, e, /, a, i/)
cannot be a vertex sequence of T above. In fact, we have v5 = f. But we have that
A(u4)  {iq,u2,u3,u4} = {d}  {b,c,d,e} =0,
A(v3)  {vuv2,v3,v4} = {c, e}  {6, c, d, e} =0,
A{v2)  {tq, u2, t>3, 14} = {6, d,5}  {6, c, d, e} = {#}, and
4(ui)  {iq,n2,u3,u4} = {a, c, /} â€” {6, c, d, e} ={a,/}.
That is, the sequence t has /, one of the elements of A(rq) â€” {iq, n2, v3, n4}, for U5
instead of g which is an element of A(v2) â€” {iq, n2, u3, tq}. It is a contradiction to
the maximality of k in the definition of vertex sequence.
Proposition 3.3.7 Let (iq, t>2,..., vn) be a vertex sequence of an ntree T. For any
Vj with j > 1 there is unique 1 < k < j such that Vj G A(vk) â€” , v2,..., tq_4}.
Proof First of all, for any j > 1, we have Vj / n, for i = 1,2,... ,j â€” 1 from the
Definition 3.3.4. Assume that there is j > 1 such that
{Vj,Vj+ i,...,Uâ€ž} D ((UfÂ¿i4(Ã¼<)) U {vi,V2,...,Vj_i}) = 0.
32
Then, for any u Â£ {vj,vj+i,...,un} and any w Â£ (UÂ¿_j A(nÂ¿)) U {rq, u2, â– .., Vj},
there is no (u,u;)path in T. That is, T is not connected. It is a contradiction
to the connectedness of a tree. Therefore there is l < k < j such that vj â‚¬
A(vk) â€” {ui, u2, â– â– â– , Vji}. If there are 1 < k\ < k2 < j such that Vj Â£ A(vkl) â€”
{ui,u2,...,Uj_i} and Vj Â£ A(vki)  {ui, v2,.. ., uÂ¿i}, then the two edges, {vj,vkl}
and {vj,vk2}i are elements of E(T). There is a path between vkl and vk2 in the
subtree {{iq, v2,..., Uj_i}, E'} where E' = {{u, u} Â£ E(T)\u, v Â£ {ui, u2,..., *b_i}}.
The tree T contains a cycle consisting of {vj,vkl}, {vj,vk2} and the (vkl, t?fc2)path.
This contradicts the fact that a tree is acyclic. Therefore there is only one l < k < j
such that vj Â£ A(vk) â€” {i>i, v2,..., Uj_x}. T
From the Definition 3.3.4 and the previous lemma we get the following corolÂ¬
lary immediately.
Corollary 3.3.8 Let (ui, u2,..., vn) be a vertex sequence of an ntree. Then
(1) v2 Â£
(2) If d(vj) > 2 for any j = 2,3,..., n â€” 1, then Vj+1 Â£ A{vj).
(3) For any j = 2,3,..., n â€” 1 if d(vj) = 1 and there is i < j such that d(vf) > 3
and d(vk) = 2 for every k = i + 1, i + 2,... ,j â€” 1, then uJ+1 Â£ A(v{) and Vj+i Â£
A(vp), for every p = 1,2,... ,i  l,i + 1,..., j.
Proposition 3.3.9 Let (ui,u2,... ,vn) be a vertex sequence of an ntree and A(ui) =
... ,ulm} with 1 < li < 12 < ... < lm. Then
{v1,v2,...,vn} = Uâ€L1Â£(v1,ult).
Proposition 3.3.10 Let s = (t?x, u2,..., vn) be a vertex sequence of a given ntree
T. If d(vj) = m for some j > 1, then there exist m vertices, Vj,Vj1,Vj2t... ,Vjm_l
such that A(vj) = {uÂ¿, vjl, vj3,..., ujm_1} with i < j < ji < j2 < ... < jm1 â€¢
Proof For any j > 1 there exists unique i < j such that {u,,Uj} Â£ E(T)
by Proposition 3.3.7; that is, Â£ A(vj). So, for any vk Â£ A{vj) â€” {u,}, j < k.
33
Since d(vj) = m, there are m â€” 1 vertices, Vj1,Vj2:.. â€¢ such that A(vj) â€”
{Vi, Vh ,vh,..., ujm_1} with i < j < ji < ji < ... < jm1 â€¢ L
Proposition 3.3.11 Let s = (v\, u2, â– â– â– , vn) be a vertex sequence of a given ntree
T. Let d(vi) = m and A(ui) = {uin vl2,..., uim} be the adjacent set of vu named
so that 1 < li < 12 < ... < lro Then, for any i = 1,2,..., m, the set of vertices in
the subbranch, B'(vi,uiÂ¿), forms a section in s starting with v\t.
Proof Clearly, the section containing Ui, starts with Vil = u2. Let B'(yi,V\f) =
{ui1? Uij+i,.. â€¢, }. Assume that there is a vertex u G V(T) â€” B\vi, UiJ such
that (uijjUij+i,... ,ui 1+j,Â«,...) is a section of s. Since u Â£ B'(v i, Vi,), A(vll+k)~
B'{vuVll) = 0 for any k = 0,1,2, ...,j from the definition. Thus U^=0A(ui1+a;) â€”
B'(v!,Â«!,) = 0. That is, U^.=0A(ui1+fc) D B'(vijUiJ. On the other hand, A(ui1+Â¿) C
B'(v i,uij) forevery k = 0,1,2,..., j. Thus we have that U^.=0A(ull+fc) C B'(vi, ui,).
Therefore U^._0yl(ui1+fc) = B'(vThat is, j = mi. Hence B'{vi,ui,) forms
a section in s. Since B'{yi,^,) is a subbranch of v\, Ui1+TOl+i 0 A(u) for any
u e B'(vi,vij). Thus ull+mi+1 G A(vi). That is, ui1+mi+1 = vl2. For any i =
1,2, ...,m â€” 1, let Â¿T(di,Ui.) form a section in s and Ui,+mi+i = By a simiÂ¬
lar method, B'(yi,Uil+1) forms a section in s and Uii+1+rn,+1+i â€” nit+2. L
The section formed by a subbranch B'(yi,ui.) is denoted by (B'(vi,UiÂ¿)).
From the previous proposition, we have the following corollaries.
Corollary 3.3.12 Let s = (ui, u2,..., vn) be a vertex sequence of a given ntree
T. Let d(vi) = m and A{vi) = V\2,..., Uim} be the adjacent set of Ui, named
so that 1 < lx < 12 < ... < lm. The section (B'(vi,fp)) precedes the section
(B'(v i,vlj+l)) in s for i â€” 1,2, ...,m â€” 1, and there is no other vertex between
these two sections.
Corollary 3.3.13 Let (ui, u2,..., vn) be a vertex sequence of a given ntree T.
For any j > 1 with d(vj) = m > 2 let A{vf) â€” {ut, Vjl,Vj2,..., Vjm_1} where i <
34
j < ji < ji < â€¢â€¢â€¢ < jmâ€”i â€¢ Then (B'ivjiVjt)) = (vjt, vji+1,..., %+1)i) for any
i = 1,2,..., m â€” 1. In particular, we have that j\ = j + 1.
Note that = {iq, vÂ¿m_uVÂ¿m_i+i,..., uJmi} is the (iq, )branch
vertex set for some jm > jm1
Corollary 3.3.14 Let (iq, u2,..., vn) be a vertex sequence of a given ntree. If
i < j with vj â‚¬ j4(i>,) and Vj Â£ A(vk) for every k = i + 1, i + 2,..., j â€” 1, then
vp Â£ A(vk) for every p > j and k = i + 1, i + 2,..., j â€” 1.
For any given vertex sequence s = (vi,U2,..., vn) and any vertex Vj ^ tq, v\
belongs to a branch of vj. We want to give a special name to this branch.
Definition 3.3.15 Let s = (iq, v2> â€¢ â€¢ â– , vn) be a vertex sequence of a given ntree
T. For any j > 1 the root branch of Vj in s is the branch of Vj containing iq.
Note that there is no root branch for iq.
Proposition 3.3.16 Let (iq, u2,..., vn) 6 S be the vertex sequence set of a given
ntree T. For any 1 < j < n, if iq+i ^ then vj is an end vertex.
Proof If Vj+1 ^ A[vj) for some 1 < j < n, then vj has a unique edge with a
vertex, uÂ¿, with i < j by Proposition 3.3.7. That is, d(vj) â€” 1. Therefore Vj is an
end vertex. X
Proposition 3.3.17 Let (tq, u2,..., vn) â‚¬ S be a vertex sequence set of a given
ntree T. For any 1 < j < n there is an integer jm such that
{ui, u2, â€¢ â€¢ â€¢, Vj} U {vjm,vjm+l,..., vn}
is the vertex set of the root branch of Vj.
Proof Let l < j < n and d(vj) = m. Then, by Proposition 3.3.10, there exist
m different vertices, uÂ¿, Vj,, tq2,..., Vjm_l such that A{vj) = , Vj,, ,..., Vjm_l}
with i < j < ji < j2 < ... < jm1 From Corollary 3.3.13, we have that
(Ufc=i Tl (vj,Vjk)^ â€” (i>jj, Ujj+i,..., vjmi).
35
That is, the union of all the subbranches of tq except the root branch is a section
from Ujj to Therefore the set of vertices in the root branch is {tq, tq,..., tq}U
{Vjm1Vjm+ L
Note that, if jm > n, then the root branch is {tq, tq,..., tq}.
Corollary 3.3.18 Let s = (tq, tq,..., vn) 6 S be the vertex sequence set of a given
ntree T. Let j / 1. Then d(vj) = 1, that is, Vj is an end vertex of T, if and only
if the set of vertices of the root branch of Vj is V(T) = {tq, tq, â€¢ â€¢ â€¢, fn}.
Proposition 3.3.19 Let s = (tq, tq,..., vn) be a vertex sequence of a given n
tree T. For any vertex vj with j > 1 the (vi,vj)path in T is a subsequence of
(Vi,V2,...,Vj).
Proof For any vertex Vj with j > 1 there is a unique vertex Vkp with kv < j
such that Vj â‚¬ A(vkP) For the vertex Vkp there is another unique vertex Vkp_,
with A;p_i < kp such that Vkp â‚¬ A(vkp_i). Then we will have that iq is the unique
adjacent vertex of a previous vertex â– Therefore the (tq, Uj)path is a subsequence
(vi,Vkl,...,Vkp_l,Vkp,Vj) of (vi,v2,...,Vj). 1
Proposition 3.3.20 For any two vertex sequences Â¿j and s2 in the vvertex seÂ¬
quence set Sv = {s â‚¬ *Sjvi = v} the root branch vertex sets of a vertex x ^ v\ â€” v
in si and in s2 are same.
Proof Let S\ = (tq, v2,..., vn) and s2 = (iq, u2,..., un) be two different vertex
sequences of a given tree T with tq = u\. Let x be an arbitrary vertex in T
with x iq. Then there are two integers k and / such that tq = x = uÂ¡. Let
A(x) = {x\,x2,... ,xm\,t/}. Since T is a tree, there is a unique (tq,x)path in T.
Then one of the elements of A{x) is in the (tq, x)path. Without loss of generality,
let y be in the (tq, a:)path. From the proof of Proposition 3.3.17 we have that
(uJLV B'(x, xr)) =
Vk\â€¢ â€¢ â€¢ 5 Vk+r}
(U^i'B'tx.x,,)) = (U ^B'(u,,xr))
â€” (Ul+11 Ul+2, â– â€¢ â€¢ 1 Ul+r)
36
for some r. Therefore we have that
{iq, ^2? â€¢ â€¢ â€¢ , Vk} U {Ufcjr+l , â€¢ â€¢ â€¢ , tyi} â€” {^1, ^2, â€¢ â€¢ â€¢ , M/} U {Â«/+r+l, â€¢ â€¢ â€¢ , ^n}â€™
That is, the root branch vertex sets of x in sj and in s2 are same. J_
Example 3.3.21 Let T be an 8tree as shown below.
v w
p q r s t u
Figure 33
Clearly, Si = (q,r,s,t,u,w,v,p), s2 = (q,p,r,s,w,t,u,v) e Sr = {s â‚¬
Sjiq = q}. Let
(vi,v2,... ,v8) = {q,r,s,t,u,w,v,p) and (iq, u2,..., u8) = (q,p,r,s,w,t,u,v).
Then the root branch vertex set of s is {p, q, r, v}. That is, {p,q,r,v} = {q, r} U
{v,p} = {ri,u2} U {v7,u8} and {p,q,r,v} = {
Therefore the root branch vertex sets are same but the position might be different.
On the other hand, d(w) = 1 and the unique adjacent vertex of w â€” v&
is s = V3. Thus s = V3 is in the root branch of w = ve. Therefore the root
branch vertex set of w = vÂ§ is {iq, v2, V3, U4, v5, u6} U {u7, u8} = {tq, v2,..., u8} =
{q,r,s,t,u,w,v,p}. That is, the root branch vertex set of an end vertex is Vâ€™(T)
itself.
3.4 Degree Sequence
In this section, we define the degree sequence of a vertex sequence.
Definition 3.4.1 Let S be the vertex sequence set of an ntree T. For any verÂ¬
tex sequence s = (tq, v2,..., vn) â‚¬ S, we call Da = (di, d2,..., dn) the degree seÂ¬
quence of s where cfi = d{vj) and di = d(v{) â€” 1 for i > 2. We call the set
D = â‚¬ 5} the degree sequence set of S. And, for any vertex v â‚¬ V, we
37
call Dv â€” {DÂ¡\s â‚¬ S',,} C D the vdegree sequence set of Sv. For any section
(Bfavjt)) = (vj,vji,vji+u...,vji+li) of s, formed by the (vj, UjJbranch vertex
set B(vj,vJt), we define Bd(vj,vjt) = {dj, dj{, dji+i,..., dJi+1i} as the (vj, vJt)
branch degree set where d^ â€” d(vk) â€” 1 for k = j, ji, ji + 1, â– â– â– , ji+i â€” 1. And we define
the (vj,Vj^branch degree section by (Bd(vj, Vjt)) = {dj,djiidji+1,..., dj1+1i) which
is denoted by (Bd(vj{)) if there is no confusion. In the same way as in Definition
3.3.1, we define {Bd'(vjt)) = (Bd'(vj,vh)) = (djitdji+1,... A+1i) .
Example 3.4.2 Occasionally, it may happen that there are two or more vertex seÂ¬
quences of a given tree with the same degree sequence. In the following example, there
are several pairs of vertex sequences with the same degree sequences, respectively.
p r s i
Figure 34
Since there are 5 vertices in V(T), there are exactly 5! different seqnences of vertices
of T having length 5 and using each vertex only once. But only 16 of them satisfy
the condition (*) in Definition 3.3.4. Therefore the vertex sequence set, S, of the
given tree, T, consists of the following sequences
Â«1 = (P,r,q,s,t), s7 = (r,p,9,5,i), s\3 = (s,r,p,q,t),
Â¿2 = (p,r,s,t,q), s8 = (r,p,s,t,q), s\4 = {s,r,q,p,t),
Â¿3 = (q,r,p,s,t), s9 = (r,q,p,s,t), sis = {s,t,r,p,q),
s4 = (q,r,s,t,p), sIo = (r,q,s,t,p), s\6 = (s,t,r,q,p),
Ss = (t,s,r,p,q), six = (r,s,t,p,q),
s6 = (t, s, r, q, p), sl2 = (r, s, t, q, p).
Then the degree sequences of these vertex sequences are
Afi = <1,2,0,1,0), 0* = <3,0,0,1,0), Dsh =<2,2,0,0,0),
Ds2 = (1,2,1,0,0), D3s = <3,0,1,0,0), DaU =<2,2,0,0,0),
Ds3 = (1,2,0,1,0), Ds9 = (3,0,0,1,0), AÂ¡Ãs = (2,0,2,0,0),
A4 = (1,2,1,0,0), Aio = (3,0,1,0,0), AfÂ« = (2,0,2,0,0),
As = (l,l,2,0,0), Dsh =(3,1,0,0,0),
Dse = (1,1,2,0,0), Dah = (3,1,0,0,0).
38
Now we have that
Ds\ = Ds3,
D=D
U Sg â€” U s 10 5
Dsâ€˜i â€” D s\ i
Dsh ^S12)
Dss â€” Ds~6 ,
Ds{4 i
Ds[3
D s7 â€” Dsg,
Dsi 5 DÂ¡Â¡i 6
That is, for some different vertex sequences, the degree sequences might be same.
And, if the degree sequences are same, then the trees denoted by these degree seÂ¬
quences are isomorphic.
Remark: The sum of the terms in any degree sequence of a given ntree is the
same, namely, n â€” 1. It is 4 in the above example, and it is same as the number of
edges in the tree. See Theorem 3.4.6 below. Since d\ = d(vi) and d, = d(u,) â€” 1 for
i = 2,3,..., n, we have the following proposition.
Proposition 3.4.3 Let DÂ¿ â€” (Â¿i, Â¿2, â€¢ â€¢ â€¢, dn) be a degree sequence of a vertex seÂ¬
quence s of an ntree T. Then 0 < d\ < n â€”1, and 0 < dt < n â€” 1 for i = 2,... ,n.
Proposition 3.4.4 If dj ^ 0 for any 1 < j < n, then Vj+\ is an adjacent vertex
of Vj.
Proposition 3.4.5 Let s = (uj, u2,..., vn) be a vertex sequence of an ntree T and
DÂ¡ = (di, d2,..., dâ€ž) the degree sequence of s. For any j â‚¬ {1,2,..., n â€” 1} with
dj â€” m > 1, if the set {itÂ¿+1, Vj+2, â€¢.., Vk} is the collection of all the vjsubbranches
except the root subbranch, then k > j + m.
Proof Since dj = m = d(vj) â€” 1, from Proposition 3.3.10, the vertex Vj has m
subbranches besides the root subbranch. Thus there are at least m endvertices in
the collection of those subbranches other than the root branch. Therefore at least m
terms in (dj, dj+i,..., dk) are 0. Hence k > j + m. J_
Theorem 3.4.6 (First Reduced Degree Sum Formula) For any degree sequence
Ds = (di, d2,..., dn) of a vertex sequence s (E S of a given ntree T we have
ELi di = n\ = \E\
and
+ EIUW â€” 1) = 0.
39
Proof d\ is the degree of a vertex, and the remaining n â€” 1 terms are each one
less than the degree of a vertex. By the degree sum formula (Proposition 3.2.7), it is
true that ]C"=1 d(wÂ¿) = 2\E\. So
Â£â€= i di = di + di
= d\v i) + E"=a(d(t,01)
= d(vl) + Â£"= 2 d(vi)  Â£â€=2 1
= + Â£IU Â«M) (Â«1)
= Â£â€=i Â¿(Â«O ("1)
= 2i?  (n  1)
= 2(n â€” 1) â€” (n â€” 1)
= n â€” 1 = Â£.
Then we have that
di + ET=2(*  !) = + Â£"=2 di â€œ Â£"=2 1
= Â£?=i^ (Â»i)
= (n 1)  (n 1)
= 0.1
Corollary 3.4.7 For any degree sequence, D~s = (di, d2,..., dn) of a vertex sequence
s â‚¬ S of a given ntree T, we have
di + Â£Â¿=2, d,jto(di â€” 1) = Â£,2, Â¿i =o 1
Proof From the first reduced degree sum formula we have that di + Â£tn_2(d,' â€” 1) = 0.
Then
0 = d\ + Â£"=2(d, â€” 1)
= Â£"_2i d,jto(di ~ 1) + Â£"=2, d,=o(di  1)
= dl + C=2, d,/oK'  1) + Â£"=2,
= d\ 1 Â£,â€”2 diÂ¿o(di â€” 1) â€” Â£Â¿=2, d,=o !â€¢
Thus we get the result. T
Note that di > 0 for any i = 1,2,..., n. So d, â€” 1 < 0 if and only if di = 0.
Corollary 3.4.7 shows that, in a degree sequence D (efi, d2,..., dn), the total
number of 0 terms is the same as d\ + X^â€=2 (di â€” 1). Therefore there are exactly
d,/0
* + Â£[ l_2 d,^o(di ~ 1) endvertices in the given tree. Because of this, we have the
following proposition.
Proposition 3.4.8 For any degree sequence = (d1? d2,..., dn) of a vertex seÂ¬
quence s 6 S of a given ntree T, we have
40
di + EÂ¿=2(dÂ¿ ~ 1) 7^ 0
/or euen/ 1 < q < n.
Proof Assume that there is q < n such that d\ + â€” 1) = 0. Since DÂ¡ =
(Â¿i, Â¿2,dn) is a degree sequence of a vertex sequence s Â£ S of the given ntree
T, we have that vq+i Â£ A(up) for some p < q. Thus at least one of the endvertices
counted by vp is one of vq+i, vg+2,..., vn. That is, in the section (di: d?,..., dq),
we have at least one less 0 terms than d\ + 0(dÂ¿ â€” 1)? that is,
d\ +
EU
diÂ¿ o(dÂ«
1)
EU, di
=0i
= di
+ EL
z2(di â€” 1)
 di
+ E?=
:2, d,Â¿o(di
1)
+ EU,
d,= 0
= di
+ EL
=2, diÂ¿o(di
1)
+ EU,
di =0
= di
+ EL
=2, diÂ¿o(di
1)
 EU
d,=0
> 1.
This is a contradiction. Therefore we have d\ + ^Â¿_2(dt â€” 1)^0 for every 1 < q <
n. X
Theorem 3.4.9 (Second Reduced Degree Sum Formula) Let T â€” (V, E)
be an ntree. Let DÂ¡ = (d\, Â¿2, â€¢ â€¢â€¢, dn) be the degree sequence of a vertex sequence
s Â£ S of the given ntree T. For any jk > 1, let B'[vjk) â€” Ã¼j*+i,uj*+ii}
be a (vj,Vjk)subbranch vertex set. Then we have
4+ a=o
and
djk T ELj*+i(dÂ¿ â€” 1) 7^ 0
for every jk < q < jk+l.
Proof Let E{B'(vjk)) = {{u,o} Â£ E(T)\u,v Â£ B'(vjk)}. Then
%) = (%j,Â£(%j))cr
is also a tree, and (B'(vjk)) = (vjk, Vjk+i,..., %+1i) is a vertex sequence of the
tree T(vjk). Note that d(vjk) = djk +1 in T. Since Vj $ B'(vjk), the edge {vj,Vjk}
is not in the edge set E(Bâ€™(vjk)). Thus we have that d(vjk) = djk in the subtree
41
T(vjk). Then the result comes from the first reduced degree sum formula (Theorem
3.4.6) and Proposition 3.4.8. JL
Corollary 3.4.10 Let s = (tq, i?2, â– â– ., vn) be a vertex sequence of an ntree T and
Ds â€” (Â¿1,^2, â€¢ â€¢ â€¢ ,dn) the degree sequence of s. For any dj > 0, if the subsequence
(vj, Vj+i,..., vp) is the collection of all the Vjbranches except the root subbranch of
Vj, then
dj + Y7i=j+i(di  1) = 0
and
di + HLj+Adi ~ 1) Â± 0
for every j < q < p.
Proof For j â€” 1, they are true by the first reduced degree sum formula (Theorem
3.4.6) and Proposition 3.4.8. Let j > 1 and Vj 6 A(vt) for some t < j.
(i) If dj = 0, then d(vj) = 1, that is, vj is an end vertex. Since j > 1, Vj v
Thus the root branch is the whole tree by Corollary 3.3.18. Therefore (Vj) is the
collection of all the Ujbranches except the root subbranch and dj + Y^=J+i(di â€” 1) =
dj = 0.
(ii) If dj > 0, then the subsequence (vj, Vj+1,..., vp) which is the collection of all
the Ujbranches except the root subbranch, is the (uÂ¿, Uj)subbranch, since Vj â‚¬ A{vf).
By the previous theorem, we get dj + J2Pi=:+i(di â€” 1) = 0. Thus the collection of all
the Ujbranches except the root subbranch of Vj is the (vt, Uj)subbranch. Therefore,
by the second reduced degree sum formula, we get dj + YH=j+i(di â€” 1) / 0 for every
j < q < p. 1
Not all of the nonnegative integer sequences with n entries can be a vertex
sequence of an ntree. There is a condition to being a degree sequence of an ntree.
Definition 3.4.11 Let D â€” (ofi, d2, â€¢ â€¢ â€¢, dn) be a nonnegative integer sequence.
We say that D satisfies the degree sequence condition, briefly DSC, if and only if it
satisfies all of the following:
42
(a) di > 0.
(b) Â£"=1^ =nh
(c) For any j < n with dj = m > 0, there are exactly m + 1 integers, j < ji =
j + 1 < ji < â€¢ â€¢ â€¢ < jrn < jm+1(= n + 1 if j = 1), such that
4 + YÃkJjl+\(di  1) = 0 for k = 1,2
(d) For any jk with djk = l > 0, there are exactly l + 1 integers, jk < jk, =
jk + l < jk2 < â€¢ â€¢ â€¢ < jk, < ifc(I+1) = jfc+i such that
+ 71(rfÂ¡i)= 0 for p = 1,2,...,/.
Theorem 3.4.12 LeÃ D â€” (dj, d2? â€¢ â€¢ â€¢, ^n) a nonnegative integer sequence. Then
D is a degree sequence of a vertex sequence of an ntree if and only if D satisfies
DSC.
Proof [=>â€¢] This follows from Proposition 3.4.3, Theorem 3.4.6 and Theorem 3.4.9.
[<=] Let V = {ui,Ã¼2, ... ,uâ€ž} be a set of n different vertices. Let D â€”
(di, o?25 â€¢ â€¢ â€¢, dn) be a nonnegative integer sequence satisfying the above conditions.
Since di > 0 and dÂ¿ > 0 for any i = 2,3,..., n, we can assign the degree of each
vertex ut 6 V by d{y\) = d\ and d(v{) = d, + 1 for i = 2,3,..., n. Let j < n and
dj = m > 0. From (c), there are exactly m + 1 integers, j < ji < j? < ... < jm <
jrn+1, such that ji = j + 1, and djk + Â£Â¿=jâ€˜+l(d,  1) = 0 for k = 1,2,..., m. Now
we assign an edge between vj and vjk for k = 1,2,..., m. That is, Vjk â‚¬ A(vj)
for i â€” 1,2,..., m.
(i) If djk = l > 0, we assign an edge between Vjk and Vjkp for p = 1,2,...,/
by (d). We do a similar procedure for each vertex, Vj, assigned to dj > 0. Therefore
vjk e A(vj) and vjkp
(ii) If dj{ = 0, then we do not assign any new edge for the vertex vJt. But it is
already joined to the vertex Vj by an edge.
Let s = (v\,V2,..., vn). Then all the vertices in s are connected. Since each vertex
Vi is assigned di new edges and Â£"_! dÂ¿ = n â€” 1 from (b), we have n vertices and
43
n â€” 1 edges. Then the set of all the edges assigned above, together with the vertex
set V, forms an ntree by Proposition 3.2.6. _L
Remark: The proof of Theorem 3.4.12 shows how to construct a tree that correÂ¬
spond to the given degree sequence satisfying DSC.
Example 3.4.13 Let D = (3,2,3,1,0,0,0,1,0,2,0, 0,0) be a nonnegative integer
sequence having 13 entries. Let V â€” {wi,t>2,... ,vn} be a set of n different vertices.
^9
fn
.t>13
, V8
v7
VlO
1^1
V2
V3 V4 V
Vu v6
Figure 35
(a) Clearly, = 12 and d\ = 3 > 0. So we expect a 13tree.
(b) Since d(yi) â€” 3, there are 4 different integers, 1 < li = 2 < I2 = 10 < I3 =
13 < I4 = 14, such that d2 + Â£f=3(dÂ¿ 1) = 2 +(2+ 0111401) =
0, d\Q + â€” 1) = 2 + (â€”1 â€” 1) = 0, and Â¿13 = 0. We assign edges between
Vi and each of u2,i>i0, U13.
(c) (i) Since v2 = 2, there 3 different integers, 2 < 3 < 8 < 10, such that Â¿3 +
Yl]=4{di â€” 1) = 3 + (0 â€” 1 â€” 1 â€” 1) = 0, and d$ + (dg â€” 1) = 1 â€” 1 = 0. We assign
edges between v2 and each of v% and u8.
(ii) Since u10 = 2, there are 3 different integers, 10 < 11 < 12 < 13, such that
dn = 0 and Â¿12 = 0. We assign edges between r>io and each of t>n and Vi2.
(iii) Since vi3 â€” 0, we do not assign any new edge for U13. But the vertex V13
is connected to iq by an edge.
44
In a similar way, we get {v3, v4}, {i>3, u6}, {u3, u7}, {u8,U9}, and {n4,n5}. Let E be
the collection of all the edges above, that is,
E = {{ui, u2}, {*>1, vio}, {vu t>i3}> {^2, u3}, {u2, us}, {uio, Un},
{uio, U12}, {u3, U4}, {u3, U6}, {u3, u7}, {u8, u9}, {u4, u5}}
Since the edge set E contains 12 elements, the set T â€” {V, E} is a 13tree.
3.5 Perfect Sequence and Main Theorem
In the previous section, we showed that, for a given ntree T, there exist many
vertex sequences depending on the starting vertex and choice among the adjacent
vertices for each successive vertex. Therefore the degree sequence set D for the
given tree T contains many different degree sequences which denote the same tree
T. We need a way to choose one degree sequence representing the tree T. First of
all we want to define an order on the set of all finite, nonnegative, integer sequences.
Definition 3.5.1 For any two finite nonnegative integer sequences D = (pi,p2, â– â€¢â– iPk)
and E â€” (qi,q2, , D < E if and only if either there is m < min{A;,/} such
that pm < qm and pi â€” qi for all i = 1,2,..., m â€” 1, or k < l and = qÂ¡ for all
i = 1,2,..., k. We write D < E if D < E or D â€” E.
Proposition 3.5.2 Let D be the set of all degree sequences of an ntree. The degree
sequence set, (D, <), is a totally ordered finite set.
Proof For any two different degree sequences the order defined above is well defined.
Since each term dt of a degree sequence is nonnegative and less than n and the
length of each degree sequence is n, the degree sequence set D is a finite set. Thus
(D, <) is a totally ordered finite set. _L
Definition 3.5.3 For a given ntree T, let S be the vertex sequence set of T and
D the degree sequence set of T. Then the maximum degree sequence in D is called
the perfect sequence of the given tree T, and it is denoted by pf(T). For the perfect
sequence, pf(T), of a given tree, T, there is a vertex sequence s = (iq, u2,..., vn) â‚¬
45
S such that the degree sequence of 5 is pf{T). We call s a perfect vertex sequence
of the tree and V\ a perfect vertex of the tree.
In the Example 3.4.2, the perfect sequence is (3,1,0,0,0} and there are two
different perfect vertex sequences, sh = (r,s,t,p,q) and sÂ¡2 = (r,s,t,q,p) for it.
But the trees represented by Sn and S12 are isomorphic.
For any given tree the perfect sequence is unique by Proposition 3.5.2. But
there may exist several different perfect vertex sequences. If the given tree is a
labeled tree, then these perfect vertex sequences denote different labeled trees. But
these labeled trees are isomorphic to each other; that is, there is a unique unlabeled
tree. Therefore, for a given perfect sequence, there exists a unique unlabeled tree.
Theorem 3.5.4 Let dÂ¡ be the perfect sequence of a tree T. If T' is any other tree
whose perfect sequence is also d$, then T' is isomorphic to T.
Proof If T has more than one vertex sequence with the same degree sequence, then
these vertex sequences all produce the isomorphic trees, by using the construction in
the proof of Theorem 3.4.12. T
Because of this theorem, we have our main theorem below.
Theorem 3.5.5 (Tree Classification Theorem) For any positive integer n, let
T(n) be the set of unlabeled ntrees and V(n) the set of perfect sequences of ntrees.
Then there is a onetoone correspondence between T(n) and V(n).
3.6 Obtaining Perfect Sequences
By Theorem 3.5.5, if we can construct the set of perfect sequences of ntrees,
P(n), without being given any tree shape, then we can classify all the ntrees by this
set. And, in the same way as in Example 3.4.13, we can get the different tree shapes
from the elements of V(n).
Proposition 3.6.1 For the perfect vertex iq of a perfect vertex sequence s =
(tq, v2,..., vn), we have d{yi) > d(vi) for Â¿ = 2,3, ...,n.
46
Proof If there is some k â‚¬ {2,3, ...,n} such that d{v > d{v i), then Ds =
(d(vi),...) < Dy = (d(vk), â– â– .) for any vertex sequence s' = (tq,...). This contraÂ¬
dicts the maximality of Ds. T
Proposition 3.6.2 If Ds â€” (di, d2,..., dn) is the perfect sequence of a given ntree
T, then n â€” 1 > d\ > dÂ¿ > dn â€” 0 for any i = 2,3,..., n â€” 1.
Proof Since T is an ntree and Ds is the perfect sequence, we have n â€” 1 > d\. For
any vertex u Â£ V(T) we have d(iq) > d(u) by Proposition 3.6.1. Since d\ = d{vi)
and d{ = d(ut) â€” 1 for every 1 < i < n, we have d\ > di for 1 < i < n. Since dn
must denote one of the end vertices, each of which has degree 1, we have dn = 0.
Therefore we have dt > dn = 0 for every 1 < i < n â€” 1. T
Definition 3.6.3 Let D = {dl5 d2,..., dn) be a degree sequence. Then we say that
D satisfies the ordered subbranch condition, denoted OSbC, if and only if D satisfies
the following condition for all j > 1 :
If d(Ã¼j) = m + 1 > 2, and S'(uJ2), ..., B'(vjm) with j < ji < j2 <
... < jm are the subbranches of the vertex Vj other than the root subbranch, then
(Bdâ€™(vjk)) > (Bd'(vjk+1)) for every k = 1,2,..., m â€” 1.
Recall that (Bd'^Vjf)) is a section of the degree sequence D. (See Definition
3.3.3 and 3.4.1.)
Remark : If j = 1, then there is no root branch of tq; instead, there is one more
subbranch, Bd'(vjm+l), such that jm < jm+1 and {B'(vjrn)) > {Bdâ€™(vjrn+1)) .
Theorem 3.6.4 Let D = (di, d2,..., dn) be a nonnegative integer sequence. If D
is the perfect sequence of an ntree, then it satisfies the DSC and the OSbC.
Proof It is clear from Theorem 3.4.12 and the maximality of D. _L
Proposition 3.6.5 Let T be a tree and v a vertex with the largest degree in T.
Then there is exactly one degree sequence Dv for T starting at v and satisfying
both DSC and the OSbC.
47
By the above proposition, the converse of Theorem 3.6.4 is not true. The
following example is a counterexample.
Example 3.6.6 Let A = (cfi, d2:..., d\2) = (3,2,2,1,0,0,2,0, 0,1,0, 0). Then
E,=i Â¿i = 11 and dx = 3 > 0. And the sequence A can be separated into 4
sections like (3), (2,2,1,0,0,2,0,0) , (1,0), (0). The second section can be sepÂ¬
arated into 3 sections, (2), (2,1,0,0), (2,0,0), and the second and third secÂ¬
tion of these can be separated by (2), (1,0), (0) and (2), (0), (0), respecÂ¬
tively. All of these subsections satisfy the second and third conditions of DSC.
Thus the sequence A satisfies the DSC. That is, the sequence A is a degree seÂ¬
quence of a vertex sequence (iq, v2,..., V12) of a 12tree by Theorem 3.4.12. On the
other hand, the sequence A satisfies the OSbC, since we have (2,2,1,0, 0,2,0,0) >
(1,0) > (0). But A is not the perfect sequence of the 12tree. In fact, we rearrange
the sequence A to get B = (d7 + 1, d2, di â€” 1, di0, dn, di2, d3, d4, d5, d&, dg, dg) =
(3, 2,2,1,0,0,2,1,0,0,0,0) . Note that d(rq) = d\ and d(vj) â€” Â¿7 + 1. Clearly the
sequence B also satisfies the DSC and the OSbC. So B is also a degree sequence
of another vertex sequence (tq, v2, rq, Uio, t>n, Vu, V3, tq, u5, vq, v$, vq) of the 12tree.
That is, A and B are degree sequences of the same 12tree (see Figure 36). And
we have
B = (3,2,2,1,0,0,2,1,0,0,0,0) > (3,2,2,1,0,0,2,0,0,1,0,0) = A.
Thus A is not the perfect sequence. Furthermore A cannot be the perfect sequence
of any tree, because A and B denote isomorphic trees, and B > A.
vn iq0 iq U12
48
In this example we have that d(vi) = 3 = d(v7) which is the largest degree
in the tree. That is, the perfect sequence must start with the largest degree. We will
see in Example 3.6.8 that the degree sequence B is the perfect sequence of this tree.
Note that (1,0) is the only possible degree sequence when d\ â€” 1. Therefore
we assume d\ > 1 from now on.
Now we show how to construct alternate degree sequences from A, similar
to B. Assume that we have a nonnegative integer sequence A = (cfi, d2, â– â– ., dn)
satisfying the DSC and the OSbC. If there is 1 < i < n such that d\ < d,, then
A cannot be the perfect sequence of any tree by Proposition 3.6.2. So we assume
that d\ > d{ for i â€” 2,3, ...,n, and there is at least one integer 1 < j < n such
that dj â€” di â€” 1. That is, there is an ntree T = (V, E) such that A is a degree
sequence of a vertex sequence s = (rq, V2,..., vn) of T, and T has at least two
vertices of maximum degree d\. Clearly, dj > 0. Thus Vj is not an end vertex in
T. By Corollary 3.3.18, the set of the vertices of the root branch of Vj is not the
set of vertices in A itself. So, from Proposition 3.3.17 there is j < m < n such
that RB(vj) = {uj, v2> â€¢ â€¢ â€¢, Vj} U {uTO+i, um+2, â€¢ â€¢ â€¢, vn} is the set of the vertices of
the root branch of vj. Clearly, (wj+i, u,+2,..., vm) is the section containing all of
the subbranches of Vj except the root branch. Therefore, if dj â€” p, then there are
p subsections of A like (Bd'^Vj^) > (Bd'(vj2)) > ... > (5d'(uJp)) with j < j\ <
J2 < ... < jp < m +1, since A satisfies OSbC. On the other hand, let E(RB'(vj)) =
{{u,u} Â£ E(T)\u,v Â£ RB'(vj)} be the set of edges in the root subbranch oÃ Vj. Then
V = {RB'(vj), E(RB'(vj))} is a subtree of T. Let S(T') be the vertex sequence
set of T'. By Proposition 3.3.7, there is a unique integer 1 < k < j such that
Vk â‚¬ A{vj). Clearly, Vk â‚¬ RB'(vj). Let SVk(T') be the v^vertex sequence set of
S(T'), in which the first term of each vertex sequence is u*.. Let DVk(T') be the
Vkdegree sequence set of SVk(T') (see Definition 3.4.1). Since DVk{T') is a finite
set, there is a maximum sequence in DVk(T'). We denote the maximum sequence in
49
DVk(T') by mdR(vk). Note that di must be changed to d\ â€” 1 in mdR(vk), since
it is not going to be used for the first term anymore in the new degree sequence.
Now we have that either mdR(vk) > (Bd'(vjx)), or there is an integer 1 < q < p
such that (Bd'(vjq)) > mdR(vk) > (Bd'(vjq+ J). Note that mdR(vk) may be the
smallest section, then Vj., would not exist.
We rearrange the given degree sequence A to a new sequence in which dj + 1
is the first term and the subbranch degree sequences of Vj,
mdR(vk), (Bd'(vh)) , (Bd'(vh)), ..., (Bd'(vjp)) ,
are listed in descending order. Denote this new degree sequence by .
Since the subbranch degree sequences of vj are listed in descending order in
My , the new degree sequence M^ satisfies the OSbC. Since Mis a rearrangement
of A with di = d(vj) = dj + 1 > 0, d, = d(vi) â€” 1 = cfi â€” 1 for some i > 1, and
A satisfies the DSC, the new degree sequence M^ satisfies the conditions (a) and
(b) of the definition of the DSC. Since mdR(vk) is the perfect sequence of T', it
satisfies the conditions (c) and (d) of the definition of the DSC. Clearly, each section,
(Bd'(vjt)), satisfies the conditions (c) and (d) of the definition of the DSC. Therefore
the new degree sequence satisfies the DSC. Thus the new degree sequence M^
satisfies both the DSC and the OSbC.
For the given nonnegative integer sequence A = (di, 0Ã2,..., dn), the set
{Myjd(vj) â€” d(vi)} is a nonempty finite set, since A 6 {MyJd(vj) â€” d{iq)}.
Therefore there is a maximum element in this set. Let Pa = max{M^d(vj) = d(iq)}.
Now A determines a unique tree T. Suppose B is another vertex sequence for T
satisfying the DSC and the OSbC. Then {MÂ®d(ug) = d(vj)} = {M^d(uÂ¿) = Â¿(iq)},
where vq is an analogue of vj, by Proposition 3.6.5. Therefore Pa is the perfect
sequence for T.
By the above explanation together with Proposition 3.6.2 and Theorem 3.4.12,
we have the following theorem.
50
Theorem 3.6.7 Let A = (di, d2,..., dn) be a nonnegative integer sequence satisfyÂ¬
ing the DSC and the OSbC with d\ > dÂ¿ for i â€” 2,3, ...,n. Then A = Pa if and
only if A is the perfect sequence of an ntree.
In the following example, we show the degree sequence B in Example 3.6.6
is the perfect sequence of the tree.
Example 3.6.8 In Example 3.6.6,
A â€” (di, d2,..., dif) â€” (3,2,2,1,0,0,2,0,0,1,0,0)
satisfies the DSC and the OSbC and d2 = d3 = d7 = d\ â€” 1. That is, d(v2) = d(v3) â€”
d{v7) = d(tq). Thus max{M^j = 2,3,7} is the perfect sequence of the tree. For
v7 we have that RB(v7) = {v\,v2,... ,v7} U {uio, i>n, tq2} and (Bd'(y3)) = (0) =
(Bd'(v9)). Thus T' = {RB'(y7), E(RB'(v7)} is the subtree produced by the root
subbranch of v7. (See Figure 37.)
Since v7 â‚¬ A(v2), we get the v7vertex sequence set
SV7(T') = { (u2,Ui,t;io,Uii,Vi2,i;3,V4,t;5,t;6),
(U2, VU U10, vn, Vi2, u3, v6, v4, v5),
(u2, Ui, t>i2, U10, t>n, V3, V4, V5, V6) ,
(u2, Ui, Ãœ12, U10, vn,v3, vG, v4, vs),
(v2, V3, V4, V5, U6, Ui, U10, Vu, V12) ,
(u2, ^3? ^4? ^5) ^6? ^15 ^12} ^10) ^ll) j
(W2, t>3, Ãœ6, U4, Â«5, Vi, 1710, V11, V12) ,
(V2, V3, V6, V4, VS, VU V12, V10, VU) }.
Then the v7degree sequence set of SV7(T') is
DV7{T) ={ (2,2,1,0,0,2,1,0,0),
(2,2,1,0,0,2,0,1,0),
(2,2,0,1,0,2,1,0,0),
(2,2,0,1,0,2,0,1,0),
(2,2,1,0,0,2,1,0,0),
(2,2,1,0,0,2,0,1,0),
(2,2,0,1,0,2,1,0,0),
(2,2,0,1,0,2,0,1,0) }
= { (2,2,1,0,0,2,1,0,0),
(2,2,1,0,0,2,0,1,0),
(2,2,0,1,0,2,1,0,0),
(2,2,0,1,0,2,0,1,0) }.
51
Therefore we have the maximum sequence mdR(v7) in DV7(T') as mdR(y7) â€”
(2,2,1,0,0,2,1,0,0). Since mdR(v7) = (2,2,1,0,0,2,1,0,0) > (0) = {Bd'{vs)) =
(fld'M), we have = (3, 2, 2,1,0,0, 2,1,0,0,0,0).
In the same way, we get
M* =(3,2,1,0,0,2,1,0,0,2,0,0)
M* =(3,2,2,1,0,0,2,0,0,1,0,0).
Therefore max{MVj\j = 2,3,7} = MV7. Since
=(3,2,2,1,0,0,2,1,0,0,0,0)
> (3,2,2,1,0,0,2,0,0,1,0,0)
= A,
MÂ¿), which is B in the Example 3.6.6, is the perfect sequence of the tree.
Vll v10 Vi U12
V2
V5 V4 V3 V6
Figure 37
By the Tree Classification Theorem (Theorem 3.5.5) and the previous example,
we have the following important corollary.
Corollary 3.6.9 Let A = (
fying the DSC and the OSbC. If d4 â€” 1 > for every i = 2,3,..., n, then A is the
perfect sequence for an ntree.
We will see in the following example how we construct a tree shape from a
given maximal degree sequence.
Example 3.6.10 Let D = (ofi, d2, â€¢ â€¢ â€¢, dw) = (3,2,1,2,0, 0,1,0,0,0) be a maximal
degree sequence. Then we expect a 10tree, T. Let s = (tq, v2,..., u10) be the vertex
sequence having D as the degree sequence. Then d(v 1) = dx = 3, so the vertex rq
52
has 3 edges. (See Figure 38.) Though there is no specific order for these edges, we
label them as ei,e2,e3 for the convenience of explanation.
e2
e3 vi ei
Figure 38
From the definition of vertex sequence (Definition 3.3.4) u2 is one of the
elements of the adjacent set of iq, A(iq). Without loss of generality, let ej =
{ui,u2}. Then u2 has 2 â€˜newâ€™ edges, e4 and e5, since d2 = 2. Let e4 = {u2,u3}.
By the same reasoning, v3 has 1 â€˜newâ€™ edge, named e$. So e& = {u3,u4}. Since
d4 â€” 2, v4 has 2 â€˜newâ€™ edges, named e7 and e8. (See Figure 39.)
e2
^5
e8
e3 V\ el l>2 e4 V3 ^6 v4 e7
Figure 39
Since d5 = 0 and d6 = 0, neither v5 nor v$ has a â€˜newâ€™ edge; that is, u5
and v6 are end vertices. On the other hand, v3 and ve are the adjacent vertices of
t>4, since d4 = 2. Thus v7 is an adjacent vertex of one of {u4, u2, u3}. Since d3 = 1
and v4 is the adjacent vertex of v3, v3 is not an adjacent vertex of v7. Since
d2 = 2 and we have only one adjacent vertex of u2, namely v3, we add an adjacent
vertex, v7. (See Figure 310.)
v7
ve
e2
e8
e3 v4 ci V2 e4 V3 e6 v4 e7 v5
Figure 310
53
Since d7 = 1, v7 has 1 â€˜newâ€™ edge which connects with vs. Since d8 = 0,
there is noâ€˜newâ€™edge from vÂ». Since we covered up to <Â¿2; the remaining two vertices,
vq and i>io, are adjacent vertices of tq which has two â€˜unusedâ€™ edges, e2 and e3.
Let e2 = {u^ug}, and e3 = {ui,uio} Since we are thinking about an unlabeled tree
and we put the edge names for our convenience, there are no specific names for the
vertices and edges. Thus we get the following 10tree. (See Figure 311.)
(Â«9)
(uio) (t>i)
M
" M
(n2) M
Figure 311
M
M
3.7 Extended Partitions and the Algorithm
In Definition 3.5.3 and the following explanation, it was shown that any tree
can be represented by a unique perfect sequence. So, in this section, we develop an
algorithm to produce all the perfect sequences for the ntrees for any positive integer
n. From these perfect sequences, we get the shape of each ntree as shown in the
previous example, as well as the total number of ntrees.
From Definition 3.4.1 and the First Reduced Degree Sum Formula (Theorem
3.4.6), we see that a degree sequence for an (n + l)tree always has n + 1 nonnegative
integers as entries and these integers always add up to n which is the number of
the edges of the given (n + l)tree. Thus, our first step in listing all possible degree
sequences is to form ordered partitions of the integer n as sums of smaller positive
integers.
54
Definition 3.7.1 An ordered partition of the positive integer n is a sequence of
positive integers, (pi,p2, â€¢ â€¢ â€¢ ,Pm), such that Yl?=i Pi = n
Note that 1 < m < n in the definition above.
Definition 3.7.2 An extended partition of an ordered partition (pi,p2, â€¢ â€¢ â€¢ iPm ) Â°f
the positive integer n is a sequence of n + 1 nonnegative integers, (eh, d2, â€¢ â€¢ â€¢> dn+i) 5
such that
(i) di=pi, di = ni and
(ii) there is a subsequence (djx , dj2,..., djm) such that djk = pk for k = 1,..., m.
Note that any entry dÂ¿ in the extended partition must be 0 unless it is a
member of the ordered partition. It is important that different extended partitions
are produced not only by the different ordered partitions but can also be produced
by the same ordered partition.
For example, let (di, d2,.., dâ€ž+i) and (d'l5 dj,..., d,+1) be ^wo extended
partitions of an ordered partition (pi,p2,... ,pm), and let (dj,, dj2,..., djm) and
Â«.< djm) be subsequences, respectively, such that djl = di = p\ = d'ti and
djk = pk = d'tk for k = 2,3, ...,m. Assume that ji ^ ti for at least one i. Then
there are sections of zeros with different lengths between and djk, and d't(k
and d[k, respectively. So the extended partitions are different. Thus the order is
very important in an extended partition.
Also note that the length of the extended partition of n must be n + 1. So,
for any given ordered partition, (pi,p2^... ,pm), we need to add n â€”m + 1 zeros to
make an extended partition.
Proposition 3.7.3 Let (ch, d2,..., dn+i) be an extended partition of an ordered
partition P = (dj,, dj2,..., djm) of the integer n. Then this extended partition satÂ¬
isfies the First Reduced Degree Sum Formula (Theorem 346).
55
Proof From the given extended partition (d\, d,2,..., dn+1) of an ordered partition
(dji ,dj2,..., djm) of the integer n, we have that Eâ€=u ^ = Â£!=i 4 = n an<^
+ Â£"Â¿2 (^Â« ~ 1) = di + Â£fcl2(4 _ 1) + Ei=i, d,
= 4 + Â£L2(4 1) + Â£K wit ~ i)
= EZU 4 + CUi) + Â£K w(i)
= n + (m  l)(l) + (nm + 1)(â€”1),
= n â€” ra + 1â€”n + m â€” 1 = 0. T
Even though we have a very large number of extended partitions of n, some of
them can not be a maximal degree sequence for an (n + l)tree. So we first eliminate
some ordered partitions that could not possibly result in maximal degree sequences.
To do so, we need to check some properties of perfect sequences.
Proposition 3.7.4 Let DÂ¡ = (di, d2) â€¢ â€¢ â€¢, 4+i) be the perfect sequence of a given
(n f l)Ãree T. If d{ / 0 and d,+1 = 0, then 0 = dt+i = d,+2 = = d,+d,.
Proof Since dÂ¿ ^ 0, the vertex vÂ¡ has dÂ¿ â€˜newâ€™edges; that is, there are dÂ¿ adjacent
vertices of vÂ¿ and they are labeled by dÂ¿ consecutive integers after i. Since DÂ¡ is
a perfect sequence, it satisfies the OSbC. That is, the vertex ut+x is a vertex of the
largest degree among the adjacent vertices of vÂ¿. But c/,+1 = 0. Therefore we have
dj = 0 for j = i + 2, i + 3,..., i + d,. J_
This proposition gives us a method to make an extended partition from an
ordered partition by adding zeros. That is, we know certain extended partitions
cannot be perfect sequences just by checking this property.
Proposition 3.7.5 Let Ds = (cfi, d2, â€¢ â€¢ â€¢, 4+i) be the perfect sequence of a given
(n + l)Ãree T. If n > 2, then d\ > 2.
Proof If d\ = 1, then dt = 0 for every i = 2,3,..., n + 1 by Proposition 3.6.2.
Since n > 2, there are at least 3 vertices in T. Thus we have that â€”1) =
1 + Â£"=2(â€”l) ^ l + (â€”1 â€” 1) = â€”1 t^O. This is a contradiction to the First Reduced
Degree Sum Formula. Therefore efi > 2. !_
56
According to the above proposition, the only trees whose perfect sequences
start with 0 or 1 are the tree with zero edge and the tree with one edge. The only
degree sequence representation of a tree with zero edge is (0), and the only degree
sequence representation of a tree with one edge is (1,0). These are easy enough to
visualize and count, so for the algorithm we work on the assumption that we are
listing trees with at least two edges. Then the first term of the perfect sequence will
be at least 2.
The following proposition shows that we have an arc shape tree if the first
term of the perfect sequence is 2.
Proposition 3.7.6 If Ds = (efi, Â¿2, â€¢ â€¢ â€¢, dn+i) is Â¿he perfect sequence of an (n + l)tree
T and d\ â€” 2, then di = 1 for i = 2,3,..., n â€” 1 and dn = dn+1 = 0.
Proof By Proposition 3.6.2 we have that â‚¬ {0,1} for any i = 2,3,..., n + 1,
since di = 2. Then the possible highest degree in T is degree 2. So T is just a
linear tree with n edges. That is, there are exactly two end vertices which cannot be
v\. Therefore there are exactly two zeros in the sequence. Write a degree sequence
for T, in the order of the vertices in the following figure.
tfn+l V1 v2 V3 Un_x Vn
Figure 312
That is, start at a vertex adjacent to an end vertex and move in the longer
direction first. Clearly the degree sequence resulting from this method contains
the two zeros for vn and un+i. Then we will get a degree sequence of the form
D = (2,1,1,..., 1,0,0). For the same tree, the numbers in a degree sequence cannot
change, but only the order in which they occur can change by changing the position
of zeros. And, clearly, the new ordered sequence < D. So the sequence D must be
the perfect sequence of T. That is, DÂ¡ = D. Â±
57
Most of the extended partitions of n start with a 1 or a 2. Since the last two
propositions classify trees whose degree sequences start with these values, we do not
need to consider ordered partitions that start with 1 or 2.
Proposition 3.7.6 shows that for any given number n of edges there is only one
tree with n edges whose perfect sequence starts with a 2. In particular, it is a linear
tree with n edges, since d,= 1 for i = 2,3,..., n â€” 1. We add this sequence at the
end of the algorithm.
By Propostion 3.6.2, we do not need the ordered partitions in which the largest
summand appears more than once or is not the first term in the ordered partition.
If an ordered partition has either one or two entries only, then we can produce
only one extended partition for each case. See the following Proposition 3.7.7 and
Corollary 3.7.8.
Proposition 3.7.7 Let Ds â€” (dx, d2, â€¢ â€¢ â€¢, dn+i) be the perfect sequence of an (n +1)
tree T with d2 = 0. Then di â€” n and d, = 0 for all i > 1.
Proof: Let s be a vertex sequence yielding a perfect sequence DÂ¡ of an (n + 1)
tree T. Since Ds is a perfect sequence and 0 = d2 = d(v2) â€” 1, all of the vertices in
the tree other than v\ have degree 1, so all the terms in Dg other than d\ must be
zero. By the First Reduced Degree Sum Formula (Theorem 3.4.6), we have d\ â€” n. J_
Corollary 3.7.8 If an ordered partition P â€” (p\,p2) of n has only two entries,
then the produced extended partition, (d1? d2,..., dn+i), from, P is such that dx =
Pi, d2 = p2 and dÂ¿ = 0 for i = 3,4,..., n + 1.
We demonstrated several necessary conditions which apply to a perfect seÂ¬
quence of a tree. Unfortunately, while these conditions are sufficient in almost all
cases, there are certain sequences which, although they satisfy all the conditions, are
not perfect sequences of the trees they represent. The problems usually arise when
the tree has two or more vertices of the largest degree. We saw a counterexample in
Example 3.6.6.
58
Note that in the Example 3.6.6 the two degree sequences A and B have the
same ordered partition, P = (3,2,2,1,2,1). That is, our algorithm produces the
degree sequence B as one of the extended partitions of the ordered partition P.
Therefore all we need to do is just drop the degree sequence A from the result of
our algorithm.
Each step described so far has involved checking each member of a list of
possible degree sequences and removing those which could not be the perfect sequence
of a tree. The last step is no exception. Given a degree sequence, we check if there is
more than one vertex of maximal degree. If there are more than one (in Example 3.6.6,
both vertices iq and vj = u\ have the same degree, 3), we write degree sequences
for the tree starting at each such vertex. If the largest of these with respect to
the ordering < is the sequence we are checking, then it is a perfect sequence. If
some other vertex gives a larger degree sequence, it is false that the sequence we are
checking is a perfect sequence. This is a surefire approach in that, given a degree
sequence, it checks for a perfect sequence by considering every possible alternative
for the maximum and seeing if the sequence in hand is the largest.
CHAPTER 4
TREES AS INVARIANTS FOR BRAIN TUMOR SHAPES
4.1 Introduction
A sphere packing treatment plan for a given brain tumor is produced in the
first chapter. This plan can be represented numerically as a list of parameters of
the dose spheres. In particular, the list consists of the coordinates of the dose center
in millimeters along the lateral (Lat), anteriorposterior (AP), and axial (Ax) axes
together with the radius of the dose sphere in millimeters. From these data, we
can assign a unique corresponding graph by matching a sphere with a vertex and
matching the adjacency of two spheres with an edge. Then, by using the notion of
cutvertex, we give an order to the vertex set. We use this order to decide which edges
to choose in order to obtain a unique maximal tree contained in the graph.
Our main theorem of this section is Theorem 4.5.6: The perfect sequences are
invariants of the shapes of arbitrary brain tumors. We then develop a Mathematica
program (see Appendix B) to get a maximal tree of the graph for a sphere packing
plan for a given brain tumor. In the near future, we plan to use this program to
classify all the treatment plans which have been done at the Brain Institute of the
University of Florida. We hope that the classification gives some insight about brain
tumor shapes.
In this chapter, we assume that a brain tumor is connected, so the graph
representation for any brain tumor is a connected graph.
59
60
4.2 Definitions and Some Properties
Definition 4.2.1 A graph G â€” (V(G), E(G)) consists of a vertex set V(G) =
{ui,..., vn} and an edge set E(G) = {ei,..., em} C {{u, v}u, v G V(G)} in which
each edge is an unordered pair of two distinct vertices u and v, and the edge {u, u}
is said to join u and v. For an edge {u,v} we say that u and v are adjacent
vertices and the vertex u and the edge {u,u} are incident with each other, as are
v and {u,t>}. From now on, P(G') denotes the number of vertices and \E(G)\
denotes the number of edges. For a vertex u â‚¬ P(G), the vertex set
Ag{u) = {u â‚¬ G\v is an adjacent vertex of u}
is called the adjacent set of u in G.
Since there is no loop represented by a single edge, in a graph in this article,
no vertex is adjacent to itself.
Definition 4.2.2 We say that a graph G is isomorphic to a graph H if and only
if there exists a bijection / : V(G) â€”>â– V(H) such that {u,u} Â£ E(G) if and only if
{/(Â«) ,f(v)}â‚¬E(H).
Definition 4.2.3 A walk is an alternating sequence (u0, e\, iq, e2,..., e*,, Vk) of verÂ¬
tices and edges s.t. u,_i / for i â€” 1,2and e, = for every i. A
cycle is a walk (i>o, ei, iq, e2,..., e*, Vk) with k > 3 in which v0 = Vk is the only
vertex repeated. A path is a walk with no repeated vertex. A (u,v)path is a path
with the first vertex u and the last vertex v. A (u,u)path is a vertex u. The length
of the (u,v)path, l(u,v), is the number of edges in the (u,v)path. So the length of
(u, u)path is 0. The distance d(u,v) between two vertices u and v is the length of
a shortest path joining them, if any; otherwise d(u,v) = oo.
Note that a path of length >1 is of linear shape.
Definition 4.2.4 Let G â€” (V, E) be a graph where V = {vi, u2,..., un}. Then,
for any vertex v â‚¬ V, we say that PG(v) = j if and only if
61
j = min{f â‚¬ {1,2,..., n} there is a (v, u,)path in G}.
We call Pg(v) the pathvalue, or simply the Pvalue of the vertex v in G.
Definition 4.2.5 Let G = (V,E) be a graph. For any nonempth subset V C V,
let F = { {u,v} Â£ E \ u,v â‚¬ V } C E. For any subset F' of F, the graph
G' = (V', F') is called a subgraph of G. A graph G is connected if and only if every
pair of vertices are joined by a path. A graph having no cycle is acyclic. Note that an
acyclic graph can be disconnected. A forest is an acyclic graph. A tree is a connected
acyclic graph. An ntree is a tree with n vertices.
Definition 4.2.6 The degree of a vertex v, denoted by d(v), is the sum of the
number of edges incident with v.
Definition 4.2.7 A cutvertex of a graph is a vertex of degree greater than 2, whose
removal increases the number of components of the graph.
Definition 4.2.8 A nonseparable graph is connected, nontrivial, and has no cutverÂ¬
tex. A block of a graph is a maximal nonseparable subgraph.
Thus if v is a cutvertex of a connected graph G, then G â€” v is disconnected.
Note that the traditional definition of cutvertex (usually called cutpoint in graph
theory) does not have the degree condition. But we do not want to separate any
linear shape subgraph into several blocks because there is no difficulty in handling
linear shape subgraphs in our algorithm. Thus we do not call a vertex with degree
2, a cutvertex in this chapter, even though it separates the graph.
And if H is a block, then none of the vertices in H is a cutvertex of H.
Note that if there is no cutvertex in G, then the graph G itself is a block. This
definition differs from the definition of block in Harary [5], because we have a different
definition of cutvertex. Because of this, we have Proposition 4.2.10 below.
Proposition 4.2.9 If a graph G is of linear shape, then the maximal tree of G is
G itself.
62
Proposition 4.2.10 [5] Let v be a vertex of a connected graph G. The following
statements are equivalent:
(1) v separates G and d(v) > 2.
(2) There exist vertices u and w distinct from v such that v is on every (u,w)
path.
(3) There exists a partition of the set of vertices V(G) â€” {t>} into subsets U and W
such that for any vertices u â‚¬ U and w â‚¬ W, the vertex v is on every (u,w)path.
Definition 4.2.11 Let G = {V, E} be a graph. For any vertex v â‚¬ V we define
the edge subset Ig(v) C E(G) by Iq{v) â€” {{v,u} â‚¬ E(G)\ u â‚¬ Ag(u)}. We call
Ig(v) the incident edge set of v in G.
Clearly, Ig(v) contains all the incident edges of the vertex v in G.
Definition 4.2.12 Let G â€” {V,E} be a graph. For any subset F of E we call the
set V(F) = {u â‚¬ V(G) there is an edge e 6 F such that v and e are incident }
the F vertex set.
Proposition 4.2.13 [5] The following statements are equivalent for a graph T :
(1) T is a tree.
(2) Every two vertices of T are joined by a unique path.
(3) T is connected and Vr(T') = + 1.
(4) T is acyclic and P(T) = \E(T)\ + 1.
Note that if a tree has n vertices, then there are exactly n â€” 1 edges.
Definition 4.2.14 Let G = {V, E} be a connected graph and E' a subset of E.
A connected subgraph T = {V, E'} is called a maximal tree of G if and only if
T' = {V, E' U {e}} contains a cycle for any edge e â‚¬ E â€” E'.
Our work in this chapter produces a particular maximal tree for any given
graph. So we have the following (wellknown) theorem.
Theorem 4.2.15 Every connected graph has a maximal tree.
63
Example 4.2.16 Let G be a graph with 14 vertices and 18 edges as in the figure
41 below.
There are 3 cutvertices, 6, c, e. So the given graph G can be separated into
6 blocks as follows:
Gi = {V1,El} = {{a,b}, {{a,b}}},
G2 = {V2,E2} = { { b,c,k,l,m,n },
{ {b,c},{c,k},{k,l},{l,m},{m,n},{n,b},{b,k},{k,m} } },
G3 = {f/3, Â£3} = { { c, d, e }, { {c, d}, {d, e} } },
G4 = {V4,E4} = { { ej,g }, { {e, /},{/, g], {g, e} } },
G5  {V5,E5} = { { e, h,i }, { {e,h},{h,i} } }, and
G6 = {V6,Â£6} = { { e,j }, { {e,j} } }.
64
4.3 Graphs for Sphere Packing Plans
The sphere packing treatment plan S having n spheres for a given brain tuÂ¬
mor, from Chapter 1, consists of the centers, ci, C2,..., Cn, and the radii, j*i, r2,..., rn,
of the spheres with c, = (xÂ¿, yÂ¿, 2,) for every i, where Â£,â€¢ is for the AP coordinate,
yÂ¡ is for the Lat coordinate and Z{ is for the Ax coordinate of the center, c,. (Note
that the physicians use diameters, rather than radii, in their data sets.)
We consider the spheres to be the vertices of a graph, and we assign the vertex,
V{, for the sphere centered at cÂ¿ with radius rÂ¿. Now we can assign an edge between
two vertices if the spheres assigned by these two vertices meet. In Chapter 1, we
noted that the collimators do not deliver clear cut boundaries for spheresthere is
some spillover. The neurosurgeons currently assume that the portion between two
spheres gets enough dosage of radiation if the distance between the centers of the
two spheres is less than or equal to eleven tenths of the sum of the radii of the two
spheres. Thus for any two vertices assigned for two spheres satisfying this condition
we assign an edge between them. Then we have the following proposition.
Proposition 4.3.1 Let G â€” {V, E} be a graph for the sphere packing plan of a
brain tumor. For any vertices and Vj in V, d(c,,Cj) < y4(rt + rf) if and only if
Â£ E. Thus we have the following edge set;
E = {{z, j}\d(ci,Cj) < (ll/10)(r, f rj) where i < j, i = 1,2,...,n â€” 1} .
Since any brain tumor is connected in this article, there is at least one (u, v)
path in G for every two vertices, u and v in G. Thus we have the following propoÂ¬
sition.
Proposition 4.3.2 The graph G = {V, E} produced by the above method for a
given tumor is a connected graph.
Since we are dealing with unlabeled graphs in this article, we have the following
proposition.
65
Proposition 4.3.3 For any sphere packing treatment plan for a given brain tumor,
there exists a unique graph that we assign to the treatment plan.
By means of the above method, we get a unique connected graph G = {V, E}
for the sphere packing treatment plan for the tumor. This graph does not have the
information about the positions of the dose spheres as an aspect of the structure and
looks only at how the spheres are attached. It is clear that the graph G does not
have a loop or multiple edges between two vertices, since every edge is assigned for
two different vertices which are assigned for two attached spheres.
4.4 Order in the Vertex Set
To get a maximal tree from a graph we may need to delete some edges in the
graph. Thus we need to label each vertex in order to choose certain edges to delete,
even though we are dealing with an unlabeled graph in this chapter. So there is no
specific meaning for this labeling except that it is only used to choose a maximal tree
for the graph.
Example 4.4.1 Let P be the sphere packing treatment plan (see Figure 43) for a
given brain tumor.
Figure 43
Then, by Proposition 4.3.3, there is a unique graph G for the sphere packing plan.
The graph is given below in Figure 44.
P
Figure 44
66
There are two different isomorphism classes, namely T\ and T2, of maximal trees
for the graph G. That is, we could choose either of these for a maximal tree to assign
the graph G. Since, in our algorithm, we want to get the same maximal tree for a
given graph every time, we need certain rules to get the same tree. (See Figure 45.)
Tx T2
Figure 45
In this chaper, the graph comes from a sphere packing treatment plan. That is,
the main shape of the brain tumor depends on the size of spheres and the connection
between spheres. Even though we cannot keep the information about the size of
spheres in the graph, it seems to us that the bigger the sphere is, the more it has a
chance to get attached to more spheres. So we assume that the larger degree vertices
play a more important role in classifying the tumor shapes into trees than the smaller
degree vertices. That is, by choosing a largest degree vertex first, the shape of the
tumor is apparently most closely preserved. Thus, in the previous example, we choose
the vertex q or s as the starting vertex; then we pick all the incident edges. Then
we get the tree Tx. Therefore we want to choose the tree Tx for the maximal tree of
G. Note that the tree T2 produces a linear graph which does not show the shape of
the tumor as closely as Tx.
But there are some vertices which are more important than the bigger degree
vertices. Recall the graph G in Example 4.2.16. Then, in the block G2, it is clear
that the vertex k has the largest degree, 5. But the cutvertices, b and c, play a
critical role in obtaining a maximal tree of G. So, to get a maximal tree for a graph
G, we want to start at the cutvertices of G first, instead of the vertices with the
largest degree. So we want to label the cutvertices first. On the other hand, there
67
are different kinds of â€˜cutverticesâ€™ in some graphs. The following example shows such
a case.
Example 4.4.2 Let G = {V, E} be the graph in Figure 46.
u9
u 7
u 5
Then there are two cutvertices of the graph G, namely ui and Uio. UsÂ¬
ing these two cutvertices, the graph G is separated into four subgraphs, namely
G\, G21 G3 and G4. (See Figure 47.)
Then the subgraphs Gi, G2, and G3 are blocks, but the subgraph G4 conÂ¬
tains its own cutvertices, u4 and w8, which are not cutvertices of the graph G. We
separate the subgraph G4 into four subgraphs by using its cutvertices u4 and ug.
68
(See Figure 48.) Note that there is no specific order in labeling the subgraphs. So
at this moment, we relabel the subgraphs of G by H\, H2, ..., H7.
u9
Figure 48
Then, again, the subgraphs Hi, H2, ..., He are blocks, but the subgraph Hj
contains its own cutvertex, ue, which is neither a cutvertex of the graph G nor a
cut vertex of the subgraph G4. We separate this subgraph H7 into two subgraphs by
using ue. (See Figure 49.) And we relabel the subgraphs of G by B\, B2, ..., Bg.
Therefore {Â«2,^10} is the set of cutvertices of the graph G. But {u4,u8} is
the set of the cutvertices of a subgraph which is produced after the first separation
using the cutvertices of the graph, and ug is the cutvertex of a subgraph which is
produced after the second separation.
For the purpose of labeling the vertices in a graph, we separate these cutvertex
sets for different levels, from each other.
Definition 4.4.3 Let G be the graph assigned for a given sphere packing plan
P. Then the vertex set V(G) of the graph G is the union of two disjoint subsets,
namely Ci(G) and C[(G), where Ci(G) = {v Â£ V(G)\ v is a cutvertex of G}
69
and C[(G) = V(G) â€” Gi(G). We call C\(G) the first step cutvertex set of G. For
every i = 2,3,..., V(G), we define a subset G,(G) of Gf_1(G) as the collection
of the vertices of C^fiG), which are cut vertices of a subgraph produced after the
(i â€” l)th separation using the elements of G,_i(G). Then we call G,(G) the ith
step cutvertex set of G, and let C(G) = Gf_x(G) â€” G,(G). We assume that there
are finitely many vertices in a given graph. If there exists at least one cutvertex,
then there exists an integer 1 < k < n such that G,(G) ^ 0 for every i < k, and
GÂ¿(G) = 0 for every i > k + 1. Then we call A; the separation step constant of the
graph G, and let GC(G) = (Ct)c(G).
If there is no cutvertex of G then the graph is a block, and we say that the
separation step constant of G is 0.
u9
Figure 49.
Proposition 4.4.4 Let (G) = n and let the separation step constant of G be
k > 0. Then there is a sequence of integers {mi, m2,..., m*} with 0 < mi < m2 <
...
70
Proof Since k > 0, the first step cutvertex set Gi(G) is not empty. Thus there is
a positive integer, namely mi, such that Gi(G) = m\. For every i = 2,3,...,A;,
let m, = m,_i + G,(G). For every i = 2,3,..., k, the t'th step cutvertex set GÂ¿(G)
is not empty. So we have that m, > mÂ¿_i, and \Ci(G)\ = mÂ¿ â€” m,_i. J.
Now we label the vertices as follows;
(1) the vertices in Gi(G) are labeled by {1,2,..., mi},
(2) for every i = 2,3,..., k, the vertices in G,(G) are labeled by {m,_i + 1, mÂ¿_i +
2,..., mÂ¡ â€” 1, mÂ¿}, and
(3) the vertices in GC(G) are labeled by {m*, + 1, m^ + 2,..., n}.
We arbitrarily name the vertices of G,(G) and GC(G), and denote that Gi(G)
}ui, U2,..., }, G,(G) â€” {^T7iÂ¿_i+i) +21 â€¢ â€¢ â€¢ 5 } for i 2,3,...,/u, and
CC(G) = {umfc+i, umjfc+2,..., un}. It is clear that for every vertex up â‚¬ G,(G) and
every vertex uq â‚¬ Cj(G) with i ^ j, we have that p < q if and only if i < j.
And also, for every vertex uv â‚¬ Ci(G) for every i = 1,2, ...,&, and every vertex
uq â‚¬ CC(G), we have that p < q. Assume that, for every vertex the center and
the radius of the sphere which is assigned to that vertex u,, are cÂ¿ = {xÂ¿, zÂ¿) and
r,, respectively.
We now relabel the elements for each of the sets, Ci(G), (^(G),..., Ck(G)
and GC(G), in the following way:
for every pair of vertices, it,, Uj, in one of Gi(G), G2(G),..., Gfc(G) and GC(G),
Â¿ < j if and only if
(i) d(ui) > d(uj)] or
(ii) If d(ui) = d(uj), then rÂ¿ > rÂ¿; or
(ii) If d(ui) = d(uj) and rÂ¿ = rÂ¿, then xt < Xj\ or
(iii) If d{u{) = d{uj), rÂ¿ = rj, and x, = Xj, then yÂ¿ < yj; or
(iv) If d(ui) = d(uj), r, = rÂ¿, x, = Xj, and y, = yj, then z, <
Note that d(u{) represents degree of the vertex it, in G.
71
It is clear that for any two different spheres, at least one of the coordinates is
different. Thus any two distinct spheres can always be compared by the above order.
Therefore the above order for the vertices of the graph is well defined.
4.5 Maximal Tree from a Graph
In this section, we develop an algorithm that will uniquely select a maximal
tree from the graph which is produced for a given brain tumor. The choice is made
so that the tree will most closely resemble, in our opinion, the shape of the tumor 
that is, of the graph obtained for the tumor.
Let G â€” (V, E) be the graph for a given sphere packing plan. Let 14(G) =
{ui, u2,..., un} for some positive integer n, where all the vertices in V are labeled
by the method in the previous section.
Define a subgraph T0 â€” (14(G), F0) with I4(F0) = V = 14(G) and F0 =
0; that is, there is no edge in T0. Then, for every i = 1,2, ...,n, we have that
Pt0(uÃ) = i. (See Definition 4.2.4.)
(1) For the vertex iq, let Fi = F0 U Ig(ui). (See Definition 4.2.11.) Let =
(14(G), Fi). The subgraph (I4(F1),F1) C T\ is a tree. Thus, if V(Fi) = 14(G), then
T\ â€” (14(G), Fi) = (14(Fi), F\) is our maximal tree of G. Note that for every vertex
v â‚¬ 14(Fj) we have that Ptx (u) = 1, and for every vertex Uj V(Fi), we have that
PF (Uj) = j.
(2) Assume that 14(Fi) ^ 14(G). For the vertex U2, there are two cases, that is,
Fji(ii2) = 1 and F7â€™1(u2) = 2. Now define the edge subset I'G(u2) by /Â¿(u2) =
IgM  {{u2,u} â‚¬ E\ Pt,{v) = 1}.
(a) If Ptx(u2) = 1, that is, there is a (ui,u2)path in the subgraph Ti, then let
F2 = FjU/Â¿(u2). Each edge in the set {{u2, v} â‚¬ E\ PTl (v) = 1} = {{u2, v) 6 EÂ¡ve
V(Fi)} makes a cycle with {uuu2} and {tii,u}, since FT,(u2) = 1. Thus there is
no cycle in the subgraph (14(F2),F2), so it is a tree. Let T2 = (14(G), F2) C G.
72
If V(F2) = V(G), then T2 = {V(G),F2) = (V(F2),F2) is our maximal tree of G.
Note that for every vertex v Â£ V(F2) we have that PtAv) = 1, and for every vertex
uj 0 V(F2) we have that Pj2(uj) â€” j.
(b) If PtÂ¡{u2) = 2, then there is no (iti,u2)path in the subgraph T\.
(i) If there is a nonempty subset {tt2l, u2i,..., u2t} C V(Ig(u2)) where PtAu 2.)
= 1 for i = 1,2,..., t and 2, < 2j for 1 < i < j < t, then let F2 = F\ U PG(u2) U
{{u2,u2l}}. Then the subgraph (V(F2),F2) C G is a tree. Let T2 = (^(G),^) C
G. If V(F2) = I/(G), then T2 = (V(G),F2) = (V(F2), F2) is our maximal tree of
G. Note that for every vertex v 6 V(F2) we have that Pt2 (v) = 1, and for every
vertex Uj V(F2) we have that PtAu])=j.
(ii) If none of the vertices in V{Ig{u2)) has pvalue 1, that is, for every vertex
v Â£ V(Ig{u2)) there is no (i
for every vertex w Â£ F\ and Ptx(v) = 2 for every vertex v Â£ Ig(u2), the subgraph
(V(F2),F2) is a forest, but not a tree. Let T2 = (V(G),F2) C G. It is clear that
the subgraph T2 cannot be our maximal tree, even if we have that V(F2) â€” V(G),
because the subgraph T2 is disconnected. Note that for every vertex v Â£ V{F2) =
V(F\ U IG(u2)) = V{Fi) U V(Ig{u2)) we have that Pt2(v) = 1 if v Â£ V(F\), and
Pt2(v) = 2 if v Â£ V(Ig(u2)). And, for every vertex Uj Â£ V(F2), we have that
ptAuj) = j
(3) By induction, at the jth step, the subgraph Tj_i = (V(G),Fji) has been
defined. We want to define the subgraph Tj C G. For the vertex iÃÂ¿, there are two
cases, that is, Ptj_Au3) â€” j and Pt}^ {uj) = q for some 1 < q < j. For every
i = 1,2,..., n, define
= min{l Â£ {1,2,... ,n}\PT]_Aui) = * and ut Â£ AG{uj) }.
If the set {/ Â£ {1,2,..., n}Â«/ Â£ V{lQ{uj)) and Pt}Aui) = * } is empty, then the
vertex (,) is not defined. So there is no edge {tij,umj_,(Â»)}.
73
(a) Let Ptv_i (uj) = j; that is, for every Â¿ = 1,2,..., j â€” 1, j + 1,..., n, there is
no (uj, u,)path in the subgraph Tji = (V(G),Fj1). Let
Fj = Fj1 U (Ut=1{ {uj, umj_i(,)} }) .
Then the subgraph (V(Fj), Fj) is a forest. Let Tj = (V(G),Fj) C G. If V(Fj) =
V(G) and \V(Fj)\ = \Fj\ + 1, then, by Proposition 4.2.13, the subgraph Tj =
(V(G), Fj) = (V(Fj), Fj) is our maximal tree of G.
(b) Let Ptj_, (uj) = q for some 1 < q < j, that is, there is a (u9, tÃÂ¿)path in the
subgraph Tji = (V(G), Fji). Now let
Fj = Fjx U (ui=1) {v,j, }) .
Then the subgraph (V(Fj), Fj) is a forest. Let Tj = (V(G),Fj) C G. If V(Fj) =
V{G) and \V(F3)\ â€” \Fj\ + 1, then by Proposition 4.2.13, the subgraph Tj =
(V(G), Fj) = (V{Fj), Fj) is our maximal tree of G.
Note that if u is an Tth step cutvertex, then d(u) in G is same as d(u) in
Ti_j. Also note that, for every i = 1,2, ...,n, {ui, tÂ¿2,..., tt,} C ^(F,), that is,
V(G) = V(Fn). We will see in Theorem 4.5.3 that Tn = (V(G),Fn) = (V(Fn), Fn)
is our maximal tree of G. By the Tree Classification Theorem (Theorem 3.6.8), this
tree Tn can be denoted by a unique perfect sequence.
Example 4.5.1 Let G = {V, E) be a graph where V = {ux, u2? â€¢ â€¢ â€¢, us}? which
are ordered as in the previous subsection, and E = {ei, e2,..., e13} as in the figure
below. Clearly, the graph G does not have a cutvertex, so the order is decided by
only the sizes of the spheres and the coordinates of the centers of the spheres.
^8 ^13 U7 e2 U\
u2 e5 u6
Figure 410
U5
74
Let To = (V(G),Fo) be a subgraph of G = (V, E) where Fo = 0, as shown in
Figure 411.
u 8
u7
U1
u3.
u4
*U5
â€¢ â€¢
u2 U6
Figure 411
Since IG(uj) = {ex,e2}, let Fx = F0U/g(ui) = {ex,e2}. Then V^Fx) = {ux,u5,u7}.
Since ^(Fi) = {ui,U5,u7} ^ V{G), the tree (V(Fi),Fi) is not a maximal tree of
G. Let Ti = (V(G),Fi) which is shown in Figure 412.
â€¢ â€¢
U2 U&
Figure 412
We have that mi(Â¿) = i for i = 1,2,3,4,6, and 8, but mx(5) and mx(7) do not
exist. Since Fr1(u2) = 2 and AG(u2) = {Â«3, u4, u6}, let
F2 = Fx U (U,=3i4i6{{u2, u,}})
= Fx U Ig(u2)
= {eii e2} U {e3, e4, es} = {ex, e2, e3, e4, es}.
Then V{F2) = {ux,u5,n7, u2,u3,u4,u6} ^ V(G) and the subgraph (V(F2),F2) isa
forest, not a tree. Thus it cannot be a maximal tree of G. Let T2 = (V(G),F2).
(See Figure 413.)
75
u2 e5 u6
Figure 413
Note that Pt2{ui) = Pt2(u5) = Pt2{u7) = 1, Pt2{u2) = Pt2{u3) = Pt2(u4) =
Pt2{u6) â€” 2, and Pt2(us) = 8. Thus we have that m2( 1) = l,m2(2) = 2, m2(8) = 8.
Note that m2(i) does not exist for Â¿ = 3,4,5,6,7. Since Ui ^ A(u3) and Pt2(u3)â€”
2, weget F3 = i?2U{{u3, u8}} = {ei, e2, e3, e4, e5, e7}. Since the subgraph (V(F3),F3)
is a forest, but not a tree, it is not a maximal tree of G. Define T3 = (V(G),F3).
(See Figure 414.)
u2 e5 u6
Figure 414
Note that Pt3(ui) = Pt3(u3)  Ptz{u7) = 1 and Pt3(u2) = Pt3{us) = Pt3(ua) =
Pt3{u6) = Pt3(us) â€” 2. Thus we have that m3(l) = 5 and m3(2) = 4. Note that
m3(?) does not exist for i = 3,4, ...,8. Since Pt3(u4) = 2, we define F4  F3 U
{{u4,u5}} = {e1,e2,e3,e4,e5,e7,e8}. Since V(F4) = {uj, v2,..., u8} = V(G) and
V(F4) = 8 = F4 + 1, the subgraph T = (V(F4), F4) is a maximal tree of G. (See
Figure 415.)
76
u8 Ur e2 u i
Figure 415
At this moment, we have that Pt(u\) = Pt{u2) = ... = Pt(u$) = 1. Then,
for every ut â‚¬ V(G) with t > 4, we have that Ft = F4. Thus, for every t > 4, the
subgraph (V(Ft),Ft) is same as T = (V(F4), F4).
The next example shows how we get the maximal tree for a given graph
containing some cutvertices.
Example 4.5.2 Let G be the graph in Example 4.2.16. (See Figure 41.) Then
Ci(G) = {6,c,e} and Cf(G) â€” {a,d, f,g,h,i,j,k,l,m,n}. In Ci(G), we have that
d(e) = 5, d(b) = 4 and d(c) = 3. On the other hand, in C'i(G), we have that
d(k) = 5, d(m) = d(n) = 3, d(d) = d(f) = d(g) = d(h) â€” d(l) â€” 2, d(a) = d(i) =
d(j) = 1. And C2 = 0 Thus we order the vertices as follows; v\ = e, v2 = b, v3 = c
and v4 = k, {v^v&} = {m,n}, {u7, u8, u9, uio, Un} = {d, f,g, h, /}, {u12,Ui3,ui4} =
{a,i,j}. Assume that v5 = m, u6 = n, u7 = d, v8 = /, u9 = g, ui0 = /i, uu =
l, vi2 = a, i>j3 = Â¿, u14 = j, which are decided by the sizes and centers of the
corresponding spheres. (See Figure 416.)
Figure 416
77
Since IG(vi) = { {ui,t>7},{ui,u8},{ui,M>{ui>uio},{^i,ui4} }, let Fi = IG(vi).
Then V(Fi) = {i7X, v7, v8, v9, u10, i>i4}. Since V(Fi) ^ V(G), the tree 7\ = (V(G), Fi)
is not a maximal tree of G. Thus, for the cut vertex i>i, we have the following subÂ¬
graph which is T\ = (V(G), Fi). (See Figure 417.)
Ãœ5Â» â€¢ Vn Ui3Â«
v6Â» â€¢ V4
vu v2 v3
Figure
Since Ptx{v2)  2, and, for every vertex v â‚¬ AG(v2) = {u3,u4,u6,^12}, ^ 1,
let F2 = Fi U Ig(v2). The subgraph (V/(F2), F2) is a forest, but not a tree. Thus it
cannot be a maximal tree of G. Now, for the cutvertex w2, we have the following
subgraph which is T2 = (1/(G),F2). (See Figure 418.)
417
â€¢ V\i Ui3Â«
Since Ag(v3) = {u2,u4,u7}, Pt2{v2) = Fr2(u4) = 2 = Pt2(v3), and Pt2(v7) = 1, let
F3 = F2 U {{u3,u7}}. Then, for the cutvertex u3, we have the following subgraph
which is (1/(F3),F3). Then the subgraph (V(F3),F3) is a tree. But V{F3) fi V(G).
78
Thus it cannot be a maximal tree of G. We have the following subgraph which is
T3 = (V(G),F3). (See Figure 419.)
V5â€¢ â€¢ Un Ui3Â«
Since AG{y4) = {u2, u3, u5, i>6, t>n}, and
Pt3{v2) = Pt3(v3) = Pt3(v6) = 1 = Pt3(v4), Pt3{v5) = 5, Pt3(vu) = 11,
we define F4 = F3 U {{^4,^5}} U {{^4,^11}}. Then the subgraph (V(F4), F4) is a
tree. But V(F4) ^ V(G). Thus it cannot be a maximal tree of G. Then, for the
cutvertex v4, we have that the following subgraph which is T4 = (V(G), F4). (See
Figure 420.)
For the vertices v5,v6,v7,vs,v9, there is no â€˜newâ€™ edge. Thus we get the same subÂ¬
graph as above, that is, T4 = T5 = T6 = T7 â€” T8 â€” T9. But, for the cutverÂ¬
tex U10, we have that U13 Â£ V(Ig(v 10)) and Pt9{v 13) = 13 ^ 1 = Pt9{vio) Thus
79
f10 = f9 U{ {^10,^13 } and V(FW) = V(G). Therefore we have the following subÂ¬
graph which is T = (V(G),Fiq) = (V(Fio), Fio). (See Figure 421.)
Clearly, the subgraph T â€” (F(Fâ€™io), Fi0) does not include a cycle. Thus it is a maxÂ¬
imal tree of G. That is, for every t â€” 11,12,13,14, we have that (F(Ft),FÂ¡) =
(V(Fio), F10). Therefore, the subgraph T = (V(Fi0), Fi0) = (V'(G), F10) is the maxÂ¬
imal tree of the graph G.
For this resulting tree T = (V(Fio), Fio), choose a vertex sequence s by
s = (vi,v7,v3,v2,v4,v5,vn,ve,vi2,vi0,vi3,vs,v9,vi4). Then the sequence
P = (5,1,1,3,2,0,0,0,0,1,0,0,0,0)
is the degree sequence of s. By Theorem 3.4.12, P satisfies DSC. It is clear that
P also satisfies DSbC. By Corollary 3.6.10, P is the perfect sequence for the tree
T = (V(F10), fio).
Theorem 4.5.3 Let G â€” {V, F} be a graph for a sphere packing plan where
V = {uj, U2,..., un} with the order described in Section f.f. Let the subgraph
(V(Fn),Fn) of G come from the above procedure. Then V(Fn) = V(G). ThereÂ¬
fore (V(Fn), Fâ€ž) is a maximal tree of G and there is a unique perfect nsequence for
the tree (V(Fn),Fn).
Proof The subgraph (V(Fn), Fn) is connected. In fact, assume that the subÂ¬
graph (V(Fn),Fn) contains two disjoint subtrees, namely Tx and T2. Then the
80
pathvalue for every vertex in T\ is different from the pathvalue for every verÂ¬
tex in T2. Since the graph G is connected, there is at least one edge e â‚¬ E(G)
which connects the two subgraphs Ti and T2 and e ^ E(T\ U T2) = Fn. Let
{ {vh,vki},{vj2ivk2}i â€¢ â€¢ â€¢ Avjqivkq} } be the collection of such edges where v]t â‚¬ T\
and Vk, â‚¬ T2 for i = 1,2, Without loss of generality, let ji < ji for every
i = 2,3,..., q. Then it is clear that {uj,, } â‚¬ Fjl by the procedure of getting the
subgraph (V(Fn), Fn). Then Â£ Fn. This is a contradiction. Therefore the
subgraph (V(Fn), Fn) is connected. The pathvalue in the procedure of getting the
subgraph (V(Fn),Fn) guarantees that the subgraph (V(Fn),Fn) is acyclic. ThereÂ¬
fore (V(Fn), Fn) is a tree. It is clear that, for every i = 1,2, ...,n, the vertex set
V(Fi) contains the vertices u\, u2,..., uÂ¿. Thus V(Fn) contains all the vertices of
G, since V{Fn) D {Ul,u2,.. = V{G). That is, V(Fn) = V = V(G). Hence
the subgraph (V(Fn), Fn) is a maximal tree of G. By the Tree Classification TheÂ¬
orem (Theorem 3.6.8), it is clear that there is a perfect nsequence for the tree
(V(Fn),Fn). 1
Definition 4.5.4 The tree shape of a brain tumor is the tree corresponding to that
tumor, or equivalently, its unique corresponding perfect sequence.
Definition 4.5.5 Two tumors have the same shape if and only if their corresponding
graphs obtained from their respective sphere packing plans are the same.
As in Chapter 2, for any brain tumor, we have a unique sphere packing plan
which can be denoted by a graph. From the previous theorem, we have a unique
perfect sequence for the graph. That is, we can assign exactly one perfect sequence to
each brain tumor. So, if two perfect sequences are distinct, then their corresponding
trees, and therefore their respective corresponding graphs and sphere packings, are
also distinct. That is, if two tumors are represented by distinct perfect sequences,
then their corresponding trees are not isomorphic. And their respective graphs and
81
sphere packing plans are not isomorphic. Thus, we may consider their shapes to be
distinct. Therefore we have the following as our main theorem of this section.
Theorem 4.5.6 The perfect sequences are invariants of the shapes of arbitrary brain
tumors.
4.6 Graph from a Maximal Tree
Let T = {V, E'} be a maximal tree for a graph G = {V, E} which is obtained
from the sphere packing plan of a given tumor or solid shape. We observe that we can
recover the original graph, G, from T. Since V(T) = V = V(G) and E(T) â€” E' C
E = E(G), we need to add the remaining edges to the edge set of the maximal tree to
get the graph G. Since the graph G comes from the sphere packing treatment plan,
each vertex v in G denotes a sphere of the sphere packing plan. Thus, for every
vertex â‚¬ G, let c, = (a;,, y,, zf) and rÂ¿ be the center and the radius of the sphere.
By Proposition 4.3.1, for any two vertices t>, and Vj in V if d(c,,Cj) < jÂ£(rÂ¿ + rÂ¿),
then the edge is in the edge set, E(G), of the graph G. Thus
E' U {{uÂ¿,i>j} â‚¬ V x V\ d(ci,Cj) < }Â¿(r, + r,)} = E.
Therefore we get the graph G = {V, E}.
4.7 Algorithm for Our Maximal Tree
Let G = (V, E) be the graph for a sphere packing plan with V = n.
STEP A Find all the Â¿th step cutvertices of the graph G, for 1 < i < n.
STEP B Label the vertices in the vertex set V by the order explained in Section
4.4.
STEP C By induction, for each i = 1,2, ...,n, get the subgraph (1/(FÂ¿),FÂ¿) of
G. Then T = (V(Fn), Fn) is our maximal tree.
APPENDIX A
SHELLING PROGRAM FOR SPHERE PACKING
In this appendix, we provide the C++ program we developed to get the sphere
packing plan for a given tumor shape. The first section shows the program translating
the data from CT scan to the data of appropriate form for the main program, which
is in the second section. Note that this program does not include the score function
which was developed by Thomas Wagner. We refer the reader Appendix E for the
complete article on the sphere packing treatment plan.
A.l Program for Data Reading
// rewrite data file into x,y,z format
// & produce the total # of data, x,y,z min & max
#include
#include
#include
#include
#include
mainO
{ const int nofdata=3370; // comes from data003.txt
int xpositionfnofdata], ypositionfnofdata], zposition[nofdata];
int q;
float a; float b; float c;
int k=0;
{
>
ifstream inClientFile("data003.txt",
while (inClientFile >> a >> b >> c)
{ xpositionfk] = int(floor(a)); //
ypositionfk] = int(floor(b)); //
zpositionfk] = int(floor(c)); //
k++;
>
ios::in);
check the number
check the number
check the number
int xmin=xposition[0];
int xmax=xposition[0];
int ymin=yposition[0];
int ymax=yposition[0];
int zmin=zposition[0];
int zmax=zposition[0];
for (int i=0; i
{ if ( xpositionfi] < xmin )
xmin = xposition[i];
82
83
>
//
{
>
else
if
( xposition[i]
xmax
=
xposition[i] ;
else
if
( yposition[i]
ymin
=
yposition[i] ;
else
if
( yposition[i]
ymax
=
yposition[i];
else
if
( zposition[i]
zmin
=
zposition[i];
else
if
( zposition[i]
zmax
=
zposition[i];
> xmax )
< ymin )
> ymax )
< zmin )
> zmax )
output the result of centers
ofstream outClientFile("mM003.txt", ios::app);
{ outClientFile << nofdata << setw(10) << xmin << setw(5)
<< xmax << setw(5) << ymin << setw(5) << ymax
<< setw(5) << zmin << setw(5) << zmax << endl;
>
cout << nofdata << setw(5) << xmin << setw(5) << xmax
<< setw(5) << ymin << setw(5) << ymax << setw(5)
<< zmin << setw(5) << zmax;
cin >> q;
>
A.2 Program for Center Searching
// searching the centers for pat 2
#include
#include
#include
Â«include
#include
int square(int); // function prototype
int MAX(int,int,int); // function prototype
main()
{ const
int
nofdata=3370;
//
check
the
number from mM003.txt
const
int
xmin
=
32;
//
check
the
number
const
int
xmax
=
4;
//
check
the
number
const
int
ymin
=
29;
//
check
the
number
const
int
ymax
=
4;
//
check
the
number
const
int
zmin
=
57;
//
check
the
number
const
int
zmax
=
37;
//
check
the
number
const
int
xf ix
=
1xmin;
const
int
yfix
=
1ymin;
const
int
zf ix
=
1zmin;
int xposition[nofdata] , yposition [nofdata], zposition [nofdata] ;
int q;
float a; float b; float c;
int k=0;
84
{ ifstream inClientFile("data003.txt", ios::in);
while (inClientFile >> a >> b >> c)
{ xposition[k] = int(floor(a))+xfix;
yposition[k] = int(floor(b))+yfix;
zposition[k] = int(floor(c))+zfix;
k++;
>
>
const int x = xmax  xmin + 3;
const int y = ymax  ymin + 3;
const int z = zmax  zmin + 3;
const int nc = x*y*z; // number of cells
int cc[nc]; // status value
for (int i=0; i
cc[i] = 3;
int xc[nc]
int yc [nc]
int zc [nc]
// xcoord. of c[nc]
// ycoord. of c[nc]
// zcoord. of c[nc]
// setting the x,y,zcoord. for each cell
for (int i=0; i
{ zc[i] = floor(i/(y*x));
xc[i] = floor((izc[i]*y*x)/x);
yc[i] = izc [i] *y*xxc[i] *x;
>
// setting the boundary of tumor by 1
for (int i=0; i
{ cc[zposition[i]*x*y + xposition[i]*y + yposition[i]] = 1;
// setting the initial of normal by 0
cc[0] = 0;
// setting the normal cell by 0
int h=0;
for (int j=0; j
{ for (int i=0; i
{ if ( cc[i] == 3 )
{ if (cc[i1]==0cc[i+l]==0Icc[iy]==0cc[i+y]==0)
{ cc[i] = 0;
h++;
>
else;
>
else;
>
>
if (h==0)
break;
else
h=0;
// setting the remaining tumor cell by
for (int i=0; i
{ if ( cc[i] == 3 )
cc[i] = 2;
else;
}
// checking
// checking
// checking
// checking
// checking
// setting the boundary cells
int maxrad;
int nofmaxrad;
for (int j=0; j<=MAX(x,y,z)/2+l; j++)
{ int k=0;
for (int i=0; i
{ if (cc[i] == 1 II cc[i] == 2)
{ k++;
if ( cc[il] == j I I cc[i+l] ==j  cc[iy]==j
II cc[i+y]==j II cc[iy*x]==j II cc[i+y*x] =
{ cc[i] = j+1;
nofmaxrad = k;
}
else;
>
else;
}
maxrad = i;
if (k==0)
break;
else;
// checking the no. of nofmaxrad
// if nofmaxrad > 100, then you have change the number
if (nofmaxrad >100)
{ cout << endl << endl << endl << endl << endl <<
<< endl << endl << endl << endl;
cin >> q;
>
else;
// pick the boxes with maxrad
int suprad[l00];
int TotalNofBox4[l00];
int TotalNofMax=0;
for (int i=0; i
{ if (cc[i]==maxrad)
{ suprad[TotalNofMax]=i;
TotalNofMax++;
>
else;
cout << TotalNofMax << endl;
cin >> q;
// Count for the number of centers of each case
for (int k=0; k
{ // reset TotalNofBox4[] to 0
for (int i=0; i<100; i++)
{ TotalNofBox4[i] = 0;
}
TotalNofBox4[maxrad] ++;
int tcc[nc]; // temporary cell notion
for (int i = 0; i
tcc[i] = cc[i]; // reassign tcc[i] for 'for' loop
// remove the biggest sphere
for (int p=0; p
{ int dist = floor(sqrt( square(yc[suprad[k]]yc[p])
+ square(xc[suprad[k]]xc[p])
+ square(zc[suprad[k]]zc[p]) ));
if ( dist < maxrad )
tcc[p] = 0;
else;
}
// setting the remaining tumor cell by 2
for (int i=0; i
{ if ( tcc[i] >= 1 )
tcc[i] = 2;
else;
>
// setting the boundary cells
for (int j=0; j<=MAX(x,y,z)/2+l; j++)
{ int 1=0;
for (int i=0; i
{ if (tcc[ij == 2)
{ if ( tcc[il] == j  tcc[i+l] ==j  tcc[iy]==j
II tcc[i+y]==j  tcc[iy*x]==j  tcc [i+y*x]==j
{ tcc[i] = j + 1;
++1;
>
else;
>
else;
>
if (1==0)
break;
87
else;
>
// check if there is more maxrad pixcel
for (int i=0; i
{ if ( tcc[i] == maxrad )
{ cout << endl << endl << endl <<
" WARNING!!! WE HAVE MORE 'MAXRAD' PIXCEL AT ("
Â« xc[i] Â« Â« yc[i] Â« Â« zc[i] Â« ")"
<< endl;
cin >> q;
break;
>
else;
>
//remove the smaller or = sphere
for (int j=maxradl; j>1; â€”j)
{ for (int i=0; i
{ if ( tcc[i] ==j )
{ TotalNofBox4[j]++;
// remove the sphere
for (int 1=0; l
{ int dist = floor(sqrt( square(ye[i]yc[1])
+ square (xc [i]xc [1] )
+ square (zc [i] zc [1] ) ));
if ( dist < j )
tcc[l] = 0;
else;
>
// setting the remaining tumor cell by 2
for (int i=0; i
{ if ( tcc[i] >= 1 )
tcc[i] = 2;
else;
>
// setting the boundary cells
for (int j=0; j<=MAX(x,y,z)/2+l; j++)
{ int 1=0;
for (int i=0; i
{ if (tcc[ij == 2)
{ if ( tcc[iâ€”1] == j  tcc[i+l] ==j 
tcc[iy]==j II tcc[i+y]==j I I
tcc[iy*x]==j  tcc[i+y*x]==j )
{ tcc[i] = j + 1;
++1;
>
else;
>
else;
}
if (1==0)
break;
88
else;
>
>
else;
>
>
// output the result of centers
{ ofstream outClientFile("pat003NofCenters.txt", ios::app);
{ outClientFile << endl << endl << endl << endl << endl <<
(k+1) << setw(5) << "(" << (xc[suprad[k]]xfix) <<
<< (yc[suprad[k]]yfix) << << (zc[suprad[k]]zfix)
<< ") with radius = " << maxrad << endl;
>
>
for (int j=maxrad; j>1; j â€”)
{ ofstream outClientFile("pat003NofCenters.txt", ios::app);
{ outClientFile << setw(4) << j << setw(4)
<< TotalNofBox4[j] << endl;
>
}
>
// Search for centers
for (int k=0; k
{ int tcc[nc]; // temporary cell notion
for (int i = 0; i
tcc[i] = cc[i]; // reassign tcc[i] for 'for' loop
// output the result of centers
{ ofstream outClientFile("pat003centers.txt", ios::app);
{ outClientFile << (k+1) << endl << setw(5) << "x"
<< setw(5) << "y" << setw(5) << "z" << setw(10)
<< "radius" << endl << setw(5) << (xc[suprad[k]]xfix)
<< setw(5) << (yc[suprad[k]]yfix) << setw(5)
<< (zc[suprad[k]]zfix) << setw(10) << maxrad << endl;
>
}
// remove the biggest sphere
for (int p=0; p
{ int dist = floor(sqrt( square(yc[suprad[k]]yc[p])
+ square (xc [suprad [k]]xc[p])
+ square (zc [suprad [k]]zc[p]) ));
if ( dist < maxrad )
tcc[p] = 0;
else;
>
// setting the remaining tumor cell by 2
for (int i=0; i
C if ( tcc[i] >= 1 )
tcc[i] = 2;
89
else;
}
// setting the boundary cells
for (int j=0; j<=MAX(x,y,z)/2+l; j++)
{ int 1=0;
for (int i=0; i
{ if (tcc[ij == 2)
{ if ( tcc[il] == j I I tcc[i+l] ==j II tcc[iy] ==j
II tcc[i+y]==j II tcc[iy*x]==j II tcc[i+y*x]==j )
{ tcc [i] = j+1;
++1;
}
else;
>
else;
}
if (1â€”0)
break;
else;
}
// check if there is more maxrad pixcel
for (int i=0; i
{ if ( tcc[i] == maxrad )
{ cout << endl << endl << endl <<
" WARNING!!! WE HAVE MORE â€™MAXRADâ€™ PIXCEL AT ("
<< xc[i] << << yc[i] << << zc[i] << ")" << endl;
cin >> q;
break;
}
else;
>
//remove the smaller or = sphere
for (int j=maxradl; j>1; â€”j)
{ for (int i=0; i
{ if ( tcc[i] ==j )
{ TotalNofBox4[j]++;
// output the result of centers
{ ofstream outClientFile("pat003centers.txt",
ios::app);
{ outClientFile << setw(5) << (xc[i]xfix)
<< setw(5) <<(yc[i]yfix) << setw(5)
<< (zc[i]zfix) << setw(10) << j << endl;
}
>
// remove the sphere
for (int 1=0; l
{ int dist = floor(sqrt( square(ye[i]yc [1])
+ square (xc[i]xc[l])
+ square(zc[i]zc[l] ) ));
if ( dist < j )
tcc[l] = 0;
90
else;
>
// setting the remaining tumor cell by 2
for (int i=0; i
{ if ( tcc[i] >= 1 )
tcc[i] = 2;
else;
>
// setting the boundary cells
for (int j=0; j<=MAX(x,y,z)/2+l; j++)
{ int 1=0;
for (int i=0; i
{ if (tcc[i] == 2)
{ if ( tcc[il] == j I I tcc[i+l] ==j II
tcc[iy]==j II tcc[i+y]==j II
tcc [iy*x] ==j  tcc[i+y*x]==j )
{ tcc [i] = j+1;
++1;
>
else;
}
else;
}
if (1â€”0)
break;
else;
}
}
else;
>
>
{ ofstream outClientFile("pat003centers.txt", ios::app);
{ outClientFile << endl;
}
>
}
>
int square(int a)
{ return a*a;
>
int MAX(int a, int b, int c)
{ int p=0;
p = a;
if (a<=b)
p = b;
else if (a<=c)
p = c;
else;
return p;
>
APPENDIX B
PROGRAM FOR THE MAXIMAL TREE OF A GRAPH
In this appendix, we provide the Mathematica program we developed in joint
work with Matthew Harvey. The program below was written by Matthew Harvey to
get the maximal tree for the graph which we assigned for a brain tumor. Note that
this program does not use the method of ordering for the vertices as in Chapter 3.
Instead this program uses some commands which are in the function package in the
Mathematica program library. If anyone wants to develop a program like this using a
different program language, we recommand the vertex ordering algorithm in Chapter
3.
<< DiscreteMath'Combinatorica'
Off[General::spell]
LexPrecedes[a_, b_] := Module[{i},
For[i = 1, i <= Min[Length[a], Length[b]], i++,
If[ a[[i]] ! = b[[i]],
If[ a[[i]] > b [ [i]] , Return[True] , Return [False] ] ]
3;
If[ Length [b] > Length [a], Return [False] ];
Return[True]]
DegreePoint[a_, k_] := Count[Flatten[ToUnorderedPairs[a]], k];
IndSubgraph[Graph[g_, v_], s_List] := Module[{i, j, newmat, sc},
newmat = g;
sc = Complement[Range[Length[v]], s] ;
For[i =1, i <= Length[g], i++,
For [j = 1, j <= Length [g], j++,
If[MemberQ[sc, i]  MemberQ[sc, j], newmat[[i, j]] = 0]]];
Graph[newmat, v]];
CutPoints[gr_] :=
Select[ArticulationVertices[gr], DegreePoint[gr, #] > 2 &]
NonCutPartitions [gr_] := Module[{partitions, cpts, bees, skeleton},
partitions = {};
bees = Select[BiconnectedComponents[gr], Length[#] > 2 &];
partitions = Union[partitions, bees] ;
skeleton =
IndSubgraph[gr,
Union[
Complement[ Range[V[gr]], Flatten[bees] ],
91
CutPoints[gr]
]
];
partitions =
Union[partitions,
Select[ConnectedComponents[
FromUnorderedPairs[
Select[ToUnorderedPairs[
skeleton], # !=
Intersection[#,
Flatten[
Select[BiconnectedCoraponents[skeleton],
Length [#] >2 &]
]
]
&]
]
], Length[#] > 1
&]
];
Table[
Select[
ToUnorderedPairs[gr] ,
# == Intersection[ #, partitions [[i]] ]
&], {i, Length[partitions]}]]
APPENDIX C
PROGRAM FOR CLASSIFYING TREES BY SEQUENCES
In the first section, we provide the Mathematica program we developed in
joint work with Matthew Harvey. This program was written by Matthew Harvey to
get the maximal degree sequences for any given integer n > 2. And we show the
program results up to n = 10 in the second section.
C.l Program
Off[General::spell]
<< DiscreteMathâ€˜Combinatorica*
DoPartitions[n_] :=
Module[{i, k, templist, returnlist},
If[n == 0, {{}},
returnlist = {};
For[k = n, k >= 1, kâ€”,
templist = DoPartitions[n  k] ;
For[i = 1, i <= Dimensions[templist] [[l]], i++,
returnlist = Append [returnlist,
Flatten[{k, templist [[i]] }]
]
]
];
returnlist
]
];
Filter[thelist_] :=
Select[thelist,
! ( MemberQ[{1, 2}, #1[[1]]]  (Max[#l] != First[#1])
II (Count [#1, Max[#l]] > 1)
)
&]
AddZerosToSeq[part_] :=
Module[{nZeros, n, i, j, k, sum, positions, seqs, withzeros},
If [Length [part] == 1, Return [{Table [If [i == 1, part[[l]], 0] ,
{i, part [[l]] + 1>]>]
];
seqs = {>;
sum = Sum[part[[i]], {i, 1, Length[part]>];
nZeros = (sum + 1)  Length[part];
n = Table[k, {k, 3, sum}];
positions = KSubsets[n, nZeros  1];
93
Ford = 1, i <= Length [positions] , i++,
k = 2;
withzeros = {First[part]};
For[j = 2, j <= sum, j++,
If[MemberQ[positions[[i]], j] ,
withzeros = Append[withzeros, 0],
withzeros = Append[withzeros, part[[k]]];
k++
]
];
withzeros = Append[withzeros, 0];
seqs = Append[seqs, withzeros]
];
seqs
];
AddZeros[seq_] :=
Flatten[Table[ AddZerosToSeq[seq[[i]]] , {i, 1, Length[seq]}],1]
SubSeqs[dseq_] :=
Module[ {branchlist, branch, workseq, branches, i},
branches = dseq[[l]];
workseq = Drop[dseq, 1];
branchlist = {};
For[i = 1, i <= branches, i++,
dsum = workseq[[1]];
workseq = Drop[workseq, 1];
branch = {dsum};
While[dsum != 0, dsum += (workseq[[l]]  1);
branch = Append[ branch, workseq[[l]] ];
workseq = Drop[workseq, 1]
] f
branchlist = Append[branchlist, branch]
^ )
branchlist
]
LexPrecedes[a_, b_] :=
Module[{i},
For[i = 1, i <= Min[Length[a], Length[b]], i++,
If[a[[i]] ! = b[[i]],
If [ a[[i]] > b[[i]] , Return [True] , Return [False]
]
]
];
If[Length[b] > Length[a], Return [False] ] ;
Return[True]
]
WeightsQ[dseq_] :=
Module[{i, subseqs, flag},
subseqs = SubSeqs[dseq];
flag = 1;
If[OrderedQ[subseqs, LexPrecedes],
95
For[i = 1, i <= Length[subseqs], i++,
If[ ! WeightsQ[subseqs[[i]]], flag=0, flag*=l ]
],
Return[False]
];
If[ flag == 1, Return[True] , Return[False] ]
]
DegreeSumQ[dseq_] :=
Module [{dsum},
dsum = dseq[[l]];
For[i = 2, i <= Length[dseq], i++,
dsum += dseq[[i]]  1;
If[dsum == 0,
If [i == Length[dseq].Return[True],Return[False]
]
]
];
Return[False]
]
ExtractSubtree[dseq_] :=
Module [{rtotal},
rtotal = dseq[[l]];
i = 1;
While[rtotal > 0,i++;
rtotal += dseq[[i]]  1;
];
Take[dseq, i]
]
RewriteTree [dseq_, i_] : =
Module[{branchesoff, subtree, brlength, seqlst},
If[i == 1, Return[dseq]];
branchesoff = i  1;
While[ ( branchesoff > 1 && Sum[dseq[[k]]  1,
{k, branchesoff, i  1}] < 0 ) 
dseq[[branchesoff]] == 0, branchesoffâ€” ];
brlength = Length[ ExtractSubtree[Drop[dseq, i  1]] ];
subtree = Drop[dseq, {i, i + brlength  1}];
subtree[[branchesoff]]â€”;
seqlst = SubSeqs[Take[dseq, {i, i + brlength  1}]];
seqlst = Append[seqlst, RewriteTree[subtree,branchesoff]];
seqlst = Sort[seqlst, LexPrecedes[#1, #2] &];
Flatten[{dseq[[i]] + 1, seqlst}]
RightOrderQ[dseq_] :=
Module[{pos, seqs, j, rewritten},
pos = Flatten[ Position[dseq, dseq[[l]]  1] ];
If[pos == {}, Return[True]];
seqs = {};
For [j = 1, j <= Length [pos] , j++,
rewritten = RewriteTree[ dseq, pos[[j]] ];
96
If[LexPrecedes[rewritten, dseq] && rewritten != dseq,
Return[False]
]
];
Return[True]
LemmasQ[dseq_] :=
DegreeSumQ[dseq] && WeightsQ [dseq] && RightOrderQ[dseq]
CheckLemmas[lstseq_] :=
Module[{retseq, twoseq, i, j, n, p},
retseq = {};
j = 0;
P = 10;
n = Length[lstseq];
For[i = 1, i <= n, i++,
If[ j > n/10,
Print [p, "'/,"];
p += 10;
j = 0
];
j++;
If [LemmasQ [ lstseq[[i]] ],
retseq = Append[ retseq, lstseq[[i]] ]
]
];
twoseq = {2};
If [Length[lstseq] > 0,
For[i = 2, i < Length[ retseq[[l]] ]  1, i++,
twoseq = Append[twoseq, 1]
]
];
twoseq = Join[twoseq, {0, 0}];
retseq = Append[retseq, twoseq];
retseq
]
DSTRecursive[dseq_, bvert_, hvert_ , trparam_ ] :=
Module[{i, v, tree, tempseq},
v = hvert;
tree = trparam;
If[dseq != {0}, tempseq = Drop[dseq, 1];
For[i = 1, i <= dseq[[l]], i++,
v++;
tree = Append[tree, {bvert , v}];
subseq = ExtractSubtree[tempseq];
tempseq = Drop[tempseq, Length[subseq]];
{tree, v} = DSTRecursive[subseq, v, v, tree];
];
{tree, v}
97
DegreeSeqToTree[dseq_] :=
First[ DSTRecursive[dseq, 1, 1, {}] ] ;
ToGraphics[ckd_] :=
For[t = 1, t <= Length[ckd], t++,
Print [ckd [ [t]]];
ShowGraph[RadialEmbedding[FromUnorderedPairs
[ DegreeSeqToTree[ckd[[t]]] ]
]
]
]
C.2 Some Program Results
The notation for a degree sequence is (d\, di,..., dn). But the Mathematica
program produces the sequence notation with { and }. So in this section, any set
notation is actually denotes the degree sequence. Note that an ntree has n vertices,
so there are n1 edges.
n=3 total : 1
{2, 0, 0}
n=4 total : 2
C3, 0, 0, 0}, {2, 1, 0, 0>
n=5 total : 3
{4, 0, 0, 0, 0>, {3, 1, 0, 0, 0},
{2
1
1, 0, 0}
98
n=6 total : 6
{5, 0, 0, 0, 0, 0}, {4, 1, 0, 0, 0, 0>, {3, 2, 0, 0, 0, 0},
{3, 1, 0, 1, 0, 0>, {3, 1, 1, 0, 0, 0}, {2, 1, 1, 1, 0, 0}
n=7 total : 11
{6,
o,
0,
0,
0,
0,
0>,
{5,
1,
o,
0,
o,
0,
0>, {4,
2,
o,
o,
o,
o,
o>,
{4,
1,
0,
1,
0,
0,
0>,
{4,
1,
1,
0,
o,
0,
0>, {3,
2,
1,
o,
0,
o,
0},
{3,
1,
2,
0,
0,
0,
0>,
{3,
1,
o,
1,
0,
1,
0>, {3,
1,
1,
0,
1,
o,
0},
{3,
1,
1,
1,
0,
0,
0>,
{2,
1,
1,
1,
1,
0,
0}
/S rS rS rS rS r"S rS rS rS rS rS rS
tocococococoÂ£Â».fÂ»4*uiai'i
PPKKIOIOHKIOl'MO
lâ€˜lk0M'00000000
(â€˜â– OiiO>kOiâ€˜OtkOOC>
(â€˜(â€˜â– OOOOOOOOOO
OOMOOOOOOOOO
oooooooooooo
S'S'Sâ€™Sâ€™S'^'rJSâ€™lrJS''rJS'
rS rS rS rS rS rS rS rS rS rS rS CO
COCOCOCOCO^>4^^tfÂ»CJ105
iâ€”lâ€”L lâ€”k lâ€”*â– lâ€”k â–ºâ€”k CD lâ€”k CD CD O
(kOlOikOH^KkOOi^0
t'i'OOOOOOOOO
Oh^OOt^Ot^OOOO
ooooooooooo
ooooooooooo
S'S'Sâ€™S'S'Sâ€™S'Sâ€™S'Sâ€™Sâ€™
CO
CO
n=8 total
total
O
o
r"S r"S rS rS rS rS rS râ€œS rS rS r"S rS rS rS rS rS r"S rS rS rS râ€œS rS rS
O
o o o o o
o o o o o
o o o o o o o
O
o
o
o
o
O
o o o o o
o o o o o
o o o o o o o
O
o
tH
o
o
o
o o o o o
O th o o o
O th O O O th O
O
o
o
tH
o
o
O O O TH o
O O 1â€”1 o th
O O tH tH O O O
tH
o
tH
tH
th
o
o o o o o
CN Tl O O O
O tH O O O O rl
o
tH
tH
O
tH
o
H O O H tâ€”1
O O O th O
(NOHOCNhh
tH
CN
O
tH
tH
o
O O th O th
O O H rH (N
tH tâ€”1 tH CN tH tâ€”1 tH
CN
tH
tH
tH
tH
tH
h CO CN H H
CN CN CN CN H
t4 rH rH CN CN CN CN
th
tH
tH
tH
tH
N
(Ã¼ininirun^'Ã'Ã^^
N1 N1 "Ã n oo n to
co
co
co
co
co
S'Sâ€™S'S'S'S'S'S'S'S'S'S'S'S'S'S'S'S'S'S'S'S'S'
rS rS rS rS rS râ€œS rS rS rS rS rS rS r*S rS rS rS rS rS rS rS rS rS rS rS
O
O O O O O
O O O O O
O O O O O O O
o
o
o
o
o
o
O
o o o o o
o o o o o
O rH O O O O O
o
o
o
tH
o
o
o
o o o o o
o o o o o
o o o o o th o
tH
o
o
O
tH
tH
o
o o o o o
o o th o o
O th iH o O O th
o
o
o
tH
O
tH
o
O O iH o th
O O th th O
O O th th O th O
O
tH
CN
O
tH
tH
o
o o o o o
o o o o o
rH tH CD tH tH O tH
Tâ€”l
tH
th
tH
tH
tH
O
O TH O CN tH
H CN O TH CO
CN O H th CN th th
CN
CN
tH
tH
tH
tH
O
CN H CM H H
CO CN CN CN H
th tâ€”i th tâ€”i CN CN CN
th
tH
tH
tH
tH
tH
co
CD CD LO LO LO
^ ^ ^ ^ ^
^ ^ ^ ro co co
co
co
co
co
co
CN
SÂ»
wv w
SÂ» S* S* SÂ» SÂ»
Sr* SÂ» SÂ» SÂ» SÂ» SÂ» SÂ»
SÂ»
SÂ»
SJ
SÂ»
SÂ»
SrJ
=10 total : 106
rS <S r** i*t i*v rS rS rS rS rS rS rS rS rS rS rS rS rS rS rS rS rS
ooooooooooooooooooooooo
OOOOOOOOOOOOOOOOOOOOOtHO
O
0,
O
o
O
O
O
o
o
o
o
o
o
o
o
0,
o
o
0,
o
o
o
o
o
o
o
o
O
O
o
o
tH
o
o
o
o
tH
o
o
o
o
tH
o
o
tH
tH
o
o
o
o
tH
o
tH
o
o
tH
o
tH
o
o
tH
o
o
o
o
o
CN
o
tH
o
o
o
o
O
o
o
CN
tH
o
o
o
o
tH
o
o
tH
o
o
o
o
tH
tH
o
tH
o
o
tH
tH
o
o
o
o
tâ€œH
o
CN
o
tH
o
o
o
o
tH
o
o
o
o
o
o
tH
O
tH
o
o
o
tH
tH
CN
tH
tH
tH
CN
tH
co
CN
CN
tH
o
o
tH
tH
co
CN
tH
tH
co
CN
CN
CN
CN
tH
tH
tH
tH
co
co
CN
CN
CN
CN
CN
CN
costDtDtott>LOioiou)ioir)ininiO'Ã^<Ã^<#^^^<
Sâ€™Sâ€™S'Sâ€™S'S'Sâ€™Sâ€™Sâ€™S'S'Sâ€™S'S'S'S'S'S'Sâ€™Sâ€™S'S'S'
r"S rS r"S rS râ€œS r"S rS rS rS r"S rS rS rS rS rS <S râ€œS rS râ€œS râ€œS rS rS rS
O
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
tH
o
o
o
o
tH
o
o
o
tH
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
tH
o
o
o
o
o
o
o
o
o
o
o
o
o
o
o
tH
o
o
o
tH
tH
o
o
o
o
tH
CN
o
tH
o
o
o
tH
o
tH
o
o
o
tH
tH
o
o
o
tH
tH
o
o
CN
o
tH
o
tH
o
o
o
o
o
o
o
o
o
o
o
o
tH
tH
o
tH
o
tH
o
o
o
CN
o
o
o
tH
o
CN
tH
o
tH
CN
o
tH
co
CN
o
tH
tH
tH
tH
o
CN
o
tH
o
o
CN
tH
CN
tH
tH
co
CN
CN
CN
tH
tH
tH
tH
tH
co
co
CN
CN
CN
CN
CN
'r^S,Sâ€™S'S,Sâ€™Sâ€™S'S'Sâ€™Sâ€™Sâ€™S,Sâ€™Sâ€™S,Sâ€™S'Sâ€™S''rJSâ€™S'
o
rS rS rS rS rS rS rS rS rS rS rS rS rS rS rS rS rS rS rS rS rS r"S rS rS rS rS rS rS rS rS
oooooooooooooooooooooooooooooo
OOOOOOOOOOOOOOOOOOOOOOOOOOOOOO
000000000O00000000OO0O0O0000OO
OOOOOOOOOOOOOOOOOOOOOOOOOOOOOO
OOOOOOOOOOOCNOOOOOOOOOOOOOCNOOOO
O O O O O O O O O O O O O O O O CN O O O O O O O o o O O O O
OrlTHC'IOHrOCSOHrlOOrlrlC'JrlHHdrH'HTHCNCSHrHHrlrl
tHHHCNC'ICNHtHHHtHCNCNCNCNâ€™HtHtHHHHCNCNhhhhhhh
CNCNCNHHHTHHrHHHCNCNCNCNCNlCNCNCNCNCNHHHHHrHrHrHH
^^^^^^^^^^^COCOCOCOCOCOCOCOCOCOCOOOCOCOCOCOCOCOCN
vvvvvvvvvvvwvvvvvvvvvvvvvvvvv
rS râ€œS rS rS rS rS râ€œS rS rS rS rS rS rS rS râ€œS rS rS rS rS rS rS rS rS rS rS rS r^Â» rS rS rS
OOOOOOOOOOOOOOOOOOOOOOOOOOOOOO
oooooooooooooooooooooooooooooo
â– o
o
o
O
â– O
O
o
o
O
O
o
o
o
O
O
â– o
o
O
O
O
O
O
â– o
O
O
O
rH
s~H
O
O
o
â– o
o
o
O
â– o
o
sH
o
O
o
o
o
O
o
O
o
â– o
O
o
o
o
O
o
tH
O
O
â€¢O
â€¢O
o
o
o
â– o
o
â– O
O
o
O
o
o
o
o
o
O
tH
O
o
O
O
o
O
o
O
â– O
O
â– o
rH
rH
O
â€¢o
o
H
o
o
o
o
â– O
o
CN
â– o
o
o
o
o
O
o
â– o
â– o
o
O
â– o
es
â– o
â– o
rH
es
O
O
â– O
â– o
o
O
H
â€¢o
o
â– O
â– O
CN
o
o
o
o
CN
â– o
o
CN
CN
o
rH
H
o
â– o
H
â– o
CN
O
O
â– o
â€¢o
â– o
o
â€¢o
â– O
co
CN
es
es
H
sH
tH
â– o
â– o
es
es
es
â– o
â€¢O
â– o
â– O
â– O
o
es
es
es
tH
â– O
â– o
â– O
â– o
â– o
CN
es
es
rH
rH
â– O
â– O
sâ€œH
tH
o
H
es
es
es
es
es
es
es
es
es
O
O
O
â€¢O
â– o
tH
rH
â€¢o
rH
^^^^^^^^^^^^cocococococococococococococococococo
WW W WVVVVVV W WWW vvw vvvv
end of data
APPENDIX D
LABELED GRAPHS AND TREES
D.l Number of Labeled Graphs and Labeled Trees
To obtain the number of labeled graphs with n vertices, we need only observe
that each of the (â€) possible edges is either present or absent. Thus we have the
following theorem:
Theorem D.1.1 [5]
The number of graphs on n labeled vertices is
In LovÃ¡sz [6], the Priifercode is used to denote a labeled tree. And, for any
given Priifercode, one can construct the tree which is denoted by that code. Then,
using the properties of the Priifercode, Cayley proved the following theorem:
Theorem D.l.2 (Cayley) [8] The number of trees on n labeled vertices is nn~2.
D.2 Sketch of the Construction of a Priifercode
Among the end vertices of a given labeled tree, pick the one whose label is
minimal. Cut it off from the tree and write down the label of its adjacent vertex.
Repeat this process until only one edge remain. At this point we have a sequence of
n â€” 1 numbers written down. We call this resulting sequence the Priifercode of the
tree.
The following example shows how we construct the Priifercode of a given
tree.
Example D.2.1 Let T be the tree shown below.
3 6
2
1
5
7
8
4
Figure Dl.
There are 5 end vertices, 2, 3,4,6,8. We cut off the smallest labeled end vertex
which is 2. Then the sequence is (1) and the remaining tree is Figure D2 below.
Then we cut off the next smallest labeled end vertex which is 3, and add the
label, 1, of the adjacent vertex of 3 to the sequence. That is, the sequence is (1,1).
From the remaining tree we cut off the next smallest labeled end vertex which is 4.
Then the sequence is (1,1,1). And we get the tree shape in Figure D3.
102
103
3 6
15 7 8
4
Figure D2.
6
15 7 8
Figure D3.
Among the 3 end vertices, 1 is the smallest labeled one. So we cut off the
vertex 1 and write down the label, 5, of the adjacent vertex of 1. The sequence then
will be (1,1,1,5) and the remaining tree is Figure D4.
6
5 7 8
Figure D4.
The remaining tree has two end vertices, that is, it is a linear shape. We cut
off the smaller labeled vertex, 6. And the sequence will be (1,1,1,5,5) . After we cut
off the next smaller labeled vertex which is 5, and get the sequence (1,1,1,5,5, 7) ,
the remaining tree has only two vertices. Itâ€™s clear we have to cut off the smaller
labeled vertex which is 7 and the adjacent vertex will be 8. Therefore we do not need
to add this number to the sequence. Hence the last sequence, (1,1,1,5,5, 7), is the
Priifercode of the given tree T.
If we assign different labels for the tree above, then we will get a different
Priifercode. But the code must consist of 6 terms which is two less than the total
number of vertices in the tree. And each term can be any number among 1,2,..., 8.
Thus we have the following theorem.
D.3 Reconstruction of a Tree from a PrÃ¼fercode
Let (al5 02,..., aâ€ž2) be a Priifercode. That is, we expect an ntree. Let
6i, Â¿>2, â€¢ â€¢ â€¢, 5n_i be the indices of the removed points. Then, for any i, 6, is obviously
different from 61,62,... ,6,_i,a,. Also, 6, ^ aj if j > i, because 6, is removed in
the ith step, so it can not be an adjacent vertex later.
Conversely, if k Â£ {61? 62,..., 6,_x, a,, a,+1,..., an_2}, then Vk is an end verÂ¬
tex of V(T) â€” {veÂ»,, v&2, â€¢ â€¢ â€¢, iâ€™fe.â€”i} Thus
104
6, = min{A:A: Â£ {61, b2,..., 6,1, a,, ai+i,..., aâ€ž_2}.
Let the sequence (1,1,1,5,5,7) be a Priifercode. That is, a\ = l,a2 =
l,a3 = l,a4 = 5,a5 â€” 5, o6 = 7. Then b\ â€” min{A;A: ^ {ai, a2,..., a6}} =
min{fcA; 0 {1,5,7}} = min{2,3,4,6,8} = 2. That is, there is an edge between ax =
1 and b\ = 2. Then 62 = min{A;A; ^ {6i, a2,..., a6}} = min{A;A; ^ {2,1,5,7}} =
min{3,4,6,8} = 3. Thus the tree T has an edge between a2 = 1 and 62 = 3. In a
similar manner, we have the following:
63 = min{A:fc ^ {6i,62,a3,... ,a6}}
= min{fcA; ^ {2,3,1,5, 7}} = min{4, 6,8} = 4,
(Thus T has an edge {Â«3,63} = {1,4})
64 = min{fcfc ^ {61,62,53,a4,a5,a6}}
= min{A:A: ^ {2,3,4,5,7}} = min{l,6,8} = 1,
(Thus T has an edge {a4,64} = {5,1})
65 = min{A:A: ^ {61,62,63,64, a5,a6}}
= min{A:A; 0 {2,3,4,1,5, 7}} = min{6,8} = 6,
(Thus T has an edge {05,65} = {5,6})
b6 = min{A;A: ^ {6l5 62,63,64,65,05}}
= min{fcfc ^ {2,3,4,6, 7}} = min{5,8} = 5,
(Thus T has an edge {a6,66} = {7,5}).
And, for the vertex a& = 7, the largest labeled vertex u8 will be the remaining
adjacent vertex, that is, there is an edge {7,8}.
So the tree T has the vertex set V = {1,2,3,4,5,6, 7,8} and the edge set
E = {{1,2}, {1,3}, {1,4}, {1,5}, {5,6}, {5, 7}, {7,8}}, which is the tree shape in the
example above. That is, we can construct a labeled tree from a Priifercode. Clearly,
different Priifercodes produce different labeled trees.
BIOGRAPHICAL SKETCH
TaeIl Yi was born December 16, 1962 in Seoul, Korea. He attended the
Dankook University from 1981 to 1988, earning a Bachelor of Science degree in mathÂ¬
ematics education in February 1988. During college, he was in the service of the
Korean Army for 27 months. After college, he entered graduate school at Dankook
University in March 1988, and completed a Master of Science degree in mathematÂ¬
ics in Feburary 1990. During that time he worked at Yongmoon High School as a
mathematics teacher.
He married MoonSil Kim on June 27, 1987. Their first son, HanYong
(David), was born April 12, 1988, and their second son, ChangYong (Peter), was
born May 3, 1994.
In August 1991 TaeIl Yi entered graduate school in the Mathematics DeÂ¬
partment at the University of Illinois at Urbana/Champaign. After receiving the
degree of Master of Science in August 1994, he moved to graduate school in the InÂ¬
struction and Curriculum Department of the College of Education at the University
of Florida, and in August 1995 transfered to the University of Florida, Mathematics
Department. He pursued a concurent degree in the College of Education and received
a degree of Master of Education in December 1997. During his doctoral work in the
Mathematics Department, he served as a graduate teaching assistant.
He received a Russell Corporation Endowed Dissertation Scholarship Spring
2000. He is a member of the Phi Kappa Phi honor society.
105
I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and is fully adequate, in scope and
quality, as a dissertation for the degree of Doctor of Philosophy.
^ ^
Beverly LHsrechner, Chairman
Professor of Mathematics
I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and is fully adequate, in scope and
quality, as a dissertation for the degree of Doctor of Philosophy.
5hilip Boyland
Associate Professor of Mathematics
I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and is full/ adequate, in scope and
quality, as a dissertation for the degree of Doctor of PhilosophyÂ»
^sling
lessor of Mathematic
I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and is fully adequate, in scope and
quality, as a dissertation for the degree of Doctor of Philosophy.
WU
LiChien Shen
Professor of Mathematics
I certify that I have read this study and that in my opinion it conforms to
acceptable standards of scholarly presentation and is fully adequate, in scope and
quality, as a dissertation for the degree of Doctor of Philosophy.
Mary G. Kanxowskr
Professor of Teaching and Learning
This dissertation was submitted to the Graduate Faculty of the Department of
Mathematics in the College of Liberal Arts and Sciences and to the Graduate School
and was accepted as partial fulfillment of the requirements for the degree of Doctor
of Philosophy.
August 2000
Dean, Graduate School
xml version 1.0 encoding UTF8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchemainstance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID EG4D341NG_QRXNCI INGEST_TIME 20140321T17:53:03Z PACKAGE AA00018872_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES

