Citation
Adsorption of substituted aromatic compounds by activated carbon

Material Information

Title:
Adsorption of substituted aromatic compounds by activated carbon : a mechanistic approach to quantitative structure activity relationships
Creator:
McElroy, Jennifer A. ( Dissertant )
Mazyck, David W. ( Thesis advisor )
Place of Publication:
Gainesville, Fla.
Publisher:
University of Florida
Publication Date:
Copyright Date:
2005
Language:
English

Subjects

Subjects / Keywords:
Activated carbon ( jstor )
Adsorption ( jstor )
Carbon ( jstor )
Carbon dioxide ( jstor )
Chemicals ( jstor )
Contaminants ( jstor )
Correlations ( jstor )
Electrons ( jstor )
Functional groups ( jstor )
Isotherms ( jstor )
Dissertations, Academic -- UF -- Environmental Engineering Sciences
Environmental Engineering Sciences thesis, M.E

Notes

Abstract:
Because of their widespread prevalence in ground and surface waters, aromatic compounds pose a significant risk to public health. Of the current biological, chemical, and physical methods for remediation of these contaminants, activated carbon has been chosen as the primary method of treatment by many water treatment facilities. Despite its widespread use, questions remain concerning the economic feasibility of activated carbon, and determining suitable carbons for specific contaminants often requires extensive experimentation, time, and cost. Previous work in developing predictors of carbon performance using quantitative structure-activity relationships (QSARs), has fallen short of providing accurate models for carbon adsorption. This study investigates the role of QSARs in predicting the adsorption of monosubstituted benzenes onto a single activated carbon and offers a general protocol for the implementation of QSARs for predictive adsorption analysis. This study has produced successful correlations among aqueous solubility, sterics, and log K. These correlations demonstrate optimistic effectiveness for using these parameters for adsorption prediction and signify the value of the functional and effective QSAR protocol utilized in this study.
Subject:
activated, adsorption, aromatic, benzene, carbon, correlation, QSAR, QSARs, substituted
General Note:
Title from title page of source document.
General Note:
Document formatted into pages; contains 81 pages.
General Note:
Includes vita.
Thesis:
Thesis (M.E.)--University of Florida, 2005.
Bibliography:
Includes bibliographical references.
General Note:
Text (Electronic thesis) in PDF format.

Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
Copyright McElroy, Jennifer A.. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
Embargo Date:
5/1/2005
Resource Identifier:
003322317 ( aleph )
71230805 ( OCLC )

Downloads

This item has the following downloads:

hobbs_j ( .pdf )

hobbs_j_Page_17.txt

hobbs_j_Page_65.txt

hobbs_j_Page_62.txt

hobbs_j_Page_37.txt

hobbs_j_Page_05.txt

hobbs_j_Page_71.txt

hobbs_j_Page_46.txt

hobbs_j_Page_24.txt

hobbs_j_Page_33.txt

hobbs_j_Page_44.txt

hobbs_j_Page_49.txt

hobbs_j_Page_53.txt

hobbs_j_Page_79.txt

hobbs_j_Page_40.txt

hobbs_j_Page_06.txt

hobbs_j_Page_15.txt

hobbs_j_Page_77.txt

hobbs_j_Page_21.txt

hobbs_j_Page_51.txt

hobbs_j_Page_08.txt

hobbs_j_Page_38.txt

hobbs_j_Page_55.txt

hobbs_j_Page_81.txt

hobbs_j_Page_02.txt

hobbs_j_Page_73.txt

hobbs_j_Page_19.txt

hobbs_j_Page_26.txt

hobbs_j_Page_78.txt

hobbs_j_Page_42.txt

hobbs_j_Page_63.txt

hobbs_j_Page_28.txt

hobbs_j_Page_27.txt

hobbs_j_Page_61.txt

hobbs_j_Page_18.txt

hobbs_j_Page_52.txt

hobbs_j_Page_36.txt

hobbs_j_Page_22.txt

hobbs_j_Page_57.txt

hobbs_j_Page_23.txt

hobbs_j_Page_01.txt

hobbs_j_Page_29.txt

hobbs_j_Page_35.txt

hobbs_j_Page_20.txt

hobbs_j_Page_10.txt

hobbs_j_Page_41.txt

hobbs_j_Page_64.txt

hobbs_j_Page_30.txt

hobbs_j_Page_43.txt

hobbs_j_Page_80.txt

hobbs_j_Page_12.txt

hobbs_j_Page_56.txt

hobbs_j_Page_67.txt

hobbs_j_Page_32.txt

hobbs_j_Page_34.txt

hobbs_j_Page_76.txt

hobbs_j_Page_69.txt

hobbs_j_Page_11.txt

hobbs_j_Page_58.txt

hobbs_j_Page_74.txt

hobbs_j_Page_59.txt

hobbs_j_Page_03.txt

hobbs_j_Page_50.txt

hobbs_j_Page_75.txt

hobbs_j_Page_25.txt

hobbs_j_Page_13.txt

hobbs_j_Page_09.txt

hobbs_j_Page_39.txt

hobbs_j_Page_07.txt

hobbs_j_Page_04.txt

hobbs_j_pdf.txt

hobbs_j_Page_66.txt

hobbs_j_Page_48.txt

hobbs_j_Page_14.txt

hobbs_j_Page_72.txt

hobbs_j_Page_54.txt

hobbs_j_Page_70.txt

hobbs_j_Page_31.txt

hobbs_j_Page_60.txt

hobbs_j_Page_68.txt

hobbs_j_Page_16.txt

hobbs_j_Page_45.txt

hobbs_j_Page_47.txt


Full Text












ADSORPTION OF SUBSTITUTED AROMATIC COMPOUNDS BY ACTIVATED
CARBON: A MECHANISTIC APPROACH TO QUANTITATIVE STRUCTURE
ACTIVITY RELATIONSHIPS
















By

JENNIFER A. MCELROY


A THESIS PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
MASTER OF ENGINEERING

UNIVERSITY OF FLORIDA


2005

































Copyright 2005

by

Jennifer A. McElroy



























Knowledge and understanding is a gift from God to be used for His glory.





"Whatever you do, work at it with all your heart, as working for the Lord, not for men,
since you know that you will receive an inheritance from the Lord as a reward. It is the
Lord Christ you are serving." Colossians 3:23-24















ACKNOWLEDGMENTS

The utmost acknowledgment goes to my Lord and Savior Jesus Christ, for without

Him nothing would have existence or find purpose. He has guided my steps and blessed

me with opportunities that I pray will be used for His glory. With all my heart I thank my

husband, Steve McElroy, for the wonderful blessing and comfort that he is in my life.

Heartfelt gratitude goes out to my parents, Gary and Sandra Hobbs, who have always

been an amazing source of strength and encouragement to me, and to my brother

Matthew who has taught me courage and perseverance.

Without the help of my professors I would have never come this far. Dr. David

Mazyck has been an astounding source of encouragement and has taught me to strive for

excellence. Dr. Angela Lindner has been a wonderful source of support and has been

crucial in the development of this research. I would also like to sincerely thank all of the

scientists who helped me in various stages of my research, Dr. Susan Sinnott, everyone at

Engineering Performance Solutions: Rick, Ron, and Matthew Tennant, my fellow

graduate researchers: Ameena Khan, Christina Ludwig, Morgana Bach, Jennifer Stokke,

Thomas Chestnutt, Jack Drwiega, and Vivek Shyamasundar. Special gratitude and

acknowledgment are extended to Maria Paituvi for her noteworthy help and dedication to

this work.
















TABLE OF CONTENTS

page

A C K N O W L E D G M E N T S ................................................................................................. iv

LIST O F TA B LE S ............................ ........... ..... ......... ............ .. vii

LIST OF FIGURES ...................................................... ................... viii

ABSTRACT .............. ......................................... ix

CHAPTER

1 IN TRODU CTION ................................................. ...... .................

2 LITER A TU R E REV IEW ............................................................. ....................... 2

A activated C arbon ....................................................... .......... .......... .... ....
The A ctivation Process ......................................... ........... ... .. ........ ..
C O 2 and H 20 A ctivation ............................................... ............................ 3
A d so rp tio n F acto rs........................................................................................................ 4
Surface A rea ................................................................................................. 4
P ore Size D istribution........ .................................................. ........ ... ...........5
p H ........................ .................... ...... ................................ .. ............ .. .. 5
Surface Chemistry ..................................... ......... ................. .6
Predicted A dsorption M echanism s...................................... ........................... ........7
7t-7t B onding Theory ..................... .. .. ...................... .. ....... .......... ...... .
Electron D onor Theory........................................................... ............9
F unctional G roup F orm ation ........................................................... .....................10
D election of Functional Groups ........................................ ....... ............... 10
Types of Functional Groups Formed........ ........................................11
From CO 2 activation ............. .................. ........ .......... ........11
From H 20 activation ............................................................................. 12
Factors controlling which groups are formed ...........................................12
Precursor and conditions of activation...................................................... 13
Optimizing the Use of Activated Carbon ........................ ......... ................15
Quantitative Structure A activity Relationships ......................................... .................23







v









3 M ATERIALS AND M ETHODS ........................................ ......................... 28

Isoth erm M eth o d s ............................................................................ ..................... 2 8
D evelopm ent of Isotherm D ata ..... .......... ........ ..............................................29
T arget C ontam inants ..................... .. .... ................... .... .. ........... 29
B e n z en e ................................................................2 9
Substituted benzenes ............................................................................ 30
Freundlich Isotherm s ......... ............ ...... ....... .... ...... .......... 31
Langmuir Isotherms......... ......... ........ ...........................32
K inetic Stu dy M methods ....................................................................... ..................32
Developm ent of Kinetic Equations.............. ...................................... .............. 33
QSAR Development .............................................. ........................... 33

4 M A N U SC R IPT ....... ....... ...... ...................................... .. .. .. ...... .............. .. 38

Introduction ........................... ..............................38
E x p e rim e n ta l .................................................................................................4 6
A dsorption Isotherm s ............................................... .............................. 46
K inetic D ata C collection ............................. ............................ ............. .49
E stablishm ent of Q SA R ........................................................... .....................49
D iscu ssion of R esults......... ............................................................ .. .... .. .. 53

5 CONCLUDING REMARKS ............. .... ................................. 64

L IST O F R E FE R E N C E S ........... ........................................................ .......................... 66

B IO G R A PH IC A L SK E T C H ..................................................................... ..................71
















LIST OF TABLES


Table pge

2.1 Elemental Analysis of Apricot Stones Before and After Steam Activation at
6 0 0 0C .............................................................................. 14

2.2 Reported Values of Freundlich K for Benzene. ................. .......... ............... 17

2.3 Example of Parameters Commonly Used in QSAR Development .....................25

3.1 Target Contaminants Used in Isotherm and Kinetic Studies. ................................29

3.2 Selected Parameters Used in the QSAR Training Set........................................... 35

4.1 Size and Aqueous Solubility of M IB and Phenol ................................................40

4.2 Selected Physical Properties of F400...... ................................. ...............42

4.3 Reported Values of Freundlich K for Benzene. ......... ....................................... 44

4.4 Experimentally Derived Values for log K, log Q, and log t...............................49

4.5 Selected Descriptors of Substituent or Compound Electronics, Sterics, and
H ydrophobicity............ .. ....... ........................................................... 51

4.6 Selected Results of QSAR Analysis with log K, log Q, and log t as the
D dependent V ariables. ............................................. .............. ... 55
















LIST OF FIGURES


Figure pge

2.1 Representation of Point of Zero Charge ......................................... ...............6

2.2 Representation of sigma and pi bonding, arrows within the atoms of
formaldehyde represent electron filling in outer orbitals ...........................................8

2.3 Representation of the Benzene Pi System (Source: Tedankara Library, 2003).........8

3.1 Craig Plot Illustrating the Parameter Spread of a Hydrophobicity Constant (7)
and the H am m ett Constant (Cind) ....................................... .................................34

4.1 Example of Isotherm Development............ ............... ..... .... ............. ...... 48

4.2 QSAR Correlation 1: Predicted versus Observed Values of log K.....................56

4.3 QSAR Correlation 2: Predicted versus Observed Values of log K.....................57

4.4 QSAR Correlation 3: Predicted versus Observed Values of log Q.....................58

4.5 QSAR Correlation 4: Predicted versus Observed Values of log K.....................59

4.6 QSAR Correlation 5: Predicted versus Observed Values of log t..........................60

4.7 QSAR Correlation 6: Predicted versus Observed Values of log t..........................61















Abstract of Thesis Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Master of Engineering

ADSORPTION OF SUBSTITUTED AROMATIC COMPOUNDS BY ACTIVATED
CARBON: A MECHANISTIC APPROACH TO QUANTITATIVE STRUCTURE
ACTIVITY RELATIONSHIPS

By

Jennifer A. McElroy

May 2005

Chair: David Mazyck
Major Department: Environmental Engineering Sciences

Because of their widespread prevalence in ground and surface waters, aromatic

compounds pose a significant risk to public health. Of the current biological, chemical,

and physical methods for remediation of these contaminants, activated carbon has been

chosen as the primary method of treatment by many water treatment facilities. Despite

its widespread use, questions remain concerning the economic feasibility of activated

carbon, and determining suitable carbons for specific contaminants often requires

extensive experimentation, time, and cost. Previous work in developing predictors of

carbon performance using quantitative structure-activity relationships (QSARs), has

fallen short of providing accurate models for carbon adsorption. This study investigates

the role of QSARs in predicting the adsorption of monosubstituted benzenes onto a single

activated carbon and offers a general protocol for the implementation of QSARs for

predictive adsorption analysis. This study has produced successful correlations among

aqueous solubility, sterics, and log K. These correlations demonstrate optimistic









effectiveness for using these parameters for adsorption prediction and signify the value of

the functional and effective QSAR protocol utilized in this study.














CHAPTER 1
INTRODUCTION

In 1969, Mattson et al. stated that "The mechanisms by which active carbon

functions to remove phenols from aqueous solutions has never been clearly defined. A

thorough understanding of the sorptive mechanisms is essential for accomplishing the

most effective use of active carbon in water and wastewater treatment" (Mattson et al.

1969). Over 3 decades later, carbon scientists are still pondering the same dilemma. It is

clear that activated carbon is an efficient, cost effective adsorbent for removing numerous

organic materials from both gaseous and aqueous solutions. The general mechanisms of

adsorption are known to be physical adsorption and chemisorption. It has also been

shown that factors, such as pH and surface chemistry, play a significant role in the

adsorption process (Kipling and Shooter 1966, Coughlin et al. 1968, Mattson et al. 1969,

Muller et al. 1985, Vidic et al. 1993, Radovic 2001). Yet, the mechanisms that directly

lead to the adsorption of an adsorbate still remain elusive. Identifying the role of

functional groups in the adsorption process is an essential component to understanding

these mechanisms. The carbon activation process is the fundamental step in the

development of surface chemistry and functional groups. Understanding the adsorption

process will enable an optimized use of activated carbon, eliminating the wasteful

practice of trial and error adsorption treatment. To help circumvent this black-box

method of carbon dosing, various relationships are currently being established between

the adsorbate and the adsorbent. Two of these relationships are adsorption isotherms and

Quantitative Structure Activity Relationships (QSARs).














CHAPTER 2
LITERATURE REVIEW

Activated Carbon

Activated carbon is an adsorbent created by treating a carbonaceous material

thermally and/or chemically so that it develops internal pores that yield its characteristic

high surface area. Thermal activation is commonly employed, where a carbonaceous

precursor (e.g., coal, wood, nutshells) is subjected to high temperatures typically ranging

from 300C to 800C (Mattson et al. 1969). In chemical activation, the precursor is first

exposed to a chemical (e.g., NaOH, H3PO4), which drives pore development and is then

subjected to heat treatment. Heating temperatures are typically lower for the production

of chemically activated carbon. Activated carbon is commonly used in separation

processes involving organic compounds, and is widely used in fields such as medicine,

water and air treatment, the food and beverage industry, and industrial processes.

The Activation Process

During the activation process, an oxidizing agent breaks down the graphite

structures of the carbonaceous precursor, a process known as gasification, resulting in the

formation of high energy areas that serve as adsorption sites. The resulting carbon

parameters (e.g., pore size distribution (PSD), functional groups) are dependent on the

oxidant used. Steam (H20) and CO2 are the most common agents used and will therefore

be the focus of this section.









CO2 and H20 Activation

The literature generally agrees-with one notable exception (Wigmans 1989)

which has been cited 134 times to date, according to the Science Citation Index-that the

use of CO2 as the oxidizing agent during carbon activation leads to an enhanced

development of micropores; while the use of steam promotes the propagation of

mesopores (Tomkow et al. 1977, Rodriguez-Reinoso et al. 1995, Molina-Sabio et al.

1996, Johns et al. 1999). Creation of pores resulting from the endothermic oxidation of

carbon with steam or CO2 follows Eqn 2.1 and 2.2:

Using C02: C + CO2 COsurface + CO(g) Eqn 2.1
Using steam: C + H20 COsurface + H2(g) Eqn 2.2

With sufficient heating, CO2 will break a carbon bond from the aromatic ring in the

precursor. To satisfy the bond, the CO2 will leave behind an oxygen molecule at the

surface, thus creating CO while the remaining CO from the initial CO2 molecule will

escape as a gas (Eqn 2.1). In the case of steam (Eqn 2.2), the H2 generated can anneal to

the aromatic ring. The annealing process is slower than the formation of the CO bond,

yet the former is dominant, thus resulting in a greater energetic reaction and the creation

of a larger pore. If the H2 does not anneal carbon active sites (i.e., those C atoms that are

not fully saturated), the H2 diffuses out of the carbon and can actually inhibit the forward

progress of gasification (Ergun and Mentser 1965).

The impact of oxygen functional groups have been shown to effect pore

development. Molina-Sabio et al. (1996) found that the evolution of oxygen functional

groups as CO from the carbon surface was correlated with the CO2 micropores and the

widening of pores from steam. It should be noted that the presence of oxygen functional









groups could also affect the adsorption of target contaminants such as phenol, which will

be discussed in following sections.

Adsorption Factors

Surface Area

Surface area is an important parameter to consider when creating or selecting an

adsorbent. In an ideal adsorption situation-where all other conditions (such as pore

size, surface chemistry, and adsorbent-adsorbate interactions) are optimal for

contaminant removal-the surface area would serve as the limiting factor for the

adsorption process. In this case, as the activated carbon surface area increased, so would

the adsorption of the target contaminant. Manufacturers of activated carbon attempt to

increase the surface area of the adsorbent with the hope of enhancing the carbon's

removal efficiency. Typically, the surface area of activated carbon reaches a maximum

around 1500 m2/g.

The activated carbon surface area is commonly found using a theory developed by

Brunauer, Emmett, and Teller (BET) for physical adsorption. While the BET theory is

inadequate as a universal equation for physical adsorption, it has been adapted to describe

surface area (Dabrowski 2001). BET surface area is determined by flowing nitrogen gas

at 77 K through a sample of activated carbon, allowing the N2 to enter the pores of the

carbon. From the amount of N2 that adsorbs to the pores, the surface area is deduced. It

should be noted that the BET surface area can at times be misleadingly when attempting

to directly correlate it to the adsorption capacity of an activated carbon (Pope 2003).

Since nitrogen gas molecules are much smaller than many target contaminants, the size of

the contaminant itself should be considered with respect to the pore size distribution of

the carbon.









Pore Size Distribution

Another essential parameter to consider for an absorbent is its pore size distribution

(PSD). Pore size distribution is usually expressed as a graphical relationship, using pore

width (A) as the independent variable and cumulative pore volume (cc/g) as the

dependent variable. Pore widths that fall under 20 A are considered to be micropores,

from 20 to 500 A are mesopores, and above 500 A are macropores. The variation of pore

widths in an activated carbon is dependent upon the activation process. Theoretically,

any precursor can have any desired PSD, yet, the degree of distribution may require more

manipulation of the activation environment, and hence may be more energy (and cost)

intensive.

pH

When the parameter of pH is discussed in an experiment, commonly it is used as a

descriptor to express the ionic conditions of a given aqueous system. However, in the

case of activated carbon adsorption, it is important to take into account the pH of the

carbon surface and internal pores in addition to the aqueous medium. An internal

measurement of the activated carbon is expressed through the point of zero charge (PZC).

The PZC is the point at which the carbon surface has no detectable charge. In Figure 2.1,

the PZC is at pH 7; therefore at a solution pH below 7, the carbon surface is positively

charged and a solution pH above 7 will promote a negatively charged carbon surface.

PZC is an important characteristic when predicting or describing the process of

adsorption, yet is only a dominant factor when the target contaminant is close to the

adsorption site. For example (assuming the contaminant is in range of the adsorption

site), if the target contaminant was a cation, a solution pH greater than 7 would be desired









for the PZC illustrated in Figure 2.1. Conversely, if the target contaminant was an anion,

a solution pH below 7 would be desired.




+ At zero charge
(+H =7
mmol charge p
pH
g

(-)
PZC
curve


Figure 2.1. Representation of Point of Zero Charge

The zeta potential, a measurement of external charge, governs the attraction of the

target contaminant to the activated carbon. Zeta potential is the measurement of electric

potential at the shearing plane-the "space" between the activated carbon surface and the

adjacent water molecules (Adamson 1990).

Surface Chemistry

The structure of activated carbon is graphitic in nature, consisting of molecular

layers of carbon which can be viewed, according to Coughlin and Ezra, much like a

polynuclear aromatic molecule (Coughlin and Ezra 1968). These layers contain carbon

atoms that are bonded together with three sigma bonds and one pi bond having sp2

hybridization. It is also possible for sp3 hybridization (tetrahedron) to occur, which may

result in cross-linking among the graphite layers (Coughlin and Ezra 1968). The carbon

within this structure is microcrystalline and is held together with the graphite layers

through van der Waals forces. When other atoms are bound within this system, they can

be present within the layers, forming "heterocyclic" rings, or at the edges of the









microcrystalline carbon molecules, thus forming functional groups (Coughlin and Ezra

1968). Edge sites located between the graphitic layers are very reactive and are therefore

prominent sites for functional groups and adsorption. It is asserted by Coughlin and Ezra

(1968), that the basal face of the benzene ring can weakly adsorb through '7 interactions.

A discussion of activated carbon surface chemistry should also include a thorough

examination of electron interactions, including electron density, electrostatics, attraction

and repulsion, as well as dispersive forces and the influence of functional groups located

on both the adsorbent and the adsorbate. Numerous discussions in the literature center

around speculation of these interactions and many of these theories have not yet been

sufficiently disproven. It is hoped that a deeper understanding of the activated carbon

surface chemistry will provide the keys for unlocking the mechanisms of the adsorption

process. Although questions remain, the literature gives significant insights into the

surface chemistry of activated carbon; these theories will be examined in the following

sections of this paper.

Predicted Adsorption Mechanisms

Despite the extensive amount of research that has advanced the field of activated

carbon, the adsorption mechanisms still remain unknown. Currently, the literature has

concentrated on two mechanistic theories-the theory of 7i-7T bonding first proposed by

Coughlin and Ezra (1968) and the electron donor theory proposed by Mattson et al.

(1969).

7t-7t Bonding Theory

The 7t-7t bonding theory takes into consideration the 7t bonds which occur between

two p-orbitals as shown in Figure 2.2. This type of bonding occurs between the 2p









orbitals in the benzene ring presented in Figure 2.3. The electrons in the 2p orbitals

travel above and below the benzene basal plane, therefore creating a delocalization of

electrons about the ring. At the basal plane, weak adsorption can occur through 7

interactions, while at edge sites adsorption is much stronger (Coughlin and Ezra 1968).












CY
Figure 2.2. Representation of sigma and pi bonding, arrows within the atoms of
formaldehyde represent electron filling in outer orbitals. Adapted from
Petrucci and Harwood (1997).





H H




Figure 2.3. Representation of the Benzene Pi System (Source: Tedankara Library, 2003)

Species that are bonded to the graphitic edges of activated carbon can cause a

disturbance in the electron density of the basal plane. For example, the presence of 02

molecules (i.e., electron-withdrawing groups) will influence the nt electron distribution by

removing electrons and creating positive holes in the conduction band of the n electron

system, thus decreasing the dispersive adsorption potential of the carbon surface

(Coughlin and Ezra 1968, Franz et al. 2000, Radovic 2001). The opposite is also true;









removing electron-withdrawing groups from the carbon surface will increase the '7

electron density and therefore result in an increased adsorption potential (Radovic 2001).

Electron Donor Theory

The electron donor theory suggests that an exchange of electrons takes place during

the adsorption process, whereas in '7-7K bonding dispersive interactions are dominant.

Mattson et al. (1969) argue that Coughlin and Ezra do not explain the effects of 7

interactions across the distance of the basal plane as well as the mechanism of interaction

of the sorbates to the basal plane. Using the nitro- group as an example, Mattson et al.

(1969) suggest that as electron-withdrawing groups reduce the 7t electron density, the

nitro-substituted aromatic would become an acceptor that would interact with a donor

such as carbonyl oxygen. In a broader sense, the carbonyl oxygen would be the major

donor and the aromatic ring would be the acceptor. Taking into consideration the 7t

system interactions, Mattson et al. (1969) state that the adsorption orientations should be

planar.

A notable paper by Franz et al. (2000) disagrees with the findings of Mattson et al.

(1969). Mattson et al. (1969) base the electron donor theory on their observation that as

surface oxygen increases so does the amount of carbonyl functional groups. As the

carbonyl group serves as the major electron donor in the donor-acceptor complex,

Mattson et al. credit it for the increase in phenol adsorption. Mattson et al. attribute

subsequent decreases in adsorption to the formation of carboxyl functional groups.

However, Franz et al. (2000) have found that an increase in oxygen functional groups

leads to an increase in the formation of both carbonyl and carboxyl groups. Radovic

(2001) also refutes the findings of Mattson et al., stating that "carbon oxidation does not









(necessarily) involve the conversion of carbonyl groups into carboxyl groups," thus

suggesting that the reasoning of Mattson et al. is flawed. Franz et al. (2000) suggest that

the decrease in adsorption is due to the presence of water molecules that create bulky

clusters on the carboxyl groups due to hydrogen bonding and block access to the

adsorption sites. To support this conclusion, Franz et al. (2000) argue that if moisture is

enough to block N2 from adsorbing during BET surface analysis, that a compound much

larger than N2 should also be expected to be blocked. In agreement with the speculations

of Coughlin and Ezra (1968), Franz et al. (2000) support an adsorption theory based on

dispersive/repulsive interactions.

Functional Group Formation

Detection of Functional Groups

Throughout the literature, the most popular form of functional group identification

is through the use of Boehm titrations (presently cited 172 times in the literature

according to the Science Citation Index). Functional group identification is made

through the applications of organic acid-base chemistry. These acid-base reactions

examine the chemisorbed oxygen on the carbon surface, defining four acidic surface

groups (Boehm 1966 and Coughlin et al. 1968):

(I) strongly acidic carboxyl group
(II) weakly acidic carboxyl group
(III) phenolic hydroxyl group
(IV) carbonyl group.
Titrations are performed to identify which groups are present through a

neutralization process using various bases. Group (I) can be neutralized by NaHCO3,

NaCO3, NaOH, and NaOC2H5. Group (II) can be neutralized by Na2CO3 or stronger









bases. Group (III) functional groups are titrated with NaOH and group (IV) reacts with

sodium ethoxide (NaOC2H5) (Boehm 1966).

In addition to Boehm titrations, other methods have been presented (e.g., X-ray

detection, and Fourier Transform Infrared Spectroscopy (FTIR)), yet the success of many

of these methods has not been universally accepted for the accurate description of the

carbon surface-especially within the pore walls.

Types of Functional Groups Formed

From CO2 activation

The formation of carbonyl and carboxyl groups from CO2 activation has been

reported in the literature (Mattson and Mark 1969, Johns et al. 1999, Guo and Lua 2002).

Through the use of infrared internal reflectance spectroscopy, Mattson et al. (1970) have

determined the presence of "significant amounts" of carboxyl and carbonyl groups which

they speculate have a strong interaction at the edges of the aromatic basal planes.

Although no rationale is offered, Mattson et al. (1970) have speculated that at some

temperatures (3000C, 4000C, 6000C, 7000C) the use of CO2 during activation can

suppress the formation of oxide surface functional groups. In addition to carboxyl and

carbonyl groups, Johns et al. (1999) have determined that CO2 activation produces more

acidic functional groups than steam activation; specifically, the presence of phenolic and

lactone groups was detected. Lua and Guo (Lua and Guo 2000, Guo and Lua 2002)

found that CO2 activation of oil palm shell produced a slightly acidic surface consisting

of quinones, aromatic rings, and oxygen atoms speculated to be bound at the edges of the

graphitic sheets.









From H20 activation

During the process of steam activation, in addition to oxidizing the carbon surface,

steam prevents secondary char formation on the carbon by removing volatiles from the

surface and can interact with both organic and inorganic matter (Petrov et al. 1994).

Through these interactions with the carbon surface, various functional groups are formed.

The formation of hydroxyl (e.g., phenol and lactone), carboxyl, and carbonyl groups have

been reported in the literature (Gergova et al. 1993, Petrov et al. 1994, Johns et al. 1999).

While both hydroxyl and carboxyl groups were detected after steam and CO2 activation,

Johns et al. (1999) reports a decreased presence of both following steam activation.

Factors controlling which groups are formed

The functionality of a given activated carbon is highly dependant on its

carbonaceous precursor and its heat treatment. Various raw materials will form varying

functional groups when activated depending upon their structure and composition. For

example, a precursor containing sulfur would be expected to exhibit sulfur containing

functional groups, while a material void of sulfur would not form such groups unless

treated with a thiochemical such as H2S. Likewise, the intensity of heat treatment can

control the degree at which surface groups are burned off, or formulated. In the previous

example, if the activation temperature was intense enough to burn off the sulfur as a gas,

it would be expected that a sulfur functional group would not form on the carbon surface.

It therefore follows that different types of activated carbon will adsorb contaminants

differently-hence, the efficiency is variable and can be tailored in the activation process.

Similarly, the same activated carbon will have unique adsorption relationships with

varying contaminants-i.e., the efficiency will vary according to the chemical and

physical characteristics of the adsorbate.









Precursor and conditions of activation

Upon examining the significance of the activation process on the surface chemistry

of an activated carbon, it is insightful to examine the functionality of the carbon both

before and after the oxidation process. In the literature, two such papers were found.

Gergova et al. (1993) have examined the surface of apricot stones prior to and following

steam activation. The results of this study are summarized in Table 2.1. From the data, it

is clear that the composition of carbon, nitrogen, and sulfur increased while hydrogen and

oxygen decreased. While both nitrogen and sulfur also show an increase in percent

weight, the difference is less than 1% (almost less than 0.5%) and would be difficult to

argue any significant difference in comparison with the other elements. The most notable

changes are between carbon and oxygen; the weight percentage of carbon increased by

nearly 40% while oxygen decreased approximately 35%. It should be noted that the

composition of oxygen was found by difference; hence, it was assumed that the material

consisted only of carbon, hydrogen, nitrogen, sulfur, and oxygen. While the authors

mention the detection of carbonyl and hydroxyl groups, they fail to draw any conclusions

about the change in elemental concentration. Furthermore, the authors do not state the

process by which the complete elemental analysis was performed, making it nearly

impossible for the reader to derive a conclusion concerning the difference in composition.

Guo and Lua (2002) reported the change in functional groups detected on oil shell

palm prior to and after CO2 activation. The raw material oxygen functional groups

consisted mainly of carbonyl groups (e.g., ketones and quinones), ethers, and phenols.

After carbonization, the presence of ketones and quinones were detected, and following

activation, only the presence of quinones remained. The authors attribute the loss of the

ketone functional groups to the intense heating (950C) (Lua and Gou 2000, Guo and Lua









2002). The effects of temperature on the formation of functional groups during the

activation process have also been confirmed by Mattson et al. (1969). For the case of the

oil shell palm, the result of the carbon surface changes resulted in a slightly acidic carbon

surface (Lua and Gou 2000, Guo and Lua 2002).

Table 2.1. Elemental Analysis of Apricot Stones Before and After Steam Activation at
6000C, Adapted from Gergova et al. 1993
Elemental Analysis Raw Material Activated Carbon
(% weight) (% weight)
C 51.45 89.67
H 6.34 2.40
N 0.20 0.58
S 0.09 0.39
0 (by difference) 41.92 6.96

In addition to the effects of temperature, the presence of impurities in the

carbonaceous precursor can lead to notable effects in the formation of surface functional

groups. Mattson and Mark (1969) state that the presence of oxygen, nitrogen, sulfur,

hydrogen, and ash are common impurities and that specifically the presence of oxygen

and hydrogen (especially when bonded with oxygen) can have distinct effects on the

adsorption process. Coughlin et al. (1968) found that the presence of chemisorbed

oxygen decreased the adsorption capacity for phenol while Vidic et al. (1993) have

shown the opposite effect. It should be noted that Coughlin et al. observed the adsorption

of phenol in a dilute solution-it can be argued that the dilute solution accounted for the

lack of adsorption and not the oxygen functional groups. In the case of a dilute solution,

the driving force for adsorption is very low, thus decreasing the probability for the

adsorbate and adsorbent to interact and adsorption to occur.









Optimizing the Use of Activated Carbon

Commonly, to estimate the adsorption capacity of an activated carbon, Freundlich

or Langmuir isotherms are developed. However, according to Weber and DiGiano

(1996) : "Despite the sound theoretical basis of the Langmuir, BET, and Gibbs models,

these isotherms often fail to describe experimental solution sorption data adequately." It

is proposed that the Freundlich equation can better describe the activated carbon

adsorption process than the afore mentioned models due to the heterogeneous nature of

the carbon surface. It will be shown in the following sections how the Freundlich

equation compares to the frequently used Langmuir model for activated carbon

adsorption of benzene and monosubstituted benzenes. Developed by Herbert (Heinrich)

Freundlich from empirical observations in the early 1900s, the Freundlich model offers

an exponential equation, based on heterogeneous surface energy distribution, which

describes the variation in adsorption heat with respect to adsorbate concentration (Weber

and DiGiano 1996):


q, = KC Eqn 2.3

Where:

qe = the mass of contaminant adsorbed per unit mass of carbon

K = "adsorption capacity", Freundlich constant

C = concentration of the adsorbate in solution at equilibrium

1/n = strength of adsorption, Freundlich constant

To validate his empirically derived equation, Freundlich utilized a special case of the

Gibb's relationship, working under the assumption that in a dilute solution Gibb's surface

excess is equivalent to the amount of contaminant adsorbed (Weber and DiGiano 1996).









The Freundlich equation can be used in designing an activated carbon system.

For example, using the Freundlich constants K and 1/n, the following equations can be

used to estimate the bed life of an activated carbon (Snoeyink 1990):


Y = PGAC Eqn 2.4
(C C1)

Where:

Y = carbon bed life

Co = initial concentration of contaminant

C1 = final concentration of contaminant

PGAC = apparent density of carbon

The Freundlich constants K and 1/n are commonly sited in the literature without a

thorough discussion of what each represents. It is one intent of this paper to attempt to

clarify the significance (and limitations) of these parameters. The so called adsorbentt

capacity" constant K represents the influence of various adsorption energies that are

associated with the heat of adsorption. However the term adsorbentt capacity" only

applies when K = qe, hence when C=l and K is dependent solely on the change in

adsorbate concentration due to adsorption (Chiou 2002). The term 1/n represents the sum

of diverse energies associated with adsorption. The notation of 1/n comes from the

thermodynamic derivation of the Freundlich equation, but, given that the equation is

typically used empirically, the literature often uses simply "n" (Weber and DiGiano

1996).

Certainly adsorbentt capacity" (K) and strength of adsorption (1/n) should vary

with varying activated carbons. Noting that no two activated carbons are identical, it

should be easily deduced that the performance of varying activated carbons would also









differ. These differences could account for the discrepancies reported for the Freundlich

constants in the literature. It should further be noted that the concentration range of the

activated carbon used in the development of isotherms may have a significant effect on

the value of K and 1/n. It is not uncommon to find K values that differ by several orders

of magnitude in the literature; for example, Table 2.2 shows three different values of K

for benzene. The variation in reported values is likely due to various definitions of

equilibrium conditions, the use of different carbons, and the use of varying concentration

ranges for both the adsorbent and adsorbate. However, if these K values are to be used as

tools in designing activated carbon systems, these differences must be accounted for and

understood.

Table 2.2. Reported Values of Freundlich K for Benzene.
K Source
0.12602 Noll 1999


1.0 Dobbs and Cohen 1980
41.7 Loll et al. 2004

An understanding of the Polanyi-Manes model (Eqn 2.5) reveals adsorption trends

which are reflected in the Freundlich constants. Using the Polanyi-Manes model, the

energy necessary for an adsorbate to displace a solvent (in order to be adsorbed on the

adsorbent) can be calculated (Chiou 2002). For example, for similar solutions under

constant conditions of temperature and pressure, more energy is required to adsorb more

adsorbate due to the increased volume of solvent that must be displaced.


ES = RTn n)


Eqn 2.5


Where:

E', = adsorption potential (energy required for a volume of solute (s) to displace a
volume of solvent (/) in the adsorption process)









R = gas constant

T = temperature

Cs= solute concentration

Ce= equilibrium concentration of adsorbate in solution

Applying the same basic principles to the Freundlich constants, as the adsorption

potential increases, the value of K should also increase, as K is a representation of the

energies required for adsorption. Likewise, as the adsorbent concentration range varies,

so will the concentration of adsorbate in the solution; thus leading to variation in the

adsorption potential and hence a change in K. It should then follow that as the adsorbent

range is varied in isotherm experiments, the resulting value of K will differ, potentially

across several magnitudes.

Developed originally for air phase adsorption, the Langmuir equation (Eqn 2.6) has

been widely adapted for aqueous systems by replacing the original pressure term with a

term representing the solute equilibrium concentration.

1 1 1
-= 0-+ Eqn 2.6
qe Q', bQ, C,

Where:

qe = the mass of contaminant adsorbed per unit mass of carbon

Qa = maximum adsorption capacity

b = net enthalpy of adsorption

Ce = equilibrium concentration of adsorbate in solution

The Langmuir equation draws its theory from the condensation and evaporation of gas

molecules on a solid adsorbent surface (Weber and DiGiano 1996). Condensation takes

into account available adsorption sites on the adsorbent surface as well as the rates at









which the adsorbates contact the adsorbent surface. Three basic assumptions are

encompassed within the Langmuir equation: (1) adsorption energy is constant throughout

the activated carbon surface and is independent of surface coverage (2) there are no

interactions between adsorbate molecules or migration of adsorbates to other adsorption

sites (3) the carbon surface will only support a monolayer adsorption of the adsorbate

(Weber and DiGiano 1996). Two constants are typically reported from the Langmuir

model, Qa and b. Qa represents a condition where the activated carbon surface is

covered by a monolayer-i.e. when the surface has reached its capacity according to the

assumptions embedded within the Langmuir model. The constant b represents the net

enthalpy of adsorption and is described by a ratio of rate constants (kadsorption/kdesorption) on

a mole (or mass) unit basis (Chiou 2002, Weber and DiGiano 1996).

Weber and DiGiano (1996) have shown that the Freundlich equation can be

represented using the general terms of the Langmuir equation:

q, = Q0,g(b gC)n Eqn. 2.7

The subscript g denotes a generalized form of the parameters where Qa,g = Qa, while bg

represents the Langmuir constant b and also accounts for site energy. The constant n

represents the Freundlich 1/n and accounts for the heterogeneity of the surface site

energies. With this relationship in mind, the Freundlich K becomes a function of Qa,g

and bg (Weber and DiGiano 1996):

K = Qo,gbgn Eqn. 2.8

The use of isotherms for estimating carbon adsorption is a typical textbook

approach that is still widely used by both carbon manufacturers and its users. When used

properly, isotherm constants are valuable tools for understanding the adsorption process









of a given system. Isotherm constants relate the equilibrium conditions of a carbon with

an adsorbate and indicate both the capacity of the adsorbent for that adsorbate and the

strength at which the adsorbate is adsorbed. When isotherm constants are known for a

given carbon and target contaminant, the removal of that adsorbate can be predicted and

optimized.

While empirically determined isotherm constants are useful, the development of

these constants are very time consuming and their applications are limited to the

conditions under which they were derived. Ultimately, to accurately represent the

adsorption process for a given system, the carbon should be tested within that system.

Use of a direct testing method will render a more accurate solution if conditions permit

its application; such a method is carbon profiling. Carbon profiling is a valuable concept

that has yet to be implemented widely in the water treatment field. By using carbon

profiling, the effectiveness of an activated carbon can be observed without relying on an

adsorption isotherm. Adsorption isotherms are always performed at equilibrium

conditions, accounting for the effects of adsorption and desorption which are inherent in

the isotherm equations. Equilibrium conditions are typically not seen on a full scale

level-such as in a water treatment facility-and therefore caution should be used when

estimating full-scale treatment parameters. It should be noted that the role of kinetics

with respect to PAC and GAC are different. For example, in a water treatment facility,

PAC would typically be added at the head of the plant (during rapid mix) so that

sufficient contact time is allotted for the contaminants to interact with the carbon. Since

PAC is smaller than GAC, more time is required for the contaminants to come into initial

contact with the carbon. However, because of the greater internal volume of GAC, an









adsorbate may take much longer to reach equilibrium within the carbon matrix than

within PAC. The following steps comprise carbon profiling:

1. Obtain a sample of the water to be treated

2. Perform dose removal studies using various amounts of the activated carbon,
mimicking the plant's processes: contact times, chlorination, pretreatments, etc.

3. Generate dose removal curves

4. Determine the most effective carbon dose for removing the targeted contaminant,
accounting for treatment objective and economics

While carbon profiling circumvents a limitation of adsorption isotherms, e.g.,

equilibrium conditions, this system still has its limitations. Much like the adsorption

isotherms, carbon profiling is not performed at full-scale and is time consuming, as

specific conditions must be met and optimized through a series of dose removal studies.

The traditional method of using previously existing isotherm constants to develop

removal schemes is outdated and inefficient. Parameters rendered from the traditional

methods likely do not emulate full-scale processes since isotherms are developed under

equilibrium conditions and often the conditions under which the constants were derived

are not addressed. Adsorption is dependent on the type of activated carbon used; yet, the

isotherm parameters that are typically referenced from the literature are developed with

unstated types of carbon (often using nanopure water) and therefore should not be relied

upon to determine the adsorption properties of another system. The constants therefore

will not reflect the interactions of natural organic matter or any other constituents that

may be present in the contaminated water. It should be noted that while adsorption

isotherms rely on equilibrium conditions for appropriate application of the isotherm

equations, they certainly can be conducted with respect to other aspects of the intended

treatment system. In this case, a similar procedure as carbon profiling can be followed









incorporating the treated water and chemicals present in the large scale process, with the

condition that equilibrium exists. The decision of using carbon profiling or adsorption

isotherms should be carefully considered with respect to the treatment system for which

its application is intended.

The development of models that incorporate adsorption system descriptors such as

kinetic rates and adsorption isotherm constants are promising tools that can incorporate

the effects of unknown mechanisms into predictive adsorption analysis. By applying an

empirically based model to an appropriate system, the mechanisms within the adsorption

process can be described and accounted for without being fully known. Such a tool,

Quantitative Structure Activity Relationships (QSAR), is becoming widely known in the

medical and biological fields, but has yet to see a dominating presence in adsorption

study. Blum et al. (1994) have developed a QSAR for activated carbon adsorption in

water using 363 organic compounds. While this model is noteworthy, it is based on

literature-derived constants-which may vary drastically with respect to experimental

procedure and analysis-without a sturdy foundation, these models will not accurately

portray the adsorption process. Blum et al. show an understanding of this vital point as

the authors mention that "...in applying QSAR based on literature data, it is imperative to

consider the relevance of the data source to one's own investigative situation." However,

when used in conjunction with accurately descriptive parameters (i.e., case specific),

QSAR can be a powerful tool for understanding activated carbon adsorption. The goal of

this study was to investigate the use of adsorption isotherm constants in QSAR and to

provide a protocol by which QSAR and adsorption isotherm constants can be effectively

used to describe the adsorption process for a given system.









Quantitative Structure Activity Relationships

While adsorption isotherms and carbon profiling are useful in developing

adsorbate-adsorbent relationships, they are time intensive and non-specific in regards to

adsorbate-adsorbent structural interactions. Relying solely on isotherm relationships and

profiling would require extensive laboratory analysis for each individual compound (and

condition) that is to be analyzed. The adsorption of a chemical compound to activated

carbon is dependent on both the chemical and physical properties of the carbon and the

compound. Therefore, by knowing the chemical and physical properties, the adsorbate-

adsorbent relationship can theoretically be predicted. Using known parameters to predict

the activity of a compound is the focus of QSAR modeling.

Introduced into the field of biology just over four decades ago, QSARs have gained

prominence in the both the biological and medical fields (Hansch and Fujita 1995).

While QSARs have proven to be powerful tools for the development of numerous

chemicals and medications, the adsorption field has yet to develop a robust QSAR for

adsorption prediction. Within an adsorption system, a QSAR focuses on predicting the

molecular interactions between the adsorbate and adsorbent (Brasquet et al. 1997). For

example, in an adsorption system, the independent variable input into the QSAR would

be the target adsorbate's intrinsic parameters (e.g., chemical, electronic, steric,

hydrophobic properties) while adsorption parameters (e.g., isotherm constants, kinetic

rates, etc.) would be the dependent variables. A modeling program based on QSAR

interactions is theoretically able to predict the behavior of a given contaminant in the

presence of an activated carbon-thus given the independent variable (or variables) the

QSAR outputs a calculated dependent variable describing the adsorption system, for

instance, an adsorption capacity.









The driving force behind a QSAR is its ability to formulate a relationship between

two compounds through the use of tools such as regression analysis or pattern recognition

(Lloyd 2002). Regression analysis is commonly based on a partial least squares

projection, which is a multivariate statistical method; while pattern recognition is usually

performed by artificial neural networks (ANNs). ANNs mathematically mimic the

pattern recognition properties within the human brain, using parallel processing and

weighted connections, which store the problem solving knowledge. These weighted

connections are trained to specific situations through repeated exposure and comparison

to truth data sets (PNNL 1997). The key advantage of the QSAR model is that a known

equation or the exact pathway of reactions is not necessary. A good QSAR model can

rely solely on the structural and derived experimental characteristics of the targeted

compounds to predict their activity. Additionally, unknown mechanisms could ultimately

be disclosed through the analysis of how substituents impact QSAR correlations.

Within the development of a QSAR, diverse parameterization is essential. Having

a robust set of carefully chosen parameters increases the probability of unlocking a

description of the mechanisms involved in a system. Three key components of a QSAR

are: hydrophobic, electronic, and steric factors (Hansch and Fujita 1995). Table 2.3 lists

various hydrophobic, electronic, steric and chemical properties that are commonly used

in QSAR analysis.

While diverse parameterization is essential to a QSAR, the selection of these

parameters is crucial for the healthy development of the QSAR. A primary weakness in

QSAR construction is often the selection of the parameters (Hansch and Fujita 1995).

Poor parameter selection can lead to collinearity problems (i.e., when two parameters are









directly affected by each other) thus creating false indications of correlation. Also, it is

essential that parameters cover a wide range of space-i.e., the compound training set

should be carefully selected so that multiple spectrums of the parameter scale are

represented. This idea will be more fully developed in the Methods section with the

introduction of Craig plots.

Table 2.3. Example of Parameters Commonly Used in QSAR Development.
Hydrophobic Properties H, LOG Kow

Electronic Properties GInd, Resonance, Field Effects

Steric Properties Molar Refractivity, Molal Refraction, Es

Chemical/Physical Properties Boiling Point, Melting Point, Density, Molecular Weight,

Aqueous Solubility, Enthalpy, Vapor Pressure

Where CInd represents inductive effects described by the Hammett Constant and Es
represents Taft's steric factor.

An important aspect of QSAR development that should be addressed is the

selection of the training set and the specific parameters that are chosen for the

correlations. There are two basic approaches to QSAR that are seen in the literature, the

first approach, developed by Hansch, is an approach favored by chemists in which the

components within the QSAR are intellectually analyzed with respect to their chemical

activity (Hansch and Fujita 1995). The second approach, described by Wold and Dunn

(1983), is based on the statistical analysis of the components for QSAR development.

Certainly, while the statistical method can prove to be insightful in some instances, a

strong QSAR should be constructed with a vast understanding of the components in the

system in order to avoid complications such as parameter collinearity, uneven spread of

parameters, and the application of implausible mechanisms. In addition, it is also









important to consider the functionality of the compounds that are within the training set;

only structural changes in functional groups that can be accurately represented by

available parameters should be selected (Hansch and Fujita 1995).

This present work approaches QSAR development for the adsorption process in

three improved ways in comparison to past studies:

1. Dependent variables used are not from the literature, but were derived
experimentally within this study

2. Degree of statistical analyses performed

3. Robust spread of parameterization

As was shown in Table 2.2, empirically derived isotherm constants selected haphazardly

from the literature can be detrimental to a study. The electronic, steric, hydrophobic, and

chemical parameters used within this study were not experimentally derived within the

scope of this work, yet each of these parameters has been well documented within the

scientific community and has strict procedures for their procurement. However, in the

case of deriving the isotherm constants, many variables are present: the type of carbon

used, how it was activated, the conditions under which the carbon was kept, how it was

added to the system, the pH of the water in the system, the solutes used within the

system, the definition of equilibrium for that system, the concentrations of carbon and

solute that were used, etc. In order to normalize these variables, each of the isotherm

constants used within this system was derived under uniform experimental conditions.

In order to measure the quality of the QSAR correlations, several statistical

methods were used: R2, adjusted R2, standard error, F-Ratio, as well as Q2 (a tool that

measures the predictability of a correlation). This combination of statistical analyses is

more diverse than other adsorption QSAR that are listed in the literature and therefore






27


provides a more rigorous analysis of the correlations. For example, the study of Brasquet

and coworkers (1997) refers to R2 values, but offers no other statistical validation for

their resulting correlation.

As was mentioned previously, the spread of parameters is an important concept in

the selection of the training set. For this study, compounds demonstrating a wide range

of steric, electronic, hydrophobic, and chemical properties were selected. This variety of

parameterization is essential to aid in the identification of mechanisms involved in the

correlations.














CHAPTER 3
MATERIALS AND METHODS

Isotherm Methods

Adsorption isotherms were developed by adding a known concentration of

contaminant, 100-[tg/L, to distilled deionized (DDI) water at a pH of 5.5 1.5, the

variation is due to the use of unaltered DDI water. The desired amount of powdered

carbon was weighed and dried overnight at 1050C. After cooling the carbon in a

dessicator, a carbon stock slurry (10,000-mg/L) was prepared by combining the carbon

and DDI water by mixing on a stir plate. The slurry was stored in a dessicator at room

temperature and mixed on a stir plate before each application. Isotherms were created for

each target contaminant using the following concentrations of carbon: 1, 3, 5, 10, 15, 25,

50, and 75-ppm. The desired amount of carbon slurry was added to a 100-mL or 50-mL

SGE gas tight Luer Lock syringe containing DDI water. The experimental data was

collected for each contaminant individually using powdered Calgon F400 activated

carbon with 100-[tg/L solutions of either benzene or a chosen monosubstituted benzene

(shown in Table 3.1). For each run, the desired amount of contaminant was added from a

stock solution to the 100-mL syringe, yielding a final volume of 100-mL. The syringe

was then mixed end over end on a rotator for 2 hours to 6 hours, depending on

equilibrium conditions for that target contaminant. Equilibrium conditions were

determined through the development and analysis of dose removal curves over various

times. The condition of equilibrium was considered achieved when the rates of

adsorption and desorption appeared to be at a steady state. After mixing, the samples









were filtered into VOC vials using 25-mm Fisherbrand nylon syringe filters which were

sized to allow for only the PAC to be removed (both 0.20-am and 0.45- jam were used,

depending on their availability). Samples were sealed with Teflon septa caps and stored

at 4C until analysis. All samples were analyzed with a Saturn 2100T GC/MS using a

Supelco VOCOLTM Fused Silica capillary column with a Supelco SPME 75-jtm

CARBOXENTM PDMS fiber, utilizing an oven temperature program that holds at 400C

for 2 minutes and then climbs to 2100C at a rate of 80C/min. Each isotherm was

replicated. Only data which produced a RSD of 20% or less were accepted.

Table 3.1. Target Contaminants Used in Isotherm and Kinetic Studies.
Target contaminants
Benzene
Isopropylbenzene
n-butylbenzene
t-butylbenzene
Isobutylbenzene
Nitrobenzene
Aniline
Benzaldehyde
Fluorobenzene
Chlorobenzene
Bromobenzene
lodobenzene
Phenol

Development of Isotherm Data

Target Contaminants

Benzene

Benzene is a six carbon structure bonded in a cyclic formation with three

resonating double bonds. The benzene ring is one of the defining characteristic of









aromatic compounds. Due to its cyclic structure, and despite its unsaturated state,

benzene is a stable compound that will undergo primarily substitution reactions, such as

H2 annealing (Ouellette 1998). H2 annealing is a critical step during steam activation

which modifies the structure of the benzene rings in the graphitic carbon sheets, creating

the porosity of the carbon. The electron orientation of benzene drives its unique reaction

mechanisms. As was shown in Figure 2.2, the benzene ring experiences a delocalization

of electrons, where the carbon atoms have three sigma bonds-two bonding with the

adjacent carbon atoms and the third is bonded with a hydrogen atom. The remaining

electron gives the delocalization characteristic as it orbits above and below the basal

plane in a 2p orbital (Ouellette 1998). For this study, benzene has been selected as the

parent compound for the QSAR training set.

Substituted benzenes

Phenol is an aromatic compound consisting of a hydroxyl functional group bonded

via a sp2 hybridization to a carbon atom in the benzene ring. Adding a functional group

to the benzene ring can result in activating or deactivating the reactivity of the ring.

Because of the addition of the hydroxyl group, phenol is a strongly activating

compound-i.e, it increases the reactivity of the benzene ring. In contrast, a benzene ring

substituted with a chloro- or nitro- group causes the benzene ring to be less reactive.

Activating groups are more reactive and donate electrons, therefore increasing the

electron density of the benzene ring. Deactivating groups are less dense (therefore less

reactive) due to their tendency to attract-or withdraw electrons (Ouellette 1998). For

example, in the reaction of an activating (electron donating) group with an arene

(aromatic hydrocarbon), the electrons in the 2p orbital are attracted towards the arene,

thus creating an increase in electron density at the benzene ring, which in turn will make









the ring more reactive (Carey 2003). In the case of phenol, which acts as a weak acid, the

presence of an electron-withdrawing group on the carbon surface will cause a decrease in

electron density at the benzene ring and result in a more stable, more acidic, and less

negative compound. Contrastingly, in the presence of an electron-donating group, the

phenol will become less acidic (Ouellette 1998). When a phenol is oxidized, it becomes

a quinone-a carbon ring having two ketone carbonyll carbon atom bonded to two other

carbons) groups (Ouellette 1998). In the case of activated carbon adsorption, the

functional groups present on the carbon surface will contribute to the nature of the

reaction between the surface and the adsorbate.

Freundlich Isotherms

Freundlich isotherms were developed using the linearized Freundlich equation:


logq = logK +-log C Eqn 5
n

Where:

qe = the mass of contaminant adsorbed per unit mass of carbon

K = adsorption capacity, Freundlich constant

n = strength of adsorption, Freundlich constant

Cf = final concentration of contaminant

The term qe was determined by taking the difference in the initial and final

concentrations of the target contaminant in each isotherm run and dividing it by the mass

of activated carbon added to the system. All target concentration changes were attributed

to adsorption to the activated carbon. The log of qe was graphed verses log Cf to develop

the Freundlich isotherm plot. A linear trendline was added to the plot using Microsoft









Excel from which the slope and y-intercept were determined. The slope represents the

value for 1/n and the y-intercept represents the value taken as K.

Langmuir Isotherms

The Langmuir isotherm parameters were found using a linearized form of the

1
Langmuir equation as was shown in Eqn 4. A graphic was created by plotting verses
q,

1
-C. Using Microsoft Excel, a linear trendline was added from which the slope and y-


1
intercept were taken. The slope represents the value and the y-intercept is taken as
bQ00


The values for Q, and b where then determined.
QO

Kinetic Study Methods

Kinetic runs were completed individually for each compound shown in Table 3.1

using powdered Calgon F400 activated carbon. For each run, an aliquot of F400 carbon

slurry (prepared as stated above) was added to a 50-mL SGE gas tight Luer Lock syringe

containing 50-mL of DDI water at a pH of 5 + 1, resulting in a final carbon concentration

of 20-ppm. Using a Hamilton 1 to 10-[tL syringe the desired target contaminant was

added from a stock solution to the 50-mL syringe, yielding a final contaminant

concentration of 100-ppb. The 50-mL syringe was then rotated end over end. Kinetic

studies were preformed for 3, 5, 10, 15, 30, and 60 minutes. Following each kinetic run,

all samples were filtered into VOC vials using 25-mm Fisherbrand 0.20 or 0.45-[im nylon

syringe filters. Samples were sealed with Teflon septa caps and stored at 4C until

analysis. All samples were analyzed with a Saturn 2100T GC/MS using a Supelco

VOCOLTM Fused Silica capillary column with a Supelco SPME 75-jtm CARBOXENTM









PDMS fiber, utilizing an oven temperature program that holds at 400C for 2 minutes and

then climbs to 2100C at a rate of 80C/min. Each kinetic run was replicated. Only data

points that were sufficiently reproducible, i.e., those which produced a RSD value of 20%

or lower, were accepted.

Development of Kinetic Equations

The time for each compound to reach 86% removal with F400 was incorporated

into the QSAR as a kinetic parameter. The value of 86% was selected as it appeared as a

removal point common to each of the compounds in the training set. The time to 86%

removal was interpolated from a trendline created in Microsoft Excel from a graph of

percent removal verses time.

QSAR Development

The first essential step in the development of the QSAR was the selection of a

suitable training set. For the parent compound, benzene was selected, and

monosubstituted functionality drove the selection of the substituents. For this study it

was desired to use single substitutions within the training set to investigate the impact of

a sole functional group on the adsorption system. In the selection of monosubstituted

benzenes it is important to consider only the addition of functional groups that are well

parameterized-i.e., that can be described by an available parameter within the QSAR.

Non-descriptive groups cannot be well represented in the QSAR and therefore should be

avoided in the selection of a training set. It is also important to include compounds that

produce a wide spread of parameter values to ensure that the training set is representing a

good range of electronic, steric, hydrophobic, and chemical properties. A diverse spread

of parameter values adds robustness to the QSAR process and can be illustrated

graphically using a Craig plot (Lindner et al. 2003). Figure 3.1 demonstrates the spread










of parameter values for both n and GId. It is important to note the spread of the data

points throughout the quadrants, as the degree of spreading increases, so does the range

of parameterization.

For this study, monosubstituted benzenes were selected to demonstrate a large range of

steric, electronic, hydrophobic, and chemical parameterization. Table 3.2 illustrates the

values of these parameters for the training set.


2.5

S2-

1.5

1

0.5


-C.2 -0.1 0.1 0.2 03 0.4 0.5 0.6 07
-0.5

-1



crind


Figure 3.1. Craig Plot Illustrating the Parameter Spread of a Hydrophobicity Constant (n)
and the Hammett Constant (aInd).

After the training set was chosen, each parameter was examined for collinearity.

When plotted, any parameters that showed a linear correlation of R2 > 0.4 were

considered collinear and were not used in multilinear regression analysis.













Table 3.2. Selected Parameters Used in the QSAR Training Set (CRC Handbook of Chemistry and Physics 1999, Physical Properties
of Chemical Compounds 1955, MOPAC 2004).

Monosubstituted Benzene Electronics Sterics Hydrophobicity Physical Properties
X Descriptors Desciptors Descriptors


Log W C TOTAL
(Mnd ACCP MR Taft Es Kow (g/ (mass (A)
mole) %)

Fluorobenzene (FB) 0.52 -0.115 0.09 -0.55 2.27 0.14 96.1 0.154 5.21
Chlorobenzene (CIB) 0.47 -0.103 0.6 -0.97 2.84 0.39 113 0.0387 5.55
Bromobenzene (BrB) 0.44 -0.097 0.89 -1.16 2.99 0.86 157 0.0387 5.7
Iodobenzene (odoB) 0.39 -0.099 1.39 -1.62 3.28 1.12 204 0.0387 5.8
Nitrobenzene (NB) 0.64 0.039 0.74 -2.52 1.85 -0.85 123 0.0193 6.07
Aniline (Ani) 0.12 -0.169 0.54 -0.61 0.9 -1.23 93.1 0.21 5.84
Phenol (Ph) 0.29 -0.145 0.28 -0.55 1.48 -1.12 94.1 3.38 5.64
Benzaldehyde (Bzal) 0.3 -0.066 0.69 --- 1.48 -0.32 106 8.66 7.02
Isopropylbenzene (IsoPB) 0.01 -0.106 1.5 -1.71 3.66 1.22 120 0.3 7.18
n-butylbenzene (nBB) -0.04 -0.11 1.96 -0.16 4.26 2.13 134 0.0056 9.71
Isobutylbenzene (IsoBB) -0.03 -0.107 1.96 -2.17 4.01 1.7 134 0.0015 8.19
t-butylbenzene (tBB) -0.07 -0.107 1.96 -2.78 4.11 1.98 134 0.001 7.27
Benzene (Benz) 0 -0.102 0.1 0 2.13 0 78.1 0.178 4.97
Where ond represents the inductive effects described by the Hammett Constant, ACCP represents the charge of the para-carbon on the
functional group, MR represents molar refractivity, Es represents Taft's steric factor, log Kow is the octanol-water partitioning
coefficient, n7 represents the hydrophobicity of the compound, MW is the molecular weight, Caq is the aqueous solubility in water, and
TOTL is the total length of the compound, calculated by the bond lengths of the compound in the longest direction.









Parameters representing intrinsic properties of the compounds (e.g., hydrophobic,

electronic, steric, chemical, physical properties) were incorporated into the QSAR as

independent variables, while the parameters that were experimentally derived in this

adsorption system (e.g., isotherm constants) were incorporated as dependent variables

using SPSS 12.0 software (SPSS., Inc., Chicago, IL, USA). Classical QSAR procedure,

as developed by Hansch, was followed for the development of correlations (Hansch and

Fujita 1995). Linear, polynomial, and multilinear correlations were employed to describe

the data.

Correlations that resulted from the QSAR analysis had to meet a series of stringent

statistical criteria to be considered valid:

1. No collinearity of independent variables (for MLR correlations)
2. R2 > 0.7
3. F-Ratio values must meet a 95% confidence limit
4. Standard error considered
5. Q2 determined for goodness of predictability

Condition 1 is only applicable in regards to multi-variable analysis. Independent

variables were linearly correlated; variables that produced a R2 value of 0.4 or higher

were considered collinear and were excluded from the QSAR analysis. Conditions 2 and

3 were helpful in identifying poor correlations. The F-Ratio was determined using the

number of compounds in each correlation (k) and two total data sets (n), observed and

calculated values. The degrees of freedom were calculated for the variance between (vi =

n-1) and within (V2 = (n*k)-n) the data sets and compared to an upper 5% distribution

table to determine the significance of the observed and calculated data sets. As the F-

Ratio values become larger than the F-distribution values, the data sets are considered to

be more similar. Therefore, high F-Ratio values indicate the predicted data from the









QSAR correlation is in agreement with the observed data. The standard error (Condition

4) represents fluctuations in the sampling set and can be seen in the form of residuals in

the regression lines. Clearly, low values of standard error reflect better correlations

within the training set.

Condition 5 is an essential step in measuring the validation of the correlations that

result from the QSAR process. When performing a Q2 analysis, it is ideal to introduce a

new compound to test the predictability of the QSAR correlation; however if the

introduction of an additional compound is not feasible, an alternative method, roughly

referred to as the "take one out" method, will suffice. For this alternative method, one

compound from the training set is removed and the correlations are recalculated. The

new correlations are then used to test the ability of the QSAR correlation to predict the

value of the compound that was removed. This process is repeated for each compound

within the training set and equation 3.1 is used to calculate the value of Q2:


Q2 = 1- ,observed predicted new )2 Eqn 3.1
1(yi,observed )

While no set limit has been established for Q2 values, larger values indicate a better

ability for the QSAR correlation to predict the observed values within the given system.

Ideally the Q2 should be close to the value for the adjusted R2, thus indicating that the

new correlation did not cause a significant shift in the predicted data. For this study a

difference of 0.2 between the Q2 and adjusted R2 was considered the maximum limit for

variation. Similar, yet less stringent, statistical procedures as those outlined here have

been recommended by Eriksson and coworkers (Eriksson et al. 2003).














CHAPTER 4
MANUSCRIPT

Introduction

The use and production of toxic synthetic aromatic compounds in applications such

as the synthesis of rubber, paint solvents, insecticides, detergents, fragrances, fuels, and

dry cleaning (USEPA 2002,USEPA 1995) continue to increase in our society, and as a

result of their increased demand, the potential for introduction of these synthetic aromatic

compounds into our environment is escalating. The contamination of water, air, and soils

from exposures such as accidental spills, inappropriate disposal, and byproduct

formation, will continue to exacerbate the condition of our surroundings without

sufficient intervention and treatment. Given the increased exposure of the public to these

chemicals, environmental and health impacts such as toxicity, carcinogenicity,

mutagenicity, and teratogenicity in humans and animals, aquatic toxicity, and degradation

of potable water quality are also anticipated to increase. To circumvent such threats,

treatment methods must be established to remove synthetic aromatics from air and water.

It is well known that activated carbon is capable of removing aromatic compounds from

contaminated water.

Traditionally, the removal of volatile organic compounds (VOCs), a class of

compounds in which many aromatics belong, from water is accomplished thorough the

use of activated carbon and/or aeration. Activated carbon and packed tower aeration are

the USEPA-recommended forms of treatment for the removal of synthetic organic

contaminants, such as benzene and chlorobenzene, from drinking water (USEPA 2002,









USEPA2 2002). The most efficient method of aeration is by means of a packed tower.

This system is effective, yet is rarely sufficient to achieve low maximum contaminant

level (MCL) standards. It should also be noted that in cooler climates, the use of the

aeration towers becomes limited-due to possible ice formation and lower kinetics

(Hammer and Hammer 2001). While the efficiency of a packed tower can be optimized,

this process requires an increase in tower diameter, high surface area packing materials,

as well as increased tower height (Noll 1999). For some compounds these adjustments

can prove costly or impossible. In order to ensure MCL regulations are met, aeration is

usually coupled with activated carbon.

Activated carbon has shown to be an extremely efficient, cost effective adsorbent.

As activated carbon adsorption can be reversible, captured materials can potentially be

desorbed and salvaged. Activated carbon can effectively adsorb VOCs in both gaseous

and aqueous media. The general mechanisms of adsorption are physical adsorption and

chemisorption. Physical adsorption is driven by weak intermolecular forces, such as Van

der Waals forces. It is important to note that physical adsorption does not entail the

formation of a chemical bond, i.e., there is no exchange of electrons. Due to the lack of

chemical bonding, physical adsorption is readily reversed (desorption). During

chemisorption, a chemical bond results, thus making desorption much more difficult.

The surface of activated carbon is generally non-polar, thus facilitating the

adsorption of non-polar hydrocarbons, such as benzene, n-butylbenzene, and

nitrobenzene, from a polar aqueous media, such as water. The functionality of a given

activated carbon is highly dependent on its carbonaceous precursor and its heat treatment.

Therefore, different types of activated carbon will adsorb contaminants differently;









hence, the efficiency is variable. Likewise, the same activated carbon will have unique

adsorption relationships with varying contaminants, i.e., the efficiency will vary

according to the chemical and physical characteristics of the adsorbate.

Activated carbon is currently an industry standard for the removal of organic

contaminants in air and water (USEPA 2002). Although industry and the USEPA are

confident in the capacity of carbon to remove organic, a clear method of how this

removal is accomplished is not known. Currently, there exists a disconnect between

carbon scientists and carbon users (e.g., treatment plant operators, engineers). For

example, in academia, it is commonly taught that activated carbon adsorption increases as

solubility decreases (the Lundelius Rule). However, in the case of MIB and phenol,

despite their similar size, MIB is much less soluble in water than MIB (see table 4.1), yet,

as is recognized by water treatment professionals, phenol is more readily adsorbed by

carbon.

Table 4.1. Size and Aqueous Solubility of MIB and Phenol (Perry and Green 1997,
Whelton 2001).
Compound Size (A) Aq. Solubility (mg/L)

MIB z6 194

Phenol z6 82,000



Obviously, one cannot universally state that adsorption increases as solubility

decreases; another mechanism (i.e., surface chemistry) may be influencing the adsorption

of these two compounds. This creates a problem. Exceptions to widely accepted

adsorption trends, though generally well known among carbon scientists, rarely seem to

enter into the field of actual carbon applications. The knowledge base created for future

engineers and scientists in academia is carried with these professionals into the work









field, into consulting firms and municipalities. These professionals in turn continue to

spread misconceptions to carbon users, such as treatment plant designers and operators.

This lack of communication between the carbon scientists and users results in the

improper use and constricts the utilization of activated carbon in the field. The result is a

cookbook method of carbon dosing in treatment plants. To alleviate this black-box

method of carbon dosing, various relationships are currently being examined between the

adsorbate and the adsorbent. Two of these relationships are isotherms and Quantitative

Structure Activity Relationships (QSAR). Once a carbon-contaminant relationship is

established, a predictive model can be produced. This model could be a powerful tool for

assisting the carbon user-helping to optimize the use of activated carbon and bridging

the gap between carbon science and application. The quest for a unified adsorption

theory may be surrendered for the use of such correlations, which can optimize the use of

activated carbon without requiring a comprehensive understanding of the mechanisms

which are involved. Ultimately, the development of such correlations may lead to an

enhanced knowledge of the adsorption process. An enhanced understanding of

adsorption mechanisms will lead to reduced experimentation, time, and cost of activated

carbon selection.

The overall short term goal of this project is to develop a protocol for QSAR

analysis of activated carbon. In order to develop such a protocol, this study will examine

the predictability of QSAR for a training set of monosubstituted benzenes using

Freundlich and Langmuir isotherm constants as well as kinetic rate data as dependent

variables. To allow for a greater focus on the changing functionality of the training set,

only one activated carbon will be used in this initial study. Other intrinsic short term









goals for this work include: examining the predictability of QSAR for this training set,

determining the most descriptive isotherm equation for this training set and carbon (i.e.,

Freundlich or Langmuir), and an examination of the impact of kinetic rates.

Because of its widespread use in the literature, Calgon Filtrasorb 400 (F400) has

been selected as the activated carbon for this study. Selected physical properties of F400

are shown in Table 4.2.

Table 4.2. Selected Physical Properties of F400 (Karanfil and Kilduff 1999).
Surface Area Ave. Pore Pore Volume Percent Percent Percent Percent

(m2/g) Radius (cm3/g) Surface Area Surface Area Surface Area Surface Area

(A) in Pores < 20 in Pores 20- in Pores 100- in Pores > 200

A 100 A 200 A A

948 12 0.566 83.7 15.0 0.90 0.40



To estimate the adsorption capability of an activated carbon, Freundlich or

Langmuir isotherms are commonly developed. However, according to Weber and

DiGiano (1996) : "Despite the sound theoretical basis of the Langmuir, BET, and Gibbs

models, these isotherms often fail to describe experimental solution sorption data

adequately." Due to its empirical derivation, it is proposed that the Freundlich isotherm

equation (Eqn 4.1) may better describe the adsorption process, since it innately accounts

for the heterogeneous surface energy distribution of activated carbon.


qe =KC" Eqn 4.1

Where:

qe = the mass of contaminant adsorbed per unit mass of carbon

K = "adsorption capacity", Freundlich constant









C = concentration of the adsorbate in solution at equilibrium

1/n = strength of adsorption, Freundlich constant



Developed originally for air phase adsorption, the Langmuir equation (Eqn 4.2) has

been widely adapted for aqueous systems by replacing the original pressure term with a

term representing the solute equilibrium concentration.

1 1 1
S-=- + Eqn 4.2
qe Q', bQ, C,

Where:

qe = the mass of contaminant adsorbed per unit mass of carbon

Qa = maximum adsorption capacity

b = net enthalpy of adsorption

Ce = equilibrium concentration of adsorbate in solution



The Langmuir equation draws its theory from the condensation and evaporation of

gas molecules on a solid adsorbent surface (Weber and DiGiano 1996). Condensation

takes into account available adsorption sites on the adsorbent surface as well as the rates

at which the adsorbates contact the adsorbent surface. Three basic assumptions are

encompassed within the Langmuir equation: (1) adsorption energy is constant throughout

the activated carbon surface and is independent of surface coverage (2) there are no

interactions between adsorbate molecules or migration of adsorbates to other adsorption

sites (3) the carbon surface will only support a monolayer adsorption of the adsorbate

(Weber and DiGiano 1996).









While adsorption isotherms are useful in developing adsorbate-adsorbent

relationships, they are time intensive and non-specific in regards to adsorbate-adsorbent

structural interactions. Relying solely on isotherm relationships would require extensive

laboratory analysis for each individual compound (and condition) that is to be analyzed.

The adsorption of a chemical compound to activated carbon is dependent on both the

chemical and physical properties of the carbon and the compound. It should then follow

that by knowing the chemical and physical properties, the adsorbate-adsorbent

relationship can theoretically be predicted. Using known parameters to predict the

activity of a compound is the focus of QSAR modeling. By applying an empirically

based model to an appropriate system, the mechanisms within the adsorption process can

be described and accounted for without being fully disclosed. QSAR is becoming widely

known in the medical and biological fields, but has yet to see a dominating presence in

adsorption study (Hansch and Fujita 1995). Blum et al. (1994) have developed a QSAR

for activated carbon adsorption in water using 363 organic compounds. While this model

is noteworthy, it is based on literature-derived constants-which may vary drastically

with respect to experimental procedure and analysis-without a sturdy foundation, these

models will not accurately portray the adsorption process. Table 4.3 illustrates the

discrepancies which can be found in the literature.

Table 4.3. Reported Values of Freundlich K for Benzene.
K Source

0.12602 Noll 1999

1.0 Dobbs and Cohen 1980

41.7 Loll et al. 2004









Blum et al. (1994) show an understanding of this vital point as the authors mention that

"...in applying QSAR based on literature data, it is imperative to consider the relevance

of the data source to one's own investigative situation." However, when used in

conjunction with accurately descriptive parameters (i.e., case specific), QSAR can be

powerful tools for understanding activated carbon adsorption.

In recent years, carbon scientists have begun to examine the potential of QSAR

analysis for gaining insight into the methods) of activated carbon adsorption (Blum et al.

1994, Brasquet et al. 1997). To date, no compelling methodologies have been established

for developing a protocol for QSAR-adsorption analysis. The studies examined in the

literature typically utilize literature derived isotherm constants which are extracted from a

variety of sources. This method of selecting parameters can become somewhat

haphazard, as many sources use different experimental conditions (and different carbons)

when developing these isotherm constants. To avoid this inefficient approach, this study

has experimentally derived all dependent variables-under standardized testing

conditions-to create uniformity between the compounds in the training set. A uniform

approach to parameter development will lead to more reliable QSAR correlations. In

addition to carefully developing the dependent variables, this study includes an array of

electrical, steric, hydrophobic, and physical parameters as independent variables to more

accurately describe and investigate the role of the adsorbate in the adsorption process.

Also in contrast to most QSAR-adsorption studies present in the literature, this study

implements a series of rigorous statistical tests before a correlation is accepted.

It is the hypothesis of this study that the development of a robust QSAR will

enhance the understanding of the adsorption process of F400 with this given training set









of monosubstituted benzenes. Ultimately, an enhanced understanding of the adsorption

process will lead to an optimized use of activated carbon.

Experimental

Adsorption Isotherms

Adsorption isotherms were developed by adding a known concentration of

contaminant, 100-[tg/L, and activated carbon to distilled deionized (DDI) water at a pH

of 5.5 + 1.5, the variation is due to the use of unaltered DDI water. The experimental

data was collected for each contaminant individually using powdered Calgon F400

activated carbon, which was dosed as a slurry so that accurate quantities could be

developed, with 100-[tg/L solutions of each compound in the training set (see Table 4.4).

The system was mixed end over end in a gastight syringe on a rotator until equilibrium

was reached. After mixing, samples were filtered into VOC vials using 25-mm

Fisherbrand nylon syringe filters which were sized to allow for only the PAC to be

removed (both 0.20-jam and 0.45- jam were used, depending on their availability).

Samples were sealed with Teflon septa caps and stored at 4C until analysis. All samples

were analyzed with a Saturn 2100T GC/MS using a Supelco VOCOLTM Fused Silica

capillary column with a Supelco SPME 75-itm CARBOXENTM PDMS fiber, utilizing an

oven temperature program that holds at 400C for 2 minutes and then climbs to 210C at a

rate of 80C/min. Each isotherm was replicated and was only considered valid if the RSD

was less than 20% and the R-square value was 0.70 or greater.

Isotherm data was formatted using the Freundlich and Langmuir isotherm

equations (Eqn 4.1 and 4.2). The Freundlich and Langmuir constants were extracted

from the graphical interpretation of the data for each compound in the training set. The






47


graphics in Figure 4.1 demonstrate (A) a dose removal curve and (B) a graphical

representation of the Freundlich isotherm using n-butylbenzene as an example.






























0





8

7.8

7.6

7.4
0


0
7.2
-
o
7

6.8

6.6

6.4


0 5 10 15 20
Cone PAC (ppm)


1 2 3 4

log Cf


Figure 4.1. Example of Isotherm Development. (A) Dose removal curve for
Isobutylbenzene and Calgon F400 Activated Carbon. (B) Freundlich Isotherm
of Isobutylbenzene and Calgon F400 Activated Carbon.


y = 0.4199x + 5.8488
R2= 0.9812










Kinetic Data Collection

Kinetic studies proceeded in a similar manner, taking each contaminant

individually at a concentration of 100-[Lg/L with F400 at a constant concentration of 20-

mg/L. Kinetic data was collected in the same manner as the isotherm data, with the

exception that sample run durations were: 3, 5, 10, 15, 30, and 60 minutes. Following

each kinetic run, the samples were filtered, stored, and analyzed using the same methods

as described for the equilibrium samples. Kinetic data was incorporated into the QSAR

by determining the time for F400 to remove 86% of the target contaminant.

Experimentally derived values for this kinetic parameter (log t) and selected isotherm

constants are shown in Table 4.4.

Table 4.4. Experimentally Derived Values for log K, log Q, and log t.
Compound log K R2 log Q R2 log t R2

Fluorobenzene 5.38 0.91 1.01 0.88 1.17 0.94
Chlorobenzene 5.34 0.99 1.19 0.95 1.08 0.94
Bromobenzene 5.38 0.99 1.33 0.93 1.03 0.98
Iodobenzene 5.34 0.99 1.35 0.93 1.20 0.77
Nitrobenzene 5.05 0.95 1.43 0.96 0.88 0.98
Aniline 4.38 0.97 1.26 0.99 1.78 0.93
Phenol 3.64 0.90 1.58 0.95 --- ---
Benzaldehyde 4.82 0.71 --- --- 0.75 0.69
Isopropylbenzene 5.38 0.92 1.53 0.87 1.06 0.85
n-butylbenzene 4.62 0.92 2.05 0.94 0.45 0.92
Isobutylbenzene 5.00 0.81 2.80 0.94 -0.01 0.96
t-butylbenzene 4.25 0.83 2.89 0.81 0.94 0.96
Benzene 4.94 0.89 0.92 0.96 1.76 0.91


Establishment of QSAR

The first essential step in the development of the QSAR was the selection of a

suitable training set. For the parent compound, benzene was selected, and

monosubstituted functionality drove the selection of the substituents. For this study it

was desired to use single substitutions within the training set to investigate the impact of

a sole functional group on the adsorption system. In the selection of monosubstituted









benzenes it is important to consider only the addition of functional groups that are well

parameterized-i.e., that can be described by an available parameter within the QSAR.

Non-descriptive groups cannot be well represented in the QSAR and therefore should be

avoided in the selection of a training set. It is also important to include compounds that

produce a wide spread of parameter values to ensure that the training set is representing a

good range of electronics, sterics, hydrophobic, and chemical properties.

For this study, monosubstituted benzenes were selected to demonstrate a large range of

steric, electronic, hydrophobic, and chemical parameterization. Table 4.5 illustrates the

values of these parameters for the training set.

After the training set was chosen, each parameter was examined for collinearity.

When plotted, any parameters that showed a linear correlation of R2 > 0.4 were

considered collinear and were not used in multilinear regression analysis. Parameters

representing intrinsic properties of the compounds (e.g., hydrophobic, electronic, steric,

chemical, physical properties) were incorporated into the QSAR as independent

variables, while the parameters that were experimentally derived in this adsorption

system (e.g., isotherm constants) were incorporated as dependent variables using SPSS

12.0 software (SPSS., Inc., Chicago, IL, USA). Classical QSAR procedure, as

developed by Hansch, was followed for the development of correlations (Hansch and

Fujita 1995). Linear, polynomial, and multilinear correlations were employed to describe

the data.













Table 4.5. Selected Descriptors of Substituent or Compound Electronics, Sterics, and Hydrophobicity (Hansch et al., 1995; Perry and
Green, 1997; Schwartzenbach et al., 2003; MOPAC, 2004).
Monosubstituted Benzene Electronics Sterics Hydrophobicity Physical Properties
X Descriptors Descriptors Descriptors


Log MW Caq TOTL
OInd ACCP MR Taft Es Ko (g/ (mass%)
o mole) %)

Fluorobenzene (FB) 0.52 -0.115 0.09 -0.55 2.27 0.14 96.1 0.154 5.21
Chlorobenzene (CIB) 0.47 -0.103 0.6 -0.97 2.84 0.39 113 0.0387 5.55
Bromobenzene (BrB) 0.44 -0.097 0.89 -1.16 2.99 0.86 157 0.0387 5.7
Iodobenzene (odoB) 0.39 -0.099 1.39 -1.62 3.28 1.12 204 0.0387 5.8
Nitrobenzene (NB) 0.64 0.039 0.74 -2.52 1.85 -0.85 123 0.0193 6.07
Aniline (Ani) 0.12 -0.169 0.54 -0.61 0.9 -1.23 93.1 0.21 5.84
Phenol (Ph) 0.29 -0.145 0.28 -0.55 1.48 -1.12 94.1 3.38 5.64
Benzaldehyde (Bzal) 0.3 -0.066 0.69 --- 1.48 -0.32 106 8.66 7.02
Isopropylbenzene (IsoPB) 0.01 -0.106 1.5 -1.71 3.66 1.22 120 0.3 7.18
n-butylbenzene (nBB) -0.04 -0.11 1.96 -0.16 4.26 2.13 134 0.0056 9.71
Isobutylbenzene (IsoBB) -0.03 -0.107 1.96 -2.17 4.01 1.7 134 0.0015 8.19
t-butylbenzene (tBB) -0.07 -0.107 1.96 -2.78 4.11 1.98 134 0.001 7.27
Benzene (Benz) 0 -0.102 0.1 0 2.13 0 78.1 0.178 4.97



Where olnd represents the inductive effects described by the Hammett Constant, ACCP represents the charge of the para-carbon on the
functional group, MR represents molar refractivity, Es represents Taft's steric factor, log Kow is the octanol-water partitioning
coefficient, n7 represents the hydrophobicity of the compound, MW is the molecular weight, Caq is the aqueous solubility in water, and
TOTL is the total length of the compound, calculated by the bond lengths of the compound in the longest direction.









Correlations that resulted from the QSAR analysis had to meet a series of stringent

statistical criteria to be considered valid:

4. No collinearity of independent variables (for MLR correlations)

5. R > 0.7

6. F-Ratio values must meet a 95% confidence limit

7. Standard error considered

8. Q2 determined for goodness of predictability

Condition 1 is only applicable in regards to multi-variable analysis. Independent

variables were linearly correlated; variables that produced a R2 value of 0.4 or higher

were considered collinear and were excluded from the QSAR analysis. Conditions 2 and

3 were helpful in identifying poor correlations. The F-Ratio was determined using the

number of compounds in each correlation (k) and two total data sets (n), observed and

calculated values. The degrees of freedom were calculated for the variance between (vi =

n-1) and within (v2 = (n*k)-n) the data sets and compared to an upper 5% distribution

table to determine the significance of the observed and calculated data sets. As the F-

Ratio values become larger than the F-distribution values, the data sets are considered to

be more similar. Therefore, high F-Ratio values indicate the predicted data from the

QSAR correlation is in agreement with the observed data. The standard error (Condition

4) represents fluctuations in the sampling set and can be seen in the form of residuals in

the regression lines. Clearly, low values of standard error reflect better correlations

within the training set.

Condition 5 is an essential step in measuring the validation of the correlations that

result from the QSAR process. When performing a Q2 analysis, it is ideal to introduce a

new compound to test the predictability of the QSAR correlation; however if the









introduction of an additional compound is not feasible, an alternative method, roughly

referred to as the "take one out" method, will suffice. For this alternative method, one

compound from the training set is removed and the correlations are recalculated. The

new correlations are then used to test the ability of the QSAR correlation to predict the

value of the compound that was removed. This process is repeated for each compound

within the training set and Equation 4.3 is used to calculate the value of Q2


Q2 1 i(Y,observed ,predicted, new)2 Eqn4.3
1(Y ,observed Y)

While no set limit has been established for Q2 values, larger values indicate a better

ability for the QSAR correlation to predict the observed values within the given system.

Ideally the Q2 should be close to the value for the adjusted R2, thus indicating that the

new correlation did not cause a significant shift in the predicted data. For this study a

difference of 0.2 between the Q2 and adjusted R2 was considered the maximum limit for

variation.

Discussion of Results

From the experimentally derived isotherm parameters, only the terms representing

the adsorption capacity (log K and log Q) yielded viable correlations for this training set;

therefore, the isotherm data listed in Table 4.5 only shows these isotherm parameters. It

is interesting to note that log K and log Q do not correlate with each other within this

training set. In fact, a plot of these two parameters yields an R2 value of 0.0804. This

lack of relationship implies that while log K and log Q both represent an adsorption

capacity, their actual derivation (and definition) is quite different. Webber and DiGiano

(1996) present a relationship between the Freundlich and Langmuir constants shown in

Eqn. 4.4:









KF = Qa,g bgn Eqn. 4.4

However, the experimental results found in this study do not support this (or any)

relationship between Langmuir and Freundlich constants. Scientists and carbon users

should take caution when using a single parameter, such as an adsorption capacity, to

describe a system.

Using the experimentally derived isotherm constants and kinetic descriptors, QSAR

analyses were performed and correlations were derived. Initially, hundreds of

correlations were developed and tested, yet only the six correlations shown in Table 4.6

satisfied all of the statistical validation requirements.

Correlation 1, shown in Table 4.6 and Figure 4.2, indicates that as aqueous

solubility and molar refractivity increase, log K, the Freundlich adsorption capacity,

decreases. This decrease in adsorption is expected, as the increasing solubility of a

compound will lower its tendency to come out of solution and sorb onto the carbon. This

trend is commonly seen in the field of adsorption and is described by the Lundelius Rule.

The increase in molar refractivity (MR), which is a frequently used indicator of molecular

volume, correlates with a decrease in adsorption capacity, which is also expected for a

microporous carbon such as F400. The largest compounds in this training set reach

nearly 10 A in length, when estimated through the summation of bond lengths in the

longest direction of the compound. When estimating adsorption, Kasaoka and

coworkers, as reported by Tennant and Mazyck (2003), have asserted that a compound

will adsorb in a pore that is 1.5 to 2 times its diameter. Using this estimator, it is

reasonable that the largest compounds in this training set (e.g., n-butylbenzene and

isobutylbenzene) will not undergo optimized adsorption with a predominately









Table 4.6. Selected Results of QSAR Analysis with log K,
Dependent Variables.ab


log Q, and log t as the


Correlations n R2 Adj. SE F- Q2

R2 Ratio


1.) Log K =

-0.196(0.0915)Caq-0.258(0.326)MR+5.334

2.) Log K =

-0.17(0.0765)Caq+0.851(0.77)Oind+4.857

3.) Log Q =

0.558MR2-0.457MR+1.242

4.) Log K =

-0.187(0.083)Caq-0.147(0.152)TOTL+6.024

5.) Log t =

0.197(0.212)Caq-0.242(0.146)TOTL+2.52

6.) Log t =

-5.836(6.68)ACCP-0.267(0.292)TOTL+2.162


0.693



0.749



0.781



0.724



0.715



0.703


0.631



0.699



0.733



0.669



0.651



0.637


a Caq = aqueous solubility (mass %), MR = molar refractivity, TOTL


0.33



0.30



0.33



0.31



0.29



0.30


11.260



14.914



16.07



13.131



11.266



10.644


estimated total


length of compound, ACCP = measure of the para-carbon charge. Numbers in
parentheses represent the 95% confidence interval (+/-) for the coefficient.
n = number of compounds used in correlation development, Adj. R2 = adjusted R2, SE
= standard error. Due to inefficient data, Correlation 3 does not include benzaldehyde
and Correlations 5 and 6 do not include phenol.


0.458



0.607



0.582



0.560



0.603



0.694










microporous carbon (average pore diameter less than 20 A) such as F400. The results of

this correlation appear to be sound, as only one compound showed a percent difference

between observed and predicted values that was greater than 10%, t-butylbenzene.


6-


5.5
Flb
Benz* NB Clb *
NB
Bzal* BrB
5- lodoB*#
tBB nBB IsoPB
'a IsoBB
S4.5
An





3.5
3.5 4 4.5 5 5.5
Observed

Figure 4.2. QSAR Correlation 1: Predicted versus Observed Values of log K.

The trends described by Correlation 2, shown in Table 4.6 and Figure 4.3, show a

decrease in adsorption capacity (log K) with respect to an increase in aqueous solubility

and a decrease in Hammett's constant, Gind. The relationship between log K and the

aqueous solubility of the training set is expected and is consistent with the results of

Correlation 1. However, it is unexpected that a decrease in GInd would lead to a reduced

adsorption capacity. A decrease in GInd indicates an increase in the electron donating

potential of a compound, due to a higher electron density about the aromatic ring of the

compounds in this training set. These results contrast with the 7n-7n bonding theory

developed by Coughlin and Ezra (1968) and accepted by numerous renowned carbon

scientists, such as Radovic (2001). According to the 7n-7n theory, an increase in the










electron density about the benzene ring should enhance the ability of the n electrons to

interact, thus enhancing the potential for adsorption. Echoing the results seen in

Correlation 1, only t-butylbenzene yielded a percent difference between predicted and

observed value that was greater than 10%. Further study is necessary to confirm this

trend will apply to a greater training set.



6-


5.5
NB CLB FB
BrB
5- Bzal IodoB
nBB
StBBIo *

S4.5

An

4-
Correlation 2: log K = -0.17Caq+O.-851sigI+4.857
h R=0.749, Adj. R2=0.699, se=0.30, F-Ratio=14.914, Q2=0.607
3.5
3.5 4 4.5 5 5.5
Observed

Figure 4.3. QSAR Correlation 2: Predicted versus Observed Values of log K.
Correlation 3, shown in Table 4.6 and Figure 4.4, relates the Langmuir adsorption

capacity, log Q, with MR. The results found in this correlation show that the adsorption

capacity increases as the MR-or molecular volume-increases, which contradicts the

results found in Correlation 1. It does not seem logical that a microporous carbon, such

as F400, would have an increasing adsorption capacity with increasing compound

volume, as the largest compounds in this training set are approximately 10 A in length. It

should be noted that the largest compound in this training set, n-butylbenzene, showed

the largest residual from the trendline in Correlation 3. The differences between










Correlation 1 and Correlation 3 confirm that the Langmuir Q and Freundlich K are

describing different mechanisms for adsorption and should not be used interchangeably.

Although more research comparing log K and log Q is necessary, in this particular

instance, it appears that the Freundlich K is most accurately describing the adsorption of

the training set with F400. Despite the R2 value and sufficient statistical results, nine of

the twelve compounds tested for this correlation showed greater than a 10% difference

between predicted and observed values. In fact, of those nine compounds, six showed

nearly or greater than a 20% difference. More research should be performed with an

expanded training set to confirm the effectiveness of log Q for describing adsorption in

relation with MR.


3-

2.5IsoBB tBB
2.5 nBB *


2-
IsoPB
IodoB *
1.5 Br
P FB NB
(- Benz B
** *Ph
1 C1B An


0.5 -
0.5 Correlation 3: log Q = 0.558MR -0.457MR+1.242
R=0.781, Adj. R=0.733, se=0.33, F-Ratio=16.067, Q 0.582
0 ---------------------------------
0 0.5 1 1.5 2 2.5 3 3.5
Observed

Figure 4.4. QSAR Correlation 3: Predicted versus Observed Values of log Q.

As shown in Table 4.6 and Figure 4.5, Correlation 4 states that as aqueous

solubility and total length increase, log K decreases. These results were expected, and

support Correlations 1 and 2 with respect to aqueous solubility. The total length










parameter is estimated from the bond lengths of the compounds in the longest dimension.

An increase in total length indicates an increase in the bulkiness of the compound, and as

expected for this microporous carbon, relates to a decrease in adsorption capacity. These

results for total length are in agreement with the MR results for Correlation 1. As was

found with Correlations 1 and 2, only one compound in Correlation 4 shows a percent

difference between predicted and observed values greater than 10%, t-butylbenzene.


6-


5.5
Benz Flb
NB
BrB
SIodoB BrB
5- tBB+ Bzal *-
.2IsoPB
'a IIsoBB

An
4.5- A. n nBB



Correlation 4: log K = -0.187Caq-0.147TOTL+6.024
Ph R2=0.724, Adj. R2=0.669, se=0.31, F-Ratio=13.131, Q2=0.560
3.5
3.5 4 4.5 5 5.5
Observed

Figure 4.5. QSAR Correlation 4: Predicted versus Observed Values of log K.

Correlation 5, as shown in Table 4.6 and Figure 4.6, shows the relationship

between aqueous solubility, total length, and log t. The value log t is a kinetic parameter

that represents the amount of time that a 20-ppb slurry of F400 took to achieve 86%

removal of a 100-ppb concentration of a target contaminant. As expected, and in support

with Correlations 1, 2, and 4, a greater aqueous solubility resulted in a longer time to

achieve 86% adsorption. However, in Correlation 5, a decrease in total length denoted an

increase in the time to reach 86% adsorption. This finding appears to contradict with the










results of Correlations 1 and 4 which show that an increase in molecular volume and an

increase in total length result in a decrease in adsorption capacity (log K). This

unexpected relationship between log K and total length could be an effect of steric

hindrances, buttressing of the molecules against the carbon surface, pore size distribution.

It should be noted that while all other correlations describe solely equilibrium conditions

(which the Freundlich and Langmuir equations require by definition); log t does not

represent equilibrium conditions between the carbon and contaminant. For Correlation 5,

all but two compounds (aniline and iodobenzene) showed a 10% or greater difference

between predicted and observed values. For these reasons, the log t results shown here

cannot be accurately compared to the isotherm constant results; therefore further research

should be done to confirm these findings.


2-
An
1.8 An

1.6

1.4 Flb
Clb b Benz
1.2 BrBt*
NB# oB
.- 1

0.8 tBB IsoPB

0.6 IsoBB

0.4
2 nBB Correlation 5: log t = 0.197Caq-0.242TOTL+2.52
0.222
SR=0.715, Adj. R =0.651, se=0.29, F-Ratio=l 1.266, Q =0.603
0
-0.05 0.45 0.95 1.45 1.95
Observed



Figure 4.6. QSAR Correlation 5: Predicted versus Observed Values of log t.











1.8
An
1.6 -
1b
1.4 *
Clb Benz
lodoB
1.2- BrB* *

1.

2 0.8 tBB IsoPB
06 Bzal
0.6 IsoBB Bz

0.4 NB

0.2 Correlation 6: log t = -5.836ACCP-0.267TOTL+2.162
R2=0.703, Adj. R2=0.637, se=0.30, F-Ratio=10.644, Q2=0.694
0 -----------------------------
-0.05 0.45 0.95 1.45 1.95
Observed

Figure 4.7. QSAR Correlation 6: Predicted versus Observed Values of log t.

The final correlation examined in this analysis is Correlation 6, shown in Table 4.6

and Figure 4.7. Correlation 6 indicates that an increase in ACCP (the charge of the

carbon in the para position on the functional group of the target compound) and total

length results in a decrease in the time to reach 86% adsorption of the target contaminant

by F400. As ACCP increases, the functional group of the compound becomes more

electron withdrawing, which indicates a reduced electron density on these carbon atoms.

Correlation 6 shows that as the para carbon becomes more electron withdrawing, the time

to 86% decreases-this indicates that a reduced electron density about the carbon is

enhancing the uptake of the compound by F400. This finding, as seen with Correlation 2,

contradicts the 7t-7t bonding theory. Yet, as was seen in Correlation 5, the total length and

log t are relating in an unexpected manner-showing a decrease in time to reach 86% as

the compound length increases. Much like Correlation 5, in Correlation 6, all values

except one (iodobenzene) showed percent differences between predicted and observed









values to be greater than 10%. As stated previously, the method for parameter

development for log t is not ideal and further study should be performed to clarify these

results.

Overall, as expected, Correlations 1, 2, and 4 support that as aqueous solubility

increases, log K (adsorption capacity) decreases. As the steric factors of the compounds

increase, Correlations 1 and 4 show that log K decreases. Though unexpected,

Correlation 3 confirms the finding that log K and log Q do not demonstrate a distinct

relationship with each other. For this reason, caution should be used when utilizing these

parameters to describe an adsorption system. The correlations incorporating log t,

Correlations 5 and 6, require further testing to verify the effectiveness of log t as a valid

parameter for describing carbon adsorption. It has also been suggested, through

Correlations 2 and 6, that 7n-7n bonding may not be the dominate mechanism for

adsorption in this training set. The successful correlations between aqueous solubility,

sterics, and log K demonstrate optimistic effectiveness for using these parameters for

adsorption prediction and signify the value of the functional and effective QSAR protocol

utilized in this study. The critical steps of this protocol include:

* The selection of a parent compound and training set

* Standardized experimentation, at equilibrium conditions, of the training set with
activated carbon(s)

* Uniform analysis of the experimental data to formulate representative dependent
variables

* Careful consideration and selection of reliable and well documented independent
variables

* Testing for collinear parameters

* Meticulous QSAR modeling






63


* Rigorous statistical validation of the QSAR correlations

The use of this protocol may be an essential step in working towards the goal of

understanding the mechanisms of carbon adsorption phenomenon.














CHAPTER 5
CONCLUDING REMARKS

As stated by Mattson et al. (1969a), it is essential for the advancement of the field

of activated carbon to identify and understand the mechanisms by which adsorption

occurs. An understanding of the role of functional groups in the adsorption process is a

key component to unveiling the mechanisms behind the process. The activation process

is pivotal in the development of the surface chemistry, and indeed in the formation of

functional groups present on the activated carbon. A recent workshop held by the

National Science Foundation has emphasized the potential of QSARs as a tool for

uncovering mechanisms in poorly understood phenomenon in the field of water treatment

(NSF 2004).

Dabrowski states in his work "Adsorption-From Theory to Practice" that "..

there is a need for close co-operation between theoretical and experimental groups, in

which the experiments and models are designed to complement each other" (Dabrowski

2001). Overwhelmingly, in the literature, researchers are either developing models using

past research or performing research using previously developed models. As Dabrowski

implies, in order to have synergy between models and the experimental data, the

researchers must integrate both into their designs. One approach that takes this into

consideration is the experimental-modeling method proposed in this study. In this

method, isotherm and kinetic data is collected for a training set of compounds and is

included along with widely accepted physical and chemical constants to form a robust

QSAR. This approach takes into consideration the effects of contaminant surface









chemistry and relies upon both experimental and modeling data to describe the

mechanisms of adsorption. Ultimately, this method shall directly incorporate the effects

of carbon surface chemistry as independent variables. The addition of such parameters

will make for a more robust QSAR which may lead to a deeper understanding of the role

of surface chemistry in the adsorption process.

If it is the intent of scientific community to advance the field of activated carbon,

the mechanisms of adsorption must be identified and understood. Without the knowledge

behind the process, the use of activated carbon will not be able to surpass other separation

technologies. An experimental-modeling method is a doorway to understanding

adsorption phenomena.















LIST OF REFERENCES


Adamson Arthur W. Physical Chemistry of Surfaces: Fifth Edition. New York, New
York: John Wiley & Sons, Inc., 1990.

Boehm H.P. Functional Groups on the Surfaces of Solids. Angewandte Chemie:
International Edition. Vol. 5, No. 6, pp. 533-622, 1966.

Blum Diane J.W., Suffet Irwin H., Duguet J.P. Quantitative Structure Activity
Relationship Using Molecular Connectivity for the Activated Carbon Adsorption of
Organic Chemicals in Water. Water Research. Vol. 28, No. 3, pp. 687-699, 1994.

Brasquet Catherine, Subrenat Etienne, Le Cloirec Pierre. Selective Adsorption on
Fibrous Activated Carbon of Organics from Aqueous Solution: Correlation
between Adsorption and Molecular Structure. Water Science Technology. Vol. 35,
No. 7, pp. 251-259, 1997.

Carey Francis A. Reactions ofArenes: Electrophilic Aromatic Substitutions. The
McGraw-Hill Companies Student Online Learning Center. Available [Online]:
http://www.mhhe.com/physsci/chemistry/carey/student/olc/graphics/carey04oc/ref/
chl2substituenteffects.html. Accessed December 6, 2003.

Chiou Cary T. Partition and Adsorption of Organic Contaminants in Environmental
Systems. Hoboken, New Jersey: Wiley-Interscience, 2002.

Coughlin Robert W. and Ezra Fouad S. Role of Surface Acidity in the Adsorption of
Organic Pollutants on the Surface of Carbon. Environmental Science and
Technology. Vol. 2, No. 4, pp. 291-297, 1968.

Coughlin Robert W., Ezra Fouad S., Tan Ricardo N. Influence of Chemisorbed Oxygen
in Adsorption onto Carbon from Aqueous Solution. Journal of Colloid and
Interface Science. Vol. 28, No. 3/4, pp. 386-396, 1968.

Dabrowski A. Adsorption-from theory to practice. Advances in Colloid andInterface
Science. Vol. 93, pp. 135-224, 2001.

Dobbs Richard A., Cohen Jesse M. Carbon Adsorption Isotherms for Toxic Organics.
EPA: Municipal Environmental Research Laboratory, Office of Research and
Development. April 1980.









Eriksson Lennart, Jaworska Joanna, Worth Andrew P., Cronin Mark T.D., McDowell
Robert M., Gramatica Paola. Methods for Reliability and Uncertainty Assessment
and for Applicability Evaluations of Classification and Regression Based QSARs.
Environmental Health Perspectives. Vol. 111, No. 10, pp. 1361-1375, August
2003.

Ergun S. and Mentser M. Reactions of carbon i/th carbon dioxide and steam. Chemistry
and Physics of Carbon. Vol. 1, New York, New York: Marcel Dekker, pp. 204-
261, 1965.

Fangmark Ingrid E., Hammarstrom Lars-Gunnar, Stromqvist Marianne E., Ness Amanda
L., Norman Paul R., Osmond Neale M. Estimation of Activated Carbon
Adsorption Efficiency for Organic Vapors: A Strategy for Selecting Test
Compounds. Carbon. Vol. 40, pp. 2861-2869, 2002.

Franz Marcus, Hassan Arafat A., Pinto Neville G. Effect of Chemical Surface
Heterogeneity on the Adsorption Mechanism of Dissolved Aromatics on Activated
Carbon. Carbon. Vol. 38, pp. 1807-1819, 2000.

Gregova K., Petrov N., Butuzova L., Minkova V., Isaeva L. Evolution of the Active
Surface of Carbons Produced from Various Raw Materials by Steam
Pyrolysis/Activation. Journal of Chemical Technology and Biotechnology. Vol.
58, pp. 321-330, 1993.

Guo J., Lua A.C. Characterization of Adsorbent Prepared from Oil-Palm Shell by CO2
Activation for Removal of Gaseous Pollutants. Materials Letters. Vol. 55, No. 5,
pp. 334-339, 2002.

Hammer Mark J. and Hammer Mark J. Jr. Water and Wastewater Technology: Fourth
Edition. Upper Saddle River, New Jersey: Prentice-Hall, 2001.

Hansch Corwin and Fujita Tosiho. Classical and Three-Dimensional QSAR in
Agrochemistry. Washington DC: American Chemical Society, 1995.

Hansch C., Leo A., Hoekman D. Exploring QSAR: Hydrophobic, Electronic, and Steric
Constants. Washington DC: American Chemical Society, 1995.

Johns Mitchell M., Marshall Wayne E., Toles Christopher A. The Effect of Activation
Method on the Properties of Pecan Shell-Activated Carbons. Journal of Chemical
Technology andBiotechnology. Vol. 74, pp. 1037-1044, 1999.

Karanfil Tanju and Kilduff James E. Role of Granular Activated Carbon Surface
Chemistry on the Adsorption of Organic Compounds. 1. Priority Pollutants.
Environmental Science and Technology. Vol. 33, No. 18, pp. 3217-3224, 1999.

Kipling J.J. and Shooter P.V. The Adsorption of Iodine Vapor by Graphon and Spheron.
Journal of Colloid and interface Science. Vol. 21, pp. 238-244, 1966.









Lindner Angela S., Whitfield Callie, Chen Ning, Semrau Jeremy D., Adriaens Peter.
Quantitative Structure-Biodegradation Relationships for Ortho-Substituted
Biphenyl Compounds Oxidized by Methylosinus Trichosporium OB3b.
Environmental Toxicology and Chemistry. Vol. 22, No. 10, pp. 2251-2257, 2003.

Lloyd Ted. Quantitative Structure Activity Relationship Glossary. [Available Online]:
http://www.chem.swin.edu.au/modules/mod4/glossq.html. Accessed December 27,
2004.

Loll P., Hesselsoe M., Moldrup P., Henriksen K., Larsen C., and Dahlstrom K. GAC
Treatment of MTBE and Benzene Contaminated Groundwater. Submittedfor The
Fourth International Conference on Remediation of Chlorinated and Recalcitrant
Compounds. May 24-27, 2004, Monterey, CA. [Available Online]:
http://www.dmr.as/download_get.php?file=31. Accessed December 27, 2004.

Lua A.C. and Guo J. Activated Carbon Prepared from Oil Palm Stone by One-Step CO2
Activation for Gaseous Pollutant Removal. Carbon. Vol. 38, pp. 1089-1097,
2000.

Mattson James S., Mark Harry B. Jr., Malbin Michael D., Weber Walter J. Jr., Crittenden
John C. Surface Chemistry of Active Carbon: Specific Adsorption of Phenols.
Journal of Colloid andInterface Science. Vol. 31, No. 1, pp. 116-130, 1969.

Mattson James S. and Mark Harry B. Jr. Infrared Internal Reflectance Spectroscopic
Determination of Surface Functional Groups on Carbon. Journal of Colloid and
Interface Science. Vol. 31, No. 1, pp. 131-144, 1969.

Mattson James S., Lyon Lee, Mark Harry B. Jr., Weber Walter J. Jr. Surface Oxides of
Activated Carbon: Internal Reflectance Spectroscopic Examination of Activated
Sugar Carbons. Journal of Colloid and Interface Science. Vol. 33, No. 2, pp. 284-
293, 1970.

Molina-Sabio M., Gonzalez M.T., Rodriguez-Reinoso F., Sepulveda-Escribano A. Effect
of Steam and Carbon Dioxide Activation in the Micropore Size Distribution of
Activated Carbon. Carbon. Vol. 34, No. 4, pp. 505-509, 1996.

MOPAC 2002. CAChe Software, Fujita, 2002.

Muller Gunther, Radke C.J., Prausnitz J.M. Adsorption of Weak Electrolytes from Dilute
Aqueous Solution onto Activated Carbon: Part I. Single-Solute Systems. Journal
of Colloid and Interface Science. Vol. 103, No. 2, pp. 466-483, 1985.

National Science Foundation (NSF). Advancing the Quality of Water (AWQA): Expert
Workshop to Formulate a Research Agenda. March 10-12, 2004. [Available
Online]: http://enve.coe.drexel.edu/AQWA/report/report.htm. Accessed December
27, 2004.









Noll Kenneth, E. Fundamentals ofAir Quality Systems: Design of Air Pollution Control
Devises. Annapolis, MD: American Academy of Environmental Engineers, 1999.

Ouellette Robert J. Organic Chemistry: A BriefIntroduction, Second Edition. Upper
Saddle River, New Jersey: Prentice Hall, pp. 138-170 and 217-242, 1998.

Pacific Northwest National Laboratory (PNNL). Artificial Neural Networks. Battelle
Memorial Institute. Available [Online]:
http://www.emsl.pnl.gov:2080/proj/neuron/neural/what.html. Accessed December
2003.

Perry R.H. and Green D.W. Perry's Chemical Engineers' Handbook. Seventh Edition.
New York, New York: McGraw-Hill, 1997.

Petrov N., Gergova K., Eser S. Effect of Water Vapour on the Porous Structure of
Activated Carbon from Lignite. Fuel. Vol. 73, No. 7, pp. 1197-1201, 1994.

Petrucci R.H. and Harwood W.S. General Chemistry: Principles and Modern
Applications, Seventh Edition. Upper Saddle River, New Jersey: Prentice Hall, pp.
387-417, 1997.

Pope Jason P. Activated Carbon and Some Applications for the Remediation of Soil and
Groundwater Pollution. Available [Online]:
http://www.cee.vt.edu/program_areas/environmental/teach/gwprimer/group23/web
page.htm. Accessed December 6, 2003.

Radovic Ljubisa. Chemistry andPhysics of Carbon: Volume 27. New York, New York:
Marcel Dekker, pp. 312-405, 2001.

Rodriguez-Reinoso F., Molina-Sabio M., Gonzalez M.T. The Use of Steam and CO2 as
Activating Agents in the Preparation of Activated Carbons. Carbon. Vol. 33,
No.1, pp. 15-23, 1995.

Schwartzenbach, R.P., Gschwend, P.M., and Imboden, D.M. Environmental Organic
Chemistry, Second Edition. Hoboken, New Jersey: Wiley-Interscience, 2003.

Snoeyink, Vernon L. Adsorption of Organic Compounds. Water Quality and Treatment:
A Handbook of Community Water Supplies. New York, New York: American
Water Works Association, 1990.

Tendankara Library. Chemistry: Delocalized Electrons in the Benzene Pi System.
Available [Online]:
http://library.tedankara.k12.tr/chemistry/vol3/Delocalised%20molecular%20orbials
%20(Part%201)/zl04.htm. Accessed 6 December 2003.

Tennant M.F. and Mazyck D.W. Steam-pyrolysis activation of wood char for superior
odorant removal. Carbon. Vol. 41 No. 12, pp. 2195-2202, 2003.









Tomkow Kazimierz, Siemieniewska Teresa, Czechowski Franciszek, Jankowska Anna.
Formation of Porous Structures in Activated Brown-Coal Chars Using 02, CO2,
and H20 as Activating Agents. Fuel. Vol. 56, pp. 121-124, 1977.

United States Environmental Protection Agency (USEPA). Pollution Prevention and
Toxics: Nitrobenzene Factsheet. [Available Online]:
http://www.epa.gov/opptintr/chemfact/nitro-sd.txt. Accessed December 27, 2004.

United States Environmental Protection Agency (USEPA). Ground Water & Drinking
Water. Technical Factsheet on Benzene. [Available Online]:
http://www.epa.gov/OGWDW/dwh/c-voc/benzene.html. Accessed December 27,
2004.

Vidic Radisav D., Suldan Makram T., Brenner Richard C. Oxidative Coupling of
Phenols on Activated Carbon: Impact on Adsorption Equilibrium. Environmental
Science and Technology. Vol. 27, No. 10, pp. 2079-2085, 1993.

Weber Walter J. Jr. and DiGiano Francis A. Process Dynamics in Environmental
Systems. New York, New York: Environmental Science and Technology: A Wiley
Interscience Series of Texts and Monographs, 1996.

Whelton Andrew J. Temperature Effects on Drinking Water Odor Perception. Master
Thesis. Virginia Polytechnic Institute and State University, 2001.

Wigmans T. Industrial Aspects of Production and Use of Activated Carbons. Carbon.
Vol. 27, No. 1, pp. 13-22, 1989.

Wold Svante and Dunn William J. III. Multivariate Quantitative Structure-Activity
Relationships (QSAR): Conditions for Their Applicability. Journal of Chemical
Information and Computer Science. Vol. 23, pp. 6-13, 1983.

Wu Jufang, Stromqvist Marianne E., Claesson Ola, Fangmark Ingrid E., Hammarstrom,
Lars-Gunnar. A Systematic Approach for Modeling the Affinity Coefficient in the
Dubinin-Radushkevich Equation. Carbon. Vol. 40, pp. 2587-2596, 2002.















BIOGRAPHICAL SKETCH

Born and raised in Pensacola, Florida, Jennifer graduated with an International

Baccalaureate Diploma from the International Baccalaureate Program at Pensacola High

School in May 1997. Continuing her education at the University of Florida, she defended

for highest honors and received a Bachelor of Science degree in environmental

engineering sciences in May 2002. For her master's work, Jennifer studied under Dr.

David Mazyck with a focus on the adsorption of benzenes and monosubstituted benzenes

onto activated carbon for water treatment applications. Jennifer currently resides in

Newberry, Florida, with her husband Steve and in August 2004 she joined the

water/wastewater engineering department in the consulting firm of Jones Edmunds &

Associates.




Full Text

PAGE 1

ADSORPTION OF SUBSTITUTED AROMATIC COMPOUNDS BY ACTIVATED CARBON: A MECHANISTIC APPROACH TO QUANTITATIVE STRUCTURE ACTIVITY RELATIONSHIPS By JENNIFER A. MCELROY A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF ENGINEERING UNIVERSITY OF FLORIDA 2005

PAGE 2

Copyright 2005 by Jennifer A. McElroy

PAGE 3

Knowledge and understanding is a gift fr om God to be used for His glory. “Whatever you do, work at it with all your h eart, as working for the Lord, not for men, since you know that you will receive an inheritance from the Lord as a reward. It is the Lord Christ you are serving.” Colossians 3:23-24

PAGE 4

iv ACKNOWLEDGMENTS The utmost acknowledgment goes to my Lord and Savior Jesus Christ, for without Him nothing would have existence or find purpos e. He has guided my steps and blessed me with opportunities that I pray will be used for His glor y. With all my heart I thank my husband, Steve McElroy, for the wonderful blessi ng and comfort that he is in my life. Heartfelt gratitude goes out to my parents, Gary and Sandra Hobbs, who have always been an amazing source of strength and en couragement to me, and to my brother Matthew who has taught me courage and perseverance. Without the help of my professors I woul d have never come this far. Dr. David Mazyck has been an astounding source of encouragement and ha s taught me to strive for excellence. Dr. Angela Lindner has been a wonderful source of support and has been crucial in the development of this research. I would also like to sincerely thank all of the scientists who helped me in various stages of my research, Dr. Susan Sinnott, everyone at Engineering Performance Solutions: Ric k, Ron, and Matthew Tennant, my fellow graduate researchers: Ameena Khan, Chris tina Ludwig, Morgana Bach, Jennifer Stokke, Thomas Chestnutt, Jack Drwiega, and Vi vek Shyamasundar. Special gratitude and acknowledgment are extended to Maria Paituvi for her noteworthy help and dedication to this work.

PAGE 5

v TABLE OF CONTENTS page ACKNOWLEDGMENTS.................................................................................................iv LIST OF TABLES............................................................................................................vii LIST OF FIGURES.........................................................................................................viii ABSTRACT....................................................................................................................... ix CHAPTER 1 INTRODUCTION........................................................................................................1 2 LITERATURE REVIEW.............................................................................................2 Activated Carbon..........................................................................................................2 The Activation Process.................................................................................................2 CO2 and H2O Activation.......................................................................................3 Adsorption Factors........................................................................................................4 Surface Area..........................................................................................................4 Pore Size Distribution............................................................................................5 pH..........................................................................................................................5 Surface Chemistry.................................................................................................6 Predicted Adsorption Mechanisms...............................................................................7 Bonding Theory..............................................................................................7 Electron Donor Theory..........................................................................................9 Functional Group Formation......................................................................................10 Detection of Functional Groups..........................................................................10 Types of Functional Groups Formed...................................................................11 From CO2 activation.....................................................................................11 From H2O activation....................................................................................12 Factors controlling which groups are formed..............................................12 Precursor and conditions of activation.........................................................13 Optimizing the Use of Activated Carbon...................................................................15 Quantitative Structure Activity Relationships............................................................23

PAGE 6

vi 3 MATERIALS AND METHODS...............................................................................28 Isotherm Methods.......................................................................................................28 Development of Isotherm Data...................................................................................29 Target Contaminants...........................................................................................29 Benzene........................................................................................................29 Substituted benzenes....................................................................................30 Freundlich Isotherms...........................................................................................31 Langmuir Isotherms.............................................................................................32 Kinetic Study Methods...............................................................................................32 Development of Kinetic Equations.............................................................................33 QSAR Development...................................................................................................33 4 MANUSCRIPT...........................................................................................................38 Introduction.................................................................................................................38 Experimental...............................................................................................................46 Adsorption Isotherms..........................................................................................46 Kinetic Data Collection.......................................................................................49 Establishment of QSAR......................................................................................49 Discussion of Results..................................................................................................53 5 CONCLUDING REMARKS......................................................................................64 LIST OF REFERENCES...................................................................................................66 BIOGRAPHICAL SKETCH.............................................................................................71

PAGE 7

vii LIST OF TABLES Table page 2.1 Elemental Analysis of Apricot Stones Before and After Steam Activation at 600 C........................................................................................................................14 2.2 Reported Values of Freundlich K for Benzene........................................................17 2.3 Example of Parameters Commonl y Used in QSAR Development..........................25 3.1 Target Contaminants Used in Isotherm and Kinetic Studies...................................29 3.2 Selected Parameters Used in the QSAR Training Set..............................................35 4.1 Size and Aqueous Solubility of MIB and Phenol....................................................40 4.2 Selected Physical Properties of F400.......................................................................42 4.3 Reported Values of Freundlich K for Benzene........................................................44 4.4 Experimentally Derived Values for log K, log Q, and log t.....................................49 4.5 Selected Descriptors of Substituent or Compound Electronics, Sterics, and Hydrophobicity.........................................................................................................51 4.6 Selected Results of QSAR Analysis w ith log K, log Q, and log t as the Dependent Variables.a,b............................................................................................55

PAGE 8

viii LIST OF FIGURES Figure page 2.1 Representation of Point of Zero Charge.....................................................................6 2.2 Representation of sigma and pi bondi ng, arrows within the atoms of formaldehyde represent electron filling in outer orbitals...........................................8 2.3 Representation of the Benzene Pi Sy stem (Source: Tedankara Library, 2003).........8 3.1 Craig Plot Illustrating the Parameter Spread of a Hydrophobicity Constant ( ) and the Hammett Constant ( Ind)..............................................................................34 4.1 Example of Isotherm Development..........................................................................48 4.2 QSAR Correlation 1: Predicted vers us Observed Values of log K..........................56 4.3 QSAR Correlation 2: Predicted vers us Observed Values of log K..........................57 4.4 QSAR Correlation 3: Predicted vers us Observed Values of log Q..........................58 4.5 QSAR Correlation 4: Predicted vers us Observed Values of log K..........................59 4.6 QSAR Correlation 5: Predicted vers us Observed Values of log t............................60 4.7 QSAR Correlation 6: Predicted vers us Observed Values of log t............................61

PAGE 9

ix Abstract of Thesis Presen ted to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Engineering ADSORPTION OF SUBSTITUTED AROMATIC COMPOUNDS BY ACTIVATED CARBON: A MECHANISTIC APPROACH TO QUANTITATIVE STRUCTURE ACTIVITY RELATIONSHIPS By Jennifer A. McElroy May 2005 Chair: David Mazyck Major Department: Environmental Engineering Sciences Because of their widespread prevalence in ground and surface waters, aromatic compounds pose a significant risk to public health. Of the current biological, chemical, and physical methods for remediation of thes e contaminants, activated carbon has been chosen as the primary method of treatment by many water treatment facilities. Despite its widespread use, questions remain con cerning the economic feas ibility of activated carbon, and determining suitable carbons fo r specific contaminants often requires extensive experimentation, time, and cost. Pr evious work in developing predictors of carbon performance using quantitative struct ure-activity relationships (QSARs), has fallen short of providing accurate models for carbon adsorption. This study investigates the role of QSARs in predicting the adsorpti on of monosubstituted benzenes onto a single activated carbon and offers a general prot ocol for the implementation of QSARs for predictive adsorption analysis. This study has produced succ essful correlations among aqueous solubility, sterics, and log K. These correlations demonstrate optimistic

PAGE 10

x effectiveness for using these parameters for adsorption prediction and signify the value of the functional and effective QSAR pr otocol utilized in this study.

PAGE 11

1 CHAPTER 1 INTRODUCTION In 1969, Mattson et al. stated that “T he mechanisms by which active carbon functions to remove phenols from aqueous so lutions has never been clearly defined. A thorough understanding of the sorptive mechan isms is essential for accomplishing the most effective use of active carbon in water and wastewater treatment” (Mattson et al. 1969). Over 3 decades later, carbon scientists are still pondering the same dilemma. It is clear that activated carbon is an efficient, cost effective adsorbent for removing numerous organic materials from both gaseous and aque ous solutions. The general mechanisms of adsorption are known to be physical adsorpti on and chemisorption. It has also been shown that factors, such as pH and surf ace chemistry, play a significant role in the adsorption process (Kipling and Shooter 1966, Coughlin et al. 1968, Mattson et al. 1969, Muller et al. 1985, Vidic et al 1993, Radovic 2001). Yet, the mechanisms that directly lead to the adsorption of an adsorbate stil l remain elusive. Identifying the role of functional groups in the adsorption process is an essential compone nt to understanding these mechanisms. The carbon activation pr ocess is the fundam ental step in the development of surface chemis try and functional groups. Understanding the adsorption process will enable an optim ized use of activated carbon, eliminating the wasteful practice of trial and error ad sorption treatment. To help circumvent this black-box method of carbon dosing, various relationships are currently being established between the adsorbate and the adsorbent. Two of th ese relationships are ad sorption isotherms and Quantitative Structure Activity Relationships (QSARs).

PAGE 12

2 CHAPTER 2 LITERATURE REVIEW Activated Carbon Activated carbon is an adsorbent created by treating a carbonaceous material thermally and/or chemically so that it develops internal pores that yield its characteristic high surface area. Thermal activation is commonly employed, where a carbonaceous precursor (e.g., coal, wood, nutshells) is subjec ted to high temperatures typically ranging from 300 C to 800 C (Mattson et al. 1969). In chemical activation, the precursor is first exposed to a chemical (e.g., NaOH, H3PO4), which drives pore development and is then subjected to heat treatment. Heating temper atures are typically lower for the production of chemically activated carbon. Activated carbon is commonly used in separation processes involving organic com pounds, and is widely used in fields such as medicine, water and air treatment, the food and beve rage industry, and industrial processes. The Activation Process During the activation process, an oxidizing agent breaks down the graphite structures of the carbonaceous precursor, a pr ocess known as gasificati on, resulting in the formation of high energy areas that serve as adsorption sites. The resulting carbon parameters (e.g., pore size distribution (PSD ), functional groups) are dependent on the oxidant used. Steam (H2O) and CO2 are the most common agents used and will therefore be the focus of this section.

PAGE 13

3 CO2 and H2O Activation The literature generally agrees—with one notable exception (Wigmans 1989) which has been cited 134 times to date, acco rding to the Science Citation Index—that the use of CO2 as the oxidizing agent during carbon activation leads to an enhanced development of micropores; while the use of steam promotes the propagation of mesopores (Tomkow et al. 1977, Rodriguez-Re inoso et al. 1995, Molina-Sabio et al. 1996, Johns et al. 1999). Creati on of pores resulting from the endothermic oxidation of carbon with steam or CO2 follows Eqn 2.1 and 2.2: Using CO2: C + CO2 COsurface + CO(g) Eqn 2.1 Using steam: C + H2O COsurface + H2(g) Eqn 2.2 With sufficient heating, CO2 will break a carbon bond from the aromatic ring in the precursor. To satisfy the bond, the CO2 will leave behind an oxygen molecule at the surface, thus creating CO while the remaining CO from the initial CO2 molecule will escape as a gas (Eqn 2.1). In the case of steam (Eqn 2.2), the H2 generated can anneal to the aromatic ring. The annealing process is slower than the formation of the CO bond, yet the former is dominant, thus resulting in a greater energetic r eaction and the creation of a larger pore. If the H2 does not anneal carbon active si tes (i.e., those C atoms that are not fully saturated), the H2 diffuses out of the carbon and can actually inhi bit the forward progress of gasification (Ergun and Mentser 1965). The impact of oxygen functional groups have been shown to effect pore development. Molina-Sabio et al. (1996) found that the evolution of oxygen functional groups as CO from the carbon surf ace was correlated with the CO2 micropores and the widening of pores from steam. It should be noted that the presence of oxygen functional

PAGE 14

4 groups could also affect the adsorption of targ et contaminants such as phenol, which will be discussed in follo wing sections. Adsorption Factors Surface Area Surface area is an important parameter to consider when creating or selecting an adsorbent. In an ideal ad sorption situation—where all ot her conditions (such as pore size, surface chemistry, and adsorbent-ad sorbate interactions) are optimal for contaminant removal—the surface area would serve as the limiting factor for the adsorption process. In this case, as the act ivated carbon surface ar ea increased, so would the adsorption of the target contaminant. Manufacturers of activat ed carbon attempt to increase the surface area of the adsorbent with the hope of enhancing the carbon’s removal efficiency. Typically, the surface area of activated carbon reaches a maximum around 1500 m2/g. The activated carbon surface area is comm only found using a theory developed by Brunauer, Emmett, and Teller (BET) for physical adsorption. While the BET theory is inadequate as a universal equation for physical adsorption, it has been adapted to describe surface area (Dabrowski 2001). BET surface area is determined by flowing nitrogen gas at 77 K through a sample of activated carbon, allowing the N2 to enter the pores of the carbon. From the amount of N2 that adsorbs to the pores, th e surface area is deduced. It should be noted that the BET surface area can at times be misleadingly when attempting to directly correlate it to the adsorption capacity of an activated carbon (Pope 2003). Since nitrogen gas molecules are much smaller than many target contaminants, the size of the contaminant itself should be considered wi th respect to the pore size distribution of the carbon.

PAGE 15

5 Pore Size Distribution Another essential parameter to consider for an absorbent is its pore size distribution (PSD). Pore size distribution is usually e xpressed as a graphical relationship, using pore width () as the independent variable and cumulative pore volume (cc/g) as the dependent variable. Pore widths that fall under 20 are considered to be micropores, from 20 to 500 are mesopores, and above 500 are macropores. The variation of pore widths in an activated carbon is dependent upon the activation process. Theoretically, any precursor can have any de sired PSD, yet, the degree of distribution may require more manipulation of the activation environment, and hence may be more energy (and cost) intensive. pH When the parameter of pH is discussed in an experiment, commonly it is used as a descriptor to express the ioni c conditions of a given aqueous system. However, in the case of activated carbon adsorpti on, it is important to take into account th e pH of the carbon surface and internal pores in additi on to the aqueous medium. An internal measurement of the activated ca rbon is expressed through the poi nt of zero charge (PZC). The PZC is the point at which the carbon surf ace has no detectable charge. In Figure 2.1, the PZC is at pH 7; therefore at a soluti on pH below 7, the carbon surface is positively charged and a solution pH above 7 will promote a negatively charged carbon surface. PZC is an important characteristic when predicting or describing the process of adsorption, yet is only a dominant factor wh en the target contamin ant is close to the adsorption site. For example (assuming the co ntaminant is in range of the adsorption site), if the target contaminant was a cation, a solution pH greater than 7 would be desired

PAGE 16

6 for the PZC illustrated in Figure 2.1. Conversel y, if the target contaminant was an anion, a solution pH below 7 would be desired. Figure 2.1. Representation of Point of Zero Charge The zeta potential, a measurem ent of external charge, gove rns the attraction of the target contaminant to the activated carbon. Ze ta potential is the measurement of electric potential at the shearing plane—the “space” between the activated carbon surface and the adjacent water molecules (Adamson 1990). Surface Chemistry The structure of activated carbon is graphi tic in nature, consisting of molecular layers of carbon which can be viewed, acco rding to Coughlin and Ezra, much like a polynuclear aromatic molecule (Coughlin and Ezra 1968). These layers contain carbon atoms that are bonded together with three sigma bonds and one pi bond having sp2 hybridization. It is also possible for sp3 hybridization (tetrahedron) to occur, which may result in cross-linking among the graphite la yers (Coughlin and Ezra 1968). The carbon within this structure is microcrystalline and is held together with the graphite layers through van der Waals forces. When other at oms are bound within this system, they can be present within the layers, forming “het erocyclic” rings, or at the edges of the (+) (-) g e ch mmol arg mmol charge pH PZC curve At zero charge pH = 7

PAGE 17

7 microcrystalline carbon molecules, thus fo rming functional groups (Coughlin and Ezra 1968). Edge sites located between the graphitic layers are very reac tive and are therefore prominent sites for functional groups and adsorp tion. It is asserted by Coughlin and Ezra (1968), that the basal face of the be nzene ring can weakly adsorb through interactions. A discussion of activated carbon surface chemistry should also include a thorough examination of electron interactions, includi ng electron density, elec trostatics, attraction and repulsion, as well as dispersive forces and the influence of f unctional groups located on both the adsorbent and the adsorbate. Nume rous discussions in the literature center around speculation of these inte ractions and many of these theories have not yet been sufficiently disproven. It is hoped that a deeper understand ing of the activated carbon surface chemistry will provide the keys for unlocking the mechanisms of the adsorption process. Although questions remain, the lite rature gives significant insights into the surface chemistry of activated carbon; these th eories will be examined in the following sections of this paper. Predicted Adsorption Mechanisms Despite the extensive amount of research th at has advanced the field of activated carbon, the adsorption mechanisms still rema in unknown. Currently, the literature has concentrated on two mechanistic theories—the theory of bonding first proposed by Coughlin and Ezra (1968) a nd the electron donor theory pr oposed by Mattson et al. (1969). Bonding Theory The bonding theory takes in to consideration the bonds which occur between two p-orbitals as shown in Figure 2.2. This type of bonding occurs between the 2p

PAGE 18

8 orbitals in the benzene ring presented in Fi gure 2.3. The electrons in the 2p orbitals travel above and below the benzene basal pl ane, therefore creating a delocalization of electrons about the ring. At the basal plane, weak adsorption can occur through interactions, while at edge si tes adsorption is much stronge r (Coughlin and Ezra 1968). Figure 2.2. Representation of sigma and pi bonding, arrows with in the atoms of formaldehyde represent elect ron filling in outer or bitals. Adapted from Petrucci and Harwood (1997). Figure 2.3. Representation of the Benzene Pi System (Source: Teda nkara Library, 2003) Species that are bonded to the graphitic edges of activated carbon can cause a disturbance in the electron density of the basa l plane. For example, the presence of O2 molecules (i.e., electron-withdr awing groups) will influence the electron distribution by removing electrons and creating positive holes in the conduction band of the electron system, thus decreasing the dispersive ad sorption potential of the carbon surface (Coughlin and Ezra 1968, Franz et al. 2000, Radovic 2001). The opposite is also true;

PAGE 19

9 removing electron-withdrawing groups fr om the carbon surface will increase the electron density and therefore re sult in an increased adsorption potential (Radovic 2001). Electron Donor Theory The electron donor theory suggests that an exchange of electrons takes place during the adsorption process, whereas in bonding dispersive interactions are dominant. Mattson et al. (1969) argue that Coughlin and Ezra do not explain the effects of interactions across the distance of the basal plane as well as the mechanism of interaction of the sorbates to the basal plane. Using the nitrogroup as an example, Mattson et al. (1969) suggest that as electronwithdrawing groups reduce the electron density, the nitro-substituted aromatic would become an acceptor that would interact with a donor such as carbonyl oxygen. In a broader se nse, the carbonyl oxygen would be the major donor and the aromatic ring would be the acc eptor. Taking into consideration the system interactions, Mattson et al. (1969) stat e that the adsorption or ientations should be planar. A notable paper by Franz et al. (2000) disa grees with the findings of Mattson et al. (1969). Mattson et al. (1969) base the electron donor theory on their observation that as surface oxygen increases so does the amount of carbonyl functional groups. As the carbonyl group serves as the major electron donor in the donor-acceptor complex, Mattson et al. credit it for the increase in phenol adsorption. Matts on et al. attribute subsequent decreases in adsorption to th e formation of carboxyl functional groups. However, Franz et al. (2000) have found th at an increase in oxygen functional groups leads to an increase in the formation of both carbonyl and carboxyl groups. Radovic (2001) also refutes the findings of Mattson et al., stating that “carbon oxidation does not

PAGE 20

10 (necessarily) involve the conversion of car bonyl groups into carboxyl groups,” thus suggesting that the reasoning of Mattson et al. is flawed. Fran z et al. (2000) suggest that the decrease in adsorption is due to the pr esence of water molecules that create bulky clusters on the carboxyl groups due to hydrogen bonding and block access to the adsorption sites. To support this conclusion, Fr anz et al. (2000) argue that if moisture is enough to block N2 from adsorbing during BET surface analysis, that a compound much larger than N2 should also be expected to be blocke d. In agreement with the speculations of Coughlin and Ezra (1968), Franz et al. (2000) support an adsorp tion theory based on dispersive/repulsive interactions. Functional Group Formation Detection of Functional Groups Throughout the literature, th e most popular form of f unctional group identification is through the use of Boehm titrations (pre sently cited 172 times in the literature according to the Science Citation Index). Functional group identification is made through the applications of organic acid-b ase chemistry. These acid-base reactions examine the chemisorbed oxygen on the carbon surface, defining four acidic surface groups (Boehm 1966 and C oughlin et al. 1968): (I) strongly acidic carboxyl group (II) weakly acidic carboxyl group (III) phenolic hydroxyl group (IV) carbonyl group. Titrations are performe d to identify which groups are present through a neutralization process using various bases. Group (I) can be neutralized by NaHCO3, NaCO3, NaOH, and NaOC2H5. Group (II) can be neutralized by Na2CO3 or stronger

PAGE 21

11 bases. Group (III) functional groups are titr ated with NaOH and group (IV) reacts with sodium ethoxide (NaOC2H5) (Boehm 1966). In addition to Boehm titrations, other me thods have been presented (e.g., X-ray detection, and Fourier Transform Infrared Sp ectroscopy (FTIR)), ye t the success of many of these methods has not been universally accepted for the accurate description of the carbon surface—especially within the pore walls. Types of Functional Groups Formed From CO2 activation The formation of carbonyl and carboxyl groups from CO2 activation has been reported in the literature (Mattson and Mark 1969, Johns et al. 1999, Guo and Lua 2002). Through the use of infrared internal reflect ance spectroscopy, Mattson et al. (1970) have determined the presence of “significant amounts” of carboxyl and carbonyl groups which they speculate have a strong interaction at the edges of the aromatic basal planes. Although no rationale is offere d, Mattson et al. (1970) have speculated that at some temperatures (300C, 400C, 600C, 700C) the use of CO2 during activation can suppress the formation of oxide surface func tional groups. In addition to carboxyl and carbonyl groups, Johns et al. ( 1999) have determined that CO2 activation produces more acidic functional groups than st eam activation; specifically, th e presence of phenolic and lactone groups was detected. Lua an d Guo (Lua and Guo 2000, Guo and Lua 2002) found that CO2 activation of oil palm shell produced a slightly acidic surface consisting of quinones, aromatic rings, and oxygen atoms speculated to be bound at the edges of the graphitic sheets.

PAGE 22

12 From H2O activation During the process of steam activation, in a ddition to oxidizing th e carbon surface, steam prevents secondary char formation on the carbon by removing volatiles from the surface and can interact with both organic a nd inorganic matter (Petrov et al. 1994). Through these interactions with the carbon surf ace, various functional groups are formed. The formation of hydroxyl (e.g., phenol and l actone), carboxyl, and carbonyl groups have been reported in the literatur e (Gergova et al. 1993, Petrov et al. 1994, Johns et al. 1999). While both hydroxyl and carboxyl groups we re detected after steam and CO2 activation, Johns et al. (1999) reports a decreased pres ence of both following steam activation. Factors controlling which groups are formed The functionality of a given activated carbon is highly dependant on its carbonaceous precursor and its heat treatment. Various raw materials will form varying functional groups when activated depending upon their structure and composition. For example, a precursor containing sulfur would be expected to exhi bit sulfur containing functional groups, while a material void of sulfur would not form such groups unless treated with a thiochemical such as H2S. Likewise, the intensity of heat treatment can control the degree at which surface groups are burned off, or formulated. In the previous example, if the activation temp erature was intense enough to burn off the sulfur as a gas, it would be expected that a sulfur functiona l group would not form on the carbon surface. It therefore follows that different types of activated carbon will adsorb contaminants differently—hence, the efficiency is variable an d can be tailored in the activation process. Similarly, the same activated carbon will ha ve unique adsorption relationships with varying contaminants—i.e., the efficiency will vary according to the chemical and physical characteristics of the adsorbate.

PAGE 23

13 Precursor and conditions of activation Upon examining the significance of the ac tivation process on the surface chemistry of an activated carbon, it is insightful to examine the functionality of the carbon both before and after the oxidation process. In the literature, two such papers were found. Gergova et al. (1993) have examined the su rface of apricot stones prior to and following steam activation. The results of this study are summarized in Table 2.1. From the data, it is clear that the composition of carbon, nitrogen, and sulfur increased while hydrogen and oxygen decreased. While both nitrogen and su lfur also show an increase in percent weight, the difference is less than 1% (alm ost less than 0.5%) and would be difficult to argue any significant difference in comparison with the other elements. The most notable changes are between carbon and oxygen; the we ight percentage of carbon increased by nearly 40% while oxygen decrea sed approximately 35%. It should be noted that the composition of oxygen was found by difference; hence, it was assumed that the material consisted only of carbon, hydr ogen, nitrogen, sulfur, and oxygen. While the authors mention the detection of carbonyl and hydroxyl groups, they fail to draw any conclusions about the change in elementa l concentration. Furthermore, the authors do not state the process by which the complete elemental analysis was performed, making it nearly impossible for the reader to derive a conclu sion concerning the differe nce in composition. Guo and Lua (2002) reported the change in functional groups dete cted on oil shell palm prior to and after CO2 activation. The raw materi al oxygen functional groups consisted mainly of carbonyl groups (e.g., ke tones and quinones), ethers, and phenols. After carbonization, the presence of ketone s and quinones were detected, and following activation, only the presence of quinones remain ed. The authors attr ibute the loss of the ketone functional groups to the intense heating (950 C) (Lua and Gou 2000, Guo and Lua

PAGE 24

14 2002). The effects of temperature on the formation of functional groups during the activation process have also been confirmed by Mattson et al. (1969). For the case of the oil shell palm, the result of the carbon surface ch anges resulted in a s lightly acidic carbon surface (Lua and Gou 2000, Guo and Lua 2002). Table 2.1. Elemental Analysis of Apricot St ones Before and After Steam Activation at 600 C, Adapted from Gergova et al. 1993 Elemental Analysis Raw Material (% weight) Activated Carbon (% weight) C 51.45 89.67 H 6.34 2.40 N 0.20 0.58 S 0.09 0.39 O ( by difference ) 41.92 6.96 In addition to the effects of temperatur e, the presence of impurities in the carbonaceous precursor can lead to notable e ffects in the formation of surface functional groups. Mattson and Mark (1969) state that the presence of oxygen, nitrogen, sulfur, hydrogen, and ash are common impurities and that specifically the presence of oxygen and hydrogen (especially when bonded with oxyge n) can have distinct effects on the adsorption process. Coughlin et al. ( 1968) found that the presence of chemisorbed oxygen decreased the adsorption capacity for phenol while Vi dic et al. (1993) have shown the opposite effect. It should be noted that Coughlin et al. ob served the adsorption of phenol in a dilute soluti on—it can be argued that the d ilute solution accounted for the lack of adsorption and not th e oxygen functional groups. In th e case of a dilute solution, the driving force for adsorption is very lo w, thus decreasing the probability for the adsorbate and adsorbent to interact and adsorption to occur.

PAGE 25

15 Optimizing the Use of Activated Carbon Commonly, to estimate the adsorption capacity of an activated carbon, Freundlich or Langmuir isotherms are developed. Ho wever, according to Weber and DiGiano (1996) : “Despite the sound theoretical basi s of the Langmuir, BET, and Gibbs models, these isotherms often fail to describe experime ntal solution sorption da ta adequately.” It is proposed that the Freundlich equation can better describe the activated carbon adsorption process than the afore mentioned m odels due to the heterogeneous nature of the carbon surface. It will be shown in the following sections how the Freundlich equation compares to the frequently us ed Langmuir model for activated carbon adsorption of benzene and monosubstituted benzenes. Developed by Herbert (Heinrich) Freundlich from empirical observations in the early 1900s, the Freundlich model offers an exponential equation, based on heter ogeneous surface energy distribution, which describes the variation in adsorption heat w ith respect to adsorbate concentration (Weber and DiGiano 1996): n eKC q1 Eqn 2.3 Where: qe = the mass of contaminant ad sorbed per unit mass of carbon K = “adsorption capacity”, Freundlich constant C = concentration of the adsorbat e in solution at equilibrium 1/n = strength of adsorption, Freundlich constant To validate his empirically derived equation, Freundlich utilized a special case of the Gibb’s relationship, working under the assumption that in a dilute so lution Gibb’s surface excess is equivalent to the amount of contam inant adsorbed (Weber and DiGiano 1996).

PAGE 26

16 The Freundlich equation can be used in designing an activated carbon system. For example, using the Freundlich constants K and 1/n, the following equations can be used to estimate the bed life of an activated carbon (Snoeyink 1990): GAC l o eC C q Y) ( Eqn 2.4 Where: Y = carbon bed life Co = initial concentration of contaminant Cl = final concentration of contaminant GAC = apparent density of carbon The Freundlich constants K and 1/n are comm only sited in the li terature without a thorough discussion of what each represents. It is one intent of this paper to attempt to clarify the significance (and limitations) of th ese parameters. The so called “adsorbent capacity” constant K represents the influen ce of various adsorpti on energies that are associated with the heat of adsorption. However the term “adsorbent capacity” only applies when K = qe, hence when C=1 and K is depe ndent solely on the change in adsorbate concentration due to adsorption (Chiou 2002). The term 1/n represents the sum of diverse energies associated with adso rption. The notation of 1/n comes from the thermodynamic derivation of the Freundlich eq uation, but, given that the equation is typically used empirically, th e literature often uses simply “n” (Weber and DiGiano 1996). Certainly “adsorbent capac ity” (K) and strength of adsorption (1/n) should vary with varying activated carbons. Noting that no two activated carbons are identical, it should be easily deduced that the performan ce of varying activated carbons would also

PAGE 27

17 differ. These differences could account for the discrepancies reported for the Freundlich constants in the literatu re. It should further be noted th at the concentration range of the activated carbon used in the development of isotherms may have a significant effect on the value of K and 1/n. It is not uncommon to find K values that differ by several orders of magnitude in the literature ; for example, Table 2.2 shows three different values of K for benzene. The variation in reported valu es is likely due to various definitions of equilibrium conditions, the use of different ca rbons, and the use of varying concentration ranges for both the adsorbent and adsorbate. Ho wever, if these K values are to be used as tools in designing activated carbon systems, these differences must be accounted for and understood. Table 2.2. Reported Values of Freundlich K for Benzene. K Source 0.12602 Noll 1999 1.0 Dobbs and Cohen 1980 41.7 Loll et al. 2004 An understanding of the Polanyi-Manes m odel (Eqn 2.5) reveals adsorption trends which are reflected in the Freundlich consta nts. Using the Polanyi-Manes model, the energy necessary for an adsorb ate to displace a solvent (in order to be adsorbed on the adsorbent) can be calculated (Chiou 2002). For example, for similar solutions under constant conditions of temperat ure and pressure, more energy is required to adsorb more adsorbate due to the increased volume of solvent that must be displaced. e s slC C RT ln Eqn 2.5 Where: sl = adsorption potential (energy re quired for a volume of solute (s) to displace a volume of solvent (l) in the adsorption process)

PAGE 28

18 R = gas constant T = temperature Cs= solute concentration Ce= equilibrium concentration of adsorbate in solution Applying the same basic principles to th e Freundlich constants, as the adsorption potential increases, the value of K should also increase, as K is a representation of the energies required for adsorption. Likewise, as the adsorbent concen tration range varies, so will the concentration of adsorbate in the solution; thus leading to variation in the adsorption potential and hence a change in K. It should then follow that as the adsorbent range is varied in isotherm experiments, the resulting value of K will differ, potentially across several magnitudes. Developed originally for air phase adso rption, the Langmuir equation (Eqn 2.6) has been widely adapted for aqueous systems by re placing the original pressure term with a term representing the solute equilibrium concentration. e a a eC bQ Q q 1 1 1 Eqn 2.6 Where: qe = the mass of contaminant ad sorbed per unit mass of carbon Qa = maximum adsorption capacity b = net enthalpy of adsorption Ce = equilibrium concentration of adsorbate in solution The Langmuir equation draws its theory from the condensation and evaporation of gas molecules on a solid adsorbent surface (Web er and DiGiano 1996). Condensation takes into account available adsorp tion sites on the adsorbent surf ace as well as the rates at

PAGE 29

19 which the adsorbates contact the adsorben t surface. Three basic assumptions are encompassed within the Langmuir equation: (1 ) adsorption energy is constant throughout the activated carbon surface and is independent of surface coverage (2) there are no interactions between adsorbate molecules or migration of adsorbates to other adsorption sites (3) the carbon surface w ill only support a monolayer adsorption of the adsorbate (Weber and DiGiano 1996). Two constants ar e typically reported from the Langmuir model, Qa and b Qa represents a condition where the activated carbon surface is covered by a monolayer—i.e. when the surface has reached its capacity according to the assumptions embedded within the Langmuir model. The constant b represents the net enthalpy of adsorption and is descri bed by a ratio of rate constants (kadsorption/kdesorption) on a mole (or mass) unit basis (Chiou 2002, Weber and DiGiano 1996). Weber and DiGiano (1996) have shown that the Freundlich equation can be represented using the general te rms of the Langmuir equation: n e g g a eC b Q q ) (, Eqn. 2.7 The subscript g denotes a generalized form of the parameters where Qa,g = Qa, while bg represents the Langmuir constant b and also accounts for site energy. The constant n represents the Freundlich 1/n and accounts fo r the heterogeneity of the surface site energies. With this relationship in mind, the Freundlich K becomes a function of Qa,g and bg (Weber and DiGiano 1996): n g g ab Q K, Eqn. 2.8 The use of isotherms for estimating ca rbon adsorption is a typical textbook approach that is still widely used by both carbon manufacturer s and its users. When used properly, isotherm constants are valuable tools for understanding the adsorption process

PAGE 30

20 of a given system. Isotherm constants rela te the equilibrium condi tions of a carbon with an adsorbate and indicate both the capacity of the adsorbent for that adsorbate and the strength at which the adsorb ate is adsorbed. When isotherm constants are known for a given carbon and target contaminant, the rem oval of that adsorbate can be predicted and optimized. While empirically determined isotherm constants are useful, the development of these constants are very time consuming a nd their applications are limited to the conditions under which they were derived. Ultimately, to accura tely represent the adsorption process for a given system, the carbo n should be tested within that system. Use of a direct testing method will render a more accurate solution if conditions permit its application; such a method is carbon prof iling. Carbon profiling is a valuable concept that has yet to be implemented widely in the water treatment field. By using carbon profiling, the effectiven ess of an activated carbon can be observed without relying on an adsorption isotherm. Adsorption isotherm s are always performed at equilibrium conditions, accounting for the e ffects of adsorption and desorp tion which are inherent in the isotherm equations. Equilibrium conditi ons are typically not seen on a full scale level—such as in a water treatment facil ity—and therefore caution should be used when estimating full-scale treatment parameters. It should be noted that the role of kinetics with respect to PAC and GAC are different. For example, in a water treatment facility, PAC would typically be added at the head of the plant ( during rapid mix) so that sufficient contact time is allotte d for the contaminants to inte ract with the carbon. Since PAC is smaller than GAC, more time is required for the contaminants to come into initial contact with the carbon. However, because of the greater internal volume of GAC, an

PAGE 31

21 adsorbate may take much longer to reach e quilibrium within the carbon matrix than within PAC. The following steps comprise carbon profiling: 1. Obtain a sample of the water to be treated 2. Perform dose removal studies using vari ous amounts of the activated carbon, mimicking the plantÂ’s processes: contact times, chlorination, pretreatments, etc. 3. Generate dose removal curves 4. Determine the most effective carbon dose for removing the targeted contaminant, accounting for treatment objective and economics While carbon profiling circumvents a limita tion of adsorption isotherms, e.g., equilibrium conditions, this system still has its limitations. Much like the adsorption isotherms, carbon profiling is not performed at full-scale and is time consuming, as specific conditions must be met and optimized through a series of dos e removal studies. The traditional method of using previously existing isotherm constants to develop removal schemes is outdated and inefficient. Parameters rendered from the traditional methods likely do not emulate full-scale pro cesses since isotherms are developed under equilibrium conditions and ofte n the conditions under which the constants were derived are not addressed. Adsorption is dependent on the type of act ivated carbon used; yet, the isotherm parameters that are typically referenced from the literature are developed with unstated types of carbon (often using nanopure water) and therefore should not be relied upon to determine the adsorption properties of another system. The constants therefore will not reflect the interactions of natural organic matter or any other constituents that may be present in the contaminated water. It should be noted that while adsorption isotherms rely on equilibrium conditions for appropriate application of the isotherm equations, they certainly can be conducted w ith respect to other aspects of the intended treatment system. In this case, a similar procedure as carbon profiling can be followed

PAGE 32

22 incorporating the treated water and chemicals pr esent in the large scale process, with the condition that equilibrium exists. The d ecision of using carbon profiling or adsorption isotherms should be carefully c onsidered with respect to th e treatment system for which its application is intended. The development of models that incorporat e adsorption system descriptors such as kinetic rates and adsorption isotherm constant s are promising tools that can incorporate the effects of unknown mechanisms into predictive adsorption analysis. By applying an empirically based model to an appropriate sy stem, the mechanisms within the adsorption process can be described and accounted for without being fully known. Such a tool, Quantitative Structure Activity Relationships (QSAR), is becoming widely known in the medical and biological fields, but has yet to see a dominating presence in adsorption study. Blum et al. (1994) have develope d a QSAR for activated carbon adsorption in water using 363 organic compounds. While this model is noteworthy, it is based on literature-derived constants—which may vary drastically with respect to experimental procedure and analysis—without a sturdy f oundation, these models will not accurately portray the adsorption process. Blum et al. show an understanding of this vital point as the authors mention that “…in applying QSAR ba sed on literature data, it is imperative to consider the relevance of the data source to one’s own investigative situation.” However, when used in conjunction with accurately de scriptive parameters (i.e., case specific), QSAR can be a powerful tool for understandi ng activated carbon adsorption. The goal of this study was to investigate the use of adso rption isotherm consta nts in QSAR and to provide a protocol by which QSAR and adsorp tion isotherm constant s can be effectively used to describe the adsorpti on process for a given system.

PAGE 33

23 Quantitative Structure Activity Relationships While adsorption isotherms and carbon profiling are useful in developing adsorbate-adsorbent relationships, they are ti me intensive and non-specific in regards to adsorbate-adsorbent structural interactions. Relying solely on isot herm relationships and profiling would require extensive laboratory analysis for each individual compound (and condition) that is to be analyzed. The adso rption of a chemical compound to activated carbon is dependent on both the chemical and physical properties of the carbon and the compound. Therefore, by knowing the chemi cal and physical properties, the adsorbateadsorbent relationship can theore tically be predicted. Usi ng known parameters to predict the activity of a compound is th e focus of QSAR modeling. Introduced into the field of biology just ove r four decades ago, QSARs have gained prominence in the both the biological and medical fields (Hansch and Fujita 1995). While QSARs have proven to be powerful tools for the development of numerous chemicals and medications, the adsorption fiel d has yet to develop a robust QSAR for adsorption prediction. Within an adsorpti on system, a QSAR focu ses on predicting the molecular interactions between the adsorbate and adsorbent (Brasquet et al. 1997). For example, in an adsorption system, the inde pendent variable input into the QSAR would be the target adsorbate’s intrinsic parame ters (e.g., chemical, electronic, steric, hydrophobic properties) while adsorption paramete rs (e.g., isotherm constants, kinetic rates, etc.) would be the dependent vari ables. A modeling program based on QSAR interactions is theoretically able to predic t the behavior of a given contaminant in the presence of an activated carbon—thus given th e independent variable (or variables) the QSAR outputs a calculated dependent variab le describing the adsorption system, for instance, an adsorption capacity.

PAGE 34

24 The driving force behind a QSAR is its ab ility to formulate a relationship between two compounds through the use of tools such as regression analysis or pattern recognition (Lloyd 2002). Regression analysis is commonly based on a partial least squares projection, which is a multivariate statistical method; while pattern recognition is usually performed by artificial ne ural networks (ANNs). ANNs mathematically mimic the pattern recognition properties within the hum an brain, using para llel processing and weighted connections, which store the pr oblem solving knowledge. These weighted connections are trained to sp ecific situations through repe ated exposure and comparison to truth data sets (PNNL 1997). The key a dvantage of the QSAR model is that a known equation or the exact pathway of reactions is not necessary. A good QSAR model can rely solely on the structural and derived e xperimental characteristics of the targeted compounds to predict their activity. Add itionally, unknown mechanisms could ultimately be disclosed through the analysis of how substituents impact QSAR correlations. Within the development of a QSAR, diverse parameterization is essential. Having a robust set of carefully chosen parameters increases the probability of unlocking a description of the mechanisms involved in a system. Three key components of a QSAR are: hydrophobic, electro nic, and steric factors (Hansch and Fujita 1995). Table 2.3 lists various hydrophobic, electronic, steric and ch emical properties th at are commonly used in QSAR analysis. While diverse parameterization is essential to a QSAR, the selection of these parameters is crucial for the healthy devel opment of the QSAR. A primary weakness in QSAR construction is often the selection of the parameters (Hansch and Fujita 1995). Poor parameter selection can lead to collinearity problems (i.e., when two parameters are

PAGE 35

25 directly affected by each othe r) thus creating false indications of correlation. Also, it is essential that parameters cover a wide ra nge of space—i.e., the compound training set should be carefully selected so that multip le spectrums of the parameter scale are represented. This idea will be more fully developed in the Met hods section with the introduction of Craig plots. Table 2.3. Example of Parameters Commonly Used in QSAR Development. Hydrophobic Properties LOG KOW Electronic Properties Ind, Resonance, Field Effects Steric Properties Molar Refractivity, Molal Refraction, Es Chemical/Physical Properties Boiling Point, Melting Point, Density, Molecular Weight, Aqueous Solubility, Enthalpy, Vapor Pressure Where Ind represents inductive effects descri bed by the Hammett Constant and Es represents Taft’s steric factor. An important aspect of QSAR developm ent that should be addressed is the selection of the training set and the speci fic parameters that are chosen for the correlations. There are two basic approaches to QSAR that are seen in the literature, the first approach, developed by Hansch, is an a pproach favored by chemists in which the components within the QSAR ar e intellectually analyzed with respect to their chemical activity (Hansch and Fujita 1995). The second approach, described by Wold and Dunn (1983), is based on the statistic al analysis of the components for QSAR development. Certainly, while the statistical method can pr ove to be insightful in some instances, a strong QSAR should be constructed with a va st understanding of the components in the system in order to avoid complications such as parameter collineari ty, uneven spread of parameters, and the application of implausibl e mechanisms. In addition, it is also

PAGE 36

26 important to consider the functionality of the compounds that are w ithin the training set; only structural changes in functional groups that can be accurately represented by available parameters should be selected (Hansch and Fujita 1995). This present work approaches QSAR de velopment for the adsorption process in three improved ways in comparison to past studies: 1. Dependent variables used are not from the literature, but were derived experimentally within this study 2. Degree of statistical analyses performed 3. Robust spread of parameterization As was shown in Table 2.2, empirically deri ved isotherm constants selected haphazardly from the literature can be detrimental to a study. The electronic, steric, hydrophobic, and chemical parameters used within this study were not experimentally derived within the scope of this work, yet each of these parameters has been well documented within the scientific community and has strict procedures for their procurement. However, in the case of deriving the isotherm constants, many variables are present: the type of carbon used, how it was activated, the conditions under which the carbon was kept, how it was added to the system, the pH of the water in the system, the solutes used within the system, the definition of equilibrium for that system, the concentrations of carbon and solute that were used, etc. In order to normalize these variables, each of the isotherm constants used within this system was deri ved under uniform experi mental conditions. In order to measure the quality of the QSAR correlations, several statistical methods were used: R2, adjusted R2, standard error, F-Ratio, as well as Q2 (a tool that measures the predictability of a correlation). This combination of statistical analyses is more diverse than other adsorption QSAR that are listed in the lite rature and therefore

PAGE 37

27 provides a more rigorous analysis of the corre lations. For example, the study of Brasquet and coworkers (1997) refers to R2 values, but offers no othe r statistical validation for their resulting correlation. As was mentioned previously, the spread of parameters is an important concept in the selection of the training set. For this study, compounds demons trating a wide range of steric, electronic, hydrophobic, and chemical properties were selected. This variety of parameterization is essential to aid in the identification of mechan isms involved in the correlations.

PAGE 38

28 CHAPTER 3 MATERIALS AND METHODS Isotherm Methods Adsorption isotherms were develope d by adding a known concentration of contaminant, 100g/L, to distilled deionized (DDI) water at a pH of 5.5 1.5, the variation is due to the use of unaltered DDI water. Th e desired amount of powdered carbon was weighed and dried overnight at 105C. After cooling the carbon in a dessicator, a carbon stock slu rry (10,000-mg/L) was prepared by combining the carbon and DDI water by mixing on a stir plate. The sl urry was stored in a dessicator at room temperature and mixed on a stir plate before each application. Isotherms were created for each target contaminant using the following conc entrations of carbon: 1, 3, 5, 10, 15, 25, 50, and 75-ppm. The desired amount of car bon slurry was added to a 100-mL or 50-mL SGE gas tight Luer Lock syringe containing DDI water. The experimental data was collected for each contaminant individually using powdered Calgon F400 activated carbon with 100g/L solutions of either benzene or a chosen monosubstituted benzene (shown in Table 3.1). For each run, the desi red amount of contaminant was added from a stock solution to the 100-mL syringe, yielding a final volume of 100-mL. The syringe was then mixed end over end on a rotato r for 2 hours to 6 hours, depending on equilibrium conditions for that target c ontaminant. Equilibrium conditions were determined through the development and anal ysis of dose removal curves over various times. The condition of equilibrium was considered achieved when the rates of adsorption and desorption appeared to be at a steady state. After mixing, the samples

PAGE 39

29 were filtered into VOC vials using 25-mm Fi sherbrand nylon syringe filters which were sized to allow for only the PAC to be rem oved (both 0.20-m and 0.45m were used, depending on their availability). Samples we re sealed with Teflon septa caps and stored at 4C until analysis. All samples were anal yzed with a Saturn 2100T GC/MS using a Supelco VOCOL™ Fused Silica capillar y column with a Supelco SPME 75m CARBOXEN™ PDMS fiber, uti lizing an oven temperature program that holds at 40C for 2 minutes and then climbs to 210C at a rate of 8C/min. Each isotherm was replicated. Only data which produced a RSD of 20% or less were accepted. Table 3.1. Target Contaminants Us ed in Isotherm and Kinetic Studies. Target contaminants Benzene Isopropylbenzene n-butylbenzene t-butylbenzene Isobutylbenzene Nitrobenzene Aniline Benzaldehyde Fluorobenzene Chlorobenzene Bromobenzene Iodobenzene Phenol Development of Isotherm Data Target Contaminants Benzene Benzene is a six carbon structure bonded in a cyclic formation with three resonating double bonds. The benzene ring is one of the defini ng characteristic of

PAGE 40

30 aromatic compounds. Due to its cyclic structure, and desp ite its unsaturated state, benzene is a stable compound that will under go primarily substitution reactions, such as H2 annealing (Ouellette 1998). H2 annealing is a critical step during steam activation which modifies the structure of the benzene rings in the gr aphitic carbon sheets, creating the porosity of the carbon. Th e electron orientation of ben zene drives its unique reaction mechanisms. As was shown in Figure 2.2, the benzene ring experi ences a delocalization of electrons, where the carbon atoms have three sigma bonds—two bonding with the adjacent carbon atoms and the third is bonde d with a hydrogen atom. The remaining electron gives the delocalizati on characteristic as it orbits above and below the basal plane in a 2p orbital (Ouellette 1998). For th is study, benzene has b een selected as the parent compound for the QSAR training set. Substituted benzenes Phenol is an aromatic compound consis ting of a hydroxyl functional group bonded via a sp2 hybridization to a carbon atom in the benzene ring. Adding a functional group to the benzene ring can result in activating or deactivating the reactivity of the ring. Because of the addition of the hydroxyl group, phenol is a strongly activating compound—i.e, it increases the reactivity of th e benzene ring. In contrast, a benzene ring substituted with a chloroor nitrogroup causes the benzen e ring to be less reactive. Activating groups are more reactive and dona te electrons, therefore increasing the electron density of the benzene ring. Deactiv ating groups are less dense (therefore less reactive) due to their tenden cy to attract—or withdraw el ectrons (Ouellette 1998). For example, in the reaction of an activating (electron donating) group with an arene (aromatic hydrocarbon), the electrons in the 2p orbital are attracted towards the arene, thus creating an increase in electron density at the benzene ring, which in turn will make

PAGE 41

31 the ring more reactive (Carey 2003). In the case of phenol, which acts as a weak acid, the presence of an electron-withdrawing group on the carbon surface will cause a decrease in electron density at the benzene ring and result in a more stable, more acidic, and less negative compound. Contrastin gly, in the presence of an electron-donating group, the phenol will become less acidic (Ouellette 1998) When a phenol is oxidized, it becomes a quinone—a carbon ring having two ketone (c arbonyl carbon atom bonded to two other carbons) groups (Ouellette 1998). In the cas e of activated carbon adsorption, the functional groups present on the carbon surf ace will contribute to the nature of the reaction between the surface and the adsorbate. Freundlich Isotherms Freundlich isotherms were developed usi ng the linearized Freundlich equation: f eC n K q log 1 log log Eqn 5 Where: qe = the mass of contaminant ad sorbed per unit mass of carbon K = adsorption capacity, Freundlich constant n = strength of adsorptio n, Freundlich constant Cf = final concentration of contaminant The term qe was determined by taking the difference in the initial and final concentrations of the target contaminant in each isotherm run and dividing it by the mass of activated carbon added to the system. All target concentra tion changes were attributed to adsorption to the activated carbon. The log of qe was graphed verses log Cf to develop the Freundlich isotherm plot. A linear tre ndline was added to the plot using Microsoft

PAGE 42

32 Excel from which the slope and y-intercept we re determined. The slope represents the value for 1/n and the y-intercept repr esents the value taken as K. Langmuir Isotherms The Langmuir isotherm parameters were found using a linearized form of the Langmuir equation as was shown in Eqn 4. A graphic was created by plotting eq 1 verses eC 1 Using Microsoft Excel, a linear trendlin e was added from which the slope and yintercept were taken. The slope represents the value abQ 1 and the y-intercept is taken as aQ 1 The values for aQ and bwhere then determined. Kinetic Study Methods Kinetic runs were completed individually for each compound shown in Table 3.1 using powdered Calgon F400 activated carbon. For each run, an aliquot of F400 carbon slurry (prepared as stated above) was added to a 50-mL SGE gas tight Luer Lock syringe containing 50-mL of DDI water at a pH of 5 1, resulting in a final carbon concentration of 20-ppm. Using a Hamilton 1 to 10-L sy ringe the desired target contaminant was added from a stock soluti on to the 50-mL syringe, yielding a final contaminant concentration of 100-ppb. The 50-mL syringe was then rotated end over end. Kinetic studies were preformed for 3, 5, 10, 15, 30, and 60 minutes. Following each kinetic run, all samples were filtered into VOC vials using 25-mm Fisherbrand 0.20 or 0.45 m nylon syringe filters. Samples were sealed w ith Teflon septa caps and stored at 4C until analysis. All samples were analyzed w ith a Saturn 2100T GC/MS using a Supelco VOCOL™ Fused Silica capillary co lumn with a Supelco SPME 75m CARBOXEN™

PAGE 43

33 PDMS fiber, utilizing an oven temp erature program that holds at 40C for 2 minutes and then climbs to 210C at a rate of 8C/min. Each kinetic run was replicated. Only data points that were sufficiently reproducible, i.e., those whic h produced a RSD value of 20% or lower, were accepted. Development of Kinetic Equations The time for each compound to reach 86% removal with F400 was incorporated into the QSAR as a kinetic parameter. The va lue of 86% was selected as it appeared as a removal point common to each of the compounds in the training set. The time to 86% removal was interpolated from a trendline cr eated in Microsoft Excel from a graph of percent removal verses time. QSAR Development The first essential step in the developm ent of the QSAR was the selection of a suitable training set. For the parent compound, benzene was selected, and monosubstituted functionality dr ove the selection of the subs tituents. For this study it was desired to use single substi tutions within the training set to investigate the impact of a sole functional group on the adsorption syst em. In the selecti on of monosubstituted benzenes it is important to consider only the addition of functiona l groups that are well parameterized—i.e., that can be described by an available parameter within the QSAR. Non-descriptive groups cannot be well represented in the QSAR and therefore should be avoided in the selection of a tr aining set. It is also impor tant to include compounds that produce a wide spread of parameter values to ensure that the training set is representing a good range of electronic, steric hydrophobic, and chemical prope rties. A diverse spread of parameter values adds robustness to the QSAR process and can be illustrated graphically using a Craig plot (Lindner et al. 2003). Figure 3.1 demonstrates the spread

PAGE 44

34 of parameter values for both and Ind. It is important to note the spread of the data points throughout the quadrants, as the degree of spreading increases, so does the range of parameterization. For this study, monosubstituted benzenes were selected to demonstr ate a large range of steric, electronic, hydrophobic, and chemical parameterization. Table 3.2 illustrates the values of these parameters for the training set. Figure 3.1. Craig Plot Illu strating the Parameter Spread of a Hydrophobicity Constant ( ) and the Hammett Constant ( Ind). After the training set was chosen, each pa rameter was examined for collinearity. When plotted, any parameters that showed a linear correlation of R2 0.4 were considered collinear and were not used in multilinear regression analysis. -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5 -0.2-0.100.10.20.30.40.50.60.7 Ind

PAGE 45

35Table 3.2. Selected Parameters Used in the QSAR Training Se t (CRC Handbook of Chemistry and P hysics 1999, Physical Properties of Chemical Compounds 1955, MOPAC 2004). Electronics Descriptors Sterics Descriptors Hydrophobicity Descriptors Physical Properties Monosubstituted Benzene X Ind ACCP MR Taft Es Log Kow MW (g/ mole) Caq (mass %) TOTL () Fluorobenzene (FB) 0.52 -0.115 0.09 -0.55 2.27 0.14 96.1 0.154 5.21 Chlorobenzene (ClB) 0.47 -0.103 0.6 -0.97 2.84 0.39 113 0.0387 5.55 Bromobenzene (BrB) 0.44 -0.097 0.89 -1.16 2.99 0.86 157 0.0387 5.7 Iodobenzene (IodoB) 0.39 -0.099 1.39 -1.62 3.28 1.12 204 0.0387 5.8 Nitrobenzene (NB) 0.64 0.039 0.74 -2.52 1.85 -0.85 123 0.0193 6.07 Aniline (Ani) 0.12 -0.169 0.54 -0.61 0.9 -1.23 93.1 0.21 5.84 Phenol (Ph) 0.29 -0.145 0.28 -0.55 1.48 -1.12 94.1 3.38 5.64 Benzaldehyde (Bzal) 0.3 -0.066 0.69 --1.48 -0.32 106 8.66 7.02 Isopropylbenzene (IsoPB) 0.01 -0.106 1.5 -1.71 3.66 1.22 120 0.3 7.18 n-butylbenzene (nBB) -0.04 -0.11 1.96 -0.16 4.26 2.13 134 0.0056 9.71 Isobutylbenzene (IsoBB) -0.03 -0.107 1.96 -2.17 4.01 1.7 134 0.0015 8.19 t-butylbenzene (tBB) -0.07 -0.107 1.96 -2.78 4.11 1.98 134 0.001 7.27 Benzene (Benz) 0 -0.102 0.1 0 2.13 0 78.1 0.178 4.97 Where Ind represents the inductive effects descri bed by the Hammett Constant, ACCP represen ts the charge of the para-carbon on the functional group, MR represents molar refractivity, Es represents TaftÂ’s steric factor, log Kow is the octanol-water partitioning coefficient, represents the hydrophobici ty of the compound, MW is the molecular weight, Caq is the aqueous solubility in water, and TOTL is the total length of the compound, calculated by the bond lengths of the compound in the longest direction.

PAGE 46

36 Parameters representing intrinsic prop erties of the compounds (e.g., hydrophobic, electronic, steric, chemical, physical properties) were inco rporated into the QSAR as independent variables, while the parameters that were experimentally derived in this adsorption system (e.g., isotherm constants) were incorporated as dependent variables using SPSS 12.0 software (SPSS., Inc., Chica go, IL, USA). Classical QSAR procedure, as developed by Hansch, was followed for th e development of correlations (Hansch and Fujita 1995). Linear, polynomial, and multilinear correlations were employed to describe the data. Correlations that resulted from the QSAR an alysis had to meet a series of stringent statistical criteria to be considered valid: 1. No collinearity of independent variables (for MLR correlations) 2. R2 0.7 3. F-Ratio values must meet a 95% confidence limit 4. Standard error considered 5. Q2 determined for goodness of predictability Condition 1 is only applicable in regards to multi-variable analysis. Independent variables were linearly correlate d; variables that produced a R2 value of 0.4 or higher were considered collinear and were excluded from the QSAR analysis. Conditions 2 and 3 were helpful in identifying poor correlations The F-Ratio was determined using the number of compounds in each correlation (k) and two total data sets (n), observed and calculated values. The degrees of freedom were calculated for the variance between ( 1 = n-1) and within ( 2 = (n*k)-n) the data sets and comp ared to an upper 5% distribution table to determine the significance of the obs erved and calculated data sets. As the FRatio values become larger than the F-distribu tion values, the data sets are considered to be more similar. Therefore, high F-Ratio values indicate the predicted data from the

PAGE 47

37 QSAR correlation is in agreement with the observed data. The st andard error (Condition 4) represents fluctuations in the sampling set and can be seen in the form of residuals in the regression lines. Clearly, low values of standard error reflect better correlations within the training set. Condition 5 is an essential step in measuri ng the validation of th e correlations that result from the QSAR process. When performing a Q2 analysis, it is ideal to introduce a new compound to test the predictability of the QSAR correlation; however if the introduction of an additional compound is not feasible, an alternative method, roughly referred to as the “take one out” method, will suffice. For this alternative method, one compound from the training set is removed and the correlations are recalculated. The new correlations are then used to test the ab ility of the QSAR correlation to predict the value of the compound that was removed. Th is process is repeated for each compound within the training set and equation 3.1 is used to calculate the value of Q2: ) ( ) ( 1, 2 , 2 i observed i new predicted i observed iy y y y Q Eqn 3.1 While no set limit has been established for Q2 values, larger values indicate a better ability for the QSAR correlation to predict th e observed values with in the given system. Ideally the Q2 should be close to the value for the adjusted R2, thus indicating that the new correlation did not cause a significant shift in the predicted data. For this study a difference of 0.2 between the Q2 and adjusted R2 was considered the maximum limit for variation. Similar, yet less stringent, statis tical procedures as t hose outlined here have been recommended by Eriksson and coworkers (Eriksson et al. 2003).

PAGE 48

38 CHAPTER 4 MANUSCRIPT Introduction The use and production of toxi c synthetic aromatic compound s in applications such as the synthesis of rubber, pa int solvents, insectic ides, detergents, fragrances, fuels, and dry cleaning (USEPA 2002,USEP A 1995) continue to increase in our society, and as a result of their increased demand, the potential for introduction of th ese synthetic aromatic compounds into our environment is escalating. The contamination of water, air, and soils from exposures such as accidental spi lls, inappropriate disposal, and byproduct formation, will continue to exacerbate the condition of our surroundings without sufficient intervention and treatment. Given the increased exposure of the public to these chemicals, environmental and health im pacts such as toxicity, carcinogenicity, mutagenicity, and teratogenicity in humans a nd animals, aquatic toxicity, and degradation of potable water quality are also anticipated to increase. To circumvent such threats, treatment methods must be established to rem ove synthetic aromatics from air and water. It is well known that activa ted carbon is capable of removing aromatic compounds from contaminated water. Traditionally, the removal of volatile organic compounds ( VOCs), a class of compounds in which many aromatics belong, from water is accomplished thorough the use of activated carbon and/or aeration. Activated carbon and packed tower aeration are the USEPA-recommended forms of treatmen t for the removal of synthetic organic contaminants, such as benzene and chlorobe nzene, from drinking water (USEPA 2002,

PAGE 49

39 USEPA2 2002). The most effici ent method of aeration is by means of a packed tower. This system is effective, yet is rarely sufficient to achieve low maximum contaminant level (MCL) standards. It should also be not ed that in cooler climates, the use of the aeration towers becomes limited—due to po ssible ice formation and lower kinetics (Hammer and Hammer 2001). While the effici ency of a packed tower can be optimized, this process requires an increase in tower di ameter, high surface area packing materials, as well as increased tower height (Noll 1999). For some compounds these adjustments can prove costly or impossible. In order to ensure MCL regulations are met, aeration is usually coupled with activated carbon. Activated carbon has shown to be an extremel y efficient, cost effective adsorbent. As activated carbon adsorption can be reversible, captured ma terials can potentially be desorbed and salvaged. Activated carbon can effectively adsorb VOCs in both gaseous and aqueous media. The general mechanisms of adsorption are phys ical adsorption and chemisorption. Physical adso rption is driven by weak interm olecular forces, such as Van der Waals forces. It is im portant to note that physical adsorption does not entail the formation of a chemical bond, i.e., there is no exchange of electrons. Due to the lack of chemical bonding, physical adsorption is r eadily reversed (des orption). During chemisorption, a chemical bond results, thus making desorption much more difficult. The surface of activated carbon is gene rally non-polar, thus facilitating the adsorption of non-polar hydrocarbons, such as benzene, n-butylbenzene, and nitrobenzene, from a polar aqueous media, such as water. The functionality of a given activated carbon is highly dependent on its carb onaceous precursor and it s heat treatment. Therefore, different types of activated car bon will adsorb contaminants differently;

PAGE 50

40 hence, the efficiency is variable. Likewi se, the same activated carbon will have unique adsorption relationships with varying contaminants, i.e., the efficiency will vary according to the chemical and physical characteristics of the adsorbate. Activated carbon is currently an industry standard fo r the removal of organic contaminants in air and water (USEPA 2002). Although industry and the USEPA are confident in the capacity of carbon to rem ove organics, a clear method of how this removal is accomplished is not known. Cu rrently, there exists a disconnect between carbon scientists and carbon users (e.g., trea tment plant operators, engineers). For example, in academia, it is commonly taught th at activated carbon adsorption increases as solubility decreases (the Lundelius Rule). However, in the case of MIB and phenol, despite their similar size, MIB is much less soluble in water than MIB (see table 4.1), yet, as is recognized by water treatment professi onals, phenol is more readily adsorbed by carbon. Table 4.1. Size and Aqueous Solubility of MIB and Phenol (Perry and Green 1997, Whelton 2001). Compound Size () Aq. Solubility (mg/L) MIB 6 194 Phenol 6 82,000 Obviously, one cannot universally state th at adsorption increases as solubility decreases; another mechanism (i.e., surface ch emistry) may be influencing the adsorption of these two compounds. This creates a problem. Exceptions to widely accepted adsorption trends, though generally well know n among carbon scientists rarely seem to enter into the field of actual carbon applicati ons. The knowledge base created for future engineers and scientists in academia is carried with these professionals into the work

PAGE 51

41 field, into consulting firms a nd municipalities. These profe ssionals in turn continue to spread misconceptions to carbon users, such as treatment plant desi gners and operators. This lack of communication between the ca rbon scientists and users results in the improper use and constricts the utilization of activ ated carbon in the field. The result is a cookbook method of carbon dosing in treatment plants. To alleviate this black-box method of carbon dosing, various relationships are currently being examined between the adsorbate and the adsorbent. Two of these relationships are isotherms and Quantitative Structure Activity Relationships (QSAR). Once a carbon-contaminant relationship is established, a predictive model can be produced. This model could be a powerful tool for assisting the carbon user—helping to optimize the use of activated carbon and bridging the gap between carbon science and application. The quest for a unified adsorption theory may be surrendered for the use of such correlations, which can optimize the use of activated carbon without requi ring a comprehensive understanding of the mechanisms which are involved. Ultimately, the development of such correlations may lead to an enhanced knowledge of the adsorption pr ocess. An enhanced understanding of adsorption mechanisms will lead to reduced experimentation, time, and cost of activated carbon selection. The overall short term goal of this proj ect is to develop a protocol for QSAR analysis of activated carbon. In order to de velop such a protocol, this study will examine the predictability of QSAR for a training set of monos ubstituted benzenes using Freundlich and Langmuir isotherm constants as well as kinetic rate data as dependent variables. To allow for a greater focus on th e changing functionality of the training set, only one activated carbon will be used in this initial study. Other intrinsic short term

PAGE 52

42 goals for this work include: examining the pr edictability of QSAR for this training set, determining the most descriptive isotherm equation for this training set and carbon (i.e., Freundlich or Langmuir), and an examina tion of the impact of kinetic rates. Because of its widespread use in the l iterature, Calgon Filtrasorb 400 (F400) has been selected as the activated carbon for this study. Selected phys ical properties of F400 are shown in Table 4.2. Table 4.2. Selected Physical Propert ies of F400 (Karanfil and Kilduff 1999). Surface Area (m2/g) Ave. Pore Radius () Pore Volume (cm3/g) Percent Surface Area in Pores < 20 Percent Surface Area in Pores 20100 Percent Surface Area in Pores 100200 Percent Surface Area in Pores > 200 948 12 0.566 83.7 15.0 0.90 0.40 To estimate the adsorption capability of an activated carbon, Freundlich or Langmuir isotherms are commonly develope d. However, according to Weber and DiGiano (1996) : “Despite th e sound theoretical basis of th e Langmuir, BET, and Gibbs models, these isotherms often fail to desc ribe experimental solution sorption data adequately.” Due to its empi rical derivation, it is proposed that the Freundlich isotherm equation (Eqn 4.1) may better describe the ad sorption process, since it innately accounts for the heterogeneous surface energy distribution of activated carbon. n eKC q1 Eqn 4.1 Where: qe = the mass of contaminant ad sorbed per unit mass of carbon K = “adsorption capacity”, Freundlich constant

PAGE 53

43 C = concentration of the adsorbat e in solution at equilibrium 1/n = strength of adsorption, Freundlich constant Developed originally for air phase adsorp tion, the Langmuir equation (Eqn 4.2) has been widely adapted for aqueous systems by re placing the original pressure term with a term representing the solute equilibrium concentration. e a a eC bQ Q q 1 1 1 Eqn 4.2 Where: qe = the mass of contaminant ad sorbed per unit mass of carbon Qa = maximum adsorption capacity b = net enthalpy of adsorption Ce = equilibrium concentration of adsorbate in solution The Langmuir equation draws its theory fr om the condensation and evaporation of gas molecules on a solid adsorbent surface (Weber and DiGiano 1996). Condensation takes into account available adsorption sites on the adsorben t surface as well as the rates at which the adsorbates contact the adsorb ent surface. Three basic assumptions are encompassed within the Langmuir equation: (1 ) adsorption energy is constant throughout the activated carbon surface and is independent of surface coverage (2) there are no interactions between adsorbate molecules or migration of ad sorbates to other adsorption sites (3) the carbon surface w ill only support a monolayer adsorption of the adsorbate (Weber and DiGiano 1996).

PAGE 54

44 While adsorption isotherms are useful in developing adsorbate-adsorbent relationships, they are time intensive and nonspecific in regards to adsorbate-adsorbent structural interactions. Relying solely on is otherm relationships w ould require extensive laboratory analysis for each i ndividual compound (and condition) that is to be analyzed. The adsorption of a chemical compound to ac tivated carbon is dependent on both the chemical and physical properties of the carbon and the compound. It should then follow that by knowing the chemical and physical properties, the ad sorbate-adsorbent relationship can theoretically be predicted. Using known parameters to predict the activity of a compound is the focus of QSAR modeling. By appl ying an empirically based model to an appropriate system, the m echanisms within the adsorption process can be described and accounted for without being fu lly disclosed. QSAR is becoming widely known in the medical and biologi cal fields, but has yet to s ee a dominating presence in adsorption study (Hansch and Fujita 1995). Blum et al. (1994) have developed a QSAR for activated carbon adsorption in water usi ng 363 organic compounds. While this model is noteworthy, it is based on literature-deri ved constants—which may vary drastically with respect to experimental procedure a nd analysis—without a st urdy foundation, these models will not accurately portray the adsorp tion process. Table 4.3 illustrates the discrepancies which can be found in the literature. Table 4.3. Reported Values of Freundlich K for Benzene. K Source 0.12602 Noll 1999 1.0 Dobbs and Cohen 1980 41.7 Loll et al. 2004

PAGE 55

45 Blum et al. (1994) show an understanding of th is vital point as the authors mention that “…in applying QSAR based on literature data, it is imperative to consider the relevance of the data source to one’s own investiga tive situation.” However, when used in conjunction with accurately descriptive para meters (i.e., case specific), QSAR can be powerful tools for understanding ac tivated carbon adsorption. In recent years, carbon scientists have begun to examine the potential of QSAR analysis for gaining insight into the method(s) of activated carbon adsorption (Blum et al. 1994, Brasquet et al. 1997). To date, no comp elling methodologies have been established for developing a protocol for QSAR-adsorpti on analysis. The studies examined in the literature typically util ize literature derived isotherm constants which are extracted from a variety of sources. This method of sele cting parameters can become somewhat haphazard, as many sources use different expe rimental conditions (a nd different carbons) when developing these isotherm constants. To avoid this inefficient approach, this study has experimentally derived all dependen t variables—under standardized testing conditions—to create uniformity between the compounds in the training set. A uniform approach to parameter development will lead to more reliable QSAR correlations. In addition to carefully developing the dependent variables, this study includes an array of electrical, steric, hydro phobic, and physical parameters as independent variables to more accurately describe and investigate the role of the adsorbate in the adsorption process. Also in contrast to most QSAR-adsorption studies present in the literature, this study implements a series of rigorous statistical tests before a correlation is accepted. It is the hypothesis of this study that the development of a robust QSAR will enhance the understandi ng of the adsorption process of F 400 with this given training set

PAGE 56

46 of monosubstituted benzenes. Ultimately, an enhanced understanding of the adsorption process will lead to an optimi zed use of activated carbon. Experimental Adsorption Isotherms Adsorption isotherms were develope d by adding a known concentration of contaminant, 100g/L, and activated carbon to distille d deionized (DDI) water at a pH of 5.5 1.5, the variation is due to the us e of unaltered DDI water. The experimental data was collected for each contaminant individually using powdered Calgon F400 activated carbon, which was dosed as a slur ry so that accurate quantities could be developed, with 100g/L solutions of each compound in th e training set (see Table 4.4). The system was mixed end over end in a gas tight syringe on a rotator until equilibrium was reached. After mixing, samples were filtered into VOC vials using 25-mm Fisherbrand nylon syringe filters which were sized to allow for only the PAC to be removed (both 0.20-m and 0.45m were used, depending on their availability). Samples were sealed with Tefl on septa caps and stored at 4C until analysis. All samples were analyzed with a Saturn 2100T GC/M S using a Supelco VOCOL™ Fused Silica capillary column with a Supelco SPME 75m CARBOXEN™ PDMS fi ber, utilizing an oven temperature program that holds at 40C for 2 minutes and then climbs to 210C at a rate of 8C/min. Each isotherm was replicated a nd was only considered valid if the RSD was less than 20% and the R-square value was 0.70 or greater. Isotherm data was formatted using the Freundlich and Langmuir isotherm equations (Eqn 4.1 and 4.2). The Freundlic h and Langmuir constants were extracted from the graphical interpretati on of the data for each compou nd in the training set. The

PAGE 57

47 graphics in Figure 4.1 dem onstrate (A) a dose removal curve and (B) a graphical representation of the Freund lich isotherm using n-butylbenzene as an example.

PAGE 58

48 Figure 4.1. Example of Isotherm Deve lopment. (A) Dose removal curve for Isobutylbenzene and Calgon F400 Activated Carbon. (B) Freundlich Isotherm of Isobutylbenzene and Cal gon F400 Activated Carbon. 0 20 40 60 80 100 120 0510152025Conc PAC (ppm)% Removal n-butylbenzene y = 0.4199x + 5.8488 R2 = 0.9812 6.4 6.6 6.8 7 7.2 7.4 7.6 7.8 8 012345 log Cflog qe A B

PAGE 59

49 Kinetic Data Collection Kinetic studies proceeded in a sim ilar manner, taking each contaminant individually at a c oncentration of 100g/L with F400 at a consta nt concentration of 20mg/L. Kinetic data was collected in the sa me manner as the isotherm data, with the exception that sample run durations were: 3, 5, 10, 15, 30, and 60 minutes. Following each kinetic run, the samples were filtered, stored, and analyzed using the same methods as described for the equilibrium samples. Ki netic data was incorporated into the QSAR by determining the time for F400 to remove 86% of the target contaminant. Experimentally derived values for this kine tic parameter (log t) and selected isotherm constants are shown in Table 4.4. Table 4.4. Experimenta lly Derived Values for log K, log Q, and log t. Compound log K R2 log Q R2 log t R2 Fluorobenzene Chlorobenzene Bromobenzene Iodobenzene Nitrobenzene Aniline Phenol Benzaldehyde Isopropylbenzene n-butylbenzene Isobutylbenzene t-butylbenzene Benzene 5.38 5.34 5.38 5.34 5.05 4.38 3.64 4.82 5.38 4.62 5.00 4.25 4.94 0.91 0.99 0.99 0.99 0.95 0.97 0.90 0.71 0.92 0.92 0.81 0.83 0.89 1.01 1.19 1.33 1.35 1.43 1.26 1.58 --1.53 2.05 2.80 2.89 0.92 0.88 0.95 0.93 0.93 0.96 0.99 0.95 --0.87 0.94 0.94 0.81 0.96 1.17 1.08 1.03 1.20 0.88 1.78 --0.75 1.06 0.45 -0.01 0.94 1.76 0.94 0.94 0.98 0.77 0.98 0.93 --0.69 0.85 0.92 0.96 0.96 0.91 Establishment of QSAR The first essential step in the developm ent of the QSAR was the selection of a suitable training set. For the parent compound, benzene was selected, and monosubstituted functionality dr ove the selection of the subs tituents. For this study it was desired to use single substi tutions within the training set to investigate the impact of a sole functional group on the adsorption syst em. In the selecti on of monosubstituted

PAGE 60

50 benzenes it is important to consider only the addition of functiona l groups that are well parameterized—i.e., that can be described by an available parameter within the QSAR. Non-descriptive groups cannot be well represented in the QSAR and therefore should be avoided in the selection of a tr aining set. It is also impor tant to include compounds that produce a wide spread of parameter values to ensure that the training set is representing a good range of electronics, sterics, hydr ophobic, and chemical properties. For this study, monosubstituted benzenes were selected to demonstr ate a large range of steric, electronic, hydrophobic, and chemical parameterization. Table 4.5 illustrates the values of these parameters for the training set. After the training set was chosen, each pa rameter was examined for collinearity. When plotted, any parameters that showed a linear correlation of R2 0.4 were considered collinear and were not used in multilinear regression analysis. Parameters representing intrinsic proper ties of the compounds (e.g., hydrophobic, electronic, steric, chemical, physical properties) were incor porated into the QSAR as independent variables, while the parameters that were experimentally derived in this adsorption system (e.g., isotherm constants) were inco rporated as dependent variables using SPSS 12.0 software (SPSS., Inc., Chicago, IL USA). Classical QSAR procedure, as developed by Hansch, was followed for the development of correlations (Hansch and Fujita 1995). Linear, polynomial, and multilinear correlations were employed to describe the data.

PAGE 61

51Table 4.5. Selected Descriptors of Substituent or Compound Electronics, Sterics, and Hydrophobicity (H ansch et al., 1995; Perry and Green, 1997; Schwartzenbach et al., 2003; MOPAC, 2004). Electronics Descriptors Sterics Descriptors Hydrophobicity Descriptors Physical Properties Monosubstituted Benzene X Ind ACCP MR Taft Es Log Kow MW (g/ mole) Caq (mass %) TOTL () Fluorobenzene (FB) 0.52 -0.115 0.09 -0.55 2.27 0.14 96.1 0.154 5.21 Chlorobenzene (ClB) 0.47 -0.103 0.6 -0.97 2.84 0.39 113 0.0387 5.55 Bromobenzene (BrB) 0.44 -0.097 0.89 -1.16 2.99 0.86 157 0.0387 5.7 Iodobenzene (IodoB) 0.39 -0.099 1.39 -1.62 3.28 1.12 204 0.0387 5.8 Nitrobenzene (NB) 0.64 0.039 0.74 -2.52 1.85 -0.85 123 0.0193 6.07 Aniline (Ani) 0.12 -0.169 0.54 -0.61 0.9 -1.23 93.1 0.21 5.84 Phenol (Ph) 0.29 -0.145 0.28 -0.55 1.48 -1.12 94.1 3.38 5.64 Benzaldehyde (Bzal) 0.3 -0.066 0.69 --1.48 -0.32 106 8.66 7.02 Isopropylbenzene (IsoPB) 0.01 -0.106 1.5 -1.71 3.66 1.22 120 0.3 7.18 n-butylbenzene (nBB) -0.04 -0.11 1.96 -0.16 4.26 2.13 134 0.0056 9.71 Isobutylbenzene (IsoBB) -0.03 -0.107 1.96 -2.17 4.01 1.7 134 0.0015 8.19 t-butylbenzene (tBB) -0.07 -0.107 1.96 -2.78 4.11 1.98 134 0.001 7.27 Benzene (Benz) 0 -0.102 0.1 0 2.13 0 78.1 0.178 4.97 Where Ind represents the inductive effects descri bed by the Hammett Constant, ACCP represen ts the charge of the para-carbon on the functional group, MR represents molar refractivity, Es represents TaftÂ’s steric factor, log Kow is the octanol-water partitioning coefficient, represents the hydrophobici ty of the compound, MW is the molecular weight, Caq is the aqueous solubility in water, and TOTL is the total length of the compound, calculated by the bond lengths of the compound in the longest direction.

PAGE 62

52 Correlations that resulted from the QSAR an alysis had to meet a series of stringent statistical criteria to be considered valid: 4. No collinearity of independent variables (for MLR correlations) 5. R2 0.7 6. F-Ratio values must meet a 95% confidence limit 7. Standard error considered 8. Q2 determined for goodness of predictability Condition 1 is only applicable in regards to multi-variable analysis. Independent variables were linearly correlate d; variables that produced a R2 value of 0.4 or higher were considered collinear and were excluded from the QSAR analysis. Conditions 2 and 3 were helpful in identifying poor correlations The F-Ratio was determined using the number of compounds in each correlation (k) and two total data sets (n), observed and calculated values. The degrees of freedom were calculated for the variance between ( 1 = n-1) and within ( 2 = (n*k)-n) the data sets and comp ared to an upper 5% distribution table to determine the significance of the obs erved and calculated data sets. As the FRatio values become larger than the F-distribu tion values, the data sets are considered to be more similar. Therefore, high F-Ratio values indicate the predicted data from the QSAR correlation is in agreement with the observed data. The st andard error (Condition 4) represents fluctuations in the sampling set and can be seen in the form of residuals in the regression lines. Clearly, low values of standard error reflect better correlations within the training set. Condition 5 is an essential step in measuri ng the validation of th e correlations that result from the QSAR process. When performing a Q2 analysis, it is ideal to introduce a new compound to test the predictability of the QSAR correlation; however if the

PAGE 63

53 introduction of an additional compound is not feasible, an alternative method, roughly referred to as the “take one out” method, will suffice. For this alternative method, one compound from the training set is removed and the correlations are recalculated. The new correlations are then used to test the ab ility of the QSAR correlation to predict the value of the compound that was removed. Th is process is repeated for each compound within the training set and Equation 4.3 is used to calculate the value of Q2: ) ( ) ( 1, 2 , 2i observed i new predicted i observed iy y y y Q Eqn 4.3 While no set limit has been established for Q2 values, larger values indicate a better ability for the QSAR correlation to predict th e observed values with in the given system. Ideally the Q2 should be close to the value for the adjusted R2, thus indicating that the new correlation did not cause a significant shift in the predicted data. For this study a difference of 0.2 between the Q2 and adjusted R2 was considered the maximum limit for variation. Discussion of Results From the experimentally derived isotherm parameters, only the terms representing the adsorption capacity (log K and log Q) yielde d viable correlations for this training set; therefore, the isotherm data listed in Table 4.5 only shows these isotherm parameters. It is interesting to note that log K and log Q do not correlate with each other within this training set. In fact, a plot of these two parameters yields an R2 value of 0.0804. This lack of relationship implies that while l og K and log Q both represent an adsorption capacity, their actual derivation (and definitio n) is quite different. Webber and DiGiano (1996) present a relationship between the Fr eundlich and Langmuir constants shown in Eqn. 4.4:

PAGE 64

54 n g g a Fb Q K, Eqn. 4.4 However, the experimental results found in this study do not support this (or any) relationship between Langmuir and Freundlich constants. Scientists and carbon users should take caution when using a single parameter, such as an adsorption capacity, to describe a system. Using the experimentally derived isotherm constants and kinetic descriptors, QSAR analyses were performed and correlations were derived. Initially, hundreds of correlations were developed and tested, yet onl y the six correlations shown in Table 4.6 satisfied all of the statistical validation requirements. Correlation 1, shown in Table 4.6 and Fi gure 4.2, indicates that as aqueous solubility and molar refractivity increase, log K, the Freundlich adsorption capacity, decreases. This decrease in adsorption is expected, as th e increasing solubility of a compound will lower its tendency to come out of solution and sorb onto the carbon. This trend is commonly seen in the field of adso rption and is described by the Lundelius Rule. The increase in molar refractivity (MR), which is a frequently used indicator of molecular volume, correlates with a decrease in adsorp tion capacity, which is also expected for a microporous carbon such as F400. The larg est compounds in this training set reach nearly 10 in length, when estimated th rough the summation of bond lengths in the longest direction of the compound. Wh en estimating adsorp tion, Kasaoka and coworkers, as reported by Tennant and Mazy ck (2003), have asserted that a compound will adsorb in a pore that is 1.5 to 2 time s its diameter. Using this estimator, it is reasonable that the largest compounds in this training set (e.g., n-butylbenzene and isobutylbenzene) will not undergo optimi zed adsorption with a predominately

PAGE 65

55 Table 4.6. Selected Results of QSAR Analysis with log K, log Q, and log t as the Dependent Variables.a,b Correlations n R2 Adj. R2 SE FRatio Q2 1.) Log K = -0.196(0.0915)Caq-0.258(0.326)MR+5.334 2.) Log K = -0.17(0.0765)Caq+0.851(0.77) ind+4.857 3.) Log Q = 0.558MR2-0.457MR+1.242 4.) Log K = -0.187(0.083)Caq-0.147(0.152)TOTL+6.024 5.) Log t = 0.197(0.212)Caq-0.242(0.146)TOTL+2.52 6.) Log t = -5.836(6.68)ACCP-0.267(0.292)TOTL+2.162 13 13 12 13 12 12 0.693 0.749 0.781 0.724 0.715 0.703 0.631 0.699 0.733 0.669 0.651 0.637 0.33 0.30 0.33 0.31 0.29 0.30 11.260 14.914 16.07 13.131 11.266 10.644 0.458 0.607 0.582 0.560 0.603 0.694 a Caq = aqueous solubility (mass %), MR = molar refractivity, TOTL = estimated total length of compound, ACCP = measure of the para-carbon charge. Numbers in parentheses represent the 95% confidence interval (+/-) for the coefficient. b n = number of compounds used in correlation development, Adj. R2 = adjusted R2, SE = standard error. Due to inefficient data Correlation 3 does not include benzaldehyde and Correlations 5 and 6 do not include phenol.

PAGE 66

56 microporous carbon (average pore diameter less th an 20 ) such as F400. The results of this correlation appear to be sound, as only one com pound showed a percent difference between observed and predicted values that was greater than 10%, t-butylbenzene. Figure 4.2. QSAR Correlation 1: Predicte d versus Observed Values of log K. The trends described by Correlation 2, show n in Table 4.6 and Figure 4.3, show a decrease in adsorption capacity (log K) with respect to an increase in aqueous solubility and a decrease in HammettÂ’s constant, Ind. The relationship between log K and the aqueous solubility of the training set is exp ected and is consistent with the results of Correlation 1. However, it is un expected that a decrease in Ind would lead to a reduced adsorption capacity. A decrease in Ind indicates an increase in the electron donating potential of a compound, due to a higher elect ron density about the aromatic ring of the compounds in this training set. These results contrast with the bonding theory developed by Coughlin and Ezra (1968) and accepted by numerous renowned carbon scientists, such as Radovic (2001). According to the theory, an increase in the 3.5 4 4.5 5 5.5 6 3.544.555.5 ObservedPredicted Correlation 1: log K = -0.196Caq-0.258MR+5.334 R2=0.693, Adj. R2=0.631, se=0.33, F-Ratio=11.26, Q2=0.458 tBB Ph N B Bzal An nBB Clb IodoB IsoPB IsoBB Benz BrB Flb

PAGE 67

57 electron density about th e benzene ring should enha nce the ability of the electrons to interact, thus enhancing the potential for ad sorption. Echoing th e results seen in Correlation 1, only t-butylbenzene yielded a percent difference between predicted and observed value that was greater than 10%. Further study is necessary to confirm this trend will apply to a greater training set. Figure 4.3. QSAR Correlation 2: Predicte d versus Observed Values of log K. Correlation 3, shown in Table 4.6 and Fi gure 4.4, relates the Langmuir adsorption capacity, log Q, with MR. The results found in this correlation show that the adsorption capacity increases as the MR—or molecular volume—increases, which contradicts the results found in Correlation 1. It does not seem logical that a microporous carbon, such as F400, would have an increasing adso rption capacity with increasing compound volume, as the largest compounds in this traini ng set are approximately 10 in length. It should be noted that the largest compound in this training set, nbutylbenzene, showed the largest residual from the trendline in Correlation 3. The differences between 3.5 4 4.5 5 5.5 6 3.544.555.5 ObservedPredicted Correlation 2: log K = -0.17Caq+0.851sigI+4.857 R2=0.749, Adj. R2=0.699, se=0.30, F-Ratio=14.914, Q2=0.607 tBB Ph N B Bzal An nBB ClB IodoB IsoPB IsoBB Benz BrB FB

PAGE 68

58 Correlation 1 and Correlation 3 confirm th at the Langmuir Q and Freundlich K are describing different mechanisms for adsorpti on and should not be used interchangeably. Although more research comparing log K and log Q is necessary, in this particular instance, it appears that the Freundlich K is most accurately describing the adsorption of the training set with F400. Despite the R2 value and sufficient statistical results, nine of the twelve compounds tested for this correla tion showed greater than a 10% difference between predicted and observed values. In fact, of those nine compounds, six showed nearly or greater than a 20% difference. More research should be performed with an expanded training set to confirm the effectiv eness of log Q for de scribing adsorption in relation with MR. Figure 4.4. QSAR Correlation 3: Predicte d versus Observed Values of log Q. As shown in Table 4.6 and Figure 4.5, Correlation 4 states that as aqueous solubility and total length increase, log K d ecreases. These results were expected, and support Correlations 1 and 2 with respect to aqueous solubility. The total length 0 0.5 1 1.5 2 2.5 3 00.511.522.533.5 ObservedPredicted Benz tBB IsoBB nBB IsoPB IodoB BrB N B ClBAn Ph FBCorrelation 3: log Q = 0.558MR2-0.457MR+1.242 R2=0.781, Adj. R2=0.733, se=0.33, F-Ratio=16.067, Q2=0.582

PAGE 69

59 parameter is estimated from the bond lengths of the compounds in the longest dimension. An increase in total length indicates an increa se in the bulkiness of the compound, and as expected for this microporous carbon, relates to a decrease in adsorption capacity. These results for total length are in agreement with the MR results for Correlation 1. As was found with Correlations 1 and 2, only one compound in Correlation 4 shows a percent difference between predicted and observed va lues greater than 10%, t-butylbenzene. Figure 4.5. QSAR Correlation 4: Predicte d versus Observed Values of log K. Correlation 5, as shown in Table 4.6 and Figure 4.6, shows the relationship between aqueous solubility, total length, and l og t. The value log t is a kinetic parameter that represents the amount of time that a 20-ppb slurry of F400 took to achieve 86% removal of a 100-ppb concentration of a target contaminant. As expected, and in support with Correlations 1, 2, and 4, a greater aqueous solubility resulted in a longer time to achieve 86% adsorption. However, in Correlati on 5, a decrease in total length denoted an increase in the time to reach 86% adsorption. This finding appears to contradict with the 3.5 4 4.5 5 5.5 6 3.544.555.5 ObservedPredicted Correlation 4: log K = -0.187Caq-0.147TOTL+6.024 R2=0.724, Adj. R2=0.669, se=0.31, F-Ratio=13.131, Q2=0.560 tB B Ph NB Bzal An nBB Clb IodoB IsoPB IsoBB Benz BrB Flb

PAGE 70

60 results of Correlations 1 and 4 which show that an increase in molecular volume and an increase in total length result in a decr ease in adsorption capacity (log K). This unexpected relationship between log K and total length could be an effect of steric hindrances, buttressing of the molecules agains t the carbon surface, por e size distribution. It should be noted that while all other correlations describe solely equilibrium conditions (which the Freundlich and Langmuir equations require by definition); log t does not represent equilibrium conditions between the carbon and contaminant. For Correlation 5, all but two compounds (aniline and iodobenzen e) showed a 10% or greater difference between predicted and observed values. For these reasons, the log t results shown here cannot be accurately compared to the isotherm constant results; ther efore further research should be done to confirm these findings. 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 -0.050.450.951.451.95 ObservedPredicted Correlation 5: log t = 0.197Caq-0.242TOTL+2.52 R2=0.715, Adj. R2=0.651, se=0.29, F-Ratio=11.266, Q2=0.603 tB B N B Bzal An nBB Clb IodoB IsoPB IsoBB Benz BrB Flb Figure 4.6. QSAR Correlation 5: Predicte d versus Observed Values of log t.

PAGE 71

61 Figure 4.7. QSAR Correlation 6: Predicte d versus Observed Values of log t. The final correlation examined in this anal ysis is Correlation 6, shown in Table 4.6 and Figure 4.7. Correlation 6 indicates that an increase in ACCP (the charge of the carbon in the para position on the functional group of the target compound) and total length results in a decrease in the time to re ach 86% adsorption of the target contaminant by F400. As ACCP increases, the functi onal group of the co mpound becomes more electron withdrawing, which indicates a redu ced electron density on these carbon atoms. Correlation 6 shows that as th e para carbon becomes more elect ron withdrawing, the time to 86% decreases—this indicates that a re duced electron densit y about the carbon is enhancing the uptake of the compound by F400. This finding, as seen with Correlation 2, contradicts the bonding theory. Yet, as was seen in Correlation 5, the total length and log t are relating in an unexpected manner—s howing a decrease in time to reach 86% as the compound length increases. Much like Co rrelation 5, in Correlation 6, all values except one (iodobenzene) showed percent di fferences between predicted and observed 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 -0.050.450.951.451.95 ObservedPredicted Correlation 6: log t = -5.836ACCP-0.267TOTL+2.162 R2=0.703, Adj. R2=0.637, se=0.30, F-Ratio=10.644, Q2=0.694 tB B N B Bzal An nBB Clb IodoB IsoPB IsoBB Benz BrB Flb

PAGE 72

62 values to be greater than 10%. As st ated previously, the method for parameter development for log t is not id eal and further study should be performed to clarify these results. Overall, as expected, Correl ations 1, 2, and 4 support th at as aqueous solubility increases, log K (adsorption cap acity) decreases. As the ster ic factors of the compounds increase, Correlations 1 and 4 show th at log K decreases. Though unexpected, Correlation 3 confirms the finding that log K and log Q do not demonstrate a distinct relationship with each other. For this reason, caution should be used when utilizing these parameters to describe an adsorption system. The correlations incorporating log t, Correlations 5 and 6, require further testing to verify the effectiveness of log t as a valid parameter for describing carbon adsorption. It has also been suggested, through Correlations 2 and 6, that bonding may not be the dominate mechanism for adsorption in this training set. The succe ssful correlations between aqueous solubility, sterics, and log K demonstrate optimistic e ffectiveness for using these parameters for adsorption prediction and signify the value of the functional and effective QSAR protocol utilized in this study. The critical steps of th is protocol include: The selection of a parent compound and training set Standardized experimentation, at equilib rium conditions, of the training set with activated carbon(s) Uniform analysis of the experimental data to formulate representative dependent variables Careful consideration and selection of reliable and well documented independent variables Testing for collinear parameters Meticulous QSAR modeling

PAGE 73

63 Rigorous statistical validati on of the QSAR correlations The use of this protocol may be an essent ial step in working towards the goal of understanding the mechanisms of carbon adsorption phenomenon.

PAGE 74

64 CHAPTER 5 CONCLUDING REMARKS As stated by Mattson et al. (1969a), it is essential for the advancement of the field of activated carbon to iden tify and understand the mech anisms by which adsorption occurs. An understanding of the role of f unctional groups in the ad sorption process is a key component to unveiling the mechanisms be hind the process. The activation process is pivotal in the development of the surface chemistry, and indeed in the formation of functional groups present on the activated carbon. A recent workshop held by the National Science Foundation has emphasized the potential of QS ARs as a tool for uncovering mechanisms in poorly understood ph enomenon in the field of water treatment (NSF 2004). Dabrowski states in his work “Adsorpti on—From Theory to Practice” that “. . there is a need for close co-operation betw een theoretical and e xperimental groups, in which the experiments and models are design ed to complement each other” (Dabrowski 2001). Overwhelmingly, in the literature, res earchers are either deve loping models using past research or performing re search using previously deve loped models. As Dabrowski implies, in order to have synergy between models and the experimental data, the researchers must integrate both into their de signs. One approach that takes this into consideration is the experime ntal-modeling method proposed in this study. In this method, isotherm and kinetic data is collect ed for a training set of compounds and is included along with widely accepted physical and chemical constants to form a robust QSAR. This approach takes into consid eration the effects of contaminant surface

PAGE 75

65 chemistry and relies upon both experiment al and modeling data to describe the mechanisms of adsorption. Ultimately, this method shall directly incorporate the effects of carbon surface chemistry as independent variables. The addition of such parameters will make for a more robust QSAR which may lead to a deeper understanding of the role of surface chemistry in the adsorption process. If it is the intent of scie ntific community to advance the field of activated carbon, the mechanisms of adsorption must be iden tified and understood. Without the knowledge behind the process, the use of activated carbon will not be able to surpass other separation technologies. An experimental-mode ling method is a doorway to understanding adsorption phenomena.

PAGE 76

66 LIST OF REFERENCES Adamson Arthur W. Physical Chemistry of Surfaces: Fifth Edition. New York, New York: John Wiley & Sons, Inc., 1990. Boehm H.P. Functional Groups on the Surfaces of Solids. Angewandte Chemie: International Edition Vol. 5, No. 6, pp. 533-622, 1966. Blum Diane J.W., Suffet Irwin H., Duguet J.P. Quantitative Structure Activity Relationship Using Molecular Connectivity for the Activated Carbon Adsorption of Organic Chemicals in Water. Water Res earch. Vol. 28, No. 3, pp. 687-699, 1994. Brasquet Catherine, Subrenat Etienne, Le Cloirec Pierre. Selective Adsorption on Fibrous Activated Carbon of Organics from Aqueous Solution: Correlation between Adsorption and Molecular Structure. Water Science Technology Vol. 35, No. 7, pp. 251-259, 1997. Carey Francis A. Reactions of Arenes: Electr ophilic Aromatic Substitutions The McGraw-Hill Companies Student Online Le arning Center. Available [Online]: http://www.mhhe.com/physsci/chemistry/ca rey/student/olc/gra phics/carey04oc/ref/ ch12substituenteffects.html Accessed December 6, 2003. Chiou Cary T. Partition and Adsorption of Organic Contaminants in Environmental Systems. Hoboken, New Jersey: Wiley-Interscience, 2002. Coughlin Robert W. and Ezra Fouad S. Ro le of Surface Acidity in the Adsorption of Organic Pollutants on the Surface of Carbon. Environmental Science and Technology Vol. 2, No. 4, pp. 291-297, 1968. Coughlin Robert W., Ezra Fouad S., Tan Ricar do N. Influence of Chemisorbed Oxygen in Adsorption onto Carbon from Aqueous Solution Journal of Colloid and Interface Science Vol. 28, No. 3/4, pp. 386-396, 1968. D browski A. Adsorption—from theory to practice. Advances in Colloid and Interface Science Vol. 93, pp. 135-224, 2001. Dobbs Richard A., Cohen Jesse M. Carbon Adsorption Isotherms for Toxic Organics. EPA: Municipal Environmental Research Laboratory, Office of Research and Development. April 1980.

PAGE 77

67 Eriksson Lennart, Jaworska Joanna, Worth Andrew P., Cronin Mark T.D., McDowell Robert M., Gramatica Paola. Methods for Reliability and Uncertainty Assessment and for Applicability Evaluations of Cla ssification and Regre ssion Based QSARs. Environmental Health Perspectives Vol. 111, No. 10, pp. 1361-1375, August 2003. Ergun S. and Mentser M. Reactions of carbon with carbon dioxide and steam. Chemistry and Physics of Carbon. Vol. 1, New York, New York: Marcel Dekker, pp. 204261, 1965. Fangmark Ingrid E., Hammarstrom Lars-Gunna r, Stromqvist Marianne E., Ness Amanda L., Norman Paul R., Osmond Neale M. Estimation of Activated Carbon Adsorption Efficiency for Organic Va pors: A Strategy for Selecting Test Compounds. Carbon Vol. 40, pp. 2861-2869, 2002. Franz Marcus, Hassan Arafat A., Pinto Neville G. Ef fect of Chemical Surface Heterogeneity on the Adsorption Mechanis m of Dissolved Aromatics on Activated Carbon. Carbon Vol. 38, pp. 1807-1819, 2000. Gregova K., Petrov N., Butuzova L., Minkova V., Isaeva L. Evolution of the Active Surface of Carbons Produced from Various Raw Materials by Steam Pyrolysis/Activation. Journal of Chemical T echnology and Biotechnology Vol. 58, pp. 321-330, 1993. Guo J., Lua A.C. Characterization of Adso rbent Prepared from Oil-Palm Shell by CO2 Activation for Removal of Gaseous Pollutants. Materials Letters Vol. 55, No. 5, pp. 334-339, 2002. Hammer Mark J. and Hammer Mark J. Jr. Water and Wastewater Technology: Fourth Edition. Upper Saddle River, Ne w Jersey: Prentice-Hall, 2001. Hansch Corwin and Fujita Tosiho. Classical and Three-Dimensional QSAR in Agrochemistry. Washington DC: American Chemical Society, 1995. Hansch C., Leo A., Hoekman D. Exploring QSAR: Hydrophobic, Electronic, and Steric Constants Washington DC: American Chemical Society, 1995. Johns Mitchell M., Marshall Wayne E., Toles Ch ristopher A. The E ffect of Activation Method on the Properties of Pecan Shell-Activated Carbons. Journal of Chemical Technology and Biotechnology Vol. 74, pp. 1037-1044, 1999. Karanfil Tanju and Kilduff James E. Ro le of Granular Activated Carbon Surface Chemistry on the Adsorption of Organic Co mpounds. 1. Priority Pollutants. Environmental Science and Technology Vol. 33, No. 18, pp. 3217-3224, 1999. Kipling J.J. and Shooter P.V. The Adsorpti on of Iodine Vapor by Gr aphon and Spheron. Journal of Colloid and Interface Science Vol. 21, pp. 238-244, 1966.

PAGE 78

68 Lindner Angela S., Whitfield Callie, Chen Ning, Semrau Jeremy D., Adriaens Peter. Quantitative Structure-Biodegradation Relationships for Ortho-Substituted Biphenyl Compounds Oxidized by Me thylosinus Trichosporium OB3b. Environmental Toxicology and Chemistry Vol. 22, No. 10, pp. 2251-2257, 2003. Lloyd Ted. Quantitative Structure Activity Relationship Glossary. [Available Online]: http://www.chem.swin.edu.au/modules/mod4/glossq.html Accessed December 27, 2004. Loll P., Hesselsoe M., Moldrup P., Henriksen K., Larsen C., and Dahlstrom K. GAC Treatment of MTBE and Benzene Contaminated Groundwater. Submitted for The Fourth International Conference on Reme diation of Chlorinated and Recalcitrant Compounds. May 24-27, 2004, Monterey, CA. [Available Online]: http://www.dmr.as/download_get.php?file=31 Accessed December 27, 2004. Lua A.C. and Guo J. Activated Carbon Prep ared from Oil Palm Stone by One-Step CO2 Activation for Gaseous Pollutant Removal. Carbon Vol. 38, pp. 1089-1097, 2000. Mattson James S., Mark Harry B. Jr., Malbin Michael D., Weber Walter J. Jr., Crittenden John C. Surface Chemistry of Active Ca rbon: Specific Adsorption of Phenols. Journal of Colloid and Interface Science Vol. 31, No. 1, pp. 116-130, 1969. Mattson James S. and Mark Harry B. Jr. In frared Internal Reflectance Spectroscopic Determination of Surface Functional Groups on Carbon. Journal of Colloid and Interface Science Vol. 31, No. 1, pp. 131-144, 1969. Mattson James S., Lyon Lee, Mark Harry B. Jr ., Weber Walter J. Jr. Surface Oxides of Activated Carbon: Internal Reflectance Spectroscopic Examination of Activated Sugar Carbons. Journal of Colloid and Interface Science Vol. 33, No. 2, pp. 284293, 1970. Molina-Sabio M., Gonzalez M.T., Rodriguez-Re inoso F., Sepulveda-Escribano A. Effect of Steam and Carbon Dioxide Activation in the Micropore Size Distribution of Activated Carbon. Carbon Vol. 34, No. 4, pp. 505-509, 1996. MOPAC 2002. CAChe Software, Fujita, 2002. Muller Gunther, Radke C.J., Prausnitz J.M. Adsorption of Weak Electrolytes from Dilute Aqueous Solution onto Activated Carbon: Part I. Single-Solute Systems. Journal of Colloid and Interface Science Vol. 103, No. 2, pp. 466-483, 1985. National Science Foundation (NSF). Advanci ng the Quality of Water (AWQA): Expert Workshop to Formulate a Research Ag enda. March 10-12, 2004. [Available Online]: http://enve.coe.drexel.e du/AQWA/report/report.htm Accessed December 27, 2004.

PAGE 79

69 Noll Kenneth, E. Fundamentals of Air Quality System s: Design of Air Pollution Control Devises Annapolis, MD: American Academy of Environmental Engineers, 1999. Ouellette Robert J. Organic Chemistry: A Brief Introduction, Second Edition. Upper Saddle River, New Jersey: Prentice Hall, pp. 138-170 and 217-242, 1998. Pacific Northwest National Laboratory (PNNL). Artificial Neural Networks. Battelle Memorial Institute. Available [Online]: http://www.emsl.pnl.gov:2080/pr oj/neuron/neural/what.html Accessed December 2003. Perry R.H. and Green D.W. Perry's Chemical Engineers' Handbook: Seventh Edition. New York, New York: McGraw-Hill, 1997. Petrov N., Gergova K., Eser S. Effect of Water Vapour on the Porous Structure of Activated Carbon from Lignite. Fuel Vol. 73, No. 7, pp. 1197-1201, 1994. Petrucci R.H. and Harwood W.S. General Chemistry: Principles and Modern Applications, Seventh Edition. Upper Saddle Rive r, New Jersey: Prentice Hall, pp. 387-417, 1997. Pope Jason P. Activated Carbon and Some Ap plications for the Remediation of Soil and Groundwater Pollution. Available [Online]: http://www.cee.vt.edu/program_areas/envi ronmental/teach/gwprimer/group23/web page.htm Accessed December 6, 2003. Radovic Ljubisa. Chemistry and Physics of Carbon: Volume 27 New York, New York: Marcel Dekker, pp. 312-405, 2001. Rodriguez-Reinoso F., Molina-Sabio M., G onzalez M.T. The Use of Steam and CO2 as Activating Agents in the Prepar ation of Activated Carbons. Carbon Vol. 33, No.1, pp. 15-23, 1995. Schwartzenbach, R.P., Gschwend, P.M., and Imboden, D.M. Environmental Organic Chemistry, Second Edition. Hoboken, New Jersey: Wiley-Interscience, 2003. Snoeyink, Vernon L. Adsorp tion of Organic Compounds. Water Quality and Treatment: A Handbook of Community Water Supplies New York, New York: American Water Works Association, 1990. Tendankara Library. Chemis try: Delocalized Electrons in the Benzene Pi System. Available [Online]: http://library.tedankara.k12.tr/chemist ry/vol3/Delocalised%2 0molecular%20orbials %20(Part%201)/z104.htm Accessed 6 December 2003. Tennant M.F. and Mazyck D.W. Steam-pyrolys is activation of wood char for superior odorant removal. Carbon Vol. 41 No. 12, pp. 2195-2202, 2003.

PAGE 80

70 Tomkow Kazimierz, Siemieniewska Teresa Czechowski Franciszek, Jankowska Anna. Formation of Porous Structures in Activated Brown-Coal Chars Using O2, CO2, and H2O as Activating Agents. Fuel Vol. 56, pp. 121-124, 1977. United States Environmental Protection Ag ency (USEPA). Pollution Prevention and Toxics: Nitrobenzene Factshee t. [Available Online]: http://www.epa.gov/opptintr/chemfact/nitro-sd.txt Accessed December 27, 2004. United States Environmental Protection Ag ency (USEPA). Ground Water & Drinking Water. Technical Factsheet on Be nzene. [Available Online]: http://www.epa.gov/OGWDW/d wh/c-voc/benzene.html Accessed December 27, 2004. Vidic Radisav D., Suldan Makram T., Bre nner Richard C. Oxidative Coupling of Phenols on Activated Carbon: Impact on Adsorption Equilibrium. Environmental Science and Technology Vol. 27, No. 10, pp. 2079-2085, 1993. Weber Walter J. Jr. and DiGiano Francis A. Process Dynamics in Environmental Systems New York, New York: Environmen tal Science and Technology: A Wiley Interscience Series of Te xts and Monographs, 1996. Whelton Andrew J. Temperature Effects on Drinking Water Odor Perception. Master Thesis. Virginia Polytechnic In stitute and State University, 2001. Wigmans T. Industrial Aspects of Pr oduction and Use of Activated Carbons. Carbon Vol. 27, No. 1, pp. 13-22, 1989. Wold Svante and Dunn William J. III. Mul tivariate Quantitative Structure-Activity Relationships (QSAR): Conditions for Their Applicability. Journal of Chemical Information and Computer Science Vol. 23, pp. 6-13, 1983. Wu Jufang, Stromqvist Marian ne E., Claesson Ola, Fangmar k Ingrid E., Hammarstrom, Lars-Gunnar. A Systematic Approach for Modeling the Affinity Coefficient in the Dubinin-Radushkevich Equation. Carbon Vol. 40, pp. 2587-2596, 2002.

PAGE 81

71 BIOGRAPHICAL SKETCH Born and raised in Pensacola, Florida, Jennifer graduated with an International Baccalaureate Diploma from the Internationa l Baccalaureate Program at Pensacola High School in May 1997. Continuing her education at the University of Florida, she defended for highest honors and received a Bachelor of Science degree in environmental engineering sciences in May 2002. For he r masterÂ’s work, Jennifer studied under Dr. David Mazyck with a focus on the adsorption of benzenes and monosubstituted benzenes onto activated carbon for water treatment appli cations. Jennifer currently resides in Newberry, Florida, with her husband St eve and in August 2004 she joined the water/wastewater engineering department in the consulti ng firm of Jones Edmunds & Associates.