Professional Documents
Culture Documents
Solvent Effect Modeling in Heterogenous Catalysis
Solvent Effect Modeling in Heterogenous Catalysis
Scholar Commons
Summer 2021
Recommended Citation
Zare, M.(2021). Solvent Effect Modeling in Heterogenous Catalysis. (Doctoral dissertation). Retrieved
from https://scholarcommons.sc.edu/etd/6474
This Open Access Dissertation is brought to you by Scholar Commons. It has been accepted for inclusion in
Theses and Dissertations by an authorized administrator of Scholar Commons. For more information, please
contact digres@mailbox.sc.edu.
SOLVENT EFFECT MODELING IN HETEROGENOUS CATALYSIS
by
Mehdi Zare
Bachelor of Science
Shiraz University, 2011
Master of Science
University of Tehran, 2013
Chemical Engineering
2021
Accepted by:
Tracey L. Weldon, Interim Vice Provost and Dean of the Graduate School
© Copyright by Mehdi Zare, 2021
All Rights Reserved.
ii
DEDICATION
To my wife, Mozhdeh Parizad.
“You are the best mercy of God I have ever received. I found love, tranquility and image
of God in you.”
iii
ACKNOWLEDGEMENTS
I wish to thank my committee members, Dr. Chen, Dr. Regalbuto, Dr. Monnier,
and Dr. Moss for their time and help. I would like to offer my special thanks to Dr.
Mohammad Saleheen, and Adam Yonge for their contribution to my thesis with their
knowledge and insightful comments and suggestions. I am thankful to all the Heyden Lab
group members, Dr. Salai Ammal, Dr. Osman Mamun, Dr. Yongjie Xi, Dr. Biplab
Rajbanshi, Dr. Wenqiang Yang, Charles Fricke, Kyung-Eun You, Subrata Kundu, Dia
Sahsah, Nicholas Szaro, and Olajide Bamidele for being great coworkers and friends. I am
Hardcastle, Vernon Dorrell, Brian Loggans, and Shawn Hagan for assisting me with
administrative and technical issues. Finally, I would like to thank my family for their
iv
ABSTRACT
In recent years, the biorefining industry and biofuels have emerged as a major
American energy sector. Biofuels are fuels produced from plant and animal material, also
referred to as biomass. This includes wood products, manure, and corn, among other
friendly and thus pose less of a threat to environmental health. In 2019, the United States
consumed around 14.54 billion gallons of ethanol and around 1.81 billion gallons of
biodiesel. By 2030, the United States is expected to consume around 95 Mtoe of biofuels.
In order to meet current demand and expected growth in demand, it is necessary to advance
feedstock (highly reactive and water soluble, aqueous, and thermally unstable) demands
liquid-phase processing technologies for higher product selectivity and lower cost. As a
result, to design an efficient liquid-phase process, it is critical to understand the root causes
investigating reactions occurring at solid-liquid interfaces are less developed; primarily due
The objectives of this study are to develop and validate a hierarchy of multi-scale
methods for computing reaction and activation free energies of elementary processes
v
occurring at metal-solvent interfaces and to apply these methods to the rational design of
novel heterogeneous catalysts with exceptional activity and selectivity for the liquid-phase
the solvent effect, we studied the aqueous-phase effect on the initial C-H and O-H bond
derived polyols) over the (111) facet of six transition metal surfaces (Ni, Pd, Pt, Cu, Ag,
Au) using our explicit solvation method, eSMS. We found a significant metal dependence
on aqueous solvation effects that can be traced back to a different amount of charge-transfer
between the adsorbed species and metals in the reaction and transition states for the
different metal surfaces. In addition, we developed a new hybrid QM/MM approach for
computing solvent effects on the free energy of adsorption and desorption processes that
Pt(111) surface indicates that adsorption free energies in the aqueous phase can be
vi
TABLE OF CONTENTS
Acknowledgements ............................................................................................................ iv
Abstract ................................................................................................................................v
vii
LIST OF TABLES
viii
facet of six transition metals at 423 K. QM/MM-FEP
indicates a free energy calculation in water between the
critical points identified by gas-phase calculations
(using the gas-phase vibrational partition function for
the reactant and transition states). QM/MM-MFEP-
OPT represents a free energy calculation in water for
the different cleavages between the respective reactant
and transition states that have been optimized in an
aqueous-phase environment. Here, the vibrational
partition functions are computed in the aqueous phase
assuming the timescale for reorientation of the solvent
molecules is much larger than the timescale for
molecular vibrations. For comparison, implicit
solvation calculations have also been performed using
both nonperiodic (iSMS) and periodic (VASPsol)
approaches. All numbers are in eV ...........................................................................65
ix
Table 4.2 Contributions to aqueous-phase effect on free
energy (in eV) of low coverage desorption
𝑙𝑖𝑞→𝑔𝑎𝑠 𝑔𝑎𝑠→𝑙𝑖𝑞
(∆∆𝐺𝐴𝑑𝑠𝑜𝑟𝑏𝑎𝑡𝑒 = −∆∆𝐺𝐴𝑑𝑠𝑜𝑟𝑏𝑎𝑡𝑒 ) of ethylene glycol at
423 K (see Figure 4.2c), of carbon monoxide at 298 K
(see Figure 4.2b), and of phenol at 298 K (see Figure
4.2d) over the (111) facets of Pt and Cu catalysts. All
numbers also contain a 95% confidence interval based
on limited water sampling obtained by performing
multiple independent simulations ...........................................................................107
x
energy of the corresponding transition state in the
presence and absence of water ................................................................................123
xi
K in the presence of two different solvents, water and
1,4-dioxane, using iSMS methodology. ΔΔGrxn
indicates the difference in the adsorption free energy
of the corresponding intermediates in the presence and
absence of solvent. Asterisk (*) represents a surface
adsorption site and multiple asterisks are indicative of
the number of occupied active sites. Note that
calculations for solvents were performed with the help
of the COSMO-RS package with three different Ni
cavity radii: with default value, with a 10% increased
value and a 10% decreased value relative to the default
.................................................................................................................................131
xii
𝑛
pressure/fugacity of component i and 𝑋𝑟𝑐 indicates
Campbell’s degree of rate control for reaction step n .............................................136
Table B.1 Simulation box size used for (111) facet of each
metal surface ...........................................................................................................143
xiii
linear correlation. ∆𝐺 𝑎𝑐𝑡,𝑔𝑎𝑠 shows the free energy of
activation in vapor phase, H-bond denotes the change
in mean of total hydrogen bonding (acceptor + donor)
going from RS to TS, MC (molecular charge-transfer)
represents the change in the absolute sum of partial
charges on the reacting moiety going from RS to TS,
and finally BC (cleaving-bond charge-transfer)
expresses the change in sum of partial charges on the
cleaving bond going from RS to TS (see Table 3.2 in
the main text for the value of descriptors) ..............................................................148
xiv
metal surfaces and ~68% of them are in “parallel”
orientation, ~23% in H-up orientation, and only 9% in
H-down orientation (which lets the water hardly form
hydrogen bonds with the condensed phase)............................................................151
xv
transition between an atom present on the surface and
filling the cavity after the molecule is annihilated. Ow
and Hw represent the oxygen and hydrogen of the
TIP3P water model, respectively. Ho in the phenol
molecule indicates the hydrogen of the OH functional
group .......................................................................................................................177
xvi
LIST OF FIGURES
xvii
partition functions. See Table 3.1 for corresponding
data that include vibrational contributions. The number
of intermediate states between the reactant and
transition states for eSMS calculations is determined
by our desire to have an energy difference between
intermediates smaller than twice the thermal energy (<
2𝑘𝐵 𝑇). The aqueous-phase profiles portray the average
of three or more independent eSMS calculations
possessing 95% confidence intervals smaller than
±0.05 eV (see Table 3.1). The analogous plot for C-H
bond cleavage reaction is provided in the Appendix B ............................................69
xviii
the QM/MM system. In other words “gree circle”
symbolizes the adsorbate being only present in the
classical DL_POLY simulation and not in the quantum
system. Van der Waals interactions between the
solvent and the adsorbate are slowly removed between
state III and IV. The gas-phase desorption is described
between state IV and V. The (free) energy difference
between state IV and I is the solvent effect on the
stability of the adsorbate .........................................................................................109
xix
approximation), 𝑛 represents the number of sites a
phenol molecule occupies when adsorbed on the
𝑙𝑖𝑞
surface, 𝛼 (𝛼𝑃ℎ,𝑃ℎ ) is a constant associated to the
phenol-phenol lateral interactions on the surface in
𝑔𝑎𝑠→𝑙𝑖𝑞
water, and ∆∆𝐺 (∆∆𝐺𝑃ℎ (𝜃𝑃ℎ = 0)) is the
aqueous-phase effect on the low coverage phenol
adsorption on the surface. SSR is the sum of squared
residuals, which shows the error associated with each
fit .............................................................................................................................111
xx
(top right), and coadsorbed CO and PAC on a Ni (111)
catalyst surface model. In the coadsorbed state, CO is
a Lewis acid, which accepts electrons, while PAC is a
Lewis base, which donates electrons. This coadsorbed
Lewis acid and base pair explains the attractive
interaction between adsorbed CO and PAC............................................................140
xxi
z-coordinate of the H and O atom is larger than half the
H-H distance of the water molecule; this corresponds
for TIP3P water to an angle q show above larger than
52.26⁰. Water molecules that are not considered H-up
or H-down are labeled “parallel” ............................................................................154
xxii
surface from Singh et al. for two different regions: low
and high phenol concentrations. The “Const”
parameter represents the hydrogen surface coverage
relative to the free site coverage (const=0 shows zero
H coverage approximation), 𝑛 represents the number
of sites a phenol molecule occupies when adsorbed on
𝑙𝑖𝑞
the surface, 𝛼 (𝛼𝑃ℎ,𝑃ℎ ) is a constant associated to
phenol-phenol lateral interactions on the surface in
𝑔𝑎𝑠→𝑙𝑖𝑞
water, and ∆∆𝐺 (∆∆𝐺𝑃ℎ (𝜃𝑃ℎ = 0)) is the solvent
effect on the aqueous-phase low coverage phenol
adsorption on the surface. SSR is the sum of squared
residuals, which shows the error associated with each
fit. Fits with higher 𝑛 predict the lower concentration
region better, while fits with smaller n predict the
higher concentration region better. However, the fits
𝑔𝑎𝑠→𝑙𝑖𝑞
suggest that ∆∆𝐺𝑃ℎ (𝜃𝑃ℎ = 0) is quite
independent of these two fit regions .......................................................................183
xxiii
Figure C.9 Fits of equation 4.9 in the main text (adsorption
isotherm) to the experimental data of aqueous-phase
phenol adsorption on a Pt(111) catalyst surface at 314
K from Singh et al.62 See the caption of Figure C.7 for
detailed explanations ...............................................................................................186
xxiv
CHAPTER 1
INTRODUCTION
The critical role of solvents in catalytic transformations occurring at a solid-liquid
interfaces has given rise to an extensive range of studies aimed at understanding and
predicting the role of solvents. However, the inherent complexity of a reaction containing
both a complex heterogenous catalyst and a condensed phase has resulted in only a few
systematic experimental studies. In this context, theoretical calculations have the advantage
of being able to systematically investigate the root causes of such effect. There are
generally two major types of techniques to capture solvent effects computationally: implicit
(continuum) and explicit solvation methods. Implicit methods have the advantage of
computing free energies of reactions rapidly, but their reliability has often been questioned
because of their inability to capture the anisotropic site-specific interactions between the
solute and the solvent molecules. On the other hand, describing such system quantum
mechanically is limited to a few hundred atoms and a time scale of tens or a few hundred
The objective of my research is to reveal how a solvent can affect the energetics of a
catalytic reaction using implicit and QM/MM solvation approaches and to test the accuracy
1
This dissertation has been written following the manuscript style formatting,
first publication (chapter 2), which was published in July 2020 in The Journal of Physical
Ni(111) catalyst surface model in both vacuum and solvent environments with the help of
periodic DFT calculations coupled with our implicit solvation scheme for metal surfaces
network. Next, a microkinetic reaction model (MKM) was developed that permits
and 1,4-dioxane as solvents that are typically used protic and aprotic polar solvents. We
found that under all conditions the DCN is favored over the DCX. The dominant pathway
in gas phase starts with the removal of the hydroxyl group (OH) of PAC, followed by two
the dominant pathway in the liquid phases shifts after first step (removal of OH group, i.e.,
C-OH bond cleavage of PAC). It proceeds with a dehydrogenation step to form a CH3CCO
sensitivity analysis we found C-OH bond cleavage of PAC is the rate controlling step in
gas phase, while its dominance will be replaced with reaction step 9, where 𝛼-carbon of
CH3CHCO undergoes dehydrogenation. Finally, our computational results indicate that the
(111) facet of Ni is likely not the active facet for the HDO of propionic acid in both phases;
hence, other facets need to be investigated to identify the experimentally relevant active
site.
2
Chapter 3 of my dissertation contains my second publication, which has been
study, we investigated the aqueous-phase effects on the initial C–H and O–H bond
cleavages of ethylene glycol (EG) over the (111) facet of six transition metal surfaces (Ni,
Pd, Pt, Cu, Ag, Au) using our explicit solvation method, eSMS, to gain an understanding
very different aqueous-phase effect on the activation free energy barriers across different
metal, indicating that our hypothesis that the nature of metal plays a key role for solvation
effects on surface reactions has been confirmed. Next, to understand the origin of this
including the difference in charge-transfer effects in the reactant and transition states, the
change of hydrogen bonding going from reactant to transition state, and the gas-phase free
energy of activation. The correlation results suggest that the aqueous-phase effect on the
free energy of activation of O–H and C–H bond cleavages of EG over the investigated
metal surfaces originates primarily from some form of a charge-transfer effect (more
charge transfer in the TS relative to the RS leads to more stabilization) and partly on another
descriptor that could be related to hydrogen bonding and gas-phase activation barrier.
Finally, we compared the eSMS result against two implicit solvation models, iSMS and
VASPsol and found that the implicit solvation models failed to capture the full solvent
stabilization during the O–H bond cleavage of EG. For the C–H bond cleavage, implicit
and explicit solvation models anticipate comparable aqueous-phase effects for the metals
3
In my third publication in chapter 4, we developed a hybrid QM/MM methodology
verified the applicability and precision of this scheme by computing aqueous-phase effect
‡ ‡ ‡
on the free energy activation barrier (∆∆𝐺𝑙𝑖𝑞 = ∆𝐺𝑙𝑖𝑞 − ∆𝐺𝑔𝑎𝑠 ) of O-H bond cleavage of
ethylene glycol (EG) over a Cu(111) surface at 423 K with two different approaches, eSMS
‡,𝑒𝑆𝑀𝑆
methodology for surface reactions (∆∆𝐺𝑙𝑖𝑞 = −0.56 ± 0.04 𝑒𝑉) and the adsorption
‡
scheme proposed here (∆∆𝐺𝑙𝑖𝑞 = −0.50 ± 0.06 𝑒𝑉). We then applied this scheme to
compute solvent effect on adsorption of EG, carbon monoxide (CO), and phenol over (111)
facet of Cu and Pt. Our computation indicates that the aqueous phase has destabilized all
the three adsorbed species and has impacted phenol the most while CO the least
electrostatic, van der Waals, hydrogen bonding, and size of the adsorbate. Finally, to
confirm the accuracy of this proposed scheme, we computed the solvent effect on
𝑔𝑎𝑠→𝑙𝑖𝑞
adsorption free energy of phenol at low coverage (∆∆𝐺𝑃ℎ (𝜃𝑃ℎ = 0)) over Pt(111) at
experimental study of phenol adsorption on Pt (111) at 298 K within an error of less than
0.10 eV.
4
CHAPTER 2
CATALYST MODEL
Zare, M.; Solomon R. V.; Yang, W.; Yonge, A.; Heyden, A. The Journal of Physical
Chemistry C, 2020, 124, 16488-16500.
5
2.1 Abstract
The effect of two solvents, liquid water and 1,4-dioxane, has been studied from first
effects were observed on the decarbonylation rate, a substantial solvent stabilization of two
Furthermore, a significant solvent stabilization of the transition state of C-H bond cleavage
of the 𝛼-carbon of CH3CHCO, relative to the stabilization of the C-C bond cleavage of the
environments. Finally, a sensitivity analysis shows that the C-OH bond cleavage of
propionic acid and C-C bond cleavage of the 𝛼-carbon of CH3CHCO are the most rate
CH3CHCO becomes the most influential step. This shift in rate controlling state is
in aqueous phase. Overall, it is likely that the investigated (111) facet of Ni is not active
for the hydrodeoxygenation of propionic acid in neither the gas nor liquid phase and other
6
2.2 Introduction
originating from the release of carbon dioxide during fossil fuel combustion have drawn
increased attention toward utilization of alternative fuel sources and technologies such as
promising routes for upgrading pyrolysis bio-oils for producing liquid transportation fuels.2
The hydrodeoxygenation of organic acids, present in e.g. pyrolysis oils, is often found to
be a rate controlling HDO process such that there is a desire to design supported transition
metal catalysts that can efficiently convert the organic acids into alcohols and alkanes. The
HDO catalysts must have good activity, be of low cost, and reasonably stable against coke
formation. Some studies have shown Ni-based catalysts to be promising candidates for the
HDO of bio-oils to fuels.3-6 For example, Bykova et al.3 tested a series of Ni-based catalysts
with different stabilizing components for the HDO of guaiacol as a bio-oil model
compound. They found Ni-based catalysts prepared by the sol-gel method and stabilized
with SiO2 and ZrO2 are highly active, and the high activity of these catalysts correlates
with nickel loading. In another study, Yin et al.5 used a bimetallic Ni-Cu catalyst with high
Ni loading (up to 50%) for catalytic hydrotreatment of fast pyrolysis liquids. Their results
show a low rate of undesired gas and coke/char formation when using a high Ni loading.
Similarly, in Ardiyanti et al.’s work4 the catalyst with the highest Ni loading (58 wt % Ni)
promoted with Pd (0.7 wt %) was the most active, yielding oil products with improved
properties such as low oxygen content and a lower tendency for coke formation.
Another factor playing an important role in the HDO of organic acids is the solvent
environment. Experimental studies use various solvents, yet how and why the solvent alters
7
the reaction mechanism remains unknown. For example, Hoelderich et al.7 reported that in
the catalytic deoxygenation of oleic acid (C18) over Pd/C, the presence of water can change
the selectivity towards C17 hydrocarbons by up to 20%. Also, Wan et al.8 discovered that
when water is replaced with n-heptane at otherwise similar conditions, the esterification
formation of ethyl acetate. Although the solvent effect in a heterogenous catalyst can
qualitatively be explained by the impact of solvent polarity, more work remains to be done
In the present study, periodic DFT calculations coupled with our implicit solvation
scheme for metal surfaces (iSMS)13 are used for the investigation of the HDO of propionic
acid (PAC) over a Ni(111) catalyst surface model in both vacuum and solvent
considered in the reaction network bases on our prior studies of the HDO of PAC on various
elementary reactions investigated in this study. Next, a microkinetic reaction model was
developed under various reaction conditions that permits quantification of solvent effects
apparent activation barriers. This study specifically focuses on water and 1,4-dioxane as
solvents that are typically used protic and aprotic polar solvents. Finally, it is noted that the
lateral interaction effect of high surface coverage species such as adsorbed hydrogen and
8
CO are considered explicitly in our microkinetic models for all surface intermediates and
transition states.
2.3 Methods
All calculations for the catalyst slab model were performed using the DFT
implantation in the Vienna Ab Initio Simulation Package (VASP)19-20 code. The periodic
surface model was constructed using the optimized lattice constant of fcc-Ni bulk, 3.518
and its periodic images. Each Ni layer contained 12 Ni atoms with (3 × 2√3) periodicity,
the bottom two layers were fixed, while the top two layers, as well as the adsorbates, were
relaxed.
grid using Methfessel-Paxton smearing (σ=0.2 eV) with an energy cut off for plane waves
of 400 eV and a convergence criterion of a self-consistent field (SCF) of 1.0 × 10-7 eV.
The projector-augmented wave method (PAW) was utilized to describe the electron–ion
interactions along with the GGA-PW9121-22 functional for describing exchange and
correlation effects in the energy calculations. Transition state (TS) structures are optimized
using the NEB23 and Dimer methods.24-25 We note that all reported adsorption energies,
TURBOMOLE 7.0 program package.26-28 The cluster models were obtained by removal of
the periodic boundary conditions from the periodic slabs that were constructed from
9
previous plane-wave (VASP) calculations. A cluster model consisting of two layers and 51
Ni atoms was initially used for all surface reactions. For adsorption reactions that are
somewhat more sensitive to the size of the cluster, a 3-layer model containing 86 Ni atoms
was used. A convergence test of the size of the cluster is presented in the Appendix A
(Figure A.1). All adsorbates and metal atoms were represented by all-electron TZVP29-31
basis sets. The Coulomb potential was approximated with the RI-J approximation using
auxiliary basis sets.32-34 Single point energy calculations were performed with a self-
consistent field (SCF) energy convergence criterion of 1.0 × 10-7 Hartree. Energy
calculations for different spin states were carried out to identify the lowest energy spin
state. Afterwards, COSMO calculations for the cluster models in the liquid phase were
performed on the lowest energy spin state configuration at the same level of theory. The
dielectric constant was set to infinity as required for the Conductor-like Screening Model
for Real Solvents (COSMO-RS)35-36 calculations. Default cavity radii as well as 10%
increased and decreased cavity radii were used for the Ni atoms. We note that the cavity
radius is the most important solvent parameter in COSMO-RS that can, however, not be
accurately determined for Ni due to insufficient experimental data, explaining why are
Solvent effects presented in this work were investigated using the implicit solvation
model for solid surfaces (iSMS) method.13 The Gibbs free energy of an intermediate
𝑙𝑖𝑞𝑢𝑖𝑑
adsorbed on the surface in a liquid phase environment, 𝐺𝑠𝑢𝑟𝑓𝑎𝑐𝑒+𝑖𝑛𝑡𝑒𝑟𝑚𝑒𝑑𝑖𝑎𝑡𝑒 , is calculated
as:
10
𝑙𝑖𝑞𝑢𝑖𝑑 𝑣𝑎𝑐𝑢𝑢𝑚 𝑙𝑖𝑞𝑢𝑖𝑑
𝐺𝑠𝑢𝑟𝑓𝑎𝑐𝑒+𝑖𝑛𝑡𝑒𝑟𝑚𝑒𝑑𝑖𝑎𝑡𝑒 = 𝐺𝑠𝑢𝑟𝑓𝑎𝑐𝑒+𝑖𝑛𝑡𝑒𝑟𝑚𝑒𝑑𝑖𝑎𝑡𝑒 + (𝐺𝑐𝑙𝑢𝑠𝑡𝑒𝑟+𝑖𝑛𝑡𝑒𝑟𝑚𝑒𝑑𝑖𝑎𝑡𝑒 −
𝑣𝑎𝑐𝑢𝑢𝑚
𝐸𝑐𝑙𝑢𝑠𝑡𝑒𝑟+𝑖𝑛𝑡𝑒𝑟𝑚𝑒𝑑𝑖𝑎𝑡𝑒 ) (2.1)
𝑣𝑎𝑐𝑢𝑢𝑚
where 𝐺𝑠𝑢𝑟𝑓𝑎𝑐𝑒+𝑖𝑛𝑡𝑒𝑟𝑚𝑒𝑑𝑖𝑎𝑡𝑒 is the free energy in the absence of a solvent, computed here
within the harmonic approximation using plane-wave DFT calculations for periodic slab
𝑙𝑖𝑞𝑢𝑖𝑑
models, 𝐺𝑐𝑙𝑢𝑠𝑡𝑒𝑟+𝑖𝑛𝑡𝑒𝑟𝑚𝑒𝑑𝑖𝑎𝑡𝑒 is the free energy (without vibrational contributions) of the
surface species when the surface cluster model is immersed in an implicit solvent (which
is obtained by extracting selected metal atoms and removing the periodic boundary
𝑙𝑖𝑞𝑢𝑖𝑑
conditions). We note that 𝐺𝑐𝑙𝑢𝑠𝑡𝑒𝑟+𝑖𝑛𝑡𝑒𝑟𝑚𝑒𝑑𝑖𝑎𝑡𝑒 does not contain vibrational contributions
𝑣𝑎𝑐𝑢𝑢𝑚
of the adsorbate that are already considered in the first term. Finally, 𝐸𝑐𝑙𝑢𝑠𝑡𝑒𝑟+𝑖𝑛𝑡𝑒𝑟𝑚𝑒𝑑𝑖𝑎𝑡𝑒
is the DFT energy of the same cluster in the absence of the solvent. To compute the
𝑙𝑖𝑞𝑢𝑖𝑑
𝐺𝑐𝑙𝑢𝑠𝑡𝑒𝑟+𝑖𝑛𝑡𝑒𝑟𝑚𝑒𝑑𝑖𝑎𝑡𝑒 term, COSMO-RS calculations are performed using the
In addition to implicit solvation calculations performed with the iSMS method, implicit
solvation calculations were also performed at 473 K using VASPsol38-39 with a relative
permittivity of water of 34.82 at reaction conditions.40 We used the default values for the
cutoff charge density, 𝑛𝐶 , and for the width of the diffuse cavity, 𝜎.38 We also employed
the default effective surface tension parameter, 𝜏, for describing the cavitation, dispersion,
and repulsive interaction between the solute and the solvent that are not captured by the
electrostatic terms.38 While the surface tension parameter is likely most accurate only for
simulations at 298 K and not at 473 K, it is an optimized parameter of the solvent model
that cannot easily be obtained at other temperatures. Due to the absence of a large number
of experimental solvation data at 473 K, we decided that the default parameter is likely the
11
most meaningful choice. All other computational details for periodic implicit solvation
calculations were kept the same as in our periodic vapor phase calculations. We note that
For surface reactions, the forward rate constant (𝑘𝑓𝑜𝑟 ) of each reaction was
calculated as:
‡
𝑘𝐵 𝑇 −∆𝐺
𝑘𝑓𝑜𝑟 = 𝑒 𝑘𝐵 𝑇 (2.2)
ℎ
where 𝑘𝐵 is the Boltzmann constant, T represents the reaction temperature, h is the Planck
constant, and ∆𝐺 ‡ denotes the activation free energy at the corresponding reaction
‡
condition. In the solvated environment, the activation free energy (∆𝐺𝑠𝑜𝑙𝑣𝑒𝑛𝑡 ) and the
𝑟𝑥𝑛
reaction free energy (∆𝐺𝑠𝑜𝑙𝑣𝑒𝑛𝑡 ) are calculated as:
‡ ‡ 𝑇𝑆 𝐼𝑆
∆𝐺𝑠𝑜𝑙𝑣𝑒𝑛𝑡 = ∆𝐺𝐺𝑎𝑠 + [𝐺𝑠𝑜𝑙𝑣𝑒𝑛𝑡 − 𝐺𝑠𝑜𝑙𝑣𝑒𝑛𝑡 ] (2.3)
𝑟𝑥𝑛 𝑟𝑥𝑛 𝐹𝑆 𝐼𝑆
∆𝐺𝑠𝑜𝑙𝑣𝑒𝑛𝑡 = ∆𝐺𝐺𝑎𝑠 + [𝐺𝑠𝑜𝑙𝑣𝑒𝑛𝑡 − 𝐺𝑠𝑜𝑙𝑣𝑒𝑛𝑡 ] (2.4)
𝐼𝑆 𝑇𝑆 𝐹𝑆
where 𝐺𝑠𝑜𝑙𝑣𝑒𝑛𝑡 , 𝐺𝑠𝑜𝑙𝑣𝑒𝑛𝑡 , and 𝐺𝑠𝑜𝑙𝑣𝑒𝑛𝑡 represent the solvation free energies of initial,
‡ 𝑟𝑥𝑛
transition, and final states, respectively, and ∆𝐺𝐺𝑎𝑠 and ∆𝐺𝐺𝑎𝑠 are the respective activation
and reaction free energies under gas phase conditions. The reverse rate constants (𝑘𝑟𝑒𝑣 ) are
𝑘𝑓𝑜𝑟
𝑘𝑟𝑒𝑣 = (2.5)
𝐾
12
For an adsorption reaction, 𝐴(𝑔) + ∗ → 𝐴∗ , the adsorption rate is given by collision
1
𝑘𝑓𝑜𝑟 = (2.6)
𝑁0 √2𝜋𝑚𝐴 𝑘𝐵 𝑇
where 𝑚𝐴 is the molecular weight of the adsorbent 𝐴 and 𝑁0 denotes the number of sites
per unit area (1.866 × 1019 𝑚−2). While the use of collision theory with a sticking
controlling in our models such that the choice of sticking coefficient has likely no effect
on the reported results. Also, Zhang et al.43 recently showed that collision theory is a
reasonable approximation for adsorption processes from a liquid as long as mass transfer
−∆𝐺 𝑎𝑑𝑠
𝐾= 𝑒 𝑘𝐵 𝑇 (2.7)
where ∆𝐺 𝑎𝑑𝑠 is the adsorption free energy which is obtained in the presence of a solvent
as,
𝑎𝑑𝑠 𝑎𝑑𝑠 𝐴 ∗ ∗
∆𝐺𝑠𝑜𝑙𝑣𝑒𝑛𝑡 = ∆𝐺𝐺𝑎𝑠 + [𝐺𝑠𝑜𝑙𝑣𝑒𝑛𝑡 − 𝐺𝑠𝑜𝑙𝑣𝑒𝑛𝑡 ] (2.8)
𝑎𝑑𝑠 𝐴 ∗ ∗
where ∆𝐺𝐺𝑎𝑠 is the adsorption free energy in the gas phase and 𝐺𝑠𝑜𝑙𝑣𝑒𝑛𝑡 and 𝐺𝑠𝑜𝑙𝑣𝑒𝑛𝑡 are
the solvation free energies of the adsorbent 𝐴 on the surface and a clean surface immersed
in the solvent, respectively. With the forward and reverse rate constant calculated, all
nonlinear steady-state reactor equations are solved simultaneously under realistic reaction
13
2.3.4 Lateral interaction effects
Preliminary results of the microkinetic model indicate that CO and H are the most
abundant surface species. Hence, a method which considers the lateral interaction effects
of CO and H on the stability of each adsorbed species and transition state in the
Step 1: choose a reasonable coverage of high coverage species and perform DFT
calculations for all intermediates in the presence of the abundant species. We define surface
coverage of an adsorbate as the ratio of the number of sites it occupies to total number of
(12 atoms) leads to 0.25 ML surface coverage of H or CO, respectively, since H occupies
1 site while CO occupies 3 sites (see below in this section for a detailed discussion).
where 𝐺𝑖 (𝜃𝑗 ) represents the free energy of species 𝑖 as a function of surface coverage of
abundant species, 𝜃𝑗 , 𝐺𝑖 (0) is the free energy of species 𝑖 on the clean surface without
considering the lateral interaction effect, and 𝑎𝑖𝑗 is the correction factor of species 𝑖 in the
presence of the abundant species 𝑗 which can be acquired from equation (2.9) when the
𝐺𝑖 (𝜃𝑗 ) − 𝐺𝑖 (0)
𝑎𝑖𝑗 = (2.10)
𝜃𝑗
where 𝐺𝑖 (𝜃𝑗 ) is the free energy of species 𝑖 in the presence of only abundant species 𝑗.
14
Step 3: compute analogous to equation (2.9) for each transition state the free energy
𝑇𝑆
𝐺𝑛𝑇𝑆 (𝜃𝑗 ) = 𝐺𝑛𝑇𝑆 (0) + ∑ 𝑎𝑛𝑗 𝜃𝑗 (2.11)
𝑗
where superscript 𝑇𝑆 indicates the transition state, subscript 𝑛 represents the reaction
𝑇𝑆
number and 𝑎𝑛𝑗 is the adsorption correction factor for the lateral interaction effect of the
transition states in the presence of various adsorbates, we assumed that the transition states
𝑇𝑆
are affected as half the reactant and half the product state and hence, the coefficient 𝑎𝑛𝑗 is
estimated as
𝑇𝑆 1 𝑛
𝑎𝑛𝑗 = ∑ 𝑎𝑖𝑗 (2.12)
2
𝑖
𝑛
where 𝑎𝑖𝑗 represents the adsorption correction factor for the lateral interaction effect of the
𝑟𝑥𝑛
∆𝐺𝑛𝑟𝑥𝑛 (𝜃𝑗 ) = ∆𝐺𝑛𝑟𝑥𝑛 (0) + ∑ ∆𝑎𝑛𝑗 𝜃𝑗 ,
𝑗
𝑟𝑥𝑛
∆𝑎𝑛𝑗 = (𝑎𝐴∗𝑗 + 𝑎𝐵∗𝑗 − 𝑎𝐴𝐵∗𝑗 − 𝑎∗𝑗 ) (2.13)
15
𝑎𝑑𝑠 𝑎𝑑𝑠
∆𝐺𝐴𝑎𝑑𝑠 (𝜃𝑗 ) = ∆𝐺𝐴𝑎𝑑𝑠 (0) + ∑ ∆𝑎𝐴𝑗 𝜃𝑗 , ∆𝑎𝐴𝑗 = (𝑎𝐴∗𝑗 − 𝑎∗𝑗 ) (2.15)
𝑗
where ∆𝐺𝑛𝑟𝑥𝑛 (𝜃𝑗 ) and ∆𝐺𝑛‡ (𝜃𝑗 ) are the reaction and activation free energies of reaction 𝑛,
respectively, and ∆𝐺𝐴𝑎𝑑𝑠 (𝜃𝑗 ) is the coverage dependent adsorption free energy of species
𝐴. Similarly, ∆𝐺𝑛𝑟𝑥𝑛 (0) and ∆𝐺𝑛‡ (0) are the reaction and activation free energies of reaction
𝑛, respectively, and ∆𝐺𝐴𝑎𝑑𝑠 (0) is the adsorption free energy of species 𝐴 in the dilute limit.
interactions, show that the most abundant surface species are CO and H. Thus, we
computed the adsorption energy of all surface species in the presence of 0.25 ML H and
𝑟𝑥𝑛 𝑟𝑥𝑛
∆𝐺𝑛𝑟𝑥𝑛 (𝜃𝐻 , 𝜃𝐶𝑂 ) = ∆𝐺𝑛𝑟𝑥𝑛 (0) + [∆𝑎𝑛𝐻 × 𝜃𝐻 ] + [∆𝑎𝑛𝐶𝑂 × 𝜃𝐶𝑂 ] (2.16)
Figures A.2 and A.3 illustrate the correction terms for individual intermediates and
transition states, respectively. We note that we use a mean-field microkinetic model in this
study despite significant lateral interactions that can even be attractive, such as in the case
of CO and adsorbed PAC. This attractive interaction can be rationalized by the fact that the
interaction between a Lewis acid and Lewis base coadsorbed on the surface is surprisingly
strong.45 Charges on adsorbed CO and PAC, separately, on the surface, and on coadsorbed
CO and PAC are presented in the Appendix A (Figure A.4). Strong attractive lateral
16
Figure A.5 displays the differential zero-point corrected energy of adding an extra
increase in energy such that practically only four CO molecules fit on this surface while up
to 12 hydrogen atoms fit. Thus, it can be concluded that on Ni (111) CO and H occupy 3
and 1 sites, respectively. We used analogue procedures to count occupied site(s) for other
surface species that are bonded through carbon atoms to the surface. Given that the
Appendix A (Tables A.6 and A.7) all simulation results for the case that CO occupies a
single site.
facilitate certain types of cleavages. Hence, understanding the impact of the solvent
environment on each individual intermediate could shed light on how a solvent could
reshape the reaction mechanism. The change in the adsorption strength of different
intermediates can be quantified by considering the adsorption processes in the absence and
17
where, ∆𝐺𝑎𝑑𝑠,𝐴 (𝑙𝑖𝑞) and ∆𝐺𝑎𝑑𝑠,𝐴 (𝑔) are the adsorption free energies of a “gas” molecule
𝐴 in the liquid (or solvent) and in the vacuum, respectively. 𝐺 𝐴∗ (𝑙𝑖𝑞) and 𝐺 𝐴∗ (𝑔) are the
free energies of 𝐴 adsorbed on the surface in liquid and in vacuum, respectively, and
𝐺 ∗ (𝑙𝑖𝑞) and 𝐺 ∗ (𝑔) are the free energies of the clean surface in liquid and in vacuum
environments, respectively.
Table 2.1 displays the change in the adsorption free energy of each adsorbed
intermediate in the presence and absence of the solvent molecules using the iSMS method.
The most stabilized adsorbed intermediates are CH3CCOO and CH3CHCOO. In water,
their adsorption strengths are increased by 0.55 and 0.31 eV, respectively, due to adsorption
through the carbon atom and the charged carboxyl group pointing in the direction of the
solvent. This trend can also be seen in the presence of 1,4-dioxane with a slightly lower
In general, the solvent effect on the adsorption strength is pertinent to the strength
CH3CCOO and CH3CHCOO moieties in water, the negatively charged carboxyl group has
a strong electrostatic interaction with water molecules that accounts for the stabilization of
ascribed to the fact that 1,4-dioxane is a non-polar solvent and hence incapable of strong
Moreover, Table A.1 shows the aqueous phase effect on the adsorption energy of
intermediates at two different temperatures of 298 K and 473 K based on two implicit
18
solvation schemes: iSMS and VASPsol. The choice of room temperature is due to the fact
that all VASPsol parameters have been optimized based on experimental result at room
temperature. This means VASPsol results are most accurate at room temperature. The only
water. The data suggest that both iSMS and VASPsol predict (de)stabilization of adsorbed
By changing the stability of various surface intermediates and transition states, the
solvent environment plays an important role in altering the overall reaction pathway.
Therefore, to elucidate the solvent effect on each elementary reaction, the reaction and
activation free energies of all elementary steps in vacuum and in solvents are computed at
Propionic acid is consumed directly through three main reactions, which produce
𝑤𝑎𝑡𝑒𝑟 ) 1,4−𝑑𝑖𝑜𝑥𝑎𝑛𝑒
PAC in step 1 (∆(∆𝐺𝑟𝑥𝑛 = 0.00,∆(∆𝐺𝑟𝑥𝑛 ) = 0.03), CH3CHCOOH as the
𝑤𝑎𝑡𝑒𝑟 )
product of dehydrogenation of the 𝛼-carbon of PAC in step 2 (∆(∆𝐺𝑟𝑥𝑛 =
1,4−𝑑𝑖𝑜𝑥𝑎𝑛𝑒
−0.07,∆(∆𝐺𝑟𝑥𝑛 ) = −0.03), and propionate (CH3CH2COO) as the product of O-H
𝑤𝑎𝑡𝑒𝑟 ) 1,4−𝑑𝑖𝑜𝑥𝑎𝑛𝑒
bond dissociation of PAC in step 28 (∆(∆𝐺𝑟𝑥𝑛 = −0.03, ∆(∆𝐺𝑟𝑥𝑛 ) = 0.00).
Based on the solvent effect values on the reaction and activation free energies of these three
paths, it appears that the solvents do not impact the starting reactions of HDO of propionic
acid notably.
19
Furthermore, Table 2.2 illustrates a small change in the reaction and activation free
decarbonylation steps that produce CO, the reaction energies in water and 1,4-dioxane are
around 0.1 eV more exergonic than in the gas phase as a result of CO adsorption being
stabilized in water and 1,4-dioxane by 0.25 and 0.18 eV, respectively (Table 2.1). In
contrast to the small effect of the liquid phase environments on the DCN reactions, the
solvent effect is pronounced on the elementary reactions of the DCX. For convenience of
comparison of the solvent effect on important elementary reaction steps of the DCX, we
classified them into three different classes of similar types of bond dissociations.
exergonic solvent effect in water compared to in 1,4-dioxane originates from its polarity
that in turn stabilizes CH3CHCOO adsorption (the main product of this step) more in water
than in 1,4-dioxane. Another important dehydrogenation reaction occurs in step 34, where
CH3CHCOO, the second most stabilized intermediate in the liquid phases, is converted to
CH3CCOO, the most stabilized intermediate in the presence of the solvents. In particular,
a decline of 0.26 eV in the reaction free energy in water turns the endergonic reaction in
vacuum into an exergonic reaction in aqueous phase which facilitates the reaction
this step by 0.12 and 0.09 eV in water and 1,4-dioxane, respectively, makes this reaction
20
Class II: O-H bond cleavage: The O-H bond dissociation of CH3CHCOOH to
Class III: C-C bond cleavage to produce CO2: Production of CH3CH and CO2 by
C-C bond cleavage of CH3CHCOO (step 33) and formation of CH3C and CO2 from
CH3CCOO (step 38) are some of the most influenced DCX elementary reactions by a liquid
phase environment. The solvents have endergonic effects on the reaction and activation
free energies of these steps since the reactant states of these steps (CH3CHCOO and
Overall, both solvents altered the elementary reactions in the DCX mechanism
significantly while the solvent effect is small on the elementary reaction steps in the DCN
mechanism. In addition, the same calculations were performed using the VASPsol program
package and the results are compared to iSMS in Table A.3 and A.4 at 298 K and 473K.
In the case of the above-mentioned O-H bond cleavage reactions (step 31 and step 35),
VASPsol predicts endergonic solvent effects on activation barriers which contradicts both
To predict the overall TOF of our reaction network in both the presence and absence
partial pressures were set to 0.1 and 0.001 bar, respectively, and the partial pressure of
21
propionic acid, H2O (in the absence of liquid water), and CO2 were set to 1 bar. In liquid
water, the water chemical potential and partial pressure were computed assuming
where 𝑥𝐻2 𝑂 is the mole fraction of water in the liquid phase (assumed to be close
to 1), 𝑓𝐻𝐿2 𝑂 is the pure water fugacity at 473 K, 𝑦𝐻2 𝑂 is the water mole fraction in the vapor
phase, 𝑃𝑡𝑜𝑡 is the total pressure of the system, and 𝑃𝐻2 𝑂 indicates the water partial pressure
assuming an ideal gas phase. The pure water fugacity at 473 K was obtained from a steam
table and a Lee/Kesler generalized-correlation table,54 the partial pressure of water was
calculated to be 14.36 bar and this value was used in all liquid water microkinetic models.
In the following sections, we represent our microkinetic results both in the gas phase and
in the solvent environments and clarify the solvent influence on important parameters of
To determine the surface coverage, rate of each elementary surface reaction, and
the overall turnover frequency (defined here as the consumption of propionic acid), a
microkinetic model was developed taking the lateral interaction effects of the most
dominant surface species, CO and H, into account in both the gas phase and condensed
phase models. Table 2.3 summarizes the overall TOF, the decarboxylation and
decarbonylation rate, and the dominant species coverages during the HDO of propionic
acid over the Ni (111) catalyst surface in various reaction environments. In addition, the
TOF of each individual elementary reaction is listed in Table 2.4. Next, the dependence of
22
the overall vapor phase TOF on CO and H2 partial pressures is provided in Table A.5.
Finally, Figure 2.2 exhibits a schematic of the dominant pathways in the gas and liquid
phase environments.
The microkinetic model predicts that the dominant pathway begins with the
hydrogenation of CH3CH in step 25, and finally dehydrogenation of the 𝛽-carbon in step
27 to produce ethylene. The most abundant species on the surface are H and CO with
coverages of 0.631 and 0.357, respectively. Moreover, the free site coverage is 0.011. The
overall TOF, which includes both DCX and DCN, is determined to be 3.46 × 10−8 𝑠 −1.
The decarbonylation rate is 3.46 × 10−8 𝑠 −1 , while the decarboxylation rate is predicted
to be 8.72 × 10−13 𝑠 −1 . Hence, the selectivity towards DCN is close to 100% in the vapor
phase. We note that an experimental study of catalytic HDO of propionic acid over
supported group VIII noble metals performed by Lugo-Jose et al.55 found a TOF of
1.5 × 10−4 𝑠 −1 over a Ni/SiO2 catalyst at 473 K. Therefore, it is possible that the
investigated (111) facet of Ni is not the most relevant active site for this catalytic
transformation and other Ni facets or phases could be responsible for the experimentally
observed kinetics.
The dominant pathway in the liquid phases is the same as in the gas phase until
23
form the CH3CCO intermediate, a C-C bond cleavage in step 14 to produce CH3C, and
further hydrogenations through steps 24, 25, and the dehydrogenation of the 𝛽-carbon from
CH3CH2 in step 27 to produce ethylene. This shift in the dominant pathway is attributed to
the activation barriers of C-C bond dissociation of CH3CHCO (step 8) and C-H bond
cleavage of the 𝛼-carbon of CH3CHCO (step 9), which are 0.15 and 0.19 eV, respectively,
in the gas phase. We conclude that due to the lower activation barrier of step 8 relative to
step 9, step 8 is preferred in the gas phase. On the other hand, the activation barrier of step
8 is barely changed in the condensed phases while that of step 9 is decreased by 0.06 and
the condensed phases and hence forces the catalytic pathway toward the production of
CH3CHCO rather than CH3CH. As shown in Table 2.4, the overall TOF in the liquid
phases is very similar to the one in the gas phase and does not strongly depend on the cavity
radius of Ni. However, the abundant adsorbed intermediates change in liquid phase. The
the free site coverage decreases in the presence of the liquid phase. The overall effect on
the turnover frequency is not significant since the decarbonylation pathway is the dominant
HDO pathway and it is not strongly affected by the solvent. In contrast, the DCX pathways
are more affected by liquid water and the DCX rate is increased by approximately two
orders of magnitude in liquid water and one order of magnitude in liquid 1,4-dioxane.
Campbell’s degree of rate control,56-60 𝑋𝑅𝐶,𝑖 . This criterion describes which transition state
24
𝑘𝑖 𝜕𝑟
𝑋𝑅𝐶,𝑖 = ( ) (2.23)
𝑟 𝜕𝑘𝑖 𝐾 ,𝑘
𝑖 𝑗 ≠𝑘𝑖
where 𝑟 is the overall reaction rate, 𝑘𝑖 is the forward rate constant for step 𝑖, and 𝐾𝑖 is the
Next, the apparent activation barriers were computed in the temperature ranges of
𝜕ln (𝑟)
𝐸𝑎 = 𝑅𝑇 2 ( ) (2.24)
𝜕𝑇 𝑝𝑖
Figure 2.3 demonstrates the overall TOF as a function of inverse temperature in different
Finally, the reaction order with respect to hydrogen, CO and propionic acid were
calculated at 473 K and a pressure ranges of 0.9 to 1.1 bar for H2, 7.5 × 10−4 to
12.5 × 10−4 bar for CO, and 0.9 to 1.1 bar for PAC.
𝜕 ln(𝑟)
𝑎𝑖 = ( ) (2.25)
𝜕 ln(𝑝𝑖 ) 𝑇,𝑝
𝑗≠𝑖
At 473 K, microkinetic modeling results suggest that the initial C-OH bond
dissociation in step 1 is the most sensitive transition state in the gas phase with an 𝑋𝑅𝐶
value of 0.65. Another sensitive transition state belongs to step 8 (C-C bond cleavage of
𝛼-carbon of CH3CHCO), 𝑋𝑅𝐶 = 0.26. Also, step 9 (C-H bond cleavage of 𝛼-carbon of
CH3CHCO) with 𝑋𝑅𝐶 = 0.03 is the third most important step given that step 8 and 9 are
two competing steps in the reaction network. Finally, the apparent activation energy is
25
predicted to be 2.38 eV and the reaction rate is independent of CO partial pressure (𝑎𝐶𝑂 =
0), drops almost with a negative square of the hydrogen partial pressure (𝑎𝐻2 = −2.2), and
The overall turnover frequency is highly sensitive to the reaction barrier of steps 9
in the dominant pathway leading to step 9 becoming more favorable in the condensed
phase. The values of Campbell’s degree of rate control are 0.88 for step 9, 0.10 for step 1,
and 0.01 for step 8. The apparent activation energy obtained from the microkinetic model
is 2.71 eV. In contrast to the gas phase, the reaction rate decreases with CO partial pressure
(𝑎𝐶𝑂 = −0.7) in water. The reaction orders with respect to hydrogen and PAC remain
approximately the same as those in the gas phase (𝑎𝐻2 = −1.4, 𝑎𝑃𝐴𝐶 = 1.0).
In liquid 1,4-dioxane, at 473 K, steps 9 and 1 are the most sensitive transition states
with 𝑋𝑅𝐶 = 0.55 𝑎𝑛𝑑 𝑋𝑅𝐶 = 0.39, respectively, and step 8 has 𝑋𝑅𝐶 = 0.03. The decrease
in 𝑋𝑅𝐶 from 0.88 in water to 0.55 in 1,4-dioxane is related to a stronger stabilization of the
transition state of step 9 in liquid water compared to that in 1,4-dioxane. The apparent
activation barrier is 2.44 eV in the presence of 1,4-dioxane and the reaction orders are
almost similar to those in the liquid water environment (𝑎𝐶𝑂 = −0.4 𝑎𝐻2 = −1.7, 𝑎𝑃𝐴𝐶 =
1.0).
26
2.4.3.3 Comparison of implicit solvation models
While we have more confidence in our iSMS solvation model relative to VASPsol,
we also solved the microkinetic model using free energies computed with VASPsol in
(implicit) liquid water at 473 K. Table A.11 in the Appendix A provides a detailed
comparison between these two different solvent models. Overall, both solvation schemes
predict a similar solvent effect on the decarbonylation and decarboxylation of PAC. The
overall rate and the DCN rate are decreased in liquid water over Ni(111) while the DCX
rate is increased. While iSMS predicts a decrease of ~80% relative to the vapor phase for
the DCN rate, VASPsol only predicts a decrease of ~50%. Next, iSMS predicts an increase
in DCX rate by a factor 17 relative to the vapor phase, while VASPsol predicts an increase
by a factor 142. Nevertheless, the DCX hardly contributes to the overall rate in both
solvation schemes. The key differences between the solvation models are the surface
coverages of CO and H and a change in the rate controlling step being predicted by the
Table A.1), explaining the difference in surface coverage. Next, iSMS predicts the C-H
bond cleavage of CH3CHCO in step 9 to be the rate determining step in aqueous phase,
while VASPsol predicts the C-C bond cleavage of CH3CHCO in step 8 to be the rate
determining step. This difference in rate controlling step is a result of the opposite solvent
𝑎𝑐𝑡
effect predicted with iSMS and VASPsol on the activation barrier of step 9 ( ∆∆𝐺𝑖𝑆𝑀𝑆 =
𝑎𝑐𝑡
−0.06 𝑒𝑉, ∆∆𝐺𝑉𝐴𝑆𝑃𝑠𝑜𝑙 = 0.07 𝑒𝑉, see Table A.4). We note here that all iSMS calculations
with different cavity radius for Ni predict a solvent stabilization in water, leading us to
27
2.5 Conclusion
decarbonylation of propanoic acid over Ni(111) that considers the lateral interaction effect
of the most dominant surface species, CO and H. In addition, the effect of two solvents,
liquid water and 1,4-dioxane, has been investigated with the help of periodic DFT
microkinetic models were developed for each solvent at a temperature of 473 K and a
hydrogen partial pressure of 0.1 bar. Under all conditions, the decarbonylation is favored
over the decarboxylation. The dominant pathway in gas phase begins with the removal of
the hydroxyl group of propionic acid in step 1 (C-OH bond cleavage), followed by two
pathway in the condensed phases shifts after production of CH3CHCO. It continues with a
liquid water and 1,4-dioxane increase the decarboxylation rate by two orders of magnitude
and one order of magnitude, respectively, relative to the gas phase. This noticeable solvent
effect on the decarboxylation rate can be explained by the significant solvent stabilization
28
CH3CHCOO. Next, a sensitivity analysis shows that C-OH bond cleavage of propionic
acid is the most rate controlling step in the gas phase (𝑋𝑅𝐶1 = 0.65), In solvent
environments, its dominance will be replaced with reaction step 9 (C-H bond cleavage of
likely not the active facet for the HDO of propionic acid neither in gas phase nor the studied
solvents, and therefore other facets need to be investigated to identify the experimentally
2.6 Acknowledgements
Office of Basic Energy Science, Catalysis Science program under Award DE-SC0007167.
In addition, this work was partially supported by the South Carolina Smart State Center for
have been provided by the National Energy Research Scientific Computing Center
(NERSC) which is supported by the Office of Science of the U.S. Department of Energy
and in part by XSEDE under grant number TG-CTS090100. Computational resources from
the CASCADE cluster from the Environmental Molecular Sciences Laboratory (EMSL)
under Pacific Northwest National Laboratory (PNNL) are also used for selected DFT
calculations. Finally, computing resources from the USC High Performance Computing
29
2.7 Tables and Figures
Table 2.1: Solvent effect on the stability of various adsorbed species in the HDO of PAC
to ethane and ethylene over a Ni(111) catalyst surface model at a reaction temperature of
473 K. ΔΔGrxn indicates the difference in the adsorption free energy of the corresponding
intermediate in the presence and the absence of the solvent. Asterisk (*) represents a
surface adsorption site and multiple asterisks are indicative of the number of occupied
active sites.
ΔΔGrxn, eV
Adsorbed species
Water 1,4-Dioxane
CH2C*** -0.08 -0.05
CH2CH*** -0.01 0.00
CH2CH2** 0.10 0.12
CH2CHCO**** -0.09 -0.03
CH2CHCOOH**** -0.11 -0.04
CH3C*** -0.01 -0.01
CH3CCO**** -0.12 -0.06
CH3CCOO*** -0.55 -0.33
CH3CCOOH*** -0.12 -0.07
CH3CH*** 0.00 -0.01
CH3CH2** 0.01 -0.01
CH3CH2CO*** -0.05 -0.04
CH3CH2COO** -0.06 -0.04
CH3CH2COOH* -0.03 0.02
CH3CH3* 0.02 0.02
CH3CHCO** -0.09 -0.05
CH3CHCOO*** -0.31 -0.17
CH3CHCOOH** -0.10 -0.06
CHCH**** 0.05 0.08
CHCHCO**** -0.15 -0.05
CHCHCOOH**** -0.13 -0.06
CO*** -0.25 -0.18
CO2* -0.04 0.01
COOH** -0.16 -0.08
H* -0.01 0.00
H2O* 0.02 0.07
OH* 0.00 0.02
30
Table 2.2: Reaction and activation free energies of all elementary steps in the HDO of PAC to ethane and ethylene over a Ni(111)
catalyst surface model at 473 K. ΔΔGrxn and ΔΔGact indicate the reaction and the activation free energy differences between
corresponding reaction in the presence of solvent and in the gas phase, respectively.
Vacuum Water 1,4-Dioxane
# Reaction
ΔGrxn ΔGact ΔΔGrxn ΔΔGact ΔΔGrxn ΔΔGact
0 CH3CH2COOH + * → CH3CH2COOH* 0.69 -0.03 0.02
1 CH3CH2COOH* + 3* → CH3CH2CO*** + OH* -0.25 0.60 0.00 0.00 0.03 0.00
2 CH3CH2COOH* + 2* → CH3CHCOOH** + H* -0.05 0.51 -0.07 -0.03 -0.03 -0.01
3 CH3CH2CO***+ 2* → CH3CH2** + CO*** -0.61 0.78 -0.14 -0.03 -0.10 0.01
4 CH3CH2CO*** → CH3CHCO** + H* 0.19 0.49 -0.06 -0.03 -0.02 0.00
5 CH3CHCOOH** + * → CH3CHCO** + OH* -0.01 0.90 0.01 -0.03 0.04 0.00
6 CH3CHCOOH** + 3* → CH2CHCOOH**** + H* -0.42 0.57 -0.02 0.00 0.01 0.04
7 CH3CHCOOH** + 2*→CH3CCOOH*** + H* -0.19 0.35 -0.03 0.00 -0.01 0.01
8 CH3CHCO** + 4* → CH3CH*** + CO*** -1.14 0.15 -0.10 -0.01 -0.09 0.01
9 CH3CHCO** + 3* → CH3CCO**** + H* -0.73 0.19 -0.05 -0.06 -0.02 -0.03
31
Water 1,4-Dioxane
Properties Gas
Default +10% -10% Default +10% -10%
DCN TOF 3.46×10-08 6.70×10-09 4.07×10-08 4.64×10-09 1.84×10-08 1.23×10-07 1.15×10-08
DCX TOF 8.72×10-13 1.47×10-11 1.47×10-10 2.12×10-11 5.24×10-13 6.64×10-12 1.02×10-12
Overall TOF (s-1) 3.46×10-08 6.71×10-09 4.09×10-08 4.66×10-09 1.84×10-08 1.23×10-07 1.15×10-08
θ* 0.011 0.003 0.003 0.003 0.004 0.005 0.004
θH* 0.631 0.206 0.154 0.149 0.296 0.253 0.207
θCO*** 0.357 0.789 0.834 0.849 0.699 0.741 0.789
θCH3C*** 0.000 0.000 0.000 0.000 0.000 0.000 0.000
33
TOF (s-1)
# Reaction
Gas Water 1,4-dioxane
0 CH3CH2COOH + * → CH3CH2COOH* 3.45×10-08 6.69×10-09 1.84×10-08
1 CH3CH2COOH* + 3* → CH3CH2CO*** + OH* 3.46×10-08 6.70×10-09 1.84×10-08
2 CH3CH2COOH* + 2* → CH3CHCOOH** + H* 1.37×10-12 8.98×10-12 4.04×10-13
3 CH3CH2CO***+ 2* → CH3CH2** + CO*** 2.65×10-09 2.48×10-11 2.75×10-10
4 CH3CH2CO*** → CH3CHCO** + H* 3.19×10-08 6.67×10-09 1.82×10-08
5 CH3CHCOOH** + * → CH3CHCO** + OH* 5.26×10-13 4.04×10-14 2.85×10-14
6 CH3CHCOOH** + 3* → CH2CHCOOH**** + H* 2.47×10-17 2.92×10-15 2.53×10-16
7 CH3CHCOOH** + 2*→CH3CCOOH*** + H* 7.63×10-13 1.24×10-12 1.87×10-13
8 CH3CHCO** + 4* → CH3CH*** + CO*** 2.91×10-08 5.84×10-11 8.26×10-10
9 CH3CHCO** + 3* → CH3CCO**** + H* 2.75×10-09 6.61×10-09 1.73×10-08
10 CH3CHCO** + 3* → CH2CHCO**** + H* 8.35×10-13 8.27×10-14 4.01×10-13
34
36
Figure 2.2: Schematic representation of TOF (s −1 ), of various elementary reactions
involved in the dominant pathways. The left dashed box illustrates the dominant pathway
in the gas phase while the right blue drop displays the dominant pathways in the solvent
environments. Numbers in parentheses () depict reaction step numbers. The numbers in
square brackets [] in the right blue drop are the TOFs (s −1) in liquid water and the numbers
without square brackets are the TOFs (s−1 ) in liquid 1,4-dioxane.
37
Figure 2.3: Apparent activation energy plot in the temperature range of 473 to 623 K in gas
and condensed phase environments. The apparent activation barrier is 2.38 eV in the gas
phase, 2.71 eV in liquid water, and 2.44 eV in liquid 1,4-dioxane.
38
2.8 Bibliography
1. Simonetti, D. A.; Dumesic, J. A., Catalytic Production of Liquid Fuels from
Biomass‐Derived Oxygenated Hydrocarbons: Catalytic Coupling at Multiple
Length Scales. Catalysis Reviews 2009, 51 (3), 441-484.
4. Ardiyanti, A. R.; Bykova, M. V.; Khromova, S. A.; Yin, W.; Venderbosch, R. H.;
Yakovlev, V. A.; Heeres, H. J., Ni-Based Catalysts for the Hydrotreatment of Fast
Pyrolysis Oil. Energy & Fuels 2016, 30 (3), 1544-1554.
5. Yin, W.; Kloekhorst, A.; Venderbosch, R. H.; Bykova, M. V.; Khromova, S. A.;
Yakovlev, V. A.; Heeres, H. J., Catalytic hydrotreatment of fast pyrolysis liquids
in batch and continuous set-ups using a bimetallic Ni–Cu catalyst with a high
metal content. Catalysis Science & Technology 2016, 6 (15), 5899-5915.
7. Arend, M.; Nonnen, T.; Hoelderich, W. F.; Fischer, J.; Groos, J., Catalytic
deoxygenation of oleic acid in continuous gas flow for the production of diesel-
like hydrocarbons. Applied Catalysis A: General 2011, 399 (1), 198-204.
9. Saleheen, M.; Verma, A. M.; Mamun, O.; Lu, J.; Heyden, A., Investigation of
solvent effects on the hydrodeoxygenation of guaiacol over Ru catalysts.
Catalysis Science & Technology 2019, 9 (22), 6253-6273.
39
10. Saleheen, M.; Zare, M.; Faheem, M.; Heyden, A., Computational Investigation of
Aqueous Phase Effects on the Dehydrogenation and Dehydroxylation of Polyols
over Pt(111). The Journal of Physical Chemistry C 2019, 123 (31), 19052-19065.
11. Car, R.; Parrinello, M., Unified Approach for Molecular Dynamics and Density-
Functional Theory. Physical Review Letters 1985, 55 (22), 2471-2474.
12. Carloni, P.; Rothlisberger, U.; Parrinello, M., The Role and Perspective of Ab
Initio Molecular Dynamics in the Study of Biological Systems. Accounts of
Chemical Research 2002, 35 (6), 455-464.
13. Faheem, M.; Suthirakun, S.; Heyden, A., New Implicit Solvation Scheme for
Solid Surfaces. The Journal of Physical Chemistry C 2012, 116 (42), 22458-
22462.
14. Lu, J.; Behtash, S.; Heyden, A., Theoretical Investigation of the Reaction
Mechanism of the Decarboxylation and Decarbonylation of Propanoic Acid on
Pd(111) Model Surfaces. The Journal of Physical Chemistry C 2012, 116 (27),
14328-14341.
15. Behtash, S.; Lu, J.; Faheem, M.; Heyden, A., Solvent effects on the
hydrodeoxygenation of propanoic acid over Pd(111) model surfaces. Green
Chemistry 2014, 16 (2), 605-616.
16. Lu, J.; Faheem, M.; Behtash, S.; Heyden, A., Theoretical investigation of the
decarboxylation and decarbonylation mechanism of propanoic acid over a
Ru(0001) model surface. Journal of Catalysis 2015, 324, 14-24.
17. Yang, W.; Solomon, R. V.; Lu, J.; Mamun, O.; Bond, J. Q.; Heyden, A.,
Unraveling the mechanism of the hydrodeoxygenation of propionic acid over a Pt
(1 1 1) surface in vapor and liquid phases. Journal of Catalysis 2020, 381, 547-
560.
18. Behtash, S.; Lu, J.; Mamun, O.; Williams, C. T.; Monnier, J. R.; Heyden, A.,
Solvation Effects in the Hydrodeoxygenation of Propanoic Acid over a Model
Pd(211) Catalyst. The Journal of Physical Chemistry C 2016, 120 (5), 2724-2736.
19. Kresse, G.; Furthmüller, J., Efficiency of ab-initio total energy calculations for
metals and semiconductors using a plane-wave basis set. Computational
Materials Science 1996, 6 (1), 15-50.
40
20. Kresse, G.; Hafner, J., Ab initio molecular dynamics for liquid metals. Physical
Review B 1993, 47 (1), 558-561.
21. Perdew, J. P.; Wang, Y., Accurate and simple analytic representation of the
electron-gas correlation energy. Physical Review B 1992, 45 (23), 13244-13249.
22. Perdew, J. P.; Yue, W., Accurate and simple density functional for the electronic
exchange energy: Generalized gradient approximation. Physical Review B 1986,
33 (12), 8800-8802.
23. Henkelman, G.; Uberuaga, B. P.; Jónsson, H., A climbing image nudged elastic
band method for finding saddle points and minimum energy paths. The Journal of
Chemical Physics 2000, 113 (22), 9901-9904.
24. Henkelman, G.; Jónsson, H., A dimer method for finding saddle points on high
dimensional potential surfaces using only first derivatives. The Journal of
Chemical Physics 1999, 111 (15), 7010-7022.
25. Heyden, A.; Bell, A. T.; Keil, F. J., Efficient methods for finding transition states
in chemical reactions: Comparison of improved dimer method and partitioned
rational function optimization method. The Journal of Chemical Physics 2005,
123 (22), 224101.
26. Ahlrichs, R.; Bär, M.; Häser, M.; Horn, H.; Kölmel, C., Electronic structure
calculations on workstation computers: The program system turbomole. Chemical
Physics Letters 1989, 162 (3), 165-169.
27. Treutler, O.; Ahlrichs, R., Efficient molecular numerical integration schemes. The
Journal of Chemical Physics 1995, 102 (1), 346-354.
29. Weigend, F., Accurate Coulomb-fitting basis sets for H to Rn. Physical Chemistry
Chemical Physics 2006, 8 (9), 1057-1065.
30. Weigend, F.; Ahlrichs, R., Balanced basis sets of split valence, triple zeta valence
and quadruple zeta valence quality for H to Rn: Design and assessment of
accuracy. Physical Chemistry Chemical Physics 2005, 7 (18), 3297-3305.
41
31. Weigend, F.; Häser, M.; Patzelt, H.; Ahlrichs, R., RI-MP2: optimized auxiliary
basis sets and demonstration of efficiency. Chemical Physics Letters 1998, 294
(1), 143-152.
32. Eichkorn, K.; Treutler, O.; Öhm, H.; Häser, M.; Ahlrichs, R., Auxiliary basis sets
to approximate Coulomb potentials. Chemical Physics Letters 1995, 240 (4), 283-
290.
33. Eichkorn, K.; Weigend, F.; Treutler, O.; Ahlrichs, R., Auxiliary basis sets for
main row atoms and transition metals and their use to approximate Coulomb
potentials. Theoretical Chemistry Accounts 1997, 97 (1), 119-124.
34. Von Arnim, M.; Ahlrichs, R., Performance of parallel TURBOMOLE for density
functional calculations. Journal of Computational Chemistry 1998, 19 (15), 1746-
1757.
35. Klamt, A., Conductor-like Screening Model for Real Solvents: A New Approach
to the Quantitative Calculation of Solvation Phenomena. The Journal of Physical
Chemistry 1995, 99 (7), 2224-2235.
36. Klamt, A.; Jonas, V.; Bürger, T.; Lohrenz, J. C. W., Refinement and
Parametrization of COSMO-RS. The Journal of Physical Chemistry A 1998, 102
(26), 5074-5085.
38. Mathew, K.; Sundararaman, R.; Letchworth-Weaver, K.; Arias, T. A.; Hennig, R.
G., Implicit solvation model for density-functional study of nanocrystal surfaces
and reaction pathways. The Journal of Chemical Physics 2014, 140 (8), 084106.
39. Fishman, M.; Zhuang, H. L.; Mathew, K.; Dirschka, W.; Hennig, R. G., Accuracy
of exchange-correlation functionals and effect of solvation on the surface energy
of copper. Physical Review B 2013, 87 (24), 245402.
42
41. Lugo-José, Y. K.; Monnier, J. R.; Heyden, A.; Williams, C. T.,
Hydrodeoxygenation of propanoic acid over silica-supported palladium: effect of
metal particle size. Catalysis Science & Technology 2014, 4 (11), 3909-3916.
42. Lugo-José, Y. K.; Behtash, S.; Nicholson, M.; Monnier, J. R.; Heyden, A.;
Williams, C. T., Unraveling the mechanism of propanoic acid
hydrodeoxygenation on palladium using deuterium kinetic isotope effects.
Journal of Molecular Catalysis A: Chemical 2015, 406, 85-93.
43. Zhang, X.; Savara, A.; Getman, R. B., A Method for Obtaining Liquid–Solid
Adsorption Rates from Molecular Dynamics Simulations: Applied to Methanol on
Pt(111) in H2O. Journal of Chemical Theory and Computation 2020.
44. Demir, B.; Kropp, T.; Rivera-Dones, K. R.; Gilcher, E. B.; Huber, G. W.;
Mavrikakis, M.; Dumesic, J. A., A self-adjusting platinum surface for acetone
hydrogenation. Proceedings of the National Academy of Sciences 2020, 117 (7),
3446.
45. McFarland, E. W.; Metiu, H., Catalysis by Doped Oxides. Chemical Reviews
2013, 113 (6), 4391-4427.
46. I. Chorkendorff, J. W. N., Concepts of Modern Catalysis and Kinetics. third ed.;
Wiley-VCH, Weinheim: 2005.
47. Desai, S. K.; Neurock, M., First-principles study of the role of solvent in the
dissociation of water over a Pt-Ru alloy. Phys Rev B 2003, 68 (7), 075420:1-7.
48. Hibbitts, D. D.; Loveless, B. T.; Neurock, M.; Iglesia, E., Mechanistic Role of
Water on the Rate and Selectivity of Fischer-Tropsch Synthesis on Ruthenium
Catalysts. Angew. Chem. Int. Ed. Engl. 2013, 52 (47), 12273-12278.
49. Zhang, X. H.; Sewell, T. E.; Glatz, B.; Sarupria, S.; Getman, R. B., On the water
structure at hydrophobic interfaces and the roles of water on transition-metal
catalyzed reactions: A short review. Catal Today 2017, 285, 57-64.
50. Santana, J. A.; Mateo, J. J.; Ishikawa, Y., Electrochemical Hydrogen Oxidation on
Pt(110): A Combined Direct Molecular Dynamics/Density Functional Theory
Study. J Phys Chem C 2010, 114 (11), 4995-5002.
51. Skachkov, D.; Rao, C. V.; Ishikawa, Y., Combined First-Principles Molecular
Dynamics/Density Functional Theory Study of Ammonia Electrooxidation on
Pt(100) Electrode. J Phys Chem C 2013, 117 (48), 25451-25466.
43
52. Nie, X. W.; Luo, W. J.; Janik, M. J.; Asthagiri, A., Reaction mechanisms of CO2
electrochemical reduction on Cu(111) determined with density functional theory.
J Catal 2014, 312, 108-122.
53. Huang, Z. Q.; Long, B.; Chang, C. R., A theoretical study on the catalytic role of
water in methanol steam reforming on PdZn(111). Catal Sci Technol 2015, 5 (5),
2935-2944.
56. Campbell, C. T., Future Directions and Industrial Perspectives Micro- and macro-
kinetics: Their relationship in heterogeneous catalysis. Topics in Catalysis 1994, 1
(3), 353-366.
58. Stegelmann, C.; Andreasen, A.; Campbell, C. T., Degree of Rate Control: How
Much the Energies of Intermediates and Transition States Control Rates. Journal
of the American Chemical Society 2009, 131 (23), 8077-8082.
59. Kozuch, S.; Shaik, S., Kinetic-Quantum Chemical Model for Catalytic Cycles:
The Haber−Bosch Process and the Effect of Reagent Concentration. The Journal
of Physical Chemistry A 2008, 112 (26), 6032-6041.
60. Kozuch, S.; Shaik, S., A Combined Kinetic−Quantum Mechanical Model for
Assessment of Catalytic Cycles: Application to Cross-Coupling and Heck
Reactions. Journal of the American Chemical Society 2006, 128 (10), 3355-3365.
44
CHAPTER 3
SURFACE REACTIONS
Zare, M.; Saleheen, M.; Kundu, S. K.; Heyden, A. Communications Chemistry 3, 187,
2020.
45
3.1 Abstract
Solvent interactions with adsorbed moieties involved in surface reactions are often
believed to be similar for different metal surfaces. However, solvents alter the electronic
structures of surface atoms, which in turn affects their interaction with adsorbed moieties.
catalysis, we studied solvent effects on the activation free energies of the O-H and C-H
bond cleavages of ethylene glycol over the (111) facet of six transition metals (Ni, Pd, Pt,
Cu, Ag, Au) using an explicit solvation approach based on a hybrid quantum
A significant metal dependence on aqueous solvation effects was observed that suggests
solvation effects must be studied in detail for every reaction system. The main reason for
this dependence could be traced back to a different amount of charge-transfer between the
adsorbed moieties and metals in the reactant and transition states for the different metal
surfaces.
3.2 Introduction
The widespread use of solvents in applications varying from pharmaceutical1, 2, 3, 4,
5, 6
to electrochemistry7, 8, 9, 10 and catalysis11, 12, 13, 14, 15, 16 has given rise to an extensive
range of studies aimed at understanding and predicting the role of solvents. The concept
that a solvent can alter the performance of a catalyst, including its rate, selectivity, and
stability, is well known; yet predicting a specific solvation effect remains a challenge.17, 18,
19, 20, 21, 22, 23, 24
While impressive progress has been made in understanding the effects of
solvents in homogenous catalysis,25, 26, 27, 28, 29 for heterogeneously catalyzed processes that
46
processes,30 the role of solvents is hardly understood and only rarely studied. The inherent
condensed phase at a finite, often elevated, temperature has resulted in only few systematic
experimental (in situ and in operando)13, 15, 31, 32 and/or theoretical studies11, 12, 33, 34 of
is typically ascribed to: heightened importance of mass transfer effects, nature of solvent
(polarity etc.)35, 36, 37, competitive adsorption between solvent molecules and adsorbed
moieties,32, 38, 39
direct participation of the solvent in the reaction coordinate,40 and/or
selectivity, and overall catalyst lifetime. In short, understanding and predicting solvent
effects on surface reactions requires detailed investigations of the direct and indirect
interactions between the solvent, catalyst, and reacting moieties on the surface under
reaction conditions.
systemically study the effect of an individual parameter on the effect of a solvent. Ab initio
molecular dynamics (AIMD) simulations have been used;44, 45, 46 however, due to the great
computational cost associated with the quantum mechanical calculations and the large
amount of phase space sampling necessary, AIMD simulations are currently limited to
simulation systems of a few hundred atoms and a time scale of tens or a few hundred
picoseconds.10, 47, 48
47
An alternative approach is to use implicit solvation models.49, 50 While they can
compute free energies of reactions at solid-liquid interfaces rapidly, their reliability has
often been questioned14 because of their inability to capture the anisotropic site-specific
interactions between the solute and the solvent molecules. A compromise in efficiency and
approach.51, 52, 53 In this class of simulations, the adsorbate and metal atoms involved in the
reaction are considered as a QM sub-system described from first principles, while the bulk
of the solvent and metal atoms distant to the active site are considered as an MM sub-
system described using classical molecular mechanics force fields. We have previously
developed such a hybrid QM/MM model, named eSMS (Explicit Solvation Model for
Metal Surfaces),54 that considers the long-range electrostatic interaction of the solvent
molecules in the electronic structure calculation of the active site and applied it to the free
and selectivity, the interrelation of energetic changes on variation of catalyst surface, nature
of solvent, and reacting moiety must be disclosed. In this context, several studies have been
dedicated to understanding how the nature of the solvent and reacting moiety direct
catalysis.23, 26, 56, 57, 58 Nonetheless, the role of the metal identity (catalyst) on the solvent
effect has to our knowledge not been investigated yet. While it could be argued that
solvation effects should be similar for the same bond cleavages or for the same adsorbates
on different metal surfaces such as in recent studies by Greely et al.59, 60, we hypothesize
that the electronic structure modification of the metal surface and reacting moiety as a
48
result of the nearby solvent is sufficiently significant that solvation effects can differ
significantly for different metals. To confirm this hypothesis, we have investigated the
aqueous-phase effects on the initial C-H and O-H bond cleavages of EG over the (111)
facet of six transition metal surfaces (Ni, Pd, Pt, Cu, Ag, Au) using our explicit solvation
method, eSMS. The choice of reaction systems is motivated by (i) EG being a commonly
studied surrogate molecule of biomass-derived polyols, (ii) the selected transition metals
are relatively stable and commonly used for aqueous-phase processing of biomass-derived
oxygenates,61 (iii) early dehydrogenation steps of EG over Pt and Ni/Pt catalysts have
previously been found to control the overall reaction rate,62, 63, 64 (iv) at least over Pt(111)
in the vapor phase, initial C-H and O-H bond cleavage are competitive (although O-H bond
approaches have recently been used to demonstrate that for bond cleavage reactions of
alcohols over Pt(111), aqueous solvation effects are large and can currently not be
(see Table 3.1 for specific numbers). In addition, a graphical representation of the free-
energy (potential of mean force) profiles for the O-H bond cleavages is illustrated in Figure
3.2 and for the C-H bond cleavages in Figure B.1. We note that although some of the metals
(such as Ni) might get partially oxidized in liquid water environments with low reduction
potential, we chose to study the (111) facet of all metal surfaces for better comparison.
49
Generally, both O-H and C-H bond cleavages are somewhat facilitated in the presence of
water. For the O-H bond cleavage, Pt is the most active catalyst in both phases. However,
for the C-H bond cleavage, Ni is the most active catalyst in the presence of liquid water
because of strong aqueous-phase effect that is more than twice as large over Ni than Pt.
(∆𝐺 𝑎𝑐𝑡,𝑙𝑖𝑞 = 0.50 𝑒𝑉) as Ni (∆𝐺 𝑎𝑐𝑡,𝑙𝑖𝑞 = 0.55 𝑒𝑉) and Pd (∆𝐺 𝑎𝑐𝑡,𝑙𝑖𝑞 = 0.58 𝑒𝑉). In short,
evidenced by very different aqueous-phase effects on the activation free energy barriers
across different metals, our hypothesis that the nature of metal plays a key role for solvation
least a three-body problem involving solvent, catalyst, and reacting moiety. Thus, the most
from direct and indirect interactions of solvent, catalyst, and reacting moiety. To explain
the origin for the variability of aqueous-phase effects on the free energy of activation
(∆∆𝐺 𝑎𝑐𝑡 = ∆𝐺 𝑎𝑐𝑡,𝑙𝑖𝑞 − ∆𝐺 𝑎𝑐𝑡,𝑔𝑎𝑠 ) of O-H and C-H bond cleavages of EG across six
One possible descriptor that could explain the role of the metal identity, which leads
to different ∆∆𝐺 𝑎𝑐𝑡 values across different metals and bond cleavages, is the difference in
charge-transfer effects (and thus solvent stabilization) in the reactant (RS) and transition
states (TS) (see Table B.5). This charge transfer effect is attributed to the indirect influence
of the water environment causing an electronic modification of the metal atoms and the
effective charge distribution in the reacting moieties. We note that, using AIMD
50
simulations, Siemer et al.17 have recently shown that water induced local charge transfer is
interface sites. We choose two suitable descriptors for quantifying this charge-transfer
effect. One is the cleaving-bond charge transfer (BC) defined as the change in sum of
partial NPA66 charges on the cleaving bond going from RS to TS; for example in O-H bond
change in charge from, e.g., -0.1 to +0.1, +0.1 to +0.1, or -0.1 to -0.1, should all result in
no significant net stabilization since water stabilizes the RS and TS similarly. Another
descriptor describing a charge-transfer effect is the molecular charge transfer (MC) defined
here as the change in absolute sum of charges on the reacting moiety going from RS to TS,
i.e., this descriptor describes the change in charge transfer from the metal surface to the
Table 3.2 lists BC and MC for the O-H and C-H bond cleavages over the (111)
facet of six transition metal surfaces (see Table B.5 for partial charges). According to our
definition of BC and MC, smaller BC values (more negative) or larger MC values (more
transition state. For instance, in the C-H bond cleavage, the charge-transfer effect over
−0.03 𝑒 , 𝑀𝐶 = 0.08 𝑒). As shown in Table 3.2, ∆∆𝐺 𝑎𝑐𝑡 is directly related to the charge-
transfer effect descriptors in the C-H bond cleavage; that is, solvent effects increase (more
negative ∆∆𝐺 𝑎𝑐𝑡 ) with increasing charge-transfer effect (more negative BC or more
positive MC). In contrast, for O-H bond cleavage, there is no linear relationship between
∆∆𝐺 𝑎𝑐𝑡 and the charge-transfer effect. Nevertheless, after investigating other solvation
51
effect descriptors, we will show that the charge-transfer effect is likely also here a key
aqueous phase is hydrogen bonding.23, 41 Two main classes of hydrogen bonding definitions
commonly used in the literature are based on an energy criterion and a geometric definition.
Herein, we employed a geometric definition in which a hydrogen bond exists if the distance
between the donor oxygen (Od) and the acceptor oxygen (Oa), ROO, is less than 3.2 Å and
the angle ∠HOdOa is smaller than 20⁰67 (see Figure B.2). We note that EG with OH
functional groups can be either a donor or acceptor of hydrogen bonding (see Table B.4).
Hence, the change in mean of total hydrogen bonding (acceptor + donor) going from RS
to TS was chosen as a descriptor, named in the following H-bond, and is included in Table
3.2. Next, the gas-phase free energy of activation (∆𝐺 𝑎𝑐𝑡,𝑔𝑎𝑠 ), a rough measure of change
descriptor for describing the variability in solvation free energy effects across metal
surfaces. Given the absence of a linear ∆∆𝐺 𝑎𝑐𝑡 dependence on H-bond or ∆𝐺 𝑎𝑐𝑡,𝑔𝑎𝑠 , we
also studied all pairwise combinations of descriptors. Finally, we note here that we also
analyzed the water orientation (H-up, H-down, parallel – see Figure B.3) within the first
water layer of the different surfaces in the reactant state (see height distribution function of
water O in Figure B.4); however, we did not observe any significant variation in water
orientations across metals, explaining why we did not further study water orientation as a
descriptor (see Table B.10). Also, we attempted to use the standard electrode potential of
the metal elements as descriptor but again no meaningful correlation could be obtained
52
Next, we first examined the pairwise correlations between the descriptors using the
Pearson correlation coefficient (PCC).68 The results (see Table B.6) indicate that BC and
MC are totally correlated (PCC ~ -1.0) which is expected given that they both describe
charge transfer. In addition, H-bond is more correlated with MC and BC in the O-H bond
cleavage than in the C-H bond cleavage. We attribute this to the fact that hydrogen bonding
is obtained from the interaction of water molecules with the OH functional groups of EG,
and the O-H bond cleavage reaction causes a significant change in the charges on one of
the two O-H functional groups in EG in the TS. Finally, ∆𝐺 𝑎𝑐𝑡,𝑔𝑎𝑠 is hardly correlated with
Second, using a simple linear model (equation 3.1), we investigated the relation
between the descriptors and the aqueous-phase effects on the free energies of activation
(∆∆𝐺 𝑎𝑐𝑡 ). We emphasize that we do not intend to quantitatively predict ∆∆𝐺 𝑎𝑐𝑡 using a set
what physical phenomena could explain the changes in solvation effects on the kinetics of
̅̅̅̅̅̅̅̅̅
(∆∆𝐺 𝑎𝑐𝑡 − ∆∆𝐺 𝑎𝑐𝑡 ) ̅ ̅
𝑚𝑜𝑑𝑒𝑙 = 𝛼1 (𝑓1 − 𝑓1 ) + 𝛼2 (𝑓2 − 𝑓2 ) (3.1)
In equation 3.1, bar signs show the mean of the corresponding variable, 𝛼1 and 𝛼2
are model parameters, and 𝑓1 and 𝑓2 represent descriptors (see Table 3.2). The best-fitting
parameters and mean absolute error (𝑀𝐴𝐸) of the linear model for different combinations
of descriptors are listed in Table B.7. Interestingly, the linear model can estimate ∆∆𝐺 𝑎𝑐𝑡
very well for the C-H bond cleavage but not for the O-H bond cleavage. As expected, the
dominant factor in C-H bond cleavage is the charge-transfer effect; that is, 𝛼2 ≫ 𝛼1 when
one of charge-transfer effect descriptors (BC or MC) used as 𝑓2 (see Table B.7). Finally, a
53
quadratic model with two descriptors was employed to explain ∆∆𝐺 𝑎𝑐𝑡 of the O-H bond
cleavage.
̅̅̅̅̅̅̅̅̅
(∆∆𝐺 𝑎𝑐𝑡 − ∆∆𝐺 𝑎𝑐𝑡 ) ̅ ̅ ̅ 2 ̅ 2
𝑚𝑜𝑑𝑒𝑙 = 𝛼1 (𝑓1 − 𝑓1 ) + 𝛼2 (𝑓2 − 𝑓2 ) + 𝛼3 (𝑓1 − 𝑓1 ) + 𝛼4 (𝑓2 − 𝑓2 ) (3.2)
The results of this model for different combinations of descriptors are included in Table
B.8.
Overall, when comparing the model parameters of the linear and quadratic fits, we
conclude that the aqueous-phase effect on the free energy of activation of O-H and C-H
bond cleavages of EG over the investigated metal surfaces originates primarily from some
form of a charge-transfer effect (BC or MC) (more charge transfer in the TS relative to the
RS leads to more stabilization) and partly on another descriptor that could be related to H-
bond and ∆𝐺 𝑎𝑐𝑡,𝑔𝑎𝑠 (larger barriers are more stabilized). Figure 3.3 presents the computed
aqueous-phase effect by eSMS versus the estimated one from a linear fit for C-H bond
cleavage and a quadratic fit for O-H bond cleavage, in which BC and ∆𝐺 𝑎𝑐𝑡,𝑔𝑎𝑠 were used
as descriptors.
the full solvent stabilization during the O-H bond cleavage of EG (see Table 3.1). For the
C-H bond cleavage, implicit and explicit solvation models anticipate comparable aqueous-
phase effects for the metals that display a smaller charge transfer effect (Pd, Pt, Au). Table
B.9 illustrates that this observation holds even when considering typical uncertainties in
cavity radius of ±10% for transition metal elements. In short, the reliability of implicit
54
(unknown) due to the very limited availability of experimental data that can be used in the
advantage for explicit solvation models that rely “only” on a meaningful potential energy
description.
3.4 Conclusions
In summary, we hypothesized that since solvents can modify the electronic
structure of a metal surface, thereby affecting the stability of reacting moieties, the metal
investigated aqueous-phase effects on the free energy of activation of the initial O-H and
C-H bond cleavages of ethylene glycol over the (111) surface of six transition metals,
including Ni, Pd, Pt, Cu, Ag, and Au, to disclose the role of metal identity on solvation
effects. To compute the free energy of activation in the presence of water, we utilized our
the activation barrier (∆∆𝐺 𝑎𝑐𝑡 = ∆𝐺 𝑎𝑐𝑡,𝑙𝑖𝑞 − ∆𝐺 𝑎𝑐𝑡,𝑔𝑎𝑠 ) for the same bond cleavage across
different metals, suggesting that solvation effects have to be studied in detail for any
descriptors, the charge-transfer effect, hydrogen bonding, and the gas-phase activation free
energy barrier, and studying correlations between the aqueous-phase effects and these
descriptors, we can conclude that the aqueous-phase effects originate primarily from
various charge-transfer effects across the different metals, and to a lesser extent to another
descriptor that could be related to a different amount of hydrogen bonding and the gas-
55
phase activation barriers across the different metals. This observation agrees with previous
models are currently not able to capture these charge-transfer effects that can lead to
changes in preference of initial O-H versus C-H bond cleavage in alcohols over various
3.5 Methods
core, all-electron projector augmented-wave (PAW)74 method was utilized to avoid the
electrons considered for Ni, Pd, Pt, Cu, Ag, Au, C, O, and H are 10 ,10, 10, 11, 11, 11, 4,
electrons, i.e., the exchange-correlation functional, was expressed using the Perdew-Burke-
Ernzerhof (PBE)75, 76
functional within the semi-local generalized gradient
expanded using a discrete plane-wave basis set with kinetic energies limited to 400 eV. A
first order smearing method (Methfessel-Paxton)79 with 0.10 eV smearing width was
employed, allowing to accurately calculate the entropic contributions due to the smearing.
Dipole and quadrupole corrections (along the surface normal) to the total energy have been
calculated using a modified version of the Makov-Payne80 method, and Harris corrections,
56
stress-tensor and forces. A 4×4-unit cell with four layers of metal atoms (bottom two layers
fixed in their bulk positions) has been employed. By introducing a 15 Å vacuum on top of
the surface, the interaction between the periodic images along the surface normal has been
curtailed. A self-consistent field (SCF) convergence criterion for the electronic degrees of
freedom of the valence electrons was set to 1.0×10-7 eV. Transition state structures for the
band83, 84 and dimer85, 86 methods. Finally, the minima and the first order saddle points were
validated by computing the Hessian matrix and vibrational spectra. We note that spin-
TURBOMOLE 7.2 program package.87, 88, 89 To model the cluster surfaces, two layers of
metal atoms with a hexagonal shaped geometry (51 atoms) were chosen. The convergence
of the total QM/MM energy with respect to the lateral size and depth of the cluster
basis sets (def-bases) with split valence and polarization functions (def2-SVP)90, 91 were
employed to represent the adsorbate atoms and the metal atoms of Ni and Cu. Furthermore,
Ag, Pd, Pt, and Au atoms were represented using scalar relativistic effective core potentials
(ECPs) in conjunction with split valence basis sets augmented by polarization functions.91,
92
Electron exchange and correlation effects were accounted for by employing the PBE
approximation with auxiliary basis sets was used to approximate the coulomb integrals.93,
57
94
An SCF convergence criterion of 1.0 × 10−7 Hartree was established and a Gauss-
Chebyshev type spherical grid, m4, was employed to perform the numerical integrations.88
program package.95 The initial 4×4 unit cell for each metal surface was augmented laterally
to a 16×20 surface with further vacuum added in the Z-direction resulting in a simulation
box comprising of 1280 metal atoms. The simulation box size for each metal is reported in
Table B.1. The simulation box height was selected based on the work from Behler et al.67
finding that simulations of metal-water interfaces should contain a water layer of ~40 Å
height. The experimental saturated liquid water density of ~0.9 g/cm3 at 423 K was
achieved by packing the simulation box of Ni, Pd, Pt, Cu, Ag, and Au with 2258, 2758,
2800, 2398, 3045, and 2985 water molecules, respectively. All metal and adsorbate atoms
were kept fixed while the geometry of water molecules was constricted to that of TIP3P 96
geometry with the RATTLE algorithm97. To solve the Newton’s equations of motion, a
velocity version of the SHAKE algorithm,98 in conjunction with the velocity Verlet (VV)
integrator99 were used. The TIP3P model was employed for the force field parameters of
liquid water while the van der Waals parameters for adsorbate atoms were obtained from
parameters from the Combined B3LYP/6-31_G*/AMBER Potential102 were used for the
hydrogen atoms of the adsorbed moieties. Lennard-Jones parameters for hydrogen atoms
are important in QM/MM optimizations that permit hydrogen atoms to approach water
molecules and leave the protective environment of a neighboring carbon or oxygen atom.
58
interaction. The LJ cross-term of intermolecular parameters were calculated by Lorentz-
𝜎𝑖 +𝜎𝑗
Berthelot mixing rules through equations 𝜎𝑖𝑗 = and 𝜀𝑖𝑗 = √𝜀𝑖 𝜀𝑗 . All Lennard-Jones
2
parameters are included in Table B.2. The charges for the QM atoms were estimated using
the natural population analysis (NPA).66 To describe the interaction of the TIP3P water
point charges with the quantum chemically described cluster model, we employed the
TURBOMOLE. Simulations were carried out in a canonical ensemble (NVT) with Nosé-
Hoover thermostat.105, 106 A 1 ps relaxation time constant for temperature fluctuations was
used to maintain the average system temperature. Electrostatic interactions were accounted
for by using the Smoothed Particle Mesh Ewald (SPME) method107 with automatic
parameter optimization for default SPME precision and a 12 Å cutoff radius was adopted
for the van der Waals interactions and the transition between short and long range
electrostatic interactions. Unless specified otherwise, for each free energy perturbation
step, all systems were equilibrated for 250 ps and sampled for 1000 ps (1ns) using a 1 fs
2.9 ms were performed for this study. To optimize structures in an aqueous reaction
optimizing the intrinsic reaction coordinate on a potential of mean force (PMF) description
of the reaction system has been implemented in our program packages. A full description
of this methodology, eSMS (Explicit Solvation for Metal Surfaces) can be found
59
elsewhere.54. Briefly, the total energy function formulation of our eSMS method is given
by
where the first term is evaluated for a periodic slab using the VASP program
calculations), and the third term is a QM cluster calculation in a periodic mean field of
MM water molecules computed using the periodic electrostatic embedded cluster method
approximation has been validated for our eSMS approach54). We note that the number 100
in the equation indicates that 100 MM conformations, selected equally spaced from
equilibrated 1 ns molecular dynamic (MD) simulations (10 ps apart), were used to represent
the mean field of the MM water molecules. Finally, the last three terms account for the
classical (MM level of theory) electrostatic and van-der-Waals interaction energy of the
molecules with the QM cluster subsystem. We note that all solvent (water) molecules are
measurements of the system and ideally the energy estimator should also be capable of
minimizing the statistical bias and variance of the free energy differences of the physical
system being studied. Exponential averaging (EXP), also known as the Zwanzig
60
relationship108 has long been applied to study a variety of problems such as amino acid
etc. However, the EXP has been shown to represent poor efficiency and phase space
overlap,112, 113 and is also largely dependent on the distribution of the QM/MM energy.114
Here, we employed the Bennett Acceptance Ratio (BAR)115 as the free energy estimator
which uses both the forward and reverse distributions simultaneously in a more efficient
way than simply averaging the forward and reverse exponential estimators. BAR has been
demonstrated to benefit from a lower bias and variance of the free energy estimates in
(TI).112, 116 Finally, the whole free energy estimation procedure has been repeated three
times independently to establish 95% confidence intervals for evaluating the free energy
carried out only if these three experiments were not resulting in 95% confidence interval
smaller than 0.05 eV. All uncertainties reported in this study are 95% confidence intervals.
of the system is of great importance in any QM/MM approach for computing the liquid-
phase effect on the free energy of elementary processes114. Owing to a lack of consensus
on how much sampling of the configurational space is sufficient for a solvated adsorbed
carbohydrate species on a metal surface for an error smaller than 0.05 eV, we computed
the average rotational correlation time for water molecules in close proximity (up to 5 Å)
to adsorbed ethylene glycol in the reactant and transition states over the (111) facet of six
transition metals and reported them in Table B.3. The longest average correlation time is
61
computed to be ~200 ps. Hence, we decided to sample for 1000 ps to make sure that
relevant configurations of the systems are sampled adequately. As discussed above, for
free-energy calculations, the procedure was repeated at least three times with independent
MD trajectories to establish the confidence interval of the computation of the free energy
‡ ‡ 𝑇𝑆 𝐼𝑆
∆𝐺𝑠𝑜𝑙𝑣𝑒𝑛𝑡 = ∆𝐺𝐺𝑎𝑠 + [𝐺𝑠𝑜𝑙𝑣𝑒𝑛𝑡 − 𝐺𝑠𝑜𝑙𝑣𝑒𝑛𝑡 ] (3.4)
‡ 𝐼𝑆
where ∆𝐺𝐺𝑎𝑠 is the respective activation free energy under gas phase conditions, and 𝐺𝑠𝑜𝑙𝑣𝑒𝑛𝑡 and
𝑇𝑆
𝐺𝑠𝑜𝑙𝑣𝑒𝑛𝑡 represent the solvation free energies of initial and transition states, respectively, computed
as
𝑙𝑖𝑞𝑢𝑖𝑑 𝑣𝑎𝑐𝑢𝑢𝑚 𝑙𝑖𝑞𝑢𝑖𝑑
𝐺𝑠𝑢𝑟𝑓𝑎𝑐𝑒+𝑖𝑛𝑡𝑒𝑟𝑚𝑒𝑑𝑖𝑎𝑡𝑒 = 𝐺𝑠𝑢𝑟𝑓𝑎𝑐𝑒+𝑖𝑛𝑡𝑒𝑟𝑚𝑒𝑑𝑖𝑎𝑡𝑒 + (𝐺𝑐𝑙𝑢𝑠𝑡𝑒𝑟+𝑖𝑛𝑡𝑒𝑟𝑚𝑒𝑑𝑖𝑎𝑡𝑒 −
𝑣𝑎𝑐𝑢𝑢𝑚
𝐸𝑐𝑙𝑢𝑠𝑡𝑒𝑟+𝑖𝑛𝑡𝑒𝑟𝑚𝑒𝑑𝑖𝑎𝑡𝑒 ) (3.5)
𝑣𝑎𝑐𝑢𝑢𝑚
where 𝐺𝑠𝑢𝑟𝑓𝑎𝑐𝑒+𝑖𝑛𝑡𝑒𝑟𝑚𝑒𝑑𝑖𝑎𝑡𝑒 is the free energy of an intermediate (e.g. IS or TS) in the
absence of a solvent, computed here within the harmonic approximation using plane-wave
𝑙𝑖𝑞𝑢𝑖𝑑
DFT calculations for periodic slab models, 𝐺𝑐𝑙𝑢𝑠𝑡𝑒𝑟+𝑖𝑛𝑡𝑒𝑟𝑚𝑒𝑑𝑖𝑎𝑡𝑒 is the free energy of the
metal cluster and surface intermediate in liquid without vibrational contributions (that are
𝑣𝑎𝑐𝑢𝑢𝑚
already considered in the first term), and 𝐸𝑐𝑙𝑢𝑠𝑡𝑒𝑟+𝑖𝑛𝑡𝑒𝑟𝑚𝑒𝑑𝑖𝑎𝑡𝑒 is the DFT energy of the
𝑙𝑖𝑞𝑢𝑖𝑑
same cluster in the absence of the solvent. To compute the 𝐺𝑐𝑙𝑢𝑠𝑡𝑒𝑟+𝑖𝑛𝑡𝑒𝑟𝑚𝑒𝑑𝑖𝑎𝑡𝑒 term,
COSMO-RS118, 119 (Conductor-like Screening Model for Real Solvents) calculations are
performed using the COSMOtherm program with latest FINE parameterization.120 The
62
cavity radius for transition metal elements, calculations are performed with default and
10% increased and decreased cavity radius for the transition metal elements (see Table
B.9).
solvation calculations were also performed with VASPsol70, 71 using a relative permittivity
of water of 44.07 at 423 K.121 For VASPsol, we used the default values for the parameter
𝑛𝐶 that defines the value at which the dielectric cavity forms and for the width of the diffuse
cavity, 𝜎, and for effective surface tension parameter, 𝜏, describing the cavitation,
dispersion, and repulsive interaction between the solute and the solvent that are not
captured by the electrostatic terms.71 While the parameters are likely most accurate only
for simulations at 298 K and not at 423 K, they are the optimized parameters of the solvent
model that cannot easily be obtained at other temperatures. Due to the absence of adequate
experimental solvation data at 423 K, we decided that the default parameters are likely
most meaningful, i.e., the relative permittivity is the only temperature dependent solvent
parameter in our VASPsol model. All other computational details for periodic implicit
solvation calculations were kept the same as in our periodic vapor-phase calculations.
3.6 Acknowledgements
We gratefully acknowledge financial support from the U.S. Department of Energy,
Office of Basic Energy Science, Catalysis Science program under Award DE-SC0007167.
In addition, this work was partially supported by the South Carolina Smart State Center for
have been provided by the National Energy Research Scientific Computing Center
63
(NERSC) which is supported by the Office of Science of the U.S. Department of Energy
and in part by XSEDE under grant number TG-CTS090100. Computational resources from
the CASCADE cluster from the Environmental Molecular Sciences Laboratory (EMSL)
under Pacific Northwest National Laboratory (PNNL, Ringgold ID 130367, Grant Proposal
51163) are also used for selected DFT calculations. Finally, computing resources from the
64
3.7 Tables and Figures
65
VASPsol 1.65 1.81
iSMS 1.57 1.88
Ag(111)
QM/MM-FEP 1.22 ± 0.04 1.57 ± 0.03
QM/MM-MFEP-OPT 1.14 ± 0.04 1.56 ± 0.03
Vapor Phase 1.67 1.69
VASPsol 1.75 1.43
Au(111) iSMS 1.64 1.52
QM/MM-FEP 1.55 ± 0.00 1.43 ± 0.01
QM/MM-MFEP-OPT 1.51 ± 0.01 1.41 ± 0.03
66
Table 3.2: Average aqueous-phase effects on the activation free energy (∆∆𝐺 𝑎𝑐𝑡 =
∆𝐺 𝑎𝑐𝑡,𝑙𝑖𝑞 − ∆𝐺 𝑎𝑐𝑡,𝑔𝑎𝑠 ) as well as three solvation effect descriptors for the O-H and C-H
bond cleavages of ethylene glycol over the (111) surface facet of six transition metals at
423 K (see Table 3.1 for 95% confidence intervals). ∆𝐺 𝑎𝑐𝑡,𝑔𝑎𝑠 is the free energy of
activation in vapor phase, H-bond denotes the change in mean of total hydrogen bonding
(acceptor + donor) going from RS to TS, MC (molecular charge transfer) represents the
change in the absolute sum of partial charges on the reacting moiety going from RS to TS,
and finally BC (cleaving-bond charge transfer) is the change in sum of partial charges on
the cleaving bond atoms going from RS to TS.
∆𝑮𝒂𝒄𝒕,𝒈𝒂𝒔,
Cleavage Surface ∆∆𝑮𝒂𝒄𝒕 , eV H-bond MC, e BC, e
eV
Ni -0.21 0.70 0.99 0.64 -0.61
Pd -0.24 0.76 0.34 0.13 -0.36
Pt -0.45 0.68 0.54 -0.10 -0.24
O-H
Cu -0.55 1.08 0.77 0.69 -0.74
Ag -0.41 1.63 0.92 0.78 -0.77
Au -0.12 1.67 0.95 0.62 -0.63
Ni -0.34 0.79 0.26 0.41 -0.36
Pd -0.02 0.83 -0.26 -0.03 0.06
Pt -0.15 0.71 0.12 0.08 0.03
C-H
Cu -0.37 1.43 0.21 0.53 -0.46
Ag -0.34 1.91 -0.27 0.28 -0.32
Au -0.26 1.69 -0.38 0.06 -0.12
67
Figure 3.1: Aqueous-phase effects on the activation free energy barrier of the O-H and C-
H bond cleavages of ethylene glycol over the (111) surface facet of six transition metals at
423 K computed by eSMS (all structures optimized in liquid water; see Table 3.1 for
specific numbers).
68
Figure 3.2: Free-energy profiles for O-H bond cleavage of ethylene glycol in vapor and
aqueous phase over the (111) surface facets of six transition metals at 423 K without
considering vibrational contributions to the partition functions. See Table 3.1 for
corresponding data that include vibrational contributions. The number of intermediate
states between the reactant and transition states for eSMS calculations is determined by our
desire to have an energy difference between intermediates smaller than twice the thermal
energy (< 2𝑘𝐵 𝑇). The aqueous-phase profiles portray the average of three or more
independent eSMS calculations possessing 95% confidence intervals smaller than ±0.05
eV (see Table 3.1). The analogous plot for C-H bond cleavage reaction is provided in the
Appendix B.
69
(a) (b)
Figure 3.3: Parity plot of the aqueous-phase effect on the free energy of activation
(∆∆𝐺 𝑎𝑐𝑡 = ∆𝐺 𝑎𝑐𝑡,𝑙𝑖𝑞 − ∆𝐺 𝑎𝑐𝑡,𝑔𝑎𝑠 ) computed by eSMS (QM/MM-FEP in Table 3.1) versus
the model-predicted aqueous-phase effect. The predicted values are calculated using the
gas-phase free energy of activation (∆𝐺 𝑎𝑐𝑡,𝑔𝑎𝑠 ) and cleaving-bond charge transfer (BC) as
two descriptors in (a) a linear model (equation 3.1) for C-H bond cleavage and (b) a
quadratic model (equation 3.2) for O-H bond cleavage of ethylene glycol over (111)
̅̅̅̅̅̅̅̅̅
surface facets of six transition metals at 423 K. The average aqueous-phase effect, ∆∆𝐺 𝑎𝑐𝑡 ,
computed by eSMS, is -0.33 eV for O-H bond cleavage and -0.25 eV for C-H bond
cleavage. The mean of absolute errors (𝑀𝐴𝐸) between eSMS and model-predicted values
is 0.03 eV in the linear model (a) and 0.02 eV in the quadratic model (b). The corresponding
model parameters are listed in Tables B.7 and B.8.
70
3.8 Bibliography
5. Beilin E, Baker LJ, Aikins J, Baryla NE. Effect of incomplete removal of the tert-
butoxycarbonyl protecting group during synthesis of a pharmaceutical drug
substance on the residual solvent analysis. Journal of Pharmaceutical and
Biomedical Analysis 52, 316-319 (2010).
6. Mota FL, Carneiro AR, Queimada AJ, Pinho SP, Macedo EA. Temperature and
solvent effects in the solubility of some pharmaceutical compounds:
Measurements and modeling. European Journal of Pharmaceutical Sciences 37,
499-507 (2009).
7. Chen LD, Urushihara M, Chan KR, Norskov JK. Electric Field Effects in
Electrochemical CO2 Reduction. Acs Catal 6, 7133-7139 (2016).
71
10. Otani M, Hamada I, Sugino O, Morikawa Y, Okamoto Y, Ikeshoji T. Structure of
the water/platinum interface - a first principles simulation under bias potential.
Phys Chem Chem Phys 10, 3609-3612 (2008).
11. Zope BN, Hibbitts DD, Neurock M, Davis RJ. Reactivity of the Gold/Water
Interface During Selective Oxidation Catalysis. Science 330, 74-78 (2010).
12. Yoon Y, Rousseau R, Weber RS, Mei DH, Lercher JA. First-Principles Study of
Phenol Hydrogenation on Pt and Ni Catalysts in Aqueous Phase. J Am Chem Soc
136, 10287-10298 (2014).
13. Tupy SA, et al. Correlating Ethylene Glycol Reforming Activity with In Situ
EXAFS Detection of Ni Segregation in Supported NiPt Bimetallic Catalysts. Acs
Catal 2, 2290-2296 (2012).
15. Karim AM, et al. In Situ X-ray Absorption Fine Structure Studies on the Effect of
pH on Pt Electronic Density during Aqueous Phase Reforming of Glycerol. Acs
Catal 2, 2387-2394 (2012).
16. Miller KL, Lee CW, Falconer JL, Medlin JW. Effect of water on formic acid
photocatalytic decomposition on TiO2 and Pt/TiO2. J Catal 275, 294-299 (2010).
18. Singh N, Sanyal U, Fulton JL, Gutiérrez OY, Lercher JA, Campbell CT.
Quantifying Adsorption of Organic Molecules on Platinum in Aqueous Phase by
Hydrogen Site Blocking and in Situ X-ray Absorption Spectroscopy. Acs Catal 9,
6869-6881 (2019).
20. Behtash S, Lu JM, Mamun O, Williams CT, Monnier JR, Heyden A. Solvation
Effects in the Hydrodeoxygenation of Propanoic Acid over a Model Pd(211)
Catalyst. J Phys Chem C 120, 2724-2736 (2016).
72
21. Rajadhyaksha RA, Karwa SL. Solvent Effects in Catalytic-Hydrogenation. Chem
Eng Sci 41, 1765-1770 (1986).
23. Akpa BS, et al. Solvent effects in the hydrogenation of 2-butanone. J Catal 289,
30-41 (2012).
25. Knowles WS. Asymmetric hydrogenation. Acc Chem Res 16, 106-112 (1983).
26. Valgimigli L, Banks JT, Ingold KU, Lusztyk J. Kinetic Solvent Effects on
Hydroxylic Hydrogen-Atom Abstractions Are Independent of the Nature of the
Abstracting Radical - 2 Extreme Tests Using Vitamin-E and Phenol. J Am Chem
Soc 117, 9966-9971 (1995).
29. Dyson PJ, Jessop PG. Solvent effects in catalysis: rational improvements of
catalysts via manipulation of solvent interactions. Catal Sci Technol 6, 3302-3316
(2016).
30. Cornils B, Herrmann WA, Eckl RW. Industrial aspects of aqueous catalysis. J
Mol Catal a-Chem 116, 27-33 (1997).
31. Zhang L, Karim AM, Engelhard MH, Wei ZH, King DL, Wang Y. Correlation of
Pt-Re surface properties with reaction pathways for the aqueous-phase reforming
of glycerol. J Catal 287, 37-43 (2012).
32. He R, Davda RR, Dumesic JA. In situ ATR-IR spectroscopic and reaction
kinetics studies of water-gas shift and methanol reforming on Pt/Al2O3 catalysts
in vapor and liquid phases. J Phys Chem B 109, 2810-2820 (2005).
73
33. Taylor CD, Neurock M. Theoretical insights into the structure and reactivity of
the aqueous/metal interface. Curr Opin Solid St M 9, 49-65 (2005).
35. Bertero NM, Trasarti AF, Acevedo MC, Marchi AJ, Apesteguia CR. Solvent
effects in solid acid-catalyzed reactions: The case of the liquid-phase
isomerization/cyclization of citronellal over SiO2-Al2O3. Molecular Catalysis
481, (2020).
38. Zhao JJ, et al. Suppressing Metal Leaching in a Supported Co/SiO2 Catalyst with
Effective Protectants in the Hydroformylation Reaction. Acs Catal 10, 914-920
(2020).
39. Kulal AB, Kasabe MM, Jadhav PV, Dongare MK, Umbarkar SB. Hydrophobic
WO3/SiO2 catalyst for the nitration of aromatics in liquid phase. Applied
Catalysis a-General 574, 105-113 (2019).
42. Mellmer MA, Sener C, Gallo JMR, Luterbacher JS, Alonso DM, Dumesic JA.
Solvent Effects in Acid-Catalyzed Biomass Conversion Reactions. Angewandte
Chemie International Edition 53, 11872-11875 (2014).
74
44. Car R, Parrinello M. Unified Approach for Molecular-Dynamics and Density-
Functional Theory. Phys Rev Lett 55, 2471-2474 (1985).
47. Mattsson TR, Paddison SJ. Methanol at the water-platinum interface studied by
ab initio molecular dynamics. Surf Sci 544, L697-L702 (2003).
48. Yang J, Dauenhauer PJ, Ramasubramaniam A. The role of water in the adsorption
of oxygenated aromatics on Pt and Pd. J Comput Chem 34, 60-66 (2013).
49. Klamt A. Conductor-Like Screening Model for Real Solvents - a New Approach
to the Quantitative Calculation of Solvation Phenomena. J Phys Chem-Us 99,
2224-2235 (1995).
51. Zhang YK, Liu HY, Yang WT. Free energy calculation on enzyme reactions with
an efficient iterative procedure to determine minimum energy paths on a
combined ab initio QM/MM potential energy surface. J Chem Phys 112, 3483-
3492 (2000).
52. Hu H, Lu ZY, Parks JM, Burger SK, Yang WT. Quantum mechanics/molecular
mechanics minimum free-energy path for accurate reaction energetics in solution
and enzymes: Sequential sampling and optimization on the potential of mean
force surface. J Chem Phys 128, 034105:034101-034118 (2008).
53. Hu H, Lu ZY, Yang WT. QM/MM minimum free-energy path: Methodology and
application to triosephosphate isomerase. J Chem Theory Comput 3, 390-406
(2007).
75
55. Saleheen M, Zare M, Faheem M, Heyden A. Computational Investigation of
Aqueous Phase Effects on the Dehydrogenation and Dehydroxylation of Polyols
over Pt(111). The Journal of Physical Chemistry C 123, 19052-19065 (2019).
58. Ebbesen SD, Mojet BL, Lefferts L. In situ ATR-IR study of CO adsorption and
oxidation over Pt/Al2O3 in gas and aqueous phase: Promotion effects by water
and pH. J Catal 246, 66-73 (2007).
60. Clayborne A, Chun H-J, Rankin RB, Greeley J. Elucidation of Pathways for NO
Electroreduction on Pt(111) from First Principles. Angewandte Chemie
International Edition 54, 8255-8258 (2015).
64. Salciccioli M, Yu W, Barteau MA, Chen JG, Vlachos DG. Differentiation of O–H
and C–H Bond Scission Mechanisms of Ethylene Glycol on Pt and Ni/Pt Using
Theory and Isotopic Labeling Experiments. J Am Chem Soc 133, 7996-8004
(2011).
65. Zhang X, DeFever RS, Sarupria S, Getman RB. Free Energies of Catalytic
Species Adsorbed to Pt(111) Surfaces under Liquid Solvent Calculated Using
76
Classical and Quantum Approaches. Journal of Chemical Information and
Modeling 59, 2190-2198 (2019).
66. Reed AE, Weinstock RB, Weinhold F. Natural-Population Analysis. J Chem Phys
83, 735-746 (1985).
67. Natarajan SK, Behler J. Neural network molecular dynamics simulations of solid-
liquid interfaces: water at low-index copper surfaces. Phys Chem Chem Phys 18,
28704-28725 (2016).
68. Pearson’s Correlation Coefficient. In: Encyclopedia of Public Health (ed Kirch
W). Springer Netherlands (2008).
69. Faheem M, Suthirakun S, Heyden A. New Implicit Solvation Scheme for Solid
Surfaces. J Phys Chem C 116, 22458-22462 (2012).
74. Blochl PE. Projector Augmented-Wave Method. Phys Rev B 50, 17953-17979
(1994).
75. Perdew JP, Yue W. Accurate and Simple Density Functional for the Electronic
Exchange Energy - Generalized Gradient Approximation. Phys Rev B 33, 8800-
8802 (1986).
76. Perdew JP, Wang Y. Accurate and Simple Analytic Representation of the
Electron-Gas Correlation-Energy. Phys Rev B 45, 13244-13249 (1992).
77
77. Perdew JP, Burke K, Ernzerhof M. Generalized gradient approximation made
simple. Phys Rev Lett 77, 3865-3868 (1996).
78. Monkhorst HJ, Pack JD. Special Points for Brillouin-Zone Integrations. Phys Rev
B 13, 5188-5192 (1976).
81. Harris J. Simplified Method for Calculating the Energy of Weakly Interacting
Fragments. Phys Rev B 31, 1770-1779 (1985).
83. Henkelman G, Jonsson H. Improved tangent estimate in the nudged elastic band
method for finding minimum energy paths and saddle points. J Chem Phys 113,
9978-9985 (2000).
84. Henkelman G, Uberuaga BP, Jonsson H. A climbing image nudged elastic band
method for finding saddle points and minimum energy paths. J Chem Phys 113,
9901-9904 (2000).
85. Henkelman G, Jonsson H. A dimer method for finding saddle points on high
dimensional potential surfaces using only first derivatives. J Chem Phys 111,
7010-7022 (1999).
86. Heyden A, Bell AT, Keil FJ. Efficient methods for finding transition states in
chemical reactions: Comparison of improved dimer method and partitioned
rational function optimization method. J Chem Phys 123, 224101:224101-224114
(2005).
78
89. Von Arnim M, Ahlrichs R. Performance of parallel TURBOMOLE for density
functional calculations. J Comput Chem 19, 1746-1757 (1998).
91. Weigend F, Ahlrichs R. Balanced basis sets of split valence, triple zeta valence
and quadruple zeta valence quality for H to Rn: Design and assessment of
accuracy. Phys Chem Chem Phys 7, 3297-3305 (2005).
92. Russo TV, Martin RL, Hay PJ. Effective Core Potentials for Dft Calculations. J
Phys Chem-Us 99, 17085-17087 (1995).
93. Eichkorn K, Weigend F, Treutler O, Ahlrichs R. Auxiliary basis sets for main row
atoms and transition metals and their use to approximate Coulomb potentials.
Theor Chem Acc 97, 119-124 (1997).
94. Weigend F. Accurate Coulomb-fitting basis sets for H to Rn. Phys Chem Chem
Phys 8, 1057-1065 (2006).
95. Todorov IT, Smith W, Trachenko K, Dove MT. DL_POLY_3: new dimensions in
molecular dynamics simulations via massive parallelism. J Mater Chem 16, 1911-
1918 (2006).
96. Jorgensen WL, Chandrasekhar J, Madura JD, Impey RW, Klein ML. Comparison
of Simple Potential Functions for Simulating Liquid Water. J Chem Phys 79, 926-
935 (1983).
97. Andersen HC. Rattle - a Velocity Version of the Shake Algorithm for Molecular-
Dynamics Calculations. J Comput Phys 52, 24-34 (1983).
98. Smith W, Forester TR. Parallel Macromolecular Simulations and the Replicated
Data Strategy .2. The Rd-Shake Algorithm. Comput Phys Commun 79, 63-77
(1994).
99. Swope WC, Andersen HC, Berens PH, Wilson KR. A Computer-Simulation
Method for the Calculation of Equilibrium-Constants for the Formation of
Physical Clusters of Molecules - Application to Small Water Clusters. J Chem
Phys 76, 637-649 (1982).
79
101. Geerke DP, Van Gunsteren WF. The performance of non-polarizable and
polarizable force-field parameter sets for ethylene glycol in molecular dynamics
simulations of the pure liquid and its aqueous mixtures. Mol Phys 105, 1861-1881
(2007).
102. Freindorf M, Shao Y, Furlani TR, Kong J. Lennard–Jones parameters for the
combined QM/MM method using the B3LYP/6-31G*/AMBER potential. Journal
of Computational Chemistry 26, 1270-1278 (2005).
103. Heinz H, Vaia RA, Farmer BL, Naik RR. Accurate Simulation of Surfaces and
Interfaces of Face-Centered Cubic Metals Using 12-6 and 9-6 Lennard-Jones
Potentials. J Phys Chem C 112, 17281-17290 (2008).
104. Burow AM, Sierka M, Dobler J, Sauer J. Point defects in CaF2 and CeO2
investigated by the periodic electrostatic embedded cluster method. J Chem Phys
130, 174710:174711-174711 (2009).
107. Essmann U, Perera L, Berkowitz ML, Darden T, Lee H, Pedersen LG. A Smooth
Particle Mesh Ewald Method. J Chem Phys 103, 8577-8593 (1995).
111. Best SA, Merz KM, Reynolds CH. Free energy perturbation study of
octanol/water partition coefficients: Comparison with continuum GB/SA
calculations. J Phys Chem B 103, 714-726 (1999).
80
112. Shirts MR, Pande VS. Comparison of efficiency and bias of free energies
computed by exponential averaging, the Bennett acceptance ratio, and
thermodynamic integration. J Chem Phys 122, 144107:144101-144116 (2005).
113. Lu ND, Singh JK, Kofke DA. Appropriate methods to combine forward and
reverse free-energy perturbation averages. J Chem Phys 118, 2977-2984 (2003).
115. Bennett C. Efficient Estimation of Free Energy Differences from Monte Carlo
Data. J Comput Phys 22, 245-268 (1976).
116. Shirts MR, Bair E, Hooker G, Pande VS. Equilibrium free energies from
nonequilibrium measurements using maximum-likelihood methods. Phys Rev Lett
91, (2003).
117. Kreyszig E. Advanced Engineering Mathematics, 9th edn. John Wiley & Sons,
Inc. (2006).
119. Klamt A. Conductor-like Screening Model for Real Solvents: A New Approach to
the Quantitative Calculation of Solvation Phenomena. The Journal of Physical
Chemistry 99, 2224-2235 (1995).
121. Fernandez DP, Goodwin ARH, Lemmon EW, Sengers JMHL, Williams RC. A
formulation for the static permittivity of water and steam at temperatures from
238 K to 873 K at pressures up to 1200 MPa, including derivatives and Debye-
Huckel coefficients. J Phys Chem Ref Data 26, 1125-1166 (1997).
81
CHAPTER 4
HETEROGENOUS CATALYSIS
Zare, M.; Saleheen, M.; Singh, N.; Uline, M. J.; Faheem, M.; Heyden, A.
82
4.1 Abstract
Aqueous solvation free energies of adsorption have recently been measured for
solvents dramatically influence a metal catalysts activity with significant implications for
catalyst design. However, measurements are indirect and involve adsorption isotherm
implicit solvation models predict exergonic solvation effects for phenol adsorption, failing
to agree with measurements even qualitatively. In this study, an explicit, hybrid QM/MM
approach for computing solvation free energies of adsorption is developed, solvation free
energies of phenol adsorption are computed, and experimental data for solvation free
energies of phenol adsorption are reanalyzed using multiple adsorption isotherm models.
for phenol adsorption on Pt(111), that agrees with measurements to within the experimental
glycol, and phenol over the (111) facet of Pt and Cu suggests that liquid water destabilizes
all adsorbed species, with the largest impact on the largest adsorbates.
4.2 Introduction
A heterogeneously catalyzed process starts with the adsorption of at least one of the
reactants and ends with the desorption of the products.1 Activation barriers have also been
correlated to adsorption energies for about ninety years through, e.g., the well-known
catalysis, and adsorption processes at gas-solid interfaces have been studied for many
83
decades.6, 7, 8, 9, 10 For solid-liquid interfaces, less progress has been made, largely due to
the added complexity of the solvent molecules that can both directly and indirectly interact
with the adsorbate through, e.g., hydrogen bonding and by changing the electronic structure
of the adsorbent. Nevertheless, it is well-known that solvents can affect important figures
of merit in catalysis, such as activity and selectivity, that go beyond reduced mass transfer
rates and non-ideality effects that determine a reactants activity/fugacity at a given liquid
concentration.11, 12, 13, 14 These intrinsic solvent effects originate from one or a combination
of the followings: nature of solvent (pH, polarity, H-bonding ability etc.),15, 16, 17 leading to
catalytic cycle or reaction coordinate,21 and competitive adsorption between solvent and
selectivity, and stability, it is critical to understand the root causes of solvent effects on
important the larger the solvation effects are on the adsorption free energy of the various
species in a catalytic system. Recently, aqueous solvation effects have been measured on
the free energy of adsorption for phenol, benzyl alcohol, benzaldehyde, and cyclohexanol
over Pt catalyst at 298 K25 and for phenol over Pt and Rh catalysts at 4 temperatures (283,
288, 298, and 314 K).26, 27 Importantly, for phenol adsorption on supported Pt catalysts, it
has become possible to deconvolute the adsorption isotherm data for the different surface
facets, which enabled the measurement of the aqueous solvent effect on the phenol
adsorption free energy on the Pt(111) facet, which again facilitates a correlation to
computational studies. Experimentally measured aqueous solvation effects are large (> 1.4
eV at 298 K) and endergonic for phenol on Pt(111), suggesting that the kinetic properties
84
of Pt nanoparticles for phenol catalysis are a strong function of solvent properties with
significant implications for catalyst design in condensed media. We highlight that at 298
K and at 500 K a change in a rate controlling free energy barrier of 0.1 eV can change the
overall rate by a factor 50 and factor 10, respectively. However, measurements are indirect
and involve adsorption isotherm models which potentially reduces the reliability of the
and exergonic solvation effects on the free energy of phenol adsorption on Pt(111), failing
Despite the availability of fairly established methods for computing the energetics
adsorption (free) energies at solid-liquid interfaces are less developed. This is partly due to
the limited availability of experimental24, 28, 29, 30 and theoretical studies31, 32, 33, 34, 35, 36, 37, 38,
39, 40, 41, 42, 43
and partly due to inherent intricacy of a reaction system comprised of both a
complex heterogeneous catalyst and a condensed phase. Although “on the fly” electronic
simulations are well-suited for this task, as a result of the large computational cost
AIMD simulations for computing adsorption free energies from a condensed phase is
currently, practically impossible with small confidence interval from configuration space
sampling. Typically, the high computational cost of AIMD constrains both the size of the
simulation system to a few hundred atoms and the time scale of simulations to a few
picoseconds.47, 48, 49 Alternatively, implicit solvation schemes50, 51 and classical force field
85
field simulations require a reliable potential for all fluid components with the adsorption
site/metal surface which, except for perhaps water, is hardly available due to a lack of
adsorption on metal surfaces with implicit solvation models due to the need to parameterize
the implicit solvation models against experimental data that are again hardly available or
possess unknown error bars. We highlight that for homogeneous systems, implicit
solvation models have been exceptionally successful despite their limitations in capturing
predicting free energy changes for processes on solid-liquid interfaces.36, 38, 39, 54, 55, 56 In
this class of simulations, the essential region of interest that can often not be properly
described by classical force fields is described quantum mechanically, while the remainder
cost and accuracy. For example, in a metal-catalyzed adsorption reaction, adsorbate, active
site and its immediate metal neighbors are treated from first principles, while the bulk of
the solvent molecules and the nonreactive part of the simulation system are treated using
and reliable method for computing free energy changes in the presence of solvents using a
hybrid (QM/MM) approach with electrostatic embedding, named eSMS (Explicit Solvation
for Metal Surfaces)57 and applied it to several surface reactions over various transition
metal surfaces.58, 59
86
A limitation of eSMS has been that it can only be applied for computing free energy
differences of surface reactions, that is, it cannot be employed to compute free energies of
adsorption processes from a liquid-phase environment due to (i) the long reaction
coordinate in adsorption processes that increases the computational expense, and (ii) the
need to be able to compute the free energy of the quantum mechanically described part of
the simulation system within the harmonic approximation, i.e., only the classically
described part of the simulation system is sampled extensively within eSMS. In this study,
we identify a reaction coordinate for adsorption processes that enables our eSMS approach
to predict adsorption free energies at a solid-liquid interface. Next, we validated our novel
computational scheme against previous eSMS calculations of the aqueous solvation effect
on the activation free energy barrier of O-H bond cleavage of ethylene glycol, EG
(C2H6O2), over a Cu(111) catalyst surface model59 by computing the adsorption free energy
of both the reactant and transition state. Next, we compute the aqueous solvation effect on
the free energy of phenol adsorption on the Pt(111) surface, and we reanalyzed the
experimental data for this solvation free energy using multiple adsorption isotherm models.
Computational predictions at 298 K are in excellent agreement with the experimental data
with an error less than 0.10 eV which is on the order of the experimental error. Thus, these
results confirm a large endergonic aqueous solvation free energy for phenol adsorption
with significant implications for catalyst design for conversion of aromatic molecules. To
elucidate the significance of these large, computed and measured, solvation effects for the
Pt-phenol system for various other catalytic processes, we computed the aqueous solvation
effect on the adsorption free energies of carbon monoxide (CO), ethylene glycol (EG), and
87
4.3 QM/MM free energy scheme for adsorption processes at a solid-liquid
interface
The key challenge for computing the free energy of adsorption in a liquid-phase
requires extensive configuration space sampling only for the classically described part of
the simulation system, i.e., the harmonic approximation is meaningful for the quantum
mechanically described subsystem in the reactant and product state, and that involves a
relatively smooth reaction coordinate that does not involve overcoming large free energy
barriers, such that, only few free energy perturbation (FEP) steps are required for
computing the overall free energy difference. Figure 4.1 illustrates such a thermodynamic
cycle that is explained below. We note that a detailed computational methods section is
presented in the Supporting Information. The key idea of our thermodynamic cycle is to
not compute the desorption of an adsorbed moiety into a liquid phase but to only compute
the solvent effect on the desorption process. In this way, free energy differences for the
overall process are smaller, and desorption itself is only the gas-phase desorption which
can be described using the harmonic approximation or correction formulas to the harmonic
approximation.60, 61 Thus, a potential reaction coordinate for computing the solvent effect
QM/MM description of the reaction system the electrostatic interaction is an electric force
between the adsorbed species and the metal atoms in direct vicinity of the adsorbed moiety
with the solvent molecules, governed by magnitude and distance of partial charges. To
88
remove the electrostatic potential, first, partial charges on the “metal-adsorbate cluster”
(see Figure C.3) and “metal cluster” (obtained by removing adsorbate atoms from the
metal-adsorbate cluster) are computed using a suitable charge model (we used NPA62 and
DDEC663 in this study). We note that the metal cluster is a “non-physical” state in which
the adsorbate is present but does not electrostatically interact with the surrounding water
molecules that are treated classically, i.e., the partial charges on the adsorbate are set to
zero (𝑄𝑎𝑑𝑠 = 0). Next, we insert adequate “non-physical” states in between by linearly
reducing the magnitude of partial charges on the QM atoms going from the metal-adsorbate
to the metal cluster. Table C.3 summarizes the charges on the QM atoms of these two states
We can use our conventional eSMS methodology for computing the solvent effect
owing to the removal of the classical electrostatic interactions between the adsorbed
species and the solvent molecules. The total energy function of our eSMS method is given
𝑆𝑢𝑟𝑓𝑎𝑐𝑒 𝑐𝑙𝑢𝑠𝑡𝑒𝑟
𝐸𝑇 (𝑟𝑄𝑀 , 𝑟𝑀𝑀 ) = 𝐸𝑄𝑀 (𝑟𝑄𝑀 ) − 𝐸𝑄𝑀 (𝑟𝑄𝑀 ) +
𝑀𝑒𝑎𝑛𝐹𝑖𝑒𝑙𝑑(100)
〈𝛹 |𝐻𝑒𝑓𝑓 (𝑟𝑄𝑀 , 𝑟𝑀𝑀 )| 𝛹〉 +
1 𝑒𝑙𝑒𝑐+𝑣𝑑𝑊 𝑀𝑒𝑎𝑛𝐹𝑖𝑒𝑙𝑑(100)
[100 ∑100
𝑗=1 [𝐸𝑗,𝑀𝑀+𝑄𝑀⁄𝑀𝑀 (𝑟𝑄𝑀 , 𝑟𝑀𝑀 )] −
𝑄𝑖 =0,𝑖∈𝑄𝑀
where the first two terms are evaluated for a periodic slab, using the VASP65, 66 program
package, and a QM cluster in vacuum, with the help of the TURBOMOLE program
package,67, 68, 69 respectively. The combination of the third to fifth term is a QM calculation
89
in a mean field of MM water molecules using the periodic electrostatic embedded cluster
method (PEECM)70 with the fixed-charge approximation (the fixed charge approximation
has been validated for our eSMS approach in reference64). We note that the number 100 in
the equation indicates that 100 MM conformations, selected equally spaced from 1 ns
molecular dynamic (MD) simulations (10 ps apart), were used to represent the mean field
of the MM water molecules. In other words, the first term constitutes the conventional gas
phase energy computed for a periodic slab model and the combination of the second to fifth
term is the non-classical part of the electrostatic interaction energy of a mean-field liquid
phase environment with the adsorbed moiety and metal surface described quantum
adsorption energies on metal particles, this non-classical part of the electrostatic interaction
energy is fast converging with metal cluster size such that reliable energies can be
computed for relatively small metal clusters of ~50 metal atoms. Finally, the last term
accounts for the classical interaction energy within the MM subsystem and its interaction
with the QM subsystem. Since the QM geometry remains the same during the removal of
𝑒𝑙𝑒𝑐+𝑣𝑑𝑊
the electrostatic interaction, all terms except 𝐸𝑗,𝑀𝑀+𝑄𝑀 ⁄𝑀𝑀 (𝑟𝑄𝑀 , 𝑟𝑀𝑀 ) remain the same in
the energy function of every intermediate state and hence, they do not need to be computed
in a FEP step calculation. In other words, removing the classical electrostatic interaction
The QM atom coordinates of the metal-adsorbate cluster state (state I in Figure 4.1)
are ideally those optimized in liquid water and not those optimized in vapor phase. At one
point in the thermodynamic cycle the coordinates need to be changed to those in vapor.
The free energy change for this change in QM atom coordinates involves standard eSMS
90
free energy calculations that involve a number of QM and MM calculations but that are
reasonably fast given the small change in QM atom coordinates during adsorption of
typical adsorbates on a metals surface.11, 57, 58, 59 We typically perform these calculations
of the solvent molecules changing the electronic structure of the quantum system transition
from those of a metal-adsorbate system to those without adsorbate, i.e., this energy
difference constitutes the main difference between electrostatic and mechanical embedding
𝑚𝑒𝑡𝑎𝑙−𝑠𝑜𝑙𝑣𝑒𝑛𝑡 𝑚𝑒𝑡𝑎𝑙−𝑎𝑑𝑠−𝑠𝑜𝑙𝑣𝑒𝑛𝑡
𝐸 𝐼𝐼𝐼 − 𝐸 𝐼𝐼 = (𝐸𝑃𝐸𝐸𝐶𝑀 − 𝐸𝑃𝐸𝐸𝐶𝑀 )+
1
[(100 ∑100 𝑚𝑒𝑡𝑎𝑙−𝑠𝑜𝑙𝑣𝑒𝑛𝑡
𝑗=1 (𝐸𝑗,𝐷𝐿𝑃𝐿𝑂𝑌 ) 𝑚𝑒𝑡𝑎𝑙−𝑠𝑜𝑙𝑣𝑒𝑛𝑡
− 𝐸𝑗,𝐷𝐿 𝑃𝐿𝑂𝑌
)−
𝑄𝑖 =0,𝑖∈𝑄𝑀
1
(100 ∑100 𝑚𝑒𝑡𝑎𝑙−𝑎𝑑𝑠−𝑠𝑜𝑙𝑣𝑒𝑛𝑡
𝑗=1 (𝐸𝑗,𝐷𝐿𝑃𝐿𝑂𝑌 ) 𝑚𝑒𝑡𝑎𝑙−𝑎𝑑𝑠−𝑠𝑜𝑙𝑣𝑒𝑛𝑡
− 𝐸𝑗,𝐷𝐿 𝑃𝐿𝑂𝑌
𝑚𝑒𝑡𝑎𝑙−𝑠𝑜𝑙𝑣𝑒𝑛𝑡
)] − (𝐸𝑐𝑙𝑢𝑠𝑡𝑒𝑟 −
𝑄𝑖 =0,𝑖∈𝑄𝑀
𝑚𝑒𝑡𝑎𝑙−𝑎𝑑𝑠−𝑠𝑜𝑙𝑣𝑒𝑛𝑡
𝐸𝑐𝑙𝑢𝑠𝑡𝑒𝑟 ) (4.2)
where metal-solvent and metal-ads-solvent superscripts represent state III and II,
respectively (the first and last terms in parenthesis are performed using the TURBOMOLE
program package). This step involves a QM calculation; however, the free energy change
description changes of a “non-physical” state changes) and although the free energy
differences are significant, multiple independent calculations suggest that this free energy
change can efficiently be computed without introducing intermediate states (which would
practically be impossible). Here, we also note that given the magnitude of the free energy
91
(that does not consider these non-classical electrostatic contributions) for QM/MM
Step III to IV in Figure 4.1: Van der Waals interaction removal. Another interaction
between solvent and adsorbate-metal moiety are the classical van der Waals (vdW)
scaling of the distance shifted Lennard-Jones (LJ) potentials.71 This approach allows for a
smooth transition in molecular simulations between real and dummy atoms, for which all
atomic interactions have vanished. The LJ potential that we implemented in the DL_POLY
program package to annihilate or create an atom has the following functional form:
𝐴 𝐵
𝑉𝑙𝑗 = (1 − 𝜆) [(𝑟2+𝛿𝜆)6 − (𝑟 2+𝛿𝜆)3] (4.3)
where 𝑟 is the interatomic distance between two atoms, and A and B are repulsive and
each atom type of the interacting particles (𝐴 = 4𝜀𝜎12 , 𝐵 = 4𝜀𝜎 6 ; see Table C.1 for 𝜀 and
𝜎 values we used in this work). For 𝜆 = 0, the vdW interactions fully exist and an increase
interaction between the solvent molecules and the adsorbate is zero everywhere,
corresponding to the final state where the adsorbate atom is transformed to a dummy
particle. The shift parameter 𝛿 comes into play for 𝜆 larger than zero. It allows for a smooth
et al.71, the parameter 𝛿 was chosen as the square of the vdW radius of the interacting atoms
to allow the smoothest transition between an atom present on the surface and filling the
cavity after molecule annihilation. Table C.4 summarizes all 𝛿 parameter values for all pair
92
interaction terms and, as an example, Figure C.2 illustrates the LJ potential as a function
of distance between the oxygen atom of an adsorbed CO and the oxygen atom of a TIP3P
To enable a reliable free energy difference calculation, the removal of the vdW
interactions between the solvent and the adsorbate is done in multiple steps by linearly
𝑒𝑙𝑒𝑐+𝑣𝑑𝑊
interactions, in the energy function, equation 4.1, only the 𝐸𝑗,𝑀𝑀+𝑄𝑀 ⁄𝑀𝑀 (𝑟𝑄𝑀 , 𝑟𝑀𝑀 ) term
changes between two consecutive states and no QM calculations are required for this step,
i.e., only classical free energy calculations are performed at the MM level of theory. At this
point, all classical interactions between the adsorbate and the solvent have vanished.
Given that the quantum atom coordinates of the metal cluster are still those in the
gas phase adsorbed state and not those in the liquid-phase optimized metal cluster (free
site), the free energy change for this change in QM atom coordinates needs to be computed.
This involves again standard eSMS free energy calculations.11, 57, 58, 59 We note that for all
free energy calculations we employ the Bennett acceptance ratio (BAR) as free energy
estimator for all FEP steps.72 For practical convenience, we typically perform these
reference state the adsorbate in a gas-phase in equilibrium with the adsorbate in the liquid
phase. Thus, only a typical gas phase metal adsorption calculation at the QM level of theory
is required for this step. The free energy change from state I to state V constitutes the free
energy change from an adsorbed molecule surrounded by a solvent to the adsorbent (metal
93
surface) surrounded by a solvent and the adsorbate in a gas phase state. Thus, the free
energy change from state I to state IV constitutes the solvent effect on an adsorption
dependent as the gas phase adsorption energy.73 Also, the activity/fugacity calculation of
the adsorbate in vapor phase is often significantly easier than such a calculation in a solvent.
In this paper, we assumed the vapor phase to behave as an ideal gas. For adsorbates with
very low vapor pressure, where it is challenging to measure the fugacity, the gas-liquid
‡ ‡
processes, the aqueous-phase effect on the free energy of activation (∆∆𝐺𝑙𝑖𝑞 = ∆𝐺𝑙𝑖𝑞 −
‡
∆𝐺𝑔𝑎𝑠 ) of the O-H bond cleavage of ethylene glycol over a Cu(111) surface at 423 K is
computed using two different approaches, the eSMS methodology for surface reactions59
and the adsorption scheme proposed here. In our proposed adsorption scheme, the solvent
‡ 𝑔𝑎𝑠→𝑙𝑖𝑞 𝑔𝑎𝑠→𝑙𝑖𝑞
∆∆𝐺𝑙𝑖𝑞 = ∆∆𝐺𝑇𝑆 − ∆∆𝐺𝑅𝑒𝑎𝑐𝑡𝑎𝑛𝑡 (4.4)
‡ 𝑔𝑎𝑠→𝑙𝑖𝑞 𝑔𝑎𝑠→𝑙𝑖𝑞
where ∆∆𝐺𝑙𝑖𝑞 , ∆∆𝐺𝑇𝑆 , and ∆∆𝐺𝑅𝑒𝑎𝑐𝑡𝑎𝑛𝑡 indicate the solvent effect on the activation
barrier, on the adsorption of the transition state (TS), and on the adsorption of the reactant
94
Figure 4.2a illustrates the free-energy profile of the solvation effects on the
𝑙𝑖𝑞→𝑔𝑎𝑠 𝑔𝑎𝑠→𝑙𝑖𝑞
desorption of the RS and TS (∆∆𝐺𝑃ℎ = −∆∆𝐺𝑃ℎ ). For both the RS and TS, the
transition of the QM atom coordinates leads only to a small free energy change relative to
the overall solvent effect. The reactant state is more destabilized in liquid water than the
transition state leading to a significantly reduced activation free energy barrier. Both, the
RS and TS are similarly destabilized through the creation of a solvent cavity on the surface
as shown in the change in free energy during the removal of the van-der-Waals/Lennord-
Jones interactions (see Table C.5). In contrast, the classical electrostatic interactions
𝑙𝑖𝑞→𝑔𝑎𝑠 𝑙𝑖𝑞→𝑔𝑎𝑠
stabilize the RS and TS in solution (∆∆𝐺𝐸𝐿−𝑅𝑚𝑣,𝑅𝑆 = 0.49, ∆∆𝐺𝐸𝐿−𝑅𝑚𝑣,𝑇𝑆 = 1.25 𝑒𝑉).
The difference in classical stabilization can be attributed to the significant change in partial
charges of the QM system going from RS to TS (see Table C.3a). In particular, the charge
of the cleaved hydrogen atom has transitioned from 0.47 e in RS to -0.16 e in TS, explaining
Overall, the electrostatic interaction effect is, however, reduced by the non-classical
electrostatic interactions from section “II-to-III”. The latter effect is ignored in molecular
𝑙𝑖𝑞→𝑔𝑎𝑠 𝑙𝑖𝑞→𝑔𝑎𝑠
embedding calculations (∆∆𝐺𝐼𝐼−𝑡𝑜−𝐼𝐼𝐼,𝑅𝑆 = −0.47, ∆∆𝐺𝐼𝐼−𝑡𝑜−𝐼𝐼𝐼,𝑇𝑆 = −0.68 𝑒𝑉).
Finally, when considering the vibrational contributions, the overall aqueous solvent
𝑔𝑎𝑠→𝑙𝑖𝑞
effect on the adsorption free energy (∆∆𝐺𝐴𝑑𝑠𝑜𝑟𝑏𝑎𝑡𝑒 ) is 0.60 ± 0.03 eV for the RS and
‡
0.10 ± 0.05 eV for the TS, such that the solvent effect on the activation barrier (∆∆𝐺𝑙𝑖𝑞 ) is
95
−0.50 ± 0.06 eV. This agrees very well with our previous study, where we used eSMS for
‡,𝑒𝑆𝑀𝑆
surface reactions and found ∆∆𝐺𝑙𝑖𝑞 = −0.56 ± 0.04 𝑒𝑉.59
𝑔𝑎𝑠→𝑙𝑖𝑞
the adsorption free energy of phenol at low coverage (∆∆𝐺𝑃ℎ (𝜃𝑃ℎ = 0)) and at 298 K
to be 1.48 ± 0.04 eV. To compare our result to the experimental work from Singh et al. 26,
74
, we used the adsorption model from Nitta et al.75, 76
that considers the ability of
adsorbates to cover multiple active/metal sites and that can be derived from statistical
mechanics. According to the experimental work, only hydrogen and phenol are present in
2 2
𝑔−𝑙 𝜃𝐻 𝜃𝐻
𝐻2 (𝑔) + 2 ∗ (𝑙𝑖𝑞) ↔ 2𝐻 ∗ (𝑙𝑖𝑞) 𝐾𝑎𝑑𝑠,𝐻2 = 𝑓𝐻 = 𝑃𝐻 (4.6)
2 )𝜃 2 ( 2 )𝜃 2
( 1 𝑏𝑎𝑟 ∗
1 𝑏𝑎𝑟 ∗
1 = 𝜃∗ + 𝜃𝐻 + 𝑛𝜃𝑃ℎ (4.7)
where 𝜃𝑃ℎ , 𝜃𝐻 , and 𝜃∗ are the surface coverage of phenol, hydrogen and free site (water
everywhere), respectively, variable 𝑛 indicates the number of sites that a phenol molecule
𝑔−𝑙
can occupy when adsorbed on the surface, 𝐾𝑎𝑑𝑠 is the equilibrium constant of adsorption
of an adsorbate on the surface from its free state in the gas phase, 𝑓𝑃ℎ , and 𝑓𝐻 are phenol
and hydrogen fugacity in the gas phase, respectively, that can be reduced to the ideal gas
partial pressure under experimental conditions. We note that we have defined the “surface
coverage of species 𝑖” as
96
𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑠𝑢𝑟𝑓𝑎𝑐𝑒 𝑠𝑝𝑒𝑐𝑖𝑒𝑠 𝑖
𝜃𝑖 ≡ (4.8)
𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑠𝑢𝑟𝑓𝑎𝑐𝑒 𝑎𝑡𝑜𝑚𝑠
i.e., the “real” surface coverage is 𝑛𝑖 𝜃𝑖 . By performing some algebra, we reach as the
equation for the adsorption isotherm (see the Supporting Information for the full
derivation)
𝑛 𝑔𝑎𝑠→𝑙𝑖𝑞 𝑙𝑖𝑞
𝑔𝑎𝑠 1−𝑛𝜃 −{∆∆𝐺𝑃ℎ (𝜃𝑃ℎ =0)+𝛼𝑃ℎ,𝑃ℎ 𝜃𝑃ℎ }
𝑃ℎ
𝜃𝑃ℎ = 𝐾𝑎𝑑𝑠,𝑃ℎ (𝜃𝑃ℎ = 0)𝐾𝐻,𝑃ℎ 𝐶𝑃ℎ (1+𝑐𝑜𝑛𝑠𝑡 ) 𝑒𝑥𝑝 [ ] (4.9)
𝑅𝑇
𝑔𝑎𝑠
where 𝐾𝑎𝑑𝑠,𝑃ℎ (𝜃𝑃ℎ = 0) is gas-phase adsorption equilibrium constant of phenol at low
𝑔𝑎𝑠 77 𝑔𝑎𝑠 78
coverage (at 298 K: ∆𝐻𝑃ℎ = −200 𝑘𝐽⁄𝑚𝑜𝑙 , ∆𝑆𝑃ℎ = −14.7𝑅 ), 𝐾𝐻,𝑃ℎ is phenol
concentration in the aqueous phase, 𝑐𝑜𝑛𝑠𝑡 is a constant taking into account a hydrogen
𝑙𝑖𝑞
coverage effect (𝑐𝑜𝑛𝑠𝑡 = 0 indicates zero hydrogen coverage approximation), 𝛼𝑃ℎ,𝑃ℎ is a
constant associated with adsorbed phenol-phenol lateral interactions in liquid, 𝑅 is the gas
𝑔𝑎𝑠→𝑙𝑖𝑞
constant, 𝑇 is the system temperature, and finally ∆∆𝐺𝑃ℎ (𝜃𝑃ℎ = 0) is the solvent
𝑔𝑎𝑠→𝑙𝑖𝑞 𝑔−𝑙 𝑔
∆∆𝐺𝑃ℎ = ∆𝐺𝑎𝑑𝑠,𝑃ℎ − ∆𝐺𝑎𝑑𝑠,𝑃ℎ (4.10)
𝑔−𝑙 𝑙 𝑠𝑜𝑙𝑣𝑎𝑡𝑖𝑜𝑛
∆𝐺𝑎𝑑𝑠,𝑃ℎ = ∆𝐺𝑎𝑑𝑠,𝑃ℎ + ∆𝐺𝑃ℎ (4.11)
𝑙 𝑔−𝑙
We note here that ∆𝐺𝑎𝑑𝑠,𝑃ℎ and ∆𝐺𝑎𝑑𝑠,𝑃ℎ are the adsorption free energy of phenol in the
aqueous phase from its solvated state in solution and from its free state in the gas phase,
𝑠𝑜𝑙𝑣𝑎𝑡𝑖𝑜𝑛
respectively, and ∆𝐺𝑃ℎ is the solvation free energy of phenol which can be obtained
97
Next, we fitted equation 4.9 to the experimental data at different phenol
𝑔𝑎𝑠→𝑙𝑖𝑞 𝑙𝑖𝑞
concentration at 298 K to extract ∆∆𝐺𝑃ℎ (𝜃𝑃ℎ = 0) and 𝛼𝑃ℎ,𝑃ℎ . Figure 4.3 shows the
best fits for different 𝑛 and 𝑐𝑜𝑛𝑠𝑡 that are experimentally not known. A change of 𝑐𝑜𝑛𝑠𝑡
𝑔𝑎𝑠→𝑙𝑖𝑞
from 0 to 1 fortunately only changes ∆∆𝐺𝑃ℎ (𝜃𝑃ℎ = 0) by less than 0.10 eV; however,
for a large 𝑐𝑜𝑛𝑠𝑡 (high hydrogen coverage) such as 10, this change is about 0.25 eV.
Experimental work suggests that 𝑐𝑜𝑛𝑠𝑡 < 1.0 since the partial pressure of hydrogen is
quite low such that the assumption of no hydrogen coverage (𝑐𝑜𝑛𝑠𝑡 = 0) is acceptable.
Furthermore, the plots in Figure 4.3 indicate that the fits with lower 𝑛 predict the higher
concentration data better, while a higher 𝑛 leads to better fits at low phenol concentration
(see also Figure C.5). We attribute this observation and the fact that, for a fixed 𝑛 value,
orientations on the surface at low and high coverages. At low coverage, phenol adsorbs
horizontally parallel to the surface, while increasing the phenol coverage forces the
adsorbed phenol molecules to slant upwards to occupy less space on the surface for
𝑔𝑎𝑠→𝑙𝑖𝑞
additional adsorbed phenol molecules (see Figure C.6). Fortunately, ∆∆𝐺𝑃ℎ (𝜃𝑃ℎ = 0)
is not very sensitive to the 𝑛 value and we conclude that independent of these intricacies,
𝑔𝑎𝑠→𝑙𝑖𝑞
the experimental ∆∆𝐺𝑃ℎ (𝜃𝑃ℎ = 0) value is in an excellent agreement with our
computational prediction to within an error of less than 0.10 eV. Aqueous solvation effects
are indeed large for phenol adsorption on Pt(111) with significant consequences for metal
98
4.4.3 Aqueous-phase effect across adsorbates and metals
effects for the Pt-phenol system for various other catalytic processes, we computed the
𝑔𝑎𝑠→𝑙𝑖𝑞 𝑔−𝑙
aqueous solvation effect on the adsorption free energies (∆∆𝐺𝐴𝑑𝑠𝑜𝑟𝑏𝑎𝑡𝑒 = ∆𝐺𝐴𝑑𝑠𝑜𝑟𝑏𝑎𝑡𝑒 −
𝑔𝑎𝑠
∆𝐺𝐴𝑑𝑠𝑜𝑟𝑏𝑎𝑡𝑒 ) of three different adsorbates, ethylene glycol (EG), carbon monoxide (CO),
and phenol (Ph), over the (111) surface facet of Cu and Pt. In addition to our QM/MM-
FEP calculations, we also used two different well-known implicit solvation methods,
iSMS80 and VASPsol81, 82 (see Table 4.1). Positive solvent effects on the adsorption free
energies indicate that liquid water destabilizes the adsorbed moiety. In this regard, we
found that an aqueous phase destabilizes all three adsorbed species and that the impact is
most pronounced for phenol and smallest for CO, irrespective of the metal surface
adsorbates on the surface primarily to the solvent cavity formation energy and the desire
Furthermore, the metal identity also plays a key role in the solvation effect,
structure of the metal atoms that in turn affects the stability of the adsorbed moieties59.
Table 4.1 shows that all three adsorbed moieties are more destabilized on the Pt(111)
surface relative to the Cu(111) surface. To elucidate the root cause for the different solvent
effects across the metals and the adsorbates, we developed a linear model for describing
𝑦−𝑦̅ ̅̅̅
𝑓1 −𝑓 1 ̅̅̅
𝑓2 −𝑓 2
the variability of the solvation free energy of adsorption, = 𝛼1 + 𝛼2 (𝑠 is
𝑠𝑦 𝑠𝑓1 𝑠 𝑓2
standard deviation and bar sign shows mean), in which 𝑓1 and 𝑓2 are the net charge on the
99
adsorbate and number of sites occupied by the adsorbate (#Site), respectively (see Table
C.6). Based on the best fit parameters (𝛼1 = 0.45, 𝛼2 = 0.62), we conclude that different
solvent effects arise almost equally from different charge-transfer effects (net charge on
adsorbate) across metals and adsorbates and the size of the molecule (number of sites
occupied). We note that we also tested hydrogen bond formation between the adsorbate
and TIP3P water molecules as well as gas phase adsorption energy as possible descriptors,
but we found the best fit using the net charge and #Site.
Figure 4.2 illustrates the free-energy profiles of the aqueous solvent effect on the
desorption of the various adsorbed moieties. The profiles are comprised of five sub-
electrostatic interactions between the adsorbate and the liquid water molecules generally
increases the total free energy of the system, i.e., the classical electrostatic interactions
stabilize all adsorbed species. As expected, this effect is most dominant for ethylene glycol,
somewhat smaller for phenol and least relevant for CO, which is attributed to the fact that
attractive forces between charged species decrease the energy of the system while repulsive
forces between charged species have opposite effect. In this context, perturbing charges of
the adsorbate atoms from its original magnitude to zero decreases the attractive forces
100
between the adsorbate and the water molecules and in turn increases the energy of the total
system.
changing the electronic structure of the quantum system generally reduces the free energy
contributions.
between the adsorbed species and the water molecules generally decreases the free energy
of the system as shown in Table 4.2 and Figure 4.2. Slow removal of the LJ interaction
energy leads to a reduction of the water cavity for the adsorbates. Cavity formation is
generally an endergonic process and given that any adsorbate requires a similar-sized water
independent of metal surface (see Table 4.2). This again explains why larger adsorbates
that require larger water cavities are often also more destabilized on a metal surface.
Transition of metal cluster coordinates (To OPT site in Gas) contribution to the
solvent effect. This contribution is significant for adsorption of molecules that bind strongly
to the surface in the gas phase and thus distort the coordinates of the surface metal atoms
upon adsorption. For EG, CO, and phenol on Pt(111), the zero-point corrected adsorption
energies, -0.35, -1.73,and -2.29 eV, respectively, correlate with this contribution to the
101
4.4.5 Implicit (iSMS and VASPsol) vs. explicit (QM/MM-FEP) solvation
models
Finally, we compare the results of the solvent effect on the adsorption free energies
of EG, CO and phenol over Pt(111) and Cu(111) computed with our QM/MM-FEP scheme
against two different implicit solvation schemes: VASPsol and our iSMS methodology.
Table 4.1 shows that both implicit solvation methods fail to capture the true solvent effect,
probably because they underestimate the endergonic cavity formation energy. In contrast,
our explicit solvation scheme agrees quite well with experimental data probably because
the metal-water force field is specifically optimized for this free energy contribution.83
Interestingly, both implicit solvation schemes predict similar solvation free energies at 298
K, the temperature for which the VASPsol parameters have been optimized. While iSMS
permits a straightforward change in system temperature, we could only adjust the dielectric
constant for water with temperature in VASPsol and otherwise used the default VASPsol
parameters. Thus, at 423 K the solvation free energy prediction deviate significantly
between the two implicit solvation methods. To improve implicit solvation models for
description.
To compute the enthalpic and entropic contributions to the solvent effect on the
for CO and phenol on Pt(111) at five different temperatures within the temperature interval
of 285 to 353 K and constructed van’t Hoff plots shown in Figure 4.4. The results indicate
that the enthalpic and entropic contributions to the solvation free energy of adsorption are
102
of equal size for phenol and that phenol adsorption on Pt(111) in liquid water is destabilized
𝑔𝑎𝑠→𝑙𝑖𝑞
both enthalpically and entropically (∆∆𝐻𝑃ℎ = 0.71 ± 0.12 𝑒𝑉,
𝑔𝑎𝑠→𝑙𝑖𝑞
−𝑇∆∆𝑆𝑃ℎ @298𝐾 = 0.77 ± 0.09 𝑒𝑉). For CO adsorption, our error bars for the free
energy of solvation are too large to unambiguously determine the enthalpic and entropic
𝑔𝑎𝑠→𝑙𝑖𝑞
contributions with a van’t Hoff plot (∆∆𝐻𝐶𝑂 = 0.11 ± 0.22 𝑒𝑉,
𝑔𝑎𝑠→𝑙𝑖𝑞
−𝑇∆∆𝐻𝐶𝑂 @298𝐾 = 0.38 ± 0.19 𝑒𝑉). Still it appears that the entropic solvation
Phenol on Pt(111), experimental vs. computational: Singh et al.25, 27, 74 recently also
studied phenol adsorption in water at four different temperatures (283, 288, 298, 314 K) to
equation 4.9 to their measured data (see Figures 4.3 and C.7 to C.9) and Table 4.3 shows
𝑔𝑎𝑠→𝑙𝑖𝑞
the extracted ∆∆𝐺𝑃ℎ (𝜃𝑃ℎ = 0) values at the different temperatures. Overall, the
agreement between the computed (QM/MM-FEP) and experimental solvation free energies
of adsorption is excellent (an error of ~0.10 eV) for the low coverage adsorption free energy
and entropies of phenol adsorption in water disagree with our computational data. Very
large enthalpic solvation effects are predicted that are reduced by entropic contributions
𝑔𝑎𝑠→𝑙𝑖𝑞 𝑔𝑎𝑠→𝑙𝑖𝑞
(∆∆𝐻𝑃ℎ = 2.33 𝑒𝑉, ∆∆𝑆𝑃ℎ = 3.10 𝑚𝑒𝑉/𝐾). Intuitively, we fear that the
solvation effect on the enthalpy of adsorption is too large and an artifact of the small
experimental temperature interval in the Van’t Hoff plot together with the sizable error
bars of the experimental solvation effects on the Gibbs free energies of adsorption.
103
4.5 Conclusion
for heterogeneous catalysis applications. To numerically verify the proposed scheme, the
solvent effect on the activation free energy of O-H bond cleavage in EG over Cu(111) has
been computed through two adsorption free energy of solvation calculations (reactant and
transition state) and by performing conventional eSMS calculations for surface reactions.
Using the adsorption calculations, we predict a solvation effect on the activation free
energy of −0.50 ± 0.06 eV which agrees very well with our conventional eSMS
‡
calculations (∆∆𝐺𝑙𝑖𝑞 = −0.56 ± 0.04 𝑒𝑉) and which highlights that the sampling error is
similarly small in the new scheme that has the added benefit of being able to describe
adsorption events.
reanalyzed experimental data for solvation free energies of phenol adsorption on this
surface using multiple adsorption isotherm models. Our explicit solvation calculations
schemes such as VASPsol and iSMS fail to capture the true solvent effect, probably because
they underestimate the endergonic cavity formation energy. To elucidate the significance
of these large, computed and measured, solvation effects for the Pt-phenol system for
various other catalytic processes, we computed the aqueous solvation effect on the
adsorption free energies of carbon monoxide, ethylene glycol, and phenol over the Pt(111)
104
and Cu(111) surfaces. Irrespective of metal surface, the water environment destabilized all
𝑔𝑎𝑠→𝑙𝑖𝑞
three adsorbed species with the largest impact on the largest adsorbate (∆∆𝐺𝑃ℎ >
𝑔𝑎𝑠→𝑙𝑖𝑞 𝑔𝑎𝑠→𝑙𝑖𝑞
∆∆𝐺𝐸𝐺 > ∆∆𝐺𝐶𝑂 ) which is attributed to both the size of the molecule (number of
sites occupied) and corresponding solvent cavity formation and the different charge-
transfer effects (net charge on adsorbate) across the metals and adsorbates.
4.6 Acknowledgements
Office of Basic Energy Science, Catalysis Science program under Award DE-
SC0007167. This work was also supported by the South Carolina State Center for
have been provided by XSEDE facilities located at Texas Advanced Computing Center
(TACC) and San Diego Supercomputer Center (SDSC) under grant number TG-
CTS090100 and High Performance Computing clusters located at the University of South
Carolina.
105
4.7 Tables and Figures
Table 4.1: Aqueous-phase effect on the free energy of low coverage adsorption
𝑔𝑎𝑠→𝑙𝑖𝑞 𝑔−𝑙 𝑔𝑎𝑠
(∆∆𝐺𝐴𝑑𝑠𝑜𝑟𝑏𝑎𝑡𝑒 = ∆𝐺𝐴𝑑𝑠𝑜𝑟𝑏𝑎𝑡𝑒 − ∆𝐺𝐴𝑑𝑠𝑜𝑟𝑏𝑎𝑡𝑒 ) in eV for ethylene glycol over Cu(111) and
Pt(111) at 423 K, for CO on Cu(111) and Pt(111) at 298 K, and for phenol over Cu(111)
and Pt(111) at 298 K. The aqueous phase results are based on the proposed scheme in this
work (QM/MM-FEP) and two implicit solvation schemes: iSMS and VASPsol. For the
QM/MM-FEP calculations, the 95% confidence interval (based on limited water sampling
and multiple independent simulations) is also given.
𝒈𝒂𝒔→𝒍𝒊𝒒
∆∆𝑮𝑨𝒅𝒔𝒐𝒓𝒃𝒂𝒕𝒆
Adsorbate / 𝒈𝒂𝒔 ⁄𝑴𝑴−𝑭𝑬𝑷 QM/MM- VASPs
∆𝑮𝑨𝒅𝒔𝒐𝒓𝒃𝒂𝒕𝒆 ∆𝑮𝒈−𝒍,𝑸𝑴
𝑨𝒅𝒔𝒐𝒓𝒃𝒂𝒕𝒆
iSMS
Surface / T(K) FEP ol
Ethylene Glycol /
0.45 1.05 ± 0.03 0.60 ± 0.03 0.02 -0.15
Cu(111) / 423
Ethylene Glycol /
0.35 1.30 ± 0.05 0.95 ± 0.05 -0.19 -0.21
Pt(111) / 423
Carbon Monoxide /
-0.22 0.16 ± 0.06 0.38 ± 0.06 -0.03 -0.02
Cu(111) / 298
Carbon Monoxide /
-1.24 -0.71 ± 0.04 0.53 ± 0.04 0.01 -0.05
Pt(111) / 298
Phenol / Cu(111) /
-0.59 0.69 ± 0.05 1.28 ± 0.05 -0.19 -0.22
298
Phenol / Pt(111) /
-1.61 -0.13 ± 0.04 1.48 ± 0.04 -0.41 -0.47
298
106
𝑙𝑖𝑞→𝑔𝑎𝑠 𝑔𝑎𝑠→𝑙𝑖𝑞
Table 4.2: Contributions to aqueous-phase effect on free energy (in eV) of low coverage desorption (∆∆𝐺𝐴𝑑𝑠𝑜𝑟𝑏𝑎𝑡𝑒 = −∆∆𝐺𝐴𝑑𝑠𝑜𝑟𝑏𝑎𝑡𝑒 )
of ethylene glycol at 423 K (see Figure 4.2c), of carbon monoxide at 298 K (see Figure 4.2b), and of phenol at 298 K (see Figure 4.2d)
over the (111) facets of Pt and Cu catalysts. All numbers also contain a 95% confidence interval based on limited water sampling
obtained by performing multiple independent simulations.
Remove
To OPT Remove II-to-III
Lennard- To OPT site in
Surface Adsorbate adsorbate in Electrostatic (Electrostatic
Jones Gas
Liquid potential Embedding)
potential
Ethylene glycol
0.07 ± 0.00 0.49 ± 0.00 -0.47 ± 0.02 -0.64 ± 0.00 -0.05 ± 0.00
@ 423K
Carbon Monoxide
Cu(111) 0.00 ± 0.00 0.00 ± 0.00 -0.05 ± 0.07 -0.30 ± 0.02 -0.04 ± 0.00
@ 298 K
107
Phenol
-0.02 ± 0.00 0.13 ± 0.00 -0.45 ± 0.05 -0.90 ± 0.01 -0.01 ± 0.00
@ 298 K
Ethylene glycol
-0.01 ± 0.00 0.46 ± 0.00 -0.51 ± 0.05 -0.72 ± 0.01 -0.14 ± 0.00
@ 423 K
Carbon monoxide
Pt(111) 0.00 ± 0.00 0.04 ± 0.00 -0.04 ± 0.05 -0.37 ± 0.02 -0.16 ± 0.01
@ 298 K
Phenol
-0.01 ± 0.00 0.40 ± 0.00 -0.40 ± 0.03 -1.02 ± 0.01 -0.45 ± 0.00
@ 298 K
𝑔𝑎𝑠→𝑙𝑖𝑞 𝑔𝑎𝑠→𝑙𝑖𝑞
Table 4.3: Temperature dependence of the aqueous-phase effect on free energy (∆∆𝐺𝑃ℎ ), enthalpy (∆∆𝐻𝑃ℎ ), and entropy
𝑔𝑎𝑠→𝑙𝑖𝑞
(∆∆𝑆𝑃ℎ ) of low coverage phenol adsorption on the Pt(111) catalyst surface. Experimental data obtained by fitting equation 4.9 (𝑛 =
4 and 𝑐𝑜𝑛𝑠𝑡 = 0.1) to the raw data from Singh et al.25, 26, 74 Computational data obtained by the QM/MM-FEP scheme proposed in this
𝑔−𝑙 𝑔−𝑙
study. ∆𝐻𝑎𝑑𝑠,𝑃ℎ and ∆𝑆𝑎𝑑𝑠,𝑃ℎ are the enthalpy and entropy of phenol adsorption in liquid water from its free state in the gas phase
𝑔𝑎𝑠→𝑙𝑖𝑞 𝑔−𝑙 𝑔𝑎𝑠
(𝑃ℎ(𝑔) + 𝑛 ∗ (𝑙𝑖𝑞) ↔ 𝑃ℎ𝑛∗ (𝑙𝑖𝑞)), respectively (∆∆𝑍𝑃ℎ = ∆𝑍𝑃ℎ − ∆𝑍𝑃ℎ , 𝑍 ≡ 𝐺, 𝐻, 𝑆). Numbers in bracket [] in the QM/MM-
FEP row were calculated using the experimental gas-phase enthalpy and entropy of phenol adsorption on the Pt(111) catalyst surface
𝑔𝑎𝑠 𝑔𝑎𝑠
(∆𝐻𝑃ℎ = −200 𝑘𝐽⁄𝑚𝑜𝑙 77, ∆𝑆𝑃ℎ = −14.7𝑅 78).
𝒈−𝒍
𝒈𝒂𝒔→𝒍𝒊𝒒 𝒈𝒂𝒔→𝒍𝒊𝒒 𝒈𝒂𝒔→𝒍𝒊𝒒 𝒈−𝒍 ∆𝑺𝒂𝒅𝒔,𝑷𝒉 (meV/K
𝑻(K) ∆∆𝑮𝑷𝒉 (eV) ∆∆𝑯𝑷𝒉 (eV) ∆∆𝑺𝑷𝒉 (meV/K) ∆𝑯𝒂𝒅𝒔,𝑷𝒉 (eV)
)
283 1.46
288 1.44
Experimental 2.33 3.10 0.26 1.80
108
298 1.42
314 1.36
285 1.45±0.02
298 1.48±0.04
-1.60±0.12 -4.90±0.3
QM/MM-FEP 315 1.50±0.00 0.71±0.12 -2.60±0.3
[-1.36±0.12] [-3.90±0.3]
335 1.56±0.02
353 1.63±0.01
I Ia II III IIIa IV V
+ + + + + + +
Figure 4.1: Free energy perturbation approach for computing the difference in the
adsorption free energy of any adsorbed species, , in the presence of liquid water ( (𝑔) +
∗ (𝑙) ↔ ∗ (𝑙)) and in a gas phase ( (𝑔) + ∗ (𝑔) ↔ ∗ (𝑔)). States Ia and IIIa are
“non-physical” states. Between states I and II, the (classical) electrostatic interaction
between the solvent and the adsorbate is slowly removed. Also, the coordinates of the
atoms described quantum chemically transition from those in liquid to those in the vapor
phase. From state II to III, the non-classical contributions of water molecules changing the
electronic structure of the quantum system transition from those of a metal-adsorbate
system to those without adsorbate, i.e., this energy difference constitutes the main
difference between electrostatic and mechanical embedding of the QM/MM system. In
other words, symbolizes the adsorbate being only present in the classical DL_POLY
simulation and not in the quantum system. Van der Waals interactions between the solvent
and the adsorbate are slowly removed between state III and IV. The gas-phase desorption
is described between state IV and V. The (free) energy difference between state IV and I
is the solvent effect on the stability of the adsorbate.
109
Figure 4.2: (Average) free-energy profiles for aqueous-phase effects on the desorption
𝑙𝑖𝑞→𝑔𝑎𝑠 𝑔𝑎𝑠→𝑙𝑖𝑞
(∆∆𝐺𝐴𝑑𝑠𝑜𝑟𝑏𝑎𝑡𝑒 = −∆∆𝐺𝐴𝑑𝑠𝑜𝑟𝑏𝑎𝑡𝑒 ) of a) the reactant and transition states of O-H bond
cleavage of ethylene glycol over a Cu(111) surface at 423 K, b) the CO molecule over the
Cu(111) and Pt(111) surfaces at 298 K, c) the ethylene glycol molecule over Cu(111) and
Pt(111) surfaces at 423 K, and d) the phenol molecule over the Cu(111) and Pt(111)
𝑙𝑖𝑞→𝑔𝑎𝑠
surfaces at 298 K. ∆∆𝐺 (∆∆𝐺𝐴𝑑𝑠𝑜𝑟𝑏𝑎𝑡𝑒 ) is the aqueous-phase effect on the low coverage
desorption of an adsorbate. “To OPT adsorbate in Liquid” represents the FEP steps
involved in the transition of the metal-adsorbate QM system in gas to that in liquid, “To
OPT site in Gas” represents the FEP steps involved in changing the coordinates of the metal
cluster atoms with dummy adsorbates that have zero interaction with the solvent to those
of the optimized metal cluster in the gas phase, and “II-to-III” shows the non-classical
contribution of water molecules changing the electronic structure of the quantum system
transitioning from those of a metal-adsorbate system to those without adsorbate (see Figure
4.1, states II and III). “Remove Electrostatic” and “Remove Lennard-Jones” represent the
removal of classical electrostatic and Lennard-Jones interactions between an adsorbate and
the water molecules, respectively. For the magnitude of each contribution, see Tables 4.2
and C.5.
110
Figure 4.3: Fits of equation 4.9 (adsorption isotherm) to the experimental data of the
aqueous-phase phenol adsorption on a Pt(111) catalyst surface at 298 K from Singh et al.26,
74
(see Figures C.7 to C.9 for similar plots at different temperatures). The “Const”
parameter represents the hydrogen surface coverage relative to the free site coverage
(const=0 shows zero H coverage approximation), 𝑛 represents the number of sites a phenol
𝑙𝑖𝑞
molecule occupies when adsorbed on the surface, 𝛼 (𝛼𝑃ℎ,𝑃ℎ ) is a constant associated to the
𝑔𝑎𝑠→𝑙𝑖𝑞
phenol-phenol lateral interactions on the surface in water, and ∆∆𝐺 (∆∆𝐺𝑃ℎ (𝜃𝑃ℎ =
0)) is the aqueous-phase effect on the low coverage phenol adsorption on the surface. SSR
is the sum of squared residuals, which shows the error associated with each fit.
111
Figure 4.4: Temperature dependence of the aqueous-phase effect on the free energy of low
coverage adsorption of phenol and carbon monoxide over a Pt(111) surface (∆∆𝐺 =
𝑔𝑎𝑠→𝑙𝑖𝑞
∆∆𝐺𝐴𝑑𝑠𝑜𝑟𝑏𝑎𝑡𝑒 (𝜃𝐴𝑑𝑠𝑜𝑟𝑏𝑎𝑡𝑒 = 0)). The error bars indicate 95% confidence intervals based
on limited water sampling and multiple (three) independent simulations.
112
4.8 Bibliography
4. Bronsted JN. Acid and Basic Catalysis. Chemical Reviews 5, 231-338 (1928).
10. Lytken O, Lew W, Campbell CT. Catalytic reaction energetics by single crystal
adsorption calorimetry: hydrocarbons on Pt(111). Chemical Society Reviews 37,
2172-2179 (2008).
113
12. Varghese JJ, Mushrif SH. Origins of complex solvent effects on chemical
reactivity and computational tools to investigate them: a review. Reaction
Chemistry & Engineering 4, 165-206 (2019).
15. Bertero NM, Trasarti AF, Acevedo MC, Marchi AJ, Apesteguia CR. Solvent
effects in solid acid-catalyzed reactions: The case of the liquid-phase
isomerization/cyclization of citronellal over SiO2-Al2O3. Molecular Catalysis
481, (2020).
19. Mellmer MA, Sener C, Gallo JMR, Luterbacher JS, Alonso DM, Dumesic JA.
Solvent Effects in Acid-Catalyzed Biomass Conversion Reactions. Angewandte
Chemie International Edition 53, 11872-11875 (2014).
114
22. Zhao JJ, et al. Suppressing Metal Leaching in a Supported Co/SiO2 Catalyst with
Effective Protectants in the Hydroformylation Reaction. Acs Catalysis 10, 914-
920 (2020).
23. Kulal AB, Kasabe MM, Jadhav PV, Dongare MK, Umbarkar SB. Hydrophobic
WO3/SiO2 catalyst for the nitration of aromatics in liquid phase. Applied
Catalysis a-General 574, 105-113 (2019).
24. He R, Davda RR, Dumesic JA. In situ ATR-IR spectroscopic and reaction
kinetics studies of water-gas shift and methanol reforming on Pt/Al2O3 catalysts
in vapor and liquid phases. J Phys Chem B 109, 2810-2820 (2005).
28. Zhang L, Karim AM, Engelhard MH, Wei ZH, King DL, Wang Y. Correlation of
Pt-Re surface properties with reaction pathways for the aqueous-phase reforming
of glycerol. Journal of Catalysis 287, 37-43 (2012).
29. Tupy SA, et al. Correlating Ethylene Glycol Reforming Activity with In Situ
EXAFS Detection of Ni Segregation in Supported NiPt Bimetallic Catalysts. Acs
Catalysis 2, 2290-2296 (2012).
30. Karim AM, et al. In Situ X-ray Absorption Fine Structure Studies on the Effect of
pH on Pt Electronic Density during Aqueous Phase Reforming of Glycerol. Acs
Catalysis 2, 2387-2394 (2012).
31. Zope BN, Hibbitts DD, Neurock M, Davis RJ. Reactivity of the Gold/Water
Interface During Selective Oxidation Catalysis. Science 330, 74-78 (2010).
32. Taylor CD, Neurock M. Theoretical insights into the structure and reactivity of
the aqueous/metal interface. Curr Opin Solid St M 9, 49-65 (2005).
115
33. Yoon Y, Rousseau R, Weber RS, Mei DH, Lercher JA. First-Principles Study of
Phenol Hydrogenation on Pt and Ni Catalysts in Aqueous Phase. J Am Chem Soc
136, 10287-10298 (2014).
36. Clabaut P, Schweitzer B, Götz AW, Michel C, Steinmann SN. Solvation Free
Energies and Adsorption Energies at the Metal/Water Interface from Hybrid
Quantum-Mechanical/Molecular Mechanics Simulations. Journal of Chemical
Theory and Computation 16, 6539-6549 (2020).
37. Steinmann SN, Sautet P, Michel C. Solvation free energies for periodic surfaces:
comparison of implicit and explicit solvation models. Physical Chemistry
Chemical Physics 18, 31850-31861 (2016).
38. Zhang X, DeFever RS, Sarupria S, Getman RB. Free Energies of Catalytic
Species Adsorbed to Pt(111) Surfaces under Liquid Solvent Calculated Using
Classical and Quantum Approaches. Journal of Chemical Information and
Modeling 59, 2190-2198 (2019).
39. Ward BM, Getman RB. Molecular simulations of physical and chemical
adsorption under gas and liquid environments using force field- and quantum
mechanics-based methods. Molecular Simulation 40, 678-689 (2014).
40. Gould NS, et al. Understanding solvent effects on adsorption and protonation in
porous catalysts. Nature Communications 11, 1060 (2020).
41. Mushrif SH, Caratzoulas S, Vlachos DG. Understanding solvent effects in the
selective conversion of fructose to 5-hydroxymethyl-furfural: a molecular
dynamics investigation. Physical Chemistry Chemical Physics 14, 2637-2644
(2012).
116
42. Bregante DT, et al. Cooperative Effects between Hydrophilic Pores and Solvents:
Catalytic Consequences of Hydrogen Bonding on Alkene Epoxidation in Zeolites.
J Am Chem Soc 141, 7302-7319 (2019).
47. Mattsson TR, Paddison SJ. Methanol at the water-platinum interface studied by
ab initio molecular dynamics. Surf Sci 544, L697-L702 (2003).
49. Yang J, Dauenhauer PJ, Ramasubramaniam A. The role of water in the adsorption
of oxygenated aromatics on Pt and Pd. J Comput Chem 34, 60-66 (2013).
50. Klamt A. Conductor-Like Screening Model for Real Solvents - a New Approach
to the Quantitative Calculation of Solvation Phenomena. J Phys Chem-Us 99,
2224-2235 (1995).
52. Dyson PJ, Jessop PG. Solvent effects in catalysis: rational improvements of
catalysts via manipulation of solvent interactions. Catalysis Science &
Technology 6, 3302-3316 (2016).
117
53. Valgimigli L, Banks JT, Ingold KU, Lusztyk J. Kinetic Solvent Effects on
Hydroxylic Hydrogen Atom Abstractions Are Independent of the Nature of the
Abstracting Radical. Two Extreme Tests Using Vitamin E and Phenol. J Am
Chem Soc 117, 9966-9971 (1995).
54. Zhang YK, Liu HY, Yang WT. Free energy calculation on enzyme reactions with
an efficient iterative procedure to determine minimum energy paths on a
combined ab initio QM/MM potential energy surface. J Chem Phys 112, 3483-
3492 (2000).
55. Hu H, Lu ZY, Parks JM, Burger SK, Yang WT. Quantum mechanics/molecular
mechanics minimum free-energy path for accurate reaction energetics in solution
and enzymes: Sequential sampling and optimization on the potential of mean
force surface. J Chem Phys 128, 034105:034101-034118 (2008).
56. Hu H, Lu ZY, Yang WT. QM/MM minimum free-energy path: Methodology and
application to triosephosphate isomerase. J Chem Theory Comput 3, 390-406
(2007).
60. Sprowl LH, Campbell CT, Árnadóttir L. Hindered Translator and Hindered Rotor
Models for Adsorbates: Partition Functions and Entropies. The Journal of
Physical Chemistry C 120, 9719-9731 (2016).
61. Bajpai A, Mehta P, Frey K, Lehmer AM, Schneider WF. Benchmark First-
Principles Calculations of Adsorbate Free Energies. ACS Catalysis 8, 1945-1954
(2018).
62. Reed AE, Weinstock RB, Weinhold F. Natural-Population Analysis. J Chem Phys
83, 735-746 (1985).
118
63. Manz TA, Limas NG. Introducing DDEC6 atomic population analysis: part 1.
Charge partitioning theory and methodology. RSC Advances 6, 47771-47801
(2016).
70. Burow AM, Sierka M, Dobler J, Sauer J. Point defects in CaF2 and CeO2
investigated by the periodic electrostatic embedded cluster method. J Chem Phys
130, 174710:174711-174711 (2009).
72. Bennett C. Efficient Estimation of Free Energy Differences from Monte Carlo
Data. J Comput Phys 22, 245-268 (1976).
73. Hensley AJR, et al. DFT-Based Method for More Accurate Adsorption Energies:
An Adaptive Sum of Energies from RPBE and vdW Density Functionals. The
Journal of Physical Chemistry C 121, 4937-4945 (2017).
119
74. Singh N, Sanyal U, Fulton JL, Gutiérrez OY, Lercher JA, Campbell CT.
Quantifying Adsorption of Organic Molecules on Platinum in Aqueous Phase by
Hydrogen Site Blocking and in Situ X-ray Absorption Spectroscopy. ACS
Catalysis 9, 6869-6881 (2019).
77. Carey SJ, Zhao W, Mao Z, Campbell CT. Energetics of Adsorbed Phenol on
Ni(111) and Pt(111) by Calorimetry. The Journal of Physical Chemistry C 123,
7627-7632 (2019).
78. Kudchadker SA, Kudchadker AP, Wilhoit RC, Zwolinski BJ. Ideal gas
thermodynamic properties of phenol and cresols. J Phys Chem Ref Data 7, 417-
423 (1978).
79. Sander R. Compilation of Henry's law constants (version 4.0) for water as solvent.
Atmospheric Chemistry and Physics 15, 4399-4981 (2015).
80. Faheem M, Suthirakun S, Heyden A. New Implicit Solvation Scheme for Solid
Surfaces. J Phys Chem C 116, 22458-22462 (2012).
83. Heinz H, Vaia RA, Farmer BL, Naik RR. Accurate Simulation of Surfaces and
Interfaces of Face-Centered Cubic Metals Using 12-6 and 9-6 Lennard-Jones
Potentials. J Phys Chem C 112, 17281-17290 (2008).
120
APPENDIX A
Zare, M.; Solomon R. V.; Yang, W.; Yonge, A.; Heyden, A. The Journal of Physical
Chemistry C, 2020, 124, 16488-16500.
121
Table A.1: Solvent effect on the stability of various adsorbed species in the propionic acid
conversion to ethane and ethylene over a Ni (111) catalyst surface model at two different
temperatures of 298 K and 473 K based on two implicit solvation schemes: iSMS and
VASPsol. ΔΔGrxn indicates the difference in the adsorption free energy of the
corresponding intermediates in the presence and absence of water. Asterisk (*) represents
a surface adsorption site and multiple asterisks are indicative of the number of occupied
active sites.
ΔΔGrxn, eV
122
Table A.2: Solvent effect on the stability of transition states in the propionic acid conversion to ethane and ethylene over a Ni (111)
catalyst surface model at two different temperatures of 298 K and 473 K based on two implicit solvation schemes: iSMS and VASPsol.
Solvent effect on TS indicates the difference in the adsorption free energy of the corresponding transition state in the presence and
absence of water.
ΔΔGrxn
# Reaction T=298 K T=473 K
iSMS VASPsol iSMS VASPsol
1 CH3CH2COOH* + 3* → CH3CH2CO*** + OH* 0.03 0.13 0.00 0.10
2 CH3CH2COOH* + 2* → CH3CHCOOH** + H* -0.04 0.14 -0.07 0.12
3 CH3CH2CO***+ 2* → CH3CH2** + CO*** -0.10 -0.04 -0.14 -0.01
4 CH3CH2CO*** → CH3CHCO** + H* -0.05 0.00 -0.06 0.01
5 CH3CHCOOH** + * → CH3CHCO** + OH* 0.02 -0.01 0.01 -0.01
6 CH3CHCOOH** + 3* → CH2CHCOOH**** + H* -0.02 -0.03 -0.02 -0.01
7 CH3CHCOOH** + 2*→CH3CCOOH*** + H* -0.03 0.02 -0.03 0.02
CH3CHCO** + 4* → CH3CH*** + CO***
125
ΔΔGact
# Reaction T=298 K T=473 K
iSMS VASPsol iSMS VASPsol
1 CH3CH2COOH* + 3* → CH3CH2CO*** + OH* 0.04 0.29 0.00 0.29
2 CH3CH2COOH* + 2* → CH3CHCOOH** + H* -0.03 0.05 -0.03 0.03
3 CH3CH2CO***+ 2* → CH3CH2** + CO*** -0.02 0.01 -0.03 0.02
4 CH3CH2CO*** → CH3CHCO** + H* -0.03 0.01 -0.03 0.05
5 CH3CHCOOH** + * → CH3CHCO** + OH* -0.02 0.13 -0.03 0.13
6 CH3CHCOOH** + 3* → CH2CHCOOH**** + H* -0.01 -0.06 0.00 -0.04
7 CH3CHCOOH** + 2*→CH3CCOOH*** + H* 0.00 0.00 0.00 0.02
CH3CHCO** + 4* → CH3CH*** + CO***
127
PCO \
PH2 0.01 0.1 1 10 30
(bar)
0.0001 3.99×10-6 1.98×10-8 8.72×10-11 7.23×10-13 7.72×10-14
PCO \ PH2
0.01 0.1 1 10 30
(bar)
129
Table A.7: Number of occupied sites used for microkinetic modeling result displayed in
Table A.6.
130
Table A.8: Solvent effect on the stability of various adsorbed species in the propionic acid
conversion to ethane and ethylene over a Ni (111) catalyst surface model at 473 K in the
presence of two different solvents, water and 1,4-dioxane, using iSMS methodology.
ΔΔGrxn indicates the difference in the adsorption free energy of the corresponding
intermediates in the presence and absence of solvent. Asterisk (*) represents a surface
adsorption site and multiple asterisks are indicative of the number of occupied active sites.
Note that calculations for solvents were performed with the help of the COSMO-RS
package with three different Ni cavity radii: with default value, with a 10% increased value
and a 10% decreased value relative to the default.
ΔΔGrxn, eV @ 473 K
131
Table A.9: Solvent effect on the reaction free energies of all elementary surface reactions of the propionic acid conversion to ethane and
ethylene over a Ni (111) catalyst surface model at 473 K based on iSMS methodology in the presence of water and 1,4-dioxane. ΔΔGrxn
indicates the reaction free energy difference between corresponding reaction in solvent and in a gas phase. Note that calculations for
solvents were performed with the help of the COSMO-RS package with three different Ni cavity radii: with default value, with a 10%
increased value and a 10% decreased value relative to the default.
ΔΔGrxn, eV @ 473 K
Water 1,4-dioxane
ΔΔGact, eV @ 473 K
Water 1,4-dioxane
136
Figure A.1: Adsorption of different sized molecules (CO, CH3CH2, and CH3CH2C) on a
Ni (111) surface model with various cluster sizes used in iSMS calculations in liquid water
at 473 K. We employed a 3x2 cluster model for all surface reactions since the
computational cost for larger models increases dramatically while the solvent effect does
not change significantly. For adsorption/desorption processes of saturated molecules such
as C2H4, C3H6, C2H2, CH3CH2COOH, CO, CO2, H, H2O, where error cancellation of
solvation effects is smaller than for surface reactions, a 3x4 model with 86 metal atoms
was used.
137
a)
b)
Figure A.2: Lateral interaction correction terms for various intermediates using a 1/4 ML
coverage of CO and H. Shown are the correction terms of the lateral interaction effect of
H (a) and CO (b) for each individual adsorbed species.
138
a)
b)
Figure A.3: Lateral interaction correction terms for the transition state of each elementary
step using 1/4 ML coverages of CO and H. Shown are the correction terms of the lateral
interaction effect of H (a) and CO (b) on the transition state of each elementary step.
139
Figure A.4: Partial charges computed using the DDEC6 charge model on adsorbed CO (top
left), adsorbed PAC (top right), and coadsorbed CO and PAC on a Ni (111) catalyst surface
model. In the coadsorbed state, CO is a Lewis acid, which accepts electrons, while PAC is
a Lewis base, which donates electrons. This coadsorbed Lewis acid and base pair explains
the attractive interaction between adsorbed CO and PAC.
140
a) b)
Figure A.5: Differential zero-point energy corrected (ZPC) energy of adding a) an extra
CO as a function of number of CO already adsorbed on a Ni(111) surface containing 12
surface atoms with (3 × 2√3) periodicity and b) an extra H as a function of number of H
already adsorbed on the same Ni surface. Adding the fifth CO increases the energy
significantly (by over 0.45 eV) which we attribute to the fact that the surface is essentially
fully covered with 4 CO and adding an extra CO weakens the bond strength of adsorbed
CO molecules on the surface due to a strong repulsive interaction between them. On the
other hand, the differential ZPC barely increases when adding an extra hydrogen atom.
Consequently, it can be concluded that a CO molecule occupies 3 Ni sites on the surface
while H adsorbs on only 1 Ni site.
141
APPENDIX B
Zare, M.; Saleheen, M.; Kundu, S. K.; Heyden, A. Communications Chemistry 3, 187,
2020.
142
Table B.1: Simulation box size used for (111) facet of each metal surface.
Metal Simulation box size, Å3
Ni 39.75×43.03×49.01
Pd 44.53×48.20×49.01
Pt 44.98×48.69×49.01
Cu 41.04×44.43×49.01
Ag 46.85×50.71×49.01
Au 47.00×50.88×49.01
Table B.2: Lennard-Jones parameters of all elements used in this study. The parameters are
𝜎 12 𝜎 6
based on 12-6 Lennard-Jones potential, 𝑉𝑙𝑗 = 4𝜀 [( 𝑟 ) − ( 𝑟 ) ]. Ow and Hw represent
oxygen and hydrogen of TIP3P water model, respectively.
143
Table B.3: Average rotational correlation time (in picosecond) of liquid water molecules
residing in a 5 Å radius of an adsorbed ethylene glycol (C2H5OH*) species on a (111) facet
of six transition metal surfaces studied in the present work. The predicted correlation
functions have been fitted with three exponential functions to acquire average rotational
correlation times.
O-H cleavage C-H cleavage
Surface RS TS TS
Ni 218 147 79
Pd 227 101 223
Pt 183 46 53
Cu 115 67 21
Ag 82 111 196
Au 73 43 32
144
Table B.4: Number of hydrogen bonds formed from the interactions of the TIP3P water model and the reacting moiety in reactant state
(RS) and transition state (TS) of O-H and C-H bond cleavage of ethylene glycol over a (111) facet of six transition metal surfaces at 423
K. Acceptor indicates that OH functional groups of ethylene glycol accepts a hydrogen bond donated by the oxygen of water while
Donor indicates vice versa. All numbers are computed based on three independent simulations and hence possess 95% confidence
intervals (assuming a normal distribution).
RS TS RS TS
RS -0.11 -0.10 -0.75 -0.79 0.21 0.21 0.21 0.15 0.45 0.48
Ni
TS -0.11 -0.09 -0.80 -0.78 0.21 0.22 0.21 0.10 -0.12 0.47
RS -0.08 -0.10 -0.75 -0.72 0.21 0.20 0.22 0.17 0.50 0.47
Pd
TS -0.06 -0.11 -0.71 -0.71 0.23 0.18 0.22 0.15 0.10 0.47
RS
146
-0.11 -0.10 -0.75 -0.71 0.22 0.23 0.22 0.19 0.49 0.51
Pt
TS -0.12 -0.06 -0.68 -0.69 0.24 0.14 0.23 0.20 0.17 0.48
O-H
RS -0.10 -0.10 -0.75 -0.79 0.21 0.21 0.20 0.17 0.47 0.48
Cu
TS -0.10 -0.10 -0.86 -0.78 0.19 0.20 0.21 0.21 -0.16 0.48
RS -0.10 -0.09 -0.75 -0.77 0.20 0.21 0.20 0.18 0.48 0.48
Ag
TS -0.11 -0.09 -0.78 -0.79 0.20 0.20 0.19 0.14 -0.27 0.47
RS -0.13 -0.12 -0.71 -0.70 0.19 0.18 0.18 0.17 0.48 0.47
Au
TS -0.13 -0.12 -0.78 -0.69 0.20 0.18 0.20 0.12 -0.08 0.48
RS -0.11 -0.10 -0.75 -0.79 0.21 0.21 0.21 0.15 0.45 0.48
Ni
TS -0.23 -0.12 -0.73 -0.77 0.22 0.20 0.22 -0.17 0.45 0.48
RS -0.08 -0.10 -0.75 -0.72 0.22 0.20 0.22 0.17 0.50 0.47
Pd
TS -0.06 -0.12 -0.75 -0.68 0.25 0.21 0.23 0.03 0.50 0.48
RS -0.11 -0.10 -0.75 -0.71 0.22 0.23 0.22 0.19 0.49 0.51
Pt
TS -0.07 -0.11 -0.72 -0.71 0.25 0.24 0.24 0.13 0.50 0.51
C-H
RS -0.11 -0.10 -0.75 -0.79 0.21 0.21 0.21 0.17 0.47 0.48
Cu
TS
147
-0.21 -0.12 -0.72 -0.77 0.21 0.21 0.21 -0.32 0.47 0.49
RS -0.10 -0.09 -0.75 -0.77 0.20 0.21 0.20 0.18 0.48 0.48
Ag
TS -0.06 -0.13 -0.68 -0.75 0.21 0.23 0.22 -0.35 0.50 0.48
RS -0.13 -0.12 -0.71 -0.70 0.19 0.18 0.18 0.17 0.48 0.47
Au
TS 0.03 -0.18 -0.64 -0.68 0.21 0.20 0.20 -0.19 0.50 0.48
Table B.6: Pearson correlation coefficient (PCC) between descriptors used in this study. It has a value between +1 and −1, where +1 is
total positive linear correlation, 0 is no linear correlation, and −1 is a total negative linear correlation. ∆𝐺 𝑎𝑐𝑡,𝑔𝑎𝑠 shows the free energy
of activation in vapor phase, H-bond denotes the change in mean of total hydrogen bonding (acceptor + donor) going from RS to TS,
MC (molecular charge-transfer) represents the change in the absolute sum of partial charges on the reacting moiety going from RS to
TS, and finally BC (cleaving-bond charge-transfer) expresses the change in sum of partial charges on the cleaving bond going from RS
to TS (see Table 3.2 in the main text for the value of descriptors).
MC,
Cleavage MC, BC H-bond, MC H-bond, BC H-bond, ∆𝑮𝒂𝒄𝒕,𝒈𝒂𝒔 BC, ∆𝑮𝒂𝒄𝒕,𝒈𝒂𝒔
∆𝑮𝒂𝒄𝒕,𝒈𝒂𝒔
O-H -0.99 0.83 -0.78 0.56 -0.67 0.65
C-H -0.96 0.66 -0.43 -0.54 -0.45 0.22
Table B.7: Data of a linear model, (𝑦 − 𝑦̅) = 𝛼1 (𝑓1 − 𝑓̅1 ) + 𝛼2 (𝑓2 − 𝑓̅2 ), for estimating the solvent effect (∆∆𝐺 𝑎𝑐𝑡 = ∆𝐺 𝑎𝑐𝑡,𝑙𝑖𝑞 −
∆𝐺 𝑎𝑐𝑡,𝑔𝑎𝑠 ), computed by eSMS, on O-H and C-H bond cleavages of ethylene glycol at 423 K. We have not considered the combination
148
of MC and BC since they are totally correlated (see Table B.7). The fits are based on 6 data points for each cleavage which correspond
to six transition metals. Also, 𝑀𝐴𝐸 represents the mean of absolute errors associated with the fit, in eV, defined as the difference between
estimated and eSMS values.
O-H cleavage C-H cleavage
𝒇𝟏 , 𝒇𝟐 𝜶𝟏 𝜶𝟐 𝑀𝐴𝐸 𝜶𝟏 𝜶𝟐 𝑀𝐴𝐸
𝑎𝑐𝑡,𝑔𝑎𝑠
∆𝐺 , BC 0.13 0.27 0.14 -0.04 0.54 0.03
𝑎𝑐𝑡,𝑔𝑎𝑠
∆𝐺 , MC 0.06 -0.03 0.14 -0.10 -0.48 0.04
𝑎𝑐𝑡,𝑔𝑎𝑠
∆𝐺 , H-bond 0.02 0.10 0.14 -0.27 -0.44 0.04
H-bond, BC 0.43 0.48 0.12 0.03 0.61 0.04
H-bond, MC 0.30 -0.15 0.14 0.19 -0.69 0.04
Table B.8: Data of a quadratic model, (𝑦 − 𝑦̅) = 𝛼1 (𝑓1 − 𝑓̅1 ) + 𝛼2 (𝑓2 − 𝑓̅2 ) + 𝛼3 (𝑓1 − 𝑓̅1 )2 + 𝛼4 (𝑓2 − 𝑓̅2 )2 , for estimating the solvent
effect (∆∆𝐺 𝑎𝑐𝑡 = ∆𝐺 𝑎𝑐𝑡,𝑙𝑖𝑞 − ∆𝐺 𝑎𝑐𝑡,𝑔𝑎𝑠 ), computed by eSMS, on O-H bond cleavage of ethylene glycol at 423 K.
𝒇𝟏 , 𝒇𝟐 𝜶𝟏 𝜶𝟐 𝜶𝟑 𝜶𝟒 𝑀𝐴𝐸
∆𝐺 𝑎𝑐𝑡,𝑔𝑎𝑠 , BC -0.08 0.49 0.92 -4.38 0.02
∆𝐺 𝑎𝑐𝑡,𝑔𝑎𝑠 , MC -0.18 -0.44 1.29 -2.11 0.01
𝑎𝑐𝑡,𝑔𝑎𝑠
∆𝐺 , H-bond 0.11 0.44 -0.44 2.12 0.08
H-bond, BC 0.56 0.59 1.82 -2.68 0.04
H-bond, MC 0.90 -0.62 2.26 -1.19 0.04
149
Table B.9: Solvent effects on the free energy of activation (∆∆𝐺 𝑎𝑐𝑡 = ∆𝐺 𝑎𝑐𝑡,𝑙𝑖𝑞 − ∆𝐺 𝑎𝑐𝑡,𝑔𝑎𝑠 ) of O-H and C-H bond cleavage of ethylene
glycol on the (111) facet of six transition metal surfaces using the iSMS solvation scheme. Note that calculations were performed with
the help of the COSMO-RS package with three different metal cavity radii: with default value (2.2230 Å for all transition metal
elements), with a 10% increased value (2.4453 Å) and a 10% decreased (2.0007 Å) value relative to the default.
default +10% -10%
Ni -0.08 -0.06 -0.15
Cu -0.09 -0.08 -0.09
Ag -0.06 -0.06 -0.08
O-H cleavage
Pt -0.06 -0.05 -0.12
Pd 0.03 -0.01 0.02
Au -0.03 0.00 -0.05
Ni -0.04 -0.04 -0.04
Cu -0.12 -0.09 -0.15
Ag -0.03 -0.03 -0.05
150
C-H cleavage
Pt -0.09 -0.12 -0.15
Pd -0.01 -0.06 0.00
Au -0.17 -0.19 -0.08
Table B.10: Average fraction of TIP3P water molecules with different orientations (see Figure B.3 for explanation of the different
orientations) within the first water layer of the surface (see first peak in height distribution function of the water O atom in Figure B.4).
The data indicate that water orients itself similar across the metal surfaces and ~68% of them are in “parallel” orientation, ~23% in H-
up orientation, and only 9% in H-down orientation (which lets the water hardly form hydrogen bonds with the condensed phase).
Water Orientation Ni Cu Ag Pt Au Pd
Parallel 68.17±0.07 68.10±0.40 67.85±0.11 68.08±0.29 68.25±0.12 67.93±0.25
H-up 22.93±0.10 23.15±0.25 22.78±0.07 23.05±0.15 22.63±0.07 22.81±0.26
H-down 8.90±0.16 8.75±0.17 9.38±0.07 8.87±0.17 9.12±0.12 9.26±0.15
151
Figure B.1: Free-energy profiles for C-H bond cleavage of ethylene glycol in vapor and
aqueous phases over the (111) facet of six transition metal surfaces at 423 K without
considering vibrational contributions to the partition function. See Table 3.1 for
corresponding data that include vibrational contributions. The aqueous phase profile
portrays the average of three or more independent eSMS calculations possessing 95%
confidence intervals smaller than ±0.05 eV. The analogous plot for O-H bond cleavage is
provided in the main text.
152
Figure B.2: Graphical representation of a geometric hydrogen bond definition used in this
study. In the picture shown, water is the donor of hydrogen bonding, and ethylene glycol
is the acceptor of hydrogen bonding.
153
Figure B.3: Graphical representation of criterion used to distinguish different water
orientations at the surface. A water molecule in the first layer next to the surface is
considered in H-up conformation if the difference in z-coordinate of the H and O atom is
larger than half the H-H distance of the water molecule; this corresponds for TIP3P water
to an angle q show above larger than 52.26⁰. Water molecules that are not considered H-
up or H-down are labeled “parallel”.
154
Figure B.4: Height distribution function of water O and H over (111) facet of six transition
metal surfaces
155
APPENDIX C
Zare, M.; Saleheen, M.; Singh, N.; Uline, M. J.; Faheem, M.; Heyden, A.
156
C.1 Computational Details
Planewave DFT Calculations. Vapor phase DFT calculations were carried out
employing periodic boundary conditions and using the Vienna Ab Initio Simulation
method was utilized to avoid the singularities of Kohn-Sham wavefunctions at the nuclear
describe the van-der-Waals interactions. Brillouin zone integrations have been performed
with a 4×4×1 Monkhorst-Pack8 k-point grid, and electronic wavefunctions at each k-point
were expanded using a discrete plane-wave basis set with kinetic energies limited to 400
eV. For phenol adsorption, we increased the energy cutoff to 450 eV. Fractional
contributions due to the smearing accurately. The modified version of the Makov-Payne10
method with Harris corrections were applied to the stress-tensor and forces to calculate
dipole and quadrupole corrections (along the surface normal) to the total energy. Using the
supercell approach, a 4×4-unit cell with four layers of metal atoms, in which the bottom
two layers were fixed while the top two layers were relaxed, has been constructed
employing a 15 Å vacuum on top of the surface, which restricted the interaction between
the periodic images along the surface normal. The self-consistent field (SCF) calculations
were converged to 1.0×10-7 eV. A force criterion of 0.02 eV Å-1 was used on relaxed atoms
for geometry optimizations. The transition state structure of the O-H bond cleavage of
157
ethylene glycol (EG) over a Cu(111) surface model was located using a combination of
climbing-image nudged elastic band11, 12 and dimer13, 14 methods. Finally, the minima and
the first order saddle points were validated by computing the Hessian matrix and
vibrational spectra.
been carried out using the TURBOMOLE 7.2 program package.15, 16, 17 We used a cluster
model consisting of two layers with a total number of 51 metal atoms. Convergence of the
total QM/MM energy with respect to the lateral size and depth of the cluster geometry can
be found elsewhere.18 To represent the adsorbate atoms and the metal atoms, an improved
version of the default TURBOMOLE basis sets (def-bases) with split valence and
polarization functions (def2-SVP)19, 20 was used. Pt atoms were represented using scalar
relativistic effective core potentials (ECPs) in conjunction with split valence basis sets
using the PBE functional5, 6 (for phenol adsorption Grimme’s DFT-D3 methodology7 was
used). Finally, the RI-J approximation with auxiliary basis sets was used to approximate
the coulomb integrals.22, 23 Single point energy calculations were performed with an SCF
energy convergence criterion of 1.0 × 10-7 Hartree, and a Gauss-Chebyshev type spherical
MD simulations. Using the supercell approach, the initial 4×4 unit cell for each clean
surface metal was augmented laterally to a 16×20 surface with further vacuum added in
the Z-direction resulting in a simulation box comprising of 1280 metal atoms. As a result,
158
a simulation box size of 44.98 Å × 48.69 Å × 49.01 Å for a Pt(111) surface and
41.04 Å × 44.43 Å × 49.01 Å for a Cu(111) surface were achieved. The QM cluster atoms
were then substituted into the constructed surface. The simulation box height was selected
based on the work from Behler et al.,25 who found that simulations of metal-water
interfaces should contain a water layer of ~40 Å height. Subsequently, the experimental
saturated liquid water density of ~1.0 g/cm3 at 298 K for CO and phenol adsorption
calculations on Pt(111) and Cu(111) surfaces and of ~0.92 g/cm3 at 423 K for O-H bond
Pt(111) were achieved by packing the simulation box of the Pt(111) and Cu(111) surfaces
with ~3000 and ~2400 water molecules, respectively. All metal and adsorbate atoms were
kept fixed while the geometry of water molecules was constricted to that of TIP3P 26
geometry with the RATTLE algorithm,27 a velocity version of the SHAKE algorithm,28 in
conjunction with the velocity Verlet (VV) integrator29 to solve Newton’s equations of
motion. The force field parameters of liquid water were obtained from the TIP3P model.
The Lennard-Jones parameters were obtained from the OPLS force field30, 31, 32 for EG and
phenol and from Straub et al.33 for CO. In addition to the OPLS parameters, the Lennard-
Jones parameters from the combined B3LYP/6-31_G*/AMBER Potential34 were used for
the hydrogen atoms of EG. Lennard-Jones parameters for hydrogen atoms are important in
QM/MM optimizations that permit hydrogen atoms to approach water molecules and leave
159
C.1. For O-H bond breaking of EG on the Cu(111) surface, and EG adsorption on the
Pt(111) surface, the charges for the QM atoms were estimated using the natural population
analysis (NPA)36 while for CO and phenol adsorptions on the Pt(111) and Cu(111) surfaces
the DDEC637 (a refinement of the Density Derived Electrostatic and Chemical (DDEC)
approach) charge model was used (to compare the performance of NPA and DDEC6 in our
scheme, we also repeated the calculations of CO adsorption on Pt(111) with the NPA
charge model). To describe the interaction of the TIP3P water point charges with the
were carried out in a canonical ensemble (NVT) with Nosé-Hoover thermostat39, 40 with 1
ps relaxation time constant. Electrostatic interactions were accounted for by using the
Smoothed Particle Mesh Ewald (SPME) method41 with automatic parameter optimization
with default SPME precision. A 12 Å cutoff radius was adopted for the van-der-Waals
interactions and the transition between short and long range electrostatic interactions. If
not specified differently, all systems were equilibrated for 250 ps and sampled for 1000 ps
(1ns) using a 1 fs timestep to obtain 1000 MM conformations, which are 1 ps apart, for
every 1 ps.
160
Table C.1: Lennard-Jones parameters of all elements in this study based on 𝑉𝑖𝑗 =
12 6
𝜎 𝜎
4𝜀𝑖𝑗 [( 𝑟𝑖𝑗 ) − ( 𝑟 𝑖𝑗 ) ], where 𝜀 is the depth of the potential well and 𝜎 is the finite distance
𝑖𝑗 𝑖𝑗
at which the inter-particle potential is zero. Ow and Hw represent oxygen and hydrogen of
TIP3P water model, respectively. ‘Ho’ in the phenol molecule indicates the hydrogen of
the OH functional group.
of the system is a key for a reliable estimate of ensemble averages and a prediction of the
liquid phase effect on free energies of elementary processes.42, 43 Yet, extensive exploration
of the configuration space for the relevant configurations is challenging due to the
substantial number of solvent molecules in the system. In addition, it is not a priori obvious
how much water sampling is required for adsorbed species on metal surfaces.
161
Consequently, the average rotational correlation time for water molecules in proximity (up
to 5 Å) of the adsorbed moieties for each free energy calculation is computed. And the
configuration space is then sampled long enough, relative to the average rotational
correlation time, to assure that relevant configurations of the system are sampled
Figure C.1. Table C.2 also lists the average rotational correlation time and constants of the
Table C.2: Average rotational correlation time (RCT)44 and constants (in pico-second) of
𝑡 𝑡 𝑡
−( ) −( ) −( )
3-Exponential fits (𝐹(𝑡) = 𝑎1 𝑒 𝜏1 + 𝑎2 𝑒 𝜏2 + (1 − 𝑎1 − 𝑎2 )𝑒 𝜏3 ) in Figure C.1. The
𝑎1 𝜏1 2 +𝑎2 𝜏2 2 +(1−𝑎1 −𝑎2 )𝜏3 2
average rotational correlation time is computed as: 𝑅𝐶𝑇 = .
𝑎1 𝜏1 +𝑎2 𝜏2 +(1−𝑎1 −𝑎2 )𝜏3
Phenol Phenol
EG on EG on CO on CO on
on on
Cu(111) Pt(111) Cu(111) Pt(111)
Cu(111) Pt(111)
𝒂𝟏 0.127 0.102 0.520 0.534 0.305 0.394
𝒂𝟐 0.447 0.451 0.354 0.360 0.512 0.436
𝝉𝟏 122.491 86.866 18.707 46.349 54.223 60.506
𝝉𝟐 2.176 2.416 2.330 2.143 4.106 4.102
𝝉𝟑 0.271 0.290 0.069 0.047 0.188 0.186
RCT 115.0 76.6 17.4 45.0 48.5 56.5
162
Figure C.1: Rotational correlation time function of liquid water molecules residing in a 5
Å radius of adsorbed ethylene glycol (C2H6O2*) species on Cu(111) and Pt(111) at 423K,
of adsorbed carbon monoxide (CO*) on Cu(111) and Pt(111) at 298 K, and of adsorbed
phenol (C6H5OH*) on Cu(111) and Pt(111) at 298 K. The correlation time functions have
been fitted using three exponential functions to obtain the average rotational correlation
times (see Table C.2).
163
C.4 QM/MM Free Energy Calculation
The explicit Solvation model for Metal Surfaces (eSMS) methodology18, based on
QM/MM-(M)FEP (free energy perturbation)45, 46, which has been implemented in our
program packages and which was successfully employed in our previous studies,47, 48 was
used to carry out the free energy calculations. Computation of the free energy difference
between two states was carried out using the Bennett Acceptance Ratio (BAR),49 which is
a more efficient and precise way than exponential averaging (EXP) when two states are
sampled50 since it uses both the forward and reverse distributions simultaneously. Finally,
the whole free energy estimation procedure has been repeated three times, independently,
by different initialization of the velocities using different seeds for the pseudorandom
confidence intervals for evaluating the adsorption free energy of the investigated
were carried out only if these three experiments could not result in a 95% confidence
interval of lower than 0.05 eV. All uncertainties reported in this study are 95% confidence
intervals.
the default values for the VASPsol parameters. While these parameters are likely most
accurate only for simulations at 298 K and not at 423 K, they are optimized parameters of
the solvent model that cannot easily be obtained at other temperatures. All other
164
computational details for periodic implicit solvation calculations were kept the same as in
statistical mechanics.
2 2
𝑔−𝑙 𝜃𝐻 𝜃𝐻
𝐻2 (𝑔) + 2 ∗ (𝑙𝑖𝑞) ↔ 2𝐻 ∗ (𝑙𝑖𝑞) 𝐾𝑎𝑑𝑠,𝐻2 = 𝑓𝐻 = 𝑃𝐻 (C.2)
( 2 )𝜃∗2 ( 2 )𝜃∗2
1 𝑏𝑎𝑟 1 𝑏𝑎𝑟
1 = 𝜃∗ + 𝜃𝐻 + 𝑛𝜃𝑃ℎ (C.3)
where 𝜃𝑃ℎ , 𝜃𝐻 , and 𝜃∗ are the surface coverage of phenol, hydrogen, and the free site,
respectively, variable 𝑛 indicates the number of sites that a phenol molecule can occupy
𝑔−𝑙
when adsorbed on the surface; 𝐾𝑎𝑑𝑠 is the equilibrium constant of adsorption of an
adsorbate on the surface from its free state in the gas phase; 𝑓𝑃ℎ , and 𝑓𝐻 are the phenol and
hydrogen fugacities in the gas phase, respectively, which can be reduced to partial
pressures with ideal gas assumption. We note that in the following, we define the “surface
coverage” as the number of surface species divided by the number of surface atoms, which
is strictly speaking only the coverage for 𝑛=1. From Henry’s law we have
𝑚𝑜𝑙
𝑓𝑃ℎ ⁄1𝑏𝑎𝑟 = 𝑃𝑃ℎ ⁄1𝑏𝑎𝑟 = 𝐾𝐻,𝑃ℎ 𝐶𝑃ℎ ⁄( 𝐿
) (C.4)
where 𝐾𝐻,𝑃ℎ is the Henry’s constant of phenol and 𝐶𝑃ℎ is the phenol concentration in the
aqueous phase. In the following, we neglect the concentration and pressure unit but work
165
𝑔−𝑙 𝜃𝑃ℎ
𝐾𝑎𝑑𝑠,𝑃ℎ 𝐾𝐻,𝑃ℎ = (C.5)
𝐶𝑃ℎ 𝜃∗𝑛
We also have
𝑔−𝑙 𝑔𝑎𝑠 𝑔𝑎𝑠→𝑙𝑖𝑞 𝑔−𝑙 𝑔 𝑔𝑎𝑠→𝑙𝑖𝑞
𝐾𝑎𝑑𝑠,𝑃ℎ = 𝐾𝑎𝑑𝑠,𝑃ℎ 𝐾𝑃ℎ or ∆𝐺𝑎𝑑𝑠,𝑃ℎ = ∆𝐺𝑎𝑑𝑠,𝑃ℎ + ∆∆𝐺𝑃ℎ (C.6)
𝑔𝑎𝑠→𝑙𝑖𝑞 𝑔−𝑙
where ∆∆𝐺𝑃ℎ is the solvent effect on phenol adsorption, ∆𝐺𝑎𝑑𝑠,𝑃ℎ is the adsorption
𝑔
free energy of phenol in the aqueous phase from its free state in the gas phase, and ∆𝐺𝑎𝑑𝑠,𝑃ℎ
is free energy of phenol adsorption in the gas phase. Hence, (C.5) becomes
2
𝑔−𝑙 𝑔𝑎𝑠 𝑔𝑎𝑠→𝑙𝑖𝑞 𝜃𝐻 𝑔𝑎𝑠 𝑔𝑎𝑠→𝑙𝑖𝑞
𝐾𝑎𝑑𝑠,𝐻2 = 𝐾𝑎𝑑𝑠,𝐻2 𝐾𝐻2 = 𝑃𝐻 → 𝜃𝐻 = √𝐾𝑎𝑑𝑠,𝐻2 𝐾𝐻2 𝑃𝐻2 𝜃∗ (C.8)
( 2 )𝜃∗2
1 𝑏𝑎𝑟
𝑔𝑎𝑠
𝑔𝑎𝑠 −∆𝐺𝑃ℎ 𝑔𝑎𝑠 𝑔𝑎𝑠 𝑔𝑎𝑠 𝑔𝑎𝑠
𝐾𝑎𝑑𝑠,𝑃ℎ = 𝑒𝑥𝑝( 𝑅𝑇
) and ∆𝐺𝑃ℎ = ∆𝐺𝑃ℎ (𝜃𝑃ℎ = 0) + 𝛼𝑃ℎ,𝑃ℎ 𝜃𝑃ℎ + 𝛼𝑃ℎ,𝐻 𝜃𝐻 (C.10)
𝑔𝑎𝑠
𝑔𝑎𝑠 −∆𝐺𝐻 𝑔𝑎𝑠 𝑔𝑎𝑠 𝑔𝑎𝑠 𝑔𝑎𝑠
2
𝐾𝑎𝑑𝑠,𝐻2 = 𝑒𝑥𝑝( 𝑅𝑇
) and ∆𝐺𝐻2 = ∆𝐺𝐻2 (𝜃𝐻 = 0) + 𝛼𝐻,𝐻 𝜃𝐻 + 𝛼𝑃ℎ,𝐻 𝜃𝑃ℎ (C.11)
𝑔𝑎𝑠 𝑔𝑎𝑠→𝑙𝑖𝑞
1 = 𝜃∗ + 𝑛[𝐾𝑎𝑑𝑠,𝑃ℎ 𝐾𝐻,𝑃ℎ 𝐾𝑃ℎ 𝐶𝑃ℎ 𝜃∗𝑛 ] (C.12)
166
𝑔𝑎𝑠
𝑔𝑎𝑠 𝑔𝑎𝑠→𝑙𝑖𝑞 −𝛼𝑃ℎ,𝑃ℎ 𝜃𝑃ℎ
𝜃𝑃ℎ = 𝐾
⏟𝑎𝑑𝑠,𝑃ℎ (𝜃𝑃ℎ = 0)𝐾𝐻,𝑃ℎ 𝐾𝑃ℎ 𝐶𝑃ℎ 𝜃∗𝑛 𝑒𝑥𝑝( 𝑅𝑇
) (C.13)
𝛽
𝑔𝑎𝑠
𝑔𝑎𝑠→𝑙𝑖𝑞 −𝛼𝑃ℎ,𝑃ℎ 𝜃𝑃ℎ
1 = 𝜃∗ + 𝑛 [𝛽𝐾𝑃ℎ 𝐶𝑃ℎ 𝜃∗𝑛 𝑒𝑥𝑝( 𝑅𝑇
)] (C.14)
𝑔−𝑙
𝑔−𝑙 −∆𝐺𝑃ℎ 𝑔−𝑙 𝑔−𝑙 𝑙𝑖𝑞 𝑙𝑖𝑞
𝐾𝑎𝑑𝑠,𝑃ℎ = 𝑒𝑥𝑝( ) and ∆𝐺𝑃ℎ = ∆𝐺𝑃ℎ (𝜃𝑃ℎ = 0) + 𝛼𝑃ℎ,𝑃ℎ 𝜃𝑃ℎ + 𝛼𝑃ℎ,𝐻 𝜃𝐻 (C.15)
𝑅𝑇
𝑔𝑎𝑠→𝑙𝑖𝑞 𝑔𝑎𝑠
𝑔𝑎𝑠→𝑙𝑖𝑞 −∆𝛼𝑃ℎ,𝑃ℎ 𝜃𝑃ℎ −𝛼𝑃ℎ,𝑃ℎ 𝜃𝑃ℎ
1 = 𝜃∗ + 𝑛 𝛽𝐾𝑃ℎ (𝜃𝑃ℎ = 0)𝐶𝑃ℎ 𝜃∗𝑛 𝑒𝑥𝑝(
⏟ 𝑅𝑇
)𝑒𝑥𝑝( 𝑅𝑇
) or
𝑙𝑖𝑞
−𝛼 𝜃
𝑃ℎ,𝑃ℎ 𝑃ℎ
[ 𝑒𝑥𝑝(
𝑅𝑇
) ]
𝑙𝑖𝑞
𝑔𝑎𝑠→𝑙𝑖𝑞 −𝛼 𝜃𝑃ℎ
1 = 𝜃∗ + 𝑛𝜃𝑃ℎ = 𝜃∗ + 𝑛 [𝛽𝐾𝑃ℎ (𝜃𝑃ℎ = 0)𝐶𝑃ℎ 𝜃∗𝑛 𝑒𝑥𝑝( 𝑃ℎ,𝑃ℎ
𝑅𝑇
)] (C.17)
𝑙𝑖𝑞
𝑔𝑎𝑠→𝑙𝑖𝑞 −𝛼𝑃ℎ,𝑃ℎ 𝜃𝑃ℎ
𝜃𝑃ℎ = 𝛽𝐾𝑃ℎ (𝜃𝑃ℎ = 0)𝐶𝑃ℎ 𝜃∗𝑛 𝑒𝑥𝑝( 𝑅𝑇
) (C.18)
Possibly the easiest way of solving equations (C.17) and (C.18) is rewriting them into a
self-consistent loop:
𝑙𝑖𝑞
𝑔𝑎𝑠→𝑙𝑖𝑞 −𝛼𝑃ℎ,𝑃ℎ 𝜃𝑃ℎ
𝜃𝑃ℎ = 𝛽𝐾𝑃ℎ (𝜃𝑃ℎ = 0)𝐶𝑃ℎ (1 − 𝑛𝜃𝑃ℎ )𝑛 𝑒𝑥𝑝( 𝑅𝑇
) (C.19)
𝑔𝑎𝑠→𝑙𝑖𝑞 𝑙𝑖𝑞
𝑔𝑎𝑠 −{∆∆𝐺𝑃ℎ (𝜃𝑃ℎ =0)+𝛼𝑃ℎ,𝑃ℎ 𝜃𝑃ℎ }
or: 𝜃𝑃ℎ = 𝐾𝑒𝑞,𝑃ℎ (𝜃𝑃ℎ = 0)𝐾𝐻,𝑃ℎ 𝐶𝑃ℎ (1 − 𝑛𝜃𝑃ℎ )𝑛 𝑒𝑥𝑝 [ 𝑅𝑇
] (C.20)
167
Equation (C.20) can be solved iteratively using fixed-point iteration that has unfortunately
only a small convergence radius. Alternatively, we can use a one-dimensional root finder
on Eq. C.20.
𝑔𝑎𝑠→𝑙𝑖𝑞 𝑙𝑖𝑞
𝑔𝑎𝑠 −{∆∆𝐺𝑃ℎ (𝜃𝑃ℎ =0)+𝛼𝑃ℎ,𝑃ℎ 𝜃𝑃ℎ }
0 = 𝐾𝑒𝑞,𝑃ℎ (𝜃𝑃ℎ = 0)𝐾𝐻,𝑃ℎ 𝐶𝑃ℎ (1 − 𝑛𝜃𝑃ℎ )𝑛 𝑒𝑥𝑝 [ ] − 𝜃𝑃ℎ (C.21)
𝑅𝑇
We know a bracket for 𝜃𝑃ℎ (between 0 and 1/n) such that any bracketing method such as
bisection or better Brent’s method should be ideal. Note that the equations cannot be solved
analytically for n≠1. Specifically, we solved Eq. C.21 for n = 1 to 5. Once we have a
𝑔𝑎𝑠→𝑙𝑖𝑞 𝑙𝑖𝑞
function that can solve for 𝜃𝑃ℎ for any given 𝐶𝑃ℎ , ∆∆𝐺𝑃ℎ (𝜃𝑃ℎ = 0), and 𝛼𝑃ℎ,𝑃ℎ , we
𝑔𝑎𝑠→𝑙𝑖𝑞 𝑙𝑖𝑞
can optimize for ∆∆𝐺𝑃ℎ (𝜃𝑃ℎ = 0) and 𝛼𝑃ℎ,𝑃ℎ .
Next, we can also test how the presence of 𝜃𝐻 invalidates the results. Since Eq. C.18
1−𝑛𝜃𝑃ℎ
𝜃∗ = 1+𝑐𝑜𝑛𝑠𝑡 and we just have to solve Eq. C.22 for possible values of the constant ‘𝑐𝑜𝑛𝑠𝑡’
𝑛 𝑔𝑎𝑠→𝑙𝑖𝑞 𝑙𝑖𝑞
𝑔𝑎𝑠 1−𝑛𝜃 −{∆∆𝐺𝑃ℎ (𝜃𝑃ℎ =0)+𝛼𝑃ℎ,𝑃ℎ 𝜃𝑃ℎ }
𝑃ℎ
0 = 𝐾𝑎𝑑𝑠,𝑃ℎ (𝜃𝑃ℎ = 0)𝐾𝐻,𝑃ℎ 𝐶𝑃ℎ (1+𝑐𝑜𝑛𝑠𝑡 ) 𝑒𝑥𝑝 [ 𝑅𝑇
] − 𝜃𝑃ℎ (C.22)
168
Table C.3: Charges of the QM system (in e) in the initial state, where there exist
electrostatic and van der Waals interactions between the solvent and the adsorbate, and in
the final state, where all interactions have vanished as if the adsorbate was removed
completely, for different adsorbates investigated in this work. The initial and final states
correspond to states I and V in Figure C.1 in the main text, respectively. a) Reactant and
transition states of O-H bond cleavage of EG on a Cu(111) surface model using the NPA
charge model, b) CO on a Pt(111) surface model using the NPA and DDEC6 charge
models, c) ethylene glycol (EG) on the Pt(111) and Cu(111) surfaces using the NPA charge
model, d) CO on Pt(111) and Cu(111) surfaces using the DDEC6 charge model, and e)
phenol on the Pt(111) and Cu(111) surfaces using the DDEC6 charge model (Ho in the
phenol molecule indicates the hydrogen of the OH functional group).
a)
169
Cu 0.040 0.025 0.036 0.020
Cu -0.058 -0.030 -0.067 -0.026
Cu 0.128 0.019 0.150 0.003
Cu -0.043 -0.028 -0.061 -0.032
Cu 0.036 0.021 0.009 0.025
Cu 0.003 -0.002 0.000 -0.003
Cu -0.031 -0.020 -0.009 -0.025
Cu -0.074 -0.034 0.197 -0.011
Cu -0.069 -0.036 0.083 -0.035
Cu -0.029 -0.018 -0.017 -0.019
Cu 0.005 -0.007 -0.001 0.008
Cu 0.014 0.015 0.014 0.015
Cu -0.030 -0.037 0.018 -0.040
Cu -0.023 0.019 0.107 0.023
Cu -0.033 -0.042 0.003 -0.045
Cu 0.007 0.021 0.009 0.018
Cu 0.011 0.012 0.017 0.014
Cu -0.041 -0.047 -0.034 -0.050
Cu -0.053 -0.046 -0.043 -0.047
Cu 0.019 0.014 0.025 0.013
Cu 0.018 0.021 0.019 0.024
Cu 0.024 0.024 0.013 0.020
Cu 0.015 0.014 0.015 0.006
C -0.106 0.000 -0.101 0.000
C -0.097 0.000 -0.096 0.000
O -0.749 0.000 -0.856 0.000
O -0.790 0.000 -0.780 0.000
H 0.208 0.000 0.193 0.000
H 0.216 0.000 0.199 0.000
H 0.206 0.000 0.212 0.000
H 0.173 0.000 0.212 0.000
H 0.472 0.000 -0.164 0.000
H 0.483 0.000 0.479 0.000
b)
NPA DDEC6
Atom
Initial Final Initial Final
Pt 0.075 0.081 0.000 -0.001
Pt 0.076 0.080 -0.003 0.000
Pt 0.084 0.078 0.007 0.006
Pt -0.093 -0.082 -0.001 -0.003
Pt 0.081 0.073 0.003 0.003
Pt 0.089 0.087 0.008 0.000
170
Pt -0.069 -0.030 -0.019 0.010
Pt -0.010 -0.018 0.010 0.006
Pt 0.077 0.083 -0.002 -0.001
Pt 0.096 0.094 0.003 0.001
Pt -0.082 -0.043 -0.019 0.009
Pt -0.054 0.002 -0.011 0.016
Pt -0.039 -0.038 0.005 0.005
Pt 0.086 0.097 -0.001 0.000
Pt 0.029 0.034 -0.005 -0.007
Pt -0.029 -0.040 0.010 0.006
Pt -0.047 -0.044 0.005 0.004
Pt 0.035 0.029 -0.010 -0.010
Pt 0.103 0.092 0.003 0.002
Pt 0.056 0.061 -0.008 -0.006
Pt 0.092 0.092 0.003 0.002
Pt -0.064 -0.049 -0.020 -0.018
Pt 0.047 0.036 -0.010 -0.009
Pt -0.054 -0.051 -0.011 -0.013
Pt 0.002 0.012 -0.042 -0.043
Pt -0.021 -0.029 0.042 0.043
Pt -0.034 -0.030 0.043 0.044
Pt 0.003 0.009 -0.041 -0.043
Pt -0.014 -0.011 -0.049 -0.046
Pt -0.051 -0.034 0.017 0.040
Pt -0.063 -0.074 -0.020 -0.002
Pt -0.026 -0.030 0.036 0.038
Pt -0.003 -0.011 -0.049 -0.047
Pt -0.035 -0.041 -0.007 -0.005
Pt -0.029 -0.022 0.026 0.045
Pt 0.048 -0.030 0.028 -0.041
Pt -0.066 -0.048 -0.016 0.004
Pt -0.023 -0.019 0.040 0.039
Pt -0.039 -0.047 -0.001 -0.002
Pt 0.003 0.001 -0.027 -0.029
Pt -0.032 -0.033 0.014 0.035
Pt -0.073 -0.062 -0.022 -0.002
Pt -0.033 -0.023 0.029 0.030
Pt 0.001 -0.006 -0.031 -0.029
Pt 0.007 0.009 -0.030 -0.028
Pt -0.012 -0.015 0.043 0.044
Pt -0.022 -0.018 0.047 0.048
Pt 0.009 0.000 -0.029 -0.029
Pt -0.050 -0.041 -0.013 -0.012
Pt 0.017 0.009 -0.039 -0.038
Pt -0.043 -0.040 -0.012 -0.014
171
C 0.480 0.000 0.241 0.000
O -0.387 0.000 -0.115 0.000
c)
EG on Pt(111) EG on Cu(111)
Atom
Initial Final Initial Final
Metal 0.076 0.079 0.042 0.046
Metal 0.074 0.074 -0.020 -0.021
Metal 0.082 0.080 0.059 0.049
Metal -0.097 -0.094 0.017 0.015
Metal 0.078 0.081 -0.009 -0.026
Metal 0.096 0.088 -0.011 -0.025
Metal -0.026 -0.029 0.011 0.012
Metal -0.010 -0.024 0.027 0.031
Metal 0.082 0.084 0.004 0.004
Metal 0.095 0.091 -0.056 -0.081
Metal -0.044 -0.045 -0.003 0.004
Metal 0.004 0.000 0.026 0.030
Metal -0.040 -0.044 -0.021 -0.010
Metal 0.093 0.096 -0.024 -0.022
Metal 0.031 0.033 -0.039 -0.025
Metal -0.028 -0.041 -0.004 -0.005
Metal -0.045 -0.044 -0.007 -0.012
Metal 0.036 0.033 0.029 0.035
Metal 0.095 0.091 -0.019 -0.015
Metal 0.067 0.075 0.063 0.065
Metal 0.094 0.096 0.057 0.062
Metal -0.051 -0.059 0.018 0.011
Metal 0.038 0.040 0.025 0.012
Metal -0.048 -0.049 -0.009 0.005
Metal 0.002 0.014 0.019 0.034
Metal -0.038 -0.024 -0.051 -0.027
Metal -0.026 -0.027 -0.013 -0.032
Metal -0.004 0.000 0.038 0.030
Metal -0.019 -0.013 0.040 0.025
Metal -0.032 -0.035 -0.058 -0.030
Metal -0.086 -0.049 0.128 0.019
Metal -0.042 -0.025 -0.043 -0.028
Metal -0.013 -0.010 0.036 0.021
Metal -0.047 -0.043 0.003 -0.002
Metal -0.006 -0.018 -0.031 -0.020
Metal 0.014 -0.039 -0.074 -0.034
Metal -0.133 -0.048 -0.069 -0.036
172
Metal -0.032 -0.019 -0.029 -0.018
Metal -0.044 -0.042 0.005 -0.007
Metal 0.000 -0.001 0.014 0.015
Metal -0.033 -0.036 -0.030 -0.037
Metal -0.097 -0.056 -0.023 0.019
Metal -0.041 -0.015 -0.033 -0.042
Metal -0.007 -0.005 0.007 0.021
Metal 0.004 0.010 0.011 0.012
Metal -0.033 -0.018 -0.041 -0.047
Metal -0.029 -0.020 -0.053 -0.046
Metal -0.003 -0.010 0.019 0.014
Metal -0.044 -0.038 0.018 0.021
Metal 0.000 0.002 0.024 0.024
Metal -0.040 -0.046 0.015 0.014
C -0.107 0.000 -0.106 0.000
C -0.096 0.000 -0.097 0.000
O -0.748 0.000 -0.749 0.000
O -0.720 0.000 -0.790 0.000
H 0.215 0.000 0.208 0.000
H 0.226 0.000 0.216 0.000
H 0.219 0.000 0.206 0.000
H 0.196 0.000 0.173 0.000
H 0.486 0.000 0.472 0.000
H 0.505 0.000 0.483 0.000
d)
CO on Pt(111) CO on Cu(111)
Atom
Initial Final Initial Final
Metal 0.000 -0.001 0.020 0.010
Metal -0.003 0.000 -0.001 -0.003
Metal 0.007 0.006 0.006 0.011
Metal -0.001 -0.003 -0.007 -0.013
Metal 0.003 0.003 0.005 0.002
Metal 0.008 0.000 -0.008 0.003
Metal -0.019 0.010 -0.003 -0.016
Metal 0.010 0.006 0.002 -0.003
Metal -0.002 -0.001 0.014 0.019
Metal 0.003 0.001 -0.034 -0.016
Metal -0.019 0.009 0.006 0.024
Metal -0.011 0.016 -0.005 -0.006
Metal 0.005 0.005 -0.012 0.000
Metal -0.001 0.000 -0.002 0.008
Metal -0.005 -0.007 -0.004 0.011
173
Metal 0.010 0.006 0.008 0.000
Metal 0.005 0.004 -0.001 0.000
Metal -0.010 -0.010 0.008 0.002
Metal 0.003 0.002 0.008 -0.002
Metal -0.008 -0.006 0.013 0.011
Metal 0.003 0.002 -0.007 0.011
Metal -0.020 -0.018 0.012 0.015
Metal -0.010 -0.009 -0.006 -0.005
Metal -0.011 -0.013 0.018 0.013
Metal -0.042 -0.043 -0.001 -0.010
Metal 0.042 0.043 -0.021 -0.017
Metal 0.043 0.044 -0.021 -0.017
Metal -0.041 -0.043 -0.003 -0.008
Metal -0.049 -0.046 0.004 0.001
Metal 0.017 0.040 -0.009 0.002
Metal -0.020 -0.002 0.016 0.004
Metal 0.036 0.038 0.006 0.006
Metal -0.049 -0.047 -0.004 0.001
Metal -0.007 -0.005 0.023 0.027
Metal 0.026 0.045 -0.026 -0.024
Metal 0.028 -0.041 -0.014 0.009
Metal -0.016 0.004 -0.002 -0.007
Metal 0.040 0.039 -0.016 -0.019
Metal -0.001 -0.002 0.020 0.026
Metal -0.027 -0.029 0.000 -0.009
Metal 0.014 0.035 -0.015 -0.011
Metal -0.022 -0.002 0.010 0.010
Metal 0.029 0.030 -0.007 -0.008
Metal -0.031 -0.029 0.000 -0.009
Metal -0.030 -0.028 -0.008 -0.014
Metal 0.043 0.044 -0.021 -0.019
Metal 0.047 0.048 -0.026 -0.020
Metal -0.029 -0.029 -0.013 -0.011
Metal -0.013 -0.012 0.033 0.029
Metal -0.039 -0.038 -0.022 -0.015
Metal -0.012 -0.014 0.036 0.028
C 0.241 0.000 0.169 0.000
O -0.115 0.000 -0.112 0.000
174
e)
Phenol on Phenol on
Pt(111) Cu(111)
Atom Initial Final Initial Final
Metal -0.006 -0.006 0.009 0.009
Metal 0.002 -0.005 0.002 -0.002
Metal -0.005 -0.006 0.015 0.008
Metal -0.013 -0.011 -0.006 -0.014
Metal 0.005 0.014 -0.009 0.001
Metal 0.011 0.015 0.002 0.002
Metal -0.017 -0.012 0.004 -0.012
Metal -0.003 -0.003 -0.004 -0.008
Metal -0.016 0.012 0.017 0.022
Metal -0.101 0.013 -0.048 -0.016
Metal -0.007 0.012 0.021 0.022
Metal -0.006 -0.004 -0.006 -0.008
Metal 0.006 -0.003 -0.002 0.000
Metal -0.047 0.009 -0.010 0.009
Metal -0.047 0.009 -0.013 0.009
Metal 0.004 -0.003 -0.005 0.001
Metal 0.009 0.007 0.000 -0.001
Metal -0.002 0.006 -0.003 0.000
Metal 0.008 0.006 -0.002 -0.001
Metal -0.014 -0.009 0.010 0.012
Metal -0.013 -0.009 0.012 0.012
Metal -0.006 -0.004 0.014 0.015
Metal -0.062 -0.056 -0.006 -0.004
Metal -0.003 -0.004 0.014 0.015
Metal -0.035 -0.028 -0.013 -0.009
Metal 0.014 0.049 -0.003 -0.016
Metal -0.005 0.047 -0.043 -0.017
Metal -0.041 -0.029 -0.021 -0.010
Metal -0.054 -0.032 -0.013 0.001
Metal 0.009 0.040 0.001 0.006
Metal 0.024 -0.028 0.017 0.004
Metal -0.052 0.044 -0.067 0.006
Metal -0.048 -0.032 -0.013 0.001
Metal -0.004 -0.001 0.021 0.026
Metal 0.003 0.047 -0.035 -0.019
Metal 0.134 -0.017 0.023 -0.003
Metal 0.086 -0.016 0.026 -0.001
Metal 0.045 0.047 -0.034 -0.021
Metal -0.004 0.000 0.020 0.027
Metal -0.069 -0.050 -0.017 -0.008
175
Metal 0.004 0.054 -0.008 -0.010
Metal 0.032 -0.038 0.037 0.015
Metal 0.013 0.055 -0.006 -0.011
Metal -0.069 -0.050 -0.020 -0.008
Metal -0.051 -0.048 -0.021 -0.011
Metal 0.016 0.053 -0.041 -0.021
Metal 0.017 0.053 -0.040 -0.022
Metal -0.051 -0.050 -0.023 -0.012
Metal -0.010 -0.007 0.025 0.027
Metal -0.033 -0.029 -0.022 -0.014
Metal -0.007 -0.004 0.026 0.028
C -0.201 0.000 -0.188 0.000
C -0.032 0.000 -0.035 0.000
C -0.053 0.000 -0.102 0.000
C -0.071 0.000 -0.026 0.000
C -0.134 0.000 -0.219 0.000
C 0.336 0.000 0.329 0.000
O -0.406 0.000 -0.417 0.000
H 0.146 0.000 0.130 0.000
H 0.127 0.000 0.106 0.000
H 0.127 0.000 0.107 0.000
H 0.133 0.000 0.107 0.000
H 0.145 0.000 0.119 0.000
Ho 0.341 0.000 0.331 0.000
176
𝐴 𝐵
Table C.4: Values for the parameter 𝛿 in 𝑉𝑖𝑗 = (1 − 𝜆) [(𝑟 2 +𝛿𝜆)6
− (𝑟 2 +𝛿𝜆)3
] for
𝑖𝑗 𝑖𝑗
interactions used in this study. As recommended by Zacharias et al.59, the parameter δ was
chosen as the square of the vdW radius of the interacting atoms to allow for a smooth
transition between an atom present on the surface and filling the cavity after the molecule
is annihilated. Ow and Hw represent the oxygen and hydrogen of the TIP3P water model,
respectively. Ho in the phenol molecule indicates the hydrogen of the OH functional group.
O-Ow 9.672
H-Ow 7.756
Phenol
Ho-Ow 4.190
H-Hw 1.988
Ho-Hw 0.452
177
Figure C.2: Scaling of the Lennard-Jones potential interaction between the oxygen of the
CO molecule and the oxygen of the TIP3P water model, O-Ow. 𝜆 = 0 indicates the regular
LJ potential, increasing 𝜆 gradually decreases the interaction, and finally, 𝜆 = 1 leads to a
complete removal of the LJ potential as if there were no adsorbate on the surface.
178
𝑙𝑖𝑞→𝑔𝑎𝑠
Table C.5: Contributions to the aqueous-phase effect on the free energy of the low coverage desorption (∆∆𝐺𝐴𝑑𝑠𝑜𝑟𝑏𝑎𝑡𝑒 =
𝑔𝑎𝑠→𝑙𝑖𝑞
−∆∆𝐺𝐴𝑑𝑠𝑜𝑟𝑏𝑎𝑡𝑒 ) of the reactant and transition states of the O-H bond cleavage of ethylene glycol over a Cu(111) surface at 423 K (see
𝑔𝑎𝑠→𝑙𝑖𝑞
Figure 4.2a in the main text). ∆∆𝐺𝐴𝑑𝑠𝑜𝑟𝑏𝑎𝑡𝑒 also shows the overall aqueous-phase effect on the free energy of the low coverage adsorption
of the states computed with QM/MM-FEP scheme proposed in this study, as well as two well-known implicit solvation schemes: iSMS
and VASPsol. For the QM/MM-FEP calculations, the 95% confidence interval (based on limited water sampling) is given, assuming a
normal distribution. All numbers are in eV.
𝒍𝒊𝒒→𝒈𝒂𝒔 𝒈𝒂𝒔→𝒍𝒊𝒒
Contributions to ∆∆𝑮𝑨𝒅𝒔𝒐𝒓𝒃𝒂𝒕𝒆 in QM/MM-FEP scheme ∆∆𝑮𝑨𝒅𝒔𝒐𝒓𝒃𝒂𝒕𝒆
Remove
To OPT Remove II-to-III
Lennard- To OPT site QM/MM- VASPs
Adsorbate adsorbate in Electrostatic (Electrostatic iSMS
Jones in Gas FEP ol
Liquid potential Embedding)
potential
Reactant
179
0.07 ± 0.00 0.49 ± 0.00 -0.47 ± 0.02 -0.64 ± 0.00 -0.05 ± 0.00 0.60 ± 0.03 0.02 -0.15
state
Transition
0.02 ± 0.00 1.25 ± 0.00 -0.68 ± 0.05 -0.56 ± 0.01 -0.11 ± 0.00 0.10 ± 0.05 0.00 -0.21
state
Table C.6: Net charge on adsorbate and number of sites occupied by adsorbate (#Site) for all adsorbates studied in this work. These two
𝑦−𝑦̅ ̅̅̅
𝑓1 −𝑓 1 ̅̅̅
𝑓2 −𝑓 2
descriptors were then used in a linear model, = 𝛼1 + 𝛼2 (𝑠 is standard deviation and bar sign shows mean), as 𝑓1 and 𝑓2
𝑠𝑦 𝑠 𝑓1 𝑠𝑓2
𝑔𝑎𝑠→𝑙𝑖𝑞 𝑔−𝑙 𝑔𝑎𝑠
respectively, for estimating aqueous-phase effects on the free energies of adsorption (∆∆𝐺𝐴𝑑𝑠𝑜𝑟𝑏𝑎𝑡𝑒 = ∆𝐺𝐴𝑑𝑠𝑜𝑟𝑏𝑎𝑡𝑒 − ∆𝐺𝐴𝑑𝑠𝑜𝑟𝑏𝑎𝑡𝑒 ),
computed by the proposed scheme in this work (QM/MM-FEP). 𝛼1 and 𝛼2 are 0.45 and 0.62, respectively, which indicates almost equal
importance of net charge and the size of the adsorbate.
∆∆𝐺𝐴𝑑𝑠𝑜𝑟𝑏𝑎𝑡𝑒 − ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
𝑔𝑎𝑠→𝑙𝑖𝑞 𝑔𝑎𝑠→𝑙𝑖𝑞
∆∆𝐺𝐴𝑑𝑠𝑜𝑟𝑏𝑎𝑡𝑒
Net charge on 𝑠∆∆𝐺𝑔𝑎𝑠→𝑙𝑖𝑞
Adsorbate / Surface / T(K) #Site 𝐴𝑑𝑠𝑜𝑟𝑏𝑎𝑡𝑒
adsorbate, e
QM/MM-FEP Fit |Error|
181
Figure C.4: Vapor-phase optimized geometries of adsorbed species studied in this work: a)
ethylene glycol (EG) over a Cu(111) surface model, b) transition state of O-H bond
cleavage of EG over the Cu(111) surface model, c) EG over a Pt(111) surface model, d)
CO over the Cu(111) surface model, e) CO over the Pt(111) surface model, f) phenol over
the Cu(111) surface model, and g) phenol over the Pt(111) surface model.
182
Figure C.5: Fits of equation 4.9 in the main text (adsorption isotherm) to the experimental
data of aqueous-phase adsorption of phenol at 298 K on a Pt(111) catalyst surface from
Singh et al.60, 61 for two different regions: low and high phenol concentrations. The “Const”
parameter represents the hydrogen surface coverage relative to the free site coverage
(const=0 shows zero H coverage approximation), 𝑛 represents the number of sites a phenol
𝑙𝑖𝑞
molecule occupies when adsorbed on the surface, 𝛼 (𝛼𝑃ℎ,𝑃ℎ ) is a constant associated to
𝑔𝑎𝑠→𝑙𝑖𝑞
phenol-phenol lateral interactions on the surface in water, and ∆∆𝐺 (∆∆𝐺𝑃ℎ (𝜃𝑃ℎ =
0)) is the solvent effect on the aqueous-phase low coverage phenol adsorption on the
surface. SSR is the sum of squared residuals, which shows the error associated with each
fit. Fits with higher 𝑛 predict the lower concentration region better, while fits with smaller
n predict the higher concentration region better. However, the fits suggest that
𝑔𝑎𝑠→𝑙𝑖𝑞
∆∆𝐺𝑃ℎ (𝜃𝑃ℎ = 0) is quite independent of these two fit regions.
Figure C.6: Adsorbed phenol orientations at different coverages in the gas phase. At low
coverage, phenols adsorb horizontally parallel to the surface, while at high coverage, they
adsorb slanted to occupy less space on the surface.
183
Figure C.7: Fits of equation 4.9 in the main text (adsorption isotherm) to the experimental
data of aqueous-phase phenol adsorption on a Pt(111) catalyst surface at 283 K from Singh
et al.62 The “Const” parameter represents the hydrogen surface coverage relative to the free
site coverage (const=0 shows the zero H coverage approximation), 𝑛 represents the number
𝑙𝑖𝑞
of sites a phenol molecule occupies when adsorbed on the surface, 𝛼 (𝛼𝑃ℎ,𝑃ℎ ) is a constant
associated to phenol/phenol lateral interactions on the surface in water, and ∆∆𝐺
𝑔𝑎𝑠→𝑙𝑖𝑞
(∆∆𝐺𝑃ℎ (𝜃𝑃ℎ = 0)) is the aqueous-phase effect on the low coverage phenol adsorption
free energy on the surface. SSR is the sum of squared residuals, which shows the error
associated with each fit.
184
Figure C.8: Fits of equation 4.9 in the main text (adsorption isotherm) to the experimental
data of aqueous-phase phenol adsorption on a Pt(111) catalyst surface at 288 K from Singh
et al.62 See the caption of Figure C.7 for detailed explanations.
185
Figure C.9: Fits of equation 4.9 in the main text (adsorption isotherm) to the experimental
data of aqueous-phase phenol adsorption on a Pt(111) catalyst surface at 314 K from Singh
et al.62 See the caption of Figure C.7 for detailed explanations.
186
C.7 Bibliography
1. Kresse G, Furthmuller J. Efficiency of ab-initio total energy calculations for
metals and semiconductors using a plane-wave basis set. Comp Mater Sci 6, 15-
50 (1996).
5. Perdew JP, Yue W. Accurate and Simple Density Functional for the Electronic
Exchange Energy - Generalized Gradient Approximation. Phys Rev B 33, 8800-
8802 (1986).
8. Monkhorst HJ, Pack JD. Special Points for Brillouin-Zone Integrations. Phys Rev
B 13, 5188-5192 (1976).
11. Henkelman G, Jonsson H. Improved tangent estimate in the nudged elastic band
method for finding minimum energy paths and saddle points. J Chem Phys 113,
9978-9985 (2000).
187
12. Henkelman G, Uberuaga BP, Jonsson H. A climbing image nudged elastic band
method for finding saddle points and minimum energy paths. J Chem Phys 113,
9901-9904 (2000).
13. Henkelman G, Jonsson H. A dimer method for finding saddle points on high
dimensional potential surfaces using only first derivatives. J Chem Phys 111,
7010-7022 (1999).
14. Heyden A, Bell AT, Keil FJ. Efficient methods for finding transition states in
chemical reactions: Comparison of improved dimer method and partitioned
rational function optimization method. J Chem Phys 123, 224101:224101-224114
(2005).
20. Weigend F, Ahlrichs R. Balanced basis sets of split valence, triple zeta valence
and quadruple zeta valence quality for H to Rn: Design and assessment of
accuracy. Physical Chemistry Chemical Physics 7, 3297-3305 (2005).
21. Russo TV, Martin RL, Hay PJ. Effective Core Potentials for Dft Calculations. J
Phys Chem-Us 99, 17085-17087 (1995).
22. Eichkorn K, Weigend F, Treutler O, Ahlrichs R. Auxiliary basis sets for main row
atoms and transition metals and their use to approximate Coulomb potentials.
Theor Chem Acc 97, 119-124 (1997).
188
23. Weigend F. Accurate Coulomb-fitting basis sets for H to Rn. Physical Chemistry
Chemical Physics 8, 1057-1065 (2006).
24. Todorov IT, Smith W, Trachenko K, Dove MT. DL_POLY_3: new dimensions in
molecular dynamics simulations via massive parallelism. J Mater Chem 16, 1911-
1918 (2006).
25. Natarajan SK, Behler J. Neural network molecular dynamics simulations of solid-
liquid interfaces: water at low-index copper surfaces. Physical Chemistry
Chemical Physics 18, 28704-28725 (2016).
26. Jorgensen WL, Chandrasekhar J, Madura JD, Impey RW, Klein ML. Comparison
of Simple Potential Functions for Simulating Liquid Water. J Chem Phys 79, 926-
935 (1983).
27. Andersen HC. Rattle - a Velocity Version of the Shake Algorithm for Molecular-
Dynamics Calculations. J Comput Phys 52, 24-34 (1983).
28. Smith W, Forester TR. Parallel Macromolecular Simulations and the Replicated
Data Strategy .2. The Rd-Shake Algorithm. Comput Phys Commun 79, 63-77
(1994).
29. Swope WC, Andersen HC, Berens PH, Wilson KR. A Computer-Simulation
Method for the Calculation of Equilibrium-Constants for the Formation of
Physical Clusters of Molecules - Application to Small Water Clusters. J Chem
Phys 76, 637-649 (1982).
31. Geerke DP, Van Gunsteren WF. The performance of non-polarizable and
polarizable force-field parameter sets for ethylene glycol in molecular dynamics
simulations of the pure liquid and its aqueous mixtures. Mol Phys 105, 1861-1881
(2007).
32. Jorgensen WL, Laird ER, Nguyen TB, Tirado-Rives J. Monte Carlo simulations
of pure liquid substituted benzenes with OPLS potential functions. Journal of
Computational Chemistry 14, 206-215 (1993).
189
33. Straub JE, Karplus M. Molecular dynamics study of the photodissociation of
carbon monoxide from myoglobin: Ligand dynamics in the first 10 ps. Chemical
Physics 158, 221-248 (1991).
34. Freindorf M, Shao Y, Furlani TR, Kong J. Lennard–Jones parameters for the
combined QM/MM method using the B3LYP/6-31G*/AMBER potential. Journal
of Computational Chemistry 26, 1270-1278 (2005).
35. Heinz H, Vaia RA, Farmer BL, Naik RR. Accurate Simulation of Surfaces and
Interfaces of Face-Centered Cubic Metals Using 12-6 and 9-6 Lennard-Jones
Potentials. J Phys Chem C 112, 17281-17290 (2008).
36. Reed AE, Weinstock RB, Weinhold F. Natural-Population Analysis. J Chem Phys
83, 735-746 (1985).
37. Manz TA, Limas NG. Introducing DDEC6 atomic population analysis: part 1.
Charge partitioning theory and methodology. RSC Advances 6, 47771-47801
(2016).
38. Burow AM, Sierka M, Dobler J, Sauer J. Point defects in CaF2 and CeO2
investigated by the periodic electrostatic embedded cluster method. J Chem Phys
130, 174710:174711-174711 (2009).
41. Essmann U, Perera L, Berkowitz ML, Darden T, Lee H, Pedersen LG. A Smooth
Particle Mesh Ewald Method. J Chem Phys 103, 8577-8593 (1995).
42. Hu H, Yang WT. Development and application of ab initio QM/MM methods for
mechanistic simulation of reactions in solution and in enzymes. J Mol Struc-
Theochem 898, 17-30 (2009).
190
44. Pal S, Chakraborty K, Khatua P, Bandyopadhyay S. Microscopic dynamics of
water around unfolded structures of barstar at room temperature. The Journal of
Chemical Physics 142, 055102 (2015).
45. Hu H, Lu ZY, Yang WT. QM/MM minimum free-energy path: Methodology and
application to triosephosphate isomerase. Journal of Chemical Theory and
Computation 3, 390-406 (2007).
46. Hu H, Lu ZY, Parks JM, Burger SK, Yang WT. Quantum mechanics/molecular
mechanics minimum free-energy path for accurate reaction energetics in solution
and enzymes: Sequential sampling and optimization on the potential of mean
force surface. J Chem Phys 128, 034105:034101-034118 (2008).
49. Bennett C. Efficient Estimation of Free Energy Differences from Monte Carlo
Data. J Comput Phys 22, 245-268 (1976).
50. Shirts MR, Bair E, Hooker G, Pande VS. Equilibrium Free Energies from
Nonequilibrium Measurements Using Maximum-Likelihood Methods. Phys Rev
Lett 91, 140601 (2003).
52. Kreyszig E. Advanced Engineering Mathematics, 9th edn. John Wiley & Sons,
Inc. (2006).
53. Faheem M, Suthirakun S, Heyden A. New Implicit Solvation Scheme for Solid
Surfaces. J Phys Chem C 116, 22458-22462 (2012).
191
55. Mathew K, Sundararaman R, Letchworth-Weaver K, Arias TA, Hennig RG.
Implicit solvation model for density-functional study of nanocrystal surfaces and
reaction pathways. J Chem Phys 140, 084106:084101-084108 (2014).
56. Fernandez DP, Goodwin ARH, Lemmon EW, Sengers JMHL, Williams RC. A
formulation for the static permittivity of water and steam at temperatures from
238 K to 873 K at pressures up to 1200 MPa, including derivatives and Debye-
Huckel coefficients. J Phys Chem Ref Data 26, 1125-1166 (1997).
60. Singh N, Sanyal U, Fulton JL, Gutiérrez OY, Lercher JA, Campbell CT.
Quantifying Adsorption of Organic Molecules on Platinum in Aqueous Phase by
Hydrogen Site Blocking and in Situ X-ray Absorption Spectroscopy. ACS
Catalysis 9, 6869-6881 (2019).
192
APPENDIX D:
COPYRIGHT PERMISSIONS
193