You are on page 1of 21

WWW.C-CHEM.

ORG FULL PAPER

A Charge Equilibration Formalism for Treating Charge


Transfer Effects in MD Simulations: Application to
Water Clusters
Sudipta Kumar Sinha,[a,b] Mohit Mehta,[a] and Sandeep Patel *[a]

Conventional classical force fields by construction do not explicitly scaling during bond dissociation due to charge transfer over
partition intermolecular interactions to include polarization and unphysical, large distances. To control charge transfer during disso-
charge transfer effects, whereas fully quantum mechanical treat- ciation of a bond in a molecular system, we introduce a distance-
ments allow a means to effect this dissection (although not dependent scaling function (QE2 model) which, controls and
uniquely due to the lack of a charge transfer operator). Considering recovers physical behavior of the homonuclear and heteronuclear
the importance of polarization in a variety of systems, a particular charge transfer between two atoms at small and large values of
class of classical models, charge equilibration models, have been internuclear separation; and the mixed QE2 model in which we
extensively developed to study those systems; since these types of combine the QE2 model under allow and disallow charge transfer
interaction models are inherently based on movement of charge situations that describe both charge transfer and polarizability in a
throughout a system, they are natural platform for including polar- distance-dependent manner. We demonstrate the utility of both
ization and charge transfer effects within the context of molecular models in the case of a water dimer, and compare the results with
simulations. Here, we present two bond-space charge equilibration other existing models, and further, we perform short molecular
models we term as QE2 and mixed QE2 treat charge transfer in dynamics simulations for few water clusters with the QE2 model to
classical molecular mechanical calculations those provide practical show the charge transfer and internuclear separation are correlat-
solutions to two major drawbacks of charge equilibration models: ed in dynamics. VC 2017 Wiley Periodicals, Inc.

(a) a nonvanishing amount of charge transfer between two heter-


oatoms at large separations, and (b) superlinear polarizability DOI: 10.1002/jcc.24789

Introduction There are now several well-established classes of models that


effectively describe polarization effects in simulations. Major
Classical force field-based molecular modeling is applied today
approaches include (but in theory are not restricted to): induc-
to a wide spectrum of physicochemical systems with the
ible dipoles, in which a set of additional variables are intro-
aim of fundamental understanding of relevant attendant
duced into the model to describe the dipole moment induced
processes.[1–6] By necessity, classical force fields only consider
by mutual polarization interaction[15,16]; Drude oscillator, where
fixed atomic charges and ignore quantum effects such as
polarization effects are considered based on the change in
polarization and its extreme form, which may be considered as
separation between an atomic nucleus and an affixed charge
charge transfer in the molecular system. However, it is physi-
(Drude particle) coupled via a harmonic potential[17,18]; fluctu-
cally more realistic to assign charges based on the electron
ating charge model, where polarization effect is described by
density, which depend on their environment in a molecular
the charge distribution in response to the changes in geome-
system. Results of studies based on ab initio and density func-
try or external perturbations[19–21]; empirical valance bond
tional theory from the past decade have begun to point to
method, which parameterize the energy contributions of indi-
the significance of charge transfer effects related to a broad
vidual valance bond configurations[22–24]; and effective frag-
class of physical, chemical, and biological systems.[7–14] Roux,
ment potential, that uses energy decomposition of ab initio
Klein, and their coworkers have demonstrated the nontrivial
effects of charge transfer in the selectivity filter of the KcsA
[a] S. Kumar Sinha, M. Mehta, S. Patel
potassium-selective ion channel.[7,8] Liu, Head-Gordon, and Department of Chemistry and Biochemistry, University of Delaware,
their coworkers showed the importance of charge transfer in Newark, Delaware, 19716
the hydrogen-bonded systems.[9,10] Charge transfer effects are E-mail: sapatel@udel.edu
[b] S. Kumar Sinha
also important in terms of treating metalloenzymes in light of
Department of Chemistry, Indian Institute of Technology Ropar, Ropar,
the importance of such physiological moieties in human cellu- 140001, India
lar function.[11,12] However, it is highly challenging to study Contract grant sponsor: National Science Foundation; Contract grant
such systems using pure quantum mechanical or QM/MM number: CAREER:MCB:1149802; Contract grant sponsor: National
Institutes of Health; Contract grant number: COBRE:P20-RR015588
methods because the degrees of freedom of those systems
(Chemical Engineering Department) and COBRE:P20-RR017716
increase rapidly and they become computationally expensive. (Department of Chemistry and Biochemistry at the University of
The importance and incorporation of electronic polarization Delaware)
effects in molecular simulations is becoming more widespread. C 2017 Wiley Periodicals, Inc.
V

Journal of Computational Chemistry 2017, 38, 1389–1409 1389


FULL PAPER WWW.C-CHEM.ORG

data to construct the parameterized effective potentials,[25,26] their model continues to incorrectly model polarizability scal-
among many others. Among these, the charge equilibration ing with system size.
model is a promising approach for the efficient inclusion of In this work, we propose a modified charge equilibration
explicit molecular polarizability and for allowing explicit treat- model in bond space, which resolves the issue of unusual dis-
ment of dynamics charge transfer. tribution of charge at asymptotic limit in a general manner.
In short, charge equilibration models typically redistribute Technically, we have introduced a scaling function into the
charges within a molecular system in an effort to model the model that equalizes the electronegativity of the system. We
polarization response due to the conformational and environ- refer the model as QE2 charge equilibration model based on
mental changes. The partial charges on each of the atoms of a bond charge variables. Although QE2 provides sufficient flexi-
molecule are found by minimizing the electrostatic energy bility to model general properties of water clusters, but it does
function subject to a constraint that the total charge of the not capture the physical behavior of polarizability scaling for
system is conserved. There are excellent parameterization systems that are far from equilibrium. Therefore, we propose
schemes available for this model in the literature.[27–30] another model termed as mixed QE2. This model predicts a
Because polarization is modeled by a redistribution of charge more physical behavior of polarizability of the system versus
over a predefined (although not necessarily unique) spatial separation of molecules; variation of molecular polarizability
extent, the approach suffer from potential superlinear scaling with varying intermolecular distance is effectively varied varia-
of polarizability with molecular size, an artifact requiring strict tion with changing system size in the charge equilibration
attention when developing/parameterizing such models. This sense. We demonstrate polarizability for a water dimer system
is not to say that these models are ineffective. Polarizability in this study. However, the QE2 model is sufficient to describe
scaling behaviors have been addressed in several ways. Chelli the dynamics of charge transfer for moderate equilibrium sys-
et al.[31] proposed an atom–atom or bond charge transfer tem, since it captures the physical behavior of charge transfer.
model, in an effort to eliminate this superlinear scaling of the Therefore, we bound our system and perform molecular
polarizability through topological control of charge transfer. dynamics simulations on water molecular clusters using the
Shimizu et al.[32] proposed a constrained charge approximation QE2 model.
method, which consists of partitioning of the molecule into In the following section (Theoretical background section),
multiple subsystems, where the net charge of a subsystem is we first review pertinent details of existing charge equilibra-
constrained to a fixed value and the transfer of charges among tion models, their drawbacks, and our proposed models. In
them are not allowed. But this method is limited, because the Computational details section, we describe computational
arbitrary partitioning of a molecule into such subsystems details for model parameterization and dynamics of the sys-
tems. In Results and discussion section, we critically analyze
offers no specific guidance into how these units should be
each of the models in a systematic manner. Summary section
constructed, particularly if one is concerned about preserving
extends the discussion on the results obtained from molecular
transferability. Nistor et al. introduced the split charge method,
dynamics simulation trajectories. We conclude in the last
which is a generalization of charge equilibration method for
section with a summary of our findings and considerations for
nonmetallic materials.[33] Smirnov et al.[34] showed that the
future work related to this type of model.
split charge model yields correct electrostatic properties at rel-
atively weak perturbations. Warren et al.[33] discussed the ori-
gin and control of superlinear scaling of the polarizability in
Theoretical Background
the electronegativity equalization method. Despite the physi- We briefly review charge equilibration (QEq) or electronegativi-
cality of charge transfer effects in condensed phase molecular ty equalization (EE) principle as described by Mortier et al.[37]
systems, due to the challenging nature of developing such and later by Rappe and Goddard.[38] In this method, partial
models, little work has advanced in this particular area of clas- atomic charges at a specific molecular site are variable so as
sical molecular modeling of force fields. In a recent work, Son- to induce a larger or smaller dipole moment of a group of
iat et al.[35] showed the importance of polarization and charge charges in response to a local electric field. The distribution of
transfer in aqueous solution of ions. charges is calculated by minimizing an energy function which
Despite significant advances in development and applica- takes the form
tion, charge equilibration models formulated in bond-charge
representation suffers from unphysical charge distributions X N
1X N
1
EðQÞ5 vi Qi 1 Jij Qi Qj 5QT v1 QT JQ (1)
predicted for geometries far from equilibrium. For example, i51
2 i;j51
2
charge transfer over nonphysical atomic separations, or asym-
metric charge separation on salt dissociation. Chen et al.[21,36] where, the singly indexed upper-case charge variable Qi repre-
address this by introducing a distance dependent bond elec- sents one of N partial atomic charges; the parameters vi and
tronegativity into the model. They point out that the electro- Jii are the electronegativity and hardness of an atom i, respec-
negativity should depend on the geometry of the molecule. tively. The terms Jij represent screened Coulomb interactions
However, their model resolves the issue of unusual charge dis- between atoms i and j that are often taken to be the molecu-
tribution at asymptotic limit of distances only for diatomic sys- lar Coulomb integral over atomic Slater or Gaussian orbitals
tem, but it does not resolve the problem in general. Moreover, located on the respective nuclei. In this formulation, the

1390 Journal of Computational Chemistry 2017, 38, 1389–1409 WWW.CHEMISTRYVIEWS.COM


WWW.C-CHEM.ORG FULL PAPER

diagonal homogeneous atomic hardness terms Jii represents


the self-Coulomb repulsion or idempotential, evaluated at zero
internuclear separation. An additional charge conservation
constraint such as

X
N
Qi 50 (2)
i51

is applied, which guarantees the neutrality of the system. Mini-


mizing the above energy function [eq. (1)] with respect to Figure 1. Illustration of atom (qi) and bond (pi) charges in molecular sys-
each of the partial atomic charges subject to the constraint tem. The solid and dash lines in the figure represent the active and inac-
tive bonds of the system. The figure shows that only four unique bond
eq. (2), yields a system of N – 1 linear equations for the Q0i s,
charges are sufficient to describe the electrostatistics of the system. The
which can be written in matrix form as bond charges are ranked in ascending order according to the increasing
order of distances between the pair of atoms. [Color figure can be viewed
J0 Q52Dv0 (3) at wileyonlinelibrary.com]

in which J0 is the charge constrained atomic hardness matrix


(hardness matrix including the charge conservation con- degrees of freedom of the system is N – 1 no matter whether
straint) having elements J1j0 51 and Jij0 5Jij 2J1j for all i 6¼ 1. A the problem is expressed in atomic charge or bond charge
positive definite J0 ensures a local minimum of the energy, space. A representative figure has been shown in the Figure 1
eq. (1). to explain such connections. This involves restricting charge
The above QEq model works well for chemically reasonable transfer topology a priori and we refer to this as the active
structures near equilibrium.[38] A surmountable drawback topology in this work. It is important to keep in mind that the
involves charge transfer over distances defined by the spatial active topology approach by its nature affords flexibility in
extent of atomic sites involved in a group over which charge defining the topological nature of charge transfer, which we
is conserved. Approaches to deal with this are discussed in the believe to be a strength of an approach of this type. Although
literature.[28,29,31,33,39] This precludes use of this current model in this work we use a fixed (constant) topology matrix, more
formulation in situations where chemical reactions are involved sophisticated implementations in the future can allow for
and bond formation and dissociation occur. To address this dynamic construction of the active topology space based on
behavior in appropriate situations (i.e., bond dissociation), appropriate cutoff functions; this would be of practical use for
recent investigations have found it advantageous to work in a MD and MC simulations with on-the-fly updates of the active
bond space built on two-site variables, qji corresponding to topology space.
pairs of atoms (atom pair 5 bond). The definition of bonds in Introducing bond charges in the atomic QEq method [eqs.
this scheme is more general, and it considers the bonds as (1) and (4)], which we will refer to as the QE scheme, yields
connections between two atoms involved in charge transfer.
X N X N
1X N X N X N X N
Numerous nomenclatures can be found in literature defining EðqÞ5 vi qji 1 Jij qki q‘j : (6)
such bonds.[21,28,29,31,33,36] i51 j51
2 i51 j51 k51 ‘51
The atomic and bond charge representations of the QEq/EE
formalism may be connected by the use of the projection By exploiting the antisymmetry, this expression can be writ-
ten[21,28] in the form
X
Qi 5 qji (4)
j XN X N
1X N X N X N X N  
EðqÞ5 ðvj 2vi Þ qji 1 Jik 2Ji‘ 2Jjk 1Jj‘ qji q‘k
i51 j>i
2 i51 j>i k51 ‘>k
where the sum runs over all atoms j for which charge transfer
(7)
to atom i is allowed/defined given that the bond charge varia-
bles (denoted by the lower-case variables qji) satisfy antisym- for which pairs of indices are subject to the summation con-
metry relation qij 52qji and that qii 5 0. The projection and ventions j > i and ‘ > k. Explicitly, the above equation in bond
antisymmetry condition are compactly written as space is

Q5Cq (5) X
Nbond

EðpÞ5 ðvatom2 2vatom1 Þ pi


where C is the connectivity or topology matrix for the charge i51
bond  
transfer. The dimension of the matrix, C, is ðN21Þ3NðN21Þ=2, 1 NX X
bond N

1 Jatomi atomj 2Jatomi atomj 2Jatomi atomj 1Jatomi atomj pi pj


without any topological constraints (i.e., charge transfer 2 i51 j51 1 1 1 2 2 1 2 2

allowed between all atomic sites). However, such connectivity (8)


or topology matrix cannot be used for practical purpose, since
it is rank deficient. Further, the number of physically significant or

Journal of Computational Chemistry 2017, 38, 1389–1409 1391


FULL PAPER WWW.C-CHEM.ORG

X
Nbond
1 NX X
bond N bond unmodified energy function in bond space and introduced p0i s
EðpÞ5 Dvi pi 1 gij pi pj (9)
2 i51 j51 into eq. (9), thus, the modified energy function becomes
i51

  NX
bond
1 NX X
bond N bond

where, gij 5 Jatomi atomj 2Jatomi atomj 2Jatomi atomj 1Jatomi atomj ; Dvi 5 Eðp0 Þ5 Dvi s21 0
i pi 1 gij s21 21 0 0
i sj pi pj (14)
1 1 1 2 2 1 2 2
i51
2 i51 j51
ðvatom2 2vatom1 Þ,
the singly indexed lower-case variable, pi rep-
resents a bond charge between atom1 and atom2; the parame- The above equation is now our working equation for this
ter Nbond is the number of charge transfer variables (bond study; this equation is referred as the QE2 model in this manu-
charges) in the system; the other parameters such as v and J script. The lowest energy in an active topology is obtained by
have their usual meaning as described before. The parameter differentiating eq. (9) w.r.t bond charges and equating it to
Nbond is N – 1 under the topological constraints. zero, resulting in Nbond number of simulation equations that
The above QE model [eq. (9)] does not yet eliminate the can be written as
drawbacks mentioned before, but rather it describes exactly
the same model in a different notation. However, it is a good oE NXbond
0
starting point for the manipulation of this model because the 0 5Dvi 1 gij s21
j pj 50 (15)
opi j51
transformed variables allow us to introduce a distance-
dependent function, f ðrÞ that penalizes long range charge
The above set of linear equations can be written in a matrix
transfer. Martinez and coworkers suggested an approach intro-
form as
ducing a distance dependent decaying function affecting the
electronegativity and they named their model QTPIE. They P0 52SG21 Dv (16)
wisely chose a function having properties of an overlap inte-
gral between orbitals on the ith and jth atoms. Their modified where S is a diagonal square matrix whose diagonal elements
energy function is are either 1 or a value determined by sðrÞ, and the off-
diagonal elements are 0, and G is the square, symmetric bond
NX
bond
1 NX X
bond Nbond
hardness matrix. The elements of the G matrix are given by
EðpÞ5 Dvi pi fi ðrÞ1 gij pi pj (10)
i51
2 i51 j51 the corresponding elements of the Hessian matrix as

The spirit of this approach provides a means to treat molecu- o2 E


5gii (17)
lar systems with extension to include intermolecular charge op02
i
transfer. In this work, we introduce scaling factors on the bond
charge (pi) that controls the value of bond charges. The scal- and
ing factor takes the value 1 if the corresponding bond is intra-
molecular. However, it takes the value from a distance- o2 E
5gij (18)
dependent function, sðrÞ, if the corresponding bond is inter- op0i op0j
molecular. Therefore, the effective bond charges, p0i , will follow
The above equation guarantees that the intermolecular bond
a relation as shown below
charges vanish in the asymptotic limit rij 5jri 2rj j ! 1 and the
p0i 5si ðrÞpi (11) results are physically meaningful. We demonstrated the valida-
tion of this new equation for heteronuclear diatomic and
or water dimer systems. The results have been presented in the
Supporting Information (Fig. S2) and Figure 3. On rearranging
0
pi 5s21
i ðrÞpi (12) the matrix equation, we obtain

We choose the function sðrÞ as a parametrized Slater overlap SGS21 P0 52SDv (19)
integral of the form
 
or
logðbrÞ
sðrÞ5a 1:02erf pffiffiffi (13)
c 2 G0 P0 52SDv (20)

where a, b, c are free parameters of the system and r is the Thus, G0 is the similarity transformed matrix of G. Therefore,
separation between two atoms. The choice of parametrized the modified energy function as proposed in QTPIE will be
Slater overlap integral over the actual Slater overlap integral is useful but we need to transform the original bond space to
beneficial since the latter is computationally expensive. The another bond space for a physically meaningful result.
function sðrÞ is chosen after performing various tests; the Although, this approach offers a way to overcome the short-
superposition between the actual and parametrized Slater comings of the original QE model, since we introduce the
overlap integral is presented in Supporting Information (Fig. inverse of a decreasing distance-dependent function into the
S1). To rederive the energy function we started from the energy expression, the potential for instabilities arises.

1392 Journal of Computational Chemistry 2017, 38, 1389–1409 WWW.CHEMISTRYVIEWS.COM


WWW.C-CHEM.ORG FULL PAPER

We turn to discussion of electrostatic properties resulting transfer conditions. The element h is a mixing function, which
from this formalism. The system dipole moment is expressed as: takes a value from another distance dependent function, hðrÞ
that smoothly varies from 0 to 1. However, it is convenient to use
XN
h 5 0.5 for intramolecular bond charge variable and use a value
~
l5 ri Qi (21)
i51
from the function hðrÞ for intermolecular bond charge variable. A
parameterized exponential type function, hðrÞ5expð2br 2 Þ, is
The above equation can be written in bond space as adopted here. We chose this function because the intermolecular
bond charges vanish almost exponentially with the increase of
XN XN
intermolecular distances. Equation (24) allows us to deduce a
~
l5 ri Cpi 5 Dri pi 5DrT P (22)
i51 i51
self-consistent representation for a modified bond charge vari-
able which is also of a linear combination form. This new bond
Therefore, one obtains the system dipole moment from the charge variable is given as (see Appendix for details).
displacement vector, Dr, and bond charge vector, P. To obtain
the expression for polarizability, we are first required to mea-
sure the response of dipole moment of the system due to an p00i 5hpai 1ð12hÞd0 pdi (25)
external electric field ~
e. If we assume that the strength of the
field is sufficiently small so that nuclear configuration of where d0 takes a value either 0 or 1 depending on the type of
the system is unchanged under ~ e, then the polarizability, a of charge transfer (i.e., intermolecular or intramolecular).
the system in bond space will be Substituting the new bond charge variable into eq. (14), we
can define the new energy expression for mixed QE2 model as
a5DrT SG21 Dr (23)

The detailed derivation of the polarizability expression [eq. X


Nbond
1 NX X
bond N bond

Eðp00 Þ5 Dvi s21 00


i pi 1 gij s21 21 00 00
i sj pi pj (26)
(23)] can be found elsewhere.[31,39] Since the polarizability of 2 i51 j5i
i51
the system is quadratically related with displacement vector,
Dr, it grows quadratically with the increase of internuclear sep- Details of this model are described in Appendix. The model is
aration, and in fact the value of polarizability will be infinite if useful to perform MD simulation for large molecular system in
some of the elements in the displacement vector ðDrÞ increase a pairwise manner.
to 1. Moreover, the intermolecular separation in a bulk molec-
ular system may be large but that does not necessarily mean Computational Details
that the polarizability of the system will be large. As Bauer
et al.[40] showed the polarizability of a system of two water We first discuss estimation of model parameters for the water
molecules, at large intermolecular separation tends to the sum cluster systems. The models possess atomic hardness (i.e.,
of the individual water polarizabilities (on a per water mole- diagonal Jii elements), electronegativity (vi), and three parame-
cule basis). ab initio calculations also show that the total polar- ters of the function, sðrÞ, namely a, b, and c. For computational
izability of a water dimer system is twice the value of the simplicity, we consider the Jij, which is often known as hetero-
individual isolated molecule polarizabilities. However, such a geneous hardness, and they are evaluated from the atom type
problem is absent in dipole moment calculation, since it is lin- values based on the combining rule[41]
early related with displacement vector, Dr. This is because the
1
bond charges vanish at asymptotic limit of the displacement 2 ðJii 1Jjj Þ
Jij ðrij ; Jii ; Jjj Þ5 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi (27)
vector, and thus the product also vanishes at that limit. This 11 14 ðJii 1Jjj Þ2 rij2
superlinear polarizability scaling has been discussed by Warren
et al.[39] and by Nistor et al.[33]. We treat the intramolecular heterogeneous hardnesses as
To address superlinear polarizability scaling, we devise a mixed parameters in this study. A set of water clusters for this study
QE2 model. The model, in essence, is a linear combination of two is taken from the Cambridge Cluster Database.[42] We first per-
states, one allowing charge transfer, and the other suppressing form single point energy calculation on those water clusters in
charge transfer over intermolecular separations. We next demon- gas phase at the mp2/aug-cc-pvdz level of theory using Gauss-
strate this model for a simple water dimer system which requires ian09 program.[43] The atomic charges of the systems are
only one intermolecular bond charge variable and captures determined by using CHELPG scheme from the current elec-
essential features of the model. The logic is extensible to a sys- tron density at that level of theory. Once we obtain QM
tem containing many molecules exchanging charge in pairwise charges, we calculate the difference of QE charges from them
manner. The mixed QE2 model defines system polarizability as and construct a cost function, Fcost, that contains all of the
above mentioned free parameters
a5haa 1ð12hÞad (24)
XN
where aa and ad represent the two independent system polar- Fcost 5 ðDei Þ2 (28)
izabilities under conditions of allowed and disallowed charge i51

Journal of Computational Chemistry 2017, 38, 1389–1409 1393


FULL PAPER WWW.C-CHEM.ORG

where Dei 5eQM QE


i 2ei ; and ei
QM
and eQE
i are atomic charges of Lennard–Jones (LJ) and intramolecular energy, and (b) QE
the system as computed using QM and QE methods; and eQE i molecular energy where the electrostatic energy is modified.
is a function of all the parameters of the system, which is giv- Additionally, a spherical harmonic boundary potential is
en by the following equation. applied to limit intermolecular separation distances. Explicitly,
the potential energy function, Uðp0 ; rÞ takes the following form
eQE 5CP0 52CSG21 Dv (29) "
XXX    #
0 rab 12 rab 6
where C is the connectivity or topology matrix for charge Uðp ; rÞ5 4Eab 2
j<i a b
ria;jb ria;jb
transfer. In the following step, we optimize the above cost
function w.r.t. each of the parameters of the system. To per- 1XXX  
2
1 Kb riab 2riab 1
form this optimization a Monte Carlo simulated annealing 2 i a b
(MCSA) method is applied to the system.[44] The MCSA is an
1XX NXbond
1 NX X
bond Nbond
iterative algorithm; we start with initial values of guess param- KE ðria 2rc Þ2 1 Dvi s21 0
i pi 1 gij s21 0 21 0
i pi sj pj
2 i a 2 i51 j5i
eter, a starting step vector whose dimension is identical to the i51

parameter vector, and a starting temperature. The initial values (30)


of the guess parameters are taken from the literature[45]; the
components of the starting step vector were set to a value of where r and E are LJ parameters; Kb and KE are bonded and

2.0; and the initial temperature was set to a value 1000 K. Dur- wall potential force constants, respectively; rab is the equilibri-
ing optimization, the MCSA algorithm generates a succession um bond length and rc is the boundary cutoff from the center
of parameters, that is, Monte Carlo (MC) moves, tending to of mass of the systems; J and v are the hardnesses and elec-
global minimum of the Fcost. New set of parameters are gener- tronegativities of the atoms. Bond charges and atomic posi-
ated around the current set of parameters by applying random tions are treated as independent dynamical variables. The total
moves along each coordinate direction, in turn. The new energy (i.e., Hamiltonian of the system) will be
parameter values are uniformly distributed in an interval cen- NX X
molec N atom
1 NXbond
1 :
tered around the current parameters. The width of the interval Hðp0 ; rÞ5 mia ria : 21 Mi p0i 21Uðp0 ; rÞ (31)
was chosen 610% of the current values of the parameter. Half i51 a51
2 i51
2
the size of these intervals along each direction is recorded in
step vector. The aim of the variation in step vector is to main- where ma is the atomic mass of atom a and Mi is the mass of
time2
tain the average percentage of accepted MC moves at about fictitious bond charge, p0i , with dimensions of energy 3 charge 2.

one-half of the total number of MC moves. If the point falls Additional constraints on the bond charges are implicitly
outside the definition domain of Fcost, a new set of parameter included in the potential energy function described by eq. (3).
is randomly generated until a point belonging to the defini- The nuclear degrees of freedom evolve according to Newton’s
tion domain is found. A candidate set of parameter is accept- equation
ed or rejected according to Metropolis criterion[44] and the
:: oUðp0 ; rÞ
process is continuing until a thermal equilibrium is reached. ma ria 52 (32)
oria
After thermal equilibrium, the temperature of the system is
reduced at a rate of 0.95 and a new sequence of moves is and the set of bond charges evolve in time according to
made starting from current set of optimized parameters, until
thermal equilibrium is reached again, and so on. The process :: oUðp0 ; rÞ
is terminated at a temperature low enough that no more use- Mi p0i 52 (33)
op0i
ful improvement of Fcost can be expected. The details of the
MCSA method can be found elsewhere.[44] Note that we sepa- The coupled equation of motion are numerically solved using
rately parameterized the Gaussian function h(r) for the mixed velocity-Verlet integrator. The detail set of parameters for these
QE2 model for water dimer using optimized parameters simulations are presented in Table 1. The LJ parameters r and
obtained from the QE2 model. Since, we compare to existing E for oxygen were set to 2.50 Å and 0.2862 kcal/mol and the
methods for treating intermolecular charge transfer, parame- corresponding values for hydrogen were set to 0.0 Å and 0.0
ters for a QTPIE-like model for the water cluster systems are kcal/mol. The initial configuration of the systems consists of 3-
also computed by the above method. But instead of using the site water clusters containing Mwat water molecules, where the
complicated overlap integral, we use the the function s(r) for values of Mwat vary from 2 to 5. We use the global optimized
computing the bond electronegativity for this model. QE parameters for the dynamics of each of the water clusters.
Once we obtain the parameters for the QE2 model, we per- The mass of a bond charge, Mi is a fictitious quantity, chosen
form molecular dynamics simulation on each of the water clus- to be small enough to guarantee that charges readjust rapidly
ters in gas phase (using only the QE2 model). To perform the to changes in the nuclear degrees of freedom. The two force
dynamics of the system, we define the potential energy func- constants Kb and KE are set to 53106 and 53105 kcal/mol,
tion of the molecular system. The molecular potential energy respectively. We are required to use such high values of force
function of Nmolec system with Ni atoms of molecule i and constants for each bond of the water molecules because we
Nbond bond charge variables consists of two parts, (a) do not apply any additional holonomic constraint force on

1394 Journal of Computational Chemistry 2017, 38, 1389–1409 WWW.CHEMISTRYVIEWS.COM


WWW.C-CHEM.ORG FULL PAPER

Table 1. The global optimized parameters for OAO and OAH charge transfer as obtained from simulated annealing method are presented in the table.

Hardness (J)
(kcal/(mol e2)) v

Atomic Bond (kcal/(mol e))


b b
Type O H OAH HAH O H a (Å21) c (Å22)
OAO 334.56 388.27 257.77 183.32 75.27 0.00 0.816 0.532 0.478 0.011
OAH 335.95 388.28 257.81 183.25 75.20 0.00 0.767 0.567 0.471 0.051
The parameters for mixed QE2 model are also tabulated.

them; we thus acknowledge that we do not model true rigid Results and Discussion
water models, but for the current purpose to demonstrate the
feasibility of models such as QE2 for molecular dynamics simu- In this section, we present and discuss the results of our calcu-
lations, we employ basic models. We further stress that since lations for the models as outlined above. We start with discus-
we use an in-house MATLAB program to implement dynamics sion of the critical aspects of the new model, its applicability,
using the QE2 models, we do not incorporate holonomic con- and its validation. Finally, we discuss the results obtained from
straint algorithms from scratch as we feel these are not essen- MD trajectories for different sizes of water clusters in gas
tial to our discussion. We first minimized total energy for each phase.
of the cluster systems using steepest descent method. Then,
we performed molecular dynamics simulation in gas phase Model validation
using those minimized coordinates and bond charges of the
We first parameterize models using the water dimer system.
system. The systems are equilibrated for about 50 ps at con-
We treat cases for intermolecular charge transfer between (a)
stant energy and then the simulations are continued at con-
two oxygen atoms on different molecules and (b) the oxygen
stant temperature. The initial velocities of the nuclei and bond
charges are picked as Gaussian random numbers from Max- on water molecule and a single hydrogen atom on the second
well–Boltzmann distribution at 300 K and 5 K. Both nuclear water molecule. Parameters determined using the simulated
and bond charge degrees of freedom are coupled with veloci- annealing methodology described earlier are used to compute
ty rescaling thermostat at those two temperatures during sim- electrostatic properties of the system as a function of inter-
ulation. We have chosen 5:031022 fs time step for doing all of atomic separation, rij, where ij5OOorOH. The set of parameters
these simulations. we obtain are presented in Table 1. The configuration of the
All the simulations, optimizations, and the analyses of the systems at different values of rij are generated by translating
trajectories have been performed using our in-house MATLAB one of the water molecules of the dimer along the OAH bond
(version:R2013)[46] code. We consider the two intramolecular vector formed between two water molecules. We illustrate the
physical bonds of water molecules and Mwat 21 intermolecular process of such translation in Figure 2. We calculate the elec-
connections as bond charge variables. The network of Mwat 21 trostatic properties of water dimer for QE2 and mixed QE2
intermolecular connections is decided by considering the mini- charge transfer models. We consider the two different scenari-
mum spacing between the center of mass of two water mole- os of allowing intermolecular charge transfer and disallowing
cules. Specifically, we determine the first pair of water intermolecular charge transfer for calculating the electrostatic
molecules that corresponds to the minimum value of intermo- properties using the QE2 model. We further consider the
lecular distance of the system. We take this pair of water mole- charge transfer between two water molecules happens
cules to correspond to the first connection of the network. through (a) homonuclear (OAO), or (b) heteronuclear (OAH)
The next connection is decided by calculating the minimum bond charge variables. In Figure 3, we illustrate the behavior
distance between a water molecule, which is outside of the of electrostatic properties of the system as a function of rij. For
network, and the two terminal molecules of the existing net-
work. The process is continued until it reaches value of total
number of water molecules of the system.
We parameterized the systems and performed dynamics by
allowing both OAO and OAH intermolecular charge transfer.
We allow such type of charge transfer by changing only the
topology matrix of the system. We do not change the topolo-
gy matrix dynamically throughout our calculation in this study,
although, this is a feature to explore in the future. The visual Figure 2. The process of translation of first water molecule in dimer has
been illustrated here. The first water molecule is translated along the vec-
molecular dynamics (VMD) package[47] was used to monitor
tor formed between the oxygen of second water and the hydrogen of first
the simulation, visualization, and graphics preparation for this water, which is oriented to the oxygen of first water. [Color figure can be
work. viewed at wileyonlinelibrary.com]

Journal of Computational Chemistry 2017, 38, 1389–1409 1395


FULL PAPER WWW.C-CHEM.ORG

Figure 3. The electrostatic properties of water dimer system as obtained from different QE and QTPIE models, and QM calculation. The four panels in the
figure describe the electrostatic energy (panel A), interwater bond charge (panel B), dipole moment (panel C), and polarizability (panel D) of the system as
a function of intermolecular distance. The solid and dash lines represent intermolecular OAO and OAH charge transfer, respectively. Smooth transition in
electrostatic properties with intermolecular separation between allow and disallow scenarios of QE2 model as obtained from mixed QE2 model is shown in
the insets. [Color figure can be viewed at wileyonlinelibrary.com]

comparison, we include results obtained from the QTPIE and variables. A significant difference in DE(r) function has been
QM calculations. found between QTPIE and QE2 models for the allowing of
Figure 3A describes the binding electrostatic energy OAH intermolecular charge transfer. This is not surprising
between two water molecules as a function of intermolecular because the elements of bond electronegativity vector (Dv)
separations, DE(r), for all models discussed. The function, DE(r) and bond hardness matrix (G) for the intermolecular bond
for a model is calculated by taking the difference in electro- charge variables for QTPIE model vanish with the increase of
static energy between water dimer and two isolated water the distances between two water molecules, and the corre-
molecules. We compare the results as obtained from different sponding intramolecular elements of the vector and matrix are
charge transfer models and QM calculation in the figure. We constants. Therefore, the values of intermolecular bond charge
find the results obtained from all of the models and QM calcu- for this model never vanish with the increase of distances
lation are similar in nature. The function, DE(r) for the allowing between two water molecules for the allowing of OAH inter-
of OAO intermolecular charge transfer is identical for both molecular charge transfer. Thus, the function DE(r) as obtained
QE2 and QTPIE models. This is because the distance depen- from this model for the allowing of homonuclear and hetero-
dent scaling function, s(r), for the allowing of OAO intermolec- nuclear intermolecular charge transfer is significantly different.
ular charge transfer in QTPIE model does not affect the Interestingly, our results show that the function DE(r) as com-
elements of bond electronegativity vector (Dv) and the bond puted from the QE2 model for the allowing of homonuclear
hardness matrix (G). Conversely, the function, s(r) acts twice in and heteronuclear intermolecular charge transfer cases follows
the calculation of DE(r) in QE2 model; first it multiplies directly similar behavior. This is because the function, s(r) controls the
for the calculation of effective bond charges (pis) and in the elements of the Dv and G in such a way that the values of
following step, it multiplies inversely for the calculation of intermolecular bond charge variables vanish at large values
energy of the system. Therefore, the effect of function, s(r) of r. As a result the function, DE(r) is almost identical for the
cancels out from the energy calculation for this case. However, cases of homonuclear or heteronuclear intermolecular charge
the identical values of energy for both models is true for the transfer. Moreover, slight difference in the values of parameters
system containing one intermolecular bond charge variable for the allowing of OAO and OAH charge transfer are also
but such fact does not follow in a general manner for the sys- reflected in the DE(r) function at small values of r. Importantly,
tems having more than one intermolecular bond charge the Gaussian function, h(r) that we introduce in the mixed QE2

1396 Journal of Computational Chemistry 2017, 38, 1389–1409 WWW.CHEMISTRYVIEWS.COM


WWW.C-CHEM.ORG FULL PAPER

charge transfer model, transforms the energy function smooth- moment of the system as obtained from the QE2 model varies
ly from allow to disallow scenarios of QE2 model. Incorpora- in a nonlinear and linear fashion at small and large values of r,
tion of the h(r) function into the model is necessary, which respectively. This is because the electrostatic field of each of
ensures that the system behaves like water dimer at small val- the water molecules influences each other at the small values
ues of r but it transforms to two isolated water molecules as of r, but it gradually vanishes with the increase of the values
the values of r increase. We demonstrate such smooth transi- of r. The function lðrÞ as calculated by allowing OAO charge
tion between the two scenarios in the inset of Figure 3A. transfer between two water molecules using QE2 model is
Figure 3B, shows the magnitude of intermolecular bond almost identical to the function as obtained by disallowing
charge as a function of intermolecular separation, pi ðrÞ. We charge transfer between them. This is because the value of
calculate the function pi ðrÞ by allowing OAO or OAH charge intermolecular OAO bond charge as calculated from the allow
transfer between two water molecules present in the dimer. scenario of QE2 charge transfer model is small at equilibrium
Since there is no physical bond charge variable in QM calcula- configuration of dimer, and it gradually vanishes with the
tion, we consider the difference in charge between the two increase of distance between two water molecules. Therefore,
water molecules as bond charge variable for this case. The the dipole moment vector, which is associated to the intermo-
function pi ðrÞ as obtained from the QE2, mixed QE2 models lecular bond charge variable of the system, vanishes quickly
and QM calculation is qualitatively similar. The magnitude of with the increase of of the values of r. However, the magni-
intermolecular bond charge for the allowing of OAH charge tude of dipole moment for QE2 model as calculated by allow-
transfer is almost similar to the results obtained from QM cal- ing OAH intermolecular charge transfer is larger as compared
culation. We show in the inset that pi ðrÞ as obtained from QE2 to the results obtained from OAO charge transfer. Since, the
model for the allowing of either OAH or OAO intermolecular value of intermolecular bond charge for intermolecular OAH
charge transfer gradually vanishes with increasing the values charge transfer is larger as compared to OAO charge transfer
of r. We have also found that the function pi ðrÞ for OAO inter- at small values of r, the magnitude of net dipole moment vec-
molecular bond charge transfer as obtained from the QE2 tor for the former case is larger as compared to later one at
model is larger as compared to the result obtained from QM that region. Moreover, the magnitude of net dipole moment
calculation. Moreover, pi ðrÞ for intermolecular OAH bond for the allowing of either OAO or OAH charge transfer
charge transfer as obtained from QTPIE model does not vanish reaches a constant value at large values of r, because, the
at large values of r. However, the introduction of s(r) into the intermolecular bond dipole moment vector for both of the
model nicely equalizes the electronegativity and hardness over cases vanishes with the increase of the values of r. The results
the system for QE2 model. The function pi ðrÞ as obtained from obtained from the mixed QE2 model for intermolecular OAO
mixed QE2 model transforms smoothly from allow to disallow and OAH charge transfer cases at small and large values of r
charge transfer scenarios of QE2 model as the values of r are similar with the results obtained from QE2 charge transfer
increase, but such smooth transition is not prominent for the model under allow and disallow scenarios. The magnitude of
dimer system, since the amount of charge transfer between net dipole moment vector in QM calculation is computed by
two water molecules is very small. We expect that such transi- adding the two dipole moment vectors of the individual mole-
tion may be visible for efficient charge conducting materi- cule present in the dimer of water. The magnitude of dipole
als.[13,14] Additionally, we mapped the bond charges of the moment as obtained from QM calculation decreases at small
system to atomic and molecular charges and the results are values of r, and it reaches to a constant value as the values of
presented in the Supporting Information (see Figs. S3 and S4). r increases. The charge distribution of individual water mole-
So far we discuss that the QE2 model produces qualitatively cule at small values of r is quite different because the electro-
correct physical behavior of charge transfer for both homonu- static field of each of the water molecules influences each
clear and heteronuclear intermolecular charge transfer cases. other at that region. We have found that the function lðrÞ as
Both QTPIE and the QE2 models redistribute atomic charges in obtained from QTPIE model for the allowing of either OAO or
a different manner on each of the individual water molecules. OAH intermolecular charge transfer is significantly different.
The distribution of atomic charges as calculated from QE2 Since the function pi ðrÞ for the allowing of OAO charge trans-
model is also different for allow and disallow charge transfer fer gradually vanishes with the increase of distance between
scenarios. The new distribution of atomic charge influences two water molecules, the magnitude of the corresponding
the electrostatic properties such as dipole moment and polar- bond dipole moment vector also vanishes. Thus, the magni-
izabilities of the system. Therefore, it is expected that the elec- tude of dipole moment vector of the system as obtained from
trostatic properties of the system is different at small and this model is sum of the magnitude of individual dipole
large values of r. We calculated the magnitude of dipole moment vector of two water molecules at large values of r.
moment (l) and total isotropic polarizability (a) of the water However, the magnitude of dipole moment as obtained from
dimer system as a function of r. We consider the trace of diag- this model for the allowing of heteronuclear intermolecular
onalized polarizability tensor as the isotropic polarizability of charge transfer (OAH) increases linearly with the increase of
the system. We presented the results in Figures 3C and 3D. the values of r. Such a situation arises because of the presence
The net dipole moment vector of the system is computed of a nonvanishing value of intermolecular bond charge vari-
by adding all the individual bond dipole moment vectors of able at large values of r for this case. The magnitude of dipole
the system. The result shows that the magnitude of dipole moment vector associated with the nonvanishing bond charge

Journal of Computational Chemistry 2017, 38, 1389–1409 1397


FULL PAPER WWW.C-CHEM.ORG

Figure 4. Representative snapshots of four water clusters obtained from MD trajectories. The top and bottom rows in the figure correspond the configura-
tions after equilibration and at the end of simulation. The intermolecular connections among water molecules for OAO intermolecular charge transfer are
represented by dashed line. The numbers in the figure correspond the intermolecular OAO distances between two water molecules in Å. [Color figure can
be viewed at wileyonlinelibrary.com]

variable causes increase the length of net dipole moment vec- We have shown in our study that the problem is severe, partic-
tor of the system. ularly for heteronuclear intermolecular charge transfer. Howev-
Figure 3D describes the polarizability of the system as a er, heteronuclear intermolecular charge transfer is common in
function of intermolecular separation, aðrÞ. The result shows molecular system; and it fails to capture the correct physical
that the magnitude of polarizability of the system increases behavior of charge transfer in general. Conversely, the QE2
with the increase of the values of r for both QTPIE and the model reproduces correct behavior of charge transfer and rein-
QE2 models. The result is inconsistent and shows the incor- forces the applicability of this model in the molecular system.
rectness of those models. However, the mixed QE2 model, Additionally, we find that the QE2 charge transfer model does
which we have demonstrated in this work, produces correct not capture the correct physical behavior of polarizability of
behavior of the polarizability of the system at all values of r. the system at large values of r. Conversely, the mixed QE2
The magnitude of polarizability of the system as obtained model reproduces qualitatively similar electrostatics as
from mixed QE2 model is comparable with the results obtained from QM calculation at all values of r. Since the QE2
obtained from QM calculation. This is an important finding model works well for the systems those are not far from equi-
and enforces the applicability of the mixed QE2 model for librium, we use this model to study the dynamics of water
general molecular systems. However, the polarizabilities as clusters in gas phase. In the following section we have
obtained from the QE2 charge transfer model are also reason- described that in detail.
able at small values of r; but the model over estimates the
polarizability with the increase of the values of r. Therefore, MD trajectory analysis
the QE2 model is also equally applicable for the systems those
are not far from equilibrium. Importantly, the magnitude of In this section, we analyze the MD trajectories of each of the
the polarizability of the system as obtained from QE2 charge water clusters. The initial and final configurations of each of
transfer model for the allowing of OAH intermolecular charge the clusters are presented in Figure 4. Since we have shown
transfer is larger as compared to the allowing of OAO charge before that the intermolecular bond charge as obtained from
transfer between two water molecules. Such difference in QE2 model gradually vanishes with the increase of distances
polarizability primarily arises because the magnitude of inter- between two water molecules, we are expecting a correlation
molecular bond charge variable of the system under those between bond charge variable and intermolecular separation
two different charge transfer cases is significantly different. in the dynamics of those water clusters. We performed the
However, the QTPIE model over estimates polarizability of the dynamics of the four water clusters by allowing either OAO or
system for the allowing of both OAO and OAH intermolecular OAH intermolecular charge transfer between two water mole-
charge transfer cases. cules present in the systems. Note that the length of the tra-
Thus, it is clear from the above discussion that there are sig- jectory for each of the water clusters is not equal because
nificant drawbacks in QTPIE model and it raises questions on such complex dynamics is computationally expensive for larger
the applicability of the model for general molecular systems. clusters. We present scatter plots of intermolecular bond

1398 Journal of Computational Chemistry 2017, 38, 1389–1409 WWW.CHEMISTRYVIEWS.COM


WWW.C-CHEM.ORG FULL PAPER

Figure 5. Correlation between intermolecular separation and the bond charges for OAO charge transfer. Four panels in the figure describe the results
obtained from the trajectory of water dimer, trimer, tetramer and pentamer. The reference zero line is included into the figure for differentiating sign of
the bond charges. Different color for each of the lines in the figure represents individual intermolecular connections. [Color figure can be viewed at
wileyonlinelibrary.com]

Figure 6. Correlation between intermolecular separation and the bond charges for OAH charge transfer. Four panels in the figure describe the results
obtained from the trajectory of water dimer, trimer, tetramer and pentamer. The reference zero line is included into the figure for differentiating sign of
the bond charges. Different color for each of the lines in the figure represents individual intermolecular connections. [Color figure can be viewed at
wileyonlinelibrary.com]

Journal of Computational Chemistry 2017, 38, 1389–1409 1399


FULL PAPER WWW.C-CHEM.ORG

Figure 7. Time evolution of intramolecular and intermolecular bond charges of water dimer, trimer, tetramer, and pentamer during OAO intermolecular
charge transfer. The left and right panels in the figure correspond the time evolution of intramolecular and intermolecular bond charges of the water clus-
ters. Unequal length of the trajectories have been marked by unequal tick space along the time axis. The color code of each of the lines represent different
bonds present in the system. [Color figure can be viewed at wileyonlinelibrary.com]

charges and their corresponding distances between two water environment between two atoms. Since the local environment
molecules in simulated clusters as obtained by MD with OAO around two atoms can be different, the same separation of
and OAH charge transfer separately in Figure 5 (for OAO two atoms can be associated with nontrivial differences in the
charge transfer) and 6 (for OAH charge transfer). We show amount of charge transferred between the two atoms. The
similar plots for results in larger clusters for each of the inter- results as obtained from larger clusters by allowing OAO or
molecular bonds present in those clusters. We find that both OAH intermolecular charge transfer in dynamics are consistent
small and large clusters display similar correlation, and we with the calculated results as demonstrated before for the QE2
expect that such correlation may even hold in bulk systems. model. However, slight increase in the value of bond charge is
The obtained results for the allowing of OAO or OAH inter- observed at small values of intermolecular separation for the
molecular charge transfer follow almost similar type of correla- allowing of OAH charge transfer between two water mole-
tion. However, we find that the amount of charge transfer cules for larger cluster. Such increase in the values of bond
under those two charge transfer situations is significantly dif- charge may happen due to the several factors such as: (a) the
ferent. Both of the figures show that the bond charge and the scaling function or parametrized overlap integral is not a per-
distance between two water molecules are correlated for the fect exponential function at small values of internuclear sepa-
allowing of OAO and OAH charge transfer cases between ration, rather it is almost constant at that region of distances,
them. In particular, the correlation as obtained for dimer fol- (b) the collective effect of all bond charge variables to the spe-
lows similar relationship as demonstrated in the previous sec- cific one, (c) equation of motions for charge dynamics are non-
tion. Therefore, the model that we propose in this work, linear in nature, and so forth.
shows the correct behavior of charge transfer in the dynamics. In the next section, we discuss the dynamics of the bond
With increasing separation between the atoms of a bond charges, atomic charges and the derived electrostatic proper-
charge, there is a general decrease of charge transferred. The ties such as dipole moment and polarizability of each of the
profiles are not single-valued; this phenomenon arises from clusters. Finally, we discuss the time averaged properties of
the fact a particular bond charge variable represents the local water clusters those we obtain from the molecular charge

1400 Journal of Computational Chemistry 2017, 38, 1389–1409 WWW.CHEMISTRYVIEWS.COM


WWW.C-CHEM.ORG FULL PAPER

Figure 8. Time evolution of intramolecular and intermolecular bond charges of water dimer, trimer, tetramer, and pentamer during OAH intermolecular
charge transfer. The left and right panels in the figure correspond the time evolution of intramolecular and intermolecular bond charges of the water clus-
ters. Unequal length of the trajectories have been marked by unequal tick space along the time axis. The color code of each of the curves represent differ-
ent bonds present in the system. [Color figure can be viewed at wileyonlinelibrary.com]

dynamics for each of the water clusters in the following intermolecular charge transfer suggests that the two water mol-
section. ecules are close to each other at that particular instant. As a
result the values of intramolecular bond charge decrease since
Dynamical properties. In Figures 7 and 8, we show the time the whole system is neutral. Since the distance between two
evolution of intramolecular and intermolecular bond charge water molecules present in the system changes during dynam-
variables as obtained by allowing OAO or OAH charge trans- ics, the amount of intramolecular and intermolecular bond
fer between two water molecules for four water clusters. The charge also follow that fact. Once again, we have seen from the
left panels show the intramolecular bond charge transfer dynamics that the amount of transferred charge through OAH
(intramolecular bond charge transfer is charge transferred intermolecular bond charge variable is larger as compared to
between the bonded atoms in the actual molecule) and the the OAO intermolecular bond charge variable. However, the
right panels show intermolecular bond charge transfer (charge molecular charge dynamics under both charge transfer situa-
transferred between molecules along the allowed bond charge tions is almost similar for each of the water clusters.
variable; in the case of the dimer, there is only 1 bond charge The new distribution of bond charge influences the atomic
variable, in the case of the trimer, 2; in the case of the tetra- charges of each of the water molecules present in different
mer, 3; and in the case of the pentamer, 4 bond charge varia- clusters. In Figures 9 and 10, we present the results as
bles). Shown are trajectories of individual bond charge obtained from four water clusters by the allowing of either
variables; only a single trajectory for each cluster has been OAO or OAH intermolecular charge transfer. The figures show
computed, and thus shown in these figures. In general both of that the atom and bond charges follow similar behavior in the
the intermolecular and intramolecular bond charge variables dynamics of the clusters. However, atomic charges are more
are dynamic and fluctuate around an equilibrium value and realistic than bond charges and give a clear understanding of
the fluctuation is similar for all the clusters. For instance, there such charge dynamics for the water cluster systems.
is a peak at about 3 ns for both intermolecular and intramo- The charge dynamics of individual water molecules is pre-
lecular bond charge of dimer. A positive peak for sented in Figures 11 and 12. The result shows that each of the

Journal of Computational Chemistry 2017, 38, 1389–1409 1401


FULL PAPER WWW.C-CHEM.ORG

Figure 9. Time evolution of oxygen (left panels) and hydrogen (right panels) atomic charges water clusters for OAO charge transfer. Unequal length of the
trajectories have been marked by unequal tick space along the time axis. The color code for each of the curves represent different atoms present in the
system. [Color figure can be viewed at wileyonlinelibrary.com]

water molecules possess a net charge and they are occasional- such as dipole moment and polarizability are also expected to
ly neutral during dynamics, however, the sum of the charges be dynamic. We computed the magnitude of dipole moment
of individual water molecules at an instant is always zero. and polarizability of each of the water clusters using the pro-
Therefore, the neutrality of system is well maintained in the tocol as described in previous section (Theoretical background
dynamics, which is also shown in the figure. The variation of section). We calculated those properties by the allowing of
water charges as a function of time for each of the clusters either OAO or OAH intermolecular charge transfer in different
are different, but they fluctuate around an equilibrium value. water clusters. The results are presented in Figures 13 and 14.
Once again we see from the figure that the fluctuation of As we mentioned before, the magnitude of dipole moment of
water charge by allowing OAO and OAH intermolecular the system depends on the length and orientation of each of
charge transfer situations is similar. However, a water molecule the bond dipole moment vectors present in the system. Since
gains more charge if it transfers through OAH intermolecular both of the intermolecular separation and the bond charge
bond charge variable. Such result is also in agreement with are dynamical variables, hence as a consequence, each of the
our previous analysis. Therefore, we expect that the QE2 mod- bond dipole moment vectors of the system is dynamic, which
el captures a realistic picture of charge dynamics for each of is reflected in the figure. Moreover, the figure shows that the
the water cluster systems. We note that the initial charges on magnitude of net dipole moment vector for each of the water
individual water molecules may or may not be identically neu- clusters is comparable in the short simulation. But the magni-
tral due to the fact that we start from classically optimized tude of dipole moment vector for the allowing of OAH inter-
geometries; at these configurations, some charge transfer is molecular charge transfer is larger as compared to the
allowed, and depending on the configuration, some molecules allowing of OAO intermolecular charge transfer. The polariz-
may be neutral, while others not. In total, the systems are all ability for each of the systems shows an interesting feature in
neutral during the entire trajectory as the total system charge the dynamics. The figures show that the magnitude of polariz-
is constrained to be zero. Since the atomic and molecular ability by allowing of either OAO or OAH intermolecular
charges are dynamic, the electrostatic properties of the system charge transfer is quite different in dynamics. This is because

1402 Journal of Computational Chemistry 2017, 38, 1389–1409 WWW.CHEMISTRYVIEWS.COM


WWW.C-CHEM.ORG FULL PAPER

Figure 10. Time evolution of oxygen (left panels) and hydrogen (right panels) atomic charges water clusters for OAH charge transfer. Unequal length of
the trajectories have been marked by unequal tick space along the time axis. The color code for each of the curves represent different atoms present in
the system. [Color figure can be viewed at wileyonlinelibrary.com]

the polarizability of the system is quadratically related with for each of the water clusters. The results are shown in Figures
the internuclear separations. Therefore, it indicates that the 15–17. It is clear from Figure 15 that there is an optimum sep-
nuclei dynamics for each of the systems may be quite different aration between two water molecules in the dynamics of
under the two charge transfer situations. Moreover, the magni- water cluster systems. We observed that such results are true
tude of polarizability increases with the increase of the size of for both OAO and OAH intermolecular charge transfer cases.
the cluster, which is again in agreement with our common We define the relative orientation between two water mole-
understanding. cules as the angle between two dipole moment vectors of
Thus, it is clear from the above discussion that the molecu- them. The distribution of relative orientation between a pair of
lar charge dynamics for the allowing of either OAO or OAH water molecule is shown in Figure 16. The results have an
charge transfer between two water molecules is quite similar. interesting feature, we observed that there is a preferential ori-
However, the magnitude of transferred charge by allowing entation between the two water molecules those are engaged
OAH intermolecular charge transfer is larger as compared to in charge transfer. The preference in orientation increased
the results obtained by the allowing of OAO intermolecular from smaller to larger water clusters. We also observed from
charge transfer. The neutrality of each of the systems is well the figure that the pattern of such distribution by allowing
maintained during dynamics. Finally, the derived electrostatic OAO and OAH intermolecular charge transfer cases is similar.
properties such a dipole moment and polarizability are consis- Therefore, such geometrical preference, which is pretty com-
mon in the molecular systems, influences on the charge trans-
tent with our proposed model in the dynamics.
fer among the water molecules in the cluster systems (see Fig.
Static properties. In this section, we further analyzed the MD 16). Once again we have seen from the dynamics that the
trajectories of water clusters as obtained by allowing either magnitude of transferred charge for the allowing of OAH
OAO or OAH charge transfer between two molecules. Specifi- charge transfer between two water molecules is larger as com-
cally, we calculated the distributions of intermolecular distance, pared to the magnitude obtained by allowing OAO charge
relative orientation between two water molecules, and charges transfer between them. The distance between two water

Journal of Computational Chemistry 2017, 38, 1389–1409 1403


FULL PAPER WWW.C-CHEM.ORG

Figure 11. Time evolution of water charges in water dimer, trimer, tetramer, and pentamer for allowing intermolecular OAO charge transfer. Different
length of the trajectory for each of the clusters has been depicted by unequal tick space along x-axis. The color codes of each curve represent different
water molecules present in the system. The neutrality line (orange) as shown in each panel is the sum of all the water charges at any instant. [Color figure
can be viewed at wileyonlinelibrary.com]

Figure 12. Time evolution of water charges in water dimer, trimer, tetramer, and pentamer for allowing intermolecular OAH charge transfer. Different
length of the trajectory for each of the clusters has been depicted by unequal tick space along x-axis. The color codes of each curve represent different
water molecules present in the system. The neutrality line (orange) as shown in each panel is the sum of all the water charges at any instant. [Color figure
can be viewed at wileyonlinelibrary.com]

1404 Journal of Computational Chemistry 2017, 38, 1389–1409 WWW.CHEMISTRYVIEWS.COM


WWW.C-CHEM.ORG FULL PAPER

Figure 13. Time evolution of system dipole moment (left panels) and polarizability (right panels) of water dimer, trimer, tetramer, and pentamer during
OAO intermolecular charge transfer. Different length of the trajectory for each of the clusters has been depicted by unequal tick space along x-axis.

molecules and orientation of each of the water molecules charge distribution for dissociative molecular systems. Our
affect the magnitude of charge transfer. We show the time new QE2 model is more general and applicable for both
average magnitude of transferred charge between two nearest homonuclear and heteronuclear charge transfer in molecular
nalyser water molecules present in four clusters in Figure 15. system. To formulate the QE2 model, we introduce a distance
Once again we found optimum values of transferred charge, dependent scaling function into the original QE model, which
which is correlated with intermolecular bond distance. ensures that the charge transfer variables vanish asymptoti-
Thus, it is clear from the above discussion that there is a cally with distances. The model works perfectly for systems
geometrical preference for the allowing of charge transfer those are not far from equilibrium configuration. Moreover,
among water molecules present in the clusters. An orientation- we show that only distance dependent pairwise electronega-
al preference for the intermolecular charge transfer increases tivity into the model does not solve the issue of unusual dis-
from smaller to larger clusters. Such geometrical preference is tribution of atomic charges over large distances in a general
pretty common in chemical and biochemical systems. Thus, manner.
our analyses reveal that the combination of the geometrical We first obtain the parameters of the system using simulat-
preference and the magnitude of bond charge determine the ed annealing method. We chose a system of clusters contain-
stability, hydrogen bonding preference, and other physico- ing 2, 3, 4, 5 water molecules in this study. Explicitly, we
chemical properties of a chemical system. Therefore, we need minimize a cost function, which is the difference between two
an extensive study on larger cluster systems and we are cur- charges as obtained from QM calculation and the QE2 model,
rently doing that in our laboratory. contains all the free parameters of the system. We consider all
of the water clusters together during the minimization. Once
we obtain the parameters of the system, we analyze the water
Summary
dimer system. In the following step, we couple the QE2 model
In this work, we develop a general charge equilibration model with molecular mechanics energy function, construct the Ham-
in bond charge space (QE) that exhibits correct behavior of iltonian and perform dynamics of each of each of the water

Journal of Computational Chemistry 2017, 38, 1389–1409 1405


FULL PAPER WWW.C-CHEM.ORG

Figure 14. Time evolution of system dipole moment (left panels) and polarizability (right panels) of water dimer, trimer, tetramer, and pentamer during
OAH intermolecular charge transfer. Different length of the trajectory for each of the clusters has been depicted by unequal tick space along x-axis.

Figure 15. Distribution of intermolecular distances between two nearest neighbor water molecules present in four clusters. The analyses have been performed
using the MD trajectories as obtained for allowing OAO and OAH intermolecular charge transfer. [Color figure can be viewed at wileyonlinelibrary.com]

1406 Journal of Computational Chemistry 2017, 38, 1389–1409 WWW.CHEMISTRYVIEWS.COM


WWW.C-CHEM.ORG FULL PAPER

Figure 16. Distribution of angles between two dipole moment vectors of two nearest neighbor water molecules present in four clusters. The analyses
have been performed using the MD trajectories as obtained for allowing OAO and OAH intermolecular charge transfer. [Color figure can be viewed at
wileyonlinelibrary.com]

clusters. Note that we perform all of our calculation using our also validate that the model is applicable for both OAO and
in-house MATLAB program. OAH intermolecular charge transfers. However, we find that
First, we critically nalyse our model by translating one of the the magnitude of intermolecular charge transfer for allowing
water molecule present in water dimer. We show that our OAH intermolecular charge transfer is larger as compared to
model works perfectly at the asymptotic limit of distances. We the OAO charge transfer. Such result is quite reasonable, and

Figure 17. Time average magnitude of transferred charge between two nearest neighbor water molecules present in four clusters. The analyses have
been performed using the MD trajectories as obtained for allowing OAO and OAH intermolecular charge transfer. [Color figure can be viewed at
wileyonlinelibrary.com]

Journal of Computational Chemistry 2017, 38, 1389–1409 1407


FULL PAPER WWW.C-CHEM.ORG

it is in agreement with our model because the large differences polarizability of the system grows quadratically with the
in electronegativity and hardness between Oxygen and Hydro- increase of internuclear separation. Therefore, we wish to con-
gen atoms drive such larger amount of charge transfer. Since the trol such increase in polarizability of the system in a systematic
atomic charges of the system change with intermolecular separa- manner. However, the central issue in this problem is that the
tion, it is expected that the electrostatic properties of the system dimension of the system in bond charge space is not identical
follow that fact. We compute both dipole moment and polariz- at short and large internuclear distances. Such situation is
ability of the dimer system as a function of distance between pretty common in chemical reactions, where the formation
two water molecules. Our analysis reveals that the model produ- and breaking of the chemical bonds take place. Therefore, we
ces correct behavior of dipole moment of the system but it does aim to remove the irrelevant bond charge variable from the
not capture the correct behavior of the polarizability of the sys- system in a systematic manner. We consider the net polariz-
tem at large values of intermolecular separation. However, the ability of the system is a mixture of two polarizabilities of the
model is reasonable to use for further study for a bound system. system under allow and disallow charge transfer conditions in
Since the QE2 model is a reasonable choice for a bound sys- an appropriate proportion.
tem, we use it to perform molecular dynamics simulation for Consider a bimolecular system and there is only one intermo-
each of the water clusters within a sphere. We show a correlation lecular bond charge variable then the polarizability tensor of
between intermolecular distances and bond charge variables the system, a, can be defined as
through the dynamics. Existence of such a correlation reinforces
the assessment of the QE2 model. Our results also reveal that the a5haa 1ð12hÞad (A1)
new distribution of bond charge modifies the the atomic and
where, h is the mixing function, which takes a value from a
water charges of the system. We also show that the neutrality of
distance dependent function, hðrÞ, where r is the intermolecu-
the system is well maintained during the dynamics. The new dis-
lar distance. It is clear from the equation that the contribution
tribution of the charges affects the electrostatic properties such
of aa is dominating if the intermolecular charge transfer is
as dipole moment and polarizability of the system. Importantly,
allowed. Conversely, ad is dominating if there is no intermolec-
we find that the dynamics of charges and derived dipole
ular charge transfer. Multiply electric field vector, ~
e, on both
moment of the systems by allowing both OAO and OAH charge
sides of the above equation, we can obtain the induced dipole
transfers follow similar behavior. However, we find a significant
moment, ~ l ind for each of the situations
difference in dynamics of the polarizability of the systems
between the OAO and OAH intermolecular charge transfer.
~ l aind 1ð12hÞ~
l ind 5h~ l dind (A2)
Since the polarizability of the system is a function of bond hard-
ness matrix and molecular configurations, hence it may be l ind can be expressed in the following way.
The ~
expected that the configurations and charges of the systems
under those two conditions are not identical. We analyze inter- DRT DP00 5DRaT HDPa 1DRdT ðI2HÞDDPd (A3)
molecular separation, orientation, and the charge for each of the
trajectories of water cluster. Our analysis reveals that the configu- where, DR and DP are the displacement and induced bond
rations of the system are similar under two charge transfer condi- charge vectors; the elements of diagonal matrix, H are either 0.5
tions, but the intermolecular charges of the systems under those or take a value from the hðrÞ function for intramolecular and
two charge transfer conditions are significantly different. intermolecular charge transfer, respectively, and I is the identity
We discuss that the QE2 model works well for chemically rea- matrix. The above equation is general and applicable for multi-
sonable structure but it overestimates the polarizability of the sys- molecular system. Since, the dimensions of both of the vectors,
tem. To overcome this problem, we demonstrate another model, DPa and DPd are not identical, a rectangular D matrix is intro-
which we refer to as mixed QE2 model. The model constructs a duced into the equation. Since the configuration of the system
new bond charge variable, which is a combination of two bond is unchanged under those three situations, hence
charge variables of the system as obtained from QE2 model under
allow and disallow scenarios. A Gaussian function, which we intro- DRT 5DRaT 5DRdT (A4)
duce in this model, combines those two bond charges. We dem-
onstrate the model and calculate the electrostatic properties of therefore,
water dimer system in this study. The mixed QE2 model over-
DP00 5HDPa 1ðI2HÞDDPd (A5)
comes drawbacks of the QE2 model. However, we need a general
formulation of the model and need to apply it on a large molecu- On rearrangement, we can obtain
lar system. We are currently doing that in our laboratory.
   
P~e 00 2P00 5H P~ea 2Pa 1ðI2HÞD P~ed 2Pd (A6)
Appendix
Mixed QE2 Model or
In this article, we introduce another new model, which we
refer as mixed QE2 model [Eq. (26)]. As discuss before, the P~e 00 2P00 5 HP~ea 2ðI2HÞDP~ed 2 HPa 2ðI2HÞDPd (A7)

1408 Journal of Computational Chemistry 2017, 38, 1389–1409 WWW.CHEMISTRYVIEWS.COM


WWW.C-CHEM.ORG FULL PAPER

where the ~e represents the external electric field. Comparing [11] D. V. Sakharov, C. Lim, J. Am. Chem. Soc. 2005, 127, 4921.
both sides [12] I. Ivanov, J. A. Tainer, J. A. McCammon, Proc. Natl. Acad. Sci. USA 2007,
104, 1465.
[13] F. May, M. Al-Helwi, B. Baumeier, W. Kowalsky, E. Fuchs, C. Lennartz, D.
P00 5HPa 1ðI2HÞDPd (A8) Andrienko, J. Am. Chem. Soc. 2012, 134, 13818.
[14] M. Bag, T. S. Gehan, D. D. Algaier, F. Liu, G. Nagarjuna, P. M. Lahti, T. P.
Therefore, the bond charge variable under mixed QE2 model, Russell, D. Venkataraman, Adv. Mater. 2013, 25, 6411.
p00i is [15] A. Warshel, M. Kato, A. V. Pisliakov, J. Chem. Theory Comput. 2007, 3,
2034.
[16] P. Ren, J. W. Ponder, J. Phys. Chem. B 2003, 107, 5933.
p00i 5hpai 1ð12hÞd0 pdi (A9)
[17] Y. Haibo, W. F. v. Gunsteren, Comput. Phys. Commun. 2005, 172, 69.
[18] G. Lamoureux, B. Roux, J. Chem. Phys. 2003, 119, 3025.
Substituting the new bond charge variable into eq. (14), we [19] S. J. Stuart, B. J. Berne, J. Phys. Chem. 1996, 100, 11934.
can define the new energy expression for mixed QE2 model as [20] S. Patel, C. L. Brooks, III, Mol. Simul. 2006, 32, 231.
[21] J. Chen, T. J. Martınez, Chem. Phys. Lett. 2007, 438, 315.
NX [22] A. Warshel, Annu. Rev. Biophys. Biomol. Struct. 2003, 32, 425.
bond
1 NX X
bond Nbond

Eðp00 Þ5 Dvi s21 00


i pi 1 gij s21 21 00 00
i sj pi pj (A10) [23] U. W. Schmitt, G. A. Voth, J. Chem. Phys. 1999, 111, 9361.
i51
2 i51 j5i [24] S. M. Valone, S. R. Atlas, Philos. Mag. 1999, 86, 9361.
[25] N. Gresh, G. A. Cisneros, T. A. Darden, J.-P. Piquemal, J. Chem. Theory
Comput. 2007, 3, 1960.
According to this model, the value of bond charge is con-
[26] M. S. Gordon, L. Slipchenko, L. Li, J. H. Jensen, D. C. Spellmeyer, R.
trolled by the mixing function. It is clear from the eqs. (A9) Wheeler, Annu. Rep. Comput. Chem. 2007, 3, 177.
and (A10) that the mixed QE2 model smoothly transforms [27] Y.-P. Liu, K. Kim, B. J. Berne, R. A. Friesner, S. W. Rick, J. Chem. Phys.
from allow to disallow QE2 charge transfer models as the 1998, 108, 4739.
intermolecular separation increases. [28] J. L. Banks, G. A. Kaminski, R. Zhou, D. T. Mainz, B. J. Berne, R. A.
Friesner, J. Chem. Phys. 1999, 110, 741.
[29] G. A. Kaminski, H. A. Stern, B. J. Berne, R. A. Friesner, Y. X. Cao, R. B.
Acknowledgment Murphy, R. Zhou, T. A. Halgren, J. Comput. Chem. 2002, 23, 1515.
[30] H. A. Stern, G. A. Kaminski, J. L. Banks, R. Zhou, B. J. Berne, R. A.
SP thanks N. Patel for fruitful discussion and encouragement for the Friesner, J. Phys. Chem. B 1999, 103, 4730.
duration of this work. [31] R. Chelli, P. Procacci, R. Righini, S. Califano, J. Chem. Phys. 1999, 111, 8569.
[32] K. Shimizu, H. Chaimovich, J. P. S. Farah, L. G. Dias, D. L. Bostick,
Keywords: charge transfer  charge equilibration  polariza- J. Phys. Chem. B 2004, 108, 4171.
tion  force field  molecular dynamics  simulation [33] R. A. Nistor, J. G. Polihronov, M. H. M€ user, N. J. Mosey, J. Chem. Phys.
2006, 125, 094108.
[34] S. K. S, Model. Simul. Mat. Sci. Eng. 2015, 23, 074006.
[35] M. Soniat, L. Hartman, S. W. Rick, J. Chem. Theory Comput. 2015, 11,
How to cite this article: S. Kumar Sinha, M. Mehta, S. Patel. J. 1658.
Comput. Chem. 2017, 38, 1389–1409. DOI: 10.1002/jcc.24789 [36] J. Chen, D. Hundertmark, T. J. Martınez, J. Chem. Phys. 2008, 129, 214113.
[37] W. J. Mortier, S. K. Ghosh, S. Shankar, J. Am. Chem. Soc. 1986, 108, 4315.
] Additional Supporting Information may be found in the [38] A. K. Rapp e, W. A. Goddard, III, J. Phys. Chem. 1991, 95, 3358.
online version of this article. [39] W. G. Lee, J. E. Davis, S. Patel, J. Chem. Phys. 2008, 128, 144110.
[40] B. A. Bauer, T. R. Lucas, A. Krishtal, C. V. Alsenoy, S. Patel, J. Phys.
Chem. B 2010, 114, 8984.
[1] X. Zhu, P. E. M. Lopes, A. D. J. MacKerell, WIREs. Comput. Mol. Sci. [41] R. F. Nalewajski, J. Korchowiec, Z. Zhou, Int. J. Quantum Chem. Symp.
2012, 2, 167. 1988, 22, 349.
[2] O. F. Lange, D. V. D. Spoel, B. L D. Groot, Biophys. J. 2010, 99, 647. [42] D. J. Wales, M. P. Hodges, Chem. Phys. Lett. 1998, 286, 65.
[3] K. A. Beauchamp, Y.-S. Lin, R. Das, V. S. Pande, J. Chem. Theory Comput. [43] M. J. Frisch, et al., Gaussian09 Revision D.01; Gaussian Inc.: Wallingford,
2012, 8, 1409. CT 2009.
[4] E. A. Cino, W.-Y. Choy, M. Karttunen, J. Chem. Theory Comput. 2012, 8, [44] A. Corana, M. Marchesi, C. Martini, S. Ridella, ACM Trans. Math. Softw.
2725. 1987, 13, 262.
[5] M. R. Shirts, J. W. Pitera, W. C. Swope, V. S. Pande, J. Chem. Phys. 2003, [45] S. W. Rick, J. Chem. Phys. 2001, 1114, 2276.
119, 5740. [46] MATLAB Release, The MathWorks, Inc.; Natick, Massachusetts, 2013.
[6] R. B. Best, N.-V. Buchete, G. Hummer, Biophys. J. 2008, 95, L07. [47] W. Humphrey, A. Dalke, K. Schulten, J. Mol. Graph. 1996, 14, 33.
[7] S. Y. Noskov, S. Berneche, B. Roux, Nature 2004, 431, 830.
[8] D. Buchera, S. Raugeib, L. Leonardo Guidonia, M. D. Perarod, U.
Rothlisbergera, M. L. Carloni, P. Klein, Biophys. Chem. 2006, 124, 292. Received: 1 September 2016
[9] H. Tan, W. Qu, G. Chen, R. Liu, J. Phys. Chem. A 2005, 109, 6303. Revised: 16 January 2017
[10] E. Ramos-Cordoba, D. S. Lambrecht, M. Head-Gordon, Faraday Discuss. Accepted: 17 January 2017
2011, 150, 345. Published online in Wiley Online Library

Journal of Computational Chemistry 2017, 38, 1389–1409 1409

You might also like