Subscriber access provided by Columbia University Libraries

Article
Machine Learning Force Field Parameters from Ab Initio Data
Ying Li, Hui Li, Frank C Pickard, Badri Narayanan, Fatih Sen, Maria K. Y.
Chan, Subramanian Sankaranarayanan, Bernard R. Brooks, and Benoît Roux
J. Chem. Theory Comput., Just Accepted Manuscript • DOI: 10.1021/acs.jctc.7b00521 • Publication Date (Web): 11 Aug 2017
Downloaded from http://pubs.acs.org on August 12, 2017

Just Accepted

“Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted
online prior to technical editing, formatting for publication and author proofing. The American Chemical
Society provides “Just Accepted” as a free service to the research community to expedite the
dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts
appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been
fully peer reviewed, but should not be considered the official version of record. They are accessible to all
readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered
to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published
in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just
Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor
changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers
and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors
or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

Journal of Chemical Theory and Computation is published by the American Chemical

Society. 1155 Sixteenth Street N.W., Washington, DC 20036
Published by American Chemical Society. Copyright © American Chemical Society.
However, no copyright claim is made to original U.S. Government works, or works
produced by employees of any Commonwealth realm Crown government in the course
of their duties.
Page 1 of 24 Journal of Chemical Theory and Computation

1
2 Machine Learning Force Field Parameters from Ab Initio Data
3
4
5 Ying Li,a Hui Li,b Frank C Pickard IV,c Badri Narayanan,d Fatih Sen,d Maria K. Y. Chan,d,e Subramanian
6 Sankaranarayanan,d,e Bernard R. Brooks,c and Benoît Rouxb,e
7
8
9 a
Leadership Computing Facility, Argonne National Laboratory, IL 60439, USA
10 b
Department of Biochemistry and Molecular Biophysics, University of Chicago, IL 60637, USA
11 c
Laboratory of Computational Biology, National Heart, Lung and Blood Institute, National Institutes of Health,
12
Bethesda, Maryland 20892, USA
13 d
14
Center for Nanoscale Materials, Argonne National Laboratory, IL 60439, USA
e
15 Computational Institute, University of Chicago, IL 60637, USA
16
17
18 ABSTRACT
19
20 Machine learning (ML) techniques with the genetic algorithm (GA) have been applied to explore a polarizable
21
force field parameters using only ab initio data from quantum mechanics (QM) calculations of molecular clusters at
22
23 the MP2/6-31G(d,p), DFMP2(fc)/jul-cc-pVDZ, and DFMP2(fc)/jul-cc-pVTZ levels to predict experimental
24
25 condensed phase properties (i.e., density and heat of vaporization). The performance of this ML/GA approach is
26 demonstrated on 4,943 dimers electrostatic potentials and 1,250 clusters interaction energies for methanol.
27
28 Excellent agreement between the training dataset from QM calculations and the optimized force field model can be
29
30 achieved. Better results are achieved by introducing an offset factor during the machine learning process to
31 compensate for the discrepancy of the QM calculated energy and the energy reproduced by optimized force field,
32
33 where the offset factor maintain the local “shape” of the QM energy surface. Throughout the machine learning
34
process, experimental observables were not involved in the objective function, but were only used for model
35
36 validation. The best model, optimized from the QM data at the DFMP2(fc)/jul-cc-pVTZ level, appears to perform
37
38 even better than the original AMOEBA force field (amoeba09.prm), which was optimized empirically to match
39 liquid properties. The present effort shows the possibility of using machine learning techniques to develop
40
41 descriptive polarizable force field using only QM data. The ML/GA strategy to optimize force field parameters
42
43 described here could easily be extended to other molecular systems.
44
45
46
47 SECTION: Molecular Mechanics, Quantum Chemistry, Machine Learning and Force Field
48
49
50
51 INTRODUCTION
52
53 Classical simulations based on detailed atomic models and molecular mechanical (MM) potential functions are
54
55 increasingly playing an important role in physics, chemistry, biochemistry, and materials science. To obtain
56
57 meaningful results, the accuracy of the potential function, or force field, underlying such molecular dynamics (MD)
58 simulations is of critical importance.1-8 The simple force fields that are widely used for biomolecular simulations, in
59
60
1
ACS Paragon Plus Environment
 Journal of Chemical Theory and Computation Page 2 of 24

1
2
particular, are empirically optimized to reproduce a number of calculated quantum mechanics (QM) and
3 experimental properties.4 While such an empirical approach can provide atomic models of useful accuracy, they can
4
5 do so only if the simulated system of interest is built from the set of previously parameterized chemical
6 functionalities. This has a critical impact on the predictive power of MD simulations in general. For instance, in the
7
8 absence of any prior knowledge from experiment, it is unclear if the properties of a neat liquid of a compound
9
10 comprising previously untreated atom types could be reliably predicted. A force field that requires an empirical
11 calibration against experimental data is expected to be inherently more limited when applied to systems that depart
12
13 from a pre-defined set of chemical functionalities.8
14
15
The primary objective of the present study is to test the ability of machine learning (ML) techniques with the
16 genetic algorithm (GA) to parameterize a polarizable force field by relying exclusively on ab initio QM data. A
17
18 central question that we are trying to answer with the present effort is whether the resulting model can accurately
19 predict neat liquid properties of a compound over a range of temperatures from only QM data, without any a priori
20
21 information from experiment. The general idea that one could develop molecular potential functions from first
22
23 principles QM data rests on ideas from early efforts by Clementi and co-workers in the 1970’s.9 They constructed
24 an additive force field model by fitting the potential energies for the water dimer in various geometrical
25
26 configurations calculated using the QM configuration-interaction method. However, the resulting MCY model did
27
28
not accurately simulate liquid water. In retrospect, a serious limitation was the total neglect of many-body (MB)
29 effects. To test our ML/GA strategy, it is important to choose a functional form that incorporates a correct physical
30
31 representation of many-body effects and that provides an accurate description of physicochemical properties of the
32 condensed phase. Force fields that explicitly account for induced electronic polarization offer a step in the right
33
34 direction.10-16 While non-polarizable additive force fields used in biomolecular simulations, which represent
35
36 polarization in an average manner with effective fixed partial charges, can achieve a reasonably accurate
37 representation of the condensed phase, they do so as the result of numerous empirical compromises.4, 17
38
39 Consequently, it is possible that the functional form of non-polarizable additive force fields might be too limited to
40
41
achieve the desired physical accuracy over a range of thermodynamic states.
42 Different functional forms of polarizable force fields have been proposed for simulating chemical and biomolecular
43
44 systems. Those polarizable force fields include models based on classical Drude oscillators,16, 18-19 the CHARMM
45
charge equilibration (CHEQ) model based on fluctuating partial charges,15 the AMBER ff02 based on inducible
46
47 dipoles,20 and the Atomic Multipole Optimized Energetics for Biomolecular Applications (AMOEBA) force
48
49 field.21-22 While most models remain at the level of atomic point charges, the AMOEBA force field further
50 incorporates static field contributions from atomic dipoles and quadrupoles. More recently, there has been
51
52 considerable progress in the design of potential energy functions to simulate water by building upon rigorous
53
54 representations of the many-body expansion of the interaction energies.23-25 Schmidt and collaborators have
55 advanced a strategy on the basis of the symmetry-adapted perturbation theory (SAPT) methodology,26-27 yielding
56
57 models with relatively good properties for dense liquid systems. Similarly, the latest SAPT models may need to
58
incorporate Axilrod−Teller three-body dispersion and exchange to yield accurate models of dense systems.28
59
60
2
ACS Paragon Plus Environment
Page 3 of 24 Journal of Chemical Theory and Computation

1
2
Similarly, the MB-pol model developed by Paesani and co-workers has demonstrated that it is possible to achieve
3 high accuracy across all phases of water through a quantitative and physically correct representation of both short-
4
5 and long-range many-body contributions.25 Of particular importance, the MB-pol model incorporates many-body
6 van der Waals interactions. Considerations of explicit multi-body van der Waals terms go back to early work by
7
8 Mezei29 and Clementi.30-31 However, it is not yet clear whether such a rigorous treatment of many-body effects
9
10 could be routinely implemented with more complicated molecules. The present effort, we have chosen to work with
11 the AMOEBA functional form because of its relative mathematical simplicity and flexibility.
12
13 We have chosen the methanol molecule to explore the feasibility of our ML/GA parameterization strategy. This
14
15
choice is motivated by the observation that methanol represents a molecule of sufficient complexity comprising
16 both a hydroxyl polar group, able to form hydrogen bonds, as well as a bulky non-polar alkyl group.32 The
17
18 competing nature of these interactions makes methanol an excellent test case to explore the feasibility of our
19 ML/GA strategy to generate force field parameters that relies exclusively on QM data.
20
21
22 METHODS
23
24 The complete functional form of the AMOEBA force field has been described in detail elsewhere.21, 33 The present
25
26 focus is mainly on the non-bonded electrostatic and van der Waals (vdW) interaction parameters. Because the
27
internal covalent interaction (bond, angle, dihedral and torsional parameters) are not expected to greatly affect the
28
29 properties of the condensed liquid phase, they remained unchanged from the original AMOEBA force field
30
31 (amoeba09.prm).22 To compute the electrostatic energy (Eelec) between methanol molecules, the AMOEBA force
32 field requires 44 independent parameters. Those parameters are monopole (q), dipole (µx, µy, µz), quadrupole
33
34 expressed as a traceless symmetric matrix with five independent elements (Qxx, Qyx, Qyy, Qzx, Qzy), atomic
35
36 polarizability (α) for the four methanol atom types (O, C, HC, and HO). A unique Thole damping factor (a) was
37
used for all atoms, following the AMOEBA convention. The AMOEBA force field represents the vdW interactions
38
39 as a buffered 14-7 potential.34 To compute the vdW energy (EvdW) between methanol molecules, the functional form
40
41 requires 10 parameters. Eight of those parameters are the Rmin, and εmin for the four atoms types of methanol.
42
Finally, two additional parameters (the so-called reduction factor λ) are required to scale the position of the two
43
44 hydrogen atoms interaction site along their corresponding covalent bond.14
45
46 Force field parameterization, i.e. the determination of the optimal parameters associated with a complex
47 functional form, is a challenging optimization problem in a space of high dimensionality. Local optimization
48
49 approaches generally carry out this task by attempting to minimize an objective function using a simple gradient-
50
51 based algorithm (e.g., steepest descent, conjugate gradient method),35-36 and are only effective if the starting point is
52 sufficiently close to a satisfactory solution. However, even with pre-existing knowledge about the objective
53
54 function and the parameter space, a number of problems can arise. For example, the convergence to optimal
55
56
parameters is not guaranteed when the objective function is rugged, non-differentiable, or when the initial value
57 deviates significantly from the global minimum. Ultimately, the range of parameter values that must be explored
58
59 grows dramatically as the complexity of the force field functional form increases. Here, to overcome the challenge
60
3
ACS Paragon Plus Environment
 Journal of Chemical Theory and Computation Page 4 of 24

1
2
encountered with force field parameterization using ML techniques, we have used the genetic algorithm (GA),
3 which is an evolutionary algorithm that mimics the process of natural selection.37 Previous successful applications
4
5 of the genetic algorithm in similar contexts include the determination of the parameters for the ReaxFF reactive
6 force field,2-3, 5 and various force fields including Morse+QEq charge transfer ionic potential (CTIP),6 and a new
7
8 hybrid bond-order potential (HyBOP) for materials system.2-3, 5-6, 38-39
9
10 In the present work, electrostatic parameters were determined in a first stage, and vdW parameters were
11 optimized in a second stage (i.e., the electrostatic parameters were kept unchanged while the vdW parameters are
12
13 optimized on the second stage). The optimization of the electrostatic parameters of methanol was carried out in two
14
15
steps. In a first step, the value of the atomic multipoles was obtained from the QM electrostatic potential on the
16 Connolly surface of a single isolated methanol molecule. The QM electrostatic potential of a methanol monomer,
17
18 φQM, was calculated at the MP2 level of theory with various basis sets including Pople-style40 and correlation
19
20
consistent41 basis sets. The results are given in Table 1. A Distributed Multipole Analysis (DMA)42-44 was carried
21 out using the Gaussian Distributed Multipole Analysis (GDMA 2.2) program42, 45-46 for the electronic density results
22
23 from different MP2 basis sets. The multipole parameters of each atomic site were identified using the Tinker21, 47-51
24 package (the Poledit and Potential program). In a second step, all 44 electrostatic parameters were refined from the
25
26 QM electrostatic potential calculated from an extensive training dataset of 4,943 methanol dimers using the genetic
27
28 algorithm. This second step differs from the standard protocol used to determine the electrostatic parameters for the
29 AMOEBA polarizable force field from QM calculations,22, 33
which relies on the unperturbed monomer. By
30
31 considering dimers in this situation, the mutual induced polarization of the two molecules is explicitly taken into
32
33
account, allowing a direct determination of the atomic polarizabilities. The methanol dimer configurations were
34 sampled through placing two methanol molecules, where one methanol was sampled over shell radius (1- 4 Å) on
35
36 another methanol’s Connolly surface. In total, 4,943 methanol dimer configurations were sampled. These 4,943
37 configurations of methanol dimers were relaxed through MP2 geometry optimization via Gaussian09 program,52
38
39 while fixing the position of the carbon atoms. The optimization of electrostatic parameters seeks to minimize the
40
41 objective function ∆ as the averaged root mean square deviation (ARMSD) between the QM result and the
42 parameterized force field calculation, shown in equation (1):
43
44 N cluster
1
45 ∆=
N cluster
∑ χ 2 ( p1, p2 ,... pm )
46 j=1
47
N cluster  n grid  (1)
48 1
= ∑  1 ∑ (φ QM − φ FF ( p1, p2 ,... pm ) ) 2 
49  ngrid k=1 k k

50 N cluster j=1  j
51
52 where
 ,  , …   represent the electrostatic parameters of the force field and  = 4,943. It is noteworthy
53
54 that the static monopole, dipoles and quadrpoles remained essentially unchanged during this second step, which
55 primarily affected the optimization of the atomic polarizabilies.
56
57 In a second stage, the parameters for the vdW interactions were optimized to best-match the interaction between a
58
59 central methanol molecule and its nearest neighbors via supermolecular QM computations for a large number of
60
4
ACS Paragon Plus Environment
Page 5 of 24 Journal of Chemical Theory and Computation

1
2
clusters extracted from classical MD simulations of liquid methanol using the original amoeba09.prm.22 The idea of
3 exploiting simulation-derived ensemble to optimize force field parameters was first suggested by. Mezei,.53 The
4
5 electrostatic parameters optimized in the previous step are kept unchanged at this stage. Second-order Møller–
6 Plesset perturbation theory54 (MP2) was employed as the QM method to calculate the interaction energy, with basis
7
8 sets superposition error (BSSE) counterpoise correction.55 This level of theory and basis sets has been widely
9
10 verified by previous studies to predict energetics and structural properties.56 One set of QM calculations was carried
11 out at the MP2/6-31G(d,p) with BSSE level using the Gaussian09 program.52 Two additional sets density-fitted
12
13 MP2 and frozen core approximation QM calculations were carried out at the DFMP2(fc)/jul-cc-pVDZ and
14
15
DFMP2(fc)/jul-cc-pVTZ level with BSSE counterpoise corrections using the Psi4 quantum chemistry package.57 A
16 total of 1,250 clusters were included in the training set (999 clusters of 9 molecules, 157 clusters of 11 molecules,
17
18 and 94 clusters of 13 molecules). The MD simulation system used to generate the snapshots from which the clusters
19 were extracted comprises 344 methanol molecules. The MD trajectory was generated under periodic boundary
20
21 conditions (PBC) in the NpT ensemble at constant pressure and temperature using the original AMOEBA force
22
23 field (amoeba09.prm).22 To acquire a wide range of clusters configurations, NpT ensemble simulations were
24 performed on the system at various temperatures and pressures, from T = 193.15 K, p = 1.0 atm, T = 293.15 K, p =
25
26 1.2 atm, to T = 393.15 K, p = 6.293 atm, with final stable density of the system as ρ = 0.933, 0.794 and 0.627 g/ml,
27
28 respectively. These MD simulations were generated with an integration time step of 1 fs using the GPU dynamics
29 program in the Tinker-OpenMM suite (http://biomol.bme.utexas.edu/tinker-openmm). Table 3 provides further
30
31 details about the clusters extracted from the different MD simulation systems. The optimization of the vdW
32
33
parameters seeks to minimize the objective function F given by,
34
F( p1, p2 ,..., pm ) = ∑ ( ∆E QM ( p1, p2 ,..., pm ))
35 2

36 j − δ − ∆E MM
j
(2)
j
37
38
39 where δ is an offset factor and ∆E QM
j and ∆E MM
j are the interaction energies between the central molecule and the
40
41 surrounding molecules for the j-th cluster in the training set calculated from the QM and MM force field,
42 respectively. The vdW parameters in the force field were optimized using the interaction energy from 1,250
43
44 methanol clusters by employing the developed ML/GA framework.6-7, 38
45
A flowchart of the ML/GA optimization protocol is depicted in Figure 1. Briefly, the optimization starts with a
46
47 set of parent parameters, which is defined as the population. For parameters optimization of the force field, the
48
49 parents may have different ranges of settings. Some of the individuals in this population present a better fit, which
50
in the context of parameters optimization means lower value in the objective function, e.g., ∆ and F in Eqs (1) and
51
52 (2). Fit parents survive and are allowed to mate, which is accomplished by crossing patterns with other fit
53
54 individuals. During crossover, random mutations in the genes are also allowed to a certain degree to avoid a
55 stagnant gene pool and a better sampling of the parameters space. The offspring individuals form the next
56
57 generation of parents and this process continues until some pre-defined criteria are met. For the ML/GA
58
59 optimization of electrostatic parameters, the process was initiated by generating a population of sets of parameters
60
5
ACS Paragon Plus Environment
 Journal of Chemical Theory and Computation Page 6 of 24

1
2
randomly, such that their values lie within physically allowable limits, which are ±1.5 times of value from Tinker
3 Poledit and Potential programs for methanol monomer at MP2/6-311G(d,p) level of QM calculations. The
4
5 searching range of the 44 independent electrostatic parameters over which the ML/GA optimization was is given in
6 Table 2.
7
8 For the ML/GA optimization of the vdW parameters, the process was initiated by generating a population a
9
10 population of Np = 120 parameters sets randomly, such that their values lie within physically allowable limits. The
11 searching range of the 11 independent vdW parameters over which the ML/GA optimization was performed is
12
13 given in Table 4. For each set of parameters, we compute the interaction energies for all structures in the training
14
15
dataset using Tinker and evaluate the objective function F given by Eq. (2). Sets of parameters were then ranked in
16 58-59
ascending order. After the ranking, non-linear roulette wheel selection was performed to select the best 60%
17
18 members, i.e., the ones with lowest values of F, which were then subjected to genetic operations: mutation and
19 crossover with crossover-rate 3%. These mutations introduce sufficient diversity into the population, and the non-
20
21 linear selection scheme helps to avoid premature convergence of the ML/GA run. After the genetic operations, both
22
23 the parent and offspring sets of parameters are ranked by their value of F. The best Np parameters sets are then
24 chosen to constitute the next generation. Such an optimization routine ensures that only satisfactory parameters sets
25
26 survive after each generation; upon repeating this workflow for sufficient generations and sampling viable regions
27
28
in the parameters space, we performed three separate ML/GA runs starting with different random populations.
29 From each of the converged ML/GA run, we chose the final parameters set corresponding the lowest value of F.
30
31 Monitoring the evolution of the parameters during the iterations of the GA algorithm shows that they initially vary
32 rapidly until they settle down and undergo very small variations (< 0.0001%) around final values. The calculated
33
34 correlation coefficients between the vdW parameters from the GA after the initial variations show that some of the
35
36 vdW parameters are coupled. For example, the parameters εmin for HO and HC are anti-correlated, which is expected
37 because the decrease in one interaction must be compensated by an increase in the other to match the target energy
38
39 during the GA optimization. This code is freely available for download at
40
https://github.com/AmYingLi/GA4AMOEBA.
41
42
43 RESULTS AND DISCUSSION
44
45 We successfully generated electrostatic and vdW parameters for the methanol molecule consistent with the
46
47 functional form of the AMOEBA force field by exclusively taking QM data of molecular clusters as training
48
49
datasets and using machine learning techniques. For the sake of clarity, we emphasize that the intended objective of
50 the present effort is not to offer a re-calibration of the original AMOEBA force field for methanol that could be
51
52 transferable to other systems. Our central goal is to test whether it is practically feasible to get parameters for a
53
force field of a physically reasonable functional form without any prior experimental data, and to be sufficiently
54
55 accurate for liquid simulations. For this reason, we refer to these results as “optimized force field models”, and
56
57 reserve the acronym AMOEBA only for the original force field: amoeba09.prm.
58
59
60
6
ACS Paragon Plus Environment
Page 7 of 24 Journal of Chemical Theory and Computation

1
2
In the following, we present the results for the electrostatic parameters and the vdW parameters from the
3 genetic algorithm optimization by calibrating the electrostatic potential and the interaction energy for methanol
4
5 molecules. Finally, a number of additional classical MD simulations were carried out using the set of optimized
6
parameters to calculate the density (ρ) and heat of vaporization (∆Hvap) of liquid methanol to verify the accuracy of
7
8 the ML/GA strategy to generate force field parameters. The condensed liquid phase system consists of 344
9
10 methanol molecules simulated with periodic boundary conditions (PBC) in the NpT ensemble at a constant pressure
11
of 1 atm as temperature varies from -5°C to 60°C. The density and heat of vaporization were averaged from three
12
13 different simulations up to 2.5 ns initiated with different random velocities.
14
15 One important methodological question for a machine learning strategy is whether the number of clusters
16 included in the training set is sufficiently large to yield a statistically meaningful sampling for parameters
17
18 determination. As a control, to ascertain the validity of the ML/GA strategy, we first attempted to recover the vdW
19
20 parameters using interaction energy generated directly from the original amoeba09.prm force field (the original
21 electrostatic parameters from amoeba09.prm were kept unchanged for this test). If the ML/GA strategy is
22
23 functioning properly, then the generated parameters should be very close to the original vdW parameters
24
amoeba09.prm. In practice, it was found that a minimum dataset of 600 clusters of 9 methanol molecules was
25
26 necessary to accurately recover the amoeba09.prm parameters, where 300 clusters were extracted from MD
27
28 trajectories generated at atmospheric pressure and corresponding density, and 300 clusters were extracted from MD
29 trajectories generated at high pressure. The resulting liquid properties from MD simulations using the ML/GA
30
31 generated vdW parameters were essentially identical to those of the original amoeba09.prm. This test confirmed
32
33 that our ML/GA protocol was effective, as long as the training set was sufficiently large (more than several hundred
34 configurations).
35
36 The electrostatic parameters were determined by training the atomic multipoles (q, , ), atomic
37
38 polarizabilities (α) and Thole’s factor (a) from the QM electrostatic potential on the Connolly surface of a methanol
39
40 molecule using the Gaussian Distributed Multipole Analysis (GDMA 2.2) tool42, 45 and the Tinker package.47, 49-50
41 We then refined the atomic polarizabilities and Thole’s factor with a set of methanol dimers using the ML/GA. The
42
43 ML/GA generated electrostatic parameters from the MP2/6-311G(d,p) level are given in Table 5. Because the
44
45 AMOEBA force field adopts a universal Thole’s damping factor, during the refinement step, the only varying
46 parameter for each atom type is the polarizability. From table 5, we can see what are the change of the
47
48 polarizability (α) and Thole’s factor (a) through learning the electrostatic potential from a monomer to 4,943
49
50 dimers (denoted as
). As shown in Table 6, the electrostatic potential calculated from this ML/GA generated
51 force field model closely matches the QM at the MP2/6-311G(d,p) level for 4,943 methanol dimers. In fact, the
52
53 deviations are smaller than those given by the original parameterization of the AMOEBA force field
54
(amoeba09.prm).22 The new optimized force fields were cross-validated by comparing the MM and QM
55
56 electrostatic potential for 502 clusters of 9 methanol molecules, which were not used in the training process. The
57
58 results, given in Table 6, show that the deviations between the force field with ML/GA generated parameters and
59 the QM calculations are smaller than those obtained with amoeba09.prm.22 It is important to note that such
60
7
ACS Paragon Plus Environment
 Journal of Chemical Theory and Computation Page 8 of 24

1
2
comparisons with AMOEBA are meaningful because essentially the same QM methods [MP2/6-311G(d,p) and
3 MP2/aug-cc-pVTZ] were originally used to determine the amoeba09.prm electrostatic parameters.22 Therefore, the
4
5 discrepancies of the resulting parameters is likely due to sampling differences. Because the operational purpose of
6 the permanent electrostatic multipoles determined here is to reproduce interaction energies and forces between
7
8 molecules, they may differ slightly from the multipoles determined via a spatial partitioning, scheme, such as based
9
10 on localized Wannier functions.60-62
11 The current procedure used to determine the electrostatic parameters, which considers the mutual induced
12
13 polarization of the two molecules in methanol dimers, differs from the standard AMOEBA protocol.22, 33 In the
14
15
latter, one commonly fits the permanent electrostatic multipoles from a QM electrostatic potential (ESP) map at a
16 series of point around the molecule of interest. However, while considering an isolated molecule certainly provides
17
18 the necessary information to fit the static atomic electrostatic multipoles of the unperturbed molecule, it also makes
19 it difficult to determine the optimal set of atomic polarizabilities that are needed. To overcome this challenge, we
20
21 exploited the mutually perturbed ESP map for homodimers in the determination of atomic polarizabilites (while
22
23 accounting explicitly for the mutual induced polarization of the molecules from one another). In practice, all sets of
24 electrostatic parameters, determined either from the monomer of homodimer QM data, remain fairly similar to one
25
26 another. This overall consistency reflects the fact that the electrostatic features of a small molecule like methanol
27
28
are accurately constrained by the QM calculations. In contrast, the determination of the vdW parameters without
29 any experimental data is much more challenging. The vdW parameters in the AMOEBA force field
30
31 (amoeba09.prm) were empirically adjusted to yield accurate liquid properties. A key question is whether a strategy
32 exclusively based on QM calculations has the ability to yield force field models of equivalent accuracy.
33
34 The vdW parameters were optimized to best match the interaction between a central methanol molecule and the
35
36 finite polarization response to its nearest neighbors via supermolecular QM computations for a large number of
37 clusters representative of the liquid phase. This approach is different than formal perturbative approaches that seek
38
39 to estimate intermolecular interactions from the wave function of an isolated monomer such as symmetry-adapted
40
41
perturbation theory (SAPT).26-28, 63-64 In principle, SAPT could be used as an alternative route to generate the target
42 data for the ML/GA optimization. A total of 1,250 methanol clusters of 9, 11 and 13 molecules were considered in
43
44 the supermolecular QM cluster computations. Typical cluster configurations from the training set are depicted in
45
Figure 2. The interactions were characterized by considering three different levels of QM calculations: MP2/6-
46
47 31G(d,p), DFMP2(fc)/jul-cc-pVDZ, and DFMP2(fc)/jul-cc-pVTZ. The distribution of the interaction energy for
48
49 1,250 methanol clusters calculated at the MP2/6-31G(d,p) is shown in Figure 2(d). The distribution converging to a
50 Gaussian-like shape suggests that the number of the configurations is adequate to provide a sufficient sampling of
51
52 these systems. The ML/GA generated vdW parameters for all force field models are given in Table 7.
53
54 First we discuss the results from force field models optimized without the offset factor δ in Eq. (2). This means
55 that the parameters are generated to match the absolute value of the QM interaction energies for the training set.
56
57 The optimized force field models generated from the different QM calculations, MP2/6-31G(d,p), DFMP2(fc)/jul-
58
59
cc-pVDZ, and DFMP2(fc)/jul-cc-pVTZ, converge reasonably well, with correlation coefficients of 0.974, 0.977
60
8
ACS Paragon Plus Environment
Page 9 of 24 Journal of Chemical Theory and Computation

1
2
and 0.979, respectively. This is confirmed by observing Figure 3 (a), (b), and (c), where the interaction energies
3 from QM and the force field models are strongly correlated. However, the ultimate test is to verify the ability of the
4
5 optimized force field model to accurately predict condensed phase properties of liquid methanol. Table 8 shows the
6 result of density and heat of vaporization from polarizable force fields optimized from QM data at the MP2/6-
7
8 31G(d,p), DFMP2(fc)/jul-cc-pVDZ and DFMP2(fc)/jul-cc-pVTZ levels. Results from the original AMOEBA force
9
10 field (amoeba09.prm)22 are included for comparison. The performance of the force field models optimized to match
11 the absolute value of the QM interaction energies (offset factor δ set to zero) is disappointing. The liquid density
12
13 from the model based on MP2/6-31G(d,p) is 0.401 g/ml, and the density from the model based on DFMP2(fc)/jul-
14
15 cc-pVDZ is 0.568 g/ml, which are in poor agreement with experiment (0.786 g/ml). The heat of vaporization from
16 these two models is also too small. The model based on DFMP2(fc)/jul-cc-pVTZ performs slightly better, with a
17
18 density of 0.686 g/ml and a heat of vaporization of 8.58 kcal/mol.
19
20 To address this shortcoming, an offset factor δ was introduced in Eq. (2). The purpose of the offset factor is to
21 free the parameter optimization from the absolute magnitude of the QM interaction energies.65-67 It should be noted
22
23 that the offset factor δ is not directly used in the force field functional form, but only acting as a hyper-parameter
24
25 for the GA. In essence, when δ is included in Eq. (2), the force field parameters are generated to reproduce the local
26 “shape” of the energy surface of the cluster as a function of the atomic coordinates, while releasing the requirement
27
28 to match the absolute magnitude of the QM interaction energies. This strategy has some similarities with force-
29
30 matching methodologies, where a force field is optimized on the basis of the local first derivative of the energy
31 function,68 though here the emphasis is put on relative interactions energies rather than on trying to match local
32
33 forces. With a non-zero offset factor (δ ≠ 0), the force field parameters generated from ML/GA based on the
34
35 different QM training datasets, MP2/6-31G(d,p), DFMP2(fc)/jul-cc-pVDZ, and DFMP2(fc)/jul-cc-pVTZ,
36 converges similarly well, with correlation coefficients of 0.955, 0.971, and 0.975, respectively. The effect of the
37
38 offset factor between the interaction energies from QM and the force field models is noticeable in Figure 3 (d), (e),
39
and (f). However, as observed in Table 8, these force field models clearly produced liquid properties that are closer
40
41 to experiment. While the inaccuracy of force field optimized without the offset δ becomes less important with
42
43 better QM data, the trend in Table 8 suggests that meaningful information about the overall shape of the energy
44
surface and the relative interaction energies is already contained within lower level QM calculations. The best
45
46 parameter set, obtained from the QM data at the DFMP2(fc)/jul-cc-pVTZ level, appears to perform even better than
47
48 the original AMOEBA force field (amoeba09.prm),22 which was optimized empirically to match liquid properties.
49 This force field model is also able to reproduce the radial distribution functions of liquid methanol determined from
50
51 neutron scattering,69 as shown in Figure 4.
52
53 Figures 3g, 3h, and 3i compare the interaction energies from QM and the original AMOEBA force field
54 (amoeba09.prm),22 which signifies that different basis sets of the MP2 method give out different accuracy. The
55
56 original AMOEBA force field (amoeba09.prm)22 is directly compared with the QM data in Figure 3 (g, h and i).
57
This comparison leads to an interesting observation; there appears to be a shift between AMOEBA and the QM
58
59 calculations that closely reflects the offset factor δ used in the force field models optimization. While the vdW
60
9
ACS Paragon Plus Environment
 Journal of Chemical Theory and Computation Page 10 of 24

1
2
parameters of the AMOEBA force field (amoeba09.prm)22 field are empirically adjusted to yield accurate liquid
3 properties, the shifts observed in Figures 3g, 3h, and 3i essentially mirror the offset factor δ used in the force field
4
5 models optimization very consistently. This observation reinforces the conclusion that seeking to match the
6
absolute interaction energies based on MP2/6-31G(d,p) and DFMP2(fc)/jul-cc-pVDZ levels leads to inaccurate
7
8 force field models.
9
10 A comparison of the interaction energy from the QM calculations and from the force field with the optimized
11 parameters is shown in Figure 3 for the 1,250 clusters. Considering Figure 3, one can see that the ML/GA generated
12
13 hyper-parameters, offset factor δ, decreases with increasingly accurate QM method, equal to 4.4 kcal/mol for
14
15 MP2/6-31G(d,p), 2.5 kcal/mol for DFMP2(fc)/jul-cc-pVDZ, and 1.2 kcal/mol for DFMP2(fc)/jul-cc-pVTZ). This is
16 also consistent with the observation that DFMP2(fc)/jul-cc-pVTZ is the QM data leading to the best force field
17
18 model optimized to match the absolute value of the QM interaction energies (offset factor δ set to zero). By
19
20 introducing a hyper-parameter in GA, an offset factor δ in Eq. (2), we relieve the parameter optimization from the
21
constraint of matching the absolute value of the interaction energy. ML in the presence of the offset factor yields
22
23 force field models that try to reproduce the “shape” of the potential energy surface in the multi-dimensional space
24
25 of the cluster coordinates. The relative accuracy of the different parameters generated by ML/GA for the AMOEBA
26 force field model in Table 8 suggests that the relative “shape” of the potential energy surface is more important than
27
28 matching the absolute value of the interaction energies. Ultimately, discrepancies (the offset factor δ) of interaction
29
30 energy calculated from MP2 and the ML/GA generated parameters for AMOEBA force field model are likely due
31 to the incompleteness of the basis sets for the particular level of QM (MP2) theory and the insufficiency sampling
32
33 of configurations in parameterization of force field.56, 65-67 This might be the reason that higher level QM data is
34
35 able to elucidate some details that are difficult to capture at the lower level. The obvious lesson from this is that the
36 QM level must be sufficiently high.
37
38 As a final test of the accuracy of the polarizable force field optimized from QM data, additional sets of MD
39
simulations were generated to examine the density and heat of vaporization of liquid methanol as a function of
40
41 temperature. For this test, we used the vdW parameters optimized from the QM data at the DFMP2(fc)/jul-cc-
42
43 pVDZ and DFMP2(fc)/jul-cc-pVTZ levels. Figure 5 shows ρ and ∆Hvap as temperature varies from -5°C to 60°C
44
calculated from NpT MD simulations. We observe that the polarizable force field optimized from QM data
45
46 accurately predicts the density and heat of vaporization, in excellent accord with experiments, over the range of
47
48 temperatures.
49 Finally, it is important to emphasize that consideration of many-body dispersion effects, even with pair-wise
50
51 long-range vdW interactions, appears to be crucial to generate accurate force field models. For instance, attempts to
52
53 use ML of the vdW parameters based on the training dataset from the interaction energy calculated at the MP2/6-
54 31G(d,p) level for a large set of methanol dimers spanning a wide range of carbon-carbon (2 to 10 Å) distance
55
56 failed to yield an accurate ML/GA generated parameters for AMOEBA force field. While the force field model
57
with ML/GA generated parameters succeeds in accurately reproducing the dimer QM training data, the resulting
58
59 force field model fails to provide an accurate representation of the bulk liquid system. MD simulations (at 25°C and
60
10
ACS Paragon Plus Environment
Page 11 of 24 Journal of Chemical Theory and Computation

1
2
1 atmosphere) showed that such a force field model, with machine learning of vdW parameters only from dimers,
3 severely underestimates the liquid density (0.572 g/ml) and heat of vaporization (7.32 kcal/mol) compared to
4
5 experiment. We also used CCSD(T)70-72 to calculate the interaction energies of methanol dimers. While a good
6 correlation between the QM target data and the interaction energy from the force field model was obtained (R=
7
8 0.998), the model did not yield accurate condensed matter properties when used in MD simulations. This shows
9
10 that using training data from sufficiently large cluster for the parameterization is important. The lesson here is that
11 machine learning of the vdW parameters using QM data on clusters larger than simple dimers was essential to
12
13 incorporate many-body dispersion effects into the force field model. This observation is consistent with previous
14
15
results from McDaniels and Schmidt, who showed that Axilrod-Teller three-body dispersion and exchange terms
16 needed to be incorporated in their SAPT force field of methanol to obtain a satisfactory model of the liquid phase.28
17
18 In force field optimization, it is often tempting to put more emphasis on training data obtained from the highest
19
possible level of QM. The challenge is that, to remain computationally tractable, one is typically limited to smaller
20
21 systems of a few molecules that are not able to capture all the pertinent information to represent the features of a
22
23 condensed phase system. This example shows that the training data must include QM calculations on sufficiently
24 large cluster sizes. Even though the functional form for the vdW interactions comprises only pair-wise terms, it
25
26 appears that many-body dispersion effects are “effectively” incorporated during the optimization of the vdW
27
28 parameters based on the cluster QM data. While this may appear surprising, the situation is analogous to the fixed
29 charges in an additive non-polarizable force field, which are empirically optimized to account for polarization in an
30
31 effective manner.
32
33
CONCLUSIONS
34
35
36 The present study demonstrates the feasibility of predicting the condensed phase properties (i.e., density and heat of
37 vaporization) of a substance by optimizing the parameters of a polarizable force field using QM data exclusively—
38
39 without any a priori information from experiment. Experimental data was only used for validating the final force
40
field models. This idea was tested on methanol, a small yet challenging molecule containing both polar and
41
42 nonpolar moieties. A genetic algorithm was utilized to overcome the challenges of parameter optimization in a high
43
44 dimensional space. For the electrostatic component of the force field, we optimized all the multipoles,
45 polarizabilities, and Thole damping factors of methanol. The ML/GA based on dimers was able to nearly reproduce
46
47 the canonical AMOEBA electrostatic parameters, the optimized parameters displaying smaller deviations relative to
48
49 the QM electrostatic potential than the AMOEBA original parameters.
50 The vdW parameters were optimized by attempting to match the interaction energy between a central molecule
51
52 and the surrounding molecules for a large number of methanol clusters. These clusters were extracted from
53
snapshots of condensed phase MD simulations to be as representative as possible of the liquid phase. Following this
54
55 protocol, excellent agreement between the optimized force field model and the QM calculations could be achieved.
56
57 However, only the QM calculation from fairly high level provides sufficiently accurate vdW parameters for MD
58 simulation to get proper condensed phase properties of methanol. Consistently better results were obtained by a
59
60
11
ACS Paragon Plus Environment
 Journal of Chemical Theory and Computation Page 12 of 24

1
2 using a parameterization strategy allowing for an offset factor δ between the force field and the QM target data.
3 Through this strategy, which bears some similarities with force-matching methodologies,68 the force field is
4
5 optimized to reproduce the local “shape” of the QM energy surface without matching its absolute value. Without
6
7 the offset factor δ, the absolute magnitude of the QM interactions is incorporated into the MM force field via the
8 optimized vdW parameters, yielding models that appear to be less accurate in simulations of dense liquids.
9
10 One of the main difficulties of parameterizing a force field without any experimental input stems from the
11
reliability of the QM calculations. Here, we consider three QM methods: MP2/6-31G(d,p), DFMP2(fc)/jul-cc-
12
13 pVDZ and DFMP2(fc)/jul-cc-pVTZ. It is well known that MP2 with different basis sets can lead to different
14
15 accuracy. Furthermore, there appears to be an additional issue with the overestimation of interaction energy from
16 MP2 calculations, when using a small basis set.56, 65-67 The present results show that QM calculations relying on
17
18 more extensive basis sets lead to a more accurate determination of the interaction energies. In principle, sampling of
19
20 larger molecular clusters would also be expected to increase accuracy. Nevertheless, the current strategy was able
21 to leverage moderate computation resource to tune the force field parameters for reasonable agreement between the
22
23 MD simulated data and experimental values.
24
25
26
ACKNOWLEGDMENTS
27
28 We thank Nichols A. Romero and Stephen Gray for their help. Y.L. was supported by the Margaret Butler
29 Postdoctoral Fellowship at Argonne Leadership Computing Facility. We gratefully acknowledge the computing
30
31 resources provided on Blues, a high-performance computing cluster operated by the Laboratory Computing
32
33
Resource Center at Argonne National Laboratory. This research used resources of the Argonne Leadership
34 Computing Facility, which is a DOE Office of Science User Facility supported under Contract DE-AC02-
35
36 06CH11357. Use of the Center for Nanoscale Materials, an Office of Science user facility, was supported by the U.
37
S. Department of Energy, Office of Science, Office of Basic Energy Sciences, under Contract No. DE-AC02-
38
39 06CH11357. This work was completed in part with resources provided by the University of Chicago Research
40
41 Computing Center. B.R. was supported by grant R01-GM072558 from the National Institutes of Health. FCP and
42 BRB acknowledge support from the intramural research program of the National Heart, Lung and Blood Institute of
43
44 the National Institutes of Health; and utilized high-performance computational capabilities of the LoBoS clusters at
45
46 the National Institutes of Health (http://www.lobos.nih.gov).
47
48
49
50
51
52
53
54
55
56
57
58
59
60
12
ACS Paragon Plus Environment
Page 13 of 24 Journal of Chemical Theory and Computation

1
REFERENCES
2
3
1. van Duin, A. C. T.; Dasgupta, S.; Lorant, F.; Goddard, W. A., ReaxFF: A reactive force field for hydrocarbons. J Phys
4
Chem A 2001, 105 (41), 9396-9409.
5
2. Islam, M. M.; Kolesov, G.; Verstraelen, T.; Kaxiras, E.; van Duin, A. C., eReaxFF: A Pseudoclassical Treatment of
6
Explicit Electrons within Reactive Force Field Simulations. J Chem Theory Comput 2016, 12 (8), 3463-72.
7
3. Pahari, P.; Chaturvedi, S., Determination of best-fit potential parameters for a reactive force field using a genetic
8
algorithm. J Mol Model 2012, 18 (3), 1049-1061.
9
4. Mackerell, A. D., Jr., Empirical force fields for biological macromolecules: overview and issues. J Comput Chem
10 2004, 25 (13), 1584-604.
11 5. Larsson, H. R.; van Duin, A. C. T.; Hartke, B., Global optimization of parameters in the reactive force field ReaxFF
12 for SiOH. J Comput Chem 2013, 34 (25), 2178-2189.
13 6. Sen, F. G.; Kinaci, A.; Narayanan, B.; Gray, S. K.; Davis, M. J.; Sankaranarayanan, S. K. R. S.; Chan, M. K. Y.,
14 Towards accurate prediction of catalytic activity in IrO2 nanoclusters via first principles-based variable charge force
15 field. J Mater Chem A 2015, 3 (37), 18970-18982.
16 7. Narayanan, B.; Sasikumar, K.; Mei, Z. G.; Kinaci, A.; Sen, F. G.; Davis, M. J.; Gray, S. K.; Chan, M. K. Y.;
17 Sankaranarayanan, S. K. R. S., Development of a Modified Embedded Atom Force Field for Zirconium Nitride Using
18 Multi-Objective Evolutionary Optimization. J Phys Chem C 2016, 120 (31), 17475-17483.
19 8. Massobrio, C.; Du, J.; Bernasconi, M.; Salmon, P. S., Molecular Dynamics Simulations of Disordered Materials:
20 From Network Glasses to Phase-Change Memory Alloys. 2015; Vol. 215, p 1-529.
21 9. Matsuoka, O.; Clementi, E.; Yoshimine, M., CI Study of water dimer potential surface. J. Chem. Phys. 1976, 64,
22 1351-1361.
23 10. Belle, D. V.; Couplet, I.; Prevost, M.; Wodak, S. J., Calculations of electrostatic properties in proteins. J. Mol. Biol.
24 1987, 198, 721-735.
25 11. Cieplak, P.; Kollman, P. A.; Lybrand, T., A new water potential including polarization: Application to gas-phase,
26 liquid and crystal properies of water. J. Chem. Phys. 1990, 92, 6755-6760.
27 12. Rick, S. W.; Stuart, S. J.; Berne, B. J., Dynamical fluctuating charge force field: Application to liquid water. J Chem
28 Phys 1994, 101, 6141-6156.
29 13. Lopes, P. E. M.; Roux, B.; Mackerell, A. D., Jr., Molecular modeling and dynamics studies with explicit inclusion of
30 electronic polarizability. Theory and applications. Theor Chem Acc 2009, 124 (1-2), 11-28.
31 14. Ponder, J. W.; Wu, C. J.; Ren, P. Y.; Pande, V. S.; Chodera, J. D.; Schnieders, M. J.; Haque, I.; Mobley, D. L.;
32 Lambrecht, D. S.; DiStasio, R. A.; Head-Gordon, M.; Clark, G. N. I.; Johnson, M. E.; Head-Gordon, T., Current
33 Status of the AMOEBA Polarizable Force Field. J Phys Chem B 2010, 114 (8), 2549-2564.
34 15. Bauer, B. A.; Patel, S., Recent applications and developments of charge equilibration force fields for modeling
35 dynamical charges in classical molecular dynamics simulations. Theor Chem Acc 2012, 131 (3).
36 16. Lemkul, J. A.; Huang, J.; Roux, B.; MacKerell, A. D., Jr., An Empirical Polarizable Force Field Based on the
37 Classical Drude Oscillator Model: Development History and Recent Applications. Chem Rev 2016, 116 (9), 4983-
38 5013.
39 17. Mackerell, A. D., Jr.; Bashford, D.; Bellott, M.; Dunbrack, R. L.; Evanseck, J. D.; Field, M. J.; Fischer, S.; Gao, J.;
40 Guo, H.; Ha, S.; Joseph-McCarthy, D.; Kuchnir, L.; Kuczera, K.; Lau, F. T. K.; Mattos, C.; Michnick, S.; Ngo, T.;
41 Nguyen, D. T.; Prodhom, B.; Reiher, W. E.; Roux, B.; Schlenkrich, M.; Smith, J. C.; Stote, R.; Straub, J.; Watanabe,
42 M.; Wiorkiewicz-Kuczera, J.; Yin, D.; Karplus, M., All-atom empirical potential for molecular modeling and
43 dynamics studies of proteins. J Phys Chem B 1998, 102 (18), 3586-3616.
44 18. Lamoureux, G.; MacKerell, A. D.; Roux, B., A simple polarizable model of water based on classical Drude
45 oscillators. J Chem Phys 2003, 119 (10), 5185-5197.
46 19. Lamoureux, G.; Roux, B., Modeling induced polarization with classical Drude oscillators: Theory and molecular
47 dynamics simulation algorithm. J Chem Phys 2003, 119 (6), 3025-3039.
48 20. Wang, Z. X.; Zhang, W.; Wu, C.; Lei, H. X.; Cieplak, P.; Duan, Y., Strike a balance: Optimization of backbone
49 torsion parameters of AMBER polarizable force field for simulations of proteins and peptides. J Comput Chem 2006,
50 27 (6), 781-790.
51 21. Ren, P. Y.; Ponder, J. W., Polarizable atomic multipole water model for molecular mechanics simulation. J Phys
52 Chem B 2003, 107 (24), 5933-5947.
53 22. Ren, P. Y.; Wu, C. J.; Ponder, J. W., Polarizable Atomic Multipole-Based Molecular Mechanics for Organic
54 Molecules. J Chem Theory Comput 2011, 7 (10), 3143-3161.
55 23. Bukowski, R.; Szalewicz, K.; Groenenboom, G. C.; van der Avoird, A., Predictions of the properties of water from
56 first principles. Science 2007, 315 (5816), 1249-1252.
57 24. Wang, Y. M.; Huang, X. C.; Shepler, B. C.; Braams, B. J.; Bowman, J. M., Flexible, ab initio potential, and dipole
58 moment surfaces for water. I. Tests and applications for clusters up to the 22-mer. J Chem Phys 2011, 134 (9).
59 25. Paesani, F., Getting the Right Answers for the Right Reasons: Toward Predictive Molecular Simulations of Water
60 with Many-Body Potential Energy Functions. Acc Chem Res 2016, 49 (9), 1844-51.
13
ACS Paragon Plus Environment
 Journal of Chemical Theory and Computation Page 14 of 24

1
26. Schmidt, J. R.; Yu, K.; McDaniel, J. G., Transferable Next-Generation Force Fields from Simple Liquids to Complex
2
Materials. Acc Chem Res 2015, 48 (3), 548-556.
3 27. McDaniel, J. G.; Schmidt, J. R., Next-Generation Force Fields from Symmetry-Adapted Perturbation Theory. Annu
4 Rev Phys Chem 2016, 67, 467-488.
5 28. McDaniel, J. G.; Schmidt, J. R., First-Principles Many-Body Force Fields from the Gas Phase to Liquid: A
6 "Universal" Approach. J Phys Chem B 2014, 118 (28), 8042-8053.
7 29. Campbell, E. S.; Mezei, M., Use of a non Pair-Additive Intermolecular Potential Function to Fit Quantum-Mechanical
8 Data on Water Molecule Interactions. J. Chem. Phys. 1977, 67, 2338-2344.
9 30. Clementi, E. C., G, Monte-Carlo study of liquid water with 2- and 3-body ab initio potentials. Int. J. Quant. Chem.
10 1983, 10 (10), 31-41.
11 31. Clementi, E.; Kistenmacher, H.; Kołos, W.; Romano, S., Non-additivity in water-ion-water interactions. Theoret.
12 Chim. Acta 1980, 55, 257-266.
13 32. Kobayashi, T.; Shishido, R.; Mizuse, K.; Fujii, A.; Kuo, J. L., Structures of hydrogen bond networks formed by a few
14 tens of methanol molecules in the gas phase: size-selective infrared spectroscopy of neutral and protonated methanol
15 clusters. Phys Chem Chem Phys 2013, 15 (24), 9523-9530.
16 33. Wu, J. C.; Chattree, G.; Ren, P. Y., Automation of AMOEBA polarizable force field parameterization for small
17 molecules. Theor Chem Acc 2012, 131 (3).
18 34. Halgren, T. A., Representation of Vanderwaals (Vdw) Interactions in Molecular Mechanics Force-Fields - Potential
19 Form, Combination Rules, and Vdw Parameters. J Am Chem Soc 1992, 114 (20), 7827-7843.
20 35. Powell, M. J. D., Nonconvex Minimization Calculations and the Conjugate-Gradient Method. Lect Notes Math 1984,
21 1066, 122-141.
22 36. Fliege, J.; Svaiter, B. F., Steepest descent methods for multicriteria optimization. Math Method Oper Res 2000, 51 (3),
23 479-494.
24 37. McCall, J., Genetic algorithms for modelling and optimisation. J Comput Appl Math 2005, 184 (1), 205-222.
25 38. Narayanan, B.; Kinaci, A.; Sen, F. G.; Davis, M. J.; Gray, S. K.; Chan, M. K. Y.; Sankaranarayanan, S. K. R. S.,
26 Describing the Diverse Geometries of Gold from Nanoclusters to Bulk—A First-Principles-Based Hybrid Bond-Order
27 Potential. The Journal of Physical Chemistry C 2016, 120 (25), 13787-13800.
28 39. Cherukara, M. J.; Narayanan, B.; Kinaci, A.; Sasikumar, K.; Gray, S. K.; Chan, M. K. Y.; Sankaranarayanan, S. K. R.
29 S., Ab Initio-Based Bond Order Potential to Investigate Low Thermal Conductivity of Stanene Nanostructures. J Phys
30 Chem Lett 2016, 7 (19), 3752-3759.
31 40. Binkley, J. S.; Pople, J. A.; Hehre, W. J., Self-Consistent Molecular-Orbital Methods .21. Small Split-Valence Basis-
32 Sets for 1st-Row Elements. J Am Chem Soc 1980, 102 (3), 939-947.
33 41. Dunning, T. H., Gaussian-Basis Sets for Use in Correlated Molecular Calculations .1. The Atoms Boron through Neon
34 and Hydrogen. Journal of Chemical Physics 1989, 90 (2), 1007-1023.
35 42. Stone, A. J., Distributed Multipole Analysis, or How to Describe a Molecular Charge-Distribution. Chem Phys Lett
36 1981, 83 (2), 233-239.
37 43. Stone, A. J.; Alderton, M., Distributed Multipole Analysis - Methods and Applications. Mol Phys 1985, 56 (5), 1047-
38 1064.
39 44. Cisneros, G. A., Application of Gaussian Electrostatic Model (GEM) Distributed Multipoles in the AMOEBA Force
40 Field. J Chem Theory Comput 2012, 8 (12), 5072-5080.
41 45. Stone, A. J.; Alderton, M., Distributed multipole analysis - Methods and applications (Reprinted from Molecular
42 Physics, vol 56, pg 1047-1064, 1985). Mol Phys 2002, 100 (1), 221-233.
43 46. Stone, A. J., Distributed multipole analysis: Stability for large basis sets. J Chem Theory Comput 2005, 1 (6), 1128-
44 1132.
45 47. Ponder, J. W.; Richards, F. M., An Efficient Newton-Like Method for Molecular Mechanics Energy Minimization of
46 Large Molecules. J Comput Chem 1987, 8 (7), 1016-1024.
47 48. Kundrot, C. E.; Ponder, J. W.; Richards, F. M., Algorithms for Calculating Excluded Volume and Its Derivatives as a
48 Function of Molecular-Conformation and Their Use in Energy Minimization. J Comput Chem 1991, 12 (3), 402-409.
49 49. Dudek, M. J.; Ponder, J. W., Accurate Modeling of the Intramolecular Electrostatic Energy of Proteins. J Comput
50 Chem 1995, 16 (7), 791-816.
51 50. Kong, Y.; Ponder, J. W., Calculation of the reaction field due to off-center point multipoles. J Chem Phys 1997, 107
52 (2), 481-492.
53 51. Pappu, R. V.; Hart, R. K.; Ponder, J. W., Analysis and application of potential energy smoothing and search methods
54 for global optimization. J Phys Chem B 1998, 102 (48), 9725-9742.
55 52. Frisch, M. J.; Trucks, G. W.; Schlegel, H. B.; Scuseria, G. E.; Robb, M. A.; Cheeseman, J. R.; Scalmani, G.; Barone,
56 V.; Mennucci, B.; Petersson, G. A.; Nakatsuji, H.; Caricato, M.; Li, X.; Hratchian, H. P.; Izmaylov, A. F.; Bloino, J.;
57 Zheng, G.; Sonnenberg, J. L.; Hada, M.; Ehara, M.; Toyota, K.; Fukuda, R.; Hasegawa, J.; Ishida, M.; Nakajima, T.;
58 Honda, Y.; Kitao, O.; Nakai, H.; Vreven, T.; Montgomery Jr., J. A.; Peralta, J. E.; Ogliaro, F.; Bearpark, M. J.; Heyd,
59 J.; Brothers, E. N.; Kudin, K. N.; Staroverov, V. N.; Kobayashi, R.; Normand, J.; Raghavachari, K.; Rendell, A. P.;
60 Burant, J. C.; Iyengar, S. S.; Tomasi, J.; Cossi, M.; Rega, N.; Millam, N. J.; Klene, M.; Knox, J. E.; Cross, J. B.;
14
ACS Paragon Plus Environment
Page 15 of 24 Journal of Chemical Theory and Computation

1
Bakken, V.; Adamo, C.; Jaramillo, J.; Gomperts, R.; Stratmann, R. E.; Yazyev, O.; Austin, A. J.; Cammi, R.; Pomelli,
2
C.; Ochterski, J. W.; Martin, R. L.; Morokuma, K.; Zakrzewski, V. G.; Voth, G. A.; Salvador, P.; Dannenberg, J. J.;
3 Dapprich, S.; Daniels, A. D.; Farkas, Ö.; Foresman, J. B.; Ortiz, J. V.; Cioslowski, J.; Fox, D. J. Gaussian 09,
4 Gaussian, Inc.: Wallingford, CT, USA, 2009.
5 53. Mezei, M., On the Possibility of Obtaining an Effective Pairwise Additive Intermolecular Pontential via an Ab-initio
6 Route by Fitting to a Cooperative Model of Condensed Phase Configurations. J. Phys. Chem. 1991, 95, 7043-7049.
7 54. Headgordon, M.; Pople, J. A.; Frisch, M. J., Mp2 Energy Evaluation by Direct Methods. Chem Phys Lett 1988, 153
8 (6), 503-506.
9 55. Boys, S. F.; Bernardi, F., The calculation of small molecular interactions by the differences of separate total energies.
10 Some procedures with reduced errors (Reprinted from Molecular Physics, vol 19, pg 553-566, 1970). Mol Phys 2002,
11 100 (1), 65-73.
12 56. Li, H.; Ngo, V.; Da Siva, M. C.; Salahub, D. R.; Callahan, K.; Roux, B.; Noskov, S. Y., Representation of Ion-Protein
13 Interactions Using the Drude Polarizable Force-Field. J Phys Chem B 2015, 119 (29), 9401-9416.
14 57. Turney, J. M. S., A. C.; Parrish, R. M.; Hohenstein, E. G.; Evangelista, F. A.; Fermann, J. T.; Mintz, B. J.; Burns, L.
15 A.; Wilke, J. J.; Abrams, M. L.; Russ, N. J.; Leininger, M. L.; Janssen, C. L.; Seidl, E. T.; Allen, W. D.; Schaefer, H.
16 F., III; King, R. A.; Valeev, E. F.; Sherrill, C. D.; Crawford, T. D., Wiley Interdiscip. Rev. Comput. Mol. Sci. 2012, 2
17 (4), 556-565.
18 58. Lipowski, A.; Lipowska, D., Roulette-wheel selection via stochastic acceptance. Physica A 2012, 391 (6), 2193-2196.
19 59. Razali, N. M.; Geraghty, J. In Genetic Algorithm Performance with Different Selection Strategies in Solving TSP,
20 Proceedings of the World Congress on Engineering, 2011.
21 60. Wannier, G. H., The Structure of Electronic Excitation Levels in Insulating Crystals. Phys. Rev. 1937, 52, 191-197.
22 61. Marzari, N.; Vanderbilt, D., Maximally localized generalized Wannier functions for composite energy bands. Phys
23 Rev B 1997, 56 (20), 12847-12865.
24 62. Sagui, C.; Pomorski, P.; Darden, T. A.; Roland, C., Ab initio calculation of electrostatic multipoles with Wannier
25 functions for large-scale biomolecular simulations. J. Chem. Phys. 2004, 120 (9), 4530-44.
26 63. Parker, T. M.; Burns, L. A.; Parrish, R. M.; Ryno, A. G.; Sherrill, C. D., Levels of symmetry adapted perturbation
27 theory (SAPT). I. Efficiency and performance for interaction energies. J Chem Phys 2014, 140 (9), 094106.
28 64. McDaniel, J. G.; Schmidt, J. R., Physically-Motivated Force Fields from Symmetry-Adapted Perturbation Theory. J
29 Phys Chem A 2013, 117 (10), 2053-2066.
30 65. Jurecka, P.; Sponer, J.; Cerny, J.; Hobza, P., Benchmark database of accurate (MP2 and CCSD(T) complete basis set
31 limit) interaction energies of small model complexes, DNA base pairs, and amino acid pairs. Phys Chem Chem Phys
32 2006, 8 (17), 1985-1993.
33 66. Cybulski, S. M.; Lytle, M. L., The origin of deficiency of the supermolecule second-order Moller-Plesset approach for
34 evaluating interaction energies. J Chem Phys 2007, 127 (14).
35 67. Tkatchenko, A.; DiStasio, R. A.; Head-Gordon, M.; Scheffler, M., Dispersion-corrected Moller-Plesset second-order
36 perturbation theory (vol 131, 094106, 2009). J Chem Phys 2009, 131 (12).
37 68. Izvekov, S.; Parrinello, M.; Burnham, C. J.; Voth, G. A., Effective force fields for condensed phase systems from ab
38 initio molecular dynamics simulation: A new method for force-matching. J Chem Phys 2004, 120 (23), 10896-10913.
39 69. Yamaguchi, T.; Hidaka, K.; Soper, A. K., The structure of liquid methanol revisited: a neutron diffraction experiment
40 at −80 °C and +25 °C. Mol. Phys. 1999, 96 (8), 1159- 1168.
41 70. Woon, D. E., Accurate Modeling of Intermolecular Forces - a Systematic Moller-Plesset Study of the Argon Dimer
42 Using Correlation Consistent Basis-Sets. Chem Phys Lett 1993, 204 (1-2), 29-35.
43 71. Pitoňák, M.; Neogrády, P.; Černý, J.; Grimme, S.; Hobza, P., Scaled MP3 Non-Covalent Interaction Energies Agree
44 Closely with Accurate CCSD(T) Benchmark Data. ChemPhysChem 2009, 10 (1), 282-289.
45 72. Řezáč, J.; Hobza, P., Describing Noncovalent Interactions beyond the Common Approximations: How Accurate Is the
46 “Gold Standard,” CCSD(T) at the Complete Basis Set Limit? J Chem Theory Comput 2013, 9 (5), 2151-2155.
47 73. Yamaguchi, T.; Hidaka, K.; Soper, A. K., The structure of liquid methanol revisited: A neutron diffraction experiment
48 at -80 degrees C and +25 degrees C. Mol Phys 1999, 96 (8), 1159-1168.
49
50
51
52
53
54
55
56
57
58
59
60
15
ACS Paragon Plus Environment
 Journal of Chemical Theory and Computation Page 16 of 24

1
Table 1. The electrostatic potential energy ( EESP ) for methanol monomer from MP2 calculations using different
2
basis sets and from the corresponding fitted multipole parameters, with the corresponding root mean square
3
4
deviation and relative error.
5
6 ESP from
ESP from
7 fitted RMSD Relative Error
Basis MP2
8 multipole (kcal/mol) (%)
(kcal/mol)
9 (kcal/mol)
10 6-31G(d, p) 4.878 4.865 0.165 0.267
11 6-31+G* 5.675 5.649 0.172 0.458
12 5.046
6-31G* 5.059 0.170 0.257
13
14 6-311G(d, p) 5.947 5.935 0.172 0.202
15 6-311G(2df, 2pd) 4.462 4.447 0.154 0.336
16 6-311G* 5.187 5.172 0.156 0.289
17
6-311+G* 5.648 5.626 0.167 0.390
18
19 6-311++G** 5.307 5.282 0.165 0.471
20 6-311+G** 5.309 5.285 0.166 0.452
21 4.729
Aug-CC-pvDz 4.758 0.177 0.609
22
23 Aug-CC-pvTz 4.748 4.723 0.165 0.527
24 Aug-CC-pvQz 4.745 4.720 0.165 0.527
25
26
27
28
29
Table 2. Range of the 44 independent electrostatic parameters over which the ML/GA was performed.
30
31
32
Atom types
Parameters
33 O H (-O) C H (-C)
34 Monopole (q) -1.0 ~ 0 0 ~ 0.4 0 ~ 0.3 0 ~ 0.06
35 Dipole
(0 ~ 0.5, 0.0, 0 ~ 0.4) (−0.14 ~ 0, 0.0, −0.4 ~ 0) (−0.4 ~ 0, 0.0, 0 ~ 0.9) (−0.1 ~ 0, 0.0, −0.1 ~ 0)
36 (µx,µy,µz)
37 Quadrupole
38    0 ~ 0.6 *   0 ~ 0.28 *   0 ~ 0.06 *   0 ~ 0.15 * 
39
EESP  Qxx * *  
*
 
*
 
*
 
*

 0.0 −0.8 ~ 0 *   0.0 0 ~ 0.06 *   0.0 −0.11 ~ 0 *   0.0 −0.17 ~ 0 * 
 Qyx Qyy *   0 ~ 0.06  −0.08 ~ 0 * 
40 0.0 *   −0.2 ~ 0 0.0 *   −0.6 ~ 0 0.0 *  0.0
     
41  Qzx Qzy * 

42
Polarizability (α) 0.5 ~ 1.0 0.2 ~ 0.7 1.0 ~ 1.6 0.2 ~ 0.7
43
44 Thole’s factor (a) 0.3 ~ 0.5 0.3 ~ 0.5 0.3 ~ 0.5 0.3 ~ 0.5
45
46
47
48
49 Table 3. Number distribution of the extracted clusters (9 molecules, 11 molecules and 13 molecules) from different
50 liquid MD systems.
51
52 MD simulations 9 methanol 11 methanol 13 methanol
53 ρ = 0.933 g/ml, T = 193.15 K, p = 1.000 atm 372 157 94
54 ρ = 0.794 g/ml, T = 293.15 K, p = 1.200 atm 502 N/A N/A
55 ρ = 0.627 g/ml, T = 393.15 K, p = 6.293 atm 125 N/A N/A
56
57
58
59
60
16
ACS Paragon Plus Environment
Page 17 of 24 Journal of Chemical Theory and Computation

1
Table 4. Range of the 11 independent vdW parameter over which the ML/GA optimization were performed.
2
3
4 Atom types
Parameters
5 O H (-O) C H (-C)
6 Minimum energy
3.5 ~ 4.0 2.5 ~ 3.0 2.2 ~ 2.8 3.5 ~ 4.0
7 depth (εmin)
8 Minimum energy
0.12 ~ 0.15 0.001 ~ 0.0015 0.4 ~ 0.6 0.001 ~ 0.009
9 EvdW distance (Rmin)
10 H reduction
11
N/A 0.9 ~ 0.95 N/A 0.9 ~ 0.95
factor (λ)
12 Energy offset (δ) 0.05 ~ 8.0
13
14
15
16
17 Table 5. ML/GA generated the electrostatic parameters.
18
19 Atom types
Parameters
O H (-O) C H (-C)
20
21 Monopole (q) -0.45883 0.21357 0.15811 0.02905
22 Dipole
(0.23922, 0.0, 0.16079) (-0.06982, 0.0, -0.20347) (-0.19328, 0.0, 0.45462) (-0.04887, 0.0, -0.04808)
23 (µx,µy,µz)
Quadrupole
24    0.31570 0.0 0.03229   0.14001 0.0 -0.10133   0.02941 0.0 -0.30849   0.07423 -0.04155 
25  Qxx * *        
0.0

26 EESP  Qyx Qyy *   0.0 -0.40859 0.0   0.0 0.03002 0.0   0.0 -0.05306 0.0   0.0 - 0.08433 0.0 
   0.03229
 0.0 0.09289   -0.10133
 0.0 -0.17003   -0.30849
 0.0 0.02365   - 0.04155 0.0 0.01010 
27  Qzx Qzy * 

28
Polarizability
29 (α)
0.986
0.852 0.573
0.438 1.240
1.351 0.573
0.438
30 Thole’s factor
0.3908
0.390 0.3908
0.390 0.3908
0.390 0.3908
0.390
31 (a)
32
33
34
35
Table 6. Root mean square deviation (RMSD) of electrostatic potential between different force field models and
36
37 the QM calculations at the MP2/6-311G(d,p) level. The force fields include the new optimized electrostatic
38 parameters and the original AMOEBA force field (amoeba09.prm). The methanol clusters include 4943 dimers
39 and 502 nonamers (9 molecules cluster).
40
41 RMSD of Eelec Optimized force field
AMOEBA
42 (kcal/mol) model
43 4,943 dimers 0.723 0.305
44 502 nonamers 0.718 0.495
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
17
ACS Paragon Plus Environment
 Journal of Chemical Theory and Computation Page 18 of 24

1
Table 7. ML/GA generated van der Waals parameters.
2
3
4 vdW
MP2/6-31G(d,p) MP2/jul-cc-pVDZ MP2/jul-cc-pVTZ CCSD(T)*
5 parameters
Amoeba09
6 Atom
δ=0 δ≠0 δ=0 δ≠0 δ=0 δ≠0 δ=0
7 Type
8 εmin 3.4050 3.7132 3.4745 3.6377 3.4687 3.5617 3.4372 3.6165
O
9 Rmin 0.1100 0.1274 0.1484 0.1505 0.1483 0.2037 0.1484 0.1026
10 εmin 2.6550 3.1559 2.1463 2.4495 2.3580 2.9640 2.2861 2.0481
11 HO Rmin 0.0135 0.0026 0.0100 0.0045 0.0100 0.0026 0.0103 0.0100
12 λ 0.910 0.9131 0.9049 0.9198 0.9001 0.9399 0.9014 0.9090
13
14
εmin 3.7600 2.8536 3.9092 3.1137 3.8046 3.4887 3.8005 3.8744
C
15 Rmin 0.1010 0.2803 0.1018 0.1373 0.1185 0.1920 0.1198 0.1051
16 εmin 2.8700 3.8178 2.9984 3.3133 2.9583 3.3784 2.7496 2.0066
17 HC Rmin 0.0240 0.0048 0.0200 0.0118 0.0215 0.0068 0.0212 0.0203
18 λ 0.910 0.9120 0.9333 0.9077 0.9085 0.9312 0.9490 0.9357
19 εmin is the minimum energy depth; Rmin is the minimum energy distance; λ is the hydrogen reduction factor
20 *Parameters optimized only from QM data on methanol dimer
21
22
23
24 Table 8. Density (ρ) and heat of vaporization (∆
∆Hvap) of liquid methanol calculated from MD simulations (T = 25
25
°C and p = 1 atm) using different optimized polarizable force fields. The optimized vdW parameters were
26
27 determined from according to the Eq (1) without δ and with δ offset. For comparison, the results from the
28 AMOEBA force field (amoeba09.prm) are given. The MD results were averaged from 3 simulations of 2.5 ns.
29
30 MD
31 DFMP2(fc)/jul- DFMP2(fc)/jul-
Property Exp. MP2/6-31G(d,p)
32 cc-pVDZ cc-pVTZ AMOEBA
33 δ=0 δ≠0 δ=0 δ≠0 δ=0 δ≠0
34 ρ (g/ml) 0.786 0.401 0.759 0.568 0.763 0.686 0.781 0.774
35 ∆Hvap
36 8.95 6.54 9.44 7.23 9.43 8.58 8.98 9.17
(kcal/mol)
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
18
ACS Paragon Plus Environment
Page 19 of 24 Journal of Chemical Theory and Computation

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46 Figure 1. Flowchart of the machine learning with genetic algorithm ( ML/GA) strategy describing the sequence of steps
47 employed in this work.
48
49
50
51
52
53
54
55
56
57
58
59
60
19
ACS Paragon Plus Environment
 Journal of Chemical Theory and Computation Page 20 of 24

1
2
3
4 (a) (b)
5
6
7
8
9
10
11
12
13
14
15
16
17
(c)
18
19
20
21
22
23
24
25
26
27
28
29 Figure 2. Illustration of the methanol cluster configurations used for the force field parameterization. Typical
30 configurations of the (a) 9 molecules, (b) 11 molecules, and (c) 13 molecules of methanol are shown (the H atom is
31 displayed in white, C atom in cyan and O atom in red). The interaction energy between the central molecule (circled by a
32 dashed line) and the surrounding molecules is used in the optimization of the vdw parameters. (d) Distribution of
33 interaction energies from MP2/6-31G(d,p) of 1,250 methanol clusters (999 clusters of 9 molecules, 157 clusters of 11
34
molecules, and 94 clusters of 13 molecules).
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
20
ACS Paragon Plus Environment
Page 21 of 24 Journal of Chemical Theory and Computation

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
Figure 3. Comparison of the interaction energy ( ∆E ) for 1,250 methanol clusters, including 999 clusters of 9 molecules,
33
157 clusters of 11 molecules, and 94 clusters of 13 molecules, computed from (a) MP2/6-31G(d,p) (b) DFMP2(fc)/jul-
34
35 cc-pVDZ, and (c) DFMP2(fc)/jul-cc-pVTZ and the optimized force field model fitted without the offset parameter δ in
36 Eq. (2). (d), (e) and (f) are the optimized force field model fitted with the offset parameter δ in Eq. (2).
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
21
ACS Paragon Plus Environment
 Journal of Chemical Theory and Computation Page 22 of 24

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35 Figure 4. Liquid structure of methanol. The radial distribution function of methanol determined from neutron scattering69
36 compares favorably with the present force field model.73 The O-H hydrogen bonding peak is, however, higher than the
37 experiment. This may reflect the neglect of quantum effects of the nuclei in the simulations,25 as well as inaccuracies in
38 the experimentally determined pair correlation functions.
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
22
ACS Paragon Plus Environment
Page 23 of 24 Journal of Chemical Theory and Computation

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33 Figure 5. Density (ρ) and heat of vaporization (∆Hvap) of methanol as a function of temperature from MD simulations
34 using the optimized polizable force field optimized from the DFMP2(fc)/jul-cc-pVDZ and DFMP2(fc)/jul-cc-pVTZ data.
35 For comparison the results from experiments and the original AMOEBA force field (amoeba09.prm) are also shown.
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
23
ACS Paragon Plus Environment
 Journal of Chemical Theory and Computation Page 24 of 24

1
TOC figure
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
24
ACS Paragon Plus Environment