betaFeedback

WorksPhysics and Astronomy

Polarizable Embedding QM/MM for Periodic Systems

Abstract

A general polarizable embedded (PE) quantum mechanics/molecular mechanics scheme for periodic systems is presented, describing mutual polarization of the two subsystems. The QM system, described with density functional theory (DFT), is coupled to a single center multipole expansion (SCME) model, characterising H2_2O molecules in the MM region. In SCME the H2_2O molecules are ascribed anisotropic dipole and quadrupole polarizabilities and permanent multipoles up to and including the hexadecapole. Our embedding scheme illustrates a smooth and efficient convergence pattern of the periodic interaction potential by introducing a single and clustered multipole expansion points in the far-field. By choosing the near- and far-field expansion of the potential carefully the PE-QM/MM calculation matches the level of accuracy of a the QM calculation. In the short range, the electrostatic interaction between the QM and MM subsystems is damped with a real-space and pair-wise isotropic damping functions - resulting in a screened interaction and preventing over-polarization. In molecular dynamics simulations the two subsystems are separated with the elastic scattering assisted flexible inner region [Kirchhoff et. al. JCTC, 2021, 17, 9, 5863] - ensuring a smooth transition in the radial distribution at the boundary between the two subsystems.

main.pdf

Introduction

Modelling electrocatalytic reactions at the interface between solid and liquid requires capturing both the quantum-mechanical processes at the electrode, such as bond breaking and formation, and charge transfer, as well as the complex and fluctuating response of the surrounding electrolyte, which includes long-range electrostatic interaction with e.g., mobile ions, and the polarization response of the liquid. This presents a considerable challenge, and, therefore, different aspects of the electrocatalytic process are often modeled independently with established methodological frameworks [1, 2, 3] (for a recent review see [4]).

In most cases the workhorse is Kohn-Sham density functional theory [5, 6] (KS-DFT). Standard methods based on DFT, while powerful, must be extended or coupled with additional frameworks to account for the rich complexity of electrocatalytic systems. While ab initio molecular dynamics (AIMD) simulations have provided valuable insight into the fluctuation of the complex environment they are computationally demanding, particularly when the goal is to obtain a quantitative estimate of solvation energy, and the effect of applied voltage and charge transfer at the electrode-electrolyte interface [7, 8, 9]. AIMD simulations are typically limited to a timescale of only a few picoseconds, whereas sampling over nanoseconds is often necessary to achieve proper equilibration and convergence of statistical averaging [10, 11].

To overcome these challenges, the hybrid quantum mechanics/molecular mechanics (QM/MM) [12] strategy is a promising approach. In a QM/MM simulation, the system is partitioned into subsystems. The chemically active QM subsystem, where bond rearrangements and electronic effects take place, is treated using an electronic structure method. The remaining part is the MM subsystem and is described using a potential energy function depending only on atomic coordinates. This partitioning scheme dramatically reduces computational effort for large systems while retaining the accuracy needed to describe reactive events at the catalytic site. QM/MM can be a powerful tool for simulating electrocatalytic reactions in a realistic way.

QM/MM approaches are categorized by how the QM and MM subsystems are coupled: mechanical embedding (ME), electrostatic embedding (EE), and polarizable embedding (PE). In ME, there is no explicit QM/MM interaction, so long-range electrostatic effects are absent. In EE [13–15], the MM subsystem is represented by fixed point charges, allowing the QM electronic density to respond to the electrostatic field of the MM region, partially capturing electrolyte effects on interfacial reactions. EE-QM/MM has been applied to estimate the free energy of water at the Pt(111)/water interface [16] and to study the hydrogen evolution reaction (HER) on MoS2_2 electrodes [17], demonstrating its practicality for long-timescale dynamics otherwise inaccessible to AIMD. However, due to rapidly fluctuating near-surface electric fields and the electrical double layer extending well into the bulk liquid, non-mutually responsive EE-QM/MM models are inadequate for describing the solid/aqueous interface.

The PE-QM/MM approach overcomes the limitations of ME and EE by explicitly accounting for polarization response in both the QM and MM subsystems. The QM charge distribution induces dipoles, and possibly higher order moments, in the MM subsystem, and these in turn contribute to the Hamiltonian of the QM subsystem as an external potential, resulting in a self-consistent polarization field. Consequently, the energy of the system is a functional of the total polarization field. Sophisticated PE-QM/MM implementations including electrical dipoles have been developed in the context of electronic excitations of solvated molecules [18–26]. Higher order expansion up to the hexadecapole, which has been found to be adequate for typical intermolecular distances in water [27], have also been presented [28, 29].

While the PE-QM/MM approach provides a more rigorous treatment of interfacial polarization than ME or EE, it has so far not been applied extensively to electrocatalytic system. PE-QM/MM frameworks have only recently been extended to solid/liquid interfaces and their implementation in standard simulation packages is still maturing, partly due to the lack of transferability of polarizable force fields. A step in the direction of electrocatalysis is a calculation of the Raman frequencies of some surface-bound intermediates in CO2_2 electroreduction [30].

On another facet of hybrid simulations, machine learning interatomic potentials (MLIPs) have recently emerged as a powerful complement to both AIMD and classical QM/MM approaches for modeling solid/liquid interfaces, offering near-first-principles accuracy at a fraction of the computational cost and enabling access to nanosecond timescales and system sizes that are otherwise inaccessible [31–33]. Architectures such as NequIP [34], MACE [35], and Allegro [36] have been applied to metal/water interfaces, successfully reproducing structural and dynamical properties [37].

In an ML/MM framework, the QM Hamiltonian is replaced by a neural network potential (NNP) embedded in a classical MM environment, dramatically accelerating free-energy sampling while retaining an accurate description of the reactive region [38]. Delta-learning strategies have proven particularly effective for condensed-phase simulations [39], and recent work has extended this paradigm to electrochemical systems by combining field-dependent MLIPs with ML electron-density response models to simulate metal–electrolyte interfaces under applied potentials. However, most MLIPs rely on a locality assumption that fundamentally limits their ability to capture long-range electrostatic interactions essential at charged interfaces, in polar solvents, and for charge-transfer reactions [40, 41]. While frameworks such as Latent Ewald Summation (LES) offer a route to incorporating long-range Coulomb interactions without requiring ambiguous DFT partial charge labels, their extension to heterogeneous interfaces with spatially varying dielectric permittivity remains an open challenge [42, 43], and proposed remedies for describing global charge redistribution at metallic electrodes under applied bias are still maturing [42].

At the QM–MM boundary, the NNP inherits limitations from both the reference method and the embedding scheme, including inconsistent charge partitioning and the absence of mutual polarization under mechanical embedding [44]. Transferability remains a persistent concern, as ML models trained on specific electrode–electrolyte combinations frequently fail when extrapolating to unseen chemical environments or configurations outside the training distribution, producing unphysical energy drifts and trajectory instabilities [45]. Collectively, these shortcomings — the locality assumption, inadequate treatment of long-range electrostatics, and lack of transferable polarizable MM force fields for the electrolyte — further motivate the development of PE-QM/MM frameworks.

Potential energy functions with high-level description of the electrostatics and thereby transferable to different environments have been developed for water and acetonitrile [28, 29, 46–48]. These potential functions are typically parameterized by fitting results of quantum mechanical calculations (such as energy and atomic forces, multipole moments, and polarizabilities) of the solvent molecules at a specified level of theory.

In the following, we describe the polarizable embedding methodology, including a brief overview of the single center multipole expansion model (SCME) [47, 48] and the QM/MM interface implemented in the grid-based projector augmented wave code GPAW [49]. To validate the PE-QM/MM approach, we first analyze 2D periodicity in energy and electrostatic potential by calculating the Coulomb potential of a graphene sheet with a water molecule, comparing the influence of the multipole expansion against a pure QM reference. We then examine the QM/MM interaction energy of 2D periodic ice layers using different grid expansions to assess the precision of the embedding scheme. Next, MD simulations are employed to investigate the QM/MM interaction in a dynamic setting, where an isotropic real-space damping function is introduced at the QM/MM boundary to prevent over-polarization. Gold-water MD simulations demonstrate that the damping value significantly impacts the PE-QM/MM interaction, confirming its necessity. Finally, a graphene-water MD simulation is performed to analyze the solvent distribution along the zz axis relative to a pure QM calculation.

Theory

In Kohn-Sham density functional theory [5, 6] the energy of an electronic system is given by:

EKS=Ts+EC[ρ]+Exc[n]+∫Vext[ρ]dr(1)E^{\mathrm{KS}} = T_s + E_{\mathrm{C}}[\rho] + E_{\mathrm{xc}}[n] + \int V_{\mathrm{ext}}[\rho] \mathrm{d}r \tag*{(1)}

Here, TsT_s is the kinetic energy of the non-interacting electrons whose total density corresponds to the ground-state density of the interacting electrons. EC[ρ]E_{\mathrm{C}}[\rho] is the Coulomb energy of the charge density, ρ\rho. Exc[n]E_{\mathrm{xc}}[n] is the exchange-correlation energy functional of the electron density, nn, a functional form which is not known exactly and must be approximated in practice. It is typical in KS–DFT studies of electrocatalytic systems that a local or semilocal approximation of the exchange-correlation energy is used. This transforms the problem of N interacting electrons to a problem of three spatial variables to describe the total electron density. Therefore KS–DFT provides a feasible way to simulate systems with thousands of electrons. Finally, the VextV_{\mathrm{ext}} term is an external potential and is include in QM/MM to couple the QM subsystem to the MM subsystem.

In QM/MM the total energy of the system is split into the energy of each subsystem (QM and MM) and the explicit interaction between them (QM/MM)

Etot=EQM+EQM/MM+EMM.(2)E^{\mathrm{tot}} = E^{\mathrm{QM}} + E^{\mathrm{QM/MM}} + E^{\mathrm{MM}}. \tag*{(2)}

The functional derivative of the EQM/MME^{\mathrm{QM/MM}} term with respect to the QM charge density defines the external potential in the QM subsystem due to the MM multipoles, and the functional derivative with respect to the MM multipoles defines the potential fields in the MM subsystem due to the QM charge density. An iterative scheme can then be used to solve for the self-consistent polarization field in the total system. The total energy of the system is therefore expressed as a functional of both the QM charge density and multipole moments of the MM system. The resulting energy functional can be expressed as

Etot[ρ,M′]=EQM[ρ]+EQM/MM[ρ,M′]+EMM[M′](3)E^{\mathrm{tot}}[\rho,M^{\prime}] = E^{\mathrm{QM}}[\rho] + E^{\mathrm{QM/MM}}[\rho,M^{\prime}] + E^{\mathrm{MM}}[M^{\prime}] \tag*{(3)}

where M′M^{\prime} are the total moments (permanent plus polarized) of the MM system. The electrostatic coupling be-

tween the QM charge density and the MM environment can be expressed as an interaction energy functional,

Eelst.QM/MM=∫ρ(r)VMM(r) dr=∑iNMM∑s(Msi+ΔMsi)VsiQM(4)E_{\mathrm{elst.}}^{\mathrm{QM/MM}} = \int\rho(\mathbf{r})V^{\mathrm{MM}}(\mathbf{r})\,d\mathbf{r} = \sum_{i}^{N_{\mathrm{MM}}}\sum_s \left(M_s^i+\Delta M_s^i\right)V_{s}^{i\mathrm{QM}} \tag*{(4)}

where ρ(r)\rho(\mathbf{r}) is the charge density of the QM subsystem, VMM(r)V^{\mathrm{MM}}(\mathbf{r}) the electrostatic potential in the QM region due to the MM sites, including contributions from both permanent and polarized moments, VsiQM(r)V_s^{i\mathrm{QM}}(\mathbf{r}) is the potential field of rank ss due to the QM charge density evaluated at MM site ii. MsiM_s^i are the permanent moments and ΔMsi\Delta M_s^i are polarized moments of rank ss, for MM site ii. The MM potential field can be written as a sum over MM sites ii, which in Einstein notation is

VMM(r)=∑i∑sTsri(Msi+ΔMsi)(5)V^{\mathrm{MM}}(\mathbf{r}) = \sum_i \sum_s T_s^{ri}\left(M_s^i+\Delta M_s^i\right) \tag*{(5)}

where TsriT_s^{ri} is the Coulomb interaction tensor of rank ss (see Supporting Information S2) and r\mathbf{r} is a coordinate in the QM region (see Figure 1). The coordinate r\mathbf{r} and center of mass of MM site ii share a common origin, which defines the global reference frame of the total system. The on-site potential field, due to the QM charge density, of rank ss are given by

Schematic of an example 2D periodic QM/MM interface

Figure 1. Schematic of an example 2D periodic QM/MM interface. The QM region is composed of a slab (grey circles) and H2_2O molecules (red colored oxygen). The MM region is composed of H2_2O molecules (blue colored oxygen). The orange squares represent the real space regular mesh grid points, rg\mathbf{r}^{g}, on which the QM system is described. Crossing of H2_2O molecules between the two subsystems during e.g. MD simulations is prevented by the elastic scattering assisted flexible inner region (SAFIRES [50]) partitioning scheme.

VsiQM=∫ρ(r)Tsri dr(6)V_s^{i\mathrm{QM}} = \int\rho(\mathbf{r})T_s^{ri}\,d\mathbf{r} \tag*{(6)}

The EMM[M′]E^{\mathrm{MM}}[M^{\prime}] potential function requires solving self-consistently for ΔM\Delta M, and is in turn similarly affected by the presence of the QM charge density via the EQM/MM[ρ,M′]E^{\mathrm{QM/MM}}[\rho,M^{\prime}] term. The details of this coupling depends on the polarizable potential function.

Single Center Multipole Expansion

In the SCME model the energy is a functional of the polarized moments, ΔMsi\Delta M_s^i, ascribed to the COM of H2_2O molecule ii and is given by

EMM[{ΔMsi}]=Eelst[{ΔMsi}]+Enon−elst(7)E^{\mathrm{MM}}\left[\{\Delta M_s^i\}\right] = E_{\mathrm{elst}}\left[\{\Delta M_s^i\}\right] + E_{\mathrm{non-elst}} \tag*{(7)}

where the terms on the right hand side are, Eelst[{ΔMsi}]E_{\mathrm{elst}}[\{\Delta M_s^i\}], the total electrostatic energy functional and the non-electrostatic terms, Enon−elstE_{\mathrm{non-elst}}, which includes terms such as a pair-wise repulsive and short to intermediate range potential functions. In the case of the flexible variant, SCME/f, a EmonE_{\mathrm{mon}} term is also included which is a sum of the internal energies described by the Patridge-Schwenke [51] potential energy surface (PS–PES) of the water monomer.

The electrostatic term can be split up into three contributions

Eelst[{ΔMsi}]=Eperm[{ΔMsi}]+Epol[{ΔMsi}]+Eself[{ΔMsi}](8)E_{\mathrm{elst}}\left[\{\Delta M_s^i\}\right] = E_{\mathrm{perm}}\left[\{\Delta M_s^i\}\right] + E_{\mathrm{pol}}\left[\{\Delta M_s^i\}\right] + E_{\mathrm{self}}\left[\{\Delta M_s^i\}\right] \tag*{(8)}

accounting for the inter- and intramolecular contribution to the total electrostatic energy in the system. In the SCME model each MM site ii is assigned a permanent dipole up to, and including, a hexadecapole (Mi={μαi,θαβi,Ωαβγi,Φαβγδi}M^{i} = \{\mu^{i}_{\alpha}, \theta^{i}_{\alpha\beta}, \Omega^{i}_{\alpha\beta\gamma}, \Phi^{i}_{\alpha\beta\gamma\delta}\}), as well as a dipole-dipole, dipole-quadrupole, and quadrupole-quadrupole polarizability (resulting in ΔMi={Δμαi,Δθαβi}\Delta M^{i} = \{\Delta\mu^{i}_{\alpha}, \Delta\theta^{i}_{\alpha\beta}\}). At self-consistency the first two terms combine to give

Eperm+pol[{ΔMsi}]=12∑iNMM(μαi+Δμαi)Vαi+(θαβi+Δθαβi)Vαβi+ΩαβγiVαβγi+ΦαβγδiVαβγδi(9)E_{\mathrm{perm+pol}}[\{\Delta M^{i}_{s}\}] = \frac{1}{2}\sum_{i}^{N_{\mathrm{MM}}} \left( \mu^{i}_{\alpha}+\Delta\mu^{i}_{\alpha} \right)V^{i}_{\alpha} + \left(\theta^{i}_{\alpha\beta}+\Delta\theta^{i}_{\alpha\beta}\right)V^{i}_{\alpha\beta} + \Omega^{i}_{\alpha\beta\gamma}V^{i}_{\alpha\beta\gamma} + \Phi^{i}_{\alpha\beta\gamma\delta}V^{i}_{\alpha\beta\gamma\delta} \tag*{(9)}

EselfE_{\mathrm{self}} is the on-site self-energy, given by

Eself[{ΔMαi}]=−12∑iNMM(ΔμαiVαi+13ΔθαβiVαβi)(10)E_{\mathrm{self}}[\{\Delta M^{i}_{\alpha}\}] = -\frac{1}{2}\sum_{i}^{N_{\mathrm{MM}}}\left(\Delta\mu^{i}_{\alpha}V^{i}_{\alpha}+\frac{1}{3}\Delta\theta^{i}_{\alpha\beta}V^{i}_{\alpha\beta}\right) \tag*{(10)}

and accounts for the cost in energy to polarize the molecules. In the SCME model, the Coulomb interaction tensor is damped using a Gaussian-based scheme [52], such that the rank-0 damped Coulomb interaction tensor becomes

T0ij,d=1rijλ0(rij)(11)T^{ij,d}_{0} = \frac{1}{r^{ij}}\lambda_{0}(r^{ij}) \tag*{(11)}

where λs(rij)\lambda_{s}(r_{ij}) is a rank-ss damping function which depends on the distance between the center of mass of the ii and jj MM sites (see Supplementary Information S2). The potential at site ii is given by

Vi=∑j≠iNMM∑sTsij,d(Ms+ΔMs)(12)V^{i} = \sum_{j\ne i}^{N_{\mathrm{MM}}}\sum_{s}T^{ij,d}_{s}(M_{s}+\Delta M_{s}) \tag*{(12)}

and, in the case of SCME with dipole-dipole, α\alpha, dipole-quadrupole, AA, and quadrupole-quadrupole polarizabilities, CC, the induced moments are

Δμαi=ααβiVβi+Aα,βγiVβγi(13)\Delta\mu^{i}_{\alpha} = \alpha^{i}_{\alpha\beta}V^{i}_{\beta}+A^{i}_{\alpha,\beta\gamma}V^{i}_{\beta\gamma} \tag*{(13)}

References

  1. [1]Jónsson, E. Ö.; Dohn, A. O.; Jónsson, H. Polarizable embedding with a transferable H2O potential function I: Formulation and tests on dimer. Journal of Chemical Theory and Computation 2019, 15, 6562–6577.DOI
  2. [2]Blöchl, P. E. Projector augmented-wave method. Phys. Rev. B 1994, 50, 17953–17979.pubmed.ncbi.nlm.nih.gov/9976227
  3. [3]Mortensen, J.; Hansen, L.; Jacobsen, K. W. Real-space grid implementation of the projector augmented wave method. Phys. Rev. B 2005, 71, 035109.DOI
  4. [4]Dohn, A. O.; Jónsson, E. Ö.; Levi, G.; Mortensen, J. J.; Lopez-Acevedo, O.; Thygersen, K. S.; Jacobsen, K. W.; Ulstrup, J.; Henriksen, N. E.; Møller, K. B.; Jónsson, H. Grid-Based Projector Augmented Wave (GPAW) Implementation of Quantum Mechanics/Molecular Mechanics (QM/MM) Electrostatic Embedding and Application to a Solvated Diplatinum Complex. Journal of Chemical Theory and Computation 2017, 13, 6010–6022.DOI
  5. [5]Thole, B. Molecular polarizabilities calculated with a modified dipole interaction. Chem. Phys. 1981, 59, 341 – 350.DOI
  6. [6]Masia, M.; Probst, M.; Rey, R. On the performance of molecular polarization methods. II. Water and carbon tetrachloride close to a cation. J. Chem. Phys 2005, 123, 164505.pubmed.ncbi.nlm.nih.gov/16268710
  7. [7]Masia, M.; Probst, M.; Rey, R. Polarization damping in halide–water dimers. Chemical Physics Letters 2006, 420, 267 – 270.DOI
  8. [8]Burnham, C. J.; Li, J.; Xantheas, S. S.; Leslie, M. The parametrization of a Thole-type all-atom polarizable water model from first principles and its application to the study of water clusters (n=2–21) and the phonon spectrum of ice Ih. J. Chem. Phys 1999, 110, 4566–4581.DOI
  9. [9]Stone, A. J. Electrostatic damping functions and the penetration energy. J. Phys. Chem. A 2011, 115, 7017–7027.DOI
  10. [10]Sala, J.; Gürdia, E.; Masia, M. The polarizable point dipoles method with electrostatic damping: Implementation on a model system. J. Chem. Phys 2010, 133, 234101.DOI
  11. [11]Stone, A. The Theory of Intermolecular Forces; The Theory of Intermolecular Forces; OUP Oxford, 2013.
  12. [12]Arthur, D.; Vassilvitskii, S. k-means++: The Advantages of Careful Seeding. Proceedings of the 18th Annual ACM-SIAM Symposium on Discrete Algorithms (SODA). 2007; pp 1027–1035.

Paper details

vlnc:W_7s7ghe6d
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents