Ab initio potential-energy surfaces for complex, multichannel systems using modified novelty sampling and feedforward neural networks

A neural network/trajectory approach is presented for the development of accurate potential-energy hypersurfaces that can be utilized to conduct ab initio molecular dynamics (AIMD) and Monte Carlo studies of gas-phase chemical reactions, nanometric cutting, and nanotribology, and of a variety of mechanical properties of importance in potential microelectromechanical systems applications. The method is sufficiently robust that it can be applied to a wide range of polyatomic systems. The overall method integrates ab initio electronic structure calculations with importance sampling techniques that permit the critical regions of configuration space to be determined. The computed ab initio energies and gradients are then accurately interpolated using neural networks (NN) rather than arbitrary parametrized analytical functional forms, moving interpolation or least-squares methods. The sampling method involves a tight integration of molecular dynamics calculations with neural networks that employ early stopping and regularization procedures to improve network performance and test for convergence. The procedure can be initiated using an empirical potential surface or direct dynamics. The accuracy and interpolation power of the method has been tested for two cases, the global potential surface for vinyl bromide undergoing unimolecular decomposition via four different reaction channels and nanometric cutting of silicon. The results show that the sampling methods permit the important regions of configuration space to be easily and rapidly identified, that convergence of the NN fit to the ab initio electronic structure database can be easily monitored, and that the interpolation accuracy of the NN fits is excellent, even for systems involving five atoms or more. The method permits a substantial computational speed and accuracy advantage over existing methods, is robust, and relatively easy to implement.     

Theoretical investigation of the dissociation dynamics of vibrationally excited vinyl bromide on an ab initio potential-energy surface obtained using modified novelty sampling and feed-forward neural networks

The reaction dynamics of vibrationally excited vinyl bromide have been investigated using classical trajectory methods on a neural network potential surface that is fitted to an ab initio database of 12 122 configuration energies obtained from electronic structure calculations conducted at the MP4(SDQ) level of theory using a 6-31G(d,p) basis set for the carbon and hydrogen atoms and Huzinaga's (43334334) basis set augmented with split outer s and p orbitals (4332143214) and a polarization f orbital with an exponent of 0.5 for the bromine atom. The sampling of the 12-dimensional configuration hyperspace of vinyl bromide prior to execution of the electronic structure calculations is accomplished by combining novelty-sampling methods, chemical intuition, and trajectory sampling on empirical and neural network surfaces. The final potential is obtained using a two-layer feed-forward neural network comprising 38 and 1 neurons, respectively, with hyperbolic tangent sigmoid and linear transfer functions in the hidden and output layers, respectively. The fitting is accomplished using the Levenberg-Marquardt algorithm with early stopping and Bayesian regularization methods to avoid overfitting. The interpolated potentials have a standard deviation from the ab initio results of 0.0578 eV, which is within the range generally regarded as "chemical accuracy" for the purposes of electronic structure calculations. It is shown that the potential surface may be easily and conveniently transferred from one research group to another. The files required for transfer of the vinyl bromide surface can be obtained from the Electronic Physics Auxiliary Publication Service. Total dissociation rate coefficients for vinyl bromide are obtained at five different excitation energies between 4.50 and 6.44 eV. Branching ratios into each of the six open reaction channels are computed at 24 vibrational energies in the range between 4.00 and 6.44 eV. The distribution of vibrational energies in HBr formed via three-center dissociation from vinyl bromide is determined and compared with previous theoretical and experimental results. It is concluded that the combination of ab initio electronic structure calculations, novelty sampling with chemical intuition and trajectories on empirical analytic surfaces, and feed-forward neural networks provides a viable framework in which to execute purely ab initio molecular-dynamics studies on complex systems with multiple open reaction channels.     

Ab initio molecular dynamics using hybrid density functionals

Ab initio molecular dynamics simulations with hybrid density functionals have so far found little application due to their computational cost. In this work, an implementation of the Hartree-Fock exchange is presented that is specifically targeted at ab initio molecular dynamics simulations of medium sized systems. We demonstrate that our implementation, which is available as part of the CP2K/Quickstep program, is robust and efficient. Several prescreening techniques lead to a linear scaling cost for integral evaluation and storage. Integral compression techniques allow for in-core calculations on systems containing several thousand basis functions. The massively parallel implementation respects integral symmetry and scales up to hundreds of CPUs using a dynamic load balancing scheme. A time-reversible multiple time step scheme, exploiting the difference in computational efficiency between hybrid and local functionals, brings further time savings. With extensive simulations of liquid water, we demonstrate the ability to perform, for several tens of picoseconds, ab initio molecular dynamics based on hybrid functionals of systems in the condensed phase containing a few thousand Gaussian basis functions.     

Ab initio molecular dynamics of protonated dialanine and comparison to infrared multiphoton dissociation experiments

Finite temperature Car-Parrinello molecular dynamics simulations are performed for the protonated dialanine peptide in vacuo, in relation to infrared multiphoton dissociation experiments. The simulations emphasize the flexibility of the different torsional angles at room temperature and the dynamical exchange between different conformers which were previously identified as stable at 0 K. A proton transfer occurring spontaneously at the N-terminal side is also observed and characterized. The theoretical infrared absorption spectrum is computed from the dipole time correlation function, and, in contrast to traditional static electronic structure calculations, it accounts directly for anharmonic and finite temperature effects. The comparison to the experimental infrared multiphoton dissociation spectrum turns out very good in terms of both band positions and band shapes. It does help the identification of a predominant conformer and the attribution of the different bands. The synergy shown between the experimental and theoretical approaches opens the door to the study of the vibrational properties of complex and floppy biomolecules in the gas phase at finite temperature.     

Ab initio mass tensor molecular dynamics

Mass tensor molecular dynamics method was first introduced by Bennett [J. Comput. Phys. 19, 267 (1975)] for efficient sampling of phase space through the use of generalized atomic masses. Here, we show how to apply this method to ab initio molecular dynamics simulations with minimal computational overhead. Test calculations on liquid water show a threefold reduction in computational effort without making the fixed geometry approximation. We also present a simple recipe for estimating the optimal atomic masses using only the first derivatives of the potential energy.     

Theoretical investigation of the dissociation dynamics of vibrationally excited vinyl bromide on an ab initio potential-energy surface obtained using modified novelty sampling and feedforward neural networks. II. Numerical application of the method

A previously reported method for conducting molecular dynamics simulations of gas-phase chemical dynamics on ab initio potential-energy surfaces using modified novelty sampling and feedforward neural networks is applied to the investigation of the unimolecular dissociation of vinyl bromide. The neural network is fitted to a database comprising the MP4(SDQ) energies computed for 71 969 nuclear configurations using an extended basis set. Dissociation rate coefficients and branching ratios at an internal excitation energy of 6.44 eV for all six open reaction channels are reported. The distribution of vibrational energy in HBr formed in three-center dissociation is computed and found to be in excellent accord with experimental measurements. Computational requirements for the electronic structure calculations, neural network training, and trajectory calculations are given. The weight and bias matrices required for implementation of the neural network potential are made available through the Supplementary Material.     

Ab initio folding of helix bundle proteins using molecular dynamics simulations

We have demonstrated that ab initio fast folding simulations at 400 K using a GB implicit solvent model with an all-atom based force field can describe the spontaneous formation of nativelike structures for the 36-residue villin headpiece and the 46-residue fragment B of Staphylococcal protein A. An implicit solvent model combined with high-temperature MD makes it possible to perform direct folding simulations of small- to medium-sized proteins by reducing the computational requirements tremendously. In the early stage of folding of the villin headpiece and protein A, initial hydrophobic collapse and rapid formation of helices were found to play important roles. For protein A, the third helix forms first in the early stage of folding and exhibits higher stability. The free energy profiles calculated from the folding simulations suggested that both of the helix-bundle proteins show a two-state thermodynamic behavior and protein A exhibits rather broad native basins.     

Ab initio quantum dynamics using coupled-cluster

The curse of dimensionality (COD) limits the current state-of-the-art ab initio propagation methods for non-relativistic quantum mechanics to relatively few particles. For stationary structure calculations, the coupled-cluster (CC) method overcomes the COD in the sense that the method scales polynomially with the number of particles while still being size-consistent and extensive. We generalize the CC method to the time domain while allowing the single-particle functions to vary in an adaptive fashion as well, thereby creating a highly flexible, polynomially scaling approximation to the time-dependent Schro?dinger equation. The method inherits size-consistency and extensivity from the CC method. The method is dubbed orbital-adaptive time-dependent coupled-cluster, and is a hierarchy of approximations to the now standard multi-configurational time-dependent Hartree method for fermions. A numerical experiment is also given.     

Ab initio molecular dynamics study of liquid methanol

We present a density-functional theory based molecular-dynamics study of the structural, dynamical, and electronic properties of liquid methanol under ambient conditions. The calculated radial distribution functions involving the oxygen and hydroxyl hydrogen show a pronounced hydrogen bonding and compare well with recent neutron diffraction data, except for an underestimate of the oxygen–oxygen correlation. We observe that, in line with infrared spectroscopic data, the hydroxyl stretching mode is significantly red-shifted in the liquid. A substantial enhancement of the dipole moment is accompanied by significant fluctuations due to thermal motion. Our results provide valuable data for improvement of empirical potentials.     

Ab initio molecular dynamics simulation of ionic liquids

Ab initio Car-Parinnello molecular dynamics is used to simulate the structure and the dynamics of 1-butyl-3-methylimidazolium iodide ([bmim]I) ionic liquid at 300 K. Site-site pair correlation functions reveal that the anion has a strong interaction with any three C-H's of the imidazolium ring. The ring bends over and wraps around the anion such that the two nitrogen atoms take a distance to the anion. Electron donating butyl group contributes the electronic polarization in addition to geometrical (out-of-plane) polarization of the ring due to the liquid environment. This facilitates bending of the ring along the axis passing through nitrogen atoms. The average bending angle depends largely on the alkyl chain length and slightly on the anion type. Redistribution of electron density over the ring caused by the electron donating alkyl group provides additional independent evidence to the instability of lattice structure, hence the low melting point of the ionic liquid. Simulated viscosity and diffusion coefficients of [bmim]I are in quite agreement with the experiments.     

Ab initio molecular dynamics calculations to study catalysis

The modern versions of the density functional theory (DFT), especially those using the generalized gradient approximation (GGA), have reached (almost) chemical accuracy and thus can be applied to study problems of real chemical interest such as catalysis. The important equations for the DFT, the local density approximation (LDA), and GGA are given. The full-potential linearized augmented plane wave method (LAPW) is used to check the accuracy of GGA in solids. The basic concepts of the ab initio molecular dynamics (MD) method by Car and Parrinello and its recent implementation using the projector augmented Wave (PAW) method which use a similar augmentation as LAPW are described. PAW applications to ferrocene and beryllocene are summarized, which illustrate that vibrational or fluxional behavior are well described. Sodalite, an aluminosilicate, is discussed as a generic zeolite in comparison with gmelinite. A study of the dynamics of such a system allows the determination of, e.g., the proton stretch vibrations which can be related to infrared spectra. This is illustrated for the OH stretch vibration of the acid site in silicon-rich sodalite. With this methodology, we are able to study the interaction of methanol trapped inside the cage structure of silicon-rich sodalite and to gain new insight into crucial steps of catalytic reactions, namely, the hydrogen-bonding and the possible protonation in this system, or a proton-exchange reaction. The strategies for parallelizing the PAW code are outlined. © 1997 John Wiley & Sons, Inc.     

Ab initio molecular dynamics of liquid 1,3-dimethylimidazolium chloride

Density-functional-based Car-Parrinello molecular dynamics (CPMD) simulations have been performed for the ionic liquid 1,3-dimethylimidazolium chloride, [dmim]Cl, at 438 K. The local structure of the liquid is described in terms of various partial radial distribution functions and anisotropic spatial distributions, which reveal a significant extent of hydrogen bonding. The cation-anion distribution simulated with the BP86 functional is in qualitative agreement with the structural model derived from neutron diffraction data for the liquid, whereas the theoretical cation-cation distribution shows less satisfactory accord. Population analyses indicate noticeable charge transfer from anions to cations, and specific CH...Cl hydrogen bonds are characterized in terms of donor-acceptor interactions between lone pairs on Cl and antibonding sigma(CH) orbitals.     

Ab initio molecular dynamics study of formate ion hydration

We perform ab initio molecular dynamics simulations of the aqueous formate ion. The mean number of water molecules in the first solvation shell, or the hydration number, of each formate oxygen is found to be consistent with recent experiments. Our ab initio pair correlation functions, however, differ significantly from many classical force field results and hybrid quantum mechanics/molecular mechanics predictions. They yield roughly one less hydrogen bond between each formate oxygen and water than force field or hybrid methods predict. Both the BLYP and PW91 exchange correlation functionals give qualitatively similar results. The time dependence of the hydration numbers are examined, and Wannier function techniques are used to analyze electronic configurations along the molecular dynamics trajectory.     

Ab initio molecular dynamics of heme in cytochrome c

Ab initio molecular dynamics (AIMD) calculations, based on the Car-Parrinello method, have been carried out for three models of heme c that is present in cytochrome c. Both the reduced (Fe(II)) and oxidized (Fe(III)) forms have been analyzed. The simplest models (1R and 1O, respectively) consist of a unsubstituted porphyrin (with no side chains) and two axially coordinated imidazole and ethylmethylthioether ligands. Density functional theory optimizations of these models confirm the basic electronic features and are the starting point for building more complex derivatives. AIMD simulations were performed after reaching the thermal stability at T = 300 K. The evolution of the Fe-L(ax) bond strengths is examined together with the relative rotations of the imidazole and methionine about the axial vector, which appear rather independent from each other. The next models (2R and 2O) contain side chains at the heme to better simulate the actual active site. It is observed that two adjacent propionate groups induce some important effects. The axial Fe-Sdelta bond is only weakened in 2R but is definitely cleaved in the oxidized species 2O. Also the mobility of the Im ligand seems to be reduced by the formation of a strong hydrogen bond that involves the Im Ndelta1-Hdelta1 bond and one carboxylate group. In 2O the interaction becomes so strong that a proton transfer occurs and the propionic acid is formed. Finally, the models 3 include a free N-methyl-acetamide molecule to mimic a portion of the protein backbone. This influences the orientation of carboxylate groups and limits the amount of their hydrogen bonding with the Im ligand. Residual electrostatic interactions are maintained, which are still able to modulate the dissociation of the methionine from the heme.     

Ab initio molecular dynamics with dual basis set methods

On-the-fly, ab initio classical molecular dynamics are demonstrated with an underlying dual basis set potential energy surface. Dual-basis self-consistent field (Hartree-Fock and density functional theory) and resolution-of-the-identity second-order M?ller-Plesset perturbation theory (RI-MP2) dynamics are tested for small systems, including the water dimer. The resulting dynamics are shown to be faithful representations of their single-basis analogues for individual trajectories, as well as vibrational spectra. Computational cost savings of 58% are demonstrated for SCF methods, even relative to Fock-extrapolated dynamics, and savings are further increased to 71% with RI-MP2. Notably, these timings outperform an idealized estimate of extended-Lagrangian molecular dynamics. The method is subsequently demonstrated on the vibrational absorption spectrum of two NO(+)(H?O)? isomers and is shown to recover the significant width of the shared-proton bands observed experimentally.     

Ab initio molecular dynamics study of methanol adsorption on copper clusters

The preferential structures of small copper clusters Cun (n=2-9) and the adsorption of methanol molecules on these clusters are examined with first principles, molecular dynamics simulations. The results show that the copper clusters undergo systematic changes in bond length and bond order associated with altering their preferential structures from one-dimensional structures, to two-dimensional and three-dimensional structures. The results also indicate that low coordination number sites on the copper clusters are both the most favorable for methanol adsorption and have the greatest localization of electronic charge. The simulations predict that charge transfer between the neutral copper clusters and the incident methanol molecules is a key process by which adsorption is stabilized. Importantly, the changes in the dimensionality of the copper clusters do not significantly influence methanol adsorption.     

Ab initio studies of a water layer at transition metal surfaces

This paper presents a detailed study of a water adlayer adsorbed on Pt(111) and Rh(111) surfaces using periodic density functional theory methods. The interaction between the metal surface and the water molecules is assessed from molecular dynamics simulation data and single point electronic structure calculations of selected configurations. It is argued that the electron bands around the Fermi level of the metal substrate extend over the water adlayer. As a consequence in the presence of the water layer the surface as a whole still maintains its metallic conductivity-a result of a crucial importance for understanding the process of electron transfer through the water/metal interface and electrochemical reactions in particular. Our results also indicate that there exists a weak bond between the hydrogen of the water and the Rh metal atoms as opposed to the widespread (classical) models based on purely repulsive interaction. This suggests that the commonly used classical interactions potentials adopted for large scale molecular dynamics simulations of water/metal interfaces may need revision. Two adsorption models of water on transition metals with the OH bonds pointing towards or away of the surface are also examined. It is shown that due to the very close values of their adsorption energies one should consider the real structure of water on the surface as a mixture of these simple "up" and "down" models. A model for the structure of the adsorbed water layer on Rh(111) is proposed in terms of statistical averages from molecular dynamics simulations.     

Ab initio static and molecular dynamics study of 4-styrylpyridine.

We report an in‐depth theoretical study of 4‐styrylpyridine in its singlet S₀ ground state. The geometries and the relative stabilities of the trans and cis isomers were investigated within density functional theory (DFT) as well as within Hartree–Fock (HF), second‐order Møller–Plesset (MP2), and coupled cluster (CC) theories. The DFT calculations were performed using the B3LYP and PBE functionals, with basis sets of different qualities, and gave results that are very consistent with each other. The molecular structure is thus predicted to be planar at the energy minimum, which is associated with the trans conformation, and to become markedly twisted at the minimum of higher energy, which is associated with the cis conformation. The results of the calculations performed with the post‐HF methods approach those obtained with the DFT methods, provided that the level of treatment of the electronic correlation is high enough and that sufficiently flexible basis sets are used. Calculations carried out within DFT also allowed the determination of the geometry and the energy of the molecule at the biradicaloid transition state associated with the thermal cis?trans isomerization and at the transition states associated with the enantiomerization of the cis isomer and with the rotations of the pyridinyl and phenyl groups in the trans and cis isomers. Car–Parrinello molecular dynamics simulations were also performed at 50, 150, and 300 K using the PBE functional. The studies allowed us to evidence the highly flexible nature of the molecule in both conformations. In particular, the trans isomer was found to exist mainly in a nonplanar form at finite temperatures, while the rotation of the pyridinyl ring in the cis isomer was incidentally observed to take place within ≈1 ps during the simulation carried out at 150 K on this isomer.