The Experts below are selected from a list of 134913 Experts worldwide ranked by ideXlab platform
Klausrobert Muller - One of the best experts on this subject based on the ideXlab platform.
-
Molecular Force fields with gradient domain machine learning construction and application to dynamics of small molecules with coupled cluster Forces
Journal of Chemical Physics, 2019Co-Authors: Huziel E Sauceda, Stefan Chmiela, Igor Poltavsky, Klausrobert Muller, Alexandre TkatchenkoAbstract:We present the construction of Molecular Force fields for small molecules (less than 25 atoms) using the recently developed symmetrized gradient-domain machine learning (sGDML) approach [Chmiela et al., Nat. Commun. 9, 3887 (2018) and Chmiela et al., Sci. Adv. 3, e1603015 (2017)]. This approach is able to accurately reconstruct complex high-dimensional potential-energy surfaces from just a few 100s of Molecular conformations extracted from ab initio Molecular dynamics trajectories. The data efficiency of the sGDML approach implies that atomic Forces for these conformations can be computed with high-level wavefunction-based approaches, such as the “gold standard” coupled-cluster theory with single, double and perturbative triple excitations [CCSD(T)]. We demonstrate that the flexible nature of the sGDML model recovers local and non-local electronic interactions (e.g., H-bonding, proton transfer, lone pairs, changes in hybridization states, steric repulsion, and n → π* interactions) without imposing any restriction on the nature of interatomic potentials. The analysis of sGDML Molecular dynamics trajectories yields new qualitative insights into dynamics and spectroscopy of small molecules close to spectroscopic accuracy.We present the construction of Molecular Force fields for small molecules (less than 25 atoms) using the recently developed symmetrized gradient-domain machine learning (sGDML) approach [Chmiela et al., Nat. Commun. 9, 3887 (2018) and Chmiela et al., Sci. Adv. 3, e1603015 (2017)]. This approach is able to accurately reconstruct complex high-dimensional potential-energy surfaces from just a few 100s of Molecular conformations extracted from ab initio Molecular dynamics trajectories. The data efficiency of the sGDML approach implies that atomic Forces for these conformations can be computed with high-level wavefunction-based approaches, such as the “gold standard” coupled-cluster theory with single, double and perturbative triple excitations [CCSD(T)]. We demonstrate that the flexible nature of the sGDML model recovers local and non-local electronic interactions (e.g., H-bonding, proton transfer, lone pairs, changes in hybridization states, steric repulsion, and n → π* interactions) without imposing any res...
-
Molecular Force fields with gradient domain machine learning construction and application to dynamics of small molecules with coupled cluster Forces
Journal of Chemical Physics, 2019Co-Authors: Huziel E Sauceda, Stefan Chmiela, Igor Poltavsky, Klausrobert Muller, Alexandre TkatchenkoAbstract:We present the construction of Molecular Force fields for small molecules (less than 25 atoms) using the recently developed symmetrized gradient-domain machine learning (sGDML) approach [Chmiela et al., Nat. Commun. 9, 3887 (2018) and Chmiela et al., Sci. Adv. 3, e1603015 (2017)]. This approach is able to accurately reconstruct complex high-dimensional potential-energy surfaces from just a few 100s of Molecular conformations extracted from ab initio Molecular dynamics trajectories. The data efficiency of the sGDML approach implies that atomic Forces for these conformations can be computed with high-level wavefunction-based approaches, such as the "gold standard" coupled-cluster theory with single, double and perturbative triple excitations [CCSD(T)]. We demonstrate that the flexible nature of the sGDML model recovers local and non-local electronic interactions (e.g., H-bonding, proton transfer, lone pairs, changes in hybridization states, steric repulsion, and n → π* interactions) without imposing any restriction on the nature of interatomic potentials. The analysis of sGDML Molecular dynamics trajectories yields new qualitative insights into dynamics and spectroscopy of small molecules close to spectroscopic accuracy.
-
Molecular Force fields with gradient domain machine learning construction and application to dynamics of small molecules with coupled cluster Forces
arXiv: Chemical Physics, 2019Co-Authors: Huziel E Sauceda, Stefan Chmiela, Igor Poltavsky, Klausrobert Muller, Alexandre TkatchenkoAbstract:We present the construction of Molecular Force fields for small molecules (less than 25 atoms) using the recently developed symmetrized gradient-domain machine learning (sGDML) approach [Chmiela et al., Nat. Commun. 9, 3887 (2018); Sci. Adv. 3, e1603015 (2017)]. This approach is able to accurately reconstruct complex high-dimensional potential-energy surfaces from just a few 100s of Molecular conformations extracted from ab initio Molecular dynamics trajectories. The data efficiency of the sGDML approach implies that atomic Forces for these conformations can be computed with high-level wavefunction-based approaches, such as the "gold standard" CCSD(T) method. We demonstrate that the flexible nature of the sGDML model recovers local and non-local electronic interactions (e.g. H-bonding, proton transfer, lone pairs, changes in hybridization states, steric repulsion and $n\to\pi^*$ interactions) without imposing any restriction on the nature of interatomic potentials. The analysis of sGDML Molecular dynamics trajectories yields new qualitative insights into dynamics and spectroscopy of small molecules close to spectroscopic accuracy.
-
towards exact Molecular dynamics simulations with machine learned Force fields
Nature Communications, 2018Co-Authors: Stefan Chmiela, Huziel E Sauceda, Klausrobert Muller, Alexandre TkatchenkoAbstract:Molecular dynamics (MD) simulations employing classical Force fields constitute the cornerstone of contemporary atomistic modeling in chemistry, biology, and materials science. However, the predictive power of these simulations is only as good as the underlying interatomic potential. Classical potentials often fail to faithfully capture key quantum effects in molecules and materials. Here we enable the direct construction of flexible Molecular Force fields from high-level ab initio calculations by incorporating spatial and temporal physical symmetries into a gradient-domain machine learning (sGDML) model in an automatic data-driven way. The developed sGDML approach faithfully reproduces global Force fields at quantum-chemical CCSD(T) level of accuracy and allows converged Molecular dynamics simulations with fully quantized electrons and nuclei. We present MD simulations, for flexible molecules with up to a few dozen atoms and provide insights into the dynamical behavior of these molecules. Our approach provides the key missing ingredient for achieving spectroscopic accuracy in Molecular simulations.
-
machine learning of accurate energy conserving Molecular Force fields
Science Advances, 2017Co-Authors: Stefan Chmiela, Alexandre Tkatchenko, Huziel E Sauceda, Igor Poltavsky, Kristof T Schutt, Klausrobert MullerAbstract:Using conservation of energy—a fundamental property of closed classical and quantum mechanical systems—we develop an efficient gradient-domain machine learning (GDML) approach to construct accurate Molecular Force fields using a restricted number of samples from ab initio Molecular dynamics (AIMD) trajectories. The GDML implementation is able to reproduce global potential energy surfaces of intermediate-sized molecules with an accuracy of 0.3 kcal mol −1 for energies and 1 kcal mol −1 A −1 for atomic Forces using only 1000 conformational geometries for training. We demonstrate this accuracy for AIMD trajectories of molecules, including benzene, toluene, naphthalene, ethanol, uracil, and aspirin. The challenge of constructing conservative Force fields is accomplished in our work by learning in a Hilbert space of vector-valued functions that obey the law of energy conservation. The GDML approach enables quantitative Molecular dynamics simulations for molecules at a fraction of cost of explicit AIMD calculations, thereby allowing the construction of efficient Force fields with the accuracy and transferability of high-level ab initio methods.
Agílio A. H. Pádua - One of the best experts on this subject based on the ideXlab platform.
-
Molecular Force field for ionic liquids iii imidazolium pyridinium and phosphonium cations chloride bromide and dicyanamide anions
Journal of Physical Chemistry B, 2006Co-Authors: Jose Canongia N Lopes, Agílio A. H. PáduaAbstract:This is the third set of parameters of a Force field for the Molecular simulation of ionic liquids, developed within the spirit of the OPLS-AA model and thus oriented toward the calculation of equilibrium thermodynamic and structural properties. The parameter sets reported here concern the cations alkylimidazolium, tetra-alkylphosphonium, and N-alkylpyridinium, and the anions chloride, bromide, and dicyanamide. The Force field is built in a stepwise manner that allows the construction of models for an entire family of cations, with alkyl side chains of different length, for example. Due to the transferability of the present Force field, the ions studied here can be combined with those reported in our two previous publications to create a large variety of ionic liquids that can be studied by Molecular simulation. The parameters reported were obtained through different series of ab initio calculations concerning the geometry, Force constants, torsion energy profiles, and electrostatic charge distributions o...
-
using spectroscopic data on imidazolium cation conformations to test a Molecular Force field for ionic liquids
Journal of Physical Chemistry B, 2006Co-Authors: Jose Canongia N Lopes, Agílio A. H. PáduaAbstract:A Molecular Force field for the computer simulation of ionic liquids is evaluated a posteriori by confrontation against Raman spectroscopic data, published after the Force field had been formulated. Specifically, the terms in the Force field describing the conformational aspects of dialkylimidazolium cations, which were specifically developed for these compounds using high level ab initio calculations, are those affecting the distribution of conformers in simulated ionic liquids. Those distributions are compared with analyses of the liquid-phase Raman spectra, and the features of a series of dihedral torsions along the alkyl side chains in 1-alkyl-3-methylimidazolium cations in several ionic liquids are discussed.
-
Molecular Force field for ionic liquids III: Imidazolium, pyridinium, and phosphonium cations; chloride, bromide, and dicyanamide anions
Journal of Physical Chemistry B, 2006Co-Authors: José N. Canongia Lopes, Agílio A. H. PáduaAbstract:This is the third set of parameters of a Force field for the Molecular simulation of ionic liquids, developed within the spirit of the OPLS-AA model and thus oriented toward the calculation of equilibrium thermodynamic and structural properties. The parameter sets reported here concern the cations alkylimidazolium, tetra-alkylphosphonium, and N-alkylpyridinium, and the anions chloride, bromide, and dicyanamide. The Force field is built in a stepwise manner that allows the construction of models for an entire family of cations, with alkyl side chains of different length, for example. Due to the transferability of the present Force field, the ions studied here can be combined with those reported in our two previous publications to create a large variety of ionic liquids that can be studied by Molecular simulation. The parameters reported were obtained through different series of ab initio calculations concerning the geometry, Force constants, torsion energy profiles, and electrostatic charge distributions of the ions under study. Validation of the Force field consisted of comparison with experimental crystal structure and liquid density data.
-
Molecular Force field for ionic liquids composed of triflate or bistriflylimide anions
Journal of Physical Chemistry B, 2004Co-Authors: Jose Canongia N Lopes, Agílio A. H. PáduaAbstract:A set of Force field parameters is proposed for the Molecular simulation of ionic liquids containing the anions trifluoromethylsufate and bis(trifluoromethylsulfonyl)imide, also known as triflate and bistriflylimide, respectively. The new set can be combined with existing Force fields for cations in order to simulate common room-temperature ionic liquids, such as those of the dialkylimidazolium family, and can be integrated with the OPLS-AA or similar Force fields. Ab initio quantum chemical calculations were employed to obtain Molecular geometry, torsional energy profiles, and partial charge distribution in the triflate and bistriflylimide anions. One of the torsions in bistriflylimide, corresponding to the dihedral angle S−N−S−C, has a complex energy profile which is precisely reproduced by the present parameter set. A new set of partial electrostatic charges is also proposed for the pyrrolidinium and tri- and tetra-alkylammonium cations. Again, these parameters can be combined with the OPLS-AA specific...
Alexandre Tkatchenko - One of the best experts on this subject based on the ideXlab platform.
-
Molecular Force fields with gradient domain machine learning construction and application to dynamics of small molecules with coupled cluster Forces
Journal of Chemical Physics, 2019Co-Authors: Huziel E Sauceda, Stefan Chmiela, Igor Poltavsky, Klausrobert Muller, Alexandre TkatchenkoAbstract:We present the construction of Molecular Force fields for small molecules (less than 25 atoms) using the recently developed symmetrized gradient-domain machine learning (sGDML) approach [Chmiela et al., Nat. Commun. 9, 3887 (2018) and Chmiela et al., Sci. Adv. 3, e1603015 (2017)]. This approach is able to accurately reconstruct complex high-dimensional potential-energy surfaces from just a few 100s of Molecular conformations extracted from ab initio Molecular dynamics trajectories. The data efficiency of the sGDML approach implies that atomic Forces for these conformations can be computed with high-level wavefunction-based approaches, such as the “gold standard” coupled-cluster theory with single, double and perturbative triple excitations [CCSD(T)]. We demonstrate that the flexible nature of the sGDML model recovers local and non-local electronic interactions (e.g., H-bonding, proton transfer, lone pairs, changes in hybridization states, steric repulsion, and n → π* interactions) without imposing any restriction on the nature of interatomic potentials. The analysis of sGDML Molecular dynamics trajectories yields new qualitative insights into dynamics and spectroscopy of small molecules close to spectroscopic accuracy.We present the construction of Molecular Force fields for small molecules (less than 25 atoms) using the recently developed symmetrized gradient-domain machine learning (sGDML) approach [Chmiela et al., Nat. Commun. 9, 3887 (2018) and Chmiela et al., Sci. Adv. 3, e1603015 (2017)]. This approach is able to accurately reconstruct complex high-dimensional potential-energy surfaces from just a few 100s of Molecular conformations extracted from ab initio Molecular dynamics trajectories. The data efficiency of the sGDML approach implies that atomic Forces for these conformations can be computed with high-level wavefunction-based approaches, such as the “gold standard” coupled-cluster theory with single, double and perturbative triple excitations [CCSD(T)]. We demonstrate that the flexible nature of the sGDML model recovers local and non-local electronic interactions (e.g., H-bonding, proton transfer, lone pairs, changes in hybridization states, steric repulsion, and n → π* interactions) without imposing any res...
-
Molecular Force fields with gradient domain machine learning construction and application to dynamics of small molecules with coupled cluster Forces
Journal of Chemical Physics, 2019Co-Authors: Huziel E Sauceda, Stefan Chmiela, Igor Poltavsky, Klausrobert Muller, Alexandre TkatchenkoAbstract:We present the construction of Molecular Force fields for small molecules (less than 25 atoms) using the recently developed symmetrized gradient-domain machine learning (sGDML) approach [Chmiela et al., Nat. Commun. 9, 3887 (2018) and Chmiela et al., Sci. Adv. 3, e1603015 (2017)]. This approach is able to accurately reconstruct complex high-dimensional potential-energy surfaces from just a few 100s of Molecular conformations extracted from ab initio Molecular dynamics trajectories. The data efficiency of the sGDML approach implies that atomic Forces for these conformations can be computed with high-level wavefunction-based approaches, such as the "gold standard" coupled-cluster theory with single, double and perturbative triple excitations [CCSD(T)]. We demonstrate that the flexible nature of the sGDML model recovers local and non-local electronic interactions (e.g., H-bonding, proton transfer, lone pairs, changes in hybridization states, steric repulsion, and n → π* interactions) without imposing any restriction on the nature of interatomic potentials. The analysis of sGDML Molecular dynamics trajectories yields new qualitative insights into dynamics and spectroscopy of small molecules close to spectroscopic accuracy.
-
Molecular Force fields with gradient domain machine learning construction and application to dynamics of small molecules with coupled cluster Forces
arXiv: Chemical Physics, 2019Co-Authors: Huziel E Sauceda, Stefan Chmiela, Igor Poltavsky, Klausrobert Muller, Alexandre TkatchenkoAbstract:We present the construction of Molecular Force fields for small molecules (less than 25 atoms) using the recently developed symmetrized gradient-domain machine learning (sGDML) approach [Chmiela et al., Nat. Commun. 9, 3887 (2018); Sci. Adv. 3, e1603015 (2017)]. This approach is able to accurately reconstruct complex high-dimensional potential-energy surfaces from just a few 100s of Molecular conformations extracted from ab initio Molecular dynamics trajectories. The data efficiency of the sGDML approach implies that atomic Forces for these conformations can be computed with high-level wavefunction-based approaches, such as the "gold standard" CCSD(T) method. We demonstrate that the flexible nature of the sGDML model recovers local and non-local electronic interactions (e.g. H-bonding, proton transfer, lone pairs, changes in hybridization states, steric repulsion and $n\to\pi^*$ interactions) without imposing any restriction on the nature of interatomic potentials. The analysis of sGDML Molecular dynamics trajectories yields new qualitative insights into dynamics and spectroscopy of small molecules close to spectroscopic accuracy.
-
towards exact Molecular dynamics simulations with machine learned Force fields
Nature Communications, 2018Co-Authors: Stefan Chmiela, Huziel E Sauceda, Klausrobert Muller, Alexandre TkatchenkoAbstract:Molecular dynamics (MD) simulations employing classical Force fields constitute the cornerstone of contemporary atomistic modeling in chemistry, biology, and materials science. However, the predictive power of these simulations is only as good as the underlying interatomic potential. Classical potentials often fail to faithfully capture key quantum effects in molecules and materials. Here we enable the direct construction of flexible Molecular Force fields from high-level ab initio calculations by incorporating spatial and temporal physical symmetries into a gradient-domain machine learning (sGDML) model in an automatic data-driven way. The developed sGDML approach faithfully reproduces global Force fields at quantum-chemical CCSD(T) level of accuracy and allows converged Molecular dynamics simulations with fully quantized electrons and nuclei. We present MD simulations, for flexible molecules with up to a few dozen atoms and provide insights into the dynamical behavior of these molecules. Our approach provides the key missing ingredient for achieving spectroscopic accuracy in Molecular simulations.
-
machine learning of accurate energy conserving Molecular Force fields
Science Advances, 2017Co-Authors: Stefan Chmiela, Alexandre Tkatchenko, Huziel E Sauceda, Igor Poltavsky, Kristof T Schutt, Klausrobert MullerAbstract:Using conservation of energy—a fundamental property of closed classical and quantum mechanical systems—we develop an efficient gradient-domain machine learning (GDML) approach to construct accurate Molecular Force fields using a restricted number of samples from ab initio Molecular dynamics (AIMD) trajectories. The GDML implementation is able to reproduce global potential energy surfaces of intermediate-sized molecules with an accuracy of 0.3 kcal mol −1 for energies and 1 kcal mol −1 A −1 for atomic Forces using only 1000 conformational geometries for training. We demonstrate this accuracy for AIMD trajectories of molecules, including benzene, toluene, naphthalene, ethanol, uracil, and aspirin. The challenge of constructing conservative Force fields is accomplished in our work by learning in a Hilbert space of vector-valued functions that obey the law of energy conservation. The GDML approach enables quantitative Molecular dynamics simulations for molecules at a fraction of cost of explicit AIMD calculations, thereby allowing the construction of efficient Force fields with the accuracy and transferability of high-level ab initio methods.
Thomas A Halgren - One of the best experts on this subject based on the ideXlab platform.
-
MERCK Molecular Force FIELD. V. EXTENSION OF MMFF94 USING EXPERIMENTAL DATA, ADDITIONAL COMPUTATIONAL DATA, AND EMPIRICAL RULES
Journal of Computational Chemistry, 1996Co-Authors: Thomas A HalgrenAbstract:This article describes the extension of the Merck Molecular Force Field (MMFF94) to a much broader range of organic systems. It also describes a preliminary parameterization of MMFF94 for the hydronium and hydroxide ions and for various halide, alkalai, and alkalai earth ions as well as for such “protein” metals as Zn2+, Ca2+, Cu2+, Cu+, Fe2+, and Fe3+. The extension employed computational data on charge distributions, Molecular geometries, and conformational energies for a series of oxysulfur (particularly sulfonamide) and oxyphosphorous compounds and for a diverse set of small molecules and ions not covered in the core parameterization. It also employed experimental data for approximately 2800 good-quality structures extracted from the Cambridge Structural Database (CSD). Some of the additional computational data were used to extend the explicit parameterization of electrostatic interactions and to more widely define a useful additive approximation for the “bond polarity” parameters (bond charge increments) used in MMFF94. Both the experimental and computational data served to define reference bond lengths and angles that the extended Force field uses in conjunction with Force constants obtained from carefully calibrated empirical rules. The extended torsion parameters consist partly of explicit parameters derived to reproduce MP2/6-31G* conformational energies and partly of “default parameters” provided by empirical rules patterned after those used in DREIDING and UFF but calibrated, where possible, against computationally derived MMFF94 torsion parameters. Comparisons to experimental data show that MMFF94 reproduces crystallographic bond lengths and bond angles with relatively modest root mean square (rms) deviations of approx. 0.02 A and 2°, respectively. © John Wiley & Sons, Inc.
-
merck Molecular Force field ii mmff94 van der waals and electrostatic parameters for interMolecular interactions
Journal of Computational Chemistry, 1996Co-Authors: Thomas A HalgrenAbstract:This article defines the parameterization and performance of MMFF94 for interMolecular interactions. It specifies the novel “buffered” functional forms used for treating van der Waals (vdW) and electrostatic interactions, and describes the use of : (1) high quality ab initio data to parameterize vdW interactions involving aliphatic hydrogens; and (2) HF/6-31G* calculations on hydrogen-bonded complexes to parameterize nonbonded interactions in polar systems. Comparisons show that appropriate trends in the HF/6-31G* data are well reproduced by MMFF94 and that interMolecular interaction energies and geometries closely parallel those given by the highly regarded OPLS Force field. A proper balance between solvent–solvent, solvent–solute, and solute–solute interactions, critically important for prospective success in aqueous simulations, thus appears to be attained. Comparison of MMFF94, OPLS, CHELPG electrostatic potential fit, QEq, Gasteiger, and Abraham charges for 20 small molecules and ions also shows the close correspondence between MMFF94 and OPLS. As do OPLS and all current, widely used Force fields, MMFF94 employs “effective pair potentials” which incorporate in an averaged way the increases in polarity which occur in high dielectric media. Some limitations of this approach are discussed and suggestions for possible enhancements to MMFF94's functional form are noted. © 1996 John Wiley & Sons, Inc.
-
merck Molecular Force field i basis form scope parameterization and performance of mmff94
Journal of Computational Chemistry, 1996Co-Authors: Thomas A HalgrenAbstract:This article introduces MMFF94, the initial published version of the Merck Molecular Force field (MMFF). It describes the objectives set for MMFF, the form it takes, and the range of systems to which it applies. This study also outlines the methodology employed in parameterizing MMFF94 and summarizes its performance in reproducing computational and experimental data. Though similar to MM3 in some respects, MMFF94 differs in ways intended to facilitate application to condensed-phase processes in Molecular-dynamics simulations. Indeed, MMFF94 seeks to achieve MM3-like accuracy for small molecules in a combined “organic/protein” Force field that is equally applicable to proteins and other systems of biological significance. A second distinguishing feature is that the core portion of MMFF94 has primarily been derived from high-quality computational data—ca. 500 Molecular structures optimized at the HF/6-31G* level, 475 structures optimized at the MP2/6-31G* level, 380 MP2/6-31G* structures evaluated at a defined approximation to the MP4SDQ/TZP level, and 1450 structures partly derived from MP2/6-31G* geometries and evaluated at the MP2/TZP level. A third distinguishing feature is that MMFF94 has been parameterized for a wide variety of chemical systems of interest to organic and medicial chemists, including many that feature frequently occurring combinations of functional groups for which little, if any, useful experimental data are available. The methodology used in parameterizing MMFF94 represents a fourth distinguishing feature. Rather than using the common “functional group” approach, nearly all MMFF parameters have been determined in a mutually consistent fashion from the full set of available computational data. MMFF94 reproduces the computational data used in its parameterization very well. In addition, MMFF94 reproduces experimental bond lengths (0.014 A root mean square [rms]), bond angles (1.2° rms), vibrational frequencies (61 cm−1 rms), conformational energies (0.38 kcal/mol/rms), and rotational barriers (0.39 kcal/mol rms) very nearly as well as does MM3 for comparable systems. MMFF94 also describes interMolecular interactions in hydrogen-bonded systems in a way that closely parallels that given by the highly regarded OPLS Force field. © 1996 John Wiley & Sons, Inc.
-
merck Molecular Force field iv conformational energies and geometries for mmff94
Journal of Computational Chemistry, 1996Co-Authors: Thomas A Halgren, Robert B NachbarAbstract:This article describes the parameterization and performance of MMFF94 for conformational energies, rotational barriers, and equilibrium torsion angles. It describes the derivation of the torsion parameters from high-quality computational data and characterizes MMFF94's ability to reproduce both computational and experimental data, the latter particularly in relation to MM3. The computational data included: (i) ∼ 250 comparisons of conformational energy based on “MP4SDQ/TZP” calculations (triple-zeta plus polarization calculations at a defined approximation to the highly correlated MP4SDQ level) at MP2/6-31G* geometries; and (ii) ∼ 1200 MP2/TZP comparisons of “torsion profile” structures at geometries derived from MP2/6-31G* geometries. The torsion parameters were derived in restrained least-squares fits that used the complete set of available computational data, thereby ensuring that a fully optimal set of parameters would be obtained. The final parameters reproduce the “MP4SDQ/TZP” and MP2/TZP computational data with root mean square (rms) deviations of 0.31 and 0.50 kcal/mol, respectively. In addition, MMFF94 reproduces a set of 37 experimental gas-phase and solution conformational energies, enthalpies, and free energies with a rms deviation of 0.38 kcal/mol; for comparison, the “MP4SDQ/TZP” calculations and MM3 each gives a rms deviation of 0.37 kcal/mol. Furthermore, MMFF94 reproduces 28 experimentally determined rotational barriers with a rms deviation of 0.39 kcal/mol. Given the diverse nature of the experimental conformational energies and rotational barriers and the clear indications of experimental error in some cases, the MMFF94 results appear excellent. Nevertheless, MMFF94 encounters somewhat greater difficulty in handling multifunctional compounds that place highly polar functional groups in close proximity, probably because it, like other commonly used Force fields, too greatly simplifies the description of electrostatic interactions. Some suggestions for enhancements to MMFF94's functional form are discussed. © 1996 John Wiley & Sons, Inc.
-
merck Molecular Force field iii Molecular geometries and vibrational frequencies for mmff94
Journal of Computational Chemistry, 1996Co-Authors: Thomas A HalgrenAbstract:This article describes the parameterization and performance of MMFF94 for Molecular geometries and deformations. It defines the form used for the valence-coordinate terms that represent variations in bond lengths and angles, and it describes the derivation of quadratic Force constants from HF/6-31G* data and the derivation of reference bond lengths and angles from fits to MP2/6-31G*-optimized geometries. Comparisons offered show that MMFF94 accurately reproduces the computational data used in its parameterization and demonstrate that its derivation from such data simultaneously confers the ability to reproduce experiment. In particular, MMFF94 reproduces experimentally determined bond lengths and angles for 30 organic molecules with root mean square (rms) deviations of 0.014 A and 1.2°, respectively. MM3 reproduces bond angles to the same accuracy, but reproduces experimental bond lengths more accurately, in part because it was fit directly to thermally averaged experimental bond lengths; MMFF94, in contrast, was fit to (usually shorter) energy-minimum values, as is proper for an anharmonic Force field intended for use in Molecular-dynamics simulations. The comparisons also show that UFF and a recent version of CHARMm (QUANTA 3.3 parameterization) are less accurate for Molecular geometries than either MMFF94 or MM3. For vibrational frequencies, MMFF94 and MM3 give comparable overall rms deviations versus experiment of 61 cm−1 and 57 cm−1, respectively, for 15 small, mostly organic molecules. In a number of instances, MM3's derivation employed observed frequencies that differ substantially—by nearly 400 cm−1 in one case—from other published frequencies which had themselves been confirmed theoretically by good-quality ab initio calculations. Overall, the comparisons to experimental geometries and vibrational frequencies demonstrate that MMFF94 achieves MM3-like accuracy for organic systems for which MM3 has been parameterized. Because MMFF94 is derived mainly from computational data, however, it has been possible to parameterize MMFF94 with equal rigor for a wide variety of additional systems for which little or no useful experimental data exist. Equally good performance can be expected for such systems. © John Wiley & Sons, Inc.
Huziel E Sauceda - One of the best experts on this subject based on the ideXlab platform.
-
Molecular Force fields with gradient domain machine learning construction and application to dynamics of small molecules with coupled cluster Forces
Journal of Chemical Physics, 2019Co-Authors: Huziel E Sauceda, Stefan Chmiela, Igor Poltavsky, Klausrobert Muller, Alexandre TkatchenkoAbstract:We present the construction of Molecular Force fields for small molecules (less than 25 atoms) using the recently developed symmetrized gradient-domain machine learning (sGDML) approach [Chmiela et al., Nat. Commun. 9, 3887 (2018) and Chmiela et al., Sci. Adv. 3, e1603015 (2017)]. This approach is able to accurately reconstruct complex high-dimensional potential-energy surfaces from just a few 100s of Molecular conformations extracted from ab initio Molecular dynamics trajectories. The data efficiency of the sGDML approach implies that atomic Forces for these conformations can be computed with high-level wavefunction-based approaches, such as the “gold standard” coupled-cluster theory with single, double and perturbative triple excitations [CCSD(T)]. We demonstrate that the flexible nature of the sGDML model recovers local and non-local electronic interactions (e.g., H-bonding, proton transfer, lone pairs, changes in hybridization states, steric repulsion, and n → π* interactions) without imposing any restriction on the nature of interatomic potentials. The analysis of sGDML Molecular dynamics trajectories yields new qualitative insights into dynamics and spectroscopy of small molecules close to spectroscopic accuracy.We present the construction of Molecular Force fields for small molecules (less than 25 atoms) using the recently developed symmetrized gradient-domain machine learning (sGDML) approach [Chmiela et al., Nat. Commun. 9, 3887 (2018) and Chmiela et al., Sci. Adv. 3, e1603015 (2017)]. This approach is able to accurately reconstruct complex high-dimensional potential-energy surfaces from just a few 100s of Molecular conformations extracted from ab initio Molecular dynamics trajectories. The data efficiency of the sGDML approach implies that atomic Forces for these conformations can be computed with high-level wavefunction-based approaches, such as the “gold standard” coupled-cluster theory with single, double and perturbative triple excitations [CCSD(T)]. We demonstrate that the flexible nature of the sGDML model recovers local and non-local electronic interactions (e.g., H-bonding, proton transfer, lone pairs, changes in hybridization states, steric repulsion, and n → π* interactions) without imposing any res...
-
Molecular Force fields with gradient domain machine learning construction and application to dynamics of small molecules with coupled cluster Forces
Journal of Chemical Physics, 2019Co-Authors: Huziel E Sauceda, Stefan Chmiela, Igor Poltavsky, Klausrobert Muller, Alexandre TkatchenkoAbstract:We present the construction of Molecular Force fields for small molecules (less than 25 atoms) using the recently developed symmetrized gradient-domain machine learning (sGDML) approach [Chmiela et al., Nat. Commun. 9, 3887 (2018) and Chmiela et al., Sci. Adv. 3, e1603015 (2017)]. This approach is able to accurately reconstruct complex high-dimensional potential-energy surfaces from just a few 100s of Molecular conformations extracted from ab initio Molecular dynamics trajectories. The data efficiency of the sGDML approach implies that atomic Forces for these conformations can be computed with high-level wavefunction-based approaches, such as the "gold standard" coupled-cluster theory with single, double and perturbative triple excitations [CCSD(T)]. We demonstrate that the flexible nature of the sGDML model recovers local and non-local electronic interactions (e.g., H-bonding, proton transfer, lone pairs, changes in hybridization states, steric repulsion, and n → π* interactions) without imposing any restriction on the nature of interatomic potentials. The analysis of sGDML Molecular dynamics trajectories yields new qualitative insights into dynamics and spectroscopy of small molecules close to spectroscopic accuracy.
-
Molecular Force fields with gradient domain machine learning construction and application to dynamics of small molecules with coupled cluster Forces
arXiv: Chemical Physics, 2019Co-Authors: Huziel E Sauceda, Stefan Chmiela, Igor Poltavsky, Klausrobert Muller, Alexandre TkatchenkoAbstract:We present the construction of Molecular Force fields for small molecules (less than 25 atoms) using the recently developed symmetrized gradient-domain machine learning (sGDML) approach [Chmiela et al., Nat. Commun. 9, 3887 (2018); Sci. Adv. 3, e1603015 (2017)]. This approach is able to accurately reconstruct complex high-dimensional potential-energy surfaces from just a few 100s of Molecular conformations extracted from ab initio Molecular dynamics trajectories. The data efficiency of the sGDML approach implies that atomic Forces for these conformations can be computed with high-level wavefunction-based approaches, such as the "gold standard" CCSD(T) method. We demonstrate that the flexible nature of the sGDML model recovers local and non-local electronic interactions (e.g. H-bonding, proton transfer, lone pairs, changes in hybridization states, steric repulsion and $n\to\pi^*$ interactions) without imposing any restriction on the nature of interatomic potentials. The analysis of sGDML Molecular dynamics trajectories yields new qualitative insights into dynamics and spectroscopy of small molecules close to spectroscopic accuracy.
-
towards exact Molecular dynamics simulations with machine learned Force fields
Nature Communications, 2018Co-Authors: Stefan Chmiela, Huziel E Sauceda, Klausrobert Muller, Alexandre TkatchenkoAbstract:Molecular dynamics (MD) simulations employing classical Force fields constitute the cornerstone of contemporary atomistic modeling in chemistry, biology, and materials science. However, the predictive power of these simulations is only as good as the underlying interatomic potential. Classical potentials often fail to faithfully capture key quantum effects in molecules and materials. Here we enable the direct construction of flexible Molecular Force fields from high-level ab initio calculations by incorporating spatial and temporal physical symmetries into a gradient-domain machine learning (sGDML) model in an automatic data-driven way. The developed sGDML approach faithfully reproduces global Force fields at quantum-chemical CCSD(T) level of accuracy and allows converged Molecular dynamics simulations with fully quantized electrons and nuclei. We present MD simulations, for flexible molecules with up to a few dozen atoms and provide insights into the dynamical behavior of these molecules. Our approach provides the key missing ingredient for achieving spectroscopic accuracy in Molecular simulations.
-
machine learning of accurate energy conserving Molecular Force fields
Science Advances, 2017Co-Authors: Stefan Chmiela, Alexandre Tkatchenko, Huziel E Sauceda, Igor Poltavsky, Kristof T Schutt, Klausrobert MullerAbstract:Using conservation of energy—a fundamental property of closed classical and quantum mechanical systems—we develop an efficient gradient-domain machine learning (GDML) approach to construct accurate Molecular Force fields using a restricted number of samples from ab initio Molecular dynamics (AIMD) trajectories. The GDML implementation is able to reproduce global potential energy surfaces of intermediate-sized molecules with an accuracy of 0.3 kcal mol −1 for energies and 1 kcal mol −1 A −1 for atomic Forces using only 1000 conformational geometries for training. We demonstrate this accuracy for AIMD trajectories of molecules, including benzene, toluene, naphthalene, ethanol, uracil, and aspirin. The challenge of constructing conservative Force fields is accomplished in our work by learning in a Hilbert space of vector-valued functions that obey the law of energy conservation. The GDML approach enables quantitative Molecular dynamics simulations for molecules at a fraction of cost of explicit AIMD calculations, thereby allowing the construction of efficient Force fields with the accuracy and transferability of high-level ab initio methods.