Towards the Modeling of Atomic and Molecular Clusters Energy by Support Vector Regression

Vitek, Ale; Stachoň, Martin; Krömer, Pavel; Snáel, Václav

doi:10.1109/incos.2013.26

Cited by 16 publications

(14 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…57,58 Originally, SVMs have been developed for the classification of data into two groups with different properties by introducing a multidimensional hyperplane separating the data. As this hyperplane usually has a very complicated non-linear shape, the data is processed by a kernel substitution, which is in essence the mapping of the original data onto a higher-dimensional feature space, in which a linear separation is possible (see Fig.…”

Section: Support Vector Machinesmentioning

confidence: 99%

Perspective: Machine learning potentials for atomistic simulations

Behler

2016

The Journal of Chemical Physics

1,149

995

View full text Add to dashboard Cite

Nowadays, computer simulations have become a standard tool in essentially all fields of chemistry, condensed matter physics, and materials science. In order to keep up with state-of-the-art experiments and the ever growing complexity of the investigated problems, there is a constantly increasing need for simulations of more realistic, i.e., larger, model systems with improved accuracy. In many cases, the availability of sufficiently efficient interatomic potentials providing reliable energies and forces has become a serious bottleneck for performing these simulations. To address this problem, currently a paradigm change is taking place in the development of interatomic potentials. Since the early days of computer simulations simplified potentials have been derived using physical approximations whenever the direct application of electronic structure methods has been too demanding. Recent advances in machine learning (ML) now offer an alternative approach for the representation of potential-energy surfaces by fitting large data sets from electronic structure calculations. In this perspective, the central ideas underlying these ML potentials, solved problems and remaining challenges are reviewed along with a discussion of their current applicability and limitations. Published by AIP Publishing. [http://dx

show abstract

Section: Support Vector Machinesmentioning

confidence: 99%

Perspective: Machine learning potentials for atomistic simulations

Behler

2016

The Journal of Chemical Physics

1,149

995

View full text Add to dashboard Cite

show abstract

“…So far, fragment-based models have been applied to charge transport in DNA using a DFTB/MM framework [ 95 ] and to excited states in photosynthetic complexes using TDDFT [ 96 ], and an extension to DNA excited states is certainly feasible. Another exciting approach is the use of machine learning, where larger systems could in principle be described with QM accuracy, but at the speed of classical force fields [ 97 , 98 , 99 , 100 , 101 , 102 ].…”

Section: Potential Energy Surfacesmentioning

confidence: 99%

Challenges in Simulating Light-Induced Processes in DNA

et al. 2016

View full text Add to dashboard Cite

In this contribution, we give a perspective on the main challenges in performing theoretical simulations of photoinduced phenomena within DNA and its molecular building blocks. We distinguish the different tasks that should be involved in the simulation of a complete DNA strand subject to UV irradiation: (i) stationary quantum chemical computations; (ii) the explicit description of the initial excitation of DNA with light; (iii) modeling the nonadiabatic excited state dynamics; (iv) simulation of the detected experimental observable; and (v) the subsequent analysis of the respective results. We succinctly describe the methods that are currently employed in each of these steps. While for each of them, there are different approaches with different degrees of accuracy, no feasible method exists to tackle all problems at once. Depending on the technique or combination of several ones, it can be problematic to describe the stacking of nucleobases, bond breaking and formation, quantum interferences and tunneling or even simply to characterize the involved wavefunctions. It is therefore argued that more method development and/or the combination of different techniques are urgently required. It is essential also to exercise these new developments in further studies on DNA and subsystems thereof, ideally comprising simulations of all of the different components that occur in the corresponding experiments.

show abstract

“…9,10 While more sophisticated (nonpolarizable and polarizable) FFs have been developed over the years and are still the most common PEFs used in MD and MC simulations, [11][12][13][14][15] in the last decade machine-learning (ML) models trained on electronic structure data have gained in popularity, enabling computer simulations with higher level of accuracy. [16][17][18][19] Various types of ML PEFs have been proposed, including neural network potentials (NNPs), [20][21][22][23][24][25][26][27][28][29] Gaussian approximation potentials (GAPs), 30 moment tensor potentials (MTPs), 31 and spectral neighbor analysis potentials (SNAPs), 32 as well as PEFs based on the atomic cluster expansion, 33 graph networks, kernel ridge regression methods, 34 gradient-domain machine learning (GDML), 35 support vector machines (SVM), 36 permutationally invariant polynomials (PIPs), [37][38][39][40][41][42] and permutation invariant polynomial neural networks (PIP-NNs). [43][44][45][46] The interested reader is referred to several excellent reviews of ML-based PEFs which have recently appeared in the literature.…”

Section: Introductionmentioning

confidence: 99%

MB-Fit: Software Infrastructure for Data-Driven Many-Body Potential Energy Functions

Bull-Vulpe

Riera

Goetz

et al. 2021

Preprint

View full text Add to dashboard Cite

Many-body potential energy functions (MB-PEFs), which integrate data-driven representations of many-body short-range quantum mechanical interactions with physics-based representations of many-body polarization and long-range interactions, have recently been shown to provide high accuracy in the description of molecular interactions, from the gas to the condensed phase. Here, we present MB-Fit, a software infrastructure for the automated development of MB-PEFs for generic molecules within the TTM-nrg ("Thole-type model energy") and MB-nrg ("many-body energy") theoretical frameworks. Besides providing all the necessary computational tools for generating TTM-nrg and MB-nrg PEFs, MB-Fit provides a seamless interface with the MBX software, a many-body energy/force calculator for computer simulations. Given the demonstrated accuracy of the MB-PEFs, we believe that MB-Fit will enable routine, predictive computer simulations of generic (small) molecules in the gas, liquid, and solid phases, including, but not limited to, the modeling of isomeric equilibria in molecular clusters, solvation processes, molecular crystals, and phase diagrams.

show abstract

Towards the Modeling of Atomic and Molecular Clusters Energy by Support Vector Regression

Cited by 16 publications

References 18 publications

Perspective: Machine learning potentials for atomistic simulations

Perspective: Machine learning potentials for atomistic simulations

Challenges in Simulating Light-Induced Processes in DNA

MB-Fit: Software Infrastructure for Data-Driven Many-Body Potential Energy Functions

Contact Info

Product

Resources

About