Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 2109.03769

Training Algorithm Matters for the Performance of Neural Network Potential: A Case Study of Adam and the Kalman Filter Optimizers

Yunqi Shao·Florian M. Dietrich·Carl Nettelblad·Chao Zhang
View original source ↗Read source document ↗Export metadata ↓

Abstract

One hidden yet important issue for developing neural network potentials (NNPs) is the choice of training algorithm. Here we compare the performance of two popular training algorithms, the adaptive moment estimation algorithm (Adam) and the Extended Kalman Filter algorithm (EKF), using the Behler-Parrinello neural network (BPNN) and two publicly accessible datasets of liquid water [Proc. Natl. Acad. Sci. U.S.A. 2016, 113, 8368-8373 and Proc. Natl. Acad. Sci. U.S.A. 2019, 116, 1110-1115]. This is achieved by implementing EKF in TensorFlow. It is found that NNPs trained with EKF are more transferable and less sensitive to the value of the learning rate, as compared to Adam. In both cases, error metrics of the validation set do not always serve as a good indicator for the actual performance of NNPs. Instead, we show that their performance correlates well with a Fisher information based similarity measure.

Explore related subjects

physics.chem-phcs.LGphysics.comp-ph

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Yunqi Shao, Florian M. Dietrich, Carl Nettelblad, Chao Zhang. 2021-11-09. Training Algorithm Matters for the Performance of Neural Network Potential: A Case Study of Adam and the Kalman Filter Optimizers. https://doi.org/10.1063/5.0070931

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

SPIBER: Reconstructing Free Energy Landscapes from Short, Unconverged Trajectories with Generative Flow Networks

Molecular systems have many degrees of freedom, but their metastable behavior can often be described by a few collective variables. Identifying these variables and estimating free energies along them from limited simulation data remains a challenging, important problem. Separate short trajectories may sample different metastable states without capturing transitions or establishing their relative equilibrium populations. For unbiased trajectories generated with the same Hamiltonian at a single temperature, alternate methods based on histogram reweighting cannot correct this imbalance. Here we present SPIBER, which combines the State Predictive Information Bottleneck (SPIB) with Generative Flow Networks (GFlowNets). SPIB uses deep learning to approximate slow degrees of freedom through a past-future information bottleneck, retaining information needed to predict future metastable states. We show that this compression limits conditional entropy variations in populated regions, allowing conditional mean potential energies, which are much easier to calculate, to be used to approximate free energy differences. Given sufficient local sampling to estimate these energies, they define the target distribution for GFlowNets, energy-based generative samplers that sample according to estimated thermodynamic stability rather than observed populations. For a particle in a radial double-well potential, for alanine dipeptide, and for the nine-residue peptide AIB9, SPIBER recovers free energy differences between sampled metastable states to within one thermal energy unit of reference values. The method combines collective-variable learning and free energy estimation in up to four latent dimensions, without requiring converged state populations or additional molecular dynamics simulations.

physics.chem-ph↗
arXiv2026

Intrinsic Matching Frustration in Fluctuating Finite Systems

We formulate intrinsic matching frustration (IMF), a fluctuation-induced, kinetics-independent reduction in the mean capacity permitted by a prescribed matching rule. For complementary one-to-one matching, the instantaneous capacity is set by the minority population, so fluctuations produce a nonzero mean deficit even when the two populations are balanced on average. At finite size, this deficit depends on the full distribution of the population difference and is determined by its variance alone only in the Gaussian limit. Compartmentalization hides matching capacity by preventing cancellation between local imbalances of opposite sign. Fusion releases this hidden capacity monotonically under coarse graining, producing a measurable recovery of product yield following local reaction to completion.

physics.chem-ph↗
arXiv2026

Phonon chirality as an additive control of CISS: a symmetry-protected law

Chirality-induced spin selectivity (CISS) is usually associated with molecular handedness. The possible contribution of chiral phonons is less established. We study a helical tight-binding model in which local phonon angular momentum modulates spin-dependent nearest-neighbor hopping. Fewest-switches surface hopping calculations give the transmitted spin polarization $\mathrm{SP}=aC+b\mathrm{PH}$. Here $C$ is the molecular chirality and $\mathrm{PH}$ is the phonon chirality. A mirror symmetry reverses $C$, $\mathrm{PH}$, and $\mathrm{SP}$ simultaneously. This symmetry excludes both a chirality-independent offset and a $C\cdot\mathrm{PH}$ term. The phonon contribution can therefore enhance, cancel, or reverse the molecular CISS signal.

physics.chem-ph↗