An official website of the United States government
Abstract Machine learning interatomic potentials (MLIPs) are a promising technique for atomic modeling. While small errors are widely reported for MLIPs, an open concern is whether MLIPs can accurately reproduce atomistic dynamics and related physical properties in molecular dynamics (MD) simulations. In this study, we examine the state-of-the-art MLIPs and uncover several discrepancies related to atom dynamics, defects, and rare events (REs), compared to ab initio methods. We find that low averaged errors by current MLIP testing are insufficient, and develop quantitative metrics that better indicate the accurate prediction of atomic dynamics by MLIPs. The MLIPs optimized by the RE-based evaluation metrics are demonstrated to have improved prediction in multiple properties. The identified errors, the evaluation metrics, and the proposed process of developing such metrics are general to MLIPs, thus providing valuable guidance for future testing and improvements of accurate and reliable MLIPs for atomistic modeling.
more »
« less
Machine‐learning interatomic potentials for pyrolysis of polysiloxanes and properties of SiCO ceramics
Abstract We present machine‐learning interatomic potentials (MLIPs) for simulations of Si–C–O–H compounds. The MLIPs are constructed from moment tensor potentials (MTPs) and were trained to a library of configurations that included polysiloxane structures, hypothetical crystalline and amorphous SiCOH structures, and trajectories of Si–C–O–H systems obtained via ab initio molecular dynamic (aiMD) simulations at elevated temperatures. Passive, active, and hybrid learning strategies were implemented to develop the MLIPs. The MLIPs reproduce vibrational properties of polymers and SiCOH structures obtained from aiMD simulations, thus providing a tool to identify chemical units and distinct structural characteristics through their vibrational properties. Simulations of the polymer‐to‐ceramic transformation show the development of mixed tetrahedra in SiCO ceramics and align with experimental observations. Million‐atom simulations for several nanoseconds highlight the precipitation of graphitic nanosheets from a carbon‐rich SiCO precursor. Atomistic simulations with the MLIPs deliver details of chemical reaction mechanisms during the pyrolysis of polysiloxanes, including methane abstraction and Kumada‐like rearrangements that transform the siloxane backbone. While the MLIPs still leave room for systematic improvement, they deliver simulations with “density functional theory (DFT)‐like” quality at low and high temperatures.
more »
« less
- Award ID(s):
- 1743701
- PAR ID:
- 10531168
- Publisher / Repository:
- Wiley-Blackwell
- Date Published:
- Journal Name:
- Journal of the American Ceramic Society
- ISSN:
- 0002-7820
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Si is an n-type dopant in Ga2O3 that can be introduced intentionally or unintentionally. The results of Secondary Ion Mass Spectrometry, Hall effect, and infrared absorption experiments show that the hydrogen plasma exposure of Si-doped Ga2O3 leads to the formation of complexes containing Si and H and the passivation of n-type conductivity. The Si-H (D) complex gives rise to an O-H (D) vibrational line at 3477.6 (2577.8) cm-1 and is shown to contain a single H (or D) atom. The direction of the transition moment of this defect has been investigated to provide structure-sensitive information. Theory suggests possible structures for an OH-Si complex that are consistent with its observed vibrational properties.more » « less
-
Density-functional theory based molecular-dynamics simulations were used to investigate high-pressure chemical reactions in liquid mixtures of CO2 with several elements (Si, Mn, and Fe) at high temperatures of 2000–3000 K. Our ab initio simulations indicate that these reactant elements can reduce CO2 to C at high pressures (20 GPa) leading to the formation of C-C chains, with Si by far the most effective carbon-reducing agent. A combined chemical analysis using Bader charge analysis and crystal orbital Hamilton population (COHP) on simulation snapshots shows that significant charge transfer from the reducing element to the C atoms creates instability in the C-O covalent bonds. COHP analysis further shows that Mn/Fe-O and Mn/Fe-C bonding interactions are weaker compared to the Si counterparts. These results further our understanding of the redox chemistry of CO2 at conditions relevant to planetary mantle interiors and demonstrate the effectiveness of high pressure in the reduction of CO2 directly to solid carbon.more » « less
-
null (Ed.)The origin in deshielding of 29 Si NMR chemical shifts in R 3 Si–X, where X = H, OMe, Cl, OTf, [CH 6 B 11 X 6 ], toluene, and O X (O X = surface oxygen), as well as i Pr 3 Si + and Mes 3 Si + were studied using DFT methods. At the M06-L/6-31G(d,p) level of theory the geometry optimized structures agree well with those obtained experimentally. The trends in 29 Si NMR chemical shift also reproduce experimental trends; i Pr 3 Si–H has the most shielded 29 Si NMR chemical shift and free i Pr 3 Si + or isolable Mes 3 Si + have the most deshielded 29 Si NMR chemical shift. Natural localized molecular orbital (NLMO) analysis of the contributions to paramagnetic shielding ( σ p ) in these compounds shows that Si–R (R = alkyl, H) bonding orbitals are the major contributors to deshielding in this series. The Si–R bonding orbitals are coupled to the empty p-orbital in i Pr 3 Si + or Mes 3 Si + , or to the orbital in R 3 Si–X. This trend also applies to surface bound R 3 Si–O X . This model also explains chemical shift trends in recently isolated t Bu 2 SiH 2 + , t BuSiH 2 + , and SiH 3 + that show more shielded 29 Si NMR signals than R 3 Si + species. There is no correlation between isotropic 29 Si NMR chemical shift and charge at silicon.more » « less
-
null (Ed.)The application of machine learning models and algorithms towards describing atomic interactions has been a major area of interest in materials simulations in recent years, as machine learning interatomic potentials (MLIPs) are seen as being more flexible and accurate than their classical potential counterparts. This increase in accuracy of MLIPs over classical potentials has come at the cost of significantly increased complexity, leading to higher computational costs and lower physical interpretability and spurring research into improving the speeds and interpretability of MLIPs. As an alternative, in this work we leverage “machine learning” fitting databases and advanced optimization algorithms to fit a class of spline-based classical potentials, showing that they can be systematically improved in order to achieve accuracies comparable to those of low-complexity MLIPs. These results demonstrate that high model complexities may not be strictly necessary in order to achieve near-DFT accuracy in interatomic potentials and suggest an alternative route towards sampling the high accuracy, low complexity region of model space by starting with forms that promote simpler and more interpretable inter- atomic potentials.more » « less
