A Deep Learning Potential for Accurate Shock Response Simulations in Tin
Yixin Chen, Xiaoyang Wang, Wanghui Li, Mohan Chen, Han Wang · 2025
Read A Deep Learning Potential for Accurate Shock Response Simulations in Tin by Yixin Chen, Xiaoyang Wang, Wanghui Li, Mohan Chen, Han Wang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-12698.txt.
Introduction
Yixin Chen and co-authors build a machine-learned model of how tin atoms push on each other in a 2025 paper. Tin is a test metal for studying how ductile materials respond to shock waves. Atom-level simulations can reveal the fine mechanics, but hand-built force models for tin often miss key behavior, mostly its switch between solid forms under pressure.
Their deep learning potential was trained through a concurrent learning loop that keeps adding the cases the model handles worst. It covers pressures and temperatures from calm conditions to extremes, in solid and liquid forms alike. The model reproduces quantum-level reference values, measured melting lines, phase boundaries, and shock data well. The authors present it as a bridge between quantum accuracy and large-scale shock simulation.
Catalog record
| CREATOR | Yixin Chen, Xiaoyang Wang, Wanghui Li, Mohan Chen, Han Wang |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 35 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2505-12698
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A Deep Learning Potential for Accurate Shock Response Simulations in Tin Yixin Chen, Xiaoyang Wang, Wanghui Li, Mohan Chen, Han Wang Abstract Tin (Sn) plays a crucial role in studying the dynamic mechanical responses of ductile metals under shock loading. Atomistic simulations serves to unveil the nano-scale mechanisms for critical behaviors of dynamic responses. However, existing empirical potentials for Sn often lack sufficient accuracy when applied in such simulation. Particularly, the solid-solid phase transition behavior of Sn poses significant challenges to the accuracy of interatomic potentials. To address these challenges, this study introduces a machine-learning potential model for Sn, specifically optimized for shock-response simulations. The model is trained using a dataset constructed through a concurrent learning framework and is designed for molecular simulations across thermodynamic conditions ranging from 0 to 100 GPa and 0 to 5000 K, encompassing both solid and liquid phases as well as structures with free surfaces. It accurately reproduces density functional theory (DFT)-derived basic properties, experimental melting curves, solid-solid phase boundaries, and shock Hugoniot results. This demonstrates the model’s potential to bridge ab initio precision with large-scale dynamic simulations of Sn. I introduction Metals subjected to dynamic loading exhibit distinctive behaviors, which have been extensively investigated in both academic research and engineering applications [ 1 , 2 , 3 ] . Among them, tin (Sn) is of particular interest as it is a typical representative of low-melting-point ductile metals, whose dynamic mechanical responses have been studied widely in experiments [ 4 , 5 , 6 , 7 ] . Previous studies have revealed a complex pressure-temperature (P-T) phase diagram of Sn. In detail, at 286 K and ambient pressure, Sn undergoes a well-known phase transition from a covalent-bonding diamond structure ( \alpha -Sn, in Fd\bar{3}m space group) to a metallic-bonding double-bct structure ( \beta -Sn, in I4_{1}/amd space group) [ 8 ] . As pressure increases, a variety of high-pressure phases have been identified, including bct ( I4/mmm ), bcc ( Im\bar{3}m ), hcp ( P6_{3}/mmc ), and a bco ( Immm ) phase whose presence is still under debate [ 9 , 10 , 11 ] . Moreover, when subjected to shock loading, two phase transitions of Sn have been observed: a \beta -bct solid phase transition occurring at 9-10 GPa [ 12 ] and a solid-liquid phase at nearly 39 GPa [ 13 ] . Beyond phase transitions under high pressure, the dynamic fracture behavior of Sn during shock loading, specifically spallation, is of significant interest to the shock physics community. The shock induced spallation has been experimentally characterized as classical-spall and micro-spall. The former is usually featured by the production of solid-state spall layers, while the later is generally characterized by dispersed liquid debris and extensive cavitation [ 14 ] . The transition pressure between classic and micro spall has been reported to be at approximately 40 GPa experimentally using optical microscopy technique [ 4 , 5 ] . Shock experiments can also estimate the spall strength via the measurement of free surface velocities [ 6 ] . The spall strength of Sn depends on the state of materials at the time of fracture. It has been reported that spall strength in classic spall is usually an order of magnitude higher than that in the micro-spall [ 7 ] . Moreover, the spall strength is sensitive to the strain rate. The spall strength of liquid Sn increases from 0.1 GPa at 10^{5} strain rate, to 1.9 GPa at 10^{9} strain rate [ 6 ] . Despite the wealth of experimental results, substantial challenges remain in fully understanding the mechanisms of the phase transitions and dynamic response of Sn which requires high-resolution diagnostic techniques. A primary challenge is the difficulty of observing ultra-fast dynamic behaviors in situ during experiments. Phenomena such as solid phase transitions and the nucleation of microscopic voids and cavities often originate from material behaviors occurring at atomic or lattice spatial scales and occur within picosecond or nanosecond timeframes [ 15 , 16 , 17 ] . The existence of technical challenges in extreme experiments calls for advanced simulations. Atomic-scale simulation has emerged as a vital alternative approach, enabling direct investigation into the underlying atomisitic mechanisms for these complex material responses. In particular, the non-equilibrium molecular dynamics (NEMD) methods have been extensively employed in studying the shock compression and spallation of materials [ 18 , 19 , 20 , 21 , 22 ] . These studies have successfully addressed numerous microscopic processes associated with mechanical responses under shock-wave loading. For example, Liao et al. [ 23 ] revealed that the degree of melting has a substantial effect on void distributions and thermodynamic pathways in both classical spallation and micro-spallations. Wang et al. [ 24 ] studied the dependence of spall strength on shock velocities. Durand et al. [ 22 ] carried out large-scale MD simulations and studied the impact of phase transitions on micro-jetting and ejecta production in shock-loaded Sn. Wu et al. [ 25 ] examined the properties of ejecta under supported and unsupported shocks and identified multiple phases of Sn using the adaptive common neighbor analysis (a-CNA) method [ 26 ] . These simulations have highlighted the complex dynamic response of Sn under shock loading and also signified effect of corresponding phase transitions in Sn. The accuracy of NEMD simulations is largely dependent on the precision of the potential energy surface (PES) model. All of the studies mentioned above employed empirical potentials to represent the PES of tin. Liao et al. , Durand et al. adopted the modified embedded atom method (MEAM) [ 27 ] model parameterized by Ravelo et al. [ 28 ] . Wang et al. and Wu et al. applied the embedded atom method (EAM) model [ 29 ] parameterized by Sapozhnikov et al. [ 30 ] . Despite the extensive application of empirical potentials in tin simulations, existing empirical models are unable to concurrently characterize both the \beta and bct phases in shock-loaded simulations of Sn. The EAM potential by Sapozhnikov et al. [ 30 ] is only fitted to the bct and bcc phases at high pressures, which fails to reproduce the \beta phase at ambient pressures. Consequently, when employing this EAM model in MD simulations, the initial structural configuration under ambient pressure conditions must be set to bct or bcc rather than the experimentally prevalent \beta phase [ 22 ] . The MEAM model by Ravelo et al. [ 28 ] and a more recent model by Ko et al. [ 31 ] , are designed to capture the low-pressure \alpha and \beta phases, whereas the bct and bcc phases are not included as fitting targets in these models. Consequently, these MEAM models may provide an inaccurate description of high-pressure phase transitions in dynamic loading simulations. The shortcomings in these models underscore the need for the development of more precise models in this area, which serves as the motivation for our research. To this end, machine learning interatomic potentials (MLIPs), which have emerged as a powerful approach for describing the inter-atomic PES, provide an alternative way to overcome the long-standing challenge [ 32 , 33 , 34 , 35 , 36 ] . Recently, two MLIPs for Sn have been proposed, i.e., DP-SCAN