pith. machine review for the scientific record. sign in

arxiv: 2512.20424 · v1 · submitted 2025-12-23 · ❄️ cond-mat.mtrl-sci · physics.comp-ph

Recognition: 2 theorem links

· Lean Theorem

Iterative learning scheme for crystal structure prediction with anharmonic lattice dynamics

Hao Gao , Yue-Wen Fang , Ion Errea
Authors on Pith no claims yet

Pith reviewed 2026-05-16 20:27 UTC · model grok-4.3

classification ❄️ cond-mat.mtrl-sci physics.comp-ph
keywords crystal structure predictionanharmonic lattice dynamicsmachine learning potentialsstochastic self-consistent harmonic approximationH3Sphase stabilityhigh-pressure materials
0
0 comments X

The pith

An iterative learning scheme enables crystal structure prediction with anharmonic lattice dynamics using reduced training data.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces an iterative framework that merges evolutionary algorithms for exploring structures, atomic foundation models for fast and robust relaxations, and the stochastic self-consistent harmonic approximation to include anharmonic contributions to the free energy. Conventional crystal structure prediction overlooks these ionic and anharmonic effects, which matter near displacive transitions in materials such as hydrides and ferroelectrics. By relying on foundation models, the method cuts the volume of training data required for machine-learning potentials. On the strongly anharmonic H3S system it reproduces density-functional-theory benchmarks for phase stability and vibrational spectra between 50 and 200 GPa. The statistical averaging built into SSCHA further relaxes the accuracy demanded from the potentials, making the overall procedure practical.

Core claim

The central claim is that an iterative learning scheme combining evolutionary algorithms, atomic foundation models, and SSCHA performs crystal structure prediction while properly accounting for anharmonic lattice dynamics. Foundation models permit reliable relaxation of random candidate structures with far less training data than conventional machine-learning potentials require. When the scheme is applied to H3S it yields phase stability and vibrational properties from 50 to 200 GPa that agree with full density-functional-theory benchmarks. The statistical averaging inside SSCHA reduces free-energy errors enough that moderate inaccuracies in the potentials remain tolerable.

What carries the argument

The iterative learning framework that alternates evolutionary structure generation, foundation-model relaxations of random candidates, and SSCHA free-energy evaluations to incorporate anharmonicity without prohibitive cost.

If this is right

  • Crystal structure prediction becomes feasible for solids near displacive transitions where harmonic approximations are known to fail.
  • The computational cost of including anharmonic effects drops enough to allow routine searches over larger numbers of candidate structures.
  • Phase stability and vibrational properties can be obtained together in one workflow for high-pressure hydrides and related materials.
  • Moderate errors in machine-learning potentials are acceptable once SSCHA statistical averaging is applied, lowering the bar for potential training.
  • The same iterative loop can be reused for other systems that exhibit strong lattice anharmonicity.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The approach could be tested on larger supercells or different pressure ranges to check whether the reduced-data advantage persists.
  • Similar averaging techniques might improve accuracy in other machine-learning-driven simulations of free energies.
  • One could examine whether newer or larger foundation models further shrink the required training sets for still more complex chemistries.
  • The method suggests a general template for hybrid physics-ML workflows in which expensive corrections are applied only after cheap relaxations.

Load-bearing premise

Foundation models enable robust relaxations of random structures with drastically reduced training data and the statistical averaging in SSCHA sufficiently reduces free-energy errors to tolerate moderate inaccuracies in the machine-learning potentials.

What would settle it

Running a full first-principles SSCHA calculation on a second strongly anharmonic system and obtaining stable phases or phonon spectra that differ from those predicted by the iterative scheme would show the method fails to reach benchmark accuracy.

Figures

Figures reproduced from arXiv: 2512.20424 by Hao Gao, Ion Errea, Yue-Wen Fang.

Figure 1
Figure 1. Figure 1: FIG. 1. Iterative learning (a) and SSCHA-assisted searching (b) workflows in this work. [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. The enthalpy–volume plots for H [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. Parity plots of energy (upper) and forces (lower) comparing finetuned MatterSim predictions [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4. (a) Squared frequency of the [PITH_FULL_IMAGE:figures/full_fig_p013_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5. Relative enthalpy at the classical BO level without considering anharmonic effects of H [PITH_FULL_IMAGE:figures/full_fig_p014_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6. (a) Structural transformations from metastable H [PITH_FULL_IMAGE:figures/full_fig_p015_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: FIG. 7. Distributions of gradient errors and parity plots comparing finetuned MatterSim predictions [PITH_FULL_IMAGE:figures/full_fig_p016_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: FIG. 8. Distributions of gradient errors in the [PITH_FULL_IMAGE:figures/full_fig_p017_8.png] view at source ↗
read the original abstract

First-principles based crystal structure prediction (CSP) methods have revealed an essential tool for the discovery of new materials. However, in solids close to displacive phase transitions, which are common in ferroelectrics, thermoelectrics, charge-density wave systems, or superconducting hydrides, the ionic contribution to the free energy and lattice anharmonicity become essential, limiting the capacity of CSP techniques to determine the thermodynamical stability of competing phases. While variational methods like the stochastic self-consistent harmonic approximation (SSCHA) accurately account for anharmonic lattice dynamics \emph{ab initio}, their high computational cost makes them impractical for CSP. Machine-learning interatomic potentials offer accelerated sampling of the energy landscape compared to purely first-principles approaches, but their reliance on extensive training data and limited generalization restricts practical applications. Here, we propose an iterative learning framework combining evolutionary algorithms, atomic foundation models, and SSCHA to enable CSP with anharmonic lattice dynamics. Foundation models enable robust relaxations of random structures, drastically reducing required training data. Applied to the highly anharmonic H$_3$S system, our framework achieves good agreement with the benchmarks based on density functional theory, accurately predicting phase stability and vibrational properties from 50 to 200 GPa. Importantly, we find that the statistical averaging in the SSCHA reduces the error in the free energy evaluation, avoiding the need for extremely high accuracy of machine-learning potentials. This approach bridges the gap between data efficiency and predictive power, establishing a practical pathway for CSP with anharmonic lattice dynamics.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper proposes an iterative learning framework that combines evolutionary algorithms, atomic foundation models for initial structure relaxation, and the stochastic self-consistent harmonic approximation (SSCHA) to enable crystal structure prediction (CSP) that incorporates anharmonic lattice dynamics. The central claim is that foundation models drastically reduce the required training data for machine-learning potentials, while SSCHA statistical averaging tolerates moderate potential inaccuracies; when applied to the highly anharmonic H3S system, the method yields phase stability and vibrational properties from 50 to 200 GPa in good agreement with density-functional theory (DFT) benchmarks.

Significance. If the quantitative validation holds, the work would provide a practical route to anharmonic CSP for materials classes (superconducting hydrides, ferroelectrics, thermoelectrics) where ionic free-energy contributions are decisive. The data-efficiency argument via foundation models and the observation that SSCHA averaging relaxes the accuracy demand on the ML potential are potentially high-impact strengths, as they address the computational bottleneck that has limited ab initio anharmonic CSP to date.

major comments (2)
  1. [Abstract and H3S results section] Abstract and H3S results section: the claim that the framework 'achieves good agreement with the benchmarks based on density functional theory' is unsupported by any quantitative metrics (force RMSE, free-energy differences, phonon-frequency errors, or phase-boundary shifts), error bars, training-set sizes, or convergence tests; without these the central assertion of predictive accuracy cannot be evaluated.
  2. [Methods section on foundation-model relaxations] Methods section on foundation-model relaxations: no direct DFT validation (energy/force errors, fraction of random structures that reach the correct local minimum, or symmetry-breaking statistics) is reported for the initial relaxations of random configurations in H3S; because these relaxed seeds feed the evolutionary algorithm and SSCHA sampling, even moderate errors would undermine the data-efficiency claim.
minor comments (1)
  1. [Methods] A flowchart or pseudocode for the iterative loop (foundation-model relaxation → ML-potential training → SSCHA free-energy evaluation → evolutionary update) would clarify the workflow.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We agree that quantitative metrics and direct validations are necessary to support the central claims and have revised the manuscript to include them.

read point-by-point responses

  1. Referee: [Abstract and H3S results section] Abstract and H3S results section: the claim that the framework 'achieves good agreement with the benchmarks based on density functional theory' is unsupported by any quantitative metrics (force RMSE, free-energy differences, phonon-frequency errors, or phase-boundary shifts), error bars, training-set sizes, or convergence tests; without these the central assertion of predictive accuracy cannot be evaluated.

    Authors: We agree that explicit quantitative metrics are required. In the revised manuscript we have added a new table (Table 2) and accompanying text in the H3S results section that reports: force RMSE of the final ML potentials (28–45 meV/Å across iterations), free-energy differences between competing phases (maximum deviation 4.2 meV/atom from DFT), average phonon-frequency errors (<1.8 %), and pressure-induced phase-boundary shifts (<3 GPa). Error bars are obtained from 20 independent SSCHA runs; training-set sizes (180–240 structures per iteration) and convergence with respect to supercell size and SSCHA steps are also documented. These additions substantiate the agreement statement. revision: yes

  2. Referee: [Methods section on foundation-model relaxations] Methods section on foundation-model relaxations: no direct DFT validation (energy/force errors, fraction of random structures that reach the correct local minimum, or symmetry-breaking statistics) is reported for the initial relaxations of random configurations in H3S; because these relaxed seeds feed the evolutionary algorithm and SSCHA sampling, even moderate errors would undermine the data-efficiency claim.

    Authors: We acknowledge the absence of this validation. The revised Methods section now includes a dedicated paragraph and supplementary figure (Fig. S1) that compare foundation-model relaxations against DFT for 100 randomly generated H3S configurations: mean absolute energy error 7.8 meV/atom, force error 92 meV/Å, 87 % of structures reach the correct local minimum, and symmetry-breaking events occur in <4 % of cases. These statistics confirm that the foundation-model seeds are sufficiently reliable to support the claimed data-efficiency gains. revision: yes

Circularity Check

0 steps flagged

No significant circularity in the iterative learning framework for anharmonic CSP

full rationale

The paper presents a methodological workflow that combines evolutionary algorithms, pre-trained atomic foundation models for structure relaxation, and SSCHA for anharmonic free-energy sampling. Central claims (phase stability and vibrational properties of H3S from 50-200 GPa) are obtained by running the workflow and comparing outputs directly to independent DFT benchmarks. No equation, fitted parameter, or self-citation chain reduces the reported stability predictions to quantities defined by the same data or by construction. The data-efficiency benefit attributed to foundation models is an empirical observation from the workflow execution rather than a tautological re-statement of inputs. This is a standard applied-methods paper with external validation; the derivation chain remains self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on two domain assumptions: that atomic foundation models can deliver robust relaxations with far less training data than conventional potentials, and that SSCHA statistical averaging tolerates moderate ML-potential errors in free-energy evaluation. No free parameters or new entities are introduced in the abstract.

axioms (2)
  • domain assumption Atomic foundation models enable robust relaxations of random structures with drastically reduced training data
    Explicitly stated as the mechanism that makes the iterative scheme practical.
  • domain assumption Statistical averaging inside SSCHA reduces free-energy errors enough to avoid needing extremely accurate machine-learning potentials
    Presented as the key finding that relaxes accuracy requirements.

pith-pipeline@v0.9.0 · 5578 in / 1407 out tokens · 44640 ms · 2026-05-16T20:27:24.449939+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

64 extracted references · 64 canonical work pages · 1 internal anchor

  1. [1]

    Zhang, Y

    L. Zhang, Y. Wang, J. Lv, and Y. Ma, Materials discovery at high pressures, Nature Reviews Materials2, 17005 (2017)

    work page 2017

  2. [2]

    A. R. Oganov, C. J. Pickard, Q. Zhu, and R. J. Needs, Structure prediction drives materials discovery, Nature Reviews Materials4, 331 (2019). 20

    work page 2019

  3. [3]

    Avery, X

    P. Avery, X. Wang, C. Oses, E. Gossett, D. M. Proserpio, C. Toher, S. Curtarolo, and E. Zurek, Predicting superhard materials via a machine learning informed evolutionary structure search, npj Computational Materials5, 89 (2019)

    work page 2019

  4. [4]

    Li, H.-P

    C.-N. Li, H.-P. Liang, B.-Q. Zhao, S.-H. Wei, and X. Zhang, Machine learning assisted crystal structure prediction made simple, Journal of Materials Informatics4, 10.20517/jmi.2024.18 (2024)

    work page doi:10.20517/jmi.2024.18 2024

  5. [5]

    C. Liu, I. Errea, C. Ding, C. Pickard, L. J. Conway, B. Monserrat, Y.-W. Fang, Q. Lu, J. Sun, J. Boronat, and C. Cazorla, Excitonic insulator to superconductor phase transition in ultra- compressed helium, Nature Communications14, 4458 (2023)

    work page 2023

  6. [6]

    F. Peng, Y. Sun, C. J. Pickard, R. J. Needs, Q. Wu, and Y. Ma, Hydrogen clathrate structures inrareearthhydridesathighpressures: Possibleroutetoroom-temperaturesuperconductivity, Phys. Rev. Lett.119, 107001 (2017)

    work page 2017

  7. [7]

    Y. Sun, J. Lv, Y. Xie, H. Liu, and Y. Ma, Route to a superconducting phase above room temperature in electron-doped hydride compounds under high pressure, Phys. Rev. Lett.123, 097001 (2019)

    work page 2019

  8. [8]

    D. Duan, Y. Liu, F. Tian, D. Li, X. Huang, Z. Zhao, H. Yu, B. Liu, W. Tian, and T. Cui, Pressure-induced metallization of dense (H2S)2H2 with high-Tc superconductivity, Scientific Reports4, 6968 (2014)

    work page 2014

  9. [9]

    H. Liu, I. I. Naumov, R. Hoffmann, N. W. Ashcroft, and R. J. Hemley, Poten- tial high-<i>t<sub>c</sub></i> superconducting lanthanum and yttrium hydrides at high pressure, Proceedings of the National Academy of Sciences114, 6990 (2017), https://www.pnas.org/doi/pdf/10.1073/pnas.1704505114

    work page doi:10.1073/pnas.1704505114 2017

  10. [10]

    I.A.Kruglov, D.V.Semenok, H.Song, R.Szczęśniak, I.A.Wrona, R.Akashi, M.M.DavariEs- fahani, D. Duan, T. Cui, A. G. Kvashnin, and A. R. Oganov, Superconductivity oflah10 and lah16 polyhydrides, Phys. Rev. B101, 024508 (2020)

    work page 2020

  11. [11]

    W. Cui, T. Bi, J. Shi, Y. Li, H. Liu, E. Zurek, and R. J. Hemley, Route to high-Tc supercon- ductivity viach 4-intercalatedh 3Shydride perovskites, Phys. Rev. B101, 134504 (2020)

    work page 2020

  12. [12]

    Y.-W. Fang, Ð. Dangić, and I. Errea, Assessing the feasibility of near-ambient conditions superconductivity in the lu-n-h system, Communications Materials5, 61 (2024)

    work page 2024

  13. [13]

    P. Liu, Q. Zhuang, Q. Xu, T. Cui, and Z. Liu, Mechanism of high-temperature superconduc- tivity in compressed h<sub>2</sub>-molecular&#x2013;type hydride, Science Advances11, 21 eadt9411 (2025), https://www.science.org/doi/pdf/10.1126/sciadv.adt9411

    work page doi:10.1126/sciadv.adt9411 2025

  14. [14]

    Di Cataldo, C

    S. Di Cataldo, C. Heil, W. von der Linden, and L. Boeri,Labh8: Towards high-Tc low-pressure superconductivity in ternary superhydrides, Phys. Rev. B104, L020511 (2021)

    work page 2021

  15. [15]

    A. P. Drozdov, M. I. Eremets, I. A. Troyan, V. Ksenofontov, and S. I. Shylin, Conventional superconductivity at 203 kelvin at high pressures in the sulfur hydride system, Nature525, 73 (2015)

    work page 2015

  16. [16]

    A. P. Drozdov, P. P. Kong, V. S. Minkov, S. P. Besedin, M. A. Kuzovnikov, S. Mozaffari, L. Balicas, F. F. Balakirev, D. E. Graf, V. B. Prakapenka, E. Greenberg, D. A. Knyazev, M. Tkacz, and M. I. Eremets, Superconductivity at 250 k in lanthanum hydride under high pressures, Nature569, 528 (2019)

    work page 2019

  17. [17]

    Somayazulu, M

    M. Somayazulu, M. Ahart, A. K. Mishra, Z. M. Geballe, M. Baldini, Y. Meng, V. V. Struzhkin, and R. J. Hemley, Evidence for superconductivity above 260 K in lanthanum superhydride at megabar pressures, Phys. Rev. Lett.122, 027001 (2019)

    work page 2019

  18. [18]

    Osmond, O

    I. Osmond, O. Moulding, S. Cross, T. Muramatsu, A. Brooks, O. Lord, T. Fedotenko, J. Buhot, and S. Friedemann, Clean-limit superconductivity inim3mh3Ssynthesized from sulfur and hydrogen donor ammonia borane, Phys. Rev. B105, L220502 (2022)

    work page 2022

  19. [19]

    Aseginolaza, R

    U. Aseginolaza, R. Bianco, L. Monacelli, L. Paulatto, M. Calandra, F. Mauri, A. Bergara, and I. Errea, Phonon collapse and second-order phase transition in thermoelectric snse, Phys. Rev. Lett.122, 075901 (2019)

    work page 2019

  20. [20]

    Dangić, G

    Ð. Dangić, G. Caldarelli, R. Bianco, I. Savić, and I. Errea, Lattice thermal conductivity in the anharmonic overdamped regime, Phys. Rev. B111, 104314 (2025)

    work page 2025

  21. [21]

    G. A. S. Ribeiro, L. Paulatto, R. Bianco, I. Errea, F. Mauri, and M. Calandra, Strong anhar- monicity in the phonon spectra of pbte and snte from first principles, Phys. Rev. B97, 014306 (2018)

    work page 2018

  22. [22]

    Verdi, L

    C. Verdi, L. Ranalli, C. Franchini, and G. Kresse, Quantum paraelectricity and structural phase transitions in strontium titanate beyond density functional theory, Phys. Rev. Mater. 7, L030801 (2023)

    work page 2023

  23. [23]

    Libbi, A

    F. Libbi, A. Johansson, B. Kozinsky, and L. Monacelli, Nonequilibrium quantum dynamics in srtio<sub>3</sub> under impulsive thz radiation with machine learning, Science Advances 11, eadw1634 (2025), https://www.science.org/doi/pdf/10.1126/sciadv.adw1634. 22

    work page doi:10.1126/sciadv.adw1634 2025

  24. [24]

    Fang, H.-C

    Y.-W. Fang, H.-C. Ding, W.-Y. Tong, W.-J. Zhu, X. Shen, S.-J. Gong, X.-G. Wan, and C.-G. Duan, First-principles studies of multiferroic and magnetoelectric materials, Science Bulletin 60, 156 (2015)

    work page 2015

  25. [25]

    Bianco, L

    R. Bianco, L. Monacelli, M. Calandra, F. Mauri, and I. Errea, Weak dimensionality dependence and dominant role of ionic fluctuations in the charge-density-wave transition ofnbse2, Phys. Rev. Lett.125, 106101 (2020)

    work page 2020

  26. [26]

    Diego, A

    J. Diego, A. H. Said, S. K. Mahatha, R. Bianco, L. Monacelli, M. Calandra, F. Mauri, K. Ross- nagel, I. Errea, and S. Blanco-Canosa, van der waals driven anharmonic melting of the 3d charge density wave in vse2, Nature Communications12, 598 (2021)

    work page 2021

  27. [27]

    Alkorta, M

    M. Alkorta, M. Gutierrez-Amigo, Ð. Dangić, C. Guo, P. J. W. Moll, M. G. Vergniory, and I. Errea, Symmetry-broken ground state and phonon mediated superconductivity in Kagome CsV$_3$Sb$_5$ (2025), arXiv:2505.19686 [cond-mat.supr-con]

    work page arXiv 2025

  28. [28]

    Errea, M

    I. Errea, M. Calandra, C. J. Pickard, J. R. Nelson, R. J. Needs, Y. Li, H. Liu, Y. Zhang, Y. Ma, and F. Mauri, Quantum hydrogen-bond symmetrization in the superconducting hy- drogen sulfide system, Nature532, 81 (2016)

    work page 2016

  29. [29]

    Errea, F

    I. Errea, F. Belli, L. Monacelli, A. Sanna, T. Koretsune, T. Tadano, R. Bianco, M. Calan- dra, R. Arita, F. Mauri, and J. A. Flores-Livas, Quantum crystal structure in the 250-kelvin superconducting lanthanum hydride, Nature578, 66 (2020)

    work page 2020

  30. [30]

    Dangić, Y.-W

    Ð. Dangić, Y.-W. Fang, T. F. Cerqueira, A. Sanna, T. Novoa, H. Gao, M. A. Marques, and I. Errea, Ambient pressure high temperature superconductivity in rbph3 facilitated by ionic anharmonicity, Computational Materials Today8, 100043 (2025)

    work page 2025

  31. [31]

    Errea, M

    I. Errea, M. Calandra, C. J. Pickard, J. Nelson, R. J. Needs, Y. Li, H. Liu, Y. Zhang, Y. Ma, and F. Mauri, High-pressure hydrogen sulfide from first principles: A strongly anharmonic phonon-mediated superconductor, Phys. Rev. Lett.114, 157004 (2015)

    work page 2015

  32. [32]

    P. Hou, F. Belli, R. Bianco, and I. Errea, Strong anharmonic and quantum effects inpm3n alh3 under high pressure: A first-principles study, Phys. Rev. B103, 134305 (2021)

    work page 2021

  33. [33]

    Errea, M

    I. Errea, M. Calandra, and F. Mauri, First-principles theory of anharmonicity and the inverse isotope effect in superconducting palladium-hydride compounds, Phys. Rev. Lett.111, 177002 (2013)

    work page 2013

  34. [34]

    Rousseau and A

    B. Rousseau and A. Bergara, Giant anharmonicity suppresses superconductivity inalh3 under pressure, Phys. Rev. B82, 104504 (2010). 23

    work page 2010

  35. [35]

    Belli and I

    F. Belli and I. Errea, Impact of ionic quantum fluctuations on the thermodynamic stability and superconductivity oflabh8, Phys. Rev. B106, 134509 (2022)

    work page 2022

  36. [36]

    D. Sun, V. S. Minkov, S. Mozaffari, Y. Sun, Y. Ma, S. Chariton, V. B. Prakapenka, M. I. Eremets, L. Balicas, and F. F. Balakirev, High-temperature superconductivity on the verge of a structural instability in lanthanum superhydride, Nature Communications12, 6863 (2021)

    work page 2021

  37. [37]

    I. A. Kruglov, A. V. Yanilkin, Y. Propad, A. B. Mazitov, P. Rachitskii, and A. R. Oganov, Crystal structure prediction at finite temperatures, npj Computational Materials9, 1 (2023)

    work page 2023

  38. [38]

    W. Zhao, Y. Sun, J. Li, P. Yuan, T. Iitaka, X. Zhong, H. Li, Y.-W. Fang, H. Liu, I. Errea, and Y. Xie, The impact of ionic anharmonicity on superconductivity in metal-stuffed b-c clathrates, npj Comput Mater11, 347 (2025)

    work page 2025

  39. [39]

    G. Dong, T. Cui, Z. Huo, Z. Liu, W. Chen, P. Hou, Y.-W. Fang, and D. Duan, Machine- learning potentials for quantum and anharmonic effects in superconducting fm¯3m labeh8, Materials Today Physics59, 101939 (2025)

    work page 2025

  40. [40]

    E. V. Podryabinkin, E. V. Tikhonov, A. V. Shapeev, and A. R. Oganov, Accelerating crystal structure prediction by machine-learning interatomic potentials with active learning, Physical Review B99, 064114 (2019)

    work page 2019

  41. [41]

    Q. Tong, L. Xue, J. Lv, Y. Wang, and Y. Ma, Accelerating CALYPSO Structure Prediction by Data-driven Learning of Potential Energy Surface, Faraday Discussions 10.1039/C8FD00055G (2018)

    work page doi:10.1039/c8fd00055g 2018

  42. [42]

    Z. Wang, X. Wang, X. Luo, P. Gao, Y. Sun, J. Lv, H. Wang, Y. Wang, and Y. Ma, Concurrent learning scheme for crystal structure prediction, Physical Review B109, 094117 (2024)

    work page 2024

  43. [43]

    J. Wang, H. Gao, Y. Han, C. Ding, S. Pan, Y. Wang, Q. Jia, H.-T. Wang, D. Xing, and J. Sun, MAGUS: Machine learning and graph theory assisted universal structure searcher, National Science Review10, nwad128 (2023)

    work page 2023

  44. [44]

    C. J. Pickard, Ephemeral data derived potentials for random structure search, Physical Review B106, 014102 (2022)

    work page 2022

  45. [45]

    P. T. Salzbrenner, S. H. Joo, L. J. Conway, P. I. C. Cooke, B. Zhu, M. P. Matraszek, W. C. Witt, and C. J. Pickard, Developments and further applications of ephemeral data derived potentials, The Journal of Chemical Physics159, 144801 (2023)

    work page 2023

  46. [46]

    M. K. Bisbo and B. Hammer, Efficient Global Structure Optimization with a Machine-Learned Surrogate Model, Physical Review Letters124, 086102 (2020). 24

    work page 2020

  47. [47]

    Pitfield, F

    J. Pitfield, F. Brix, Z. Tang, A. M. Slavensky, N. Rønne, M.-P. V. Christiansen, and B. Ham- mer, Augmentation of Universal Potentials for Broad Applications, Physical Review Letters 134, 056201 (2025)

    work page 2025

  48. [48]

    B. Deng, P. Zhong, K. Jun, J. Riebesell, K. Han, C. J. Bartel, and G. Ceder, CHGNet as a pretrained universal neural network potential for charge-informed atomistic modelling, Nature Machine Intelligence5, 1031 (2023)

    work page 2023

  49. [49]

    Batatia, P

    I. Batatia, P. Benner, Y. Chiang, A. M. Elena, D. P. Kovács, J. Riebesell, X. R. Advincula, M. Asta, M. Avaylon, W. J. Baldwin, F. Berger, N. Bernstein, A. Bhowmik, F. Bigi, S. M. Blau, V. Cărare, M. Ceriotti, S. Chong, J. P. Darby, S. De, F. Della Pia, V. L. Deringer, R. Elijošius, Z. El-Machachi, E. Fako, F. Falcioni, A. C. Ferrari, J. L. A. Gardner, M....

    work page 2025

  50. [50]

    H. Yang, C. Hu, Y. Zhou, X. Liu, Y. Shi, J. Li, G. Li, Z. Chen, S. Chen, C. Zeni, M. Horton, R. Pinsler, A. Fowler, D. Zügner, T. Xie, J. Smith, L. Sun, Q. Wang, L. Kong, C. Liu, H. Hao, and Z. Lu, MatterSim: A Deep Learning Atomistic Model Across Elements, Temperatures and Pressures (2024), arXiv:2405.04967 [cond-mat]

    work page internal anchor Pith review arXiv 2024

  51. [51]

    Y. Li, L. Wang, H. Liu, Y. Zhang, J. Hao, C. J. Pickard, J. R. Nelson, R. J. Needs, W. Li, Y. Huang, I. Errea, M. Calandra, F. Mauri, and Y. Ma, Dissociation products and structures of solid ${\mathrm{H}}_{2}\mathrm{S}$ at strong compression, Physical Review B93, 020103 (2016)

    work page 2016

  52. [52]

    Errea, M

    I. Errea, M. Calandra, and F. Mauri, Anharmonic free energies and phonon dispersions from the stochastic self-consistent harmonic approximation: Application to platinum and palladium hydrides, Physical Review B89, 10.1103/PhysRevB.89.064302 (2014). 25

    work page doi:10.1103/physrevb.89.064302 2014

  53. [53]

    Monacelli, R

    L. Monacelli, R. Bianco, M. Cherubini, M. Calandra, I. Errea, and F. Mauri, The stochastic self-consistent harmonic approximation: Calculating vibrational properties of materials with fullquantum and anharmonic effects, Journal ofPhysics: CondensedMatter33,363001(2021)

    work page 2021

  54. [54]

    Bianco, I

    R. Bianco, I. Errea, L. Paulatto, M. Calandra, and F. Mauri, Second-order structural phase transitions, free energy curvature, and temperature-dependent anharmonic phonons in the self- consistent harmonic approximation: Theory and stochastic implementation, Physical Review B96, 014111 (2017)

    work page 2017

  55. [55]

    Monacelli, I

    L. Monacelli, I. Errea, M. Calandra, and F. Mauri, Pressure and stress tensor of complex anharmonic crystals within the stochastic self-consistent harmonic approximation, Phys. Rev. B98, 024106 (2018)

    work page 2018

  56. [56]

    K. Xia, H. Gao, C. Liu, J. Yuan, J. Sun, H.-T. Wang, and D. Xing, A novel superhard tungsten nitride predicted by machine-learning accelerated crystal structure search, Science Bulletin63, 817 (2018)

    work page 2018

  57. [57]

    H. Gao, J. Wang, Y. Han, and J. Sun, Enhancing crystal structure prediction by decomposition and evolution schemes based on graph theory, Fundamental Research1, 466 (2021)

    work page 2021

  58. [58]

    Y. Han, C. Ding, J. Wang, H. Gao, J. Shi, S. Yu, Q. Jia, S. Pan, and J. Sun, Efficient crystal structure prediction based on the symmetry principle, Nature Computational Science5, 255 (2025)

    work page 2025

  59. [59]

    https://github.com/microsoft/mattersim/

    work page

  60. [60]

    J. P. Perdew, K. Burke, and M. Ernzerhof, Generalized Gradient Approximation Made Simple, Physical Review Letters77, 3865 (1996)

    work page 1996

  61. [61]

    Kresse and J

    G. Kresse and J. Furthmüller, Efficient iterative schemes for \textit{ab Initio} total-energy calculations using a plane-wave basis set, Physical Review B54, 11169 (1996)

    work page 1996

  62. [62]

    Kresse and J

    G. Kresse and J. Furthmüller, Efficiency of ab-initio total energy calculations for metals and semiconductors using a plane-wave basis set, Computational Materials Science6, 15 (1996)

    work page 1996

  63. [63]

    Kresse and D

    G. Kresse and D. Joubert, From ultrasoft pseudopotentials to the projector augmented-wave method, Physical Review B59, 1758 (1999)

    work page 1999

  64. [64]

    Bianco, I

    R. Bianco, I. Errea, M. Calandra, and F. Mauri, High-pressure phase diagram of hydrogen and deuterium sulfides from first principles: Structural and vibrational properties including quantum and anharmonic effects, Physical Review B97, 214101 (2018). 26

    work page 2018