Searcharxiv⌕ Search

arXiv · 2611.13898

Orchestr.AI: A Unified Workflow for Training and Benchmarking of Machine-Learned Interatomic Potentials

Abstract

Machine-learning force fields enable atomistic simulations at near-reference accuracy and substantially reduced computational cost, but their development remains fragmented across software frameworks with incompatible data formats, configuration schemes, dependencies, and execution procedures. Orchestr.AI addresses this limitation through a unified workflow that coordinates different MLIP software frameworks, referred to as engines, for data preparation, model training, fine-tuning, evaluation, and deployment. A common YAML interface defines each experiment, while engine-specific environment profiles preserve compatible software stacks and avoid dependency conflicts. The platform currently supports five MLIP model architectures, namely SchNet, PaiNN, NequIP, Allegro, and MACE, through their corresponding software implementations, providing automated configuration mapping and standardized dataset handling across these frameworks. Benchmarks on the rMD17 and 3BPA datasets show that Orchestr.AI reproduces the expected performance trends while enabling controlled cross-engine comparisons. Repeated executions with fixed configurations and engine profiles produce highly consistent results, demonstrating practical workflow reproducibility. Beyond training, Orchestr.AI provides modular architectural interfaces that connect trained models to ASE-based simulations, batch inference, epistemic uncertainty estimation, calibration, and active-learning workflows. The platform therefore provides a reproducible and extensible infrastructure for developing, comparing, and deploying machine-learning interatomic potentials, which we validate here through cross-engine accuracy and repeatability benchmarks.

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Zain Ul Abideen, Mario Fernández-Pendás, Muhammad Usman, Camilo Rodríguez Quintero, Masuma Suleymanova, Ignacio Arganda-Carreras, Ivan Infante. 2026-11-10. Orchestr.AI: A Unified Workflow for Training and Benchmarking of Machine-Learned Interatomic Potentials. https://searcharxiv.com/abs/2611.13898

Citations retain the original authors and DOI and include this site’s record URL. If this discovery service contributed to your research, you can also cite Searcharxiv by Indiciae: BibTeX · RIS.

KEEP EXPLORING

Related papers

Evidence for crystal-field-selected octupolar order in wolframite oxides

Magnetic order can involve higher-rank moments such as octupoles, which are difficult to identify in strongly distorted crystals. Here we present evidence for predominantly octupolar order in the monoclinic wolframites MgReO$_4$ and ZnReO$_4$. Muon spin spectroscopy detects magnetic transitions near 83 and 126 K, respectively, with similar precession frequencies in both compounds. An ab initio effective Hamiltonian shows that strong crystal fields split the spin--orbit quartet, isolating low-energy doublets with weak dipolar character. Exchange selects octupole-dominated order in distinct collinear antiferromagnetic structures, with moments along $b$ in MgReO$_4$ and tilted within the $ac$ plane in ZnReO$_4$. The estimated dipole moments of 0.06 and 0.12 $μ_{\mathrm{B}}$ per Re, respectively, are consistent with the absence of a resolved magnetic contribution in neutron powder diffraction. A single-ion model illustrates how octupolar states generate local fields with a vanishing net dipole moment. These findings support a crystal-field route to hidden magnetism beyond double perovskites.

cond-mat.mtrl-sci↗

Substrate Role in Polaron Formation on Single-layer Transition Metal Dihalides

Single-layer transition metal dihalides grown on conducting substrates were shown to host stable polarons. Here, we investigate small polarons in a single layer of the insulator MnBr$_2$, which was grown by molecular beam epitaxy on three different substrates: graphene on Ir(110), graphene on Ir(111), and Au(111). Polarons on these substrates can be created, converted, and moved by the STM tip when a tunneling current flows at a proper bias voltage. Four polaron species are observed in MnBr$_2$ grown on graphene Ir(110), three of which resemble the species previously observed for CoCl$_2$ on graphite, whereas MnBr$_2$ grown on the other substrates exhibits only two of these four species. Changing the substrate from graphene on Ir to Au(111) increases polaron number density by an order of magnitude. These large differences demonstrate that the choice of conducting substrate plays a decisive role in determining polaron phenomenology in substrate-supported single-layer insulators. Two unique observations are reported for polarons in MnBr$_2$ on graphene supported by Ir(110): First, mobile polarons are guided by the periodic potential originating from the super-moiré of MnBr$_2$ with graphene and Ir(110); second, polarons remain stable up to 300\,K.

cond-mat.mtrl-sci↗

Systematic comparison of approximations and functionals in first-principle calculations of aluminum-based III-V ferroelectric nitrides

We revisit first-principles predictions of structural, ferroelectric, and electronic properties in aluminum-based III-V nitride alloys, focusing on Al1-xScxN and Al1-xBxN. Using density functional theory within a unified 48-atom supercell framework, we systematically assess the role of chemical disorder and exchange-correlation approximations by comparing the virtual crystal approximation (VCA) and special quasirandom structures (SQS), as well as PBE, PBESol, SCAN, and SCAN+rVV10 functionals. We demonstrate that, even amongst the similar PBE and PBESol functionals, large quantitative and qualitative differences emerge. In particular, VCA or SQS PBESol modeling (a popular functional) underestimates the stability domain of the ferroelectric wurtzite phase in Al1-xScxN compared to SQS PBE or SQS SCAN. We demonstrate that the 5-fold coordinated hexagonal phase predicted in 2002 by Farrer and Bellaiche [Phys. Rev. B 66, 201203] is a metastable state whose energy lies between that of the four-fold coordinated ferroelectric wurtzite phase and the energy of the six-fold coordinated rocksalt phase below the transition point. In contrast, Al1-xBxN shows a much faster destabilization of the wurtzite ferroelectric phase, with bond breaking which strongly distorts the wurtzite structure (with enhanced polarization) and eventually favors a zincblende phase and a threefold coordinated hexagonal layer phase. Our analysis highlights the critical importance of both local disorder and exchange-correlation treatment in the quantitative prediction of the functional properties of III-V nitride ferroelectrics.

cond-mat.mtrl-sci↗