Beyond Adam: SOAP and Muon for Faster, Label-Efficient Training of Machine Learning Interatomic Potentials

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Gil Harari, Yoel Zimmermann, Ola Tangen Kulseng, Laura Zichi, Chuin Wei Tan, Marc L. Descoteaux, Boris Kozinsky
• Venue: arXiv: cs.LG
• Published: 2026-07-02
• Collected: 2026-07-03
• URL: http://arxiv.org/abs/2607.02499v1
• PDF: https://arxiv.org/pdf/2607.02499v1
• DOI: UNKNOWN

Plain-Language Brief

A study compares various matrix-structured optimizers, including SOAP, Muon, and a hybrid of the two, with Adam and its variants for training machine learning interatomic potential (MLIP) models. The new optimizers are found to outperform Adam in terms of speed and accuracy, particularly under partial force supervision.

What the Document Says

• Machine learning interatomic potentials (MLIPs) are a significant aspect of AI for scientific simulation.
• The study investigates a class of matrix-structured optimizers like Muon, SOAP, and the hybrid SOAP-Muon for training NequIP and Allegro MLIP models.
• Adam and its variants have been the standard choice for optimizers in the community.
• The new optimizers are found to outperform Adam in both convergence speed and final accuracy.
• SOAP and SOAP-Muon are robust and consistently strong methods.
• Muon only provides partial gains relative to Adam.
• The improvements are particularly pronounced under partial force supervision.
• The authors suggest that optimizer choice is an overlooked yet impactful design axis for MLIPs.

Key Concepts Mentioned

• Machine learning interatomic potential
• MLIP models
• Optimizers
• Adam
• SOAP
• Muon
• SOAP-Muon
• Convergence speed
• Final accuracy
• Partial force supervision

Claims Made by the Document

• New optimizers outperform Adam in terms of convergence speed and final accuracy.
• SOAP and SOAP-Muon are robust and consistently strong methods for training MLIP models.
• Muon only provides partial gains relative to Adam.

Evidence Used

• Results from the study comparing various optimizers for training NequIP and Allegro MLIP models.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Machine learning interatomic potentials (MLIPs) have become a hallmark of AI for scientific simulation. While efforts on new architectures and datasets have led to increasingly accurate and general models, the choice of optimizer for training has largely remained unexplored, defaulting to Adam and its variants in the community. Here, we implement and systematically compare a class of recently proposed matrix-structured optimizers, including Muon, SOAP, and the hybrid SOAP-Muon, for training NequIP and Allegro MLIP models. We find that these optimizers can substantially outperform Adam in both convergence speed and final accuracy. SOAP and SOAP-Muon emerge as robust and consistently strong methods, while Muon only provides partial gains relative to Adam. The improvements are particularly pronounced under partial force supervision. Our results indicate that optimizer choice is an overlooked yet impactful design axis for MLIPs.