aiqbench analysis
This research investigates how training objectives influence the scalability of machine learning surrogates for AC optimal power flow, revealing that physics-aware methods like the augmented Lagrangian significantly reduce constraint violations as grid size increases. The study demonstrates that while standard mean squared error training scales primarily with model capacity, incorporating physical laws balances performance gains across both model and data dimensions, offering a more robust path to real-world deployment in energy systems.
Scaling Laws for Physics-Aware ACOPF Surrogate Learning
arXiv:2609.16282v1 Announce Type: new Abstract: Learning-based surrogates for AC optimal power flow (ACOPF) promise large speedups over classical solvers, but their operational value depends on physical feasibility as much as predictive accuracy. Physics-aware objectives such as the augmented Lagrangian (AL) improve constraint satisfaction at additional per-step cost, yet how this trade-off behaves with scale is uncharacterized. We sweep model and dataset sizes under both MSE and AL training, and characterize how constraint violation changes with network size across grids. Both objectives improve as power laws, but at different rates: MSE is governed primarily by model capacity, while AL is balanced across both. Violation grows roughly twice as fast with network size under MSE as under AL. On matched hardware, AL reduces violation by nearly $30\times$ for an order of magnitude more training time, with negligible added memory. The training objective determines not only where a surrogate lands but how its quality evolves with scale.
Abstract:Learning-based surrogates for AC optimal power flow (ACOPF) promise large speedups over classical solvers, but their operational value depends on physical feasibility as much as predictive accuracy. Physics-aware objectives such as the augmented Lagrangian (AL) improve constraint satisfaction at additional per-step cost, yet how this trade-off behaves with scale is uncharacterized. We sweep model and dataset sizes under both MSE and AL training, and characterize how constraint violation changes with network size across grids. Both objectives improve as power laws, but at different rates: MSE is governed primarily by model capacity, while AL is balanced across both. Violation grows roughly twice as fast with network size under MSE as under AL. On matched hardware, AL reduces violation by nearly $30\times$ for an order of magnitude more training time, with negligible added memory. The training objective determines not only where a surrogate lands but how its quality evolves with scale.
Machine Learning (cs.LG)
arXiv:2609.16282 [cs.LG]
(or arXiv:2609.16282v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2609.16282 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 243 words
Related coverage
Lantern: Conflict-Aware Gradient Blending for Physics-Guided Diffusion Models in Calorimeter Simulation
arXiv · Jul 29, 2026
PIRAMID: Progress and Plans
LessWrong · Jul 27, 2026
Predict before you train: Scaling Laws for particle physics foundation models
arXiv · Jul 31, 2026
Synthesis Superintelligence: from Semiconductors to Superconductors — Periodic Labs’ Liam Fedus and Ekin Dogus Cubuk
Latent Space · Oct 8, 2026