Searcharxiv⌕ Search

arXiv · 2611.16106

Machine learning and high accuracy: Neural preconditioners for computational gravity

Abstract

Machine learning (ML) developments have empowered valuable calculations in many areas of theoretical physics. Certain topics involving moderate-to-low accuracy requirements have been particularly fortunate because many ML techniques, such as neural networks, introduce a noise floor that would be difficult to penetrate with stricter tolerances. For this reason, topics requiring high accuracy can be hesitant to pursue ML. One prominent example is gravitational wave modeling, which relies on traditional methods that accommodate the high sensitivities of detectors such as LISA. However, recent progress in applied mathematics has identified hybrid methods where fast-but-noisy ML mechanisms are inserted into otherwise traditional numerical schemes to accelerate convergence in a way that ultimately meets or exceeds the accuracy of standard methods. We develop an optimized hybrid ML method to tackle a class of challenging PDEs; specifically, we demonstrate ML implementations for gravitational waves from asymmetric-mass black hole binaries and elastic waves. These require solving sparse linear algebra systems with Helmholtz features that are especially difficult for traditional iterative methods. Our new approach is to train a graph neural network (GNN) as a significantly faster preconditioner using a novel loss function based on the true residual of the iterative algorithm. For the most difficult regions of parameter space, our neural preconditioner decreases iteration counts by 2-15x compared to traditional preconditioners. Because the GNN operates directly on the sparse matrix and needs only modest physical context, it generalizes beyond its training domain and can be adapted to new problems across physics by retraining rather than redesign. Hybrid ML methods such as this offer a new and compelling avenue to handle high accuracy and speed requirements in gravitational wave modeling and beyond.

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Thomas Osburn, Nils Vu. 2026-11-11. Machine learning and high accuracy: Neural preconditioners for computational gravity. https://searcharxiv.com/abs/2611.16106

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

Quantum-information bounds on black-hole evaporation and de Sitter expansion

It is well-known that the unitary nature of black hole evaporation enforces its entropy evolution to follow the Page curve. In this Letter, we further explore the outcome of the quantum speed limit on how fast a quantum system can evolve unitarily. When combined with the maximal information transmission bound, we place upper limits on the evaporation rates of dynamical black-hole entropy and mass that are otherwise divergent at the very end of the Hawking evaporation, during which the Penrose inequality from the cosmic censorship conjecture is consistently obeyed. Further applying the maximal information transmission bound to an effective-field-theory description for all sub-horizon regions of a de Sitter space would sufficiently reproduce a weaker trans-Planckian censorship conjecture up to a numeric prefactor conditionally.

gr-qc↗

Inflation in unimodular loop quantum cosmology

We study inflation in the setting of unimodular loop quantum cosmology, where time evolution is defined in unimodular time rather than with respect to a free, massless scalar field as is standard in loop quantum cosmology. The unimodular setting leads to a natural Schrödinger time evolution in a time coordinate with clear geometric meaning, defined independently of any particular matter content; an inflaton can be included but is not needed as a clock. We review the unimodular version of loop quantum cosmology and comment on possible connections to full (unimodular) loop quantum gravity. Then, focusing on semiclassical effective equations, we derive analytical solutions in simple cases such as a constant potential, emphasising the use of a unimodular time coordinate. We also discuss numerical solutions for phenomenologically interesting cases such as a quadratic potential and Starobinsky inflation, comparing different possible choices of initial conditions. In particular, we show that choosing an $α$-attractor potential allows for models of a bounce either dominated by kinetic or potential energy, which are compatible with observations while potentially including observable imprints of the quantum-gravity regime.

gr-qc↗

Quadratic effective energy--momentum tensor on uniform-density hypersurfaces during slow-roll inflation

We investigate the quadratic-order effective energy-momentum tensor (2EMT) of scalar cosmological perturbations on uniform-density hypersurfaces during slow-roll inflation. The 2EMT is a gauge-fixed effective source rather than a gauge-invariant observable. We impose the complete scalar gauge conditions $δρ=0$ and $E=0$, express all perturbations in terms of the Bardeen potential $Ψ$, and distinguish the strict infrared and ultraviolet limits from the intermediate regimes. The uniform-density time shift differs from the comoving one by an explicitly gradient-suppressed term, $Q_{\mathrm{UD}}-Q_{\mathrm{C}}=-ΔΨ/[3{\mathcal H}({\mathcal H}'-{\mathcal H}^2)]$. The two sources vanish at zero gradient, while their first nonzero infrared coefficients differ and give $w_{\rm raw}\simeq-1/3$ at leading slow-roll order. In the intermediate infrared, all four gauges have this leading ratio, with gauge-dependent amplitudes. In the ultraviolet, the longitudinal, spatially-flat, and comoving gauges give $w_{\rm raw}\simeq1/3$, whereas the uniform-density gauge gives $w_{\rm raw}\simeq1$. The common $1/ε$ enhancement of the matter-clock gauges reflects the slow evolution of their clocks; the additional uniform-density $1/σ_2^2$ enhancement originates from the Laplacian contribution to the density constraint. Neither enhancement is interpreted as a gauge-independent physical instability, and the gauge reconstruction must remain perturbatively small.

gr-qc↗