⚛️ general relativity

Neural-Spectral Discovery of Rotating Black Holes Beyond General Relativity

The paper introduces {\sc Akribeia}, a hybrid framework combining physics-informed neural networks with pseudo-spectral refinement to generate and certify precise, continuous rotating black hole solutions in higher-curvature gravity theories, enabling the first construction of such solutions with multiple angular momenta in non-perturbative regimes.

Original authors: Felipe Agurto-Sepúlveda, Marcelo Oyarzo, Anxo Biasi, Devansh Agarwal, Ethan Tregidga, James F. Steiner, Jose D. Edelstein, Gaston Giribet, Cecilia Garraffo

Published 2026-07-09

📖 4 min read🧠 Deep dive

CC BY 4.0

Original authors: Felipe Agurto-Sepúlveda, Marcelo Oyarzo, Anxo Biasi, Devansh Agarwal, Ethan Tregidga, James F. Steiner, Jose D. Edelstein, Gaston Giribet, Cecilia Garraffo

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine trying to find a specific, perfect shape in a vast, foggy mountain range. In the world of physics, this "mountain range" is the universe, and the "shape" is a spinning black hole. For decades, scientists have been able to map the mountains of General Relativity (our current best theory of gravity), finding the famous "Kerr" black hole. But when they try to add new, complex ingredients to the theory—ingredients that might explain how gravity works at the quantum level—the map becomes a tangled, foggy mess. The equations become so twisted and non-linear that traditional math tools get stuck, unable to find the path to a solution.

This paper introduces a new team of explorers called Akribeia to solve this problem. Here is how they do it, broken down into simple steps:

1. The Problem: Getting Lost in the Fog

Think of finding a rotating black hole in these new, complex theories as trying to navigate a maze where the walls move. Traditional methods are like a hiker who needs a very specific starting point (a "seed") to begin. If the hiker doesn't know exactly where to start, they can't find the exit. For a long time, scientists didn't have these starting points for spinning black holes in these advanced theories, so the solutions remained hidden.

2. The Solution: A Two-Step Discovery Team

The authors created a hybrid framework called Akribeia that combines two different tools to navigate this maze:

Step 1: The AI Scout (Physics-Informed Neural Networks)
Imagine sending out a smart drone (an AI) that doesn't need a map. Instead, it flies around randomly, learning the "rules of the terrain" (the laws of physics) as it goes. It doesn't need a perfect starting point; it just starts guessing and adjusts its flight path until it finds a general shape that fits the rules. This is the "exploratory" phase. It finds a rough, continuous path through the fog.
Step 2: The Master Cartographer (Pseudo-Spectral Refinement)
Once the drone finds a rough path, a master cartographer takes over. This cartographer uses high-precision tools (like a super-accurate ruler and calculator) to trace over the drone's path. They smooth out every bump and ensure the lines are mathematically perfect. This step turns the "rough sketch" into a "certified, high-definition map."

3. The Discovery: New Spinning Shapes

Using this team, the scientists successfully found rotating black holes in theories that include "quadratic" (squared) and "cubic" (cubed) curvature terms.

The "Quadratic" Test: They first tested their method on a known 5-dimensional universe. The AI found the path, and the cartographer polished it. The result matched all previous known solutions perfectly, proving the team works.
The "Cubic" Breakthrough: Then, they went into uncharted territory: a 7-dimensional universe with "cubic" gravity terms. No one had ever found a spinning black hole solution here before. The Akribeia team found them! They discovered families of black holes that spin with multiple angular momenta, all described by smooth, continuous functions.

4. What They Found

The new black holes look a bit different from the classic ones:

The "Squashing" Effect: The shape of the black hole's horizon (its event horizon) changes. In some cases, it gets "squashed" or stretched in a way that suggests it might be more stable than previously thought.
Stronger "Frame Dragging": Imagine spinning a bowl of honey; the honey near the spoon spins faster and pulls the surrounding honey with it. The new black holes spin the fabric of space-time even more vigorously than standard black holes. This "dragging" effect is stronger, which would change how nearby objects orbit.

5. Why It Matters (According to the Paper)

The paper claims this method is a game-changer because:

No "Seeds" Needed: You don't need to know the answer beforehand to find it. The AI can start from random guesses.
Continuous Maps: Instead of finding one solution at a time, they can generate a whole "family" of solutions that change smoothly as you tweak the physics parameters.
Future Observations: While the paper doesn't claim to have solved a specific medical or engineering problem yet, it states that having these "certified maps" is crucial for future telescopes (like the Event Horizon Telescope) and gravitational wave detectors. When these instruments look at real black holes, scientists will need these new maps to tell if the black holes they see are the standard kind or something exotic from a new theory of gravity.

In short, the paper presents a new "AI + Math" toolkit that successfully navigated a previously impossible mathematical maze, revealing the first-ever detailed maps of spinning black holes in complex, higher-dimensional gravity theories.

Technical Summary: Neural-Spectral Discovery of Rotating Black Holes Beyond General Relativity

Problem Statement
Finding stationary rotating black hole solutions in higher-curvature theories of gravity (beyond General Relativity) is a fundamental challenge. While virtually all approaches to quantum gravity predict corrections to the Einstein-Hilbert action, no systematic method exists to generate stationary solutions for these theories. Traditional numerical approaches, such as Newton-type iterative schemes, rely heavily on accurate initial seeds derived from perturbative regimes or specific limits. In the absence of such seeds, generating initial approximations for highly non-linear, non-perturbative coupled systems of ordinary differential equations (ODEs) is a significant bottleneck. Consequently, a systematic, parametric catalogue of rotating black hole solutions across beyond-GR theories remains unavailable, hindering the confrontation of these theories with upcoming observational data from facilities like the Event Horizon Telescope (EHT) and LIGO-Virgo-KAGRA.

Methodology: The Akribeia Framework
The authors introduce Akribeia, a novel hybrid framework designed to close this gap. It combines Physics-Informed Neural Networks (PINNs) with a pseudo-spectral refinement step to produce "certified neural-field" solutions. The methodology proceeds in two distinct stages:

PINN Discovery Stage:
- Representation: Unknown metric functions are represented as continuous, differentiable neural fields parametrized by network weights ( $W$ ).
- Training: The network is trained by minimizing the mean squared residuals of the field equations and boundary conditions over a set of collocation points. Exact derivatives are computed via automatic differentiation.
- Initialization: Crucially, this stage operates from random initializations, eliminating the need for perturbative seeds or parameter continuation procedures.
- Optimization Strategy: The training follows a four-stage protocol: (i) Adam optimizer with learning rate warm-up; (ii) stable learning rate for broad landscape exploration; (iii) adaptive learning rate decay for refinement; and (iv) L-BFGS quasi-Newton optimization to reduce residuals to $\sim 10^{-3}$ .
- Output: A continuous family of approximate solutions across the parameter space.
Pseudo-Spectral Refinement Stage:
- The trained neural network output serves as the initial seed for a classic pseudo-spectral Chebyshev solver equipped with extended-precision arithmetic and a Gauss-Newton optimizer.
- This stage rapidly reduces residuals far below machine double-precision, yielding high-precision approximations expressible as Chebyshev polynomials.
- This hybrid approach leverages the exploratory power of machine learning to find solution branches and the precision of spectral methods to certify them.

Key Contributions

First Rotating Solutions in Cubic Lovelock Gravity: The paper presents the first fully non-linear, stationary rotating black hole solutions in cubic Lovelock theory (specifically in $D=7$ dimensions) with arbitrary coupling constants and equal angular momenta.
Parametric Families in Quadratic Lovelock (EGB): The method constructs continuous families of rotating solutions in 5D Einstein-Gauss-Bonnet (EGB) gravity, parametrized by the coupling constant $\alpha$ , without relying on perturbative seeds.
Certified Neural Fields: The authors provide continuous, globally defined functions that are differentiable to any order, allowing for the computation of exact derivatives and residuals at arbitrary points.
Validation: The framework successfully recovers the known Myers-Perry rotating black hole solution in the General Relativity limit with a relative error of $10^{-160}$ using 300-digit extended-precision arithmetic.

Results

5D EGB Theory: The method generated a continuous family of solutions for coupling constants $\alpha \in [0, 3]$ $α \in [0, 3]$ . The results matched existing numerical solutions for $\alpha=3$ $α = 3$ and the exact Myers-Perry solution for $\alpha=0$ $α = 0$ .
- Physical Observations: The inclusion of quadratic curvature terms with rotation leads to a "squashing" of the horizon ( $h(r_H)/r_H^2 < 1$ ) for couplings above a critical value $\alpha^*$ . The frame-dragging function $w(r)$ increases monotonically with $\alpha$ , suggesting enhanced Lense-Thirring precession rates.
7D Cubic Lovelock Theory: The authors solved the highly non-linear ODE system for cubic curvature terms, varying the cubic coupling $\beta \in [0, 0.5]$ $β \in [0, 0.5]$ while fixing the quadratic coupling.
- Physical Observations: The metric functions exhibit qualitatively similar behavior to the quadratic case. The solutions approach the analytic Myers-Perry solution in the limit where higher-curvature couplings vanish.
Performance: The hybrid method achieved unprecedented precision for arbitrary coupling values, overcoming the initialization sensitivity that typically plagues spectral solvers in these regimes.

Significance and Claims
The paper claims that Akribeia provides a systematic, parametric approach to constructing rotating black hole solutions in higher-curvature gravity, a task previously limited by the lack of suitable initial seeds. By producing certified, continuous solutions, the method enables the construction of theoretical templates necessary for upcoming observational tests.

The authors emphasize that their solutions reveal physical deformations in the horizon geometry and frame-dragging effects that are directly sensitive to higher-curvature couplings. These features are precisely the observables probed by X-ray spectroscopy, polarimetry, and gravitational wave ringdown spectra. The ability to trace physical observables (such as horizon radius and angular velocity) continuously across parameter space is presented as a critical step toward quantifying or breaking degeneracies between black hole spin and higher-curvature corrections in future data analysis. The method is noted to be adaptable to other setups involving coupled non-linear partial differential equations, though the current work focuses on the co-homogeneity-1 reduction to ODEs.