Deep learning emergent spacetime from fermionic spectral functions in holography
Koji Hashimoto, Hyun-Sik Jeong, Keun-Young Kim, Daichi Takeda, Kwan Yun
hep-th · cond-mat.str-el · cs.LG
Quick verdict
Neural ODEs reconstruct an extremal RN-AdS bulk and probe charge from synthetic fermionic spectra; convincing demo, but noise-free circular data limits scope.
Summary
The paper casts the holographic inverse problem - inferring the bulk metric functions f(z), h(z) and gauge field At(z) of a charged AdS black hole from boundary fermionic spectral functions A(w,k) - as physics-informed machine learning. Hard architectural constraints enforce AdS asymptotics, horizon regularity, and the (1-z)^2 double zero enforcing T=0; three small MLPs parameterize the unknown deviations, and the exact Dirac Riccati flow equation is integrated through the learned backgrounds during training (12x5 grid, 60 points; Adam then L-BFGS). Trained on noise-free spectra generated from the extremal RN-AdS background (mu=sqrt(3)) for three probe charges realizing non-Fermi-liquid, marginal-Fermi-liquid, and Fermi-liquid-like regimes, the method recovers the exact geometry (final losses 4.65e-5 to 8.72e-3) and, in a joint run, infers the charge as q=1.5610 versus the true 1.56 (0.06% error). Relaxing the near-boundary slope constraint exposes isospectral degenerate bulks whose near-horizon AdS2xR2 invariants match the exact values (Lx^2/L2^2 approx 6, 1/ed^2 approx 12), realizing the IR-universality argument of Faulkner et al. Main limitations: purely synthetic noise-free data from the same forward solver, no bulk equations of motion imposed, very small data grids, and no systematic uncertainty or seed statistics.
Review complete
Assessment at a glance
5
risk flags identified
● high 0
● medium 3
● low 2
5 strengths · 5 weaknesses · 5 author questions
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- A hard-constrained Neural ODE embedding the exact Dirac flow reconstructs the extremal RN-AdS metric and gauge field from 60-point boundary fermionic spectra across NFL, MFL, and FL-like regimes.
- The probe fermion charge can be jointly inferred from the spectra to sub-percent accuracy (q=1.5610 versus 1.56, a 0.06% error).
- Bulk profiles differing throughout the radial direction but sharing near-horizon AdS2xR2 data (Lx^2/L2^2 approx 6, 1/ed^2 approx 12) yield identical spectra near the Fermi surface, realizing the expected zero-temperature holographic degeneracy.
- Spontaneous convergence of independent training runs onto such degenerate bulks indicates the network isolates IR CFT universality rather than overfitting a single UV completion.
Strengths
Physics encoded as architecture, not penalty terms
The ansatz Eq. (18) hard-wires f(0)=h(0)=1, the (1-z)^2 double zero enforcing T=0, f(1)=0, and At(1)=0, so no constraint-violation loss weights need tuning; the exact Dirac Riccati flow Eq. (8) is integrated through the learned background at every training step (Sec. IV).
Degeneracy finding connected to known theory and quantified
Sec. VI verifies that degenerate trained bulks reproduce the exact near-horizon invariants Lx^2/L2^2=6 and 1/ed^2=12 (trained values 5.842-6.162 and 11.75-12.32), matching the Faulkner et al. IR-universality argument, and explicitly cautions that these metrics are probe-response sources, not solutions of a fixed bulk action.
Joint geometry-parameter inference
Robustness across spectral regimes and dimensions
Interpretable companion analysis of the degeneracy
Weaknesses
Circular, noise-free validation (inverse crime)
Reconstruction claim exceeds what 60 points can determine
No bulk equations of motion constrain the ansatz
No uncertainty quantification or seed statistics
Underspecified validation protocols
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
- All 5 strengths & 5 weaknesses, analyzed in detail
- 5 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
What is the holographic inverse problem addressed in this paper?
In holography (AdS/CFT), the usual forward calculation computes boundary observables from a known gravitational bulk. The inverse problem runs the other way: given boundary data - here the fermionic spectral function A(w,k), analogous to ARPES measurements - recover the unknown bulk spacetime (metric functions and gauge field) of the dual charged AdS black hole.
Can this method be applied to real ARPES experimental data now?
Not yet. All demonstrated reconstructions use noise-free synthetic spectra generated from the exact extremal Reissner-Nordstrom AdS solution. The authors list finite temperature, non-extremal black holes, noise robustness, and real cuprate or heavy-fermion ARPES datasets as future work.
What is the isospectral bulk degeneracy found in the paper?
Different bulk geometries that share the same near-horizon AdS2xR2 data (the combinations L2^2/Lx^2 and ed^2) produce identical fermionic spectra near the Fermi surface. So at zero temperature the boundary data determine only the infrared universality class of the dual field theory, not a unique ultraviolet-complete bulk spacetime.
What is a Neural ODE and why is it used here?
A neural network whose discrete layers are replaced by a continuous ordinary differential equation integrated numerically. In this paper the ODE is the physical Dirac flow equation itself, so training adjusts network-parameterized bulk fields while the exact physics equation is solved to high precision at every optimization step.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- The scaling-exponent formula Eq. (16) and a few inline math expressions are garbled in text extraction; the review relies on the surrounding prose and standard results of Faulkner et al.
- Figures are available only through captions and axis labels; the plotted agreement quality in Figs. 2, 4, 5, and 9 could not be independently verified.
- No explicit code or data availability statement appears anywhere in the extracted text; codeAvailable=false reflects that absence.
- The concurrent related work of Xiao et al. (Ref. [33]) was not consulted; comparative judgments rest on the paper's own description.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.