Deep learning-based holography for T-linear resistivity

TL;DR

This paper uses deep learning within holographic duality to model and understand the $T$-linear resistivity characteristic of strange metals, deriving key features of these materials through neural network techniques.

Contribution

It introduces a physics-informed neural network approach to derive dilaton potentials that reproduce $T$-linear resistivity and specific heat scaling in holographic models of strange metals.

Findings

Dilaton potentials show universal exponential growth at low temperatures.

The method recovers the Gubser-Rocha model at a specific slope.

$T$-linear resistivity remains robust at higher temperatures with AdS asymptotics.

Abstract

We employ deep learning within holographic duality to investigate $T$-linear resistivity, a hallmark of strange metals. Utilizing Physics-Informed Neural Networks, we incorporate boundary data for $T$-linear resistivity and bulk differential equations into a loss function. This approach allows us to derive dilaton potentials in Einstein-Maxwell-Dilaton-Axion theories, capturing essential features of strange metals, such as $T$-linear resistivity and linear specific heat scaling. We also explore the impact of the resistivity slope on dilaton potentials. Regardless of slope, dilaton potentials exhibit universal exponential growth at low temperatures, driving $T$-linear resistivity and matching infrared geometric analyses. At a specific slope, our method rediscovers the Gubser-Rocha model, a well-known holographic model of strange metals. Additionally, the robustness of $T$-linear resistivity at higher temperatures correlates with the asymptotic AdS behavior of the dilaton coupling to the Maxwell term. Our findings suggest that deep learning could help uncover mechanisms in holographic condensed matter systems and advance our understanding of strange metals.