Reproduce “Machine-learning emergent spacetime from linear response in future tabletop quantum gravity experiments” on OpenResearch

About this paper

Koji Hashimoto, Koshiro Matsuo, Masaki Murata, Gakuto Ogiwara, Daichi Takeda

A research team from Kyoto University and Saitama Institute of Technology presents an interpretable neural network (NN) model that reconstructs two independent metric components of an emergent 3D spacetime from 2D boundary quantum field theory linear response data. The model successfully reconstructs the BTZ black hole metric and its horizon condition with high accuracy, achieving a classification accuracy of 1.0 and learning horizon parameters that closely match theoretical values.

Key insights

• Direct integration of a robust numerical solver (Runge-Kutta) into the NN architecture enables accurate propagation of field values deep into the bulk and makes the model interpretable, as trainable weights correspond to physical metric coefficients.
• Spatially and temporally inhomogeneous linear response data from the boundary is sufficient to reconstruct multiple (two) independent, spacetime-dependent metric components of the bulk.
• The problem of bulk reconstruction, particularly for black hole spacetimes, can be effectively framed as a classification task to determine if boundary data corresponds to a field satisfying the ingoing boundary condition at the horizon.

Summary by alphaXiv.