Photon-loss threshold for classical simulability of Gaussian boson sampling
In plain words
Gaussian boson sampling sends squeezed light through a large optical network and records where photons arrive; it was used to claim quantum advantage. Losing photons makes the task easier for classical computers, and the loss level at which it becomes easy is not known exactly.
Precise statement
For Gaussian boson sampling with $M$ modes, $K$ single-mode squeezed inputs of squeezing r, a Haar-random interferometer and uniform transmission $\eta$, find the boundary $\eta_c(r, K, M)$ separating classically efficient sampling to fixed total-variation error from sampling hard under standard conjectures, as M, K go to $\infty$ at fixed ratio.
What would settle it
A classical algorithm efficient below $\eta_c$ together with a hardness proof above it, for the same scaling family.
Status in the literature
Unverified note
Oh et al. (arXiv:2306.03709, Nature Physics 2024) simulated reported experiments by exploiting their loss; sufficient conditions for hardness of lossy GBS appeared in 2025 (arXiv:2511.07853).