A scalable protocol to certify analog simulator outputs
In plain words
Checking a simulator by calculation fails exactly where the simulator is useful. The task is a test that needs only modest classical computing and a reasonable number of measurements.
Precise statement
Give a protocol using $\operatorname{poly}(N)$ measurements and $\operatorname{poly}(N)$ classical post-processing (for example Hamiltonian learning, time-reversal echoes, cross-platform comparison of randomized measurements) that bounds the error of local observables of the prepared state relative to the ideal Hamiltonian, for $N > 100$ in a regime where classical simulation is believed hard. Answer: protocol with a proof of soundness and an experimental demonstration.
What would settle it
A protocol with a soundness proof applied on hardware with $N > 100$ in a classically hard regime.
Related problems
- More general than Certified accuracy of a Hubbard simulator where numerics fail