Revision 2

Accepted · core change · record

Changes against revision 1:

Title: Certified accuracy of a Hubbard simulator where numerics fail
Plain: Once a simulator goes where computers cannot follow, nobody can check its answers by calculation. The task is to bound its errors using only independent calibrations.
Precise · changed · core: Revision 1: Choose a point $(U/t, \delta, T/t)$ where unbiased numerical methods lose control (for example $U/t = 8, \delta = 0.1, T/t < 0.1$). Using Hamiltonian learning of t, t', U and harmonic confinement, entropy thermometry, and stability bounds, give a certified error bar on the nearest-neighbour spin correlator $C(1,0)$ and on the density. Answer: an error budget with all sources listed and a total below 5%.; Revision 2: Choose a point $(U/t, \delta, T/t)$ where unbiased numerical methods lose control (for example $U/t = 8, \delta = 0.1, T/t < 0.1$). Using Hamiltonian learning of $t, t', U$ and harmonic confinement, entropy thermometry, and stability bounds, give a certified error bar on the nearest-neighbour spin correlator $C(1,0)$ and on the density. Answer: an error budget with all sources listed and a total below $5\%$.
Settled by: A published error budget combining in-situ Hamiltonian calibration with a proof or numerical bound on how parameter errors propagate to local observables.
Kind: well-posed
Answer type: value
References: (empty)
Literature status: open
Status note: (empty)
Posed since: (empty)

Decisions