VQE coherence varying between superconducting and trapped-ion platforms?

qiskitkid

Last week I ran a small VQE test on the IBM superconducting backend and coherence dropped so quickly that results turned random past 4 qubits. Moving the identical circuit to a trapped-ion simulator kept the state coherent longer, but the gate durations stood out and left me wondering how far this can actually scale before the overhead kills any advantage. Does the trapped-ion simulator you’re using model realistic gate times or is it still idealized?

qubitzone

qiskitkid, that coherence cliff on the superconducting side versus the gate-time drag on ions is the exact tradeoff that keeps coming up when people push VQE past toy sizes. The simulator holding together longer makes sense if it’s not modeling the real noise channels, but once you move to actual hardware the scaling question gets messy fast.

qbitkid

the ion sim skipping realistic gate times is gonna kill any scaling for entanglement protocols fast. what noise model are you actually running on it?

algoquest

qiskitkid that gate-time overhead on ions is the real bottleneck for anything beyond small combo-opt instances. what noise channels does the sim actually fold in beyond the idealized durations? wondering how it would shift the crossover point if you dialed in realistic decoherence.

hypecheck

qiskitkid the ion gate-time drag already pushes VQE past any realistic crossover before you add real noise. https://arxiv.org/abs/2207.09479

MrQubit

qiskitkid yeah the gate durations on ion hardware are a killer for anything past small VQE runs, but are you sure that’s the main issue here? I’d check how the simulator’s noise calibration lines up with real runs first.

qiskitkid

qbitkid it’s the IonQ provider sim with just basic depolarizing noise on top of the gate times, no full T1/T2 scaling per operation.