Predictive Structure Behind Rare Outcomes in Random Quantum Circuits

Abstract

Rare outputs of random quantum circuits are usually treated as terminal statistics. We ask whether they reveal intermediate organization that remains useful beyond the selecting future. Comparing unconditioned random-normal (RN) and peak-selected random-peaked (RP) circuits from the same local generator, we find stronger pairwise correlations and probability redistribution but reduced bipartite and output spreading in RP across n=8-16. Exact full-suffix decomposition reveals stronger peak-directed interference. The structural separation and fresh-future advantage persist across four nominal depths at n=8-14, with 2.01-7.15 fold RP enrichment in the depth extensions at fixed, independently calibrated thresholds. Within the unconditioned ensemble, a structural prefix score predicts fresh-event probability without using the evaluated circuits' terminal outcomes. Population-preserving phase scrambling identifies a functional contribution from relative-phase organization across the tested settings, including primary-depth held-out confirmation. RP trajectory-guided initialization improves high-peak yield over Haar initialization under common local refinement. Intermediate multivariate guidance also improves the fresh-continuation susceptibility of constructed states beyond peak-only guidance across the same size-depth grid. Rare outputs thus reveal intermediate physical organization that remains useful under new dynamics and can guide circuit construction.