Three Hamiltonians are Sufficient for Unitary k-Design in Temporal Ensemble

Unitary $k$-designs are central to quantum information and quantum many-body physics as efficient proxies for Haar-random dynamics. We study how chaotic Hamiltonian evolution can generate unitary $k$-designs. Standard approaches typically rely on many independent Hamiltonian realizations or fine-tuning evolution times. Here we show that unitary designs can instead arise from a quenched temporal ensemble, where Hamiltonians are sampled once and held fixed, while randomness enters only through the evolution times. We analyze a two-step protocol (2SP), applying $H_1$ for time $t_1$ and $H_2$ for time $t_2$, and a three-step protocol (3SP) with an additional quench, with all times randomly drawn from a prescribed distribution. Time averaging imposes energy-index matching in the frame potential (FP), which quantifies the distance to Haar random. Analytically and numerically, we show that 2SP cannot realize a general unitary $k$-design, whereas 3SP can do so for arbitrary $k$. The advantage of 3SP is that the additional random phases impose stronger constraints, eliminating independent permutation degrees of freedom in the FP. For Gaussian unitary ensemble Hamiltonians, we prove these results rigorously and show that under imperfect time averaging, 3SP achieves the same accuracy as 2SP with a parametrically narrower time window.