Random unitaries — quantum operations drawn from distributions that mimic the Haar measure to varying degrees — underpin benchmarking protocols, quantum learning, and many-body physics simulations. Generating them efficiently is a core circuit-design problem, traditionally measured along two axes: circuit depth and qubit count. Prior low-depth constructions achieved speed by borrowing heavily from ancillary qubits, sometimes requiring as many extra qubits as the system itself.
In this unreviewed preprint posted to arXiv, the authors claim to eliminate that spatial overhead entirely. They present explicit constructions of approximate multiplicative-error k-designs that reach depth Õ(k)(log n)^(1/δ) on δ-dimensional qubit architectures and Õ(k) log log n with all-to-all connectivity — figures the authors argue are optimal — without any ancillae. They also introduce what they call an "exactification lemma," a general technique for promoting approximate designs to exact ones, which they say yields an exponential reduction in resources compared to state-of-the-art exact constructions.
If the results survive peer review, the work would tighten space-time trade-offs across a broad class of quantum protocols, from randomized benchmarking and shadow tomography to quantum-secure cryptographic primitives. The paper has not yet been peer-reviewed, and independent verification of the depth optimality claims and the exactification construction will be needed before the results can be considered established.