First-hand information for everyone
The paper gives a sampling task that very shallow quantum circuits can solve, but that is hard for any efficient classical algorithm under plausible lattice-based assumptions. At the same time, a classical computer can efficiently check whether an output sample is valid. In short, the authors present a way to get a verifiable quantum advantage using circuits that are extremely shallow in depth.
The researchers build on a previous Learning with Errors (LWE)-based proof of quantumness and “compile” it into two low-depth quantum implementations. One uses log-logarithmic depth with ordinary one- and two-qubit gates (a class the authors call QNC^0[log log]). The other uses constant depth but allows gates that act on many qubits at once (a class called QAC^0, meaning constant-depth with unbounded fan-in). Importantly, neither implementation needs mid-circuit measurements or classical feed-forward. The quantum device simply runs a shallow unitary circuit and samples its outputs once.
At a high level the construction rests on the hardness of LWE, a standard lattice-based cryptographic assumption, together with a stronger technical property the authors call a strengthened adaptive-hardcore-bit assumption (with carry predicates). The authors give supporting evidence for this strengthened property but do not prove it unconditionally. Under these assumptions, the paper shows that honest quantum samplers produce outputs that a classical polynomial-time verifier accepts, while any polynomial-time classical sampler is provably much less likely to be accepted.
Why this matters: constant- or near-constant-depth quantum circuits are among the simplest quantum devices one can build. Showing they can solve sampling tasks that are both classically hard and efficiently checkable brings us closer to practical, convincing tests of “quantumness.” It also clarifies what very shallow quantum circuits can do, and it separates the need for mid-circuit measurement or adaptive control from the ability to produce verifiable, classically hard samples.
There are important caveats. The hardness claims are conditional on lattice-based assumptions and on a strengthened adaptive-hardcore-bit property that is not standard; the paper provides evidence but not a proof for that strengthening. The result applies to a sampling problem rather than a decision problem, and the constant-depth implementation requires unbounded fan-in gates, which are a strong model assumption. In short, the advantage is compelling under the stated assumptions, but those assumptions are the main source of uncertainty.