Can shallow circuits maintain quantum advantage if depolarizing noise strength varies across different qubits or layers?

This is a sophisticated question that sits at the intersection of quantum circuit design and noise robustness. The short answer: no, not reliably — but the reason is subtle and depends on whether noise variations are engineered or adversarial.

The uniform-noise baseline

When depolarizing noise is uniform across qubits and layers, shallow circuits face a fundamental problem: noise itself forces them to remain shallow. [AG-2024.03-2009] proves that realistic uncorrected noise effectively *truncates* most quantum circuits to logarithmic depth for observable estimation. This sounds like good news for shallow circuits, but it comes with a catch—the noise also destroys the quantum advantage simultaneously. [AG-2024.03-2009] shows that any noisy circuit becomes efficiently simulable classically with constant additive error, regardless of circuit depth.

The reason: non-unital noise (which includes depolarization) prevents barren plateaus, making optimization feasible, but it also allows a classical algorithm to estimate observables in time independent of depth. Shallow circuits don't escape this; they're already in the regime where classical simulation works.

The heterogeneous-noise regime

When noise *varies* across qubits or layers, the situation becomes more nuanced:

Threshold effects exist. [AG-2024.05-1599] studies shallow-circuit learning under depolarizing noise of varying strength. They prove that quantum-classical separation persists if noise strength stays below inverse-polynomial levels (in system size), but vanishes above inverse-polylogarithmic thresholds. Critically, this analysis does not assume uniform noise—it's about an upper bound on noise strength. If different qubits have noise *below* the threshold while others exceed it, you lose advantage globally.

Spatial structure matters enormously. [AG-2024.06-1798] analyzes linear-optical circuits under photon loss (a spatial noise process) and shows that noise effectively removes vertices from the circuit graph. Using percolation theory, they prove that above a critical loss threshold, the circuit decomposes into disconnected components that can be simulated classically. Heterogeneous loss across modes would accelerate this fragmentation—regions with high noise decouple from the rest, eliminating long-range entanglement needed for advantage.

T-gate bottleneck. [AG-2024.07-2298] studies circuits mixing Clifford and T gates under depolarizing noise. They derive conditions: if noise rate exceeds the fraction of T gates, classical simulation becomes easy. Unequal noise across layers would preferentially degrade whichever layers are noisier, and if those contain most of the non-Clifford content, advantage evaporates.

The engineering loophole

There's one narrow escape: [AG-2024.01-1336] demonstrates that mid-circuit measurements with classical feedback can maintain quantum advantage even under complete depolarization in the large-depth limit. Crucially, this works because measurements actively reset the state rather than passively accumulating noise. If heterogeneous noise is *deliberately engineered* to trigger protective mid-circuit measurements in high-noise regions, advantage might persist. But this requires designing the circuit *around* the noise pattern—it's not a free pass.

Bottom line

Spatial or temporal heterogeneity in noise doesn't rescue shallow circuits for quantum advantage. Instead, it typically accelerates classical simulability by breaking entanglement structure or exceeding noise thresholds in critical layers. The archive does not contain results showing robust advantage recovery under untuned heterogeneous depolarization.