arXiv:2609.32025
Nonstabilizerness of the entangling layer in quantum circuit Born machines
Marcin Płodzień
Abstract
The entangling layer of a quantum circuit Born machine can supply both entanglement and nonstabilizerness. In this work we show that the entangling layer can be a Clifford circuit, which supplies no nonstabilizerness. We use an architecture in which a fixed, non-trainable entangler is reused in every layer and only single-qubit rotations are optimized. There, a Clifford entangler reaches the same Kullback--Leibler divergence as a near-Haar one at sufficient circuit depth, and entanglers spanning zero to Haar-typical nonstabilizerness reach the same loss, up to a small offset for a free-fermion matchgate entangler. A doping control that raises the entangler's nonstabilizerness while leaving the entangling power of every gate unchanged leaves the loss flat. With a Clifford entangler the nonstabilizerness of the trained state comes entirely from the rotations, which take the trained circuit outside the stabilizer formalism. We also show that, among all pure states whose Born distribution is a given target, the one whose amplitudes are the positive square roots of the probabilities has the smallest stabilizer Rényi entropy, so this minimum can be computed without training.
AI Summary
Does a quantum circuit Born machine need nonstabilizerness in its entangling layer, or can trainable single-qubit rotations supply it? The paper studies circuits that reuse a fixed entangler between optimized rotations, comparing random Clifford, $T$-doped Clifford, local-Haar and matchgate brickworks through exact-statevector training and KL-divergence evaluation. For Gaussian-mixture, Sherrington–Kirkpatrick Gibbs and constructed high-nonstabilizerness targets, Clifford, doped-Clifford and local-Haar entanglers reach statistically comparable losses at sufficient depth. At ten qubits, doping up to all 90 two-qubit gates raises the entangler’s probe-state stabilizer Rényi entropy (SRE) from zero to approximately eight bits while leaving loss flat; doping preserves each gate’s entangling power. Matchgates show a small persistent positive loss offset. The trained states nevertheless have substantial nonstabilizerness, supplied by the rotations with Clifford entanglers, and their SRE matches random-phase encodings of their output probabilities. A separate theorem proves that, among all pure states with Born probabilities $p$, the positive-amplitude state $\sum_x\sqrt{p_x}|x\rangle$ minimizes SRE at every integer order $\alpha\ge2$. A Walsh-transform expansion and triangle inequality establish the result, replacing phase optimization by one SRE evaluation, still exponentially costly. Training is limited to at most 12 qubits and this fixed-entangler architecture; the comparisons do not establish asymptotic equivalence. Constructed targets confound increasing intrinsic SRE with increasing sparsity. Entangler SRE is measured on one reference input. A deterministic Clifford demonstration on ten-qubit two-dimensional targets attains best-run KL divergence below $0.13$, but fails to resolve spiral arms, illustrating that low loss need not imply structural fidelity.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.