arXiv:2609.40227
Randomness of exact unitary designs under symmetry
Christopher Vairogs, Felix Leditzky
Abstract
We define the unitary design strength of a finite unitary ensemble to be the maximal integer $t$ for which its $t$-th statistical moment is identical to that of the unitary group. In the presence of physical symmetry, not all operators of an ensemble are symmetry-compatible. This motivates us to define the symmetric design strength of a finite unitary ensemble to be the maximal integer $t$ for which the $t$-th moments of its symmetry-compatible sub-ensemble match those of the group of symmetry-compatible unitaries. We show that in sufficiently high dimension and for a broad family of symmetries, which includes the global on-site $\mathrm{U}(1)$ and $\mathrm{SU}(d)$ symmetries, there exist finite unitary ensembles of arbitrarily high design strength whose symmetric design strength is strictly higher. Ensembles with this property may be more versatile under symmetry constraints for protocols that rely on unitary randomization than they are without symmetry. Our results extend the previous work [Mitsuhashi and Yoshioka, PRX Quantum 4.4 (Nov. 2023)] showing that the symmetric design strength of the Clifford group is strictly upper bounded by its unitary design strength for symmetries that do not trivialize the symmetry-compatible unitaries. Due to the special significance of unitary designs with a group structure, we also derive a variety of bounds on the unitary and symmetric design strengths of finite unitary ensembles whose operators form a group. In particular, we provide a simple proof that an arbitrary ensemble of operators forming a finite group must have a symmetric design strength of at most two under the global on-site $\mathrm{U}(1)$ and $\mathrm{SU}(d)$ symmetries. Along the way, we show that there exist uniformly weighted groups of arbitrarily high symmetric design strength in arbitrary dimension and analyze their structure.
AI Summary
The paper asks how exact unitary-design strength changes when an ensemble is restricted to its symmetry-compatible operators, with probabilities conditioned on that restriction. The reference measure then becomes Haar measure on the symmetry commutant, rather than on the full unitary group. For any integers $t'>t\geq1$, it constructs finite weighted ensembles with ordinary design strength exactly $t$ and symmetric design strength at least $t'$, provided the symmetry representation contains an irrep occurring with multiplicity one and the Hilbert-space dimension is sufficiently large. For $t\leq3$, dimension at least three suffices. Applicable symmetries include collective $\mathrm{U}(1)$ and $\mathrm{SU}(q)$, and subsystem permutations under additional size conditions. The construction uses convolutions, block-diagonal designs with relative phases, and zeros of zonal spherical functions to match moments through order $t$ while retaining a nonzero $(t+1)$-moment discrepancy. For ensembles forming finite groups, symmetric strength above five requires an abelian symmetry commutant; nonabelian examples attain strength five. Under collective $\mathrm{U}(1)$ or $\mathrm{SU}(q)$ symmetry on $N\geq4$ subsystems, every finite-group ensemble has symmetric strength at most two, regardless of weighting. These bounds follow by reducing to ordinary group designs on multiplicity spaces and applying their classification. Separate constructions raise group-design strength from one to arbitrarily high symmetric strength. The results concern exact designs. The constructions are not necessarily practical, provide no local-gate compilations, and leave open extension beyond the multiplicity-one condition and stronger bounds for group designs.
Error found: The abstract's unrestricted finite-group upper bound fails for two-site global SU(d), where independent finite phases on the symmetric and antisymmetric sectors yield arbitrarily high symmetric design strength. Theorem 3(1) uses the wrong sector multiplicities for global U(1), and its bound is false for four qubits because the charge-sector dimensions 1, 4, 6, 4, 1 support a finite symmetric 3-design group. Theorem 1 incorrectly concludes t' > t while allowing t' = t, and its proof requires symmetric-design strength at least t+1 rather than merely the stated t'.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 35 pages