Classification and Exact Local Masking in Finite-Field Clifford Dual-Unitary Circuits

Technical Summary: Classification and Exact Local Masking in Finite-Field Clifford Dual-Unitary Circuits

Problem Statement
This work addresses the classification of two-qudit finite-field Clifford dual-unitary gates and their application to exact local masking and operator transport in homogeneous brickwork circuits. While dual-unitary circuits are known for their solvability and maximal velocity of information propagation ($v_B=1$), a complete structural understanding of the finite-field Clifford subset—specifically regarding their transport phases and privacy properties—remains incomplete. The paper seeks to classify these gates under ordered local equivalence, determine the resulting transport phases in repeated circuits, and derive exact "masking distances" (the minimum input span required to hide information from local output subsystems).

Methodology
The paper employs a finite-field algebraic framework where $n$-qudit Weyl operators (modulo phase) are labeled by vectors in $\mathbb{F}_q^{2n}$. Clifford transformations are represented by symplectic matrices in $Sp(2n, q)$.

1. Gate Classification: The authors analyze the symplectic matrix representation $S = \begin{pmatrix} A & B \\ C & D \end{pmatrix}$ of two-qudit gates. They define dual-unitarity via the invertibility of blocks $B$ and $C$, and the "perfect tensor" condition via the invertibility of all four blocks.
2. Ordered Local Equivalence: Gates are classified under the action of ordered one-qudit Clifford groups ($SL(2, q)$) acting on the input and output legs. This reduces the continuous-like space of gates to discrete "cores."
3. Circuit Dynamics: The study extends from isolated gates to homogeneous brickwork circuits. The authors utilize one-site edge channels (transfer matrices $T_L, T_R$) derived from the light-cone maps to determine operator spreading and "glider" behavior.
4. Masking Analysis: Using the concept of "local masking," the paper defines $d_r(t)$ as the minimum contiguous input span such that any output Weyl string of weight $\le r$ has a backward image that cannot be supported entirely within the input. This is linked to exact decoupling of local subsystems from the input.

Key Contributions and Results

• Classification of Dual-Unitary Cores:
  Under ordered local Clifford equivalence, the dual-unitary locus in $Sp(4, q)$ contains exactly $q$ nonlocal cores:

  • Perfect Cores ($q-2$): Indexed by $\delta = \det B = \det C \in \mathbb{F}_q \setminus \{0, 1\}$. These correspond to perfect tensors (AME states) and are related to the cross-ratio $\lambda = \delta/(\delta-1)$ of an associated $[4, 2, 3]_q$ MDS configuration.
  • Rank-One Core ($1$): Contains all classes where $A$ is singular but non-zero.
  • SWAP Core ($1$): Contains all classes where $A=0$.
    The stabilizers and orbit sizes for these cores are explicitly computed.

• Transport Phases in Homogeneous Circuits:
  While the isolated gate classification yields three core types, homogeneous repetition separates the rank-one core into distinct transport behaviors, resulting in five transport phases in total:

  1. Immediate Bilateral Erasure ($\kappa=1$): Corresponds to perfect tensors. Edge channels vanish immediately.
  2. Delayed Bilateral Erasure ($\kappa=2$): Edge channels are non-zero but nilpotent of order 2.
  3. One Gliding Edge: One edge supports a persistent glider; the other erases.
  4. Two Gliding Edges: Both edges support persistent gliders.
  5. SWAP Transport: Both edges support full permutations (gliders).
     The paper establishes a "codimension ladder" in $Sp(4, q)$, where these phases correspond to algebraic loci of codimensions 0, 1, 2, 3, and 4, respectively. As $q \to \infty$, the perfect tensor phase dominates the ensemble.

• Exact Local Masking Laws:
  The paper derives exact formulas for the masking distances $d_r(t)$ (the smallest input span hidden from an output set of size $r$ at time $t$):

  • Perfect Tensor Phase ($\kappa=1$):
    $d_1(t) = 4t$, $d_2(t) = 4t - 2$.
    These values saturate the causal ceilings for depth-$2t$ nearest-neighbor circuits.
  • Delayed Erasure Phase ($\kappa=2$):
    $d_1(t) = 4t - 2$, $d_2(t) = 4t - 4$ (for $t \ge 2$).
  • Glider/SWAP Phases:
    $d_r(t) = 1$ for all $r, t \ge 1$ (no masking occurs).

  These distances imply that sufficiently short quantum messages are completely hidden (the reduced state is maximally mixed) from any one- or two-qudit observer, even if the input is entangled with a reference, while remaining exactly recoverable from the full output.

• Qubit Case ($q=2$):
  Since no AME(4, 2) state exists, the perfect tensor phase is empty for qubits. The optimal phase is "delayed erasure." The paper provides a scalar edge witness ($w_\nu$) based on tetrahedral states to experimentally distinguish between nilpotent, gliding, and SWAP phases in qubit circuits.

• Error Sensitivity:
  Numerical simulations for $q=3$ show that local leakage scales linearly with coherent perturbation strength, whereas the infidelity of recovery using the ideal inverse scales quadratically near the perfect point. Stochastic Pauli noise preserves local privacy (zero leakage) but degrades recovery fidelity.

Significance and Claims
The paper claims to provide the first ordered local classification of finite-field Clifford dual-unitary gates and to convert this classification into exact, reference-secure masking distances.

• Distinction from Data Hiding: The work distinguishes its "local-observer private encoding" from standard bipartite quantum data hiding (which relies on LOCC restrictions) and multipartite secret sharing. Here, privacy is exact subsystem decoupling for any measurement on the bounded output, generated dynamically by the circuit depth.
• Algebraic Structure: The results highlight a deep connection between the transport phases and the algebraic geometry of $Sp(4, q)$, specifically the incidence codimensions of determinantal varieties and fixed-line conditions.
• Optimality: For $q \ge 3$, the perfect tensor phase achieves the theoretical maximum masking distance allowed by the causal structure of nearest-neighbor circuits.
• Limitations: The paper explicitly notes that the classification of higher-local distances ($r \ge 3$) remains an open problem, as edge-erasure depth is insufficient to characterize three-local behavior, which depends on boundary monodromy and cross-ratio parameters.

The work does not propose new experimental hardware but offers a theoretical framework for understanding privacy and scrambling in solvable quantum circuits, with specific constructive examples for qutrits ($q=3$) and qubits ($q=2$).