Cultivating logical catalysts for fault-tolerant dyadic phase rotations

Problem Statement

Fault-tolerant quantum computation typically relies on Clifford operations supplemented by non-Clifford resource states (magic states) to achieve universality. In codes with restricted transversal gate sets, such as the rotated surface code, standard approaches to implementing fine dyadic phase rotations (e.g., $Z_{2^{-b}}$) face two primary bottlenecks:

1. Approximation Error: Standard Clifford+T synthesis and Repeat-Until-Success (RUS) circuits approximate target rotations. The synthesis precision $\epsilon$ must be chosen below the target logical error rate, causing the non-Clifford cost (T-count) to grow as the computation demands lower logical error.
2. Verification Difficulty: While "magic state cultivation" (preparing a low-distance state, verifying it, and growing it to high distance) has been successful for the canonical $|T\rangle$ state, it does not easily extend to finer dyadic phases ($b \ge 3$). Direct verification of states like $Z_{2^{-b}}|+\rangle$ requires non-Clifford measurements, creating a circular dependency where the resource needed for verification is the very resource being cultivated.
3. Catalyst Limitations: Existing catalytic methods, such as phase-gradient states, offer reusability but often require non-Clifford preparation or have online costs (depth and T-count) that scale with the precision $b$. Furthermore, phase-gradient states are eigenstates of non-Clifford adders, making them difficult to cultivate using only Clifford operations.

Methodology

The authors propose a surface-code cultivation protocol for reusable logical catalyst states that implement exact fine dyadic phase gates ($Z_{2^{-b}}$) via phase kickback. The core mechanism relies on the controlled-Clifford catalyst framework:

1. Catalyst Construction:

   • The catalyst is defined as an eigenstate of a high-period Clifford circuit $U$. Specifically, the authors utilize a "brickwork" CNOT circuit $U_n$ composed of alternating layers of disjoint CNOTs.
   • For a target phase $Z_{2^{-b}}$, they select $n = 2^b + 1$ qubits. The circuit $U_n$ has a period of $2^m$ (where $m = \lceil \log_2 n \rceil$).
   • An eigenstate $|\psi_{n,r}\rangle$ of $U_n$ with eigenvalue $\omega_m^r$ serves as the catalyst. When used in a controlled-$U_n$ gadget, phase kickback applies an exact $Z_{2^{-b}}$ rotation to the control qubit while returning the catalyst to itself.
   • Case Study: The paper focuses on the $\sqrt{T} = Z_{1/8}$ gate ($b=3$). This requires a 9-qubit brickwork circuit $U_9$ with period 16. The catalyst is the $r=1$ eigenstate $|\psi_9\rangle$ with eigenvalue $e^{i\pi/8}$.

2. Cultivation Protocol:
   The protocol prepares a fault-tolerant logical copy of $|\psi_9\rangle$ through three stages:

   • Physical Preparation & Encoding: A physical 9-qubit eigenstate is prepared and encoded into nine independent distance-three rotated surface-code blocks ($Rot(3)$).
   • Logical Verification (Phase Estimation): The encoded state undergoes a logical $U_9$ measurement via a Quantum Phase Estimation (QPE) circuit.
     • Crucially, the logical $U_9$ check is intrinsically fault-tolerant. Because $U_9$ has a discrete spectrum of 16th roots of unity, a single round of phase estimation with a trivial syndrome (reading out the target phase) projects the state exactly into the correct eigenspace.
     • The authors prove that a single verification round ($N_1=1$) provides an effective logical fault distance of $f_{logical} \approx 2.654$, sufficient to suppress leading errors without the repeated checks required for $|T\rangle$ cultivation.
   • Code Growth: Post-selected blocks are grown from distance 3 to distance 7. This involves unitary growth ($Rot(3) \to Reg(3) \to Rot(5)$) and stabilizer-measurement growth ($Rot(5) \to Rot(7)$), utilizing complementary-gap decoding to filter residual errors.

3. Simulation Strategy:
   Due to the non-Clifford nature of the verification stage (involving Toffoli gates for controlled-$U_9$), the authors employ a hybrid simulation:

   • Front End: A tensor-network (MPS) state-vector simulation tracks the noisy physical preparation and logical $U_9$ verification. It analytically calculates the probability of passing postselection (Rao–Blackwellization) rather than discarding failed shots.
   • Back End: A stabilizer simulation (Stim) with matching-based decoding (PyMatching) simulates the Clifford-only growth stages, seeded by the syndrome information extracted from the front end.

Key Contributions

• Exact Dyadic Catalysts: The paper introduces a method to cultivate reusable catalyst states for exact dyadic phases ($Z_{2^{-b}}$) that are eigenstates of Clifford circuits, bypassing the need for non-Clifford verification.
• Single-Round Verification: The authors demonstrate that for high-period Clifford circuits, a single logical phase-estimation round is sufficient to achieve fault-tolerant verification, contrasting with the multiple rounds typically needed for $|T\rangle$ states.
• Constant Online Depth: The online invocation of the catalyst uses a controlled-Clifford circuit with constant T-depth (independent of $b$) via parallelized Toffoli gates, avoiding the logarithmic depth scaling of synthesis or the linear depth of phase-gradient adders.
• Hybrid Simulation Framework: A robust simulation methodology combining tensor networks and stabilizer codes to evaluate cultivation protocols involving non-Clifford verification.

Results

• Performance: At a physical error rate of $p = 10^{-3}$, the protocol successfully cultivates a distance-seven logical catalyst with a logical leakage rate of $\sim 10^{-6}$ using approximately 7 expected attempts. With stronger postselection, this can be suppressed to $\sim 10^{-7}$.
• Comparison to $|T\rangle$ Cultivation: The single-round verification achieves a leakage rate comparable to recent single-block $|T\rangle$ cultivation protocols, despite the catalyst spanning nine surface-code blocks and supporting a finer phase.
• Resource Trade-offs:
  • Online Cost: The $\sqrt{T}$ catalyst requires 8 Toffoli gates (or 32 T gates) per invocation with constant depth. This is competitive with or better than optimized RUS and synthesis methods for high-precision targets, while being exact.
  • Offline Cost: The preparation cost is amortized over many uses. However, the catalyst size scales as $O(2^b)$ logical qubits (9 blocks for $\sqrt{T}$), which is a significant overhead compared to synthesis (0 ancillas) or phase-gradient states ($O(b)$ qubits).

Significance and Claims

The paper claims that this approach offers a distinct trade-off: it exchanges the $\epsilon$-dependence of synthesis costs (where cost grows as precision tightens) for a $b$-dependence (where resource state size grows with phase fineness).

• Exactness: Unlike synthesis or RUS, the catalyst introduces zero approximation error per use. The rotation is algebraically exact, bounded only by the residual logical infidelity of the catalyst itself.
• Latency: The constant online T-depth is a structural advantage for latency-critical applications (e.g., inner loops of Trotterized simulations) where variable or logarithmic depth is detrimental.
• Feasibility: The protocol provides a concrete, Clifford-only pathway to prepare high-fidelity, fine-phase catalysts from scratch, a capability previously obstructed by the non-Clifford nature of the required verification for finer dyadic phases.

The authors conclude that while the qubit overhead limits the protocol to "modest" values of $b$ (moderately fine phases), it is highly effective for workloads involving the repeated application of a fixed, fine dyadic phase. They identify reducing the $O(2^b)$ overhead and extending the cultivation to other structured magic states (like $|CCZ\rangle$) as key future directions.