Noise-induced classical phases in optimally-unraveled random quantum circuits

Technical Summary: Noise-Induced Classical Phases in Optimally-Unraveled Random Quantum Circuits

Problem Statement
The central problem addressed is the boundary between quantum evolutions that can be efficiently simulated on classical computers and those requiring quantum hardware. While specific classes of quantum circuits (e.g., stabilizer circuits) admit efficient classical descriptions via the Gottesman-Knill theorem or tensor networks (e.g., Matrix Product States, MPS), the presence of noise in realistic quantum devices qualitatively alters the dynamics. A crucial open question is whether noise can render quantum dynamics easier to simulate classically. Furthermore, open quantum systems admit two complementary descriptions: the unconditional evolution of a mixed density matrix (via a quantum channel or Lindblad equation) and the stochastic evolution of pure-state trajectories (unravelings). These descriptions are physically equivalent in the average but computationally distinct. The paper investigates whether the "gauge freedom" in choosing an unraveling (Kraus decomposition) can be optimized to minimize the classical resources required for simulation, potentially revealing "noise-induced classical phases" where dynamics remain simulable at arbitrary circuit depths.

Methodology
The authors study random Clifford circuits "doped" with single-qubit non-Clifford phase rotations ($U_\phi$) and subject to local noise channels ($\mathcal{N}$). The system consists of $N$ qubits evolving through $N_T$ layers, where each layer applies a global random Clifford circuit, followed by a non-Clifford rotation and a local noise channel.

To simulate these dynamics, the authors employ Clifford-augmented Matrix Product States (CAMPS). In this formalism, a many-body state $|\Psi\rangle$ is represented as $C |\Psi_{\text{MPS}}\rangle$, where $C$ is a Clifford operator (handled efficiently via the stabilizer formalism) and $|\Psi_{\text{MPS}}\rangle$ is a standard MPS carrying the residual correlations that Clifford operations cannot remove. The simulation cost is governed by the bond dimension of the inner MPS, which is controlled by the amount of "non-stabilizerness" (or "magic") that cannot be absorbed by $C$.

The core methodological innovation is the definition of an optimal unraveling. Instead of fixing the Kraus decomposition of the noise channel arbitrarily, the authors optimize the decomposition to minimize the fraction of non-Clifford operations, denoted as $\bar{c}$.

1. Geometric Representation: For single-qubit channels, the authors map the channel to a point in a geometric space (complex plane for aligned dephasing, a cone for equatorial isotropic Pauli noise). They identify the convex hull of Clifford operations (a "Clifford diamond" or "pyramid").
2. Cost Minimization: The optimal cost $\bar{c}$ is defined as the minimal weight of the non-Clifford component required to express the noisy channel as a convex combination of a Clifford channel and a non-Clifford component.
3. Analytical and Numerical Verification: Analytical solutions for $\bar{c}$ are derived for aligned dephasing and isotropic Pauli noise. These are verified against numerical CAMPS simulations tracking the maximal entanglement entropy ($S_{\text{max}}$) of the inner MPS.
4. Resource Analysis: The study correlates $\bar{c}$ with trajectory-resolved resources (entanglement entropy, Stabilizer Rényi Entropy) and unraveling-independent measures (Robustness of Magic of the Choi state).

Key Contributions and Results

1. Existence of Noise-Induced Classical Phases:
   The authors identify extended parameter regions where $\bar{c} = 0$. In these regions, the noisy channel admits an unraveling composed entirely of Clifford operations. Consequently, the quantum trajectories remain fully disentanglable by Clifford operations at arbitrary circuit depths ($N_T \to \infty$), rendering the system classically simulable. This phase is explicitly demonstrated for aligned dephasing and specific isotropic Pauli noise models (such as depolarizing noise where $p_\perp = p_z$), but is bounded by threshold probabilities $p_\pm$. Crucially, the paper notes that for generic depolarizing noise or tilted dephasing, the symmetry required for a classical phase may be broken; for instance, in depolarizing noise, the regions of nonzero cost are asymmetric, and for tilted dephasing, the interplay between the noise axis and the non-Clifford rotation can raise the simulation cost and hinder Clifford disentangling.

2. Dependence on Unraveling Choice:
   The existence of these classical phases is not an intrinsic property of the physical noise channel alone but depends critically on the choice of unraveling. A "standard" Kraus decomposition of the same physical channel may yield $\bar{c} > 0$, hiding the classical phase. Conversely, for "tilted" dephasing (where the noise axis is not aligned with the rotation axis), noise can actually increase the simulation cost ($\bar{c} > 1$ in some regimes), making the system harder to simulate than the closed-system limit. However, "twirling" the noise (randomizing the basis) can restore the classical phase by mapping the channel to an isotropic form.

3. Relationship to Quantum Resources:

   • Robustness of Magic (RoM): The optimal cost $\bar{c}$ is analytically bounded by the RoM of the channel's Choi state. For specific noise models, $\bar{c}$ is exactly proportional to the RoM. This establishes that the classical simulability is fundamentally limited by the non-stabilizerness inherent in the channel.
   • Phase Classification: The dynamics are classified into three regimes based on $\bar{c}$, inner-MPS entanglement, and Stabilizer Rényi Entropy:
     • Disentanglable Phase: $\bar{c} > 0$ but finite; non-stabilizerness is present but can be absorbed by the Clifford part; inner-MPS entanglement vanishes.
     • Classical Phase: $\bar{c} = 0$; both inner-MPS entanglement and non-stabilizerness vanish; trajectories are purely Clifford.
     • Quantum Phase: $\bar{c} > 0$ and grows with depth; residual entanglement and non-stabilizerness grow, leading to exponential simulation costs.

4. Failure in the Averaged Description:
   The authors demonstrate that these classical phases disappear when considering the averaged density matrix (unconditioned dynamics). By generalizing CAMPS to Clifford-augmented Matrix Product Operators (CAMPO), they show that the disentangling power of the optimal unraveling is lost. In the density-matrix picture, the system behaves as if it were in the closed-system limit (disentanglable only up to $N_T \sim N$), regardless of the noise strength. This highlights that the "classicality" is a property of the specific stochastic trajectory ensemble, not the unconditional state.

Significance
The paper claims to provide a general understanding of the classical simulability of open quantum systems by redefining the paradigm of optimal unraveling. Rather than optimizing solely for entanglement (as in standard tensor network approaches), the authors propose optimizing for the specific resource (non-Clifford operations) that obstructs the chosen efficient representation (CAMPS).

The results suggest that the emergence of classicality in noisy random circuits is not a universal feature of noise but depends on the interplay between the noise geometry and the measurement scheme (unraveling). The work establishes that optimal unravelings act as an additional computational degree of freedom, capable of exposing classical structures hidden in the density-operator description. This paves the way for future studies on the complexity of driven-dissipative dynamics and the potential extension of these principles to other efficiently simulable manifolds (e.g., Gaussian states for fermionic systems).