Robust Phase Shadow Scheme

• The paper introduces a robust phase-shadow protocol that estimates off-diagonal quantum state properties using controlled-Z gates and simple circuit ensembles, matching full Clifford shadow performance.
• It decomposes quantum states into diagonal and off-diagonal sectors to provide unbiased estimators with precise variance bounds and balanced measurement rounds.
• The scheme extends to a noise-robust variant using classical deconvolution for efficient post-processing of stabilizer observables on ion and neutral-atom platforms.

Robust Phase Shadow Scheme is a classical-shadow measurement framework for estimating global quantum properties under realistic noise by using random circuits in which controlled-Z is the unique entangling gate type. In its canonical formulation, the protocol targets the off-diagonal sector of an $n$-qubit state, combines phase-shadow rounds with computational-basis rounds for the diagonal sector, and attains variance guarantees that are described as “almost the same” as those of full Clifford shadows while using a substantially simpler gate set. The same framework admits a noise-robust extension based entirely on classical post-processing, with strict unbiasedness retained under gate-dependent $Z$-type noise models relevant to trapped-ion and neutral-atom platforms (Zhang et al., 17 Jul 2025).

1. Circuit ensemble and basic object of estimation

The phase-shadow ensemble is defined by circuits of the form

$$U = H^{\otimes n} U_A \in \mathcal{E}_{\mathrm{phase}},$$

with

$$U_A = \prod_{0 \le i < j \le n-1} CZ_{i,j}^{A_{i,j}} \prod_{k \in [n]} S_k^{A_{k,k}},$$

where $A$ is a random symmetric binary matrix with i.i.d. entries $A_{i,j}\in\{0,1\}$ for $i\le j$, each with probability $1/2$. For a computational-basis outcome $b\in\{0,1\}^n$, the associated rank-one operator is

$\Phi_{U,b} := U^\dagger |b\rangle\langle b| U.$

This circuit architecture is described as CZ–S–H: a random layer of two-qubit $Z$0 gates and single-qubit $Z$1 gates, followed by a fixed $Z$2 layer before computational-basis measurement (Zhang et al., 17 Jul 2025).

The protocol does not reconstruct the full state in a single algebraic step. Instead, it isolates the off-diagonal component

$Z$3

A central design feature is therefore the decomposition of observables and states into diagonal and off-diagonal sectors. For global properties such as fidelities to multipartite target states, the off-diagonal component is estimated by phase-shadow rounds, while the diagonal component is obtained from direct computational-basis measurements (Zhang et al., 17 Jul 2025).

This decomposition is specific to the algebra generated by the phase-shadow ensemble. It is also the reason the protocol is especially natural on architectures where controlled-phase interactions are native or efficiently compiled, since the entangling layer is restricted to $Z$4 gates rather than arbitrary Clifford entanglers (Zhang et al., 17 Jul 2025).

2. Moment structure and induced measurement channel

For an ensemble $Z$5 on $Z$6, with $Z$7, the $Z$8-th moment function is defined as

$Z$9

For the phase ensemble, the defining structural statement is the proposition that for $$U = H^{\otimes n} U_A \in \mathcal{E}_{\mathrm{phase}},$$0,

$$U = H^{\otimes n} U_A \in \mathcal{E}_{\mathrm{phase}},$$1

where $$U = H^{\otimes n} U_A \in \mathcal{E}_{\mathrm{phase}},$$2 is the unitary representation of the permutation $$U = H^{\otimes n} U_A \in \mathcal{E}_{\mathrm{phase}},$$3 on $$U = H^{\otimes n} U_A \in \mathcal{E}_{\mathrm{phase}},$$4, and the union is the bitwise OR operation for operators written in the computational basis (Zhang et al., 17 Jul 2025).

For $$U = H^{\otimes n} U_A \in \mathcal{E}_{\mathrm{phase}},$$5, this specializes to

$$U = H^{\otimes n} U_A \in \mathcal{E}_{\mathrm{phase}},$$6

with $$U = H^{\otimes n} U_A \in \mathcal{E}_{\mathrm{phase}},$$7 and $$U = H^{\otimes n} U_A \in \mathcal{E}_{\mathrm{phase}},$$8. The corresponding measurement channel is

$$U = H^{\otimes n} U_A \in \mathcal{E}_{\mathrm{phase}},$$9

and for the phase-shadow ensemble one obtains

$$U_A = \prod_{0 \le i < j \le n-1} CZ_{i,j}^{A_{i,j}} \prod_{k \in [n]} S_k^{A_{k,k}},$$0

Accordingly, the channel preserves the identity, retains the off-diagonal sector, and removes the diagonal sector from the effective phase-shadow output (Zhang et al., 17 Jul 2025).

This algebraic form distinguishes phase shadows from Haar or full Clifford constructions. The paper emphasizes that the relevant moment object is not a weighted sum over permutation operators, but a union-of-permutations structure. A plausible implication is that the protocol’s statistical behavior is controlled by a highly specific diagonal-circuit combinatorics rather than by approximate unitary-design arguments.

3. Unbiased estimators and statistical guarantees

The basic noiseless estimator for the off-diagonal component is

$$U_A = \prod_{0 \le i < j \le n-1} CZ_{i,j}^{A_{i,j}} \prod_{k \in [n]} S_k^{A_{k,k}},$$1

It satisfies

$$U_A = \prod_{0 \le i < j \le n-1} CZ_{i,j}^{A_{i,j}} \prod_{k \in [n]} S_k^{A_{k,k}},$$2

For any observable $$U_A = \prod_{0 \le i < j \le n-1} CZ_{i,j}^{A_{i,j}} \prod_{k \in [n]} S_k^{A_{k,k}},$$3, the off-diagonal contribution $$U_A = \prod_{0 \le i < j \le n-1} CZ_{i,j}^{A_{i,j}} \prod_{k \in [n]} S_k^{A_{k,k}},$$4 is estimated through

$$U_A = \prod_{0 \le i < j \le n-1} CZ_{i,j}^{A_{i,j}} \prod_{k \in [n]} S_k^{A_{k,k}},$$5

with variance bounded by

$$U_A = \prod_{0 \le i < j \le n-1} CZ_{i,j}^{A_{i,j}} \prod_{k \in [n]} S_k^{A_{k,k}},$$6

where $$U_A = \prod_{0 \le i < j \le n-1} CZ_{i,j}^{A_{i,j}} \prod_{k \in [n]} S_k^{A_{k,k}},$$7 is the off-diagonal part of $$U_A = \prod_{0 \le i < j \le n-1} CZ_{i,j}^{A_{i,j}} \prod_{k \in [n]} S_k^{A_{k,k}},$$8 and $$U_A = \prod_{0 \le i < j \le n-1} CZ_{i,j}^{A_{i,j}} \prod_{k \in [n]} S_k^{A_{k,k}},$$9 is the Frobenius norm (Zhang et al., 17 Jul 2025).

For projector observables $A$0, this immediately yields

$A$1

The diagonal contribution is handled separately by computational-basis sampling, with estimator

$A$2

and variance

$A$3

If $A$4 phase-shadow rounds and $A$5 computational-basis rounds are used, then

$A$6

and the protocol can be balanced so that

$A$7

with $A$8 (Zhang et al., 17 Jul 2025).

The corresponding sample-complexity statement is

$A$9

with $A_{i,j}\in\{0,1\}$0. The paper explicitly contrasts this with the $A_{i,j}\in\{0,1\}$1 bound of full Clifford shadows and states that phase shadows match the performance of global Clifford shadows up to small constant factors (Zhang et al., 17 Jul 2025).

4. Noise-robust extension and classical deconvolution

The principal noise model places the dominant imperfection in the $A_{i,j}\in\{0,1\}$2 layer: $A_{i,j}\in\{0,1\}$3 with $A_{i,j}\in\{0,1\}$4 i.i.d. $A_{i,j}\in\{0,1\}$5. Averaging gives a Pauli channel

$A_{i,j}\in\{0,1\}$6

whenever $A_{i,j}\in\{0,1\}$7 is applied. The paper also analyzes an extended biased $A_{i,j}\in\{0,1\}$8-type model,

$A_{i,j}\in\{0,1\}$9

motivated in part by native error structure on Rydberg platforms (Zhang et al., 17 Jul 2025).

Under noise, the second moment acquires a Pauli decomposition

$i\le j$0

For Pauli operators of the form

$i\le j$1

the coefficient is

$i\le j$2

A key structural fact is that $i\le j$3 for $i\le j$4, except $i\le j$5. This means that inversion is well-defined precisely on the non-$i\le j$6 sector, i.e. on the off-diagonal component targeted by the protocol (Zhang et al., 17 Jul 2025).

The robust single-shot estimator is therefore

$i\le j$7

and it obeys

$i\le j$8

For observables,

$i\le j$9

with

$1/2$0

The resulting sample complexity scales as

$1/2$1

This identifies a precise bias–variance trade-off: the estimator remains unbiased under the modeled noise, but the variance grows exponentially in $1/2$2 once the small-$1/2$3 regime is left (Zhang et al., 17 Jul 2025).

5. Efficient post-processing for stabilizer observables

A major computational feature of the robust phase-shadow scheme is an efficient post-processing algorithm for stabilizer-state observables. If $1/2$4 is a stabilizer projector and $1/2$5 is a phase-shadow snapshot, then both are stabilizer density operators. The robust estimator reduces to a sum over the phaseless intersection of the corresponding stabilizer groups. The paper states an informal proposition:

“Suppose that the observable $1/2$6 is a stabilizer state, the post-processing of robust phase estimation using Eq. (9) is efficient, specifically $1/2$7, in expectation.” (Zhang et al., 17 Jul 2025)

The mechanism behind this statement is the bound

$1/2$8

with a tail bound $1/2$9 for $b\in\{0,1\}^n$0, where $b\in\{0,1\}^n$1 is the size of the shared stabilizer structure. Expected runtime is then dominated by tableau operations rather than by an exponentially large Pauli expansion (Zhang et al., 17 Jul 2025).

The algorithmic workflow uses the $b\in\{0,1\}^n$2-Tableau of

$b\in\{0,1\}^n$3

performs Gaussian elimination to identify the subspace of $b\in\{0,1\}^n$4 whose conjugates lie in the relevant phaseless stabilizer intersection, and then accumulates only the contributing $b\in\{0,1\}^n$5 terms. For Pauli observables $b\in\{0,1\}^n$6, the estimator simplifies to

$b\in\{0,1\}^n$7

which is computable via tableau conjugation in $b\in\{0,1\}^n$8, while $b\in\{0,1\}^n$9 itself is obtained in $\Phi_{U,b} := U^\dagger |b\rangle\langle b| U.$0 (Zhang et al., 17 Jul 2025).

This efficient post-processing addresses a bottleneck that is common in shadow protocols for global properties. In the robust phase-shadow setting, the simplification relies on the random stabilizer structure induced by the CZ–S–H ensemble rather than on a generic Clifford inverse.

6. Platform relevance and relation to broader robust-shadow literature

The phase-shadow construction is explicitly tailored to architectures “such as trapped ions and neutral atoms,” where controlled-phase interactions are native or efficiently synthesized. The paper notes that in trapped ions, $\Phi_{U,b} := U^\dagger |b\rangle\langle b| U.$1 layers can be compiled via MS gates with global interactions and that over-rotation noise matches the $\Phi_{U,b} := U^\dagger |b\rangle\langle b| U.$2 model, while in neutral atoms, controlled-phase operations are native and biased $\Phi_{U,b} := U^\dagger |b\rangle\langle b| U.$3-type noise is naturally accommodated by the extended $\Phi_{U,b} := U^\dagger |b\rangle\langle b| U.$4 model (Zhang et al., 17 Jul 2025).

Within the broader robust-shadow literature, the scheme occupies a distinct position. “Robust shadow estimation” introduced a calibration stage that learns the effective noisy shadow channel and inverts it so as to obtain an unbiased estimate under gate-independent, stationary, Markovian noise (Chen et al., 2020). “Robust ultra-shallow shadows” showed that shallow noisy frames can be inverted when the effective measurement map remains locally unitarily invariant, with the noisy frame learned directly from experimental data and represented efficiently as a tensor network (Farias et al., 2024). “Local robust shadows on a trapped ion computer — a case study” experimentally demonstrated alternating calibration and estimation stages, together with Pauli-$\Phi_{U,b} := U^\dagger |b\rangle\langle b| U.$5-twirling before measurements, to mitigate measurement errors on a trapped-ion device (Wilkens et al., 30 Mar 2026).

Robust Phase Shadow Scheme differs from those constructions in two linked respects. First, it restricts the entangling layer to $\Phi_{U,b} := U^\dagger |b\rangle\langle b| U.$6, rather than starting from a global Clifford or generic shallow ensemble. Second, its robust inversion is not formulated as a generic learned-frame correction but as an explicit Pauli-sector deconvolution on the non-$\Phi_{U,b} := U^\dagger |b\rangle\langle b| U.$7 sector, where the coefficients $\Phi_{U,b} := U^\dagger |b\rangle\langle b| U.$8 are analytically characterized under the noise models considered (Zhang et al., 17 Jul 2025). This makes the protocol particularly suited to estimating global quantities such as fidelities to complex multipartite stabilizer targets while retaining strict unbiasedness under the modeled gate-dependent noise.

A plausible implication is that the protocol represents a specialization of robust-shadow methodology to controlled-phase-native hardware: the statistical guarantees remain close to those of full Clifford shadows, but the circuit ensemble, noise model, and post-processing are all aligned with the algebra of $\Phi_{U,b} := U^\dagger |b\rangle\langle b| U.$9-generated random stabilizer states rather than with the full Clifford group (Zhang et al., 17 Jul 2025).