Query Learning Nearly Pauli Sparse Unitaries in Diamond Distance

Abstract: We study the problem of learning nearly $(s,ε)$-sparse unitaries, meaning that the Pauli spectrum is concentrated on at most $s$ components with at most $ε$ residual mass in Pauli $\ell_1$-norm. This class generalizes well-studied families, including sparse unitaries, quantum $k$-juntas, $2^k$-Pauli dimensional channels, and compositions of depth $O(\log\log n)$ circuits with near-Clifford circuits. Given query access to an unknown nearly sparse unitary $U$, our goal is to efficiently (both in time and query complexity) construct a quantum channel that is close in diamond distance to $U$. We design a learning algorithm achieving this guarantee using $\tilde{O}(s^6/ε^4)$ forward queries to $U$, and running time polynomial in relevant parameters. A key contribution is an efficient quantum algorithm that, given query access to an arbitrary unknown unitary $U$, estimates all Pauli coefficients (up to a shared global phase) whose magnitude exceeds a given threshold $θ$, extending existing sparse recovery techniques to general unitaries. We also study the broader class of unitaries with bounded Pauli $\ell_1$-norm. For that class, we prove an exponential query lower bound $Ω(2^{n/2})$. We introduce a more relaxed accuracy metric which is the diamond distance restricted to a set of input states. Then, we show that, under this metric, unitaries with Pauli $\ell_1$-norm uniformly bounded by $L_1$ are learnable with $\tilde{O}(L_1^8/ε^{16})$.

• The paper introduces an efficient algorithm for learning nearly (s,ε)-sparse unitaries, achieving O(s^6/ε^4) query complexity for diamond norm closeness.
• It employs a two-step process: Bell sampling for support identification and Clifford shadow tomography for precise estimation of dominant Pauli coefficients.
• The approach leverages LCU techniques for directly implementing the learned unitary, offering practical implications for quantum device verification and benchmarking.

Efficient Query Learning of Nearly Pauli Sparse Unitaries in Diamond Distance

Problem Formulation and Theoretical Context

The paper addresses the quantum query learning of $n$-qubit unitaries whose Pauli spectrum is nearly $(s,\epsilon)$-sparse: the cumulative $\ell_1$-norm outside a subset $S$ of at most $s$ dominant Pauli strings is at most $\epsilon$. The focus is on diamond-norm closeness between the unknown target unitary $U$ (accessed via queries) and the hypothesis constructed by the learner. This class generalizes exact $s$-sparsity, quantum $k$-juntas, and unitaries arising from low-depth, near-Clifford circuits. The diamond norm is the operationally relevant metric, measuring worst-case channel distinguishability over all inputs, including those possibly entangled with external references.

Crucially, without structure, unitary process tomography is intractable due to the exponential growth in degrees of freedom—$\Omega(4^n/\epsilon)$ queries are required for general $(s,\epsilon)$0-qubit unitaries. The central question is which structural priors and error metrics allow sub-exponential query complexity and feasible runtime.

Main Algorithmic Contributions

The core technical result is an efficient algorithm for query learning nearly $(s,\epsilon)$1-sparse unitaries $(s,\epsilon)$2, outputting a quantum channel that is $(s,\epsilon)$3-close to $(s,\epsilon)$4 in diamond distance, using only forward queries to $(s,\epsilon)$5. The query complexity is $(s,\epsilon)$6 and runtime is polynomial in $(s,\epsilon)$7, $(s,\epsilon)$8, and $(s,\epsilon)$9.

Estimating Large Pauli Coefficients

A key device is a two-step process for support identification and coefficient estimation for the dominant Pauli terms:

1. Support Identification via Bell Sampling: The Choi-Jamiołkowski state of $\ell_1$0 is prepared, and Bell-basis measurements probabilistically identify Pauli strings with squared amplitudes $\ell_1$1. $\ell_1$2 samples suffice to capture all strings with $\ell_1$3.
2. Coefficient Estimation via Clifford Shadow Tomography: For each significant index, expectation values of engineered observables (constructed from Bell basis projectors and their linear combinations) on Choi-states yield the real and imaginary parts of the relative Pauli coefficients, up to a global phase. Shadow tomography with random Clifford unitaries reduces the classical post-processing to polylogarithmic overhead in the number of observables, and the total query complexity scales as $\ell_1$4 for the required coefficient precision.

Global phase ambiguity is shown to be irrelevant for the induced unitary channel, due to the phase alignment in the induced operator norm and diamond distance bounds.

Block Encoding and Efficient Implementation

The resulting estimate $\ell_1$5 is an $\ell_1$6-sparse (or nearly sparse) linear combination of Pauli operators, which may not be exactly unitary. The paper leverages the Linear Combination of Unitaries (LCU) technique, augmented by robust oblivious amplitude amplification, to promote $\ell_1$7 to an efficiently implementable quantum process. The cost is proportional in the normalization factor $\ell_1$8.

Query Complexity Bounds and Lower Bounds

For nearly $\ell_1$9-sparse unitaries, the optimal worst-case lower bound is $S$0 queries in diamond norm, matching the upper bound up to polynomial factors. For the strictly wider class of unitaries with bounded Pauli $S$1-norm $S$2, the diamond norm is too stringent: the paper proves an exponential lower bound $S$3 for this class, via Grover oracles which have fixed $S$4-norm yet are perfectly distinguishable with adversarial queries.

To remedy this, the authors propose a restricted diamond distance, confining distinguishability to input states with fixed marginal (e.g., maximally mixed). Under this weaker but still operationally meaningful metric, an algorithm is provided—again based on support identification and coefficient estimation—which learns such $S$5-bounded unitaries up to error $S$6 in restricted diamond norm with query complexity $S$7.

Strong Claims and Numerical Guarantees

• Diamond-norm learning of $S$8-nearly Pauli-sparse unitaries with query complexity $S$9.
• Learning under restricted diamond norm for $s$0-bounded unitaries with $s$1 queries.
• Separation between learnability in diamond and restricted diamond norm for the $s$2-bounded class, via explicit exponential lower bounds.
• No need for special access models (e.g., short-time Hamiltonian evolution, controlled unitaries, or subgroup sparsity)—forward queries to $s$3 suffice.

Practical and Theoretical Implications

Practical Implications

The results yield sample-optimal and computationally feasible methods for learning unitaries in quantum devices, provided the underlying process is Pauli-compressible. This includes verification, benchmarking, or characterization protocols for near-term quantum circuits that are not fully random nor highly entangled, but exhibit effective sparsity due to device constraints or algorithmic choices.

The LCU-based block encoding ensures that the learned hypothesis is not only abstractly close but also directly implementable as a quantum circuit. The methods are compatible with architectures supporting Clifford group gates and efficient Bell state preparation/measurement.

Theoretical Implications and Future Work

The study delineates the learnable frontier for quantum process learning under different structural assumptions, highlighting the criticality of the error metric (diamond vs. restricted diamond) and input model. The methods extend classical sparse Fourier learning (Kushilevitz-Mansour, Goldreich-Levin) into the quantum process domain, adapting support estimation to the non-Hermitian and phase-ambiguous setup of arbitrary unitaries.

Potential future research includes:

• Reducing the degree of polynomial dependence on $s$4 and $s$5, possibly via more aggressive isolation techniques, improved shadow tomography or Heisenberg scaling.
• Extending methods to adaptive or time-continuous query models, or leveraging controlled unitary access if available.
• Investigating learnability for new classes of operators, especially those defined by circuit depth or gate complexity rather than Pauli spectral structure.

Conclusion

This work establishes precise query complexity bounds and efficient quantum algorithms for learning nearly Pauli-sparse unitaries, optimal up to polynomial factors, in diamond or restricted diamond distance. The analysis delineates sharp phase transitions in process learnability—dictated by the interplay of structural priors and operational metrics—and provides practical tools for quantum process characterization in the regime of effective sparsity.

For full details, proofs, and algorithms, see "Query Learning Nearly Pauli Sparse Unitaries in Diamond Distance" (2604.00203).