Schur Sampling in Quantum Systems

• Schur Sampling is a quantum measurement process based on Schur–Weyl duality that extracts representation-theoretic information from multipartite qudit systems.
• It encompasses variants like weak, strong, and unitary sampling, each measuring different levels of the Schur-basis information for flexible quantum state analysis.
• Recent work demonstrates that sequential and streaming implementations achieve logarithmic memory and efficient gate usage, outperforming full Schur transforms.

Schur sampling, in the quantum-information sense, is the family of measurement tasks induced by the Schur–Weyl decomposition of multipartite qudit systems. It extracts representation-theoretic information associated with the commuting actions of the symmetric group and the unitary group, typically without requiring the full output of the quantum Schur transform. The basic form is weak Schur sampling, which measures only the Young label indexing the isotypic component; stronger variants retain additional registers, including the post-measurement unitary-irrep state. Recent work has recast Schur sampling as a streaming task with logarithmic quantum memory, generalized it to mixed Schur–Weyl settings, and connected it to both permutation-invariant SWAP-test protocols and classical-simulation results for Schur-transform-based circuit families (Cervero et al., 2023).

1. Representation-theoretic setting

For $n$ $d$-level systems, Schur sampling is grounded in Schur–Weyl duality, which decomposes the Hilbert space as

$$(\mathbb{C}^d)^{\otimes n}\cong \bigoplus_{\lambda\vdash n}\mathcal{P}_\lambda\otimes \mathcal{Q}^d_\lambda .$$

Here $\lambda$ is a Young label, i.e. a partition of $n$ with at most $d$ parts; $\mathcal{P}_\lambda$ is an irreducible representation of the symmetric group $S_n$; and $\mathcal{Q}^d_\lambda$ is an irreducible representation of the unitary group $U(d)$. A Schur-basis state is labeled by a triple $d$0, where $d$1 identifies the isotypic component, $d$2 indexes a basis vector in $d$3, and $d$4 indexes a basis vector in $d$5 (Cervero et al., 2023).

The quantum Schur transform is the unitary change of basis from the computational basis to this Schur basis. In this formulation, Schur sampling is not a distinct decomposition from the Schur transform; rather, it is the task of extracting some subset of the Schur-basis labels after, or instead of, implementing the full transform. For qubits, the Young labels reduce to two-row partitions $d$6, and the unitary irrep dimension becomes

$d$7

This qubit case is the simplest nontrivial setting and underlies several explicit resource estimates.

Formally, weak Schur sampling measures the projectors onto the isotypic components,

$d$8

so that the output distribution is

$d$9

This establishes Schur sampling as a measurement problem determined entirely by the representation structure of the input state.

2. Operational variants and task hierarchy

The standard variants differ by how much of the Schur-basis information is retained. The distinction between weak and strong Schur sampling is basic: weak Schur sampling measures only the Young label $$(\mathbb{C}^d)^{\otimes n}\cong \bigoplus_{\lambda\vdash n}\mathcal{P}_\lambda\otimes \mathcal{Q}^d_\lambda .$$0, while strong Schur sampling measures the full triple $$(\mathbb{C}^d)^{\otimes n}\cong \bigoplus_{\lambda\vdash n}\mathcal{P}_\lambda\otimes \mathcal{Q}^d_\lambda .$$1. In the contemporary literature, a further refinement is unitary Schur sampling, which keeps the unitary-group register after measuring $$(\mathbb{C}^d)^{\otimes n}\cong \bigoplus_{\lambda\vdash n}\mathcal{P}_\lambda\otimes \mathcal{Q}^d_\lambda .$$2, and a mixed analogue adapted to the mixed Schur–Weyl transform (Cervero-Martín et al., 2024).

Task	Output	Setting
Weak Schur sampling	$$(\mathbb{C}^d)^{\otimes n}\cong \bigoplus_{\lambda\vdash n}\mathcal{P}_\lambda\otimes \mathcal{Q}^d_\lambda .$$3	Measure only the Young label
Strong Schur sampling	$$(\mathbb{C}^d)^{\otimes n}\cong \bigoplus_{\lambda\vdash n}\mathcal{P}_\lambda\otimes \mathcal{Q}^d_\lambda .$$4	Measure the full Schur-basis triple
Unitary Schur sampling	$$(\mathbb{C}^d)^{\otimes n}\cong \bigoplus_{\lambda\vdash n}\mathcal{P}_\lambda\otimes \mathcal{Q}^d_\lambda .$$5 and $$(\mathbb{C}^d)^{\otimes n}\cong \bigoplus_{\lambda\vdash n}\mathcal{P}_\lambda\otimes \mathcal{Q}^d_\lambda .$$6	Retain only the post-measurement state on $$(\mathbb{C}^d)^{\otimes n}\cong \bigoplus_{\lambda\vdash n}\mathcal{P}_\lambda\otimes \mathcal{Q}^d_\lambda .$$7
Unitary mixed Schur sampling	$$(\mathbb{C}^d)^{\otimes n}\cong \bigoplus_{\lambda\vdash n}\mathcal{P}_\lambda\otimes \mathcal{Q}^d_\lambda .$$8 and $$(\mathbb{C}^d)^{\otimes n}\cong \bigoplus_{\lambda\vdash n}\mathcal{P}_\lambda\otimes \mathcal{Q}^d_\lambda .$$9	Mixed Schur–Weyl setting with staircase label

Most applications described in the weak-sampling literature require only $\lambda$0, not the finer labels. By contrast, unitary Schur sampling is motivated as the natural task in settings such as spectrum estimation and quantum state tomography, because the permutation register is operationally irrelevant after the representation label has been measured. In the mixed setting, the input space is

$\lambda$1

the commuting algebraic structure is governed by the walled Brauer algebra, and the label is a staircase $\lambda$2 rather than a partition.

A common source of confusion is the relation between these tasks and the full Schur transform. Weak and unitary Schur sampling are output tasks: they ask for the representation label, or for the label plus the unitary-side state, rather than for a coherent realization of every Schur-basis register. This distinction is central to the newer streaming algorithms.

3. Sequential weak Schur sampling

A central development is the streaming algorithm for weak Schur sampling introduced in “Weak Schur sampling with logarithmic quantum memory” (Cervero et al., 2023). Rather than first applying the full Schur transform to all $\lambda$3 qudits, the algorithm processes the input sequentially. It starts with the first qudit, where $\lambda$4; for each incoming qudit, it applies a single Clebsch–Gordan transform that updates the current Schur decomposition from $\lambda$5 to $\lambda$6 qudits; it immediately measures which new partition $\lambda$7 occurred; and it discards unused space, keeping only a small register that describes the current irrep.

The relevant branching step is

$\lambda$8

with invalid partitions omitted. The key observation is that, after one more qudit is added, only a small set of nearby partitions can occur. By measuring after each incremental Clebsch–Gordan update, the protocol never stores the full $\lambda$9-qudit state. It keeps only the irrep currently being tracked and the label of the chosen branch.

The paper proves that this sequential procedure has exactly the same output distribution as measuring the Schur label after a full Schur transform. Its proof identifies each possible sequence of intermediate branching choices, i.e. each path through partitions, with the multiplicity label $n$0, and then shows that summing over all such paths reproduces the weak Schur measurement probability. This makes the method an exact weak Schur sampling protocol rather than an approximation to the Schur transform.

The streaming formulation has an immediate operational consequence: qudits can be processed online as they arrive, and the protocol can be stopped at any time while still outputting the current weak Schur label for the prefix already seen. This suggests a natural fit with limited-memory and sequential quantum devices.

4. Resource bounds and memory efficiency

The principal technical claim of the 2023 weak-sampling algorithm is its logarithmic memory usage (Cervero et al., 2023). For $n$1 qubits and accuracy $n$2, the paper gives an explicit implementation requiring

$n$3

of memory and

$n$4

gates from the Clifford$n$5 set. For general qudits, it gives a memory requirement of

$n$6

and the detailed resource discussion also phrases the bound as $n$7 qudits.

The memory reduction is described as exponential relative to approaches that store the entire Schur-form representation of all $n$8 qudits. The reason is that the protocol stores only the current Young label $n$9, the current irrep state $d$0, and a small amount of workspace for the incremental Clebsch–Gordan update. Since, for fixed $d$1, the dimension of the relevant irrep grows only polynomially with $d$2, its encoding length is logarithmic in $d$3. This yields $d$4 memory instead of the $d$5 memory used by full-transform or Generalized Phase Estimation approaches.

The comparison with prior methods has two aspects. Relative to the full quantum Schur transform, the new procedure is streaming or online and avoids storing the whole input in Schur form. Relative to Generalized Phase Estimation, it preserves efficiency in gates while improving memory exponentially. The paper also derives the general qudit gate decomposition through a bound

$d$6

on the number of two-level unitaries needed for the relevant Clebsch–Gordan transforms, followed by Solovay–Kitaev-style approximation overhead.

5. Extensions, implementations, and applications

The 2024 paper “A memory and gate efficient algorithm for unitary mixed Schur sampling” formalizes unitary Schur sampling and extends the streaming paradigm to the mixed Schur–Weyl setting (Cervero-Martín et al., 2024). In the ordinary setting, unitary Schur sampling outputs the Young label $d$7 together with the reduced post-measurement state on $d$8, obtained by tracing out the permutation register $d$9. In the mixed setting, the label becomes a staircase $\mathcal{P}_\lambda$0, and the algorithm alternates ordinary Clebsch–Gordan and dual Clebsch–Gordan updates as the input traverses the fundamental and conjugate sectors. The measured label evolves as a path

$\mathcal{P}_\lambda$1

where each step adds a box for the first $\mathcal{P}_\lambda$2 qudits and removes a box for the last $\mathcal{P}_\lambda$3 qudits.

For the full mixed setting, the paper gives resource bounds

$\mathcal{P}_\lambda$4

with $\mathcal{P}_\lambda$5, and describes these as an exponential improvement in memory complexity in the constant-$\mathcal{P}_\lambda$6 regime, together with a polynomial improvement in gate complexity over naive implementations. Under the rank promise

$\mathcal{P}_\lambda$7

the bounds improve to

$\mathcal{P}_\lambda$8

The same paper explicitly motivates these tasks by spectrum estimation, quantum state tomography, entanglement concentration, purification, and majority voting.

A distinct implementation route appears in “Optimal Qubit Purification and Unitary Schur Sampling via Random SWAP Tests” (Brahmachari et al., 7 Aug 2025). There the task is realized, for permutation-invariant qubit states, by repeatedly applying random two-qubit SWAP tests that project onto singlet or triplet subspaces. The protocol removes detected singlet pairs and continues on the surviving qubits. The paper states that after approximately $\mathcal{P}_\lambda$9 random SWAP tests a sharp transition occurs, and that weak Schur sampling and unitary Schur sampling can be achieved with error $S_n$0 after only

$S_n$1

SWAP tests. It also proves the trace-distance bound

$S_n$2

and identifies the method as a lossless way of extracting the permutation-invariant content of the state. In this qubit setting, Schur sampling is connected directly to purification, tomography, and metrology.

6. Classical simulation and computational limits

Schur sampling has also been studied from the perspective of classical simulation, particularly in circuit families built around the $S_n$3 Schur transform. “Quantum Schur Sampling Circuits can be Strongly Simulated” shows that a broad class of Schur-sampling and permutational-quantum-computing circuits are classically tractable at the level of transition amplitudes: the relevant amplitudes can be efficiently approximated up to polynomial additive precision, including matrix elements of $S_n$4 irreducible representations in Young’s orthogonal form (Havlicek et al., 2018). The argument uses computationally tractable states, explicit computation of Clebsch–Gordan coefficients via Wigner $S_n$5 symbols and Racah’s formula, and Monte Carlo overlap estimation with Chernoff–Hoeffding bounds.

“A Classical Algorithm for Quantum $S_n$6 Schur Sampling” sharpens the simulation picture for output distributions (Havlíček et al., 2018). It proves that sufficiently large output probabilities in these circuits can be located classically in polynomial time, and that if the output distribution is $S_n$7-approximately $S_n$8-sparse, then it can be classically sampled in

$S_n$9

time to within $\mathcal{Q}^d_\lambda$0 total variation distance. In that framework, approximate sparsity rather than exact sparsity is the decisive criterion. The paper uses this to disprove an extended hardness conjecture for Permutational Quantum Computing based on the alleged difficulty of finding large amplitudes.

These simulation results do not negate the operational relevance of Schur sampling as a measurement primitive. They instead delimit the regimes in which Schur-transform-based quantum algorithms might exhibit super-classical behavior. As stated in the simulation literature, plausible quantum advantage survives only when the output distribution is not close to sparse and large probabilities cannot be resolved from polynomially many classical samples. This suggests that the central difficulty is not the mere presence of Schur structure, but the distributional regime in which that structure is accessed.