Random dimension reduction and learning symmetric properties of quantum states

arXiv:2606.23592 · quant-ph · Submitted 2026-06-22 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Quantum Radio. Generated commentary on the latest quantum physics and condensed matter papers.

Kai: Today's paper: "Random dimension reduction and learning symmetric properties of quantum states".

Mira: Random dimension reduction and learning symmetric properties of quantum states introduces a procedure called random dimension reduction that simultaneously reduces the dimensions of many,

Kai: First, who's behind it and why it matters.

Paper summary: Kai: To summarize this paper, "Random dimension reduction and learning symmetric properties of quantum states," the main thesis is introducing a procedure called random dimension reduction that simultaneously reduces the dimensions of many potentially distinct quantum states while preserving properties that are invariant under the tensor power action of an isometry.

Mira: The central claim here is that this method offers a black-box way to substitute the ambient dimension with something related to the maximum rank found in sample complexity when learning symmetric properties, which applies even when you have multiple input states.

Lev: It sounds like they are proposing a way to handle high-dimensional state spaces without needing to know every single dimension explicitly, which is crucial for scalability in quantum information tasks.

Kai: And the paper shows that applying this dimension reduction followed by full state tomography actually leads to improved upper bounds for estimating distances, fidelities, and relative entropies between pairs of states.

Mira: They also establish an efficient quantum circuit implementation of this procedure utilizing the Schur transform, which is a specific tool they use to achieve that efficiency in the process.

Lev: The authors also connect their dimension reduction channel's action to the random purification channel introduced by Tang, Wright, and Zhandry through the Choi–Jamiołkowski isomorphism, suggesting an integrated analysis of sample-optimal tomography.

Kai: So, it's essentially a framework that simplifies symmetric learning problems by reducing the complexity of the setup while maintaining the essential properties we are trying to measure.

Mira: The motivation is strong because in classical learning theory, symmetric properties don't depend on ambient dimension, and they want an analogous statement for quantum states where the ambient dimension often far exceeds what's needed for rank-based considerations.

Lev: If this framework holds up when we move from theoretical statements to actual physical systems, it opens up new avenues for how we can approach tomography experiments on large quantum systems.

Conclusion: Kai: Looking at "Random dimension reduction and learning symmetric properties of quantum states," this work by Lowe and Tan is really about creating a systematic tool to simplify complex quantum state analysis by effectively managing the ambient dimension through random projection techniques.

Mira: The implication I see is that we might be able to get much tighter, more practical bounds on how accurately we can characterize the relationship between different quantum states when those states are high-dimensional, because this method leverages rank instead of brute force dimension counting.

Lev: For error correction researchers like myself, if we can use these improved complexity bounds for estimating fidelities and distances in our simulations or experimental characterization routines, that has direct value in designing better codes.

Kai: It suggests a path forward where we don't necessarily need to measure every possible dimension of a state; instead, focusing on the rank helps us determine the necessary sample size much more effectively.

Mira: The connection they draw to random purification channels also hints at a deeper structural relationship in quantum information theory, suggesting that these two approaches are fundamentally linked through different mathematical lenses.

Lev: It's interesting how this work moves away from relying on those heavy tools like Schur polynomials for the full analysis, offering an alternative way to achieve the same goals computationally.

Kai: So, in simple terms, this paper provides a method that lets us measure important quantum relationships more efficiently by intelligently reducing the dimensionality of our input data based on its rank.

Center for Theoretical Physics — a Leinweber Institute, MIT

quant-ph

Submitted: 2026-06-22

Updated: 2026-10-02

Comments: 23 pages

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 90/100

The gist: Random dimension reduction and learning symmetric properties of quantum states introduces a procedure called random dimension reduction that simultaneously reduces the dimensions of many, potentially

Key concepts

Random Dimension Reduction (RTP)
This is a quantum channel that reduces the dimension of an operator X by averaging over Haar-random unitaries. It's designed to preserve properties invariant under the tensor power action of an isometry, effectively simplifying complex state structures into a lower-dimensional representation.
Weak Schur Sampling (WSS)
When dealing with identical copies of a single state, WSS is an optimal measurement technique for unitarily invariant properties. It involves projecting onto isotypic components of the symmetric group to efficiently extract the relevant information about the state's structure.
Schur Transform
This mathematical tool decomposes a large representation space into simpler, irreducible components indexed by partitions. The Schur transform allows researchers to apply dimension reduction by measuring these labels, which is then inverted to reconstruct the reduced state information.
Sample Complexity Bounds
These are improved upper limits on how many measurements are needed to estimate quantum properties like distances or fidelities between states. By using dimension reduction, the required sample size is shown to be significantly better than previous methods in low-rank settings.

Terminology

Summary

Random dimension reduction and learning symmetric properties of quantum states introduces a procedure called random dimension reduction that simultaneously reduces the dimensions of many, potentially distinct quantum states while preserving properties invariant under the tensor power action of an isometry. This provides a black-box method to replace the dimension with the maximum rank in the sample complexity of learning symmetric properties, even those depending on multiple input states.

The gist: Random dimension reduction simultaneously reduces the dimensions of many, potentially distinct quantum states while preserving properties invariant under the tensor power action of an isometry.

Random Dimension Reduction Procedure

The core procedure is defined by a quantum channel RTP that satisfies a specific transformation rule. The theorem establishes that for an operator X satisfying a certain structural condition, RTP(X) equals the expected value over Haar-random unitaries:

RTP(X) = E U∼U(k) U⊗nX0(U†)⊗n (1).

This channel is implemented to accuracy ε in diamond norm using poly(n, log d, log(1/ε)) elementary gates. A useful case is when k = d, where RTP equals the unitary twirling channel. When k < d, the behavior differs from simply removing ambient dimension.

Application to Learning Symmetric Properties

Dimension reduction simplifies symmetric learning problems by simplifying the setup. In the special case of identical copies of a single unknown state (X = ρ⊗n), measurements given by projections onto isotypic components of the symmetric group, known as Weak Schur sampling (WSS), are optimal for unitarily invariant properties, and this procedure recovers that fact while going a step further. For two quantum states ρ and σ, each of rank at most r, the dimension reduction channel satisfies Corollary 1.2:

RTP(ρ⊗a ⊗ σ⊗b) = E U∼U(k) h(Uρ0 U†)⊗a ⊗ (Uσ0 U†)⊗b i, where W is an isometry such that ρ = W ρ0W† and σ = W σ0W†.

Connection to Random Purifications

The analysis of the dimension reduction channel connects to the recently introduced random purification channel by Tang, Wright, and Zhandry. The action of the dimension reduction channel on states of the form ρ⊗n is recovered by first applying this random purification channel and then discarding the original system. Conversely, random purifications arise naturally from considering the action of dimension reduction on a larger Hilbert space through the Choi–Jamiokowski isomorphism, providing an end-to-end analysis without requiring a reference to the Schur transform or Schur polynomials. However, it is proven that there does not exist a random purification channel that simultaneously purifies copies of multiple, potentially different input states; hence, random dimension reduction is distinct from random purification in all generality.

Efficient Implementation via the Schur Transform

The efficient implementation of the procedure relies on representation theory and the Schur transform. The representation space HAn = (Cd)⊗n has a canonical decomposition into isotypic components LλVλ, indexed by partitions λ⊢ n of length at most l(λ) ⩽ d. The Schur transform Ud,nSchur intertwines these spaces:

Ud,nSchur FdπM⊗n (Ud,nSchur)† = Mλ⊢n: l(λ)≤d κλ(π) ⊗ νdλ(M).

Algorithm 3.2 describes the implementation of RTP using this transform: apply Ud,nSchur, measure the label λ⊢ n (weak Schur sampling), and then apply the inverse Schur transform Uk,nSchur†. This entire algorithm can be implemented to accuracy ε in diamond distance using a quantum circuit of size polynomial in n, log d, and log(1/ε).

Sample Complexity Bounds

The dimension reduction followed by full state tomography yields improved upper bounds for estimating distances, fidelities, and relative entropies between pairs of states. For example:

)&Trace distance:

O(r squared · log2(1/ε)/ε4) [WZ24a; CWZ25]

)Fidelity:

O(r squared · log2(1/ε)/ε4) [UNWT25]

These bounds are derived from the observation that for a symmetric function f, if F(ρ, ρb) ⩾ 1 − δ and F(σ, σb) ⩾ 1 − δ implies f(ρ, σ) − f(ρ, b σb) ≤ ε (continuity), then the sample complexity is O(r 2/δ) via full state tomography on the dimension-reduced states. This leads to improved bounds compared to prior work in the low rank setting.

Improvements for AI systems

Here are specific ways to improve AI systems based on the provided scientific paper, focusing on leveraging its core findings in dimension reduction and symmetric properties:

  1. Do not rely solely on full state tomography for estimating quantum properties (like distances or fidelities) between unknown quantum states. Instead, implement the Random Dimension Reduction procedure described in Theorem 1.1 and Algorithm 3.2 to estimate these properties using significantly fewer samples (sample complexity improvements shown in Table 1).

  2. Develop a method for efficiently estimating symmetric properties of multiple, potentially distinct quantum states simultaneously by applying the dimension reduction channel to a composite state (Corollary 1.2). This allows the AI system to infer relationships between several unknown quantum objects with better precision than methods requiring independent tomography on each state individually.

  3. Implement a black-box method for reducing the effective dimensionality of complex quantum data structures (like density matrices) while guaranteeing that key physical symmetries (invariant under the tensor power action of an isometry) are preserved. This can be used to simplify large-scale quantum simulations or machine learning models operating on high-dimensional Hilbert spaces.

  4. Integrate the connection between dimension reduction and Random Purification into state estimation protocols (Theorem 2.5). This allows for a more conceptually unified approach where the process of reducing dimensionality is viewed as a physical purification step, potentially leading to more robust and interpretable quantum inference algorithms.

  5. Use the Schur transform as an efficient computational tool for implementing the dimension reduction channel (Theorem 3.3). This provides a concrete, polynomial-time implementation for performing these complex reductions on quantum hardware, making the theoretically superior sampling bounds practically accessible on real quantum computers using poly(n, log d, log(1/ε)) elementary gates.

  6. Develop a rank-aware estimation strategy: if an AI system is dealing with states whose ranks are known or bounded (rank at most r), use the results in Section 1.2 to exploit the minimum rank dependence in sample complexity bounds, potentially leading to even tighter estimation guarantees than those based solely on maximum rank.

Sources

Related papers