PaperlayerView exact revision on arXiv

Source-linked AI summary

Introduction to Haar Measure Tools in Quantum Information: A Beginner's Tutorial

Antonio Anna Mele

arXiv:2307.08956v4quant-ph

TL;DR

The tutorial addresses the difficulty of learning Haar-measure tools caused by their reliance on representation theory. It develops linear-algebraic methods centered on moment operators, symmetric subspaces, tensor-network notation, and unitary designs, then applies them to quantum-information calculations. The tutorial presents accessible tools for Haar integration and illustrates their use in quantum machine learning and classical shadow tomography.

  • Problem

    Haar-measure tools are fundamental in quantum information but can be challenging for beginners because they rely on advanced representation theory.

  • Method

    The tutorial introduces Haar-measure calculations using linear algebra, including moment operators, symmetric subspaces, tensor-network notation, and unitary-design concepts.

  • Results

    The tutorial derives tools and formulas for Haar-measure calculations and connects them to applications including unitary designs and classical shadow tomography.

  • Takeaways & Limitations

    Clifford unitaries can reproduce Haar moments through the third moment while allowing efficient circuit sampling and implementation.

  • Takeaways & Limitations

    Classical-shadow sampling efficiency does not by itself guarantee computational efficiency, which also requires efficiently computing the unbiased estimator.

Abstract

from arXiv · show

The Haar measure plays a vital role in quantum information, but its study often requires a deep understanding of representation theory, posing a challenge for beginners. This tutorial aims to provide a basic introduction to Haar measure tools in quantum information, utilizing only basic knowledge of linear algebra and thus aiming to make this topic more accessible. The tutorial begins by introducing the Haar measure with a specific emphasis on characterizing the moment operator, an essential element for computing integrals over the Haar measure. It also covers properties of the symmetric subspace and introduces helpful tools like tensor network diagrammatic notation, which aid in visualizing and simplifying calculations. Next, the tutorial explores the concept of unitary designs, providing equivalent definitions, and subsequently explores approximate notions of unitary designs, shedding light on the relationships between these different notions. Practical examples of Haar measure calculations are illustrated, including the derivation of well-known formulas such as the twirling of a quantum channel. Lastly, the tutorial showcases the applications of Haar measure calculations in quantum machine learning and classical shadow tomography.

1 Introduction

The tutorial introduces Haar measure as uniform sampling over unitary matrices and makes its quantum-information tools accessible using only linear algebra. It develops moment-operator methods and previews applications including tomography, random-circuit sampling, benchmarking, unitary designs, and shadow tomography.

  • Haar measure formalizes drawing unitary matrices uniformly at random, with unitaries representing quantum evolutions commonly described by circuits.
  • The tutorial targets beginners by avoiding representation theory and using only linear-algebra concepts.
  • The tutorial characterizes the moment operator because it is crucial for computing Haar integrals.
  • The tutorial also introduces symmetric and antisymmetric subspaces, calculation tools, unitary designs, approximate designs, and Haar-measure applications.

2 Notation and Preliminaries

This section establishes notation for operators, states, norms, quantum channels, and tensor-product identities used throughout the tutorial.

  • L(Cd) denotes linear operators, Herm(Cd) Hermitian operators, and U(d) the unitary operators satisfying U†U = I.
  • Schatten p-norms include the trace and Hilbert-Schmidt norms, obey monotonicity and rank bounds, and are invariant under unitary multiplication.
  • Quantum states are density matrices, while pure states are unit vectors represented with bra-ket notation.
  • A quantum channel is a completely positive, trace-preserving linear map, representable using d^2 Kraus operators.

3 Haar measure and moment operator

The tutorial develops Haar integration through the moment operator, showing that it is an orthogonal projector onto the unitary commutant and can be evaluated using permutation operators and Weingarten coefficients.

  • Haar measure is the unique probability measure on U(d) invariant under both left and right multiplication.
  • The k-th moment operator converts Haar averages of degree-k unitary expressions into an operator whose matrix elements can explicitly evaluate integrals.
  • The moment operator is linear, trace-preserving, self-adjoint, and invariant under conjugation by tensor powers of unitaries.
  • The moment operator is the Hilbert-Schmidt orthogonal projector onto Comm(U(d), k).
  • Schur-Weyl duality identifies the commutant as the span of permutation operators associated with the symmetric group.
  • The moment operator can be expressed in the permutation basis by solving a k! equation system, with coefficients represented through the Gram-matrix pseudoinverse and Weingarten coefficients.
  • Permutation matrices are linearly independent when k ≤ d and linearly dependent when k > d.

4 Symmetric subspace

The symmetric subspace consists of tensor states invariant under subsystem permutations, and its projector and dimension support Haar-random-state calculations. The section also develops antisymmetric projectors and their relation to second moments.

  • Symmetric subspace: The symmetric subspace contains tensor states invariant under permutations of their constituent subsystems.
  • Symmetric subspace: The symmetric projector is the orthogonal projector onto Sym_k(Cd).
  • Symmetric subspace: The dimension of the symmetric subspace equals the number of ways to distribute k indices among d labels.
  • Antisymmetric subspace: The antisymmetric subspace is projected by an orthogonal projector, and for d ≥ k its dimension counts unordered k-element subsets of d labels.
  • Symmetric and antisymmetric subspaces: Symmetric and antisymmetric projectors are orthogonal, and for k = 2 they form an orthogonal basis for the unitary commutant.
  • Haar-random states: For any state |ϕ⟩, the Haar moment of |ϕ⟩⟨ϕ|⊗k is a uniform linear combination of permutation operators.

5 Vectorization formalism

Vectorization converts operators and superoperators into vectors and matrices, making Haar moment operators and commutant projectors easier to express and manipulate. The formalism also connects vectorized inner products with the Hilbert-Schmidt product.

  • Vectorization basics: The bijection vec(·) maps a d × d matrix to a d^2-dimensional vector, with |A⟩⟩ denoting vec(A).Every vector in (C^d)⊗2 corresponds uniquely to a matrix.
  • Vectorization basics: The vectorized identity |Ω⟩ is a non-normalized maximally entangled state, and the transpose-trick gives A ⊗ I|Ω⟩ = I ⊗ A^T|Ω⟩.This identity underlies the ABC-rule used in tensor manipulations.
  • Superoperators: Vectorization turns a linear superoperator Φ into a matrix vec(Φ) acting linearly on |X⟩⟩.The construction applies to all linear superoperators Φ: L(C^d) → L(C^d).
  • Moment operators: For d ≥ k, the Haar expectation E[|Tr(U)|^(2k)] equals k!, because permutation operators span the k-th commutant and are linearly independent.The permutation span characterization follows from Schur-Weyl duality.
  • Applications of the formalism: The vectorized moment-operator treatment provides an alternative proof of the commutant-projector theorem and supports explicit formulas for k = 1 and k = 2.Tensor network diagrams can visualize and manipulate these vectorized equations.

6 Tensor network diagrams

Tensor-network diagrams provide a graphical language for states, matrices, tensor products, contractions, traces, and permutation operators. They make swap and cyclic-permutation identities visually transparent.

  • Diagrammatic notation: Tensor networks represent tensors as boxes and contractions as connections, with ket states, bra states, and matrices assigned distinct graphical forms.This notation is intended to simplify the understanding and analysis of tensor operations.
  • Diagrammatic notation: The non-normalized maximally entangled state |Ω⟩ = |I⟩⟩ represents the vectorized identity and enables the transpose-trick diagrammatically.The relation |A⟩⟩ = A ⊗ I|Ω⟩ = I ⊗ A^T|Ω⟩ connects operators to diagrams.
  • Permutation diagrams: Permutation matrices rearrange tensor-product factors, while the identity and Flip operators have corresponding line-based diagrams.Permutation composition is visualized by placing the associated diagrams next to one another.
  • Haar-moment diagrams: Tensor diagrams can express Haar second-moment matrix elements and clarify calculations involving U ⊗ U* and Kronecker deltas.The notation links diagrammatic manipulations with swap-trick evaluations.
  • Swap tricks: The swap-trick gives Tr(A ⊗ B F) = Tr(AB), with partial traces yielding Tr_2(A ⊗ B F) = AB and Tr_1(A ⊗ B F) = BA.These identities provide graphical shortcuts for reducing tensor expressions.
  • Cyclic permutations: The cyclic-permutation-trick generalizes the swap trick: Tr(A_1 ⊗ ··· ⊗ A_k V_d(π_cyc)) = Tr(A_1 ··· A_k).The corresponding partial trace simplifies to the ordered product A_1 ··· A_k.

7 Unitary designs

Unitary designs replace Haar-random unitaries when only low-order moments are needed, offering equivalent characterizations through moments, frame potentials, and polynomial averages. The section also gives cardinality bounds, group-based criteria, and examples such as Pauli and Clifford ensembles.

  • Definition and motivation: Unitary k-designs are distributions whose k-th moments match those of the Haar measure, motivated by the cost of implementing generic Haar-random unitaries.Many quantum-information applications require only low-order Haar moments.
  • Frame potential: A distribution is a k-design if and only if its frame potential equals the Haar frame potential.The frame-potential difference equals the squared 2-norm of the difference between vectorized moment operators.
  • Frame potential: If k ≤ d, the Haar frame potential is k!, because the relevant commutant dimension equals the number of linearly independent permutation operators.Permutation operators are linearly independent in this regime.
  • Cardinality lower bound: A discrete k-design must have cardinality at least d^(2k)/k!, which grows at least exponentially with qubit number when d = 2^n.This lower bound applies to a uniform distribution over a discrete set of unitaries.
  • Equivalent definitions: A unitary k-design can equivalently be characterized by matching Haar averages for homogeneous degree-k polynomials in U and U* or by equality of moment operators.These equivalences are collected in the proposition on equivalent definitions.
  • Group criteria: For a uniform distribution over a unitary group, equality of commutant dimensions with U(d) suffices for the distribution to be a k-design.This gives a group-based route to establishing the design property.
  • State designs and examples: A unitary k-design induces a k-state design from the orbit U|ψ_0⟩, while a (k + 1)-design is automatically a k-design.The Pauli basis forms a 1-design, and the Clifford group forms a 3-design but not a 4-design.
  • State designs and examples: Clifford circuits can be implemented with O(n^2/log(n)) gates and sampled efficiently, making them practical for reproducing Haar moments through order three.The relevant gate set is {H, CNOT, S}.

8 Approximate unitary designs

Approximate unitary designs offer several ways to approximate Haar moments, with TPE, diamond-norm, frame-potential, and relative-error notions related by explicit bounds. TPE designs can also be amplified through products of independently sampled unitaries.

  • Approximate design notions: Approximate unitary designs are useful when exact designs are unnecessary, and multiple approximation notions have been proposed.The tutorial examines their relationships, including TPE, diamond, frame-potential, and relative-error definitions.
  • TPE amplification: A TPE λ-approximate k-design remains a TPE λ^P-approximate design after composing P independently sampled unitaries.The product distribution νP has a vectorized moment operator determined by the independent factors.
  • TPE amplification: Inverse-exponential precision ε = ˜εd^-c can be achieved with P repetitions satisfying the amplification bound, and P is linear in qubit number n when λ = O(1).Here d = 2^n and ˜ε, c > 0.
  • Diamond and TPE notions: TPE ε-approximation implies diamond εd^k-approximation, while diamond ε-approximation implies TPE εd^k/2-approximation.The two notions are therefore quantitatively related, but their precision parameters differ with dimension and moment order.
  • Frame-potential approximation: The frame-potential notion can be easier to estimate numerically because it uses traces of matrices in L(C^d) rather than tensor-product-space matrices.It is related to TPE approximation through εd^k in one direction and ε in the converse direction.
  • Frame-potential approximation: Frame-potential ε-approximation implies diamond εd^k-approximation, while diamond ε-approximation implies frame-potential εd^3k/2-approximation.These bounds connect the frame-potential and diamond formulations through dimension- and order-dependent precision factors.

9 Examples and applications

The tutorial applies Haar-moment tools to channel twirling, entanglement, concentration, barren plateaus, and classical shadows. These examples connect unitary designs and Haar calculations to randomized benchmarking, variational optimization, and quantum-state estimation.

  • Twirling of quantum channels: Twirling a quantum channel with a unitary 2-design produces an output channel resembling a depolarizing channel.The tutorial derives this formula as an application relevant to Randomized Benchmarking.
  • Moment calculations: The expected entanglement entropy of a pure state drawn from a 2-design is close to the maximum, although finite systems retain a Page-entropy gap.The finite-size gap separates the exact average from the maximal entropy value.
  • Concentration inequalities: Haar random states make Pauli expectation values concentrate, with Levy’s lemma improving the qubit-number dependence from exponential to double exponential.The stronger double-exponential bound requires Haar sampling rather than merely a state 2-design.
  • Barren Plateaus: For variational quantum algorithms, cost-function values and partial-derivative variances decay exponentially with the number of qubits, a phenomenon called Barren Plateaus.The stated bounds use concentration and moment-based arguments for 2-design distributions.
  • Classical shadow tomography: Classical shadows use moment calculations to construct unbiased estimators and bound the samples needed to estimate multiple observables.For Clifford measurements, the tutorial gives M(ρ) = (Tr(ρ) I + ρ)/(d + 1) and M^-1(ρ) = (d + 1)ρ − Tr(ρ) I.
  • Classical shadow tomography: Sampling efficiency for classical shadows does not by itself guarantee computational efficiency; efficient estimator computation is known for structured observables such as stabilizer states.For random Pauli bases, the protocol is both sample- and time-efficient for local observables like Pauli operators.
Loading 2307.08956v4…