The Clifford group fails gracefully to be a unitary 4-design

1.1 Motivation: Designs and derandomizations

• Explicit t-design constructions beyond t=3 remain limited, even though 4-designs provide optimal behavior in several applications.Random-circuit constructions exist for arbitrary degree, but are described as less explicit than desired.Source: §1.2 · pp. 2–3

1.2 Outline of result: Overcoming the “t = 3-barrier”

• The Clifford group is an explicit unitary 3-design but not a 4-design, and its orbits are generally not projective 4-designs.This creates a barrier for derandomizing applications that require fourth-order behavior.Source: §1.2 · pp. 2–3
• The paper explicitly analyzes the Clifford group’s fourth tensor power, finding one additional invariant subspace beyond the unitary-group case: a stabilizer code.The resulting framework supports explicit fourth-moment calculations and constructions of exact or arbitrarily precise approximate 4-designs from Clifford orbits.Source: §1.2 · p. 3
• The paper provides the representation-theoretic analysis, while companion papers apply it to signal analysis and quantum information.The applications are only briefly sketched here because the work targets communities using different technical languages.Source: §1.2 · p. 3

1.3.1 Application: Phase Retrieval

• In phase retrieval, the task is recovering an unknown vector from phase-insensitive measurements, motivating structured or deterministic alternatives to Gaussian measurement vectors.The paper’s companion work establishes near-optimal guarantees for measurements drawn randomly from Clifford orbits, including stabilizer measurements, and extends the results to low-rank matrix recovery.Source: §1.3.1 · pp. 3–4

1.3.2 Application: Quantum state distinguishability

• For quantum state distinguishability, the companion analysis establishes near-optimal POVM-norm results for high-purity states using all Clifford orbits, including stabilizer measurements.It also derives tighter entropic uncertainty relations for stabilizer measurements.Source: §1.3.2 · p. 4

1.4 Note added

• A contemporaneous study analyzes a closely related Clifford representation for improved randomized-benchmarking bounds, with largely equivalent main results but apparently different proof methods.The related representation is identified as τ(2,2).Source: §1.4 · p. 4

2.1 Projective t-designs

• Complex projective t-designs are evenly distributed unit vectors, equivalently characterized by polynomial averages, symmetric-subspace moments, or minimum frame potential.The minimum frame-potential bound is saturated exactly by t-designs.Source: §2.1 · pp. 5–6
• Projective t-designs require at least d^⌈t/2⌉⌊t/2⌋ vectors; multiqubit Clifford orbits provide the only known explicit infinite family of 3-designs.The lower bounds for t = 1, 2, 3, 4 are d, d^2, d^2(d + 1)/2, and d^2(d + 1)^2/4, respectively.Source: §2.1 · p. 6

2.2 Unitary t-designs

• Unitary t-designs reproduce Haar averages of degree-(t,t) polynomials, and for groups are equivalently characterized by matching unitary-group irreducible decompositions or permutation-generated commutants.Their frame-potential lower bound is attained exactly by unitary t-designs.Source: §2.2 · pp. 7, 9
• Any orbit of a unitary group t-design is a complex projective t-design, linking unitary-design representation theory to state-orbit constructions.This follows because the group acts irreducibly on the symmetric tensor-power subspace.Source: §2.2 · p. 9

2.3 Pauli group, Clifford group, and stabilizer codes

• The n-qubit Clifford group normalizes the Pauli group and maps projectively onto Sp(2n,F2), while serving major roles in quantum computation, error correction, benchmarking, and tomography.The Clifford group is generated by Hadamard, phase, and CNOT gates, and its projective quotient is isomorphic to the symplectic group.Source: §2.3 · pp. 9–10
• The multiqubit Clifford group is a unitary 3-design but not a 4-design, although its fourth frame potential remains close to the 4-design value.This establishes the background for analyzing how the Clifford group fails at fourth order.Source: §2.3 · pp. 9–10
• The fourth tensor power of the Clifford group has only a few more irreducible components than the fourth tensor power of the full unitary group, with details deferred to the next section.Source: §2.3 · p. 11
• A stabilizer group is an abelian Pauli subgroup excluding −1, and its stabilizer code is the common +1-eigenspace of that group.Source: §2.3 · p. 11
• An n-qubit stabilizer group of order 2^m defines a code of dimension 2^(n−m); maximal groups have one-dimensional codes called stabilizer states.Source: §2.3 · p. 11
• Stabilizer groups correspond closely to isotropic subspaces of the finite symplectic vector space F_2^(2n), because commutation makes the symplectic inner product vanish on the associated subspace.Source: §2.3 · p. 11

3.1 A special stabilizer code

• The stabilizer code V_n,4 is Clifford- and S_4-invariant, and is the only subspace of (C^d)^⊗4 stabilized by the Clifford group but not U(d).For k divisible by 4, V_n,k is the joint +1 eigenspace of a Clifford-invariant stabilizer group, with dimension d^(k−2).Source: §3.1 · p. 11

3.2 Main results

• The commutant of τ^4(C_n) is generated by S_4 and the stabilizer projector P_n,4, so the Clifford fourth tensor power has exactly one additional generator beyond U(d).This is the main theorem for d=2^n.Source: §3.2 · p. 12

3.3 Proof of Main Theorem

• The spaces W_λ^±, whenever nontrivial, carry irreducible and inequivalent representations of C_n, yielding the explicit C_n × S_4 decomposition used to establish the commutant theorem.The proof uses Schur–Weyl decomposition, dimension formulas, and the fourth frame potential, which equals 30 for n≥3.Source: §3.2 · p. 13, §3.3 · p. 15

3.4 Representations of the discrete symplectic group

• V_n,4 carries the complex Weil representation of Sp(2n,F_2), while W_[1,1,1,1], W_[4], and W_[2,2] carry the corresponding α_n, β_n, and ζ_1 representations.The dimensions of these spaces coincide with the degrees of the associated Weil characters.Source: §3.4 · pp. 15, 17
• For subgroups R≤Sp(2n,F_2), the stabilizer-code contribution M_k(R) equals the (k−1)th frame potential and the number of R-orbits on (F_2^n)^2.A subgroup has the same V_n,4 decomposition as Sp(2n,F_2) exactly when its Clifford preimage forms a unitary 3-design.Source: §3.4 · p. 17
• For n ≥ 2, Sp(2n, F2) is a rank-3 permutation group on the nonzero vectors of F2^2n [88].For n = 1, it is instead described as a rank-2 permutation group.Source: §3.4 · p. 17
• Except when n = 2, no proper subgroup of Sp(2n, F2) has the same property; n = 2 has a unique counterexample [88].For n ≠ 2, every proper subgroup has more irreducible components in V_n,4 and consequently in V_n,k.Source: §3.4 · p. 17
• When restricted to a proper subgroup of C_n, at least one of the representations W[1,1,1,1], W[4], and W[2,2] becomes reducible.Source: §3.4 · p. 17

4.1 Clifford covariant t-designs for one qubit

• For one qubit, a Clifford orbit is a 4-design exactly when x4 + y4 + z4 = 3/5, and under the same condition it is also a 5-design.The minimum fourth frame potential is 1/5; stabilizer states instead maximize it.Source: §4.1 · p. 18
• The one-qubit condition for a 6-design requires x2, y2, z2 to be distinct roots of the stated cubic, equivalently x6 + y6 + z6 = 3/7 or x2y2z2 = 1/105.The same condition gives a 7-design; 48 solutions form two Clifford orbits, while no Clifford orbit can be an 8-design.Source: §4.1 · p. 18

4.2 Random Clifford orbits are good approximations to 4-designs

• Exact 4-design fiducials always exist because the continuous deviation function ǫ has zero mean and therefore has roots.Equivalent fiducial criteria are expressed through the characteristic function and its ℓ4-norm, while ǫ also determines operator-, trace-, and frame-potential deviations.Source: §4.2 · pp. 19–21
• For ψ with α+(ψ) ≥6/d(d + 3), Clifford orbits can achieve almost the same POVM norm constants as 4-designs and are therefore very good approximations in this concrete setting.Separately, Eq. (83) gives a typical value of |ǫ(ψ)| around 6/d when d is large.Source: §4.2 · p. 21

4.3 Fiducial vectors of exact 4-designs up to five qubits

• Exact 4-design fiducials are constructed through five qubits using tensor products of one-qubit, magic, and Hoggar fiducials with specified Bloch-vector conditions.The constructions include n = 2, 3, 4 conditions x4 + y4 + z4 = 5/7, 7/11, 8/19, and n = 4, 5 constructions using the Hoggar fiducial with conditions 17/19 and 8/19.Source: §4.3 · p. 22
• Tensor-product exact 4-design fiducials are structurally limited: apart from the four-single-qubit case, at most three tensor factors are allowed, with only specific factor-size patterns.The permitted patterns are (1, 1, 1, 1), (3, 2, 1), (2, 2, 1), (n1, 1, 1), and (n1, n2).Source: §4.3 · p. 23

4.4 Algorithms for constructing projective 4-designs

• Two construction algorithms are provided: a tensor-product procedure that appends a qubit satisfying a Bloch-vector condition, and a bisection method that finds a zero of ǫ between states with opposite signs.The section also constructs exact weighted 4-designs by combining two or more Clifford orbits with appropriate weights.Source: §4.4 · p. 23

4.5 Approximate fiducial vectors of 4-designs from MUB cycler

• The section begins connecting approximate 4-design fiducials to eigenstates of special Clifford unitaries, including MUB cyclers and balanced states defined relative to mutually unbiased bases.This connection is presented as a preliminary observation intended to motivate further work.Source: §4.5 · p. 24
• The orbit of ψ_n ⊗ ψ_T is a very good approximate 4-design, with deviation magnitude around 1/2^(n+1), comparable to a random (n+1)-qubit state.Exact 4-design fiducials can also be constructed with Algorithm 1, while ψ_n and ψ_n ⊗ ψ_T can serve as inputs to Algorithm 2.Source: §4.5 · p. 24
• The conjectured deviation relation for any Singer-unitary eigenvector implies that Clifford orbits of ψ_n ⊗ ψ_T converge to 4-designs in operator norm as n grows.Singer unitaries are Clifford-group transformations cycling through complete mutually unbiased bases, and their eigenbases are well-defined because their spectra are nondegenerate.Source: §4.5 · p. 24

4.6 Harmonic invariants, connections to the real-valued theory, and 5-designs

• The Clifford group has no nontrivial harmonic invariants in bi-degrees (1,1), (2,2), or (3,3), while all degree-(4,4) harmonic invariants are multiples of ε; a normalized orbit is a 4-design exactly at a root of ε.This reformulates the degree-four invariant structure established through the Clifford commutant and the polynomial associated with its additional invariant operator.Source: §4.6 · p. 25
• The authors conjecture that a normalized Clifford orbit is a complex projective 5-design if and only if its vector is a root of ε, supported by n = 1, numerical n = 2, and related lower-degree arguments.The conjecture remains unverified in general because converting complex Clifford invariants to real ones can produce zero polynomials, although the problematic matrices form a measure-zero orthogonal set.Source: §4.6 · pp. 25–26

• The multi-qubit Clifford group is a unitary 3-design, but not a 4-design; accordingly, Clifford orbits are 3-designs, but generally not 4-designs.The lack of an explicit family of well-structured 4-designs has limited applications of t-designs for derandomizing constructions relying on random vectors.Source: §5 · p. 26
• The authors determine all irreducible components in the fourth tensor power of the Clifford group using Schur-Weyl duality and a special stabilizer code.They state that this representation structure is completely captured by those two ingredients.Source: §5 · p. 26
• The work provides exact 4-designs and arbitrarily high-precision approximate 4-designs constructed from Clifford orbits.Source: §5 · p. 26
• Open problems include whether Clifford orbits form t-designs for t > 4, the maximal attainable t, proving Conjecture 1, and constructing unitary 4-designs from the Clifford group.Source: §5 · p. 26
• A broader unresolved direction is an explicit description of commutants for higher tensor powers of the Clifford group.The paper suggests classifying commuting elements expressible through Clifford theory or related constructions if a complete description is impractical.Source: §5 · p. 26

A Alternative proof of Lemma 1

• The basis of V_n,4 consists of 4^n tensor-product states labeled by length-n strings, on which S4 acts by simultaneously permuting letters.Source: A Alternative proof of Lemma 1 · p. 27
• The S4 action partitions strings into type I, type II, and type III orbits, with type I containing 0^n, type II having length 3, and type III containing the remaining relevant strings.Source: A Alternative proof of Lemma 1 · p. 27
• For each irreducible representation λ, the construction uses orbit-dependent character sums over stabilizer cosets to form basis vectors and compute dimensions.Source: A Alternative proof of Lemma 1 · p. 28
• The dimension of W^[4] equals the total number of string orbits, while the dimension of W^[1,4] equals the number of type III orbits.Source: A Alternative proof of Lemma 1 · p. 28
• Character sums vanish for λ = [2,1,1] and [3] across the relevant subgroup cosets, implying that both W^[2,1,1] and W^[3,1] have dimension zero.Source: A Alternative proof of Lemma 1 · pp. 28–29

B.1 An interesting basis for the stabilizer code

• The vectorization construction produces d^2 orthogonal vectors contained in the stabilizer code, so they form an orthogonal basis of that code.Source: §B.1 · p. 30
• Real Clifford elements permute the basis, while general Clifford elements act by signed permutations; consequently, all Clifford unitaries act monomially.Source: §B.1 · p. 30
• In this basis, the stabilizer code affords a signed permutation representation of Sp(2n, F2).Source: §B.1 · p. 30
• The paper does not use this monomial basis in its present analysis, but suggests it could make the character derivation more explicit.Source: §B.1 · p. 30

B.2 An interesting orbit

• The paper presents the configuration as canonical but notes that no application is currently known for it, while distinguishing its monomial representations from earlier incomparable constructions.Source: §B.2 · p. 30
• A maximal isotropic subspace determines a stabilizer state in V_n,4, and Clifford transformations act monomially on the resulting collection.Source: §B.2 · p. 31
• The constructed Clifford orbit lies in W^[4], and Schur's lemma implies that its rank-one projectors form a tight frame on that space.Source: §B.2 · p. 31
• The orbit is overcomplete because its cardinality exceeds the dimension D+ of the frame space.Source: §B.2 · p. 31
• Squared inner products between orbit elements can be computed from intersections of their stabilizer groups and relate to squared overlaps of stabilizer states.Source: §B.2 · p. 31

C Proof of a generalization of Eq. (76)

• The generalized bound applies to four normalized state vectors in dimension d = 2^n and constrains the stabilizer code V_n,4.The result also provides insight into the entanglement properties of V_n,4.Source: C Proof of a generalization of Eq. (76) · pp. 31–32
• The upper bound implies that V_n,4 contains no product state and limits the fidelity between any pure code state and any product state to 1/d.The bound can be attained by identical stabilizer states or by suitable orthogonal stabilizer states from one basis.Source: C Proof of a generalization of Eq. (76) · p. 32
• The upper-bound proof uses positivity and repeated Cauchy or Hölder inequalities, with equality requiring each characteristic vector to have exactly d entries equal to ±1.The relevant Pauli expectations take only the values 0, ±1 at saturation.Source: C Proof of a generalization of Eq. (76) · p. 32
• Upper-bound saturation implies that all four vectors are stabilizer states belonging to a common stabilizer basis.This follows because the relevant Pauli eigenvalue sets form a maximal isotropic subspace.Source: C Proof of a generalization of Eq. (76) · p. 32

D Derivation of Eq. (81)

• The derivation evaluates the second moment of α_+(ψ) using the stabilizer-code projector, symmetric-subspace projectors, and Pauli operators.The relevant projector P_n,4 is onto the stabilizer code V_n,4.Source: D Derivation of Eq. (81) · pp. 32–33
• Table 3 records the total, balanced, and signed balanced permutation counts for each S_8 cycle type without odd cycles.Its signed count is N3 = N3+ − N3−, while N3+ + N3− = N2.Source: Table 3 · p. 33
• The Pauli cases split according to whether operators are identity, distinct commuting nonidentity operators, or anticommuting operators.Each case receives a separate polynomial contribution in d.Source: D Derivation of Eq. (81) · p. 33
• Permutation contributions are organized by cycle type, with S_4 terms involving three permutations of type (2^2) and six of type (4).All other S_4 permutations contain an odd-length cycle.Source: D Derivation of Eq. (81) · p. 33
• For the final cases, balanced permutations are partitioned into A+ and A− according to whether the induced number of transpositions between blocks is even or odd.The resulting signed counts determine the remaining terms in Eq. (126).Source: D Derivation of Eq. (81) · pp. 33–34

E Proof of Lemma 6

• The proof represents the difference of two pure states through its spectral decomposition and reduces the problem to a positive operator Q on their two-dimensional span.The trace- and Hilbert-Schmidt-norm relations provide the starting reduction.Source: E Proof of Lemma 6 · pp. 34–35
• Under the parameterization b = s sin φ, c = s cos φ, and a = 1 − s, f(a,b,c) = 2 sin φ{−s²[1−cos(φ−θ)] + s} ≤ 1 + sin θ.Source: E Proof of Lemma 6 · p. 36
• After expanding Q in the Pauli basis, the optimization becomes maximizing or minimizing f(a,b,c) over a convex feasible region defined by 0 ≤ Q ≤ I.Symmetry permits restricting the analysis to selected sign regions and boundary points.Source: E Proof of Lemma 6 · p. 36
• The resulting piecewise bounds are converted to Eq. (151) using sin θ = λ and the preceding norm inequality.The upper bound is saturated at an explicit boundary point, and the lower bound is saturated in both θ regimes.Source: E Proof of Lemma 6 · pp. 34, 36, 37
• Both the lower and upper bounds in Eq. (151) can be saturated.Explicit parameter choices are given for each saturation regime.Source: E Proof of Lemma 6 · pp. 34, 37

F Proof of Proposition 8

• For n ≥ 5, the characteristic-function norm exceeds the 4-design threshold, so the vector cannot generate a 4-design.The criterion used is that a 4-design fiducial must satisfy the corresponding ℓ4-norm condition.Source: F Proof of Proposition 8 · pp. 37–38
• When n3 = 1 with n1,n2 ≥ 3, the dimension condition n ≥ 7 again rules out a 4-design fiducial.Source: F Proof of Proposition 8 · p. 37
• When n3 = 1 with n1,n2 ≥ 2, the condition n ≥ 5 suffices to rule out a 4-design fiducial.Source: F Proof of Proposition 8 · p. 38

Polynomial–tensor correspondence

• Homogeneous degree-t polynomials correspond one-to-one with symmetric degree-t tensors, whose tensor powers span the totally symmetric space.The construction uses monomials, polarization, and the identification of symmetric multilinear forms with homogeneous polynomials.Source: §G.1 · pp. 38–39

Orthogonal-group symmetry

• The orthogonal group acts diagonally on tensor powers, and its action preserves the totally symmetric subspace.This representation agrees with the natural action on homogeneous polynomials.Source: §G.1 · p. 39

Harmonic decomposition

• Contraction of tensor indices maps symmetric degree-t tensors onto degree-(t−2) tensors, with its kernel providing an invariant harmonic subspace.Iterating contractions decomposes the symmetric space into irreducible representations, including lower-degree harmonic components.Source: §G.1 · pp. 39–40

Polynomial formulation

• In polynomial language, tensor contraction corresponds up to normalization to the Laplacian, so harmonic polynomials lie in its kernel.The resulting decomposition expresses homogeneous degree-t polynomials through harmonic polynomials of lower degrees.Source: §G.1 · pp. 40–41

Spherical-design criterion

• For a finite orthogonal-group orbit, being a spherical t-design is equivalent to the orbit representative vanishing on all invariant harmonic polynomials through degree t.If the group has no harmonic invariants in degrees 1 through t, every orbit is a spherical t-design.Source: §G.1 · p. 41

Complex polynomial framework

• Complex homogeneous polynomials and symmetric tensors extend the real correspondence, with Wirtinger derivatives and the complex Laplacian handling holomorphic and conjugate variables.The complex formalism treats coordinates and their conjugates as independent for differentiation.Source: §G.2 · pp. 41–42

Structural limitation

• The order of taking linear maps and symmetric powers cannot generally be interchanged, because the relevant map need not be onto.The left-hand side spans L(Sym^t(C^d)), whereas the corresponding symmetric operator space is generally smaller.Source: §G.2 · p. 43

Partial trace and harmonics

• The complex analogue of tensor contraction is the partial trace, which acts on tensor products of operators by tracing one factor.Its kernel is invariant under unitary conjugation and leads to complex harmonic tensors of bi-degree (t,t).Source: §G.2 · p. 43

Unitary representation theory

• Complex harmonic tensors decompose into irreducible representations of U(d), while the partial trace corresponds up to normalization to the complex Laplacian.The associated harmonic polynomials have bi-degree (t,t).Source: §G.2 · pp. 43–44

Complex projective-design criterion

• A finite orbit is a complex projective t-design when its representative is a common root of all invariant harmonic polynomials of bi-degrees (s,s) through t.If the group has no such harmonic invariants, every orbit is a complex projective t-design; equivalently, the group acts irreducibly on Sym^t(C^d).Source: §G.2 · p. 44

Relation to real designs

• Complex projective designs depend on projectors (|ψ⟩⟨ψ|)^⊗t and are therefore conceptually closer to real spherical designs of degree 2t or 2t+1.This reflects their bi-degree-(t,t) polynomial structure rather than dependence only on ψ^⊗t.Source: §G.2 · p. 44