Source-linked AI summary
Indistinguishability of Sum of Permutations: A Fourier Analytic Route to Classical and Quantum Security
Ritam Bhaumik, Chun Guo, Xiaoning Guo, Ashwin Jha
TL;DR
Permutation-to-function transformations address the need for pseudorandom functions when few dedicated candidates have been proposed or analyzed. This paper uses a unified Fourier framework to analyze classical and quantum indistinguishability, with bounds for sums of permutations and related constructions. In the binary setting, its bounds identify the query threshold for constant quantum advantage.
Problem
The paper asks how sums of random permutations can remain indistinguishable from random functions beyond the birthday bound, including against quantum superposition queries.
Source: §1 · pp. 2–3Method
A unified Fourier framework represents constructions by probability densities and distinguishers by acceptance functions, with query models restricting the supports that contribute.
Source: §1.1 · p. 3, §2.2 · p. 7Results
In the binary setting, advantage tends to zero through N/2−1 quantum queries, while an N/2-query parity attack has advantage 1/2.
Source: §1 · p. 3Takeaways & Limitations
The paper establishes the constant-advantage quantum query threshold in the binary setting and extends its analysis to linear-output variants and LXoP.
Source: §1 · p. 3, §1.1 · p. 3The results are information-theoretic and study quantum adversaries in the Q2 model, counting construction queries rather than computation time.
Source: §1 · p. 3
Abstract
from arXiv · showhide
We study classical and quantum indistinguishability of sums of independent random permutations and related transformations from permutations to functions. Let $G$ be a finite abelian group of order $N$, and let $\pi^k_+(x)=\pi_1(x)+\cdots+\pi_k(x)$ for $k\geq2$ independent uniform random permutations of $G$. We give a unified Fourier analytic treatment in which the construction is represented by its probability density and a distinguisher by its acceptance function, with the classical and quantum query models imposing different restrictions on the Fourier support of the latter. Classically, we obtain the bound $O_k(q/N^{k-1/2})$ for every $q<N$, and refine it below the birthday threshold to $O_k(q^2/N^k)$. In the quantum model, a simulation argument gives $O_k(N^{-(k-3/2)})$ for $q\leq(N-1)/2$, while Fourier interpolation gives concrete finite bounds up to $q\leq4N/15$ and the query-dependent bounds $O\left(\min\left\{N^{-1/2},q^3/N^2 + 1/N\right\}\right)$ and $O_k\left(\min\left\{q^3/N^k,N^{-(k-3/2)}\right\}\right)$, for $k=2$ and $k \geq 3$, respectively, throughout $1\leq q\leq(N-1)/2$. For $q = 1$, the first bound sharpens to $O(N^{-2})$. Over $G=\mathbb F_2^n$, a one-query Fourier attack matches the order of our one-query bound, while an $N/2$-query parity attack with advantage $1/2$ shows that our bounds reach the constant-advantage query threshold. We further study two variants of sum of permutations over binary vector spaces. First, we allow arbitrary surjective linear postprocessing, which includes truncation, and obtain classical and quantum bounds that retain the output-size dependence. Second, we analyse Dinur's variable-output single-permutation construction, $\mathsf{LXoP}$, for every fixed output width, and derive its classical and quantum security bounds; for one- and two-block outputs, we give concrete quantum security bounds.
1 Introduction
The paper asks how sums and related transformations of random permutations can remain indistinguishable from random functions, including under quantum superposition queries. It develops a shared Fourier framework and applies it to sum-of-permutations bounds, linear postprocessing, and LXoP.
1.1 Our Results
- In the binary setting, advantage tends to zero through N/2−1 quantum queries, while an N/2-query parity attack has advantage 1/2.This identifies the constant-advantage query threshold for every fixed k ≥ 2.Source: §1 · p. 3, §1.1 · p. 3
- For surjective linear postprocessing over binary vector spaces, the bounds retain output-size dependence and improve the prior truncated-construction single-user bound by a factor √q.The quantum bounds also retain the dependence on output size M.Source: §1.1 · p. 3
- The paper analyzes Dinur’s LXoP family at every fixed width, obtaining classical O_w(q/N^3/2) and quantum O_w(N^-1/2) bounds in the stated query range.The construction produces w output blocks using w+1 permutation calls; concrete quantum bounds are also given for widths one and two.Source: §1.1 · p. 4
1.2 Technical Overview and Organisation
- A unified analytic framework represents the construction by a density and the distinguisher by an acceptance function, then compares their matching support components.Classical distinguishers contribute only supports within a fixed queried set, whereas quantum distinguishers contribute supports of at most 2q inputs.Source: §1.2 · p. 4
- For sums of permutations, injection Fourier coefficients and their moments bound the classical and quantum contributions, with quantum analysis retaining only components on at most 2q inputs.The level-two collision component is treated separately; higher levels are controlled by the common moment estimate.Source: §1.2 · p. 4
- A checksum-based completion simulates the sum of permutations through N/2−1 quantum queries using a reference function that is (N−1)-wise independent.The completed reference function is indistinguishable from a uniform random function when 2q ≤ N−1.Source: §1.2 · p. 4
- For projected outputs, the analysis restricts Fourier indices to Im(τ⊤) and uses subspace averaging to retain dependence on output size M.The refined quantum estimate also tracks the dimension spanned by the character indices.Source: §1.2 · p. 5
- For LXoP, shared permutation calls impose linear relations among character indices, which the recursive coefficient reduction uses to bound second moments.An admissible reduction order lets the analysis recover removed indices and control repeated counting.Source: §1.2 · p. 5
2 Fourier Analytic Framework
This section builds a support-based framework for comparing oracle distributions, then specializes it to Fourier analysis on finite abelian groups and uniform injections.
2 Fourier Analytic Framework
- The framework compares oracle densities through their support components, with the distinguishing advantage expressed as an inner product against the density difference.The analysis uses the product reference measure and bounds the remaining inner product after applying query-support cutoffs.Source: §2 · p. 5, §2.2 · pp. 6–7
2.1 Product Spaces and Support Decomposition
- The framework decomposes functions on a product space into orthogonal components indexed by the coordinates on which they have nonconstant factors.Each component depends only on its support, and averaging over a supported coordinate with the others fixed gives zero.Source: §2.1 · pp. 5–6
2.2 Distinguishing Advantage and Query Support Cutoffs
- A classical q-query distinguisher can be restricted to a fixed q-input set, while a quantum q-query acceptance function has support only on sets of at most 2q inputs.The quantum cutoff is established for the stated function-oracle model, including mixed states, randomness, and intermediate measurements.Source: §2.2 · pp. 7–8
2.3 Character Bases over Finite Abelian Groups
- For finite abelian groups, characters form the orthonormal basis, so support components correspond to grouped Fourier terms.Fourier coefficients can be combined by convolution, and the coefficients of a k-fold convolution are powers of those of the original function.Source: §2.3 · pp. 8–9
2.4 Injection Densities and Fourier Coefficients
- The uniform injection density describes sampling distinct outputs, and its fixed-support Fourier moments depend only on support size.The paper also tracks the corresponding maximum coefficient and higher Fourier moments.Source: Uniform Density on Injections · pp. 9–10
3 Sum of Independent Permutations
The section develops Fourier bounds for sums of independent permutations, refining classical security at small query counts and treating quantum distinguishers through their restricted Fourier support. It also gives binary-group quantum attacks that match the one-query bound’s order and attain constant advantage at half the domain size.
3.1 Refined Classical Security
- O_k(q/N^(k-1/2)) bounds classical advantage for every q<N, via a smoothing argument.The section also improves the bound below the birthday threshold by computing the two-point distribution and applying a moment bound to higher levels.Source: §3.1 · p. 11, §3 · p. 11
- The below-birthday refinement computes the two-point distribution, then applies a moment bound to Fourier levels above two.The two-point distribution is treated separately from the higher levels.Source: §3.1 · p. 11
3.2 Quantum Security
- The direct Fourier calculation gives explicit bounds through q≤4N/15, extending the moment-estimate range q≤N/4.The section combines direct analysis of low-support Fourier components with a simulation argument for the full range 2q<N.Source: §3.2 · p. 13
3.3 Quantum Attacks over F𝒏
- For every k≥2, a one-query quantum attack has advantage 1/(2(N−1)^k), while an N/2-query attack has advantage 1/2.For n≥2, the one-query attack matches the asymptotic order of the upper bound at q=1.Source: §3.3 · pp. 13–14
4 Quantum Security by Completion
The section extends a quantum indistinguishability argument to function classes in which any prescribed d values uniquely determine a function. For sums of permutations, the checksum reconstructs the final output, yielding a bound below the half-domain query threshold.
4 Quantum Security by Completion
- If every restriction to d inputs uniquely determines a function in a class, its random member is indistinguishable from a uniform random function when 2q≤d, under the stated theorem.The proof completes independent uniform values on a fixed d-point set to obtain a d-wise independent reference function.Source: §4 · p. 14
- The sum-of-permutations checksum reconstructs the final output from the other N−1 outputs, placing the construction in the unique-completion case.The argument uses the checksum to complete a reference function from independent uniform values on N−1 inputs.Source: §4 · pp. 14–15
- For every fixed k≥2, advantage is O_k(N^−(k−3/2)) for q≤N/2−1, while an N/2-query algorithm has advantage at least 1/2.The two bounds meet at consecutive query counts and establish the asymptotic constant-advantage threshold for the binary construction.Source: §4 · p. 15
5 Linear-Output-Postprocessing Variants
The section studies sums of permutations after surjective linear postprocessing, including coordinate truncation, and derives classical and quantum bounds that depend on output size. Its quantum analysis combines completion, level-two treatment, and rank information about pulled-back masks.
5.1 Generalised and Truncated Sums of Permutations
- Surjective linear postprocessing includes projection onto fixed coordinates, recovering the truncated sum construction studied in [CKLL22].The resulting bounds retain dependence on the output size M.Source: §5.1 · p. 16, §5.2 · p. 16
- The classical analysis pulls output characters back through the surjective map and uses subspace averaging to control their Fourier contribution.The pulled-back characters lie in a subspace of size M.Source: §5 · p. 15, §5.1 · p. 16
5.2 Quantum Security
- The quantum bounds retain output-size dependence by combining completion with a sharper level-two analysis and rank-sensitive bounds for remaining levels.The projected checksum determines the final output from the other N−1 outputs.Source: §5.2 · p. 16
- For q=1, the k=2 quantum bound sharpens to O(N^−2), and for M=1 the advantage is zero.Concrete quantum bounds are also given for q≤4N/15 when N≥1024.Source: §5.2 · pp. 16–17
6 The Variable-Output LXoP Family
The section analyzes Dinur’s LXoP family for each fixed output width, using Fourier moments of the underlying permutation outputs. It establishes moment bounds under an admissibility condition and develops quantum bounds, including concrete estimates for one- and two-block outputs.
6 The Variable-Output LXoP Family
- LXoP produces w output blocks from w+1 domain-separated permutation calls, and the section analyzes every fixed width w under an admissibility condition.Domain separation makes all permutation inputs used by distinct construction inputs distinct, so their outputs form a uniform injection.Source: §6 · p. 17
- A nonzero construction block pulls back to between 2 and w+1 nonzero entries, so construction level t maps to injection levels from 2t through (w+1)t.The pullback map is injective; the fixed-support moment bound applies when (w+1)t≤N/8.Source: §6 · p. 18
- For levels t≥2, the classical argument bounds their total contribution by O_w(q^2/N^3) under its stated small-query condition.The estimate uses the fixed-support moments and assumes (w+1)q≤N/8.Source: §6 · pp. 18–19
6.1 Quantum Security
- The quantum analysis sums the same fixed-support moments only through level 2q, as required by the quantum support cutoff.The estimates include L_1=O_w(N^−3) and apply when (w+1)t≤N/8.Source: §6.1 · p. 19
- For one- and two-block outputs, the section derives concrete quantum bounds from sharper fixed-support moments.The two constants are not comparable because they arise from different estimates.Source: Concrete bounds for one- and two-block… · p. 20
A.1 Small Supports
- For support size two, Λ₂ and Θ₂ each equal 1/(N−1); at size three, the passage reports Λ₃ = 2(N−1)(N−2) and Θ₃ = 4(N−1)(N−2).Source: §A.1 · p. 23
B Proof of Lemma 3
The proof bounds Fourier moments across support levels, including beyond N/2, to establish the stated maximum bound. The query-dependent quantum argument combines level-two estimates with a common injection estimate for higher levels.
B Proof of Lemma 3
- An auxiliary random variable represents the numerical moment sequence, whose even moments permit interpolation between support sizes using Hölder’s inequality.It is not part of the oracle distribution.Source: B Proof of Lemma 3 · p. 24
- For 1 ≤ t < N, the proof states Θₜ₊₁ = t N−t(2Θₜ + Θₜ₋₁), with Θ₀ = 1 and Θ₁ = 0, and gives Θ₆ = 5(3N² + 86N + 120)(N−1)(N−2)(N−3)(N−4)(N−5).Source: B Proof of Lemma 3 · p. 24
- Coefficient merging controls the remaining support levels and extends the maximum bound beyond N/2.The merged character falls at level t−1 or t−2, enabling a recurrence from the sparse bound.Source: B Proof of Lemma 3 · p. 25
- At N = 1024, the normalized right-hand side of (40) is less than 3.245641 < 13/4, proving (12).The preceding bound applies to the remaining levels and combines the support-three contribution with bounds for other levels.Source: B Proof of Lemma 3 · p. 26
C Query-dependent Quantum Bounds for Sums of Permutations
- The query-dependent quantum bounds use the level-two components from Lemma 4 and the common injection estimate for higher levels.The framework applies to both unprojected and projected sums.Source: C Query-dependent Quantum Bounds for… · p. 27
C.1 A Bound for the Centred Collision Count
This section bounds a quantum algorithm’s correlation with the centred collision count. Its proof uses the support cutoff for quantum acceptance functions and polynomial interpolation over a family of random-function distributions.
C.1 A Bound for the Centred Collision Count
- The section develops a bound on the correlation between a quantum acceptance probability and the centred collision count.The count sums over unordered pairs of distinct inputs, with expectation taken under the uniform random-function distribution.Source: §C.1 · p. 27
- A q-query acceptance function has Fourier supports of size at most 2q, yielding a polynomial of degree at most 2q−1 in the random-function construction parameter.The polynomial is bounded between 0 and 1 at positive reciprocal integers, allowing application of Zhandry’s polynomial bound [Zha12a, Theorem B.1].Source: §C.1 · p. 28
- As the range size grows, the random-function construction is governed by the no-collision and single-collision cases, with other patterns having probability O_D(r−2).The surviving constant in the limiting argument does not depend on D or M.Source: §C.1 · p. 28
C.2 Proof of Theorem 3
The proof applies the centred-collision bound to the level-two component and controls higher levels with the common injection estimate. It obtains query-dependent bounds and a concrete finite-range bound.
C.2 Proof of Theorem 3
- For k = 2, the query-dependent bound is O(q^3/N^2 + 1/N); for k ≥ 3, it is O_k(q^3/N^k).The remainder in (41) is O_k(N^(−2k+3)).Source: §C.2 · p. 29
- For N ≥ 1024, the proof retains constants to give a concrete finite bound for 2 ≤ q ≤ 4N/15.The supplied corollary states the query range but does not include the bound’s expression here.Source: §C.2 · p. 29
- For q = 1, the advantage is at most π^2/(6(N−1)k); for q = 0, it is zero.Source: §C.2 · p. 29
- The proof combines the level-two estimate from the centred-collision lemma with the common injection estimate for remaining levels.For q = 1, the remainder is empty.Source: §C.2 · pp. 28–29
C.3 Proof of Theorem 6
For surjective linear postprocessing, the analysis retains dependence on output size by separating rank-one characters from higher-rank characters. For sums of three or more permutations, it establishes query-dependent and uniform bounds; the same refinement is not established for LXoP.
C.3 Proof of Theorem 6
- For the projected two-permutation sum, the local bound is O(q^3/N^2).The bound is stated for fixed-size output groups with an implicit constant allowed to depend on that size.Source: §C.3 · p. 30
- Separating rank-one characters from characters of rank at least two improves the remainder from 1/N to M/N^2.The proof retains the rank of the span of character coordinates to obtain this output-size dependence.Source: §C.3 · pp. 29–30
- For k ≥ 3, the local bound is O_k(q^3/N^k), while the uniform bound is N^(−(k−3/2)).The proof combines the local and uniform bounds over the query range under consideration.Source: §C.3 · p. 31
- The query-dependent refinement does not carry over to LXoP because its support-two component has a different linear structure.Source: §C.3 · p. 31
D A Deutsch-Type One-Query Algorithm
A one-query algorithm recovers a masked XOR of two function outputs with certainty. The section also develops recursive second-moment estimates for LXoP, including a fixed-width one-input bound.
D A Deutsch-Type One-Query Algorithm
- One query recovers ⟨λ, f(x) ⊕ f(x′)⟩ with certainty for distinct inputs and nonzero λ.Only one step queries the oracle; a final Hadamard transform and measurement return the masked XOR.Source: D A Deutsch-Type One-Query Algorithm · p. 31
E The Recursive Moment Argument for LXoP
- Proposition 7 bounds the recursive moment for w-admissible σ when (w + 1)t ≤ N/8, using d_t = ⌈t/2⌉.The recursion applies a fixed-support estimate at its leaves and requires recovery to preserve injectivity after merges.Source: E The Recursive Moment Argument for LXoP · pp. 31–32, §E.1 · pp. 32–33
E.1 From Coefficient Merging to the Moment Bound
- The structured recursion improves the direct fixed-support bound by a factor (2k0/(N − k0))^d.Its applicability requires recovery of removed entries so child maps remain injective; initial injectivity alone is insufficient.Source: §E.1 · pp. 32–33
E.2 Recovering Labels from an Untouched Block
- Untouched blocks let the LXoP recursion continue to depth d_t while keeping each child map injective.At each depth before d_t, at least one block remains untouched; the recovery formulas apply for every secondary choice.Source: §E.2 · pp. 34–35
E.3 The One-Input Moment
- The one-input second moment is O_w(N^-3), uniformly over the w-admissible maps considered.For support size three, at most N − 1 masks contribute, each with coefficient magnitude 2/((N − 1)(N − 2)).Source: §E.3 · p. 35
F Proofs of Concrete Bounds for Dinur’s LXoP
The one-block proof analyzes the complete construction function on a domain of size N/2 and bounds its Fourier levels using a zero first level and higher-level estimates.
F.1 One-Block Output
- The complete one-block construction function has an actual domain of size D = N/2.The proof denotes its density by ξ and uses input symmetry.Source: §F.1 · p. 35
- The t = 1 term vanishes by the support-two calculation.For t ≥ 2, the cited estimate applies through t ≤ N/16, covering the levels used below.Source: §F.1 · p. 36
- Proposition 16 applies for 2 ≤ t ≤ N/16, containing every level used below; for 2 ≤ t ≤ N/32, the proof uses D = N/2.Source: §F.1 · p. 36
F.2 Two-Block Output
The two-block proof analyzes a construction function on a domain of size N/4 and combines a one-input calculation with bounds on higher Fourier levels.
F.2 Two-Block Output
- The complete two-block construction function has an actual domain of size D = N/4.Exactly N − 1 construction masks (γ, γ), with γ ≠ 0, contribute to the one-input calculation; each coefficient is 2/((N − 1)(N − 2)).Source: §F.2 · p. 36
- The higher-level tail estimate uses a ratio of 0.855 separately for the even and odd tails.The supplied calculation bounds this ratio uniformly in N and t for t ≥ 4.Source: §F.2 · pp. 36–37
- For N = 1024 and q ≤ 2, direct evaluation through level t ≤ 4 gives total second Fourier moment below 46/N, hence below 74/N.The proof adds missing low levels as positive upper bounds.Source: §F.2 · p. 37