Fourier Analytic Security Bounds for Sums of Independent Random Permutations

black and white manga panel, dramatic speed lines, Akira aesthetic, bold ink work, A colossal diffraction grating of smoked glass etched with dense, hair-thin parallel lines, cracked along a violent diagonal and leaning alone in a pitch-black void. A single harsh white beam strikes the fracture at the center, scattering into only three razor-sharp color bands—red, green, blue—that pierce the darkness like searchlights. Radial speed lines streak outward from the fracture, warping the shards and the light trails. The grating's broken edges glint with metallic highlights against an empty, stark background. [Z-Image Turbo]
Every age devises its own test of whether chance is honest. The present one asks how many queries before a sum of random shuffles betrays itself—and finds the threshold holds against even the quantum questioner.
A paper entitled “Indistinguishability of Sum of Permutations: A Fourier Analytic Route to Classical and Quantum Security” circulated this week on the arXiv preprint server, offering a unified method for a question that will occupy the cryptographic record for some time: how well a construction formed by summing several independent random permutations can be distinguished from a genuinely random function, whether the interrogator works by classical or by quantum means. At the centre of the inquiry is a finite abelian group of order N, and the sum π^k_+(x)=π_1(x)+⋯+π_k(x), taken over k≥2 independent uniform random permutations. The authors represent the construction by its probability density and the distinguisher by its acceptance function, and observe that the classical and quantum query models impose different restrictions upon the Fourier support of the latter. Classically, the paper obtains a bound of order q/N^{k-1/2} for every q below N, with a sharper bound below the birthday threshold. In the quantum model, a simulation argument gives a bound of order N^{-(k-3/2)} for q at most (N-1)/2, while Fourier interpolation supplies concrete finite bounds up to q≤4N/15. For a single query, the bound sharpens to order N^{-2}. Over binary vector spaces, a one-query Fourier attack matches the order of the one-query bound, while a parity attack at N/2 queries shows that the bounds reach the constant-advantage query threshold. The same treatment extends to two further constructions, one permitting arbitrary surjective linear postprocessing, which includes truncation, and the other being Dinur’s variable-output single-permutation construction, LXoP, for which concrete quantum security bounds are given at one and two block outputs. Several terms in the announcement reward a plain definition before the reader goes further. Indistinguishability, the title's first word, is the entire question under examination: two functions are indistinguishable when no observer permitted a bounded number of questions can tell which of them is being offered. A random permutation is a rearrangement of the N elements of the group, chosen from the N! possible rearrangements with equal likelihood. To sum k such permutations is to apply each of the k permutations separately to a single input x and add the k results in the group's arithmetic. The result is no longer a permutation but a function from the group to itself, and the question is whether it can be told apart from a function chosen at random from all N^N such functions. The observer doing the telling is the distinguisher; each question it asks is a query; the rule by which it turns its answers into a decision is its acceptance function; its advantage is the amount by which its chance of deciding correctly exceeds the even chance of a coin. Because the paper's method is Fourier analysis, that term also wants its gloss. A finite abelian group of order N is a set of N elements with a commutative addition; the example best known to practitioners is the binary vector space F_2^n, whose elements are strings of n bits, added without carrying. Any function on such a group can be written as a weighted sum of the group's characters, the periodic building blocks of the group; the characters that appear with nonzero weight form the function's Fourier support. The authors' central observation is that classical and quantum distinguishers differ in which acceptance functions they may implement, and that this difference appears as a restriction on Fourier support. Two further phrases will recur. The birthday threshold is the number of queries, on the order of the square root of N, at which two outputs of a random function become likely to collide; below it the classical bounds sharpen, and above it a general bound applies. A parity attack, in the binary setting, collects the parity of selected outputs and compares it with the value a random function would give; at N/2 queries such an attack attains advantage 1/2, which the authors read as the point where a constant advantage becomes reachable. Two constructions in the later sections also ask for a word. Surjective linear postprocessing means a fixed linear map applied after the sum, with the property that every element of the smaller output space remains attainable; truncation is the plain form of it, the discarding of some output bits. The construction called LXoP draws a variable number of output blocks from a single permutation, and the paper gives concrete quantum security bounds for its one- and two-block outputs. These are the instruments with which the paper's bounds are tested, and the reader may keep them in view without needing to carry their algebra. The announcement itself supplies no calendar beyond its week of circulation. Its milestones are thresholds of query count and group size, and they may be set down as a ledger. Classically, the bound is of order q divided by N to the power k minus one half for every q below N, and is refined to order q squared divided by N to the power k below the birthday threshold. In the quantum model, the simulation argument holds for q no greater than (N-1)/2, and Fourier interpolation yields concrete finite bounds up to q no greater than 4N/15. For the range one to (N-1)/2 queries, the query-dependent bounds are of order the smaller of N to the minus one half and q cubed divided by N squared plus 1/N for k equal to two, and of order the smaller of q cubed divided by N to the power k and N to the power minus (k minus three halves) for k at least three. At the single query, the first of these sharpens to order N to the minus two. Set down as a census, the announcement is a spare but exact ledger. It names one central construction, the sum of k independent uniform random permutations, and two variants examined beside it: the sum with arbitrary surjective linear postprocessing, and Dinur's LXoP construction. It names two query models, classical and quantum, and two attacks at named query counts, a one-query Fourier attack and an N/2-query parity attack. Its thresholds are likewise given as fractions of N: the simulation argument holds to (N−1)/2 queries, Fourier interpolation to 4N/15, and the single-query bound stands alone. What the announcement does not supply is a calendar; its landmarks are counts, and the archivist records them as counts, without converting queries into weeks or advantages into outcomes. —Dr. Octavia Blythe Dispatch from The Prepared E0

This piece was written by AI.

Published October 1, 2026
ai@theqi.news