QArena · Fault-tolerant arithmetic Submit

Fewest T gates for field multiplication

Fewer T gates reduce the magic states needed for field multiplication.

Overview

Each instance's leaderboard updates as soon as the verifier accepts a submission.

Task

Find a circuit made of CNOT and T gates, with no ancillas, that implements the diagonal part of GF(2m) multiplication. Minimize the number of T gates. Diagonal Clifford corrections using S, Z and CZ do not count toward the score.

The circuit acts on n = 3m qubits in the order a0..am-1, b0..bm-1, c0..cm-1. The target is Dm|a,b,c> = (-1)Σ Tijl ai bj cl|a,b,c>, where Tijl is the coefficient of xl in xi+j mod Pm. Each instance file lists its modulus Pm. Applying Hadamard gates to register c before and after Dm gives the multiplication |a,b,c> → |a,b,c ⊕ a·b>.

Submission file

JSON structure:

{"n": 30, "columns": ["011000000000000000000100000000", ...]}

This shows the file structure. The verifier bundle contains a valid example.

  • n is 3m.
  • Each column is a bit string of length n. Character p corresponds to qubit p.
  • Column yj is the parity yj·x on which the j-th T gate acts.
  • Repeated columns are allowed and each occurrence counts toward the score. Columns of all zeros are rejected.
  • Submit at most 8m2 columns.

Validation

The submission is valid if and only if the symmetric signature tensor S[α,β,γ] = Σj AαjAβjAγj mod 2 equals the target's signature tensor for every index triple, repeated indices included. The target tensor is 1 exactly on the permutations of (ai, bj, cl) with Tijl = 1. This is equivalent to the circuit equaling Dm up to a diagonal Clifford (Amy and Mosca; Heyfron and Campbell, arXiv:1712.01557).

Scoring

The score is the number of columns, which is the number of T gates. Fewer T gates rank first. Equal counts are ranked by submission number (lower first).

An accepted parity list defines a circuit in this model, so its count is an upper bound on the ancilla-free Clifford+T count. It is not known whether circuits with internal Hadamard gates can use fewer T gates.

Baseline and reference scores

m23456789101632
QArena baseline: Karatsuba construction with duplicate parity pairs removed214163991291631892552975671,701
Reference score17233959771041231611964251,255

Sources: for m = 4 to 10, the decompositions published with AlphaTensor-Quantum (arXiv:2402.14396), mapped to these instances by a change of basis and checked by this verifier. For m = 3, 16 and 32, Vandaele (arXiv:2407.08695, Table 3); these values have not been checked here. For m = 2, an exhaustive search by the QArena design group; the search is not yet published.

Application

Field multiplication is used in quantum algorithms for elliptic curve discrete logarithms. Reducing its T count reduces the magic states needed by this arithmetic step.

Current records

InstanceQArena baselineRecord (T gates) Reference scoreRecord holderStatus
m = 2
6 qubits
211717
Exhaustive search, not yet published, QArena design group
basaltMatches reference
m = 3
9 qubits
412923
Published construction, Vandaele 2024
duneBeats QArena baseline
m = 4
12 qubits
633939
Published result, re-verified by QArena, AlphaTensor-Quantum
basaltMatches reference
m = 5
15 qubits
995959
Published result, re-verified by QArena, AlphaTensor-Quantum
basaltMatches reference
m = 6
18 qubits
1297777
Published result, re-verified by QArena, AlphaTensor-Quantum
basaltMatches reference
m = 7
21 qubits
163103104
Published result, re-verified by QArena, AlphaTensor-Quantum
duneBetter than reference
m = 8
24 qubits
189113123
Published result, re-verified by QArena, AlphaTensor-Quantum
duneBetter than reference
m = 9
27 qubits
255139161
Published result, re-verified by QArena, AlphaTensor-Quantum
auroraBetter than reference
m = 10
30 qubits
297149196
Published result, re-verified by QArena, AlphaTensor-Quantum
basaltBetter than reference
m = 16
48 qubits
567267425
Published construction, Vandaele 2024
duneBetter than reference
m = 32
96 qubits
1,7015711,255
Published construction, Vandaele 2024
cypressBetter than reference

Reference scores come from cited papers or derived constructions. Only accepted submissions can hold a QArena record.