wai-quantum 0.6.6: Clifford circuits measured in a product basis, exactly, by cut-rank #340

Merged

dcharlot merged 5 commits from wai/quantum-cutrank-0.6.6 into main 2026-10-19 20:26:27 -04:00

Conversation 0 Commits 5 Files changed 18 +1662 -11

dcharlot commented 2026-10-19 11:30:49 -04:00

Owner

Copy link

wai-quantum 0.6.6. Exact amplitudes, probabilities and samples of a Clifford circuit followed by one layer of single-qubit gates at any angles. The cost is set by the state's cut-rank along an order of its qubits, not by the number of qubits.

What it does

• Tableau::graph_form writes a stabilizer state as local Cliffords (H^h S^s Z^z per qubit) on a graph state.
• quantum_cutrank (a new feature, in full) sums ⟨x| ⊗ V_q |ψ⟩ = 2^{-n/2} Σ_z ∏_q W_q[x_q][z_q] (−1)^{Σ_{a~b} z_a z_b} one qubit at a time along an order, in a table of 2^{cut-rank} entries.
  • Each qubit is an in-place pass: a merge where the boundary space loses a dimension, the qubit's signed terms, and a doubling or an XOR-shifted sum for its edges ahead.
  • The passes depend only on the graph and the order, so they are planned once per state.
  • By Parseval, the table's norm is a prefix's probability, so the same passes sample exactly.
  • Precision::Single stores the table in f32 with f64 arithmetic.
  • anneal_order searches orders for less work (Σ 2^cut) and never returns a worse one.
  • Every reduction is summed in fixed chunks, so a result is the same bits at any thread count.
• Examples:
  • cutrank_qasm: cut-ranks, annealed orders, amplitudes and exact samples from an OpenQASM 2 Clifford bulk with a closing single-qubit layer.
  • patch_xeb: patched cross-entropy from measured shots, each patch on the dense state vector.
  • tn_cost: one amplitude's contraction cost as find_path finds it.
• The README gets a feature row and a section, the changelog an entry, and every lockfile that pins wai-quantum moves to 0.6.6.

Evidence

• Tests:
  • the graph form rebuilds the state (40 random circuits);
  • amplitudes equal the state vector's in probability and relative phase (30 circuits, up to 12 qubits);
  • forward, reverse and shuffled orders agree (24 to 54 qubits);
  • single-precision tables agree with double;
  • samples follow the exact law and are seeded;
  • annealing lowers the work and keeps the amplitudes;
  • cut-ranks match the entanglement of GHZ and product states.
• Mutants caught: the merge's signs, the sampling choice, the doubling's sign, the fold's coefficient, the graph form without its S.
• A 70-qubit graph-state sampling circuit (nearest-neighbour Clifford bulk, then a non-Clifford basis on every qubit), on the build host:
  • an annealed order keeps every cut at 29 bits or fewer (32 in the natural order);
  • one amplitude in 0.53 s, one exact sample in 1.1 to 1.5 s, largest table 4 GiB in f32;
  • amplitudes agree between orders and precisions to about 1e-7 relative.
• Checked on the build host:
  • scripts/check-lockfiles.sh;
  • cargo check --no-default-features --features quantum_cutrank --examples;
  • cargo test --release --features full --lib quantum_cutrank (7 passed) and -- quantum_tableau (10 passed);
  • clippy with -D warnings reports nothing in the files this PR touches. The rest of the library's test code has lints of its own, which CI does not check.
• Public-text checks clean: names, vanity metrics, comparative prose, tests in CI, orphan sources.

🤖 Generated with Claude Code

**wai-quantum 0.6.6.** Exact amplitudes, probabilities and samples of a Clifford circuit followed by one layer of single-qubit gates at any angles. The cost is set by the state's cut-rank along an order of its qubits, not by the number of qubits. ### What it does - **`Tableau::graph_form`** writes a stabilizer state as local Cliffords (`H^h S^s Z^z` per qubit) on a graph state. - **`quantum_cutrank`** (a new feature, in `full`) sums `⟨x| ⊗ V_q |ψ⟩ = 2^{-n/2} Σ_z ∏_q W_q[x_q][z_q] (−1)^{Σ_{a~b} z_a z_b}` one qubit at a time along an order, in a table of `2^{cut-rank}` entries. - Each qubit is an in-place pass: a merge where the boundary space loses a dimension, the qubit's signed terms, and a doubling or an XOR-shifted sum for its edges ahead. - The passes depend only on the graph and the order, so they are planned once per state. - By Parseval, the table's norm is a prefix's probability, so the same passes sample exactly. - `Precision::Single` stores the table in `f32` with `f64` arithmetic. - `anneal_order` searches orders for less work (Σ 2^cut) and never returns a worse one. - Every reduction is summed in fixed chunks, so a result is the same bits at any thread count. - **Examples:** - `cutrank_qasm`: cut-ranks, annealed orders, amplitudes and exact samples from an OpenQASM 2 Clifford bulk with a closing single-qubit layer. - `patch_xeb`: patched cross-entropy from measured shots, each patch on the dense state vector. - `tn_cost`: one amplitude's contraction cost as `find_path` finds it. - The README gets a feature row and a section, the changelog an entry, and every lockfile that pins wai-quantum moves to 0.6.6. ### Evidence - **Tests:** - the graph form rebuilds the state (40 random circuits); - amplitudes equal the state vector's in probability and relative phase (30 circuits, up to 12 qubits); - forward, reverse and shuffled orders agree (24 to 54 qubits); - single-precision tables agree with double; - samples follow the exact law and are seeded; - annealing lowers the work and keeps the amplitudes; - cut-ranks match the entanglement of GHZ and product states. - **Mutants caught:** the merge's signs, the sampling choice, the doubling's sign, the fold's coefficient, the graph form without its S. - **A 70-qubit graph-state sampling circuit** (nearest-neighbour Clifford bulk, then a non-Clifford basis on every qubit), on the build host: - an annealed order keeps every cut at 29 bits or fewer (32 in the natural order); - one amplitude in 0.53 s, one exact sample in 1.1 to 1.5 s, largest table 4 GiB in `f32`; - amplitudes agree between orders and precisions to about 1e-7 relative. - **Checked on the build host:** - `scripts/check-lockfiles.sh`; - `cargo check --no-default-features --features quantum_cutrank --examples`; - `cargo test --release --features full --lib quantum_cutrank` (7 passed) and `-- quantum_tableau` (10 passed); - clippy with `-D warnings` reports nothing in the files this PR touches. The rest of the library's test code has lints of its own, which CI does not check. - **Public-text checks clean:** names, vanity metrics, comparative prose, tests in CI, orphan sources. 🤖 Generated with [Claude Code](https://claude.com/claude-code)

dcharlot added 5 commits 2026-10-19 11:30:49 -04:00

wai-quantum: exact amplitudes and samples of a Clifford circuit measured in a product basis, by cut-rank elimination ... bf22681fae

Tableau::graph_form writes the state as local Cliffords on a graph state.
quantum_cutrank (feature quantum_cutrank, in full) then sums
⟨x|⊗V_q|ψ⟩ = 2^{-n/2} Σ_z ∏ W_q[x_q][z_q] (−1)^{Σ_{a~b} z_a z_b} one qubit at a
time along an order, in a table of 2^{cut-rank} entries: each qubit is an
in-place pass (a merge where the boundary space loses a dimension, the
qubit's signed terms, a doubling or an XOR-shifted sum for its edges
ahead). By Parseval the table's norm is the prefix probability, so the same
passes sample exactly. Reductions are summed in fixed chunks.

Tests: the graph form rebuilds the state (40 random circuits); amplitudes
equal the state vector's in probability and relative phase (30 circuits up
to 12 qubits); forward, reverse and shuffled orders agree (24–54 qubits, at
least 30 comparisons); samples follow the exact law (60k shots) and are
seeded; cut-ranks of GHZ and product states. Mutants caught: merge signs,
the sampling choice, the graph form without its S, the doubling's sign.

examples/cutrank_qasm reads an OpenQASM 2 Clifford bulk with a closing
single-qubit layer. On the tracker's 70-qubit random graph-state instance
(max cut 32 bits, matching an independent tableau): an exact amplitude in
about 7.6 s and an exact sample in about 14 s on the build host, 69 GB peak
in f64.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

wai-quantum cutrank: the steps planned once, tables stored at a chosen precision, and orders annealed for less work ... 70af6b4dcd

- The table operations (coordinate swaps, kernel merges, doublings, XOR
  sums, the sign masks) depend on the graph and the order alone, so they
  are planned once per state; an amplitude or a sample only runs passes.
- Precision::Single stores the table in f32 (arithmetic stays f64): half
  the memory and traffic, amplitudes within 1e-5 of the double ones (test).
- A merge followed by a doubling is one pass; a doubling no longer
  zero-fills. Every pass is cut into fixed chunks, so reductions do not
  depend on the threads.
- work() = Σ 2^cut, passes() tallies the passes by kind, and
  anneal_order(moves, seed) searches orders by simulated annealing, never
  returning a worse one (test: lower work, same amplitudes).

On the tracker's 70-qubit random graph-state instance (build host):
natural order, max cut 32: amplitude 6.2 s (f64) / 4.7 s (f32), sample
7.5 s (f32). Annealed order (seed 3, 20 000 moves): max cut 29, work 6.5×
less, largest table 4 GiB in f32: amplitude 0.53 s, exact sample 1.49 s,
amplitudes agreeing with the natural order's double ones to 1e-7.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

wai-quantum examples: patched cross-entropy benchmarking from shots (patch_xeb) and one amplitude's contraction cost (tn_cost) ... bdd46363b5

patch_xeb simulates each patch of a patched circuit exactly on the dense
state vector and scores the measured shots (normalised linear XEB per
patch, the product, its error). On a published 61-qubit experiment's
release it reproduces the released per-circuit fidelities to 5e-7 (their
JSON's rounding). tn_cost prints log10 of one amplitude's contraction cost
as find_path sees it, with the largest intermediate and the slices.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

wai-quantum 0.6.6: the cut-rank engine in the README and changelog ... f599f0a6fc

A feature row and a section for quantum_cutrank; the changelog entry; the
version in every lockfile that pins it.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

wai-quantum cutrank: clippy's is_multiple_of and an enumerated loop ... c1c325002f

Co-Authored-By: Claude Opus 5.5 <[email protected]>

dcharlot merged commit 9cfe619ef7 into main 2026-10-19 20:26:27 -04:00

dcharlot referenced this pull request from a commit 2026-10-19 20:26:29 -04:00

Merge pull request 'wai-quantum 0.6.6: Clifford circuits measured in a product basis, exactly, by cut-rank' (#340) from wai/quantum-cutrank-0.6.6 into main

dcharlot referenced this pull request 2026-10-19 20:29:08 -04:00

wai-quantum-service: the cutrank method, exact samples of a Clifford circuit measured in a product basis #342

Sign in to join this conversation.

Reviewers

No reviewers

Labels

Clear labels

No items

No labels

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

1 participant

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

Transaction-Science/open-standards!340

No description provided.