Transaction-Science/open-standards
Watch
1
0
Fork
You've already forked open-standards
 0
Code Issues Pull requests Projects Releases 15 Packages Wiki Activity Actions

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

Merged
dcharlot merged 1 commit from wai/quantum-service-cutrank-pr into main 2026-10-19 21:23:11 -04:00
Conversation 0 Commits 1 Files changed 8 +278 -9
dcharlot commented 2026-10-19 20:29:08 -04:00
Owner
Copy link

The service gains the cutrank method, built on wai-quantum 0.6.6's quantum_cutrank (#340). It gives exact samples, and amplitudes, of a Clifford circuit followed on each qubit by one-qubit gates at any angles. The cost is set by the state's cut-rank along an order of the qubits, not by how many qubits there are.

What it takes and answers

  • qasm::Program::product_measured splits the program in two. The Clifford part is lowered gate by gate. Each qubit's gates, from its first non-Clifford one on, are multiplied into one matrix. A gate on several qubits that comes after such a gate is refused, and its refusal names the gate and the qubit.
  • The planner (catalog::product_measured):
    • takes the graph form and searches orders for narrower cuts, with a move budget of about a second;
    • admits the job by its largest table, in double precision when that fits the job's memory and single otherwise;
    • caps the table entries its passes touch at 4e12, about ten minutes.
    • The run rebuilds the same order.
  • Counts are exact samples, seeded.
  • Amplitudes are exact up to one phase that all of the job's amplitudes share, since a stabilizer tableau does not fix a state's global phase. The result's global_phase member says so, and accuracy.max_cut gives the widest cut.
  • Expectation values are refused.
  • auto tries cutrank for counts after the state vectors and before mps. For amplitudes it stays explicit, so the methods that fix the phase keep their place.

Evidence

  • Tests:
    • amplitudes equal the state vector's times one phase (6, 9 and 11 qubits, 24 amplitudes each);
    • counts follow the exact distribution over 40,000 shots and repeat with their seed;
    • auto samples a 40-qubit ring with chords by cutrank;
    • the refusals name the gate.
  • Mutants caught: the tail multiplied in the wrong order, and samples mapped to the wrong qubits.
  • smoke.sh samples a 60-qubit instance (max cut 10 bits).
  • The README gets a table row and a bullet on how it runs, and the method enums in discovery and MCP list it.
  • Checked on the build host: cargo test --locked --release (64 passed), clippy with -D warnings, and the full smoke.sh against a local server.

🤖 Generated with Claude Code

The service gains the **`cutrank` method**, built on wai-quantum 0.6.6's `quantum_cutrank` (#340). It gives exact samples, and amplitudes, of a Clifford circuit followed on each qubit by one-qubit gates at any angles. The cost is set by the state's cut-rank along an order of the qubits, not by how many qubits there are. ### What it takes and answers - **`qasm::Program::product_measured`** splits the program in two. The Clifford part is lowered gate by gate. Each qubit's gates, from its first non-Clifford one on, are multiplied into one matrix. A gate on several qubits that comes after such a gate is refused, and its refusal names the gate and the qubit. - **The planner** (`catalog::product_measured`): - takes the graph form and searches orders for narrower cuts, with a move budget of about a second; - admits the job by its largest table, in double precision when that fits the job's memory and single otherwise; - caps the table entries its passes touch at 4e12, about ten minutes. - The run rebuilds the same order. - **Counts** are exact samples, seeded. - **Amplitudes** are exact up to one phase that all of the job's amplitudes share, since a stabilizer tableau does not fix a state's global phase. The result's `global_phase` member says so, and `accuracy.max_cut` gives the widest cut. - **Expectation values** are refused. - **`auto`** tries `cutrank` for counts after the state vectors and before `mps`. For amplitudes it stays explicit, so the methods that fix the phase keep their place. ### Evidence - **Tests:** - amplitudes equal the state vector's times one phase (6, 9 and 11 qubits, 24 amplitudes each); - counts follow the exact distribution over 40,000 shots and repeat with their seed; - `auto` samples a 40-qubit ring with chords by `cutrank`; - the refusals name the gate. - **Mutants caught:** the tail multiplied in the wrong order, and samples mapped to the wrong qubits. - **`smoke.sh`** samples a 60-qubit instance (max cut 10 bits). - **The README** gets a table row and a bullet on how it runs, and the method enums in discovery and MCP list it. - **Checked on the build host:** `cargo test --locked --release` (64 passed), clippy with `-D warnings`, and the full `smoke.sh` against a local server. 🤖 Generated with [Claude Code](https://claude.com/claude-code)
wai-quantum-service: the cutrank method, exact samples of a Clifford circuit measured in a product basis
All checks were successful
standards workspaces build + test / which workspaces this change touches (pull_request) Successful in 1m26s
Details
wai-quantum service / does this change touch it (pull_request) Successful in 31s
Details
wai-quantum service / service · linux (pull_request) Successful in 49s
Details
repository checks / lockfiles, tests in CI, workflow hygiene, spec and section references, public text (pull_request) Successful in 4m2s
Details
standards workspaces build + test / joule-code-rs (joule-code/joule-code-rs, joule-code-rs, linux-arm64, true) (pull_request) Successful in 9m51s
Details
standards workspaces build + test / macOS energy backends (pull_request) Successful in 9m16s
Details
wai-quantum service / service · macos (pull_request) Successful in 47s
Details
855822e2d7 
A Clifford circuit followed, on each qubit, by one-qubit gates at any angles
(qasm::Program::product_measured splits it: each qubit's gates from its first
non-Clifford one on, multiplied into one matrix; a gate on several qubits
after that is refused). The planner takes the graph form, searches orders for
narrower cuts with a move budget of about a second, and admits the job by its
largest table (double precision when it fits, single otherwise) and by the
entries its passes touch. The run rebuilds the same order.

Answers: counts, exact samples; amplitudes, exact up to one phase shared by
the job's amplitudes, which the result's global_phase says (a tableau does not
fix a state's global phase). Expectation values are refused. auto tries it for
counts after the state vectors and before mps.

Tests: amplitudes equal the state vector's times one phase (6, 9 and 11
qubits, 24 amplitudes each); counts follow the exact distribution (40 000
shots) and repeat with their seed; auto samples a 40-qubit ring with chords by
cutrank; refusals name the gate. Mutants caught: the tail multiplied in the
wrong order, samples mapped to the wrong qubits. smoke.sh samples a 60-qubit
instance (max cut 10 bits).

Co-Authored-By: Claude Opus 5.5 <[email protected]>
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
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!342
No description provided.