RB circuits should be made of Clifford objects, not a converted instruction #1023
Assignees
Labels
bug
Something isn't working
Comments
github-merge-queue bot
pushed a commit
that referenced
this issue
Jan 18, 2024
### Summary Make the Clifford synthesis algorithm for RB circuits pluggable (implementing it as a `HighLevelSynthesisPlugin`). Fixes #1279 and #1023. Change to accept Clifford elements consisting only of instructions supported by the backend for `interleaved_element` option in `InterleavedRB`. Speed up 2Q RB/IRB for backends with unidirectional 2q gates, e.g. IBM's 127Q Eagle processors. ### Details and comments Previously, for 3Q+ RB circuits, entire circuit is transpiled at once and hence for each of the resulting Cliffords, the initial and the final layout may differ, that means sampled Cliffords are changed during the transpilation. Also in the worst case, the resulting circuit may use physical qubits not in the supplied `physical_qubits`. To avoid that, this commit changes to transpile an RB sequence Clifford by Clifford. The Clifford synthesis algorithm (`rb_default`) is implemented as a `HighLevelSynthesisPlugin` (see `RBDefaultCliffordSynthesis` in `clifford_synthesis.py`), which forces the initial layout (i.e. guarantees the initial layout = the final layout) and physical qubits to use. As a byproduct, the performance of 2Q RB/IRB for backends with directed 2q gates (e.g. IBM's 127Q Eagle processors) is drastically improved. For those cases, previously we had to rely on `transpile` function to make generated circuits comply with the coupling map, however, after this commit, we can synthesize Cliffords with considering the 2q-gate direction and go through the fast path introduced in #982. Depends on Qiskit/qiskit#10477 (qiskit 0.45) --------- Co-authored-by: Helena Zhang <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Informations
Current StandardRB and IterleavedRB circuits are built from a Clifford converted to an instruction, which contains the decomposition inside it. In addition to making the circuits heavy, this prevents custom compiling routines on the Cliffords which were recently added to Qiskit (Qiskit/qiskit#7966).
So it's better to build the circuits from Clifford objects and not clifford.to_instruction().
The text was updated successfully, but these errors were encountered: