Draft: MBQC Module Refactor with ZX Calculus

[masa10-f]

TODO

• graph state module
  • Abstract Graph State(define API)
  • ZXGraphState(pyzx implementation) Will be implemented by ourselves in another issue bacause it's not the fundamental functionality in MBQC
    • [ ] cover API
    • optimization
  • MBQCCircuit(Unitary Construction)
    • XY-only graph state
    • YZ (and XZ) implementation
• flow module
  • flow and gflow (use fastflow instead)
  • focusing
  • xflow and zflow for non-deterministic MBQC
• transpiler module
  • incorporate extended MBQC(see transpiler.generate_m_cmd)
  • standardized pattern(transpiler.transpiler)
  • minimized pattern
• command module
  • measurement command(migrated from the original repo)
  • pattern definition(will be done in original repo?)
• pattern and circuit simulator(will be migrated from the original repo)
• Introduce more strict setting of ruff
• Write up docsting in numpy style
• Set up mypy CI
• Update data type with collections module
• add more test

Ref:

• Backens, Miriam, Hector Miller-Bakewell, Giovanni de Felice, Leo Lobski, and John van de Wetering. 2021. “There and Back Again: A Circuit Extraction Tale.” Quantum 5 (421): 421.

[masa10-f]

Test Status

• pattern object
• circuit object
• graph state object
• statevec sim
• pattern sim
• circ pattern comparison check
• transpiler
  • standardized
  • focused(checked focus property but not in a pattern form)
  • reduce space(pattern concatenation is required. I'm tackling it now)

Note

This repo is currently focusing on refactoring methods related to the pattern module in graphix. The resource optimization itself is out of scope now although this repo provides natural interface(GraphState) for it(Pauli preprocessing or ZX calculus). Currenlty I think It might be better to create ZX wrapper after the above TODO list is completed and merged into graphix.

[masa10-f]

memo: plan to create mutable and immutable pattern. The former has a measurement calculus method as in the original repo. The latter is fixed pattern generated from the new transpiler.

[masa10-f]

I have refactored pattern object. Now it splits into two steps. One is ImmutablePattern that is passed to simulator or executor. We can not modify it. The other is MutablePattern, which is the basically same as the Pattern in the original repo. I also made an abstract class BasePattern which determinies shared APIs.

I have noticed that add method is not guaranteed to preserve runnability and determinicity of the pattern.

[masa10-f]

Next, change data type of domain list[int] -> set[int] and input/output_nodes list[int] -> set[int]. I also plan to create BasePattern.q_indices variable to record the qargs, which currently preserved as indices of list but I believe it's not good practice to record two differenet information in a single object.

[masato-fuk]

memo: is minimize space necessary in mutable pattern?

[masa10-f]

Next, change data type of domain list[int] -> set[int] and input/output_nodes list[int] -> set[int]. I also plan to create BasePattern.q_indices variable to record the qargs, which currently preserved as indices of list but I believe it's not good practice to record two differenet information in a single object.

I have implemented this. It is probably necessary to add more tests to ensure it. Anyway, I proceed to test space reduction methods in resouce_opt.py

[masa10-f]

I proceed to test space reduction methods in resouce_opt.py

I have made it work and added example to generate reduced space pattern decomposed subgraph sequence. It seems to need a little refactoring. Next add an example of signal shifted pattern generation

[masa10-f]

Next things to do

• implement phase gadget in circuit
• Macro gate module in circuit class. This includes common gates in circuit libraries(e.g. Rx, Rz, Ry, U3, H etc.)
• Local Clifford on output vertices
• Graph State visualizer(unify pattern visualizer in the main repo. there's no need to specify the type of flows.)
• Macros to convert graph state into pattern easily, for those who familiar with measurement calculus seems not necessary
• Consider input format from pyzx(necessary info. is graph with phases) we will host ZX Calculus ourselves

[masa10-f]

As for local Clifford on outputs, Euler decompisition seems the most natural way on the perspective of ZX-Calculus. I will implement it after reflecting the feedback of the last offline meeting.

[masa10-f]

I encounterd 695 errors when I try to introduce the recommended ruff setting. I will improve my code step by step......

[masa10-f]

I will merge this PR once after the updated TODOs are completed, which means the fundamental modules will have been implemented sparsely and properly aligned with the underlying theory.
After that we will develop a density matrix backend with a little update, self-developed ZX backend, and more advanced resource state optimizer for our research purposes

[masa10-f]

cc. @nabe98 @d1ssk