The synthesis function synth_cnot_count_full_pmh returns invalid circuits for certain section_size values

[alexanderivrii]

Environment

• Qiskit version: 1.0
• Python version: 3.10
• Operating system: Windows

What is happening?

The following code

from qiskit.synthesis import synth_cnot_count_full_pmh
from qiskit.synthesis.linear import random_invertible_binary_matrix

mat = random_invertible_binary_matrix(5, seed=1234)
qc = synth_cnot_count_full_pmh(mat, section_size=6)

produces an empty circuit.

How can we reproduce the issue?

Run the code above.

What should happen?

Setting section_size to a large number would probably produce suboptimal results, however it should still produce a valid (non-empty) circuit.

Any suggestions?

I believe that something wrong is going on when section size is larger than the size of the matrix, for instance in the above example setting section_size to a number 5 or lower does work correctly.

[Tarun-Kumar07]

Hey @ShellyGarion and @alexanderivrii, I'm new to the Qiskit code base and currently exploring issues to contribute to. I noticed this one and I'm interested in working on it. Would that be possible?

[ShellyGarion]

thanks @Tarun-Kumar07 - I've assigned you on this issue. Please let us know when you open a PR.

[Tarun-Kumar07]

Hey @ShellyGarion, I'm uncertain about the expected behavior here.
The synth_cnot_count_full_pmh implements the algorithm mentioned in Efficient Synthesis of Linear Reversible Circuits. According to the paper, we use section_size to partition the provided matrix.
Therefore, I believe section_size > num_qubits is not correct. Should I throw an exception in this case?

[ShellyGarion]

@Tarun-Kumar07 - Indeed, using section_size > num_qubits is not correct.
There are two options:

1. Raise a QiskitError and add a test that checks it.
2. When section_size > num_qubits set section_size = num_qubits and return the QuantumCircuit obtained in this case.

[Tarun-Kumar07]

I believe option 1 is preferable, as it provides clarity to the user by immediately alerting them to any violation of the condition

[ShellyGarion]

I believe option 1 is preferable, as it provides clarity to the user by immediately alerting them to any violation of the condition

OK, so please go ahead with the fix, and let me know when you open a PR

[Tarun-Kumar07]

Hey @ShellyGarion , I have opened a PR #12166.
Can you please have a look at it.
Thanks

[Tarun-Kumar07]

Hey @ShellyGarion,

Is there another task or issue I can contribute to? I've been searching through the good first issues but haven't found something suitable to work on.

[ShellyGarion]

Hey @ShellyGarion,

Is there another task or issue I can contribute to? I've been searching through the good first issues but haven't found something suitable to work on.

Indeed, the best way is to ask in Qiskit slack channels (as you did). Good luck!

[ShellyGarion]

Following the discussion in #12166 we observed that for some values of section_size we get wrong circuits.
In addition, we would like to choose the optimal section_size as a parameter of the number of qubits.

[Abdalla01001]

@ShellyGarion @alexanderivrii pls, can you assign me? I want to give it a look.

[ShellyGarion]

@ShellyGarion @alexanderivrii pls, can you assign me? I want to give it a look

Thanks @Abdalla01001, but please note that there is already an open PR #12166 by @Tarun-Kumar07.
Could you please take a look at the discussion there and see if you can help?