Add basic error correction codes in qutip-qip algorithms

[rochisha0]

This PR aims to add basic error correction codes in qutip-qip algorithms. #264

• Create files for phase-flip, bit flip, and shor code.
• Write tests

[rochisha0]

@BoxiLi this is ready for review!

[BoxiLi]

Thanks a lot! This looks really cool. sorry for the delay.

The code looks pretty good to me. Two general questions before a more detailed review:

• Is class and three static methods the best way to define this? I understand that the point here is that all of them have encoding, correction and decoding. So they share the same structure. However, in this way, one always has to first define a class instance and then use those methods. Do you know how other packages handle this?
• This could be for this RP or a separate PR, depending on your opinion. Is there a way to make the function more general. For instance, the bit flip encoding circuit is defined for encoding qubit 0 to qubit 0,1,2. Can we define such a function like encode_circuit(qubit=0, ancilla=[1,2]) such that the resulting circuit will be constructed depending on the input parameters?. In this way one can generate multi-qubit error correction circuit based on those elementary functions.

[BoxiLi]

Another question, usually, the dencode circuit has measurement and classically controlled operations (for example here https://quantumcomputing.stackexchange.com/questions/6501/2-ways-to-do-the-three-qubits-bit-flip-code), which I don't see in the code?

Could you also add some reference links/papers as inline comments to the code?

[rochisha0]

Hello @BoxiLi , thanks for the review. I understood that the code needs to be more modular and reusable.

The two circuits given in stackoverflow are given in the code in two different parts. The first circuit is represented in decode_circuit and the other one in syndrome_measurement. I will try to make this clearer.

[rochisha0]

A lot of the structure is taken from https://github.com/qiskit-community/qiskit-community-tutorials/blob/master/awards/teach_me_quantum_2018/intro2qc/10.Quantum%20error%20correction.ipynb?utm_source=chatgpt.com and Neilsen and Chuang.

[rochisha0]

@BoxiLi , it's ready for review again!

[BoxiLi]

Thanks!, yes, indeed the major point is to make the code more reusable.

I have one more suggestions, in the current format, all the information about the qubit indices is saved in the class instance. I would rather argue that this should be given when the method is called rather than saved the class instance is initialized. That is:

qec_generator = BitFlipCode()
circ1 = qec_generator.encode_circuit(data_qubits=[0,1,2], ...=[3,4])
circ2 = qec_generator.encode_circuit(data_qubits=[5,6,7], ...=[8,9])

instead of (current code)

qec_generator = BitFlipCode(data_qubits=[0,1,2], ...=[3,4])
circ1 = qec_generator.encode_circuit()
qec_generator = BitFlipCode(data_qubits=[5,6,7], ...=[8,9])
circ2 = qec_generator.encode_circuit()

I guess the main point is why we need to define a class and make functions like encode_circuit class methods instead of just define bit_flip_encode_circuit. I understand that you define the class like this because all the QEC class share similar methods. Later if we have some meta method for all QEC circuits, we can define a parent class. This is fine. I imagine this class as a gagdet that knows how to generate different QEC methods.

But then it should not save the qubit indices, instead, it should only save information about the QEC circuit as a whole piece. Something that is needed to define this circuit, in a qubit independent way. This could be e.g., the number of data qubit used in the repetition code, right now it is three, but in principle it can be any N! Those are information about the abstract QEC circuit, but not about its embedding in a real N qubit system.

A direct consequence of this is making the code less redundant, as in the above code example.

What do you think?

[BoxiLi]

This should raise an InputError with the message "Bit-flip code requires 3 data qubits.". assert is used only for internal checking, but here it is meant to tell the user that they gave the wrong input.

[BoxiLi]

Also, in case you are not aware of it, qutip-qip also allows adding measurement operation
qc.add_measurement(...)
and classically control the following circuit depending on the measurement output
qc.add_gate(X, classical_control=..., classical_control_value...).
I can make a minimal example later. I don't have a coding environment right now.

The syndrome measurement and correction circuit can be defined using measurement and classically controlled operations I believe. Measurement operations should be added to the syndrome circuit correction circuit should be classically controlled gate instead of manually providing the syndrome measurement results to correction_circuit(self, syndrome). The resuling circuit should be like

[BoxiLi]

If you think it will help improving productivity I am also happy to have a chat online, maybe after 10th of June.

[rochisha0]

Hello @BoxiLi , thank you for your suggestions. I will try to incorporate these suggestions in my code.

I have a call going with @ajgpitch on every Tuesday, maybe you can join that one for a week or two?

[BoxiLi]

Sure, should be able to join on 10 June if you could share a link to my email.

[rochisha0]

@BoxiLi have made the discussed changes to bit_flip. If those look good I will make the same to other files and change the tests

[BoxiLi]

Thanks! Yes it is good to fisrt look at one part of it. I had a look at the BitFlipCode part.

I think you might build the code on the wrong base branch, because I do not see qpe.py, which you added last time. This could be the reason for the conflict. You can try to rebase the code upon the current upstream master.

It would be cool if we can combine three circuit part (with manually added error maybe?) and test it!

[BoxiLi]

I assume self.n_data and self.n_syndrome are things that are specific for this algorithm and we do not want people to change them, right? So I would suggest mark them as private members self._n_data and self._n_syndrome. In the Python convention, this means that those are private members and should not be accessed by users.

[rochisha0]

Yes, makes sense to make them private

[BoxiLi]

We should raise an error instead of using assert. Assert is for internal checking. A user should in principle never see assert error. Raise a ValueError saying the number is wrong is more appropriate here.

[BoxiLi]

Same, raise errors.

[BoxiLi]

Should be

qc.add_measurement(sq[0], sq[0], classical_store=0)

The first one is just the name of the measurement operation, the second one is the target qubit.

This triggers an error when plotting the circuit, you can use qc.draw() to test and see if it is what you expected.

[BoxiLi]

BTW I might find bug in classical controlled gate plotting. If you don't see the control line plotted, that is fine.