Improve the Generation of Random Circuits

[sbrandhsn]

What should we add?

The current generation of randomized circuits is fairly inflexible as it only allows to specify a limited number of circuit characteristics: the number of qubits, depth, maximum number of qubits in a quantum gate, and a couple of options allowing for measurements, conditional operations or resets in the random circuit. Ideally, the generation of random circuits would allow more parameters as an input, e.g. the structure of a circuit, the amount of parameterizable quantum gates, or a finer definition on the permissible types of quantum gates. I would like to use this issue to first discuss which parameters would be reasonable inputs and then track the progress on this.

I have a couple of parameters in mind:

• Density/sparsity
• Qubit connectivity, i.e. all-to-all connectivity, a line, 2D grid, random, clustered, sparse, dense, etc
• Distribution of parameterizable quantum gates to non-parameterizable quantum gates
• Distribution of gates not in a basis gate set
• Distribution of (non-)commuting gates
• Distrbution over n-qubit gates with n in N
• .... But please add any you would find useful.

[doublehoon]

Hi, I'm interested in this issue. I have also made modifications to the gateset directly in the library code of my local environment while generating random circuits. However, one concern is determining how precisely to implement those parameters because it might seem challenging. Could you provide any specific examples where you felt those parameters were necessary? I believe it would be beneficial to set appropriate goals and gradually add parameters as needed!

[sbrandhsn]

Hi @doublehoon awesome! I think this issue is deliberately defined in a broad way to allow and discuss all kinds of contributions. Some of the discussed parameters are also conflicting or correlating with each other, e.g. the number of CNOT gates might likely correlate with the density/sparsity of the quantum circuit. So, finding a general method that addresses all of these parameters seems difficult. Although this might be possible in a general way for small-scale instances with blackbox optimization techniques such as genetic algorithms.

Are you particularly interested in one set of parameters? I can imagine that it would be worthwhile to have different random circuit generators that each target a different aspect of the described parameters. An user could then select the best suited generator for their task or follow-up work can be done to combine the generators in a certain way.

[sbrandhsn]

Maybe https://arxiv.org/abs/2402.13352 can be used as a basis. :-)