Performance comparison

[meandmytram]

Hey guys, amazing library! Just out of curiosity, was there any benchmarking done against other TN packages like tensornetwork or quimb?

[refraction-ray]

Thanks for your good question.

TensorCircuit is built on top of TensorNetwork package with built-in differentiable circuit class and some hacks on the tensor network contraction path finder (TensorCircuit supportscotengra package as the path finder).
Therefore, TensorCircuit is definitely the choice over TensorNetwork package in terms of quantum circuit simulation with its handy API and better performance.

quimb is another fantastic package from the same author for cotengra. In terms of API design, we believe that TensorCircuit has better Circuit object with more built-in functionalities since our package is more focused on circuit simulation while quimb is a more general tensor network package. In terms of performance, careful benchmarks are required and we are on the way. For now, what I can say is that TensorCircuit package focuses more on end-to-end jit and vmap paradigms, which may bring performance advantages compared to quimb.