Statistical error?

[akio-tomiya]

Hi, now I can successfully execute the code. I want to know, how can I put statistical error for sampling? I think there are two sources, one is for p and another is from quantum part. What values in JLD2 is the mean values, right?

[wangleiphy]

What values in JLD2 is the mean values, right?

That's right.

You can estimate error bars using standard deviation of samples (compute std in addition to mean) https://github.com/wangleiphy/BetaVQE.jl/blob/master/src/free_energy.jl#L16 The error bar will contain only thermal fluctuation since we compute exact quantum expectation value there.

If needed, you can mimic quantum measurement to get additional statistical noises in the quantum part.

[akio-tomiya]

Thanks, I understand that.

If needed, you can mimic quantum measurement to get additional statistical noises in the quantum part.

Do you mean some numerical evaluation in Yao.jl? (I don't know Yao has such functionality though. It corresponds to "shot experiment" function in other framework.)

• (addition for clarification)

Do you mean exact diagonalization used in the training and/or evaluation of expectation values?

[wangleiphy]

I meant we compute quantum expectation <psi|H|psi> and its gradient exactly in the code.

Here is an example of getting stochastic gradient from "shot measurement" https://github.com/wangleiphy/YaoTutorial/blob/master/vqe.jl#L16

[akio-tomiya]

Thank you very much!