How to calculate onsite observerbles more efficiently?

ITensor Julia Questions
,
1

Hi there,

I find the function measurelocalops!() in the function function measure!(obs::DMRGObserver; kwargs...) in the source code \src\ITensorMPS\observer.jl.

I redefined this function as

function ITensors.measurelocalops!(obs::ObserverDemo, wf::ITensor, i::Int)
    for o in obs.onsite_ITensor_ops
        # Moves to GPU if needed
        # @show o
        oⱼ = ITensors.adapt(ITensors.datatype(wf), ITensors.op(obs.sites, o, i))
        m = ITensors.dot(wf, ITensors.apply(oⱼ, wf))
        imag(m) > 1e-8 && (@warn "encountered finite imaginary part when measuring $o")
        obs.measurements[o][end][i] = real(m)
    end
end

in my own code with the version of ITensors v0.3.35, and it works much more efficiently than inner(psi', MPO(onsite_mpo, obs.sites), psi). However, when I update ITensors to version v0.3.68, this redefined function breaks down, while the source code does not change much.

How can I still use the more efficient way to calculate averages of the onsite operator with the updated version of ITensors?

Thank you so much for your time!

Best wishes.

2

Can you share a fully runnable minimal example that reproduces the issue you are seeing, along with the full error message (as instructed in Please Read: Make It Easier to Help You)? Additionally, can you update to the the latest version of ITensors.jl (v0.6.13) and try on that version?

3

Thank you for your response! After running additional tests, I discovered a bug in one of my functions, so the issue isn’t connected to ITensors.