Ryan Levy

Looks like there's some assumption being made that for a sliced mps e.g. `psi[2:4]` returns a vector of itensors instead of a MPS, not sure if we want to adjust...

Re the sliced MPS, seems the assumption is setting/getting a unit range ```julia function deepcopy_ortho_center(M::AbstractMPS) M = copy(M) c = ortho_lims(M) # TODO: define `getindex(::AbstractMPS, I)` to return `AbstractMPS` M[c]...

Just pushed a hopefully better performance version of `sqrt`. The original logic is a bit strange to me, it will fail for any matrix that isn't PSD without a large...

@mtfishman now that ITensorMPS is out of ITensors I think this PR is complete pending your review and I'll move over there to fix the slice issue

Thanks, I implemented all of those suggested changes. Do you have any suggestions on how to handle the complex conversion? It seems that `map_diag` won't convert with a type change...

Thanks Matt, it seems that I'm not familiar enough with the NDTensors internals to determine what should be changed. That snippet doesn't seem to work as it gets a call...

I just ran into this problem as well. I suspect the issue is only when `krylovdim==n` (the default is 30), then something is wrong with the [HA/HB](https://github.com/Jutho/KrylovKit.jl/blob/52f1a781fc07737c52ecf507f5e45851e52ca4ee/src/eigsolve/golubye.jl#L91) matrices. For example...