implement truncation_error keyword arg for truncate!

[NuclearPowerNerd]

The purpose of this commit is to allow the user to access the truncation error that is internally calculated in a call to truncate!. This PR implements #96.

I have implemented this by allowing the user to pass a Ref object to the call to truncate!. The result of each SVD performed during a call to truncate! is then accumulated in that Ref.

I also added a new test for this functionality. I then re-ran all tests. There were 75165 passing and 33 broken for a total of 75198 tests. I also updated the docstring for truncate! to reflect this new keyword argument.

Here is an example of the new functionality.

A few differences with what was suggested in #96:

• I accumulate the error rather than storing the truncation error for each bond individually.
• I did not use the ! in my keyword argument. Was that just for stylistic reasons or is there some other convention for doing that (I'm aware of the convention in function names but not in variable names)?

truncation_error = Ref{Float64}()
truncation_error[] = 0.0
truncate!(someMPS, maxdim=4, cutoff=1E-5, truncation_error=truncation_error)
truncation_error[]

[NuclearPowerNerd]

One thing I noticed while playing around with this new feature is that sometimes the truncation error exceeds the cutoff. But I thought it should be that truncation_error[] <= cutoff.

The following MWE illustrates this. I'd be curious what you think. If you want I can open a separate issue about this if/once this has been merged.

d = 2
N = 7
n = d^N
sinds = siteinds(d, N)
x = collect(LinRange(0, 1, n))
y = @. 2 * x + 3 + sin(2π * x)
χ = 8

truncation_error = Ref{Float64}()

cutoff_ = 1E-4
Y = MPS(y, sinds, maxdim=χ, cutoff=0.0)
truncation_error[] = 0.0
truncate!(Y, maxdim=χ, cutoff=cutoff_, truncation_error=truncation_error)

truncation_error[]
truncation_error[] > cutoff_  # returns true!

cutoff_ = 1E-5
Y = MPS(y, sinds, maxdim=χ, cutoff=0.0)
truncation_error[] = 0.0
truncate!(Y, maxdim=χ, cutoff=cutoff_, truncation_error=truncation_error)

truncation_error[]
truncation_error[] > cutoff_ # returns false, as expected

[mtfishman]

The cutoff refers to the truncation of each SVD performed, if multiple SVDs are performed the total error could add up to a value larger than the cutoff. It could make sense to store a truncation error for each bond of the MPS. That's one reason why I'm a bit hesitant about this PR, since I'd prefer to think about this more generally in terms of what kinds of other information we might want to output and how it should get output.

[NuclearPowerNerd]

The cutoff refers to the truncation of each SVD performed, if multiple SVDs are performed the total error could add up to a value larger than the cutoff.

Ah, duh. Yeah so I will push an update here shortly and make it store per bond truncation error which is the more sensible thing to do. Thanks for the feedback.