Effective masses

[fzierler]

Effective mass for Numbers and DataPoints including error propagation. Still needs tests and requires #20.

[briederer]

First look seems fine to me.
Have you also tried building it with the Package mentioned in https://inspirehep.net/literature/1837727 instead of DataPoint from #20?

[briederer]

Also the mentioned package ADerrors.jl seems to be very finished.
Maybe we could contact the maintainer and join forces?

[fzierler]

Also the mentioned package ADerrors.jl seems to be very finished.
Maybe we could contact the maintainer and join forces?

I tried to use ADerrors.jl for performing the propagation of uncertainty of the effective masses, but the resulting error was always an order of magnitude smaller than with the methods in this PR. It might just be that I do not fully understand the error reduction techniques in ADerrors.jl (I will first need to read the references provided in that repo), but for now I am hesitant to start relying on that package since I have not fully understood it (yet?).

Maybe we should think of this package (at least for now) as a collection to most naive implementations for performing lattice spectroscopy. Once we have this working we can add more sophisticated ways of doing this. I will definitely continue to study the details of ADerrors.jl and I think at some point we might be able to provide error propagation using ADerrors.jl in addition to the naive implementations.

We definitely should mention this package in the README.

[briederer]

I tried to use ADerrors.jl for performing the propagation of uncertainty of the effective masses, but the resulting error was always an order of magnitude smaller than with the methods in this PR. It might just be that I do not fully understand the error reduction techniques in ADerrors.jl (I will first need to read the references provided in that repo), but for now I am hesitant to start relying on that package since I have not fully understood it (yet?).

Okay seems reasonable.

Maybe we should think of this package (at least for now) as a collection to most naive implementations for performing lattice spectroscopy.

By "this package" you mean our package (i.e. LatSpec.jl) am I right?

We definitely should mention this package in the README.

Feel free to add it immediately.

[fzierler]

By "this package" you mean our package (i.e. LatSpec.jl) am I right?

Yes

[briederer]

Found a few more things to be resolved.

[github-actions]

The Docs for this PR are deployed! 🥳
You can find them here.

Please make sure that you have documented your new feature properly:

• Added documentation (if needed).

[briederer]

Is LinearAlgebra. necessary here?
Not used in the other effective_mass_err-function.

[briederer]

Okay since you called a variable norm in this function it is reasonable to use explicitly LinearAlgebra.norm() here.
However then I suggest to change it in line 16 too.

[briederer]

Not sure about the abs() function here.
Personally I prefer to keep the minus sign explicitely. Though I am not sure what is actually the correct way according to the community.

[fzierler]

I don't think that there is a correct way of doing this, but all books seem to use the version without abs(). Of course this does not matter if you only consider the range up to N_T/2 so I am fine with dropping the abs().

[briederer]

Yeah I also think without abs() is a bit cleaner (and probably minimally faster 😉).
Especially without this you can more easily differ between a sinh()- or cosh()-behaviour wether the interpolators are sink or source ones.

[briederer]

I guess this would need an adaption to the new PR in #25.
Should I merge the current state of this PR into #25 and resolve the differences and close this one?
In the end effective masses should be a fundamental feature of LatSpec so it should be fine to have one submit to master where we rework the overall type and add this.

[fzierler]

I guess this would need an adaption to the new PR in #25. Should I merge the current state of this PR into #25 and resolve the differences and close this one? In the end effective masses should be a fundamental feature of LatSpec so it should be fine to have one submit to master where we rework the overall type and add this.

Yes, I think this is a good approach.