Refactor single ion anisotropies #93
Conversation
- Replace `set_anisotropy!` with `set_onsite!`. The new function expects a matrix built from `stevens_operators` or `spin_operators`. Provide `print_stevens_expansion` to decompose polynomials of spin operators into a linear combination of Stevens operators. - Make DynamicPolynomials an optional dependency. This provides symbolic functions `print_classical_stevens_expansion` and `print_classical_spin_polynomial`. - Functions to create tensor products of operators.
|
There are various breaking changes in this commit, but the main one is the replacement |
| Λ′ = rotate_operator(Λ, R) | ||
| @test Sunny.is_anisotropy_valid(cryst, i, Λ′) | ||
|
|
||
| latvecs = lattice_vectors(1.0, 1.1, 1.0, 90, 90, 90) | ||
| cryst = Crystal(latvecs, [[0., 0., 0.]]) | ||
| # print_site(cryst, i) | ||
| Λ = randn()*(𝒪[6,0]-21𝒪[6,4]) + randn()*(𝒪[6,2]+(16/5)*𝒪[6,4]+(11/5)*𝒪[6,6]) | ||
| Λ = randn()*(O[6,0]-21O[6,4]) + randn()*(O[6,2]+(16/5)*O[6,4]+(11/5)*O[6,6]) |
There was a problem hiding this comment.
Using O here (instead of 𝒪) doesn't look great when juxtaposed with coefficients. The tradeoff is that some people's terminals won't render 𝒪 properly, so I was thinking it might be better to use plain ASCII in the main examples.
There was a problem hiding this comment.
Probably, users will make themselves clear by inserting * for clarity. Also consider that this form of juxtaposed coefficients is a special feature of julia, so people are already wary. This term: (O[6,2]+(16/5)*O[6,4]+(11/5)*O[6,6]) looks fine to me, and even more fine with spaces (O[6,2] + (16/5)*O[6,4] + (11/5)*O[6,6]).
Definitely Unicode-optional is the way to go.
| @@ -7,7 +7,6 @@ | |||
|
|
|||
| using Sunny, GLMakie | |||
| using Statistics: mean | |||
| #md Makie.inline!(true); #hide | |||
There was a problem hiding this comment.
It was previously needed as a workaround for this bug MakieOrg/Makie.jl#2805.
| @@ -0,0 +1,126 @@ | |||
| # Produces a matrix representation of a tensor product of operators, C = A⊗B. | |||
| # Like built-in `kron` but with permutation. Returns C_{acbd} = A_{ab} B_{cd}. | |||
| function kron_operator(A::AbstractMatrix, B::AbstractMatrix) | |||
There was a problem hiding this comment.
There was a problem hiding this comment.
Good catch. The kron from LinearMaps is semantically the same. It might be lazy, which would save work at construction time, but I suspect it wouldn't significantly reduce the total amount of work required (e.g. there will eventually be an SVD operation).
Eventually we should have a chat about the cost of constructing Hamiltonians in the context of inverse modeling. If optimization and LSWT becomes really fast, it could be that building the Hamiltonian is the dominant cost (e.g. symmetry analysis can sometimes be really slow). I have some thoughts to run by you.
|
Looks great! Good to merge as-is |
Replace
set_anisotropy!withset_onsite_coupling!. The new function expects a matrix built fromstevens_operatorsorspin_operators. Provideprint_stevens_expansionto decompose polynomials of spin operators into a linear combination of Stevens operators.Make DynamicPolynomials an optional dependency. This provides symbolic functions
print_classical_stevens_expansionandprint_classical_spin_polynomial.Functions to create tensor products of operators.