precompute contraction GL*A and A*GR in JordanMPO_AC_Hamiltonian and JordanMPO_AC2_Hamiltonian - #330
precompute contraction GL*A and A*GR in JordanMPO_AC_Hamiltonian and JordanMPO_AC2_Hamiltonian#330AFeuerpfeil wants to merge 21 commits into
Conversation
|
Your PR requires formatting changes to meet the project's style guidelines. Click here to view the suggested changes.diff --git a/src/algorithms/derivatives/mpo_derivatives.jl b/src/algorithms/derivatives/mpo_derivatives.jl
index 02ec00b..587a45f 100644
--- a/src/algorithms/derivatives/mpo_derivatives.jl
+++ b/src/algorithms/derivatives/mpo_derivatives.jl
@@ -27,10 +27,10 @@ const MPO_AC2_Hamiltonian{L, O₁, O₂, R} = MPODerivativeOperator{L, Tuple{O
MPO_AC2_Hamiltonian(GL, O1, O2, GR) = MPODerivativeOperator(GL, (O1, O2), GR)
const MPO_Contracted_AC_Hamiltonian{L, R} = MPOContractedDerivativeOperator{L, R, 1}
-MPO_Contracted_AC_Hamiltonian(GL::L, GR::R) where {L, R} = MPOContractedDerivativeOperator{L,R,1}(GL, GR)
+MPO_Contracted_AC_Hamiltonian(GL::L, GR::R) where {L, R} = MPOContractedDerivativeOperator{L, R, 1}(GL, GR)
const MPO_Contracted_AC2_Hamiltonian{L, R} = MPOContractedDerivativeOperator{L, R, 2}
-MPO_Contracted_AC2_Hamiltonian(GL::L, GR::R) where {L, R} = MPOContractedDerivativeOperator{L,R,2}(GL, GR)
+MPO_Contracted_AC2_Hamiltonian(GL::L, GR::R) where {L, R} = MPOContractedDerivativeOperator{L, R, 2}(GL, GR)
# Constructors
# ------------
@@ -169,4 +169,4 @@ function (h::MPO_Contracted_AC2_Hamiltonian)(
)
@plansor y[-1 -2; -3 -4] ≔ h.leftenv[-1 -2 5; 1 2] * x[1 2; 3 4] * h.rightenv[3 4 5; -3 -4]
return y isa AbstractBlockTensorMap ? only(y) : y
-end
\ No newline at end of file
+end |
lkdvos
left a comment
lkdvos
left a comment
There was a problem hiding this comment.
Amazing, thanks for the quick work! Do you already have some idea of the performance difference for your specific case?
I'll try and come up with some reasonable set of cases to actually benchmark this on, I'd like to experiment with how much of a difference it actually makes even for the more modest MPO sizes too.
|
Thanks for your comments. |
Codecov Report❌ Patch coverage is
... and 2 files with indirect coverage changes 🚀 New features to boost your workflow:
|
…e code from mpo_derivatives
|
I moved my changes to the mpo_derivatives by creating a new struct
|
|
I know what's causing the test failures, I'll have a look later today to see if I can resolve this in a better way. |
…d derivates for MPO{<:MPOTensor}
…AFeuerpfeil/main
|
(@lkdvos: Please ignore the latest commit, I forgot to create a new branch) |
This reverts commit 6950267.
2 participants
This PR changes the way
JordanMPO_AC2_HamiltonianandJordanMPO_AC_Hamiltoniandeal with the A-block of the MPO:To reduce contraction const in the application to an MPS
MPSTensor, we contractGL*AandA*GRin the constructor for the 2-site derivative andGL*Afor the 1-site derivative (absorbing A into the left environment is purely conventional. If the mpo leg of the left and right environments have different sizes, one of the 2 conventions could be superior to the other.)