expectation value of a one-site operator for MPS using mixedcanonical form #384
Conversation
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## master #384 +/- ##
=========================================
Coverage ? 43.81%
=========================================
Files ? 25
Lines ? 1383
Branches ? 0
=========================================
Hits ? 606
Misses ? 777
Partials ? 0 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
|
yes, definitely! I just wanted to get the discussion started, since we definitely need a function to compute exp values after it disappeared in the latest refactorings |
There was a problem hiding this comment.
thanks for starting the work on this @starsfordummies
@jofrevalles yeah, the idea is to dispatch based on the CanonicalForm so that we can specialize behavior
the specialized implementations should do the following
- on
NonCanonical, i would just do a sweep contracting the tensors - on
MixedCanonical,canonize!to the range of sites of the operator- directly connect boundary virtual bonds between ket and bra (skipping the tensors outside of the range of the operator)
- call
contract
- on
BondCanonical,canonize!to the range of sites of the operator (so move toMixedCanonical) and callexpectagain but on theMixedCanonicalform - on
VidalGauge, i would directly take the range of sites of the operator (so move toMixedCanonical), absorb the lambdas and contract
if you know a more efficient way, we can do it that way
src/Components/MPS.jl
Outdated
| """ Expectation value for a one-site operator `op` on site `op_site` of mps `psi`, brings `psi` to canonical form. """ | ||
| function expect_1site!(psi::MPS, op::AbstractArray, op_site::Int) | ||
| canonize!(psi, MixedCanonical(site"$(op_site)")) | ||
| plug_ind = ind_at(psi, plug"$(op_site)") | ||
| temp_ind = Index(:_temp_bra) | ||
| ev = binary_einsum(psi[op_site], Tensor(op, [temp_ind, plug_ind])) | ||
| ev = binary_einsum(ev, conj(replace(psi[op_site], plug_ind => temp_ind))) | ||
|
|
||
| return ev | ||
| end | ||
|
|
||
| """ Expectation value for a one-site operator `op` on site `op_site` of mps `psi`. | ||
| This makes a copy of the MPS """ | ||
| function expect_1site(psi::MPS, op::AbstractArray, op_site::Int) | ||
| psi = copy(psi) | ||
| expect_1site!(psi, op, op_site) | ||
| end |
There was a problem hiding this comment.
mind moving them to a Operations/expect.jl file please?
Co-authored-by: Sergio Sánchez Ramírez <[email protected]>
|
i added a few methods for exp values of local ops and mpo - it could well be that bringing the MPS to canonical form is always the best way (even if it involves making a copy+canonizing it), need to benchmark it |
3 participants
probably not too tenet-idiomatic but should work