Additions needed for sweeping algorithms on trees #18
Conversation
Remove tree graph paths (merged into NamedGraphs.jl)
Extend `models.jl` and remove `test/utils.jl` Change lazy `inner` to explicit `inner_network`
Add support for choosing a subset of sites in `expect`, similar to MPS implementation Condition expectation value type on `leaf_eltype` of state
|
@mtfishman @leburgel : More of a minor comment, but should the |
That choice is mostly due to my bias towards being more explicit and avoiding abbreviations as much as possible. I don't see the disadvantage of providing the aliases which people can use if they want (and we can use internally if we decide that style is preferred). |
I stuck to explicit type names and short aliases since the choice had already been made for |
src/treetensornetwork/projttno.jl
Outdated
| mutable struct ProjTTNO <: AbstractProjTTNO | ||
| pos::Union{Vector{<:Tuple},NamedDimEdge{Tuple}} # TODO: cleanest way to specify effective Hamiltonian position? | ||
| H::TTNO | ||
| environments::Dictionary{NamedDimEdge{Tuple},ITensor} |
There was a problem hiding this comment.
It may be nice to make this a directed DataGraph.
There was a problem hiding this comment.
Could be a good idea, adding/invalidating environments by adding/removing directed edges sounds nice. I would have to think about what to do with the inside the operator position for a 2-site effective operator (remove the corresponding edges altogether?) and whether I could get rid of some other things by using more DataGraph-like methods instead.
src/treetensornetwork/projttnosum.jl
Outdated
| """ | ||
| ProjTTNOSum | ||
| """ | ||
| mutable struct ProjTTNOSum |
There was a problem hiding this comment.
Again, I think this should not be mutable.
I was also thinking about using the Applied system in ITensors.LazyApply to avoid having to actually define types like this. That would be a more general way to represent lazy combinations of projected TTNs (i.e. help with representing more general combinations, like sums of projected TTNS and TTNO, sums of TTNOs applied to TTNS, etc. without having to hardcode specific types).
Maybe we could chat about how that might work, this would be a good opportunity to rethink some of the type designs here (some of these designs are holdovers from porting the C++ version).
There was a problem hiding this comment.
This definitely sounds like a nice idea.
Merge updates for new `NamedGraphs.jl`/`DataGraphs.jl` interface Remove unnecessary `ITensors.convert_leaf_eltype` calls in favor of `ITensors.convert_eltype` Add `@test_broken` for `swapprime(::AbstractITensorNetwork, ...)` Use to `Graphs.degree` in `OpSum` -> `TTNO` converter Remove `siteinds(all/only, ...)` and `linkinds(all/only, ...)` syntax
|
@leburgel you bring up a lot of good questions and points in your first post but it is tough to try to discuss and resolve them from this PR. Could you split up questions that haven't been resolved yet into separate issues? |
Yes, I'll drop the things that are no longer relevant and move the rest to new issues. |
| return collect(Base.Iterators.flatten(edges)) | ||
| end | ||
|
|
||
| function internal_edges(P::AbstractProjTTNO{V})::Vector{NamedEdge{V}} where {V} |
There was a problem hiding this comment.
This is a good idea for a generic graph function. It is the complement of the edge boundary of a subgraph (https://en.wikipedia.org/wiki/Boundary_(graph_theory) and boundary_edges added in ITensor/NamedGraphs.jl#20).
As we discussed, a simple implementation would be internal_edges(g::AbstractGraph, vertices) = edges(subgraph(g, vertices)). Then this function could be defined as internal_edges(P::AbstractProjTTNO) = internal_edges(underlying_graph(P), sites(P)).
Codecov Report
@@ Coverage Diff @@
## main #18 +/- ##
===========================================
- Coverage 80.17% 65.36% -14.82%
===========================================
Files 29 37 +8
Lines 1226 2105 +879
===========================================
+ Hits 983 1376 +393
- Misses 243 729 +486
Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here. |
Work in progress, this adds some functionalities for tree tensor networks which are needed to be able to run sweeping algorithms on them. It's quite a lot, so in addition to raw code review I'll try to summarize the main additions below, as well as start a list of things I thought might be good to discuss while coding (some of which are not very relevant and should just be filed as issues...).
Summary of additions:
ITensorNetwork:ITensors.replaceinds:linkindsandsiteindsfilters similar to theITensors.MPScaselinkdimsAbstractTreeTensorNetworkTreeTensorNetworkOperator(new type)ITensors.OpSuminner)ProjTTNO,ProjTTNOApplyandProjTTNOSum(new types)