[NDTensors] [BUG] denseblocks on DiagBlockSparse does not conserve QNs

[AFeuerpfeil]

Description of bug

When performing denseblocks on a DiagBlockSparse tensor, the output is a tensor, which in general does not conserve QNs and thus has no well-defined flux.

Minimal code demonstrating the bug or unexpected behavior

julia> i=Index([QN()=>2])
julia> j=Index([QN(("N",0))=>1,QN(("N",1))=>1])
julia> T=delta(dag(i),j)
julia> x=random_itensor(i)
julia> @show norm(x*T-x*denseblocks(T))

Expected output or behavior

The code should output 0.

Actual output or behavior

The output, however, is nonzero.
The reason for that is that T conserves the QNs. However, denseblocks(T) does not, which can be seen from the block-structure. ITensors, however, still asigns denseblocks(T) and x a flux QN(), even though their flux is not well-defined.

I suspect the problem is related to how NDTensors stores a diagblocksparse tensor internally, as it only stores one matrix element. When performing denseblocks, it probably does not check which flux was used in the constructor delta([::Type{ElT} = Float64, ][flux::QN = QN(), ]is)

julia> @show norm(x*T-x*denseblocks(T))
1.0470682860833809
julia> @show flux(T)
QN("N",0)
julia> @show flux(denseblocks(T))
QN("N",0)
julia> @show flux(x*denseblocks(T))
QN("N",0)
julia> @show ITensors.matrix(T)
[1.0 0.0; 0.0 1.0]
julia> @show ITensors.matrix(denseblocks(T))
[1.0 0.0; 0.0 1.0]

Version information

• Output from versioninfo():

julia> versioninfo()
Julia Version 1.11.3
Commit d63adeda50d (2025-01-21 19:42 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 12 × 13th Gen Intel(R) Core(TM) i7-1360P
  WORD_SIZE: 64
  LLVM: libLLVM-16.0.6 (ORCJIT, goldmont)
Threads: 12 default, 0 interactive, 6 GC (on 12 virtual cores)
Environment:
  JULIA_EDITOR = code
  JULIA_NUM_THREADS = 12
  JULIA_PKG_SERVER = https://juliahub.com/

• Output from using Pkg; Pkg.status("ITensors"):

julia> using Pkg; Pkg.status("ITensors")
Status `~/.julia/environments/v1.11/Project.toml`
  [9136182c] ITensors v0.8.0

[mtfishman]

Interesting, thanks for the report, we'll look into it.

[mtfishman]

Should be fixed now if you update to the latest version of NDTensors.jl.

[AFeuerpfeil]

Thanks, that fixed the problem. The only inconsistency I still observed is in the norm function for the same example as above:

julia> @show norm(T)
1.4142135623730951
julia> @show norm(denseblocks(T))
1.0

The norm for denseblocks(T) is correct, whilst norm(T) gives the result corresponding to a tensor with all ones in the diagonal.

[mtfishman]

Ah that's too bad, I'll look into that one too...