Problem with exp and sparse matrices

DMRG and Numerical Methods
1

Hi,

For a research project of spinful fermions I need to implement TEBD, I wanted to do it with symmetries because my Hamiltonian conserves the total number of particles (not by spin but it respects the total number).

However, I got an error when I try to create the gates for the TEBD. Consider the following code and a very simple gate:

using ITensors
using ITensorMPS

s1,s2,s3,s4 = siteinds("Electron", 4, conserve_qns = true);
Gate = op("Cdagdn * Cup", s1)*op("Cdagdn * Cup", s2)*op("Id", s3)*op("Id", s4)

Gate is a NDTensors.BlockSparse{Float64, Vector{Float64}, 8}. Now, when I try to get the exponential (that I need to create the array of gates for TEBD) using

exp(-0.5im*Gate)

I got the following error:

exp currently supports only block-diagonal matrices

Stacktrace:
 [1] error(s::String)
   @ Base .\error.jl:35
 [2] exp(T::NDTensors.BlockSparseTensor{ComplexF64, 2, Tuple{…}, NDTensors.BlockSparse{…}})
   @ NDTensors C:\Users\JOP204\.julia\packages\NDTensors\wnM2t\src\blocksparse\linearalgebra.jl:413
 [3] exp(A::ITensor, Linds::NTuple{4, Index{Vector{Pair{…}}}}, Rinds::NTuple{4, Index{Vector{Pair{…}}}}; ishermitian::Bool)
   @ ITensors C:\Users\JOP204\.julia\packages\ITensors\SOADD\src\tensor_operations\matrix_algebra.jl:94
 [4] exp(A::ITensor, Linds::NTuple{4, Index{Vector{Pair{QN, Int64}}}}, Rinds::NTuple{4, Index{Vector{Pair{QN, Int64}}}})
   @ ITensors C:\Users\JOP204\.julia\packages\ITensors\SOADD\src\tensor_operations\matrix_algebra.jl:39
 [5] exp(A::ITensor; kwargs::@Kwargs{})
   @ ITensors C:\Users\JOP204\.julia\packages\ITensors\SOADD\src\tensor_operations\matrix_algebra.jl:110
 [6] exp(A::ITensor)
   @ ITensors C:\Users\JOP204\.julia\packages\ITensors\SOADD\src\tensor_operations\matrix_algebra.jl:107
 [7] top-level scope
   @ In[446]:1
Some type information was truncated. Use `show(err)` to see complete types.

It is not very clear to me what is the problem with this operator.

2

Change

siteinds("Electron", 4, conserve_qns = true);

to

siteinds("Electron", 4, conserve_nf = true);

With conserve_qns=true, the electron site conserves both number and spin. See SiteTypes Included with ITensor · ITensorMPS.jl

Also, as demonstrated by the following post you can easily run into issues with constructing fermionic operators directly from op. Problem with TEBD on Fermi-Hubbard model with Electron indices