◆ STensor

‘Smart’ tensor class.

Main features compared to the plain Tensor:

  • Hides the tensor rank behind a std::variant
  • attempts to implement the graded algebra needed for fermionic tensor networks. The fermionic ordering of legs does not have to coincide with the computational ordering.
  • Stores a complete, named STensorImpl::SBasis for every leg. Information about zero blocks is not lost.
  • basic support for autodifferentiation (see STensorImpl::Autodiff)

Incoming indices are bras, outgoing indices are kets. Rank-2 tensors can always be contracted with a single leg of any other (even) tensor as long as their outgoing leg comes fermionically first and their incoming leg comes fermionically second.

Implemented in the STensorImpl namespace with all associated functions.
To enable compilation of this class, define the SYTEN_STENSOR macro. By default, STensor objects up to and including rank-5 can then be used. To change this maximal rank, define the new maximal rank via SYTEN_STENSOR_MAX_RANK and also define SYTEN_STENSOR_RANKS to a comma-separated list of the valid ranks, starting at 0 and going up to SYTEN_STENSOR_MAX_RANK.
In the pretty-printing of an STensor, rank and autodifferentiation ID are given in brackets directly following the word STensor, e.g. STensor[3;0] signifies a rank-3 STensor with autodiff ID 0 (disabled), STensor[0;5] signifies an STensor with rank 0 and autodiff ID 5. The address of the STensor in memory is given next as a hexadecimal number. Subsequently, for higher-rank tensors, the bases on the (arbitrarily numbered) legs and the fermionic order of those legs are given together with the nonzero blocks of the tensor. Rank-0 tensors are presented only by their numerical value, where the imaginary part is left off if it is non-zero.