T3NS: An implementation of the Three-Legged Tree Tensor Network algorithm

T3NS: an implementation of the Three-Legged Tree Tensor Network algorithm Copyright (C) 2018-2019 Klaas Gunst Klaas.Gunst@UGent.be

This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, version 3.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with this program. If not, see https://www.gnu.org/licenses/.

Installation

T3NS can be built with CMake and depends on BLAS and LAPACK.

It is parallelized for shared memory architectures with the Open Multi-Processing (OpenMP) API.

In your terminal, do:

> cd /sourcefolder
> git clone 'https://github.com/klgunst/T3NS.git'
> cd T3NS
> mkdir build
> cd build
> cmake ..
> make

To build with MKL and icc:

> CC=$(which icc) cmake -DMKL=ON ..

To install:

> sudo make install

Testing the build can be done by:

> make test

The number of threads used by openMP can be specified by setting the OMP_NUM_THREADS variable. e.g.:

> export OMP_NUM_THREADS=4

To see the help for T3NS:

> T3NS --help

Running a calculation

Examples for running calculations are given in the examples sub directory. For running a T3NS calculation, one needs to specify both an input file and a network file.

Possible options for the input file can be found through T3NS --help. Mandatory options are:

• networkfile : The path to the defined network file for the tensor network
• symm : The symmetries that should be used.
• ts : The irreps of the targeted ground state. For each symmetry in symm you need one irrep.
• interaction : The type of interaction. Possibilities are:
  • If a path to a .FCIDUMP file, it will optimize this Hamiltonian.
  • If a path to a .FCIDUMP file preceded by DOCI, it will optimize within the seniority zero subspace.
  • For a nearest neighbor Hubbard calculations (nearest neighbor according to the network geometry) use the following format: NN_HUBBARD (t = ..., U = ...).

A selection of the optional options are:

• D : The maximal virtual bond dimension
• SITE_SIZE : The number of sites to optimize at once (1, 2, 3 or 4).
• SWEEPS : The maximal number of sweeps to be executed.
• E_CONV : If this energy difference between sweeps has been reached, the current optimization regime is stopped.

The network file is formatted as follows:

NR_SITES = number of branching and physical tensors
NR_PHYS_SITES = number of physical tensors
NR_BONDS = number of virtual bonds in the network. At the border of the
network there is an extra bond which connects the bordering tensor to
nothing.
&END
List of sites. 
'*' depict branching sites and numbers represent physical sites where the
number corresponds with the orbital in the FCIDUMP (counting starts from 0).
&END
List of bonds.
The bonds are specified by giving two tensors which it should connect. The
number 'n' corresponds with the n'th tensor in the previously specified list
of tensors. 
To specify the bond connecting the n'th tensor with nothing (and thus
specify that this tensor is a border), use:
-1 n
Every bond starts at a new line and the number of bonds should correspond
with the previously defined amount.

Once defined the files, a T3NS calculation can be executed by:

> T3NS inputfile

Intermediate results are stored in the current working directory in hdf5 format and can be used to continue a calculation through:

> T3NS --continue=T3NSwav.h5 inputfile

In this case, only the optimization scheme defined in the input file will be used for the continued calculation. Other specified options will be ignored and instead read from the hdf5 file.

Provided scripts

In the scripts folder, there are some python scripts provided for the automatic generation of network files and the optimization of the ordering.

One can generate a T3NS or DMRG network file by executing

> ./makenetwork.py nr_sites
> ./makenetwork.py nr_sites DMRG

This will print the network file to stdout. Afterwards one can optimize the orbital ordering by providing a FCIDUMP and the previously generated network file.

> ./optimizenetwork.py networkfile fcidump

Both scripts have a rudimentary help accessed by the -h or --help arguments.