MLPF: Multi-Layer Potfit

This is an auxiliary program for the Heidelberg MCTDH Package. It converts a potential energy surface into MLPF format, which can be used by MCTDH as a multi-layer operator in ML-MCTDH runs. Compared to potentials fitted by regular Potfit, MLPF potentials are much more compact, so that ML-MCTDH will run much more efficiently, especially if high-accuracy fits are used. For more information and details, see my paper, also available on arXiv.

Support for MLPF in MCTDH exists currently in a development branch, and is planned to be released with version 8.6.

If you use MLPF for your research, you are requested to cite the following paper:

Frank Otto, "Multi-Layer Potfit: An Accurate Potential Representation for Efficient High-Dimensional Quantum Dynamics." J. Chem. Phys. 140, 014106, (2014)

Documentation

This is the README for MLPF. It and the rest of the documentation are stored in the doc/ subdirectory in Markdown format. For convenience, the documentation is also available in HTML format. Please point your web browser to the README.html inside doc/.

Prerequisites

If you are using MCTDH, chances are very high that all of the prerequisites are already installed. For reference, this is the software required to build MLPF:

• Compilers for C and Modern Fortran. Versions of GCC since 4.8 should work fine.
• An installation of BLAS (e.g. OpenBLAS) and LAPACK.
• GNU Make.
• Perl and Bash.

Compilation

Before first compiling, copy the file local.mk.sample into local.mk and edit it, following the comments inside. This is for specifying the compiler and its options, and how to link with BLAS/LAPACK on your system.

To compile, simply run make. To compile faster, use make -j N, where N is the number of CPU cores you want to use. This will create a number of executables in the subdirectory bin/, namely:

• mlpf -- the MLPF fitting program
• mlpf2npot -- a program to convert an MLPF potential into a Potfit potential
• vinfo -- a simple test/example program which prints out information about vpot or vpot2 files, which are used to store full-grid potentials.

Usage

mlpf

To run the MLPF fitting program, use

mlpf inputfile

where the syntax of the inputfile is described in the input file documentation. The generated output files are:

• mlpf.dat -- a binary file containing the MLPF-format potential, which can be read by MCTDH using the mlpf-statement in the operator file.
• log -- a log file, containing information about estimated errors and the number of single-particle potentials for each mode.
• tree.dot -- (optional) an input file for Graphviz's dot utility, to display the mode hierarchy.

mlpf2npot

To convert a potential in MLPF format into Natpot format, use

mlpf2npot mlpf.dat modc

where mlpf.dat is the output file of mlpf, and modc is an integer which specifies the number of the contracted mode. It generates a file called natpot which can be used by several utilities in the MCTDH suite.

vinfo

To get some information about a full-grid potential, stored in a vpot or vpot2 file, use

vinfo vpotfile

where vpotfile is the filename of the full-grid potential. It prints out some statistical information about the potential: maximum, minimum, mean, standard deviation, and the L2 norm.