Installation and testing is detailed in the INSTALL.md file but at least Python 3 and an implementation of Message Passing Interface is required along with mpi4py and the NetworkX python package. Separate programs are used for evaluating integrals numerically and are provided in the integrands directory. These can have their own prerequisites as detailed in INSTALL.
To get a list of mandatory and optional arguments supported by the program run
hdintegrator.py with --help argument:
hdintegrator.py --help if you installed it through
In general the program must be run with at least two MPI processes and must include a serial program to use as an integrand as well as the number of dimensions to integrate in:
mpiexec -n 5 ./hdintegrator.py --integrand integrands/N-sphere.py --dimensions 2
The result consists of one line with the integral's value, absolute error and fraction of volume relative to total volume in which the integrand failed to return a value:
0.523598776383549 1.453150932917424e-08 0.0
Input and output formats
To define your own integral you must write a program that will be called by hdintegrator.py for evaluating that intergral within a given volume and give that program as an argument to hdintegrator.py. For an example see e.g. the integrand integrands/N-sphere.py which evaluates the integral for an N-dimensional sphere. Communication between hdintegrator.py and integrands is handled via standard input and output in ASCII format. Each line given to the integrand by hdintegrator.py consists of floating point numbers separated by spaces:
C a0 a1 b0 b1 c0 c1 ...
where C is the number of samples to use for evaluating the integral and a0, a1, b0, etc. represent the minimum and maximum extent of the integration volume respectively. Note that the number of dimensions might change from one line to another, although that is not the case currently. Each line received from the integrand by hdintegrator.py must consist of three floating point numbers separated by spaces in ASCII format:
V E S
where V is the value of the integral, E is an estimate of the absolute integration error and S is the suggested dimension starting from 0 in which to split the integration volume in order to minimize subsequent integration errors.
To seek support or report an issue in HDIntegrator please create a new issue at https://github.com/iljah/hdintegrator/issues
To contribute to HDIntegrator please create a new pull request at https://github.com/iljah/hdintegrator/pulls