Frequently Asked Questions

Answers

Why don’t you provide binaries for distribution xyz?

We aim to directly support the most widely used distributions and are continuously expanding our list of supported environments. Setting up a new environments means building a docker image for our automatic binary creation scripts. So if it is technically feasible and of interest we will add a new distribution.

Does BEM++ support Python 3?

BEM++ is continuously being tested on Python 3.4 and later versions. Full support for Python 2.7 and Python 3.4+ is of significant importance for us.

Do you have Docker images?

Internally we use Docker images as build environment. If there is strong interest we can also release full Docker images.

Why are there no binaries for Mac?

We provide Homebrew build scripts that are just as easy as installing direct binaries.

Do you have a Ubuntu ppa?

A ppa for Ubuntu is in preparation.

What is the minimum gcc version for compilation?

Generally, we need a C++11 enabled compiler. We have tested gcc 4.8+. gcc 4.7 may work but is not tested.

Compilation stops with an internal compiler error

The compilation can require a lot of memory due to the template structures in the library. An internal compiler error is often a sign that the compiler has run out of memory. In this case try to reduce the number of parallel build processes in make.

Does BEM++ provide logging facilities?

BEM++ integrates with the standard Python logging system. To enable console logging use the command bempp.api.enable_console_logging. For file logging use the command bempp.api.enable_file_logging. Console logging can also be enabled with by setting the environment variable BEMPP_CONSOLE_LOGGING to 1, e.g.:

export BEMPP_CONSOLE_LOGGING=1

Can I use the C++ core directly?

BEM++ until version 2 was essentially a C++ library with a Python interface bolted on. This has changed. We have rewritten all the top layers of BEM++ in Python directly. It is now essentially a Python library with a fast C++ core for low-level computations. Nevertheless, the C++ core can still be used independently of Python. But most higher level functions will not be available.

How do I speed up BEM computations?

The main factors influencing the assembly speed are the order of the quadrature rule and the accuracy parameter eps in the H-Matrix assembly. We provide relatively safe parameter values for many applications. Nevertheless, they are not optimal in every situation and it is worth testing various different parameter values to speed up the computation.

What parallelization does BEM++ support?

BEM++ uses Intel Threading Building Blocks for multithreading. We are currently working on integration with the IPython Parallel framework to allow distributed assembly on multiple cluster nodes.

Can I change the number of threads used by BEM++?

To control the number of threads use the environment variable BEMPP_NUM_THREADS, e.g. to set the maximum number of threads to 2 use export BEMPP_NUM_THREADS=2.

Why does the adaptive refinement not work?

In the default configuration the adaptive refinement of individual elements leaves hanging nodes in the mesh. The reason is that the grid manager used in BEM++ at the moment only supports global refinement. BEM++ can optionally be compiled with Dune-Alugrid as grid manager. This is not enabled by default for license reasons, but allows local refinement of meshes. To recompile BEM++ with Alugrid enabled use the CMake option WITH_ALUGRID=ON.

Does BEM++ have a full H-matrix algebra?

Until BEM++ 2.0 BEM++ used the AHMED Library to provide support for a full H-Matrix algebra. For license reason we discontinued the use of AHMED and developed our own H-Matrix compression. Implementation of a full H-Matrix algebra is in planning though for a later version of BEM++.

What happened to the H-LU decomposition?

H-LU was part of AHMED, which has been removed from the library. A self developed H-LU code is in planning for a later version of BEM++.

What preconditioners are offered for Maxwell?

Currently we have no good Maxwell preconditioning built in. It is therefore recommended to use Maxwell in dense mode and solve via LU. An efficient Calderon based preconditioner for Maxwell is almost finished, but did not make it into the library in time for version 3.0. It will however be integrated as soon as possible.

Release 3.0 took a long time. What is the future release schedule?

Release 3.0 is a major rewrite of the whole library after gaining some experience with version 1 and 2 of BEM++ and seeing what worked and what not. We now have a fairly complete development infrastructure in place and much of the code is stabilized. So we expect to release more frequently updates with functional enhancements.

What features are planned for the near future?

The focus of the BEM++ development is Maxwell and high-frequency problems and we are planning to make several improvements in these areas. Distributed assembly on clusters is quite far progressed and is currently being tested on separate branches.

Will there be a native Windows version?

Most likely no. We may be able with a lot of fixes to compile BEM++ in a Cygwin environment. But none of the developers uses Windows. So it is of little priority to us. We provide virtual machine images that run quite efficiently also under Windows.

Are you planning Matlab or Julia interfaces?

Matlab has growing Python support. So via the Matlab/Python bridge we may be able to build a Matlab version. However, this is not of high priority for us. Julia is a very interesting project. But before we invest time in it we would like to see it stabilise and mature more.