ExaFMM installation

ExaFMM is written in C++/CUDA OpenMP/MPI. Code is available from: https://bitbucket.org/exafmm/exafmm. A development version with higher performance but less features is available from: https://bitbucket.org/rioyokota/exafmm-dev.

The code requires CUDA. If you used Ubuntu 12.04, the steps are as follows.

1. Install the Nvidia driver.

2. Install requisite tools

sudo apt-get install freeglut3-dev build-essential libx11-dev libxmu-dev libxi-dev libgl1-mesa-glx libglu1-mesa libglu1-mesa-dev

3. Download the CUDA tools. 

4. Add requisite lines  to the end of ~/.bashrc (the nice example is here ) as follows

 export PATH=$PATH:/usr/local/cuda-5.0/bin
 export LD_LIBRARY_PATH=/usr/local/cuda-5.0/lib64:/lib
5. If a problem arises with libglut, find it with

    sudo find /usr -name libglut\*

    Now we can create a softlink to it

   sudo ln -s /usr/lib/x86_64-linux-gnu/libglut.so.3 /usr/lib /libglut.so

 
 

 

 

CAPLET - fast-field electrostatic solvers scaled up by parallelizing

Looking for a new generation of fast-field solvers capable of being scaled up by parallelizing, recently I find the interesting project CAPLET on the same MIT site we can find FASCTAP.

Project CAPLET is a capacitance extraction toolkit that extract capacitance at field-solver accuracy. This is the open-source code under GNU Lesser General Public license aimed for ultra-fast extraction of small-to-medium structures with 5% accuracy of reference solutions and efficiently parallelized.Thought the author claimes the simplified installation procedure, it demands some steps. We did it on Ubuntu 10.04. You can use any virtual machine (VirtualBox, etc.), install Ubuntu 10.04, upgrade the OS and install the following package:


apt-get update; sudo apt-get upgrade
   apt-get install g++
   apt-get install libcr-dev mpich2 mpich2-doc
   apt-get install libblas3gf libblas-doc libblas-dev
   apt-get install liblapack3gf liblapack-doc liblapack-dev
   apt-get install gfortran
   apt-get install openmpi-bin openmpi-doc libopenmpi-dev

To compile QT GUI you can download, for example, Qt Creator 2.4.1 and Qt 4.7.0.

The toolkit has three solvers:

 1. Caplet (based on instantiable basis functions), 

 2. FastCap (additional link to the solver is required), and

 3. Standard BEM solver (piecewise constant basis functions with collocation testing, obviously without any acceleration).

We compared the standard BEM solver verus FASTCAP with -d0 option on the proven configuration (two parallel 100x2000x200 plates). The results are not the same :-0.

2013: Still with FASTCAP. Towards ExaFMM?

Electrostatics. Looking for the parallel fast field solver.

I successfully used the FASTCAP fast field solver long years ago for capacitance calculation and electrostatic charge density and forces derivation. Incorporating this solver into a finite element code, solution of multiphysics coupled problems can be obtained quickly. With staggered scheme of course.
 

Recently coming back  to the electrostatic problem calculation, I tried to find a new generation of the fast field codes. My feeling that here is still no ready open source code except the premium FASTCAP.


The FASTCAP itself is acceptable from the accuracy standpoint, particularly for my problems in the hand. I have compared the BEM results with FEM reference solutions  and found it quite acceptable. Also for the science community it is a common practice to compare new and more efficient solvers versus FASTCAP. The problem is the speed of runs. The code was born in 90's, in the era of single-core computations. Nowadays impressed by "Biomolecular electrostatics using a fast multipole BEM on up to 512 gpus and a billion unknowns" I think to write a custom electrostatic code with ExaFMM library. Obviously it would take a long time and  perhaps I'll go flop. As a result the simple, open source,  parallel performance, electrostatic solver is still looking for.