4.4 The configuration file

In this section we will explain each of the parameters of the configuration file in more detail. Note that Nero2d is distributed with two configuration files, one general purpose file located in the /share/examples directory and another in the /share/benchmark example, which is the config file used for performing our parallel benchmarks.

The configuration files need to be edited by hand. We give an overview of the parameters

• OUTPUT_PROCESS Process that displays output to the screen. Value in [0 ... number of processes -1]. Default value is 0.
• QORDER The number of Gauss Quadrature Points used for numerical integration along a segment. Note that not every order is supported. Do not change unless you know what you are doing. Default value is 10.
• SEGMENT_LENGTH Maximum length of a segment in terms of wavelength. Default value is 0.1.
• OVERSHOOT Percentage that a segment can protrude from a box. This phenomena is taken into account when calculation the number of sampling points in the FMM factorization, and when calculation the effective X_DIST value at each level. Default value is 1.4.
• MINIMAL_BOX_SIZE Size of a box in terms of wavelength at the lowest level of the MLFMA tree. Default value is 0.25. Any value between 0.15 and 0.7 is probably acceptable for you.
• X_DIST Value in terms of boxsize. Boxes that are separated x_dist*boxSize at a certain level can interact using the high-frequency MLFMA. Default value is 1.9. Make this value very large if you want to solve your system in a classical manner.
• FMM_PRECISION Relative precision of the FMM factorization in the MLFMA algorithm. Default value is 1e-5 (five digits of accuracy). Note that a high precision is necessary in large 2D simulations.
• MAX_FMM_LEVEL Maximum height of an FMM tree. Default value is 1000 (in practice this value means: infinity).
• AGGREGATION_MATRICES Can be either FULL, COMPACT or COMPRESSED, which indicates the degree of symmetry that is to be used. Default value is COMPRESSED.
• INTERPOLATION_METHOD Can be either FFT, BLIF or MIXED. Default value is MIXED.
• MINIMAL_BLIF_LOCALITY Indicates the desired ratio of local points to boundary points in k-space partioning. Default value is 16.0.
• ITERATIVE_METHOD Your iterative solver of choice: can be TFQMR, BICGSTAB or RGMRES. Default of TFQMR.
• ITERATIVE_PRECISION Stopping criterion for the iterative solver. If the relative residual drops below this value: stop iterating. Default value is 1e-3.
• MAX_ITERATIONS Maximum number of iterations in the iterative solver. Once this number is reached, the solver is stopped, even though the requested precision has not yet been obtained. Default value is 50000.
• TRUENORMPARAMETER Only meaningful when the TFQMR solver is selected. This parameter determines how often the true residual is calculated. A value of 1 means: every iteration, a value of 2 means: every two iterations, etc. Default is 5.
• RESTARTPARAMETER Only meaningful when the R-GMRES solver is selected. This is the restart parameter in R-GMRES. Default is 5.
• BLOCKJACOBI_LEVEL Size of preconditioner. A negative value means: use no preconditioner. A value is 0 means, use blocks equal to the lowest-level box size. A higher value often means faster convergence, however at a cost of a longer setup stage. The default value is 3. You may want to raise this number for larger or more resonant simulations.
• SPACE_FILLING_CURVE Choice of space filling curve. Valid options are HILBERT or MORTON. Default value is HILBERT.
• PACKAGE_SIZE Size of a package in micro-seconds. Raising this value will reduce overhead, however, the asynchronous progress at non-dedicated parallel systems may suffer. Default is 10000.
• BENCHMARK_MODE If this value is equal to 1, all output is suppressed and the iterations are timed. Upon completion, a benchmark.dat file containing these times is written to disc. You may want to limit the number of iterations (to e.g. 30) when performing these benchmarks. Default is 0.
• MAX_OUTPUT_SIZE This value determines the maximum number of elements that are written to disc. Default value is 100000.
• WRITE_Z_MATRIX If this value is set to non-zero, the interaction matrix is written to disc, if its size is smaller than MAX_OUTPUT_SIZE. Default value is 0.
• WRITE_B_VECTOR If this value is set to non-zero, the right hand side vector is written to disc, if its size is smaller than MAX_OUTPUT_SIZE. Default value is 0.
• WRITE_X_VECTOR If this value is set to non-zero, the solution vector is written to disc, if its size is smaller than MAX_OUTPUT_SIZE. Default value is 0.