6.3.1. ChASE Object and Configuration¶
6.3.2. Executables¶
This examples provides multiple implementation of ChASE targeting different computing architectures with different data distribution scheme:
2_input_output_seq: sequential ChASE without MPI support, and Block Distribution.2_input_output: ChASE for distributed memory systems with MPI support (pure CPUs), and Block Distribution.2_input_output_block_cyclic: ChASE for distributed memory systems with MPI support (pure CPUs), and Block-Cyclic Distribution.2_input_output_mgpu: ChASE for distributed memory systems with MPI support (with GPUs), and Block Distribution.2_input_output_mgpu_block_cyclic: ChASE for distributed memory systems with MPI support (with GPUs), and Block-Cyclic Distribution.
This example uses Boost for parsing the parameters, thus the required parameters and configuration can be gotten by the help flag:
./2_input_output/2_input_output -h
6.3.3. Solving single problem¶
Here we utilize 2_input_output as an example to illustrate the way to use ChASE to solve single eigenproblem with loading external matrix.
The execution of this example through the command line is:
mpirun -np ${NPROCS} ./2_input_output/2_input_output --path_in=${MATRIX_BINARY_FILE} --n=${RANK_OF_MATRIX} --nev=${NB_of_WANTED_EIGENPAIRS} --nex=${EXTERNAL_SEARCHNING_SPACE} --mode=R
6.3.4. Solving a sequence of problems¶
Here we also utilize 2_input_output as an example to illustrate the way to use ChASE to solve a sequence of eigenproblems with loading external matrix.
The execution of this example through the command line is:
mpirun -np ${NPROCS} ./2_input_output/2_input_output --path_in=${DIRECTORY_STORE_MATRICES} --n=${RANK_OF_MATRIX} --nev=${NB_of_WANTED_EIGENPAIRS} --nex=${EXTERNAL_SEARCHNING_SPACE} --legacy=true --mode=R --bgn=2 --end=10 --sequence=true
In this example, for each physical system, a number (N) of matrices are available, which should be solved in sequence.
All the matrices are named gmat/ /1/ ell, with ell varying from 1 to N. The flag --legacy=true enables
searching the matrices conforming this naming policies in the directory ${DIRECTORY_STORE_MATRICES}.
In the execution example above, the first matrix to be solved is gmat/ /1/ 2, the the last matrix to be solved is gmat/ /1/ 10.
6.3.5. Parser of command-line arguments¶
Parameter (default value) |
Description |
|---|---|
-h [ –help ] |
show this message |
–n arg |
Size of the Input Matrix |
–double arg (=1) |
Is matrix complex double valued, false indicates the single type |
–complex arg (=1) |
Matrix is complex valued |
–nev arg |
Wanted Number of Eigenpairs |
–nex arg (=25) |
Extra Search Dimensions |
–deg arg (=20) |
Initial filtering degree |
–bgn arg (=2) |
Start ell |
–end arg (=2) |
End ell |
–tol arg (=1e-10) |
Tolerance for Eigenpair convergence |
–path_in arg |
Path to the input matrix/matrices |
–mode arg (=A) |
valid values are R(andom) or A(pproximate) |
–opt arg (=S) |
Optimi(S)e degree, or do (N)ot optimise |
–path_eigp arg |
Path to approximate solutions, only required when mode is Approximate, otherwise not used |
–sequence arg (=0) |
Treat as sequence of Problems. Previous ChASE solution is used,when available |
–mbsize arg (=400) |
block size for the row, it only matters for Block-Cyclic Distribution. |
–nbsize arg (=400) |
block size for the column, it only matters for Block-Cyclic Distribution. |
–dim0 arg (=0) |
row number of MPI proc grid, it only matters for Block-Cyclic Distribution. |
–dim1 arg (=0) |
column number of MPI proc grid, it only matters for Block-Cyclic Distribution. |
–irsrc arg (=0) |
The process row over which the first row of matrix is distributed. It only matters for Block-Cyclic Distribution. |
–icsrc arg (=0) |
The process column over which the first column of the array A isdistributed. It only matters for Block-Cyclic Distribution. |
–major arg (=C) |
Major of MPI proc grid, valid values are R(ow) or C(olumn). It only matters for Block-Cyclic Distribution. |
–legacy arg (=0) |
Use legacy naming scheme? |
Note
We have generated a few number of matrices defining (sequences of) eigenproblems from multiple material science simulation codes, if you want to test with these matrices, please feel free to contact us.
Note
For the fine tuning of more parameters in ChASE, please visit Configuration, in which we provide a class to set up all the parameters of eigensolvers. For the suggestion of selecting values of parameters, please visit Parameters and Configurations.