SWASH parameter study with Dakota

A vector parameter study of the SWASH wave-flow model driven by the Dakota iterative systems analysis toolkit.

Description

This study is broken into two stages.

In the first stage, Dakota, through the dakota_run_driver.py script, creates a series of independent PBS submissions, one for each iteration of the parameter study (currently 7), each using the submission script run_swash.sh. The submission script uses mpiexec to call SWASH in parallel using 8 processors on one compute node. Output from each run is collected and stored in PBS_O_WORKDIR in a directory run.N, where N = 1, 2, ..., 7.

In the second stage, Dakota analyses the results of each iteration with the dakota_analysis_driver.py script and creates the tabular output file dakota.dat, which summarizes the results of the parameter study.

Setup

On beach, add Dakota paths with:

export DAKOTA_DIR=/usr/local/dakota
PATH=$DAKOTA_DIR/bin:$DAKOTA_DIR/test:$PATH
export LD_LIBRARY_PATH=$DAKOTA_DIR/bin:$DAKOTA_DIR/lib:$LD_LIBRARY_PATH

Also, I recommend using the Anaconda Python distribution instead of the default Python:

PATH=/usr/local/anaconda/bin:$PATH

Execution

Run the first stage of the study with:

$ dakota -i dakota_run.in -o dakota_run.out &> run.log

After the first stage completes, run the second stage with:

$ dakota -i dakota_analysis.in -o dakota_analysis.out &> analysis.log

Results

View the results of the study in dakota.dat:

$ cat dakota.dat
%eval_id interface      BOT-sand Ufric_x_002800_000-mean Ufric_x_002800_000-stdev
1         0              2     0.00145875      0.0223389
2         0            2.5     0.00474198      0.0264774
3         0              3   -0.000528284      0.0221227
4         0            3.5     -0.0115249      0.0196799
5         0              4     -0.0152711      0.0199635
6         0            4.5     -0.0163003      0.0195528
7         0              5      -0.017241      0.0196005