Successful investigations of various properties of Earth materials will
result from hundreds or thousands of independent calculations on
physically distributed parallel platforms. These calculations are often
divided into sets, each set perhaps depending on the successful completion
of the previous one. Manual submission and monitoring of theses numerous
runs and pre- and post-run analysis present a major bottleneck in high-end
computational materials research. For example, the elasticity calculation
of MgSiO3-perovskite consumed in 2001/02 ~7 person-months of job
submissions and monitoring on a single IBM SP-2. Numerous other similarly
important calculations are necessary for proper interpretation of seismic
data and they should not be impaired by human availability. Besides,
researchers should be freed for other more complex decision making tasks.
A task automation system to enable submission of a multitude of runs is
indispensable. Besides, some simulations execute for days/weeks. Steering
mechanisms, such as pV3 and SCIRun are necessary to monitor the state of
runs while they are ongoing, and possibly change parameters without
restarting them.
We will develop an automated system to compute full elastic constant
tensors of Earth materials versus pressure and temperature. It will handle
several independent calculations of strained configurations with
accompanying phonon dispersions in each case. An orthorhombic crystal
requires at least 16 strains with the average number of ~6 wave-numbers
per strain for ~10 different pressures. Our task automation system must
handle ~1,000 independent calculations. Based on our gained experience, we
will automate the study of solid solutions as well. The number of
calculations necessary to parameterize the free energy approximated with
the cluster expansion method involves the systematic generation of
numerous ordered configurations determined by the crystal symmetry and
alloy composition. Configuration generation, job preparation, submission,
and monitoring must be automated for these calculations to become viable.
Virtually all Earth materials are solid solutions and most have more than
two-component. (Karki, Pierce, de Gironcoli)
