Jaime Marian: "Physics-based modeling of materials and uncertainty quantification as drivers for exascale computing"

Jaime Marian - University of California, Los Angeles (UCLA)

The availability and/or prospects of peta and exascale computing has transformed modeling of materials behavior into a new paradigm where simulations must be mined to extract useful data that can be compared to experiments on a meaningful basis. Moreover, materials modeling must now be accompanied by a minimum uncertainty quantification exercise that sets reasonable limits on the validity of the simulation/experimental comparison. This is particularly true within the standard paradigm of ‘multi scale’ or ‘parameter-passing’ modeling approach, where a conscientious efforts is made to transfer what is perceived as the most useful information emanating from a particular temporal or spatial scale to another. In this presentation we will show several examples of advanced materials modeling cases where such transformation is being made and we will discuss future directions.

Vasily Bulatov
over 1 year agoNovember 23, 2017
Hello Jaime, great talk you gave!  

Want to point out that the PDES (timewarp) method is delivering for us despite indeed having to deal with the rough time horizons.  As an example, last year we run a timewarp KMC simulation of the Potts model grain coarsening in a model containing ~200 billion spins.  The simulation run to the very end (single grain left), was exact (provably equivalent to a serial KMC) and showed very good parallel efficiency on 512 thousand cores of Vulcan.  There are plausible arguments made by Oppelstrup, Jefferson and myself that PDES methodology may in fact be indefinitely scalable.  

Your comment