Physics Department Seminar University of Alaska Fairbanks

J O U R N A L    C L U B

Application of a New Technique - the "Parareal Algorithm" - to Parallelize a Primitive Turbulence Code in Time
Debasmita Samaddar
Physics Dept., UAF

The Parareal Algorithm[1] is a new computational technique that parallelizes the time domain and can therefore significantly reduce the computational wall time. Parallelizing the space domain to computationally solve partial differential equations has been the most common approach so far. The Parareal Algorithm has been reported to have been successfully applied to many systems, but not yet to a turbulent system (to the best of our knowledge). If efficiently implimented, this algorithm will allow study of the turbulent transport dynamics on transport time scales - something that has heretofore been very difficult.

In this talk, we will present the results of applying the Parareal Algorithm to simulations of drift-wave turbulence[2] in slab geometry. These will be extended to include cases in which the relative dominance of the polarization and ExB nonlinearities are tuned artificially and the transport exponents may be explored.

The talk will comprise of an overview of the Parareal Algorithm and a description of its application to our problem. I will also present the results of a modification to the algorithm. References:

[1] Martin J. Gander et al, Siam J. Sci. Comput. Vol.29, No.2, pp.556-578
[2] D.E. Newman et al, Phys. Fluids B 5, 1140 (1993)

Friday, 9 October 2009
Globe Room, Elvey Building
3:45 PM