Computers model frozen fuel in blazing fusion reactor

(page 2 of 3)

The simulation must accurately depict what happens to the tiny pellet as it’s swallowed by a plasma cloud at least 1,000 times its size – and do it so the program runs in days, rather than years, on some of the world’s fastest computers.

Molding a mesh

Like many computer simulations, Samtaney’s code to model fuel pellet injection in a fusion reactor uses data meshes.  Meshes distribute data points throughout the domain being modeled, then calculate what’s happening at each point.  Taken together, those points simulate a continuous process.

But as the size of the reactor being modeled grows, a mesh that distributes points uniformly becomes too large to compute efficiently.  If the pellet has a 1-millimeter radius and the tokamak has a 1-meter radius at its narrowest point, “then that’s three orders of magnitude in linear dimensions” that must be modeled, Samtaney says.  That and other factors make computing using a uniform grid hugely demanding.

Samtaney turned to Adaptive Mesh Refinement (AMR) to surmount the challenge.  AMR conserves computer resources by focusing data points in areas of the most interest – around the pellet – and putting fewer points elsewhere.

Samtaney worked with the Advanced Numerical Algorithms Group at the Department of Energy’s Lawrence Berkeley National Laboratory (LBNL) to add AMR to his physics codes.  DOE’s Scientific Discovery through Advanced Computing (SciDAC) program supported the work.  Read more.

The method allowed the refueling simulation to run hundreds of times faster than it would have with a uniform mesh.  It found that where the pellet ends up largely depends on where it starts.

« Previous       1   |   2   |   3   |   Print       Next »