Argonne National Laboratory
Story by Jacob Berkowitz
Along with being home to Cornell University, Ithaca, New York is famous for the Moosewood Restaurant, America’s best-known vegetarian eatery.
The Moosewood’s fame is largely based on its spectacular and creative cookbooks. The cookbooks are renowned for complex recipes that both taste and look great. The problem is that for busy cooks with 20 minutes to whip up something for a 5:30 family dinner, the recipes are impractical — fun to read, but often impossible to cook.
It’s a conundrum Stefan Wild understands well — in a computational science sense — and he’s rolling up his sleeves in the computational kitchen to help solve the problem. During a summer practicum at Argonne National Laboratory and in his doctoral research at Cornell, Wild, a Department of Energy Computational Science Graduate Fellow (DOE CSGF), is working to streamline cordon-bleu-complex computer models and make them stove-top fast. The results will have green impacts of another kind: environmental engineers use the computer models he’s improving to find the best ways to stop the spread of pollution and to clean polluted groundwater.
Big, Not Fast
Quadratic approximation of a function using only 4 function values.
Click image for larger version and more information
Even with advances in computer hardware and modeling software, computational scientists often face a central hurdle: when the model bites off more than computers can easily chew. Many models are so detailed, contain so much information, and are so complex that they become technically unwieldy. The mathematical recipes, or algorithms, that make up the model are beautiful, but take too long to run to get timely results within budget.
“We call these computationally expensive models,” Wild says. “They often give researchers headaches because the model evaluation expense makes many of their favorite analysis tools ineffective in practice.”
Enter the world of model optimization. Model optimizers are applied mathematicians, like Wild, who are passionate about getting the most from cumbersome computational models.