Massively Parallel Shaking and Baking

The aim of Shake-and-Bake is to extend the direct method to larger molecules. It relies on a reformulated mathematical approach, devised by Hauptman in 1988, that involves minimizing the value of a function involving thousands of phases. To make the scheme workable, Charles Weeks of the Medical Foundation of Buffalo automated the solution process and computer scientist Russ Miller of SUNY Buffalo headed the effort to develop algorithms for massively parallel computers. "We need an enormous amount of computational power," says Miller, "and the only way to get it is through massively parallel computing."

The process begins with a trial molecular structure. "You essentially toss the atoms into a shoebox," says Miller, "with enough knowledge to make sure the structure makes chemical sense." The computer calculates a set of phases corresponding to these atomic positions and then randomly perturbs them, the shake part of the process, to arrive at a lower value in accord with Hauptman's formula. Then a new set of atomic positions is calculated, and the method continues, back and forth between atomic positions and phases as many as 200 times for a single trial structure. When enough trial structures have been shaked and baked, the right answer will emerge as a significantly low value for Hauptman's formula, at least that is the theory, and so far the researchers are, in Hauptman's words, "very much encouraged."