'Letting physicists be physicists,' and other goals of scalable solver research

摘要

Computational scientists and engineers possess tremendous opportunities to extend the scope and fidelity of simulation and allow it to interact as an equal partner with theory and experiment. However, no single person can any longer be the master all aspects of the vertically integrated scientific instrument of a validated, verified, algorithmically optimized, performance-optimized simulation code. Modern scientific software engineering conveys the fruits of expertise at lower levels of this vertically integrated software structure to physicists working at higher levels. There is no more fitting a way to end this chapter than with the statement of the 2005 Abel Prize winner, Professor Peter Lax: "Quality software renormalizes the difficulty of doing computation."