As a numerical scientist, all that matters is to see your codes used to tackle real-world problems.
Rob Parrish’s career is built around the idea that solving the quantum mechanical Schrödinger equation faster can help to computationally design better molecules and materials. In pursuing that, He’s worked with really talented groups using CPUs, GPUs, and even QPUs to bring the codes to the next level.
My career is based on quickly reading the first few chapters of the introductory quantum chemistry textbook “Szabo and Ostlund” and then spending nearly two decades trying to implement these equations correctly. My first research project as a 20-year-old undergrad was to implement a density-fitted Hartree-Fock code in PSI4. Earlier this year, my teammates and I at NVIDIA released the “cuEST” library for fast quantum chemistry on GPUs … based on a density-fitted Hartree-Fock code. So part of this talk will focus on the importance of hammering over and over on things that we as a field already know very well.
We’ll talk a little about tech. We will try to develop a rough idea how fast computers really are, and how quickly they improve with time. We will talk a bit about what computers are good at (hint: matrix multiplication) and what they are not good at (hint: almost anything else unless you try really hard). A huge key to exploiting the amazing power of modern silicon is the ability to transform differential equations into linear algebra problems, which is the whole field of numerical methods. So we’ll work through some quick illustrations on that.
We’ll talk a little about careers. One thing to emphasize here is that you can do what you want to do almost anywhere… if you try hard enough. A small piece of evidence for this is that my friends and I have somehow managed to pay the rent while engineering quantum chemistry codes at universities, national labs, startups, and now at Team Green - and it’s always been basically the same job.
Lastly, we will talk about the importance of friendship in science. Two of my mentors from the 20-year-old PSI4 days are my colleagues on the cuEST team at NVIDIA. We’ve been joined by many new and wonderful teammates, and every day is still that same long campaign to make the Schrodinger equation more useful.
*Recommended Reading (no quiz):
How to learn quantum chemistry: Read and code for yourself the first three chapters of Szabo and Ostlund: Modern quantum chemistry: introduction to advanced electronic structure theory.
How to learn numerical methods: Apply for the DOE CSGF fellowship. They will make you take 6x extra grad classes you didn’t want to take. If you survive you’ll be a numerical scientist. Or just do it all in one course: P. Moin, Fundamentals of engineering numerical analysis.
*How to learn CUDA:() Read and code for yourself all of Mark Harris’s blog posts, starting with this one.