From electricity generation with photovoltaics, through efficient synthesis with photocatalysis, to targeted therapeutics with light-activated drugs, photochemistry is at the heart of many of the technological challenges that society needs to overcome in the 21st century. Nonadiabatic molecular dynamics (NAMD) simulations have emerged as a powerful set of tools to provide atomistic details of ultrafast photochemical reactions that are unavailable from experiment alone. However, many important systems remain out of reach for current NAMD approaches because of deficiencies with existing excited-state electronic structure methods, and the high computational cost of NAMD simulations. I will discuss our efforts to address both problems.

As a paradigmatic problem, I will introduce our submission to the recent photochemistry prediction challenge of cyclobutanone, which showcases the promises and pitfalls of first-principles photochemistry. First, I will introduce the cumulative surface hopping method developed in my group, an advanced sampling method that dramatically reduces the number of independent trajectories that need to be sampled in surface hopping simulations. Second, I will show how we are exploiting synergies between ab initio methods and semiempirical models to design highly accurate semiempirical TDDFT models that are 200-300 times faster than TDDFT, and improved iterative algorithms that speed up ab initio TDDFT by about a factor of 3. These results are enabled by a new semiempirical model, called TDDFT-ris, that is 2-3 orders of magnitude faster than TDDFT and is defined across the periodic table. Finally, I will introduce the Resonating Hartree-Fock (ResHF) method, an excited-state electronic method tailor-made for photochemistry simulation. Combining these approaches will enable the broadly applicable NAMD simulations needed to help design the future of photochemistry.