1) PAPER TITLE Multiobjective Genetic Algorithms for Multiscaling Excited-State Direct Dynamics in Photochemistry 2) AUTHORS: Kumara Sastry Department of Industrial and Enterprise Systems Engineering 104 S. Mathews Ave 117 Transportation Building, MC-238 University of Illinois at Urbana-Champaign Urbana, IL 61801 (217) 333-2346 ksastry@uiuc.edu Duane D. Johnson 1304 W. Green St. 312E Materials Science and Engineering Building University of Illinois at Urbana-Champaign Urbana, IL 61801 (217) 265-0319 duanej@uiuc.edu Alexis L. Thompson Department of Chemistry 600 S. Mathews Ave A131 Chemical & Life Sciences Lab University of Illinois at Urbana-Champaign Urbana, IL 61801 (217) 244 7383 alexis@spawn.scs.uiuc.edu David E. Goldberg Department of Industrial and Enterprise Systems Engineering 104 S. Mathews Ave 117 Transportation Building, MC-238 University of Illinois at Urbana-Champaign Urbana, IL 61801 (217) 333-0897 deg@uiuc.edu Todd J. Martinez Department of Chemistry 600 S. Mathews Ave A131F Chemical & Life Sciences Lab University of Illinois at Urbana-Champaign Urbana, IL 61801 (217) 333-1449 tjm@spawn.scs.uiuc.edu Jeff Leiding Department of Chemistry 600 S. Mathews Ave A131F Chemical & Life Sciences Lab University of Illinois at Urbana-Champaign Urbana, IL 61801 (217) 333-1449 jeff@spawn.scs.uiuc.edu Jane Owens Department of Chemistry 600 S. Mathews Ave A131F Chemical & Life Sciences Lab University of Illinois at Urbana-Champaign Urbana, IL 61801 (217) 333-1449 jane@spawn.scs.uiuc.edu 3) CORRESPONDING AUTHOR: Kumara Sastry 4) ABSTRACT: Accurate simulation of photochemical reactions is critical in understanding most biological, chemical and technological processes. We use multiobjective genetic algorithms for multiscaling excited-state direct dynamics in photochemistry via rapid reparameterization of semiempirical methods. Using a very limited set of ab initio and experimental data, semiempirical parameters are reoptimized to provide globally accurate potential energy surfaces, thereby eliminating the need for full-fledged ab initio dynamics simulations, which are prohibitively expensive. Through reoptimization of the semiempirical methods, excited-state energetics are predicted accurately, while retaining accurate ground-state predictions. The GA-discovered parameters are significantly better---up to 230% lower error in the energy and 86.5% lower error in the energy-gradient---than the CURRENT BEST PUBLISHED results. Additionally, the parameter sets, verified with quantum dynamical calculations, show near-ideal behavior on critical and untested excited-state geometries. Finally, the GA-discovered parameters are transferable that enable direct dynamics simulations of photochemistry to multi-picosecond time scales, orders of magnitude than currently possible. 5) CRITERIA: (B) The result is equal to or better than a result that was accepted as a new scientific result at the time when it was published in a peer-reviewed scientific journal. (C) The result is equal to or better than a result that was placed into a database or archive of results maintained by an internationally recognized panel of scientific experts. (D) The result is publishable in its own right as a new scientific result 3/4 independent of the fact that the result was mechanically created. (E) The result is equal to or better than the most recent human-created solution to a long-standing problem for which there has been a succession of increasingly better human-created solutions. 6) STATEMENT: Photochemical reactions are fundamental in many biological (for example, photosynthesis and vision) and technological (for example, solar cells and LED displays) settings. These reactions and the associated dynamics are energetically subtle and require highly accurate descriptions of the relevant interatomic forces. Reliable predictions are costly even for small molecular reactions, but rapidly approach the impossible for reactions in complex environments, such as in solvents (e.g., water), in solid cages (e.g., zeolites), or with proteins (e.g., ion channels). Hence, having substantially faster semiempirical potentials that accurately reproduce higher-level quantum chemistry results would make it possible to address critical biological processes and technologically useful chemical reactions, or provide dramatic reduction in searching potentially technological useful light-activated reactions. Thus, developing and applying multiobjective evolutionary algorithm to provide efficient and highly competent, semiempirical methods that surpass all previous computer-aided and human-design attempts has great innovative potential, with an estimated reduction in computing time by a factor of 1000. Notably, semiempirical potentials have traditionally had the critical parameters hand designed and optimized so as to predict ground-state energies---not excited-state energies. Therefore, these carefully established parameter sets yield inaccurate potential energy surfaces, resulting in unphysical reaction dynamics. Additionally, previous attempts at optimizing parameters to yield globally-accurate, excited-state, potential energy surfaces have been but partially successful. In our work, we used a multiobjective evolutionary algorithm to reoptimize the parameter sets using a very limited set of ab initio and experimental data to yield globally accurate potential energy surfaces and excited-states, yielding accurate photochemical reaction dynamics. Our submission satisfies the following criteria for a human-competitive result: (B) The result is better than a result that was accepted as a new scientific result at the time when it was published in a peer-reviewed scientific journal AND (E) The result is equal to or better than the most recent human-created solution to a long-standing problem for which there has been a succession of increasingly better human-created solutions. Our multiobjective genetic algorithm (GA) discovered parameter sets that are 230% lower error in "energy", and 87% lower error in "energy gradient" over the CURRENT BEST PUBLISHED results. Moreover, unlike previous results, the GA results yields globally accurate potential energy surfaces, and near-ideal energies for untested, yet critical, excited-state configurations. The multiobjective GA optimization led to semiempirical potentials that performed exceptionally well for excited-state energetics, well beyond previous attempts, or expectation of human experts. Finally, the GA results demonstrate "transferability"---the property that allows parameters from one physical system to be used more generically for similar systems. For example, our GA parameter sets optimized for ethylene (C2H4) are applicable to benzene (C6H6) and vice versa. This critical "transferability" property enables accurate simulations of photochemistry in complex environments, such as proteins and condensed phases, without the need for complete reoptimization. Notably, transferable potentials for Molecular Dynamics simulations have been a "Holy Grail" for two decades. (C) The result is better than a result that was placed into a database or archive of results maintained by an internationally recognized panel of scientific experts Well established parameter sets in quantum chemistry databases (known by acronyms of MNDO, AM1, and PM3) give useful information on ground-state energies. However, they fall short of yielding globally accurate potential energy surfaces critical for accurate photochemical reaction simulation. For example, in ethylene, AM1 or PM3 parameter sets obtain the so-called pyramidalized structure as the excited-state minimum. Importantly, our multiobjective GA results obtain a twisted geometry to be the excited-state minimum, which is in agreement with ab initio and experimental results. Similarly, unlike the standard parameter sets, our GA results for benzene predict an observable excited-state lifetime of 100 fs, in agreement with experiment. Such results have never been obtained by any previous attempt at optimizing the semiempirical forms of MNDO, AM1, and PM3. (D) The result is publishable in its own right as a new scientific result 3/4 independent of the fact that the result was mechanically created. Results obtained via both hand-designed and reoptimized parameter sets have been previously published on numerous occasions. Since our results are significantly better than those previously published, and yields accurate quantum chemical properties, it should therefore be publishable in its own right. Indeed, we are submitting a paper detailing the chemistry results obtained through the GA-optimized parameter sets to a reputed chemistry journal. 7) CITATION: K. Sastry, D.D. Johnson, A. L. Thompson, D. E. Goldberg, T. J. Martinez, J. Leiding, J. Owens, "Multiobjective genetic algorithms for multiscaling excited state direct dynamics in photochemistry", Proceedings of the Genetic and Evolutionary Computation Conference 2006 (GECCO 2006), in press. 8) STATEMENT OF PRIZE DISTRIBUTION: Prize money, if any, is to be divided equally among the three contributing research groups of Johnson, Goldberg and Martinez. 9) STATEMENT OF COMPARISON TO OTHER "HUMAN COMPETITIVE" ENTRIES Because accurate simulation of photochemical reactions is critical in understanding most biological, chemical and technological processes, we have combined expertise in optimization, materials science and chemistry to address this outstanding problem. Our work uses a multiobjective genetic algorithm to bridge high-level quantum chemistry techniques and semiempirical methods to provide highly-accurate representation of complex molecular excited-state and ground-state behavior, along with a dramatic reduction in computational time. We also obtain fundamental insights into the physical basis for chemical reactions by use of the GA- discovered potentials that inherit the accuracy of the ab initio data. The GA-discovered potentials permit simulations to multi-picosecond time scales, which is large on reaction time scales and orders of magnitude larger than currently possible by ab initio methods, even for simple molecules. Finally, we have sufficient evidence to show that the GA- discovered parameter sets are transferable between similar atomic-based systems, which is a "Holy Grail" for materials and chemistry simulation. Most importantly, this multiobjective optimization approach may be applied generally to simulate a wide range of complex biological, chemical and materials problems within reasonable time frame and with sufficient accuracy---not just for the small molecules which we have used as "proof of principle". Our submission stands out from the combination of research disciplines to tackle an important and complex problem, the impact the GA-discovered solution may hold, and the use of multiobjective evolutionary algorithm to surpass any previous human or machine-based solution provided in the past two decades.