In recent years, quantum chemistry has become truly accurate, with uncertainties comparable to typical uncertainties in many experiments. This should be leading to a complete transformation of chemistry, but so far it has not. A major cause of this failure has been that accurate quantum chemistry calculations of interesting observables (e.g., product mixture composition in organic synthesis and heterogeneous catalysis, kinetic isotope effects and rates of low-temperature reactions) are pretty complicated, and often require the efforts of several professional quantum chemists, each a specialist in a certain step of the calculation. We are working on next-generation algorithms in which many of these calculations will be routinely performed by the scientists interested in the problem, rather than by computational chemists who do not know the real physical system.
The inclusion of quantum mechanical nuclear effects (such as zero point energy and tunneling) in the calculation of chemical reaction rates is of particular importance. The role of these effects is well-known from textbooks: changes in zero point energy between the reactants and the transition state are responsible for the observed kinetic isotope effects in a wide variety of reactions, and tunneling can increase the rate of an activated proton transfer reaction at low temperatures by several orders of magnitude.
The exact inclusion of these effects in calculations of chemical reaction rates is one of the most challenging tasks of modern theoretical physical chemistry, because even assuming that a reliable electronic potential energy surface (PES) is available the computational effort that is needed to solve the reactive scattering Schrödinger equation increases exponentially with the number of atoms in the reaction. We are working on developing approximate methods to overcome this problem and to provide a practical way to include quantum mechanical effects in reaction rate calculations.
Advanced Methods for Discovery of Elementary Chemical Reactions and Prediction of Chemical Reaction Networks
We are working on the development of advanced automated algorithms for discovering important new chemical reactions. The problem of finding unexpected reactions is very challenging because it scales exponentially with the number of atoms in the reactant(s).
The key to significantly improving the scaling is to use evolutionary algorithms which use all the information that is known about the Potential Energy Surface (PES) and chemical bonds to improve the probability that the next search step will be near a saddle point. The algorithm will then use the computed energy, gradients and Hessian at that search point to “learn” more about the PES landscape and provide better informed decisions about which points to search next.
The goal of this project is to create a collection of fast and reliable algorithms to make a prognosis of the reactivity of organic molecules from their structure employing quantum chemistry calculations. Traditionally, computational analysis of possible reaction pathways requires working with large datasets.
There are some general-purpose workflow engines that allow users to organize and schedule different tasks using a graphical user interface. However, quantum chemistry calculations yield results, which cannot be used by most organic chemists directly. The output of the calculations must be translated into language or formalism to point out their chemical relevance. Although there are billions of reactions involved, only a limited number of factors or reactivity principles exist, which apply to the vast majority of chemical reactions. For example, factors such as “Acidity”, “Basicity”, “Lewis basicity”, are important reactivity descriptors for the largest number of reactions. The main idea of this project approach is to automate analysis of proposed reaction pathways by the calculating their key parameters. We strongly believe that the project will help chemists to understand various reaction mechanisms and to discover new reactions.
Algorithms for Optimization of Heterogeneous Catalysts
The development of efficient algorithms of computational catalytic design with minimal human intervention and optimal computational expenses represents one of the main challenges of present-day theoretical chemistry and physics. We are working on a novel approach for computational screening heterogeneous catalysts with variable-composition simulations of the material. At the core of our algorithm is an evolutionary algorithm which incorporates “learning from history” done through selection of the low-energy structures and high-catalytic activity to become parents of the new generation. Combining it with automated algorithms for screening catalytic activity of heterogeneous catalysts provides systematic and exhaustive tools for screening a set of chemically varied complex compounds. With the proper choice of the descriptor of catalytic activity, all relevant parameters can be automatically analyzed and the most promising materials identified. The method has several innovative characteristics that allows for its application to probe complex materials as it is automated and requires minimum human intervention. We are working on several interesting applications.
- Automated Discovery of Elementary Chemical Reaction Steps Using Freezing String and Berny Optimization Methods, YV Suleimanov and WH Green, Journal of Chemical Theory and Computation 11, 4248-4259
- Ring-Polymer Molecular Dynamics for the Prediction of Low-Temperature Rates: An Investigation of the C(1D) + H2 Reaction, KM Hickson, J-C Loison, H Guo, YV Suleimanov, The Journal of Physical Chemistry Letters 6, 4194-4199
- Should thermostatted ring polymer molecular dynamics be used to calculate thermal reaction rates?, TJH Hele and YV Suleimanov, The Journal of Chemical Physics 143, 074107