The project aims to leverage generative AI methods, including architectures that incorporate Convolution Neural Network (CNN) to find correlation between the molecular ground state energy and violations in exact conditions for the eigenstate wave-function of a molecular Hamiltonian system. The electronic structure problem is the key problem for drug design, catalyst finding, and material development. With the development of new quantum and classical computing algorithms for this problem, there is a question of how accurate the new algorithm will be for compounds of industrial relevance. These compounds are usually large molecules or periodic systems with large unit cells, so the exact solution cannot be obtained using current computing resources. The project proposes to address the question of accuracy prediction by finding transferable correlations between quantities that one can evaluate for any approximate method and whose values are known for the exact wavefunction, the latter will be referred to as indicators. To obtain reliable energy accuracy estimates, the project team will use as a training set exactly-solvable model Hamiltonians and Hamiltonians for small molecular system where the exact answer is known. The machine learning CNNs will be trained to find correlations between energy and indicator errors on a training set. It will be used to predict the accuracy of any computational method generating a wavefunction for molecules of interest.
This record is a historical disbursement from Canada's open grants database — it documents a past payment, not an active program accepting applications.
View disbursement record →