CARACTERIZATION OF THE SLATER DETERMINANT CLOSEST TO A CORRELATED WAVE FUNCTION AND INDIRECT ESTIMATION OF THE FULL-CI ENERGY
The use of codes and algorithms in theoretical chemistry has been growing exponentially in the last few years. A large portion of these codes run calculations of electronic structure methods that can be expensive computationally, given the nature of these basis sets and systems in study. Therefore, understating the methods of electronic structure is of great importance, what can lead to development of methods that give accurate results, but simpler and with fast execution.
In this study, we worked with an algorithm developed by our group, that optimizes the wave function ΨminD , that minimizes the distance to a correlated wave function and we verified that if we start from this distance, it is possible or not to achieve (or to get the close as possible) the exact energy’s system, the full configuration interaction energy (FCI), in a cheaper and faster way than calculating the FCI directly.
We performed these evaluations in two different ways: with a geometric point of view, where we analyze the distance among wave functions (through their overlap integrals) with interpretations of graphics and illustrations. And with a second point of view, where we utilize neural network systems (machine learning) to verify how these overlap integrals can be used to estimate the FCI energy.
Among the overlap integrals used, two of them succeed to estimate the FCI energy, what suggest that the geometric analysis can be used to recover information about the FCI wave function.