Building accurate and efficient ab initio potential energy surfaces for vibrational spectroscopy calculations via permutationally invariant polynomials

I will start by briefly introducing the theory of permutationally invariant polynomials (PIPs) for the fitting of potential energy surfaces (PESs).[1] Then, I will focus on recent advances of the technique and related applications involving glycine, tropolone, and aspirin.[2-4] The Δ-machine learning approach and comparison of PIPs to other machine learning methods employed for PES construction will be discussed in detail.[5-7] In the final part of the talk, after briefly introducing the basics of semiclassical vibrational spectroscopy,[8,9] I will present the outcome of such calculations for glycine and ethanol performed employing two PIP PESs.[2,10,11] Eventually, I will demonstrate the flexibility of semiclassical spectroscopy in dealing with large dimensional systems including solvated ones.[12,13] The final goal is to demonstrate that efforts oriented to a more and more accurate and computationally affordable description of the electronic structure of complex and large dimensional systems may allow one to perform accurate spectroscopical (and other chemically significant) calculations.[14]