Delta-Machine Learning for Potential Energy Surfaces: A PIP approach to bring a DFT-based PES to CCSD(T) Level of Theory

"$\Delta$-machine learning" refers to a machine learning approach to bring a property such as a potential energy surface based on low-level (LL) DFT energies and gradients to close to a coupled-cluster (CC) level of accuracy. Here we present such an approach that uses the permutationally invariant polynomial (PIP) method to fit high-dimensional PESs. The approach is represented by a simple equation, in obvious notation, $V_{LL{\rightarrow}CC}=V_{LL}+\Delta{V_{CC-LL}}$. The difference potential $\Delta{V_{CC-LL}}$ is obtained using a relatively small number of coupled-cluster energies and precisely fit using a low-order PIP basis. The approach is demonstrated for CH$_4$ and H$_3$O$^+$. For both molecules, the LL PES, $V_{LL}$, is a PIP fit to DFT/B3LYP/6-31+G(d) energies and gradients, and $\Delta{V_{CC-LL}}$ is a PIP fit based on CCSD(T) energies. For CH$_4$ these are new calculations using an aug-cc-pVDZ basis, while for H$_3$O$^+$ previous CCSD(T)-F12/aug-cc-pVQZ energies are used. With as few as 200 CCSD(T) energies, the new PESs are in excellent agreement with benchmark CCSD(T) results.