Search papers, labs, and topics across Lattice.
This paper benchmarks the fidelity of Permutationally Invariant Polynomials (PIP) and PhysNet machine-learned potential energy surfaces (ML-PESs) for protonated oxalate by comparing vibrational energies, IR spectra, and tunneling splittings. They stress-test the ML-PESs by evaluating energies across a vast 15-dimensional configurational space, requiring ~1 billion energy evaluations. The results from both ML-PESs show excellent agreement across all tested properties, suggesting high fidelity in representing the underlying potential energy surface.
Despite using fundamentally different machine learning architectures, PIP and PhysNet potentials yield remarkably consistent results for vibrational properties and tunneling dynamics, suggesting that both can reliably represent complex potential energy surfaces.
There has been a veritable explosion of methods and software to perform machine-learned regression on datasets of electronic energies and forces to develop high-dimensional machine learned potential energy surfaces (ML-PESs). A major, but not deeply-studied aspect is how well different ML-PESs represent the same dataset on which they are trained, beyond the standard fitting precision metrics. Here, this is examined in detail using several''stress tests'', for two widely applied machine-learned potential approaches. One is based on permutationally invariant polynomial (PIP) linear least square regression and the other is the message-passing neural network PhysNet approach. These potentials and dipole moment surfaces are used in VSCF/VCI calculations of vibrational energies and wavefunctions. The energies from the two PESs are directly compared as are the IR spectra. In addition, tunneling splittings for the hydrogen transfer between two equivalent structures are reported from using three methods: ring polymer instanton theory, diffusion Monte Carlo simulations, and the $Q_{im}$ path method. These calculations require the evaluation of on the order of one billion energies that are widely dispersed in the 15-dimensional configurational space. The two PESs yield results for these quantities in excellent agreement with each other.
