Benchmarking vibrational spectra: 5000 accurate eigenstates of acetonitrile using tree tensor network states (2504.05382v2)

Abstract: Accurate vibrational spectra are essential for understanding how molecules behave, yet their computation remains challenging and benchmark data to reliably compare different methods are sparse. Here, we present high-accuracy eigenstate computations for the six-atom, 12-dimensional acetonitrile molecule, a prototypical, strongly coupled, anharmonic system. Using a density matrix renormalization group (DMRG) algorithm with a tree-tensor-network-state (TTNS) ansatz, a refinement using TTNSs as basis set, and reliable procedures to estimate energy errors, we compute up to 5,000 vibrational states with error estimates below 0.0007 $\mathrm{cm}^{-1}$. Our analysis reveals that previous works underestimated the energy error by up to two orders of magnitude. Our data serve as a benchmark for future vibrational spectroscopy methods and our new method offers a path toward similarly precise computations of large, complex molecular systems.