Electronic structure of low-dimensional inorganic/organic interfaces: Hybrid density functional theory, $G_0W_0$, and electrostatic models

TL;DR

This paper evaluates the effectiveness of various ab initio computational methods in predicting the electronic properties of hybrid inorganic/organic interfaces, highlighting limitations of DFT and proposing electrostatic models as efficient alternatives.

Contribution

It demonstrates that electrostatic screening models can reliably predict interface properties at low computational cost, outperforming traditional DFT and $G_0W_0$ approaches for weakly interacting systems.

Findings

DFT and hybrid functionals have limited accuracy for hybrid interfaces.

$G_0W_0$ starting points are unreliable for these systems.

Electrostatic models match experimental data effectively.

Abstract

First-principles simulations of electronic properties of hybrid inorganic/organic interfaces are challenging, as common density-functional theory (DFT) approximations target specific material classes like bulk semiconductors or gas-phase molecules. Taking as a prototypical example anthracene physisorbed on monolayer MoS$_2$, we assess the ability of different \textit{ab initio} schemes to describe the electronic structure using semi-local and hybrid DFT. For the latter, an unconstrained three-parameter range-separation scheme is employed. Comparisons against many-body perturbation theory results indicate that DFT is substantially unable to make reliable predictions about interfacial properties. Hybrid functionals, while improving the accuracy of the MoS$_2$ band structure, do not systematically enhance the description of hybrid systems with respect to semi-local functionals. Neither approach provides a good starting point for $G_0W_0$, which, consequently, cannot provide much information beyond the correct energy level alignment. We show that non-empirically parametrized electrostatic screening models can accomplish the same task at negligible computational costs. Such schemes can include substrates of hybrid interfaces in good agreement with experimental data. Our results indicate that currently, fully atomistic, many-body simulations of weakly interacting hybrid systems are not worth the required computational resources. In contrast, ab-initio-parametrized effective models mimicking the environment offer a scalable alternative without compromising accuracy and predictivity.