The general theme of our research consists in understanding structure, dynamics, and chemical reactions of complex molecular many-body systems—bridging the gap between chemistry and physics. Our aim is to capture nature as closely as possible by theoretical means—the basic entities being nuclei and electrons. This implies that we have to use atomistic ab initio computer simulation techniques which are capable of including dynamics and quantum mechanics - of course only approximately. The notion “ab initio” or “first principles” means for us that we neither want to fit to experimental data nor do we want to empirically adjust force field parameters.

Our most recent methodological developments in the realm of such ab initio simulations focus on computationally highly efficient (vulgo “cheap”) techniques that allow us to automatically construct highly accurate machine learning (ML) potential energy and electronic property surfaces which are exclusively based on the most accurate electronic structure methods currently available, namely on CCSD(T) coupled cluster calculations extrapolated to the complete basis set limit. This is achieved based on our in-house neural network based ML-code RubNNet4MD and its interface to the open-source CP2K molecular dynamics simulation package. Certainly a highlight is the introduction of what we call coupled cluster molecular dynamics (CCMD), which is an entirely ML-based methodology that opens up condensed matter simulations, in particular of liquids, at CCSD(T) accuracy.

Many decades ago, the field of molecular dynamics simulations that are directly based on electronic structure goes back to the pioneering work of Car and Parrinello (1985). The crucial idea of the Car-Parrinello approach to ab initio molecular dynamics consists in efficiently solving the electronic structure problem “on the fly” as the molecular dynamics trajectory is generated for a set of classical nuclei using Newtonian mechanics. Thus, within ab initio simulations it is neither required to compute a high-dimensional global potential energy surface prior to the simulation, nor is it necessary to reconstruct it approximately from local pair (or few-body) interactions.

This “classical” Car-Parrinello approach has been extended by Marx and Parrinello (1994) to include also the nuclei as quantum-mechanical degrees of freedom. In order to achieve this for “large systems” composed of the order of 100 nuclei or more, the Feynman-Kac formulation of quantum statistical mechanics in terms of path integrals is employed. This class of fully quantum-mechanical ab initio path integral techniques makes it possible to study—in a time-averaged sense—zero-point motion and tunneling effects for instance of protons in hydrogen-bonded or other complex environments. Extensions toward approximate quantum dynamics in the framework of ab initio simulations have been implemented using both centroid (path integral) molecular dynamics and ring polymer (path integral) molecular dynamics in all available flavors (1999–2010).

More recently, another extension of the original Car-Parrinello method, which assumed the electrons to stay in the electronic ground state, was developed by Doltsinis and Marx (2002). The basic idea of this nonadiabatic ab initio dynamics technique is to use Tully's surface hopping algorithm in combination with the so-called restricted open-shell Kohn-Sham Ansatz. This efficient approach “beyond the Born-Oppenheimer approximation” allows us to study photochemical reactions with particular focus on laser-induced processes in solution. More recently, multi-scale extensions from QM/MM up to coarse-grainning have been devised for such non-adiabatic simulations (2008–2010).

We have developed a multi-determinant Car-Parrinello propagation scheme, which enables the description of the dynamics of electronic states that cannot be represented using a single Kohn-Sham determinant (2007–2012). Using this strategy, we have computed Heisenberg's antiferromagnetic exchange coupling obtained from a spin-projected, Hubbard-corrected, broken-symmetry ground state. Generating the time evolution of this quantity “on the fly” provides access to magnetostructural dynamics, which arise from the intricate coupling of molecular motion and magnetic properties.

A field pioneered in Bochum is the general theory and computer simulation of covalent mechanochemistry (2002, 2009). In contrast to thermochemistry, photochemistry or electrochemistry (where temperature, light or electricity are used to trigger reactions), mechanochemistry utilizes mechanical force to activate and control chemical reactions. Advances in this field impact on areas of application currently under investigation such as molecular nanomechanics of single-molecule junctions, functionalized surface coatings, and mechanoenzymes.

Among the most recent developments is a method that allows us to solvate molecular complexes in superfluid helium droplets at sub-Kelvin temperatures (2014). It combines ab initio path integrals to treat chemically complex molecular solutes with a Monte Carlo sampling of the helium environment, in order to establish quantum mechanical indistinguishability - as required by the Bose-Einstein quantum statistics of ⁴He being a superfluid quantum liquid. This approach opens the doorway to the study of chemical reactivity in the absence of thermal energy, such as aggregation-induced dissociation phenomena and cryochemical reactions.

The ever-growing family of ab initio simulation techniques is ideally suited to the investigation of disordered systems at finite temperatures, molecular liquids being a prime example of this. As such, this set of methods provides the most direct insight into the structure and dynamics of solvation shells, the impact of hydrogen bonding on the properties of aqueous solutions, and, most importantly, the influence of solvation on chemical reactivity.

Within the emerging field of Solvation Science, theoretical infrared spectroscopy, in particular theoretical THz spectroscopy, has been developed with the aim to fully assign the far-IR lineshape in terms of distinct displacement patterns providing effective normal modes in a liquid state environment (2014). This is of particular importance in order to dissect THz spectra of aqueous solutions at the level of intermolecular solute/solvent coupling. Upon transferring and adapting these techniques to supercritical water (2020) and supercritical aqueous solutions (2022), it has been discovered that supercritical water in not hydrogen bonded and, thus, not a hydrogen-bonded solvent when it hosts solutes.

Upon combining solvation science with strong confinement effects, the field of nanoconfined water and aqueous solutions has been pioneered in the realm of charge migration processes (2013) and complex chemical reactions (2017) in ultranarrow slit pores. Subsequently, theoretical THz (2022) and VSFG (2023) spectroscopy methods have been tailored to decipher the properties of nanoconfined water and aqueous solutions down to the level of their hydrogen-bond network dynamics.

These simulation and analysis algorithms together with the required computer hardware constitute what we like to call a “Virtual Laboratory”. In this theoretician's version of a real laboratory chemical reactions of molecules can take place at finite temperature in liquids or on surfaces—solely governed by the basic laws of physics. This makes it possible to investigate “chemically complex” molecular systems—possibly in close contact with experimentalists.

To foster this, the Marx Group is or has been involved in several large-scale collaborative research projects such as SFB 558 (“Heterogeneous Catalysis”), FOR 436 (“Water at Interfaces”), FOR 618 (“Molecular Aggregation”), FOR 1979 (“Exploring the dynamical Landscape of Biomolecular Systems by Pressure Perturbations”), RTG 2376 (“Confinement-controlled Chemistry”), several projects of the Volkswagen-Stiftung (“Stress-Controlled Molecular Electronics”, “Multiscale Modelling”) as well as being the host of the Koselleck Focus Group on “Understanding Mechanochemistry”. Currently, key research of the Marx Group is carried out in the framework of the RESOLV Cluster of Excellence EXC 2033 (“Ruhr Explores Solvation”).

Much more detail on ab initio simulations can be found in the monograph
“Ab Initio Molecular Dynamics: Basic Theory and Advanced Methods”
written by Dominik Marx and Jürg Hutter
(Cambridge University Press, Cambridge 2009)

The research group itself consists (as a time average) of physicists, chemists, and biochemists and it is characterized by cross-disciplinarity. The spectral range of our interests is rather broad and covers applications to molecules, clusters, liquids, solids, surfaces, as well as to biologically relevant species. In order to be able to achieve these goals, we are constantly developing novel techniques and/or we are improving existing methods.