Domain-averaged exchange-correlation energies as a physical underpinning for chemical graphs.

A novel solution to the problem of assigning a molecular graph to a collection of nuclei (i.e. how to draw a molecular structure) is presented. Molecules are universally understood as a set of nuclei linked by bonds, but establishing which nuclei are bonded and which are not is still an empirical matter. Our approach borrows techniques from quantum chemical topology, which showed for the first time the construction of chemical graphs from wave functions, shifting the focus on energetics. This new focus resolves issues surrounding previous topological analyses, in which domain-averaged exchange-correlation energies (V(xc)), quantities defined in real space between each possible atom pair, hold the key. Exponential decay of V(xc) in non-metallic systems as the intercenter distance increases guarantees a well-defined hierarchy for all possible V(xc) values in a molecule. Herein, we show that extracting the set of atom pairs that display the largest V(xc) values in the hierarchy is equivalent to retrieving the molecular graph itself. Notably, domain-averaged exchange-correlation energies are transferable, and they can be used to calculate bond strengths. Fine-grained details resulted to be related to simple stereoelectronic effects. These ideas are demonstrated in a set of simple pilot molecules.