On the Predictive Power of Chemical Concepts

Abstract

Many chemical concepts can be well defined in the context of quantum chemical theories. Examples are the electronegativity scale of Mulliken and Jaffé and the hard and soft acids and bases concept of Pearson. The sound theoretical basis allows for a systematic definition of such concepts. However, while they are often used to describe and compare chemical processes in terms of reactivity, their predictive power remains unclear. In this work, we elaborate on the predictive potential of chemical reactivity concepts, which can be crucial for autonomous reaction exploration protocols to guide them by first-principles heuristics that exploit these concepts.

Autonomous Reaction Network Exploration in Homogeneous and Heterogeneous Catalysis

“…Our resource estimates showed that truly extensive explorations based on density functional theory calculations without activated pruning schemes (to cut deadwood in the exploration) are not feasible because of the sheer number of exploratory calculations to be carried out. This can be alleviated by suitable first-principles reactivity descriptors [ 160 – 163 ] which not only can suggest potentially reactive sites to be prioritized in the exploration process, but which can also determine those sites that are likely to be unreactive and that can therefore be given a very low priority in the exploration process.…”

“…Most existing algorithms likely struggle with solid phases that undergo structural rearrangements during reactions on their surfaces so that the reaction intermediates significantly differ from their gas phase counterparts; examples are flexible catalysts such as nanoclusters [ 157 ], anchored organometallic complexes [ 158 ], and reactions that remove and regenerate atoms at a surface [ 159 ]. The increased degree of complexity that the direct structural involvement of such catalysts adds to the problem of the elucidation of catalytic reactions networks for large reactants with a high degree of structural flexibility highlights an even more pronounced role of automated exploration procedures, which we, given the diverse nature of potentially catalytic agents, decided to base on electronic structure information only [ 160 – 163 ]. This allows us to exploit general heuristic concepts based on the first principles of quantum mechanics.…”

“…Our strategy so far has been to base this selection process on rules that may be derived for any molecular system and that are therefore not bound to specific compound classes. Accordingly, we introduced first-principles heuristics as a way to extract conceptual information on reactivity from the electronic wave function [ 160 – 163 ]. Note that it is not required to make a precise prediction on what atoms may react in some intermediate.…”

“…These estimates do not consider any restriction or constraint in the exploration process itself. Apart from the pruning options already discussed above (i.e., first-principles heuristics for reactivity descriptors [ 160 , 162 , 163 ] and exclusion of short-lived intermediates from further exploration), the exploration process may be kinetically driven by steering trial and search calculations to those parts of the network that can be reached under reaction conditions by exploiting barrier information [ 161 ] or explicit micro-kinetic modeling [ 229 ]. Hence, we may assume that broad automated reaction network explorations are within reach, provided that reliable approximate methods are available and the exploration space can be limited without excluding important reactions.…”

Chemoton 2.0: Autonomous Exploration of Chemical Reaction Networks

“…While this capability makes our exploration system well-suited for the elucidation of kinds of chemistries, it may also be exploited to advance exploration algorithms. For instance, the networks obtained in a brute-force fashion so far can now be taken as a data reservoir to assess the predictive power of chemical reactivity concepts [79 ], which in turn can then be exploited to filter elementary step trials based on first-principles heuristics. [20 , 78 ] Clearly, various directions for future development and extensions are obvious (see also the previous section); we will consider interleaving explorations with Chemoton with microkinetic modeling approaches such as KiNetX [64 , 123 ] and extending the search algorithms for elementary-step trials towards molecular dynamics approaches with tailored biasing schemes [30 , 40 , 44 , 124 -127 ].…”