Cheap but accurate calculation of chemical reaction rate constants from ab initio data, via system-specific, black-box force fields

Julien Steffen and Bernd Hartke 2017
Highlighted by Jan Jensen

Figure 1 from the paper. A flowchart of the EVB-QMDFF program implemented in this work, for the case of a DG-EVB-QMDFF calculation.

A few years ago I highlighted Grimme's General Quantum Mechanically Derived Force Field (QMDFF) - a black box approach that gives you a system-specific force field from a single QM Hessian calculation.  I missed the fact that Hartke and Grimme extended this approach to TSs using EVB, a year later. This EVB-QMDFF approach constructs EVB potentials connecting each pair of minima described by QMDFF.  To get the EVB parameters you need to supply the TS and 10-100 energies (and possibly 5-10 Hessian calculations) along the reaction path, depending on how complex an EVB potential is needed to describe the reaction.

What's the point of a system-specific reactive force field when you already have the TS and reaction path? Well, Steffen and Hartke show is that EVB-QMDFF can be used to perform the additional calculations needed for, for example, variational TS theory or ring polymer MD calculations to get more accurate rate constants.

Furthermore, just like for QMDFF for minima you could do all this for one conformation of ligands and use EVB-QMDFF for a conformer search or use the gas phase parameterized model to study the effect of explicit solvation.  It might even be possible to parameterize EVB-QMDFF using small ligands and then model the effect of larger ligands using the QMDFF parameters obtained for the minima.  However, all these potential uses still need to be tested.

I thank Jean-Philip Piquemal for bringing this paper to my attention



This work is licensed under a Creative Commons Attribution 4.0 International License.

Development of a True Transition State Force Field from Quantum Mechanical Calculations

Ádám Madarász, Dénes Berta, and Robert S. Paton (2016)
Contributed by Jan Jensen

A true TS FF (TTSFF) is a force field for which the TS is a true saddle point on the PES, i.e. an unconstrained geometry with a zero gradient and exactly one imaginary frequency.  The idea is that you develop the FF parameters for each TS of interest from QM calculations and then use the TTSFF to do a conformational search to find the lowest energy TS structure. The TTSFF itself cannot be used to compute the barrier.

The main trick to making this work is to include harmonic stretch terms between 1-3 atom pairs (i.e. Urey-Bradley) terms in the FF expression. Some trial-and-error is required in selecting which FF parameters to optimise and exactly what to optimise against.

The approach is developed primarily to study selectivity, where the relative energies of TS structures determine which product is made.  However, one could imagine several other potential uses.  For example, one could probably parameterise the TTSFF using a small structural model and then use the TTSFF to find TS structures with large substituents, which would then be used a initial guesses for a QM TS search.  Furthermore, if QM//(TTS)FF barriers turn out to be reasonably accurate then one could save a tremendous amount of CPU time.

I thank @CompChemNews for bringing this paper to my attention


This work is licensed under a Creative Commons Attribution 4.0

A General Quantum Mechanically Derived Force Field (QMDFF) for Molecules and Condensed Phase Simulations

Stefan Grimme J. Chem. Theory Comput. 2014, ASAP
Contributed by +Jan Jensen

Stefan Grimme has develop a black-box approach by which a molecule-specific polarizable force field that can be automatically generated from the equilibrium geometry, Hessian, atomic partial charges and covalent bond orders computed with standard quantum mechanical methods.  In addition to the molecule-specific parameters there are 47 fixed parameters (listed in the paper) that were parameterized by separate QM calculations and cover all elements up to Radon.

One of the key challenges for any FF parameterization is the parameterization of the torsions (and any inversions) which can be a difficult to automate.  Grimme solves this by automatically extracting all relevant four atom fragments (CABD) including first-shell substituents on atoms C and D, adding H atoms where needed, and performing a 360$^\circ$ energy scan using modified extended Hückel calculations. The resulting PES is fit to the usual $cos$ functional form.  The extended Hückel calculations are modified to avoid double counting 1,4 interaction also included in the non-bonded part of the FF.

So what can you use the QMDFF for?  Well, for example, you can perform a QM calculation on a single conformation of a flexible molecule and use the resulting QMDFF to map out the conformational PES.  A similar calculation on a single solvent molecule allows you to explicitly solvate the molecule.  Similarly, one can search for the lowest energy conformation (and compute frequencies!) of host-guest complexes based on QM calculations on the host and guest.  For very large systems some kind of fragmentation scheme is required.

Finally, the QMDFF also allows for bonds to be broken (but not formed) so it can also be used to predict mass spectra using Grimme's QCEIMS method, which I highlighted earlier.  Since bonds cannot be formed, QMDFF is not a reactive FF like ReaxFF, but it is interesting to think about how one could make it in to one.

Thanks to +Anders Steen Christensen for alerting me to this article.


This work is licensed under a Creative Commons Attribution 4.0 International License.