How well can we model chemical synthesis?

The Materials Genome Initiative held a 2017 workshop, that led to this published report: de Pablo et al., "New frontiers for the materials initiative," npj Computational Materials 5, 41 (2019). In part, this report lists various successes of the initiative, due to complementary theoretical, computational, and experimental studies. Perhaps more interestingly, it also lists several challenges for future research. One particular challenge that gets listed in several contexts is related to computational modeling of synthesis:

Improving computational support for the synthesis of soft materials would be very valuable. Although significant steps have been made using machine learning to predict outcomes of simple organic reactions,36,37 extending this capability to include a larger range of chemical variety and macromolecular synthesis would democratize chemical synthesis and accelerate our validation of theories for new chemistries. Computational guidance in the iterative experimental synthesis process is also critical to future advances.38

Recent advances, particularly in x-ray diffraction at high temperatures, pressures, and inert environments could allow a more quantitative understanding of the thermodynamics of crystal synthesis. This could improve computational models of nucleation and crystal formation, help optimize the synthesis of known materials, as well as drive the discovery of new non-equilibrium compounds that are only stable under narrow conditions. Computational models of synthesis would be helpful to encourage feedback between theory and synthesis and enable calculations of optimal synthetic conditions.

Often synthesis is referred to as an art, which exudes respect for those active in the field, but may also signify that theory is lagging significantly behind the practice. Of course, synthesis typically occurs through rapid non-equilibrium processes, and sometimes at very elevated temperatures or in unusual atmospheres. This makes modeling such reactions beyond the input/output scope of simple reaction equations rather challenging, and a very different proposal to the calculation of equilibrium properties of some material post-synthesis. However, smart people have made progress on difficult problems before, and it seems clear that such modeling could potentially bring new insight and advance the art itself – and perhaps the design of novel materials.

Hence, I wonder:

1. How well can we theoretically or computationally model the dynamics of synthesis today, whether of molecules or crystals? What is considered state-of-the art, and what are its limitations?
2. Have there been any particularly striking successes or failures in this regard?
3. Do such computations incorporate quantum physics?

Personally, I am particularly interested in the case of transition metal systems (e.g. oxides). However, I’d suspect there’s been considerable advances on the side of molecular and organic chemistry, where there might be a large interest in such methods from the medical industry. In support of this hunch, I was able to find a List of computer-assisted organic synthesis software on Wikipedia, but no matching lists for the general or inorganic cases.

Matter Modeling Asked by Anyon on January 1, 2022

There is an initiative from IBM called RoboRXN. It is a free web service for predicting chemical reactions.

From the IBM Research Blog recent post:

The magic behind the RXN for Chemistry is a state-of-the-art neural machine learning translation method that can predict the most likely outcome of a chemical reaction using neural machine translation architectures. Similar to translating Italian to English, our method translates the language of chemistry converting reactants and reagents to products, using the SMILE representation to describe chemical entities.

Some features of RoboRXN are:

• We provide a state-of-the-art trained artificial intelligence (AI) model that can be used in your daily research activities irrespective of the purpose
• Use the prediction mode to open a project and invite collaborators to collectively plan complex synthesis.
• Use the challenge mode to test your Organic Chemistry knowledge and prepare for class exams
• Design your retrosynthesis either using the automatic or the interactive mode. In the interactive mode, IBM RXN for Chemistry –just like an assistant – recommends disconnections and you choose.

Answered by Camps on January 1, 2022

Okay first some basics:

For most reactions we do not need to run molecular dynamics

Molecules are like springs and obey the principles of minimum energy. Therefore, normally the dynamics of a molecule will be distributed around the "minimum energy path" going from reactants to products.

The maximum along this minimum energy path is the transition state, and has energy $$E_A$$, which can be used to calculate the rate of the reaction with transition state theory

$$k=Ae^{-E_A/RT}$$

If you were to run dynamics simulations, first you would find that would not be easy to observe a reaction take place; the movement and vibrations of the individual particles occur on the $$10^{-15}$$ s timescale, but the reaction events are much rarer, and far between (e.g. seconds or longer) -- this is a consequence of the principle of minimum energy. Secondly, you would find that

1. the reacting particles must collide with each other
2. the reacting particles must have enough energy to break the old bonds
3. the particles must have the proper orientation

In other words, they follow the minimum energy path! This theory is a workhorse in modern computational chemistry and is used by the Nature paper linked by Susi Lehtola.

Okay, now that that is in order, one apparent area where reaction modeling breaks down is when there is a bifurcation event after the transition state, for example see here

These DO require molecular dynamics because there is recrossing and also dynamical effects with a bifurcating potential energy surface.

However, if you're not aware of this then, then the products that you predict will be wildly off from experiments.

See for example, this excellent paper by Singleton et al. "Recrossing and Dynamic Matching Effects on Selectivity in a Diels-Alder Reaction", Angew Chem Int, Ed, 2009, 48, 9156 https://onlinelibrary.wiley.com/doi/epdf/10.1002/anie.200903293

Answered by Cody Aldaz on January 1, 2022

The computations do include quantum physics, since the very bare minimum of electronic structure calculations involve the quantum mechanics of electrons.

There's been quite a lot of development in computational chemistry in synthesis. While computational chemistry was pivotal already in the late 1980s in developing ozone safe alternatives to cholorofluorocarbons, see Physics Today 61, 4, 58 (2008); nowadays computations are often leading the design effort of new syntheses since calculations are much faster to carry out than lab experiments, see Nature Communications volume 7, 10109 (2016), for example.

Cody briefly explained transition state theory below. I could also point out a recently published paper on a success of transition state theory: the H2S+Cl system where dynamical effects were though to be important is actually fully described by transition state theory, it's just that you need a very good quantum chemical description to achieve the necessary accuracy in the computations: J. Chem. Theory Comput. 2020, acs.jctc.0c00354

Answered by Susi Lehtola on January 1, 2022

Related Questions

I get this error in Gaussian09 even though I have enough memory. “Error termination in NtrErr: NtrErr called from FIOCnC.” Any suggestions?

1  Asked on January 5, 2022 by quantumx

Why are single excitations ignored in the MP2 component of double hybrid functional calculations?

1  Asked on January 5, 2022

Regarding oscillatory strength theoretical units to experimental ones

1  Asked on January 3, 2022

How well can we model chemical synthesis?

3  Asked on January 1, 2022 by anyon

Is it required to sample the full reaction coordinate in umbrella sampling?

1  Asked on January 1, 2022 by bnd

How to find good universities for matter modeling?

1  Asked on December 25, 2021

What does the Neighbor command do in LAMMPS?

1  Asked on December 22, 2021

For software that does not support FCIDUMP format, what format is used and how can we get the software to interact with FCIDUMP integrals?

2  Asked on December 22, 2021

How do the various programs read or write integrals in FCIDUMP format?

1  Asked on December 20, 2021

Is there an anisotropy factor (g factor) for TDDFT absorption and circular dichroism calculations?

1  Asked on December 20, 2021 by c-alexander

Why am I losing bonds when I use PACKMOL on two proteins?

1  Asked on December 20, 2021

Can we “invert” Density Functional Theory through sufficiently accurate experiment?

1  Asked on December 18, 2021 by kfc

How do you calculate the “true” chemical potential in classical density functional theory?

1  Asked on December 16, 2021

Converging the potential of a dilute binary alloy system using the Coherent Potential Approximation. What troubleshooting options could I take?

1  Asked on December 16, 2021 by luphys

The Hund’s J – Why can this be quantified?

1  Asked on December 16, 2021

How accurate are the most accurate calculations?

1  Asked on December 13, 2021 by roman-korol

Is there any software that can generate the electronic DOS of GaAs?

1  Asked on December 13, 2021

Which Linux distribution is best for Matter Modeling?

9  Asked on December 11, 2021

What is the role of a chemist in computational chemistry?

4  Asked on December 9, 2021 by georgios-karaiskakis

How to deduce phase transitions from a phonon calculation?

2  Asked on December 9, 2021 by koroma