Using quantum computing, Microsoft has made chemical simulations 30 times faster

Scientists from Microsoft, ETH Zurich and Pacific Northwest National Laboratory recently proposed a new automated workflow, AutoRXN, that leverages The scale of Azure to transform the R&D process in quantum chemistry and materials science. The team achieved a 30x speedup and a 10-fold cost reduction when simulating a catalytic chemical reaction.

Hongbin Liu, Chief Architect at Microsoft Azure Quantum, is responsible for the design and research of quantum applications, especially computational chemistry and materials science.

01Quantum mechanics has made major breakthroughs in chemistry and materials science

Predicting chemical synthesis and catalytic processes is a key endeavor in chemistry, but one of the science's most pressing challenges. It is almost impossible to determine the synthesis mechanism of reactions that occur in complex chemical spaces only through laboratory experiments. Computer simulations provide an alternative and complementary avenue for elucidating reaction mechanisms, but the manpower time involved so far has been so great that researchers can only consider a few key reaction pathways, which are often overlooked in traditional settings. This is because reaction modeling requires chemical intuition and human trial and error, and when reactions are fully and thoroughly considered, including all possible options, accurate simulation of the modeled system becomes difficult to achieve.

This reality is what prompted the Azure Quantum team to build a fully scalable quantum machine. Since quantum mechanics explains the nature and behavior of matter at the atomic level, quantum computers will essentially be able to understand and predict the complexity of nature. As they move toward this vision, the Microsoft team is also helping innovators accelerate today's advances in chemistry and materials science through new workflows that leverage the power of state-of-the-art research and Azure high-performance computing (HPC)[1].

By optimizing the simulation code and refactoring it into cloud native, the team achieved a 30x speedup and a 10x cost reduction when simulating a catalytic chemical reaction [2]. In addition, these powerful automation capabilities free scientists from a web of complex, heterogeneous hardware and software packages, allowing them to focus on the development of new products such as sustainable production of fertilizers, more environmentally friendly paints and coatings, new methods of carbon sequestration, and many others.

02AutoRXN for automated reaction exploration

AutoRXN is a new automated workflow designed to enable scientists to virtually explore reaction networks using HPC in the Azure cloud. With this progress, it becomes incredibly easy to spot and evaluate chemical reactions in the cloud, which in turn will enable businesses to transform their R&D processes and accelerate the development of new products. Using the AutoRXN workflow, scientists can expand the number of chemical reaction pathways explored from dozens to thousands of configurations with greater accuracy than traditional ones. The core work behind the AuthoRXN coordination is Chemoton, a chemical network exploration software developed by Hongbin Liu's collaborators at ETH Zurich. They adapted the chemical simulations used today to be cloud-native to the modern hardware and network topology of Azure datacenters, ensuring autonomy, stability, and minimal operational disruption to all components of the workflow.

Displays a reactive network view of the exploration range

This automation has enabled the research outlined in a recent paper, which the team applied to study the mechanisms of asymmetric hydrogenation catalysts. The AutoRXN workflow explores a large number of relatively low-cost quantum chemical calculations, automatically refines the results of a large number of expensive related self-learning calculations, and automatically collects and evaluates data: including backchecking the results through other simulation methods. The team has been able to coordinate highly accurate computational chemistry calculations at unprecedented speeds, which are critical for high-throughput tasks.

The AutoRXN workflow opens up a new avenue for modeling and understanding chemical reactions, where many side reactions can be discovered and studied to inform the actual performance of catalysts. The exploration took a closer look at the expected reaction mechanism and revealed the general reactivity of the different atoms and functional groups in the catalyst, which allowed people to improve the catalyst.

Ultimately, the team identified more than five hundred reactions and more than two thousand basic steps, revealing a comprehensive overview of the asymmetric hydrogenation reaction catalyzed by the iron complex. This goes far beyond the scope of traditional artificial reaction modeling, as many side reactions and degradation of catalysts are beyond the intuitive reach of chemists today. Taking advantage of modern hardware heterogeneity on Azure makes this process significantly faster and more cost-effective. The results of the simulations help us understand the reactivity of catalysts and accelerate the development of exciting new discoveries in chemistry and materials science.

Exploring catalytic reactions on Azure high-performance computing provides researchers with a powerful and reliable platform for hyperscale chemical and materials simulation without actually building systems and infrastructure.

03Accelerate the path to innovation

It is estimated that chemistry directly involves more than 96% of all manufactured goods. This means businesses have the opportunity to make new chemical and materials science discoveries that solve society's toughest problems and generate new growth. New technologies and methods like AutoRXN are emerging from advances in cloud computing and computational chemistry. Innovations in cloud computing capabilities and automation have achieved unprecedented scalability and hardware heterogeneity, and deep cooperation between industry, academia and research is promoting the development of cloud-optimized simulation codes and methods.

These techniques have advanced computational chemistry to a stage where it can solve challenging problems that scientists have been working on for decades.

Reference Links:

[1]https://azure.microsoft.com/en-us/solutions/high-performance-computing/#applications

[2]https://www.microsoft.com/en-us/research/publication/high-throughput-ab-initio-reaction-mechanism-exploration-in-the-cloud-with-automated-multi-reference-validation/