Force fields will accelerate atomic simulations by 10,000x in 2026, ushering in a new era of discovery

Written by Anders Blom and Igor Markov

“Force fields,” the invisible shields of science fiction legend that protect spaceships and superheroes from harm, have long captivated our imaginations. However, in the world of scientific discovery, force fields play a completely different role. A force field is a mathematical model that allows us to peer into the atomic core of matter itself.

Now, thanks to breakthrough advances in artificial intelligence, GPU acceleration, and advanced simulation platforms, force fields are poised to take a transformative leap from science fiction to mainstream research and development.

In fact, by 2026, machine learning force fields (MLFFs), also known as machine learning potentials (MLPs), are predicted to enable atomic simulations that are 10,000 times faster than today’s quantum-theoretic methods such as density functional theory (DFT).

Moreover, this speedup ushered in a new era of discovery and innovation.

What is a force field?

For decades, atomic simulation has led to major advances in materials science, electronics, and pharmaceuticals.

At the heart of these computer-based simulations is the accurate modeling of atomic interactions. To accomplish this, scientists often use force fields. It is a mathematical framework that defines how atoms attract, repel, combine (form molecules or crystals), and evolve (over time and under influence). Force fields provide “rules of engagement” to computer models by describing the potential energy landscape, such as the force at each point in space, or the electrostatic potential around a charged particle that guides molecular dynamics.

Atomic simulations allow researchers to investigate phenomena such as chemical reactions during catalysis, protein folding pathways, and the response of materials to heat and pressure. When developing and optimizing innovative materials, new drugs, and advanced nanodevices, it is important to be able to do this virtually before investing in expensive and time-consuming physical experiments.

However, until now these research areas have been hampered by a fundamental bottleneck: the trade-off between accuracy and computational feasibility.

Traditional force fields that use fixed equations and empirical parameters to model atomic behavior are computationally efficient and effective for simple systems. However, they struggle with complex materials and chemistries and lack the flexibility needed for cutting-edge discovery and innovation.

Conversely, DFT-based simulations that model atomic interactions at the quantum level offer much better accuracy but at enormous computational costs. Simulating thousands of atoms using traditional DFT techniques can take days to weeks, and large-scale realistic modeling involving hundreds of thousands or even millions of atoms remains completely impractical.

Fortunately, in recent years a new type of force field model has emerged that does not require such trade-offs.

Machine-learned force fields capture complex atomic interactions across vast systems.

The emergence of machine learned force fields (MLFF)

Machine learning force fields (MLFFs) represent a transformative leap forward in atomic simulation and scientific research.

Rather than relying on rigid predefined formulas, MLFF leverages artificial intelligence (usually deep neural networks) trained on quantum mechanical data from DFT calculations. This training allows you to learn the underlying patterns of atomic interactions and capture the complex energy landscape that governs molecular behavior. In a familiar chemical environment, these models can achieve near-DFT accuracy while running thousands of times faster than traditional quantum simulations.

Unlike traditional force fields, MLFFs can adapt to new chemical spaces and be retrained as additional data become available, making them much more flexible to meet evolving research needs.

The result is not only faster simulation; MLFF makes atomistic modeling practical for a wider range of problems, allowing researchers to explore more realistic scenarios and generate insights that were previously out of reach. MLFF allows researchers to:

Simulate a much larger systemscales from thousands to millions of atoms.
Explore longer time scalesfrom picoseconds to nanoseconds and beyond.
Model complex phenomena Heat transport, mechanical failure, diffusion, chemical reactions, etc.

All of these features are important for the research and development of next-generation materials and devices.

The technology engine that supports MLFF

The emergence and anticipated impact of MLFF can be traced back to recent hardware and software innovations. Additionally, the predicted 10,000x acceleration of simulation workflows is driven by four technical pillars:

artificial intelligence. MLFF learns atomic interactions from vast existing datasets in quantum computing and can rapidly adapt and fine-tune to new chemical environments using synthetic training data generated in real time.
GPU acceleration. Most MLFFs are natively designed to run on GPUs. Advances in algorithms, libraries, and hardware continue to improve both training and simulation speeds.
Network-based scaling. GPU memory remains the limiting factor for MLFF, so high-speed interconnects that allow many GPUs to work together seamlessly will be important not only to improve performance but also to scale simulations to handle much larger system sizes.
Advanced simulation platform. Platforms like Synopsys QuantumATK integrate MLFF running on multi-GPU architectures with advanced atomic simulation algorithms. These platforms integrate multiple computational approaches such as DFT within a single flexible Python interface, allowing users to easily customize workflows such as active learning and fine-tuning MLFF.

The availability of these technologies and the resulting democratization of advanced simulation will be a boon for research and innovation. Tasks that once required supercomputers and years of manual labor can now be completed in hours using GPU-accelerated systems. And scientists across the industry can access these powerful techniques through a professional software platform designed for ease of use and scalability.

As more startups, academic research institutions, and companies leverage these capabilities, we believe that atomic simulation will become routine in research and development, much like computer-aided design (CAD) was in mechanical engineering decades ago. MLFF plays a central role in this change, enabling large-scale, highly accurate simulations of chemical reactions, heat transport, and other complex phenomena that were once impractical to study at this level.

Synopsys QuantumATK accelerates semiconductor and materials research and development through large-scale atomic simulations.

Expected impacts: from semiconductors to pharmaceuticals

AI-powered, GPU-accelerated force fields are already having an impact across multiple fields. We also expect to see many breakthroughs in 2026, as it speeds up simulation workflows by a factor of 10,000 compared to quantum techniques such as DFT.

Development of advanced chip architecture. MLFF enables engineers to simulate complex interfaces and low concentrations of impurities, helping accelerate the development of next-generation chips and assess the impact of defects before manufacturing.
Discovery of new battery materials. Scientists are using MLFF to model solid electrolyte interfaces and ion transport much faster and more accurately, supporting the discovery of new chemistries for electric vehicles and grid storage.
Precision drug design. Drug developers can simulate protein-ligand interactions in atomic detail, enabling rapid and accurate screening of drug candidates and significantly reducing time to discovery.

The future of simulation has arrived

AI-powered, GPU-accelerated force fields hold great promise for research and innovation. When combined with advanced simulation platforms, it enables highly accurate modeling of atomic interactions at a fraction of the computational cost of traditional DFT methods.

2026 will be a transformative year in this sector. This is the year that force fields redefine the boundaries of atomic simulation, and previously unthinkable modeling and discoveries become commonplace. 10,000x faster workflows enable researchers and engineers to explore new frontiers in materials science, electronics, energy, and medicine.

The era of high-fidelity simulation for mainstream research and development has arrived. And the impact will be felt across the industry and around the world.