Blog

We are often taught to view materials as continuous blocks, solving Navier-Stokes for fluids or FEA for solids. But as engineering pushes into the nanoscale (think battery interfaces, nanofluidics, or crack initiation), the continuum assumption often breaks down. This is where molecular dynamics (MD) becomes indispensable.

The Physics of the Unseen

At its core, MD is a deterministic simulation technique. We aren't just visualizing atoms; we are solving Newton’s Second Law (F=ma) for a system of N particles. The magic lies in the interatomic potential (or force field), a mathematical function that defines how atoms interact (attract, repel, bond). By calculating the forces on every atom and integrating their equations of motion over discrete steps (typically 1 femtosecond, or 10^-15 s), we generate a trajectory that reveals the microscopic evolution of the system. We essentially trade the approximation of continuum mechanics for the computational cost of tracking millions of individual particles.

From Hard Spheres to Exascale Computing

The field was born in the late 1950s when Alder and Wainwright performed the first MD simulation using hard spheres—perfectly elastic balls bouncing in a box—to study phase transitions.

In 1964, Aneesur Rahman simulated liquid argon using realistic continuous potentials (Lennard-Jones), marking the true beginning of modern MD. What started as simulating 500 atoms for a few picoseconds has exploded into the "Exascale Era." Today, we routinely simulate billions of atoms for microseconds, allowing us to observe complex phenomena like protein folding, dislocation dynamics in alloys, and fluid flow through nanoporous media.

Bridging the Gap: Fluids, Solids, and Quantum Mechanics

For mechanical engineers, MD is the bridge between quantum mechanics and continuum mechanics.

In Fluids: We use it to study wetting, contact angles, and slip flow in nano-channel regimes where the "no-slip condition" of standard fluid dynamics fails.

In Solids: We simulate how materials fail before the crack is visible. It allows us to watch dislocations move, grain boundaries slide, and voids nucleate under stress.

The Quantum Frontier: While classical MD uses fixed force fields, Ab Initio MD (AIMD) calculates forces on the fly using quantum mechanics (density functional theory). This allows us to simulate chemical reactions and bond breaking, though at a much higher computational cost.

Current Trends: The AI Revolution

The biggest trend right now is the integration of machine learning (ML). Historically, we had to choose between fast but inaccurate classical potentials and accurate but slow quantum calculations. Now, researchers are training neural networks on quantum data to create machine learning potentials (MLPs). This gives us "quantum-level accuracy at classical speeds," revolutionizing materials discovery.

The Toolset

If you want to get started, LAMMPS (Large-scale Atomic/Molecular Massively Parallel Simulator) is the industry standard for materials and fluids due to its versatility. For visualization, OVITO is the go-to tool for rendering and analysis. Mastering these, along with Python for data post-processing, is a massive value-add for any simulation engineer.

References for the Curious:

1. The Origin Story: Alder, B. J., & Wainwright, T. E. (1957). Phase Transition in Elastic Disks. The paper that started it all.

https://journals.aps.org/pr/abstract/10.1103/PhysRev.127.359

2. Computer Simulation of Liquids by Allen and Tildesley

https://levich.ccny.cuny.edu/koplik/molecular_simulation/AT2.pdf

3. The Standard Code: Plimpton, S. (1995). Fast Parallel Algorithms for Short-Range Molecular Dynamics. The foundation of the LAMMPS code used worldwide.

https://aiichironakano.github.io/cs653/Plimpton-MD-JCP95.pdf

4. Visualization: Stukowski, A. (2010). Visualization and analysis of atomistic simulation data with OVITO–the Open Visualization Tool

https://iopscience.iop.org/article/10.1088/0965-0393/18/1/015012?utm_source=researchgate.net&utm_medium=article

5. Behler, J. (2016). Perspective: Machine learning potentials for atomistic simulations. A great overview of how AI is changing the field. https://pubs.aip.org/aip/jcp/article/145/17/170901/195141/Perspective-Machine-learning-potentials-for 

“Big whirls have little whirls, which feed on their velocity,

And little whirls have lesser whirls, and so on to viscosity.” 

— Lewis F. Richardson (1881-1953), Mathematician, Physicist and meteorologist.

This simple poem beautifully explains the heart of turbulence, the process by which energy flows from large, swirling motions to smaller and smaller ones until it finally dissipates as heat. This transfer of energy, called the energy cascade, is one of the most fascinating and challenging problems in fluid mechanics. Over the decades, scientists have used Direct Numerical Simulations (DNS) to uncover the hidden structures and behaviors within turbulent flows, pushing the limits of both physics and computing.

In 1993, Jiménez et al. explored “The Structure of Intense Vorticity in Isotropic Turbulence” using a 130-million-point simulation. They discovered thin, tube-like vortices nicknamed “worms” where the vorticity was extremely strong. These “worms” scale with the Kolmogorov microscale, the smallest size of motion before viscosity dominates. Their finding showed that turbulence is not just random motion but contains organized, coherent structures that carry intense rotational energy.

A decade later, Kaneda et al. (2003) expanded this understanding with 700 million grid points in “Energy Dissipation Rate and Energy Spectrum in High Resolution DNS”. They confirmed that, at high Reynolds numbers, the mean energy dissipation rate becomes independent of viscosity, a major prediction of Kolmogorov’s theory. This work also emphasized how challenging DNS is, as increasing turbulence requires exponentially more computational resources.

Then came the era of extreme events. In 2015, Yeung et al. published “Extreme Events in Computational Turbulence”, using a staggering 5.5×10⁵ million grid points to study rare, violent fluctuations in turbulence. These extreme event bursts, where energy dissipation and vorticity exceed average values by tens of thousands, revealed that turbulence is highly intermittent and unpredictable at small scales.

Further studies, such as Iyer, Sreenivasan, and Yeung (2020) in “Scaling Exponents Saturate in Three-Dimensional Isotropic Turbulence”, examined how velocity increments deviate from self-similarity. They found that certain scaling laws “saturate” at high Reynolds numbers, hinting at the presence of vortex sheets, thin, energetic layers rather than simple eddies, making turbulence even more complex than previously thought.

Most recently, in 2024, Yeung et al. achieved an extraordinary leap with “GPU-Enabled Extreme-Scale Turbulence Simulations”, running DNS on the Frontier exascale supercomputer with 35 trillion grid points. Using Fourier pseudo-spectral methods and GPU acceleration, they demonstrated how next-generation computing can simulate turbulence with unprecedented detail, a step toward truly understanding the physics behind chaotic flows.

Across all these milestones, one key principle endures: in the inertial subrange, the energy carried by eddies of diameter D follows a power law proportional to D⁵⁄³. For air, this range spans roughly from 0.1 cm to 1 km, a stunning reminder of how turbulence connects the smallest ripples to the largest atmospheric motions.

From Richardson’s poetic insight to exascale simulations, the study of turbulence remains a beautiful blend of physics, mathematics, and computational power, revealing the hidden order within chaos.

DOI links of the papers mentioned above:

1993: 10.1017/S0022112093002393

2003: 10.1063/1.1539855

2015: 10.1073/pnas.1517368112

2020: 10.1103/PhysRevFluids.5.054605 / 10.1103/PhysRevFluids.7.104605

2024: 10.1016/j.cpc.2024.109364

[Image 1 Reference: Yeung, P. K., Zhai, X. M., & Sreenivasan, K. R. (2015). Extreme events in computational turbulence. Proceedings of the National Academy of Sciences, 112(41), 12633–12638. https://doi.org/10.1073/pnas.1517368112 

Image 2 Reference: Ghira, A. A., Elsinga, G. E., & Da Silva, C. B. (2022). Characteristics of the intense vorticity structures in isotropic turbulence at high Reynolds numbers. Physical Review Fluids, 7(10). https://doi.org/10.1103/physrevfluids.7.104605 ]