Phase Transition for a Hard Sphere System

B. J. Alder, T. E. WainwrightThe Journal of Chemical Physics 1957 · DOI

TL;DR

In this seminal 1957 paper, Alder and Wainwright used early electronic computers to perform some of the first-ever molecular dynamics simulations, proving that a simple system of hard spheres—with no attractive forces—undergoes a first-order phase transition. By observing the system "jump" between two distinct pressure states, they provided the first direct evidence that purely repulsive interactions are sufficient to drive the transition from a fluid to a crystalline solid.

What Problem Does This Paper Address?

In the mid-20th century, a fundamental question in statistical mechanics was whether the simplest possible model of matter—hard spheres that only interact by colliding—could exhibit a phase transition like melting or freezing.

Prevailing theories were divided. Some analytical approaches, like the superposition theory, suggested a transition might occur, but others argued that attractive forces (like Van der Waals forces) were strictly necessary for a system to organize into a solid. Previous numerical attempts using Monte Carlo methods had yielded inconclusive or conflicting results. Alder and Wainwright sought to settle this by simulating the actual trajectories of particles over time, a method we now call Molecular Dynamics (MD).

Methods

The authors pioneered the use of "fast electronic computers" (specifically the UNIVAC and the IBM-704) to solve the simultaneous classical equations of motion for systems of several hundred hard spheres.

  • Simulation Setup: Particles were placed in a rectangular box with periodic boundary conditions to simulate an infinite bulk. They were initially arranged in an ordered lattice (like a crystal) with random velocity orientations.
  • Algorithmic Innovation: Unlike modern MD which often uses fixed time steps, this was an event-driven simulation. The computer calculated the exact time until the next collision between any two spheres, moved the particles to that point, and updated their velocities based on conservation of momentum.
  • Thermodynamic Mapping: After a short period to reach a Maxwell-Boltzmann velocity distribution, the authors calculated pressure using the virial theorem. This allowed them to relate the microscopic rate of momentum change during collisions directly to macroscopic pressure. They also cross-verified these results using the radial distribution function.

Datasets

While not a "dataset" in the modern sense, the authors evaluated the scaling of their findings across several system sizes:

  • Primary Systems: 32 and 108 particles. These provided the most stable data for the transition region.
  • Larger Systems: 256 and 500 particles. These were simulated but excluded from the final results due to "inadequate statistics" caused by the immense computing time required (the IBM-704 could only process 500 collisions per hour for a 500-particle system).
  • Comparison: A 96-particle system in a hexagonal arrangement was used to check for geometry-induced artifacts.

Key Insights and Ideas

  • The Two-Branch Phenomenon: The most striking discovery was that in a specific density range, the system did not settle on a single average pressure. Instead, the pressure "jumped" between two distinct branches.
  • Microscopic Caging: By studying particle positions, the authors found that on the lower branch (higher density/lower pressure), particles were "confined to the narrow region in space determined by their neighbors"—essentially trapped in a crystal lattice. On the upper branch, particles acquired enough freedom to exchange positions, characteristic of a fluid.
  • Spontaneous Transitions: The system would start in an ordered state (lower branch) and, after a variable number of collisions, spontaneously fluctuate into the fluid state (upper branch). At certain densities, it would even return to the ordered state, demonstrating the equilibrium nature of the fluctuation.
  • Entropy-Driven Ordering: This confirmed that the transition to a solid is driven by the gain in "local" entropy (more room for individual particles to move within their cages) even if the "global" configurational entropy decreases.

Key Results

  • Transition Range: The phase transition was observed when the volume ($v$) was approximately 1.5 to 1.6 times the volume of close packing ($v_0$).
  • Convergence: The 32-particle and 108-particle systems yielded identical pressure branches, suggesting that the periodic boundary conditions did not introduce significant errors for these small samples.
  • Fluctuation Scaling: In the 108-particle system, the jump to the fluid branch occurred much faster (fewer collisions per particle) than in the 32-particle system. This indicated that larger systems experience larger density fluctuations, making the transition easier to trigger.
  • Computational Benchmarks: The paper provides a fascinating snapshot of early computing: 32 particles required one hour for 7,000 collisions on an IBM-704.

Limitations and Future Work

  • Statistical Precision: The authors acknowledged that a truly "satisfactory determination" of the transition region would require much higher-speed equipment to average the pressure over many more jumps.
  • Coexistence Curve: Because the samples were so small, the two states could not exist simultaneously in equilibrium. The authors hypothesized that for an infinite system, these two branches would be connected by a horizontal line on a P-V diagram, representing a standard first-order transition.
  • Dimensionality: The authors noted they intended to investigate whether a similar transition occurs for hard spheres in two dimensions (hard disks), a study that would later become equally famous.

Summary updated

BibTeX
@article{alder1957phase,
  author = {B. J. Alder and T. E. Wainwright},
  title = {{Phase Transition for a Hard Sphere System}},
  journal = {The Journal of Chemical Physics},
  year = {1957},
  doi = {10.1063/1.1743957},
}
Figure