Computational Physics

Computational physics is the branch of physics that treats computation as a primary instrument of discovery, alongside theory and experiment.

10 frameworks, from 1945 to today · Physics · Natural Sciences

still a working view ended · each leads to its entry below

Overview

Computational physics is the branch of physics that treats computation as a primary instrument of discovery, alongside theory and experiment1. Its practitioners build numerical models of physical systems—a protein folding, a galaxy colliding, a quark-gluon plasma cooling—and run them on computers to predict what experiments will see or to explain what they have already seen1. What is at stake is reach: many systems of deep interest are analytically intractable, either because their equations have no closed-form solution or because the number of interacting parts is astronomically large1. Computation is how physics gets at those systems1. The field's history is a sequence of methods invented to break specific impasses, each one opening a new class of problems to attack.

Numerical solution of differential equations: the founding problem

Physics has been written in differential equations since Newton. For most real systems—three gravitating bodies, a fluid with viscosity, a quantum particle in an arbitrary potential—those equations cannot be solved in closed form1. The founding move of computational physics was to solve them approximately, step by step, on a machine1.

The key idea is discretization: replace continuous variables with a grid of points and derivatives with finite differences, turning a differential equation into a large system of algebraic equations a computer can grind through12. The Euler method, forward-stepping a solution by small increments, dates to Leonhard Euler in the eighteenth century; the more accurate Runge–Kutta methods were developed by Carl Runge and Wilhelm Kutta around 190034. What changed in the mid-twentieth century was that these methods became practical at scale5. The 1940s and 1950s brought electronic computers—ENIAC, then the stored-program machines—and with them the realization that a physicist could now integrate equations that had previously only been approximated by hand.

A concrete early landmark: in 1955, Enrico Fermi, John Pasta, Stanislaw Ulam, and Mary Tsingou ran a numerical experiment on the MANIAC computer at Los Alamos, simulating a chain of coupled nonlinear oscillators. They expected the energy to spread evenly among the modes—thermalize—but found it returned nearly periodically to its starting configuration. This "Fermi–Pasta–Ulam–Tsingou" result, published in 1955, was an early demonstration that numerical experiment could reveal physics no one had predicted analytically, and it helped seed the later study of solitons and nonlinear dynamics.

Monte Carlo methods: sampling instead of solving

A different impasse arose in statistical mechanics and nuclear physics: computing averages over enormous numbers of configurations, where direct summation is impossible. The answer was to sample randomly rather than enumerate exhaustively6.

The Monte Carlo method was developed at Los Alamos in the late 1940s by Stanislaw Ulam, John von Neumann, and Nicholas Metropolis, motivated partly by neutron transport calculations where particles scatter through matter in complicated ways. The core idea: if you want the average of some quantity over a probability distribution, draw random samples from that distribution and average the results67. The Metropolis algorithm (1953), which generates samples from a distribution using a random walk that accepts or rejects moves according to a simple rule, made this practical for problems in statistical physics—Ising models, lattice gases, phase transitions.

Monte Carlo's power is that its error shrinks as the square root of the number of samples regardless of the dimension of the problem, which is why it became indispensable for high-dimensional integrals and for systems with many degrees of freedom. It also became a workhorse of particle physics: simulating the detector response and the cascade of particles produced in collisions, so that experimentalists could compare data to theory.

Molecular dynamics: watching atoms move

Molecular dynamics (MD) simulates the motion of atoms and molecules by integrating Newton's equations of motion for each particle, with forces computed from a model of how particles interact8. The approach was pioneered in the late 1950s: Aneesur Rahman and others simulated liquids of interacting particles, and by the 1960s MD was established as a way to study liquids, solids, and biomolecules at the atomic level.

The method's promise is that it gives you the trajectory—not just equilibrium averages but the actual dance of atoms, from which you can extract diffusion coefficients, vibrational spectra, and mechanisms of conformational change. Its limitation is time scale: a simulation must take steps small enough to resolve atomic vibrations (femtoseconds), so reaching biologically relevant timescales (microseconds to milliseconds) requires enormous numbers of steps. This tension between accuracy and reach has driven decades of method development—faster force fields, multiple-time-step integrators, and enhanced sampling techniques that bias the simulation to explore rare events.

Density functional theory: quantum mechanics made tractable

The quantum mechanics of many electrons is formally exact but computationally hopeless: the wavefunction of N electrons depends on 3N coordinates, and N is large for any real material910. Density functional theory (DFT) offered a way out by reformulating the problem in terms of the electron density—a function of just three coordinates9.

The theoretical foundation was laid by Pierre Hohenberg and Walter Kohn in 1964, who proved that the ground-state energy of a many-electron system is a unique functional of the electron density. In 1965, Kohn and Lu Jeu Sham gave a practical scheme: replace the interacting electrons with a fictitious system of non-interacting electrons that has the same density, and solve single-particle equations911. The catch is that the exact functional is unknown; practical DFT uses approximations, most famously the local density approximation (LDA) and later the generalized gradient approximation (GGA).

DFT's impact was enormous because it made quantum-mechanical calculations of real materials—metals, semiconductors, molecules, surfaces—feasible on ordinary computers9. By the 1990s it had become the dominant method in condensed-matter physics and materials science, and Walter Kohn shared the 1998 Nobel Prize in Chemistry for it. Its limitations—the band-gap problem, the treatment of strongly correlated electrons—remain active research areas.

Ab initio molecular dynamics: forces from quantum mechanics

Classical MD needs a force field, a pre-specified model of interatomic forces12. Ab initio molecular dynamics (AIMD) removes that crutch by computing forces on the fly from quantum mechanics, typically DFT, at each time step13. The approach was developed in the 1980s and 1990s, notably by Roberto Car and Michele Parrinello, whose 1985 method unified the electronic and ionic degrees of freedom in a single dynamical scheme. AIMD made it possible to simulate chemical reactions, bond breaking, and materials under extreme conditions without empirical input, at the cost of much greater computation per step138.

N-body and hydrodynamical simulation: the cosmos on a computer

Astrophysics presented some of the earliest and most demanding computational problems: how do galaxies form, how do stars cluster, how does a supernova explode? These are N-body problems—many gravitating masses—and hydrodynamical problems—fluids under gravity and pressure.

Early N-body simulations in the 1960s and 1970s, such as those by Sverre Aarseth and by the Peebles group, followed hundreds or thousands of particles to study gravitational clustering. As computers grew, so did the ambition: by the 1980s and 1990s, cosmological simulations followed millions of particles to model the large-scale structure of the universe, and hydrodynamical codes incorporated gas dynamics, cooling, and star formation. The field drove the development of tree algorithms and particle-mesh methods for computing forces efficiently, and later of adaptive mesh refinement, which concentrates resolution where it is needed.

Lattice gauge theory: quantum field theory on a grid

Quantum chromodynamics (QCD), the theory of the strong force, is notoriously hard to solve analytically in the low-energy regime where quarks are confined14. Kenneth Wilson proposed in 1974 a lattice formulation: discretize spacetime onto a grid and compute the path integral numerically1514. This turned QCD into a problem amenable to Monte Carlo methods, and by the 1980s lattice QCD was producing quantitative results—hadron masses, the confinement phenomenon—that matched experiment. It remains one of the largest consumers of supercomputer time in physics, and its predictions for quantities like the mass of the proton are now precise enough to test the Standard Model.

Machine learning and quantum computing: the current frontier

Since the 2010s, machine learning has entered computational physics as both a tool and an object of study. Neural networks are being trained to approximate interatomic potentials, to accelerate DFT calculations, to identify phases of matter from simulation data, and to generate configurations in Monte Carlo. The appeal is speed and flexibility; the challenge is interpretability and reliability. Separately, quantum computing promises a different kind of computation—one that simulates quantum systems natively rather than approximating them on classical hardware. Quantum simulation of physics is an active experimental and theoretical frontier, though practical advantage over classical methods for problems of real physical interest remains an open question16.

The present

Computational physics today is defined less by a single method than by a toolkit: differential-equation solvers, Monte Carlo sampling, molecular dynamics, DFT, lattice field theory, and increasingly machine-learned surrogates, all running on hardware from laptops to exascale supercomputers. The field's central tension—between the accuracy of a model and the size or time scale it can reach—has not disappeared; it has simply moved to new problems, from the folding of proteins to the merger of black holes to the behavior of quantum materials. What the history shows is that each new method was invented to break a specific barrier, and that the barriers keep moving.

References

  1. Computational physics (introduction, Status in physics, Overview, Challenges in computational physics, Methods and algorithms), Wikipedia
  2. Finite difference method, Wikipedia
  3. Numerical methods for ordinary differential equations (History), Wikipedia
  4. Runge–Kutta methods, Wikipedia
  5. Numerical analysis (introduction, History), Wikipedia
  6. Monte Carlo method, Wikipedia
  7. Metropolis–Hastings algorithm, Wikipedia
  8. Molecular dynamics (introduction, Potentials in ab initio methods), Wikipedia
  9. Density functional theory (Origins, introduction), Wikipedia
  10. Quantum Monte Carlo (Background), Wikipedia
  11. Kohn–Sham equations (History), Wikipedia
  12. Force field (chemistry), Wikipedia
  13. Car–Parrinello molecular dynamics, Wikipedia
  14. Lattice QCD, Wikipedia
  15. Wilson action, Wikipedia
  16. Quantum computing, Wikipedia

The frameworks of Computational Physics

Each framework is a way of working the field has used: when it held, and the question it answers.

  1. 1945-Present

    Numerical Solution of Differential Equations

    How can equations be solved step by step on a computer?

    Fly to its tower in Nousfarer

  2. 1949-Present

    Monte Carlo method

    How can random sampling estimate averages too large to compute?

    Fly to its tower in Nousfarer

  3. 1957-Present

    Molecular Dynamics

    How do atoms move over time in a material?

    Fly to its tower in Nousfarer

  4. 1963-Present

    N-Body and Hydrodynamical Simulation

    How do galaxies and fluids evolve under gravity?

    Fly to its tower in Nousfarer

  5. 1964-Present

    Density Functional Theory

    How can electron density replace the many-body wavefunction?

    About Density Functional Theory
    Density Functional Theory (DFT) addresses a central computational challenge in physics: how to predict the properties of materials and molecules from first principles without solving the full, intractable many-electron Schrödinger equation. Its distinctive insight is to replace the complex many-body wavefunction—a function of 3N variables for N electrons—with the much simpler electron density, a function of only three spatial coordinates. This shift in focus from wavefunction to density makes DFT the most widely used electronic structure method in computational physics, enabling accurate calculations for systems ranging from individual atoms to solids, surfaces, and nanostructures.

    Core Curriculum

    The first course in Density Functional Theory typically builds around these fundamental concepts, taught in a logical progression:
    1. Electron density as the central variable: The ground-state electron density n(r)n(\mathbf{r}) determines all properties of a many-electron system. This is the foundational claim that distinguishes DFT from wavefunction-based methods.
    1. Hohenberg-Kohn theorems: Two theorems from 1964 establish the rigorous mathematical foundation. The first theorem proves a one-to-one mapping between the external potential and the ground-state electron density. The second theorem provides a variational principle: the true ground-state density minimizes the total energy functional E[n]E[n].
    1. Kohn-Sham equations: The practical workhorse of DFT. These equations map the interacting many-electron system onto a fictitious system of non-interacting electrons moving in an effective potential. The Kohn-Sham equations are:
    $\left[-\frac{\hbar^2}{2m}\nabla^2 + v_{\text{eff}}(\mathbf{r})\right] \psi_i(\mathbf{r}) = \varepsilon_i \psi_i(\mathbf{r})$ where veff(r)v_{\text{eff}}(\mathbf{r}) includes the external potential, the Hartree potential from the electron density, and the exchange-correlation potential.
    1. Exchange-correlation functional: The unknown functional Exc[n]E_{\text{xc}}[n] that captures all many-body effects beyond the non-interacting approximation. This is the only part of DFT that must be approximated, and the choice of approximation determines accuracy.
    1. Local Density Approximation (LDA): The simplest and historically first approximation, which assumes the exchange-correlation energy at each point depends only on the local density, as in a uniform electron gas.
    1. Generalized Gradient Approximation (GGA): An improvement over LDA that includes dependence on the gradient of the density, capturing some effects of density inhomogeneity.
    1. Self-consistent field (SCF) iteration: The standard algorithm for solving the Kohn-Sham equations. Starting from an initial guess for the density, one constructs the effective potential, solves for the Kohn-Sham orbitals, computes a new density, and repeats until convergence.
    1. Plane-wave basis sets and pseudopotentials: Common choices for periodic systems (solids). Plane waves are natural for crystals, while pseudopotentials replace the strong core-electron potential with a weaker effective potential, reducing computational cost.
    1. Bloch's theorem and k-point sampling: For periodic solids, Bloch's theorem allows calculations to be performed in the primitive cell, with integrals over the Brillouin zone approximated by sums over a finite set of k-points.
    1. Band structure and density of states: Key outputs from DFT calculations for solids. The Kohn-Sham eigenvalues approximate the electronic band structure, and the density of states gives the number of electronic states per energy interval.
    1. Total energy and forces: The total energy from DFT allows comparison of different atomic configurations. Forces on atoms, obtained from the Hellmann-Feynman theorem, enable structural relaxation and molecular dynamics.
    1. Van der Waals corrections: Standard LDA and GGA functionals fail to capture long-range dispersion interactions. Empirical corrections or specialized functionals are needed for systems where van der Waals forces are important.

    Intellectual History

    Density Functional Theory emerged from a long-standing problem in quantum mechanics: the many-electron problem. The immediate precursor was Hartree-Fock Theory (1928), which introduced the mean-field approximation but treated the wavefunction explicitly, making calculations expensive and missing electron correlation. In the 1950s, the Thomas-Fermi model provided an early, crude density-based approach, but it lacked the accuracy needed for chemical bonding.
    The true foundation of modern DFT was laid in 1964 by Pierre Hohenberg and Walter Kohn at the Université de Paris-Sud and the University of California, San Diego. Their seminal paper, "Inhomogeneous Electron Gas," proved the Hohenberg-Kohn theorems, establishing that the ground-state electron density uniquely determines all properties of a system. This was a profound conceptual shift: the density, not the wavefunction, became the fundamental variable.
    The practical breakthrough came in 1965, when Walter Kohn and Lu Jeu Sham published "Self-Consistent Equations Including Exchange and Correlation Effects." The Kohn-Sham scheme provided a practical computational method by introducing a fictitious system of non-interacting electrons. This work made DFT computationally tractable and remains the standard approach today.
    Early applications focused on simple solids and the uniform electron gas. The Local Density Approximation (LDA), developed in the 1960s and 1970s by workers including John Perdew and Alex Zunger, used quantum Monte Carlo results for the uniform electron gas to parameterize the exchange
    The 12 concepts its tower teaches
    • Band structure
    • Bloch's theorem
    • Electron density
    • Exchange-correlation functional
    • Generalized Gradient Approximation
    • Hohenberg-Kohn theorems
    • k-point sampling
    • Kohn-Sham equations
    • Local Density Approximation
    • Plane-wave basis sets
    • Pseudopotentials
    • Self-consistent field iteration

    Fly to its tower in Nousfarer

  6. 1974-Present

    Lattice Gauge Theory

    How can the strong force be computed on a grid?

    Fly to its tower in Nousfarer

  7. 1985-Present

    Car–Parrinello molecular dynamics

    How can atomic motion be simulated with quantum forces?

    Fly to its tower in Nousfarer

  8. 2000-Present

    Quantum Computing for Physics

    How can quantum machines simulate physics?

    Fly to its tower in Nousfarer

  9. 2010-Present

    Quantum simulator

    How can controlled quantum systems mimic other quantum systems?

    Fly to its tower in Nousfarer

  10. 2012-Present

    Machine Learning for Physics Simulation

    How can learning algorithms speed up physical simulation?

    Fly to its tower in Nousfarer

How the ideas connect

Each arrow runs from an earlier idea to the later one it shaped. The list below says what passed between them.

Numerical Solution of Differential Eq…Monte Carlo methodMolecular DynamicsN-Body and Hydrodynamical SimulationDensity Functional TheoryLattice Gauge TheoryCar–Parrinello molecular dynamicsQuantum Computing for PhysicsQuantum simulatorMachine Learning for Physics Simulati…

Beyond the field

Where Computational Physics's ideas came from in other fields, and what they led to there: the strongest 12 of 75 links.

A city in Nousfarer: a red-carpeted stair climbing terraces to a crowned tower on an acropolis, more towers round it, forest and snowy mountains beyond.
A city in Nousfarer, the game played in the browser.

Computational Physics is a planet in Nousfarer

Its city has a tower for each of these 10 frameworks. Land, walk in, and climb one concept at a time.

Land on Computational Physics

Related fields

Fields its ideas are linked to

More in Physics

See where Computational Physics sits on the Map of Human Knowledge