- Atlas
- Natural Sciences
- Computational Physics
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
- Numerical Solution of Differential Equations 1945-Present
- Monte Carlo method 1949-Present
- Molecular Dynamics 1957-Present
- N-Body and Hydrodynamical Simulation 1963-Present
- Density Functional Theory 1964-Present
- Lattice Gauge Theory 1974-Present
- Car–Parrinello molecular dynamics 1985-Present
- Quantum Computing for Physics 2000-Present
- Quantum simulator 2010-Present
- Machine Learning for Physics Simulation 2012-Present
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
- Computational physics (introduction, Status in physics, Overview, Challenges in computational physics, Methods and algorithms), Wikipedia
- Finite difference method, Wikipedia
- Numerical methods for ordinary differential equations (History), Wikipedia
- Runge–Kutta methods, Wikipedia
- Numerical analysis (introduction, History), Wikipedia
- Monte Carlo method, Wikipedia
- Metropolis–Hastings algorithm, Wikipedia
- Molecular dynamics (introduction, Potentials in ab initio methods), Wikipedia
- Density functional theory (Origins, introduction), Wikipedia
- Quantum Monte Carlo (Background), Wikipedia
- Kohn–Sham equations (History), Wikipedia
- Force field (chemistry), Wikipedia
- Car–Parrinello molecular dynamics, Wikipedia
- Lattice QCD, Wikipedia
- Wilson action, Wikipedia
- 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.
- 1945-Present
Numerical Solution of Differential Equations
How can equations be solved step by step on a computer?
- 1949-Present
Monte Carlo method
How can random sampling estimate averages too large to compute?
- 1957-Present
- 1963-Present
N-Body and Hydrodynamical Simulation
How do galaxies and fluids evolve under gravity?
- 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 CurriculumThe first course in Density Functional Theory typically builds around these fundamental concepts, taught in a logical progression:- Electron density as the central variable: The ground-state electron density determines all properties of a many-electron system. This is the foundational claim that distinguishes DFT from wavefunction-based methods.
- 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 .
- 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 includes the external potential, the Hartree potential from the electron density, and the exchange-correlation potential.- Exchange-correlation functional: The unknown functional 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.
- 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.
- Generalized Gradient Approximation (GGA): An improvement over LDA that includes dependence on the gradient of the density, capturing some effects of density inhomogeneity.
- 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.
- 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.
- 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.
- 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.
- 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.
- 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 HistoryDensity 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 exchangeThe 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
- 1974-Present
- 1985-Present
Car–Parrinello molecular dynamics
How can atomic motion be simulated with quantum forces?
- 2000-Present
Quantum Computing for Physics
How can quantum machines simulate physics?
- 2010-Present
Quantum simulator
How can controlled quantum systems mimic other quantum systems?
- 2012-Present
Machine Learning for Physics Simulation
How can learning algorithms speed up physical simulation?
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.
Lattice Gauge Theory influenced by Monte Carlo method
Kenneth Wilson's lattice formulation of gauge theory turned the QCD path integral into a high-dimensional sum that could be evaluated by Monte Carlo sampling. Lattice QCD calculations from the 1980s onward relied on the Metropolis algorithm and related sampling techniques developed in statistical physics.
Carried by Kenneth Wilson.
Car–Parrinello molecular dynamics grew out of Density Functional Theory
Car–Parrinello molecular dynamics computes interatomic forces on the fly from density functional theory rather than from empirical force fields. The 1985 method by Car and Parrinello unified the electronic density-functional equations with ionic motion in a single dynamical scheme.
Carried by Roberto Car; Michele Parrinello.
Car–Parrinello molecular dynamics influenced by Molecular Dynamics
Car–Parrinello molecular dynamics extends the classical molecular-dynamics framework of integrating equations of motion for nuclei. It keeps the MD time-stepping structure but replaces the empirical force field with forces obtained from quantum mechanics.
Carried by Roberto Car; Michele Parrinello.
Quantum simulator grew out of Quantum Computing for Physics
Quantum simulators grew out of the quantum-computing program as special-purpose devices designed to emulate quantum systems that are hard to simulate classically. The idea of using controlled quantum hardware for physics simulation is a direct branch of quantum computing for physics.
Machine Learning for Physics Simulation influenced by Monte Carlo method
Machine-learning approaches to physics simulation often use Monte Carlo sampling to generate training data and to evaluate partition functions. Techniques such as variational Monte Carlo and Markov-chain sampling provided the statistical framework that machine-learning models later accelerated or replaced.
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.
Numerical Solution of Differential Equations influenced Numerical Climate Modeling in Climatology
Numerical solution of differential equations supplied the finite-difference and later spectral methods that climate modeling adopted to integrate the atmospheric equations. The computational techniques developed for solving PDEs were carried directly into general circulation models.
Carried by John von Neumann.
Numerical Solution of Differential Equations influenced Numerical Modeling and Simulation in Planetary Science
Numerical solution of differential equations is the mathematical foundation of numerical modeling and simulation; the discretization methods were taken over directly.
Numerical Solution of Differential Equations influenced Numerical Weather Prediction in Atmospheric Science
Numerical solution of differential equations provided the discretisation methods, time-stepping schemes and stability analysis that numerical weather prediction applied to the primitive equations. The barotropic and baroclinic forecast models of the 1950s were direct applications of these numerical techniques to atmospheric flow.
Carried by Jule Charney; John von Neumann.
Numerical Solution of Differential Equations influenced Eulerian Grid Fluid Simulation in Physical Simulation
Eulerian grid fluid simulation solves the Navier-Stokes equations with finite-difference and finite-volume discretisations taken from numerical solution of differential equations. The advection, projection and pressure-solve steps derive from those numerical methods.
Monte Carlo method influenced Monte Carlo localization in Robot Perception And Slam
Monte Carlo localization samples particle sets to represent the belief over robot pose, directly applying the Monte Carlo method of drawing random samples to estimate quantities. The particle filter used in MCL is a sequential Monte Carlo method.
Carried by Dieter Fox; Wolfram Burgard; Sebastian Thrun.
Monte Carlo method influenced Monte Carlo Light Transport in Graphics Rendering
Metropolis and Ulam's Monte Carlo method, developed at Los Alamos in the 1940s, supplied the sampling and estimator machinery that Kajiya and others applied to light transport in 1986. Path tracing is a direct application of Monte Carlo integration to the rendering equation.
Carried by Nicholas Metropolis; Stanley Ulam; James Kajiya.
Monte Carlo method influenced by Statistical Mechanics in Physical Chemistry
Nineteenth-century statistical mechanics, from Maxwell and Boltzmann through Gibbs, defined the equilibrium averages and partition functions that Monte Carlo sampling was devised to compute. The physics problem of evaluating high-dimensional integrals over configurations motivated the method.
Monte Carlo method influenced by Statistical mechanics in Theoretical Chemistry
Statistical mechanics supplied the ensemble averages and Boltzmann weights that Monte Carlo methods estimate by sampling. Metropolis and Ulam's 1949 paper and the 1953 Metropolis algorithm were framed explicitly as computing statistical-mechanical averages.
Carried by Nicholas Metropolis; Stanislaw Ulam.
Monte Carlo method influenced Physically based rendering and Monte Carlo light transport in Computer Graphics
Physically based rendering solves the light-transport integral by Monte Carlo sampling, taking the estimator and variance-reduction techniques of the Monte Carlo method directly. Kajiya's path tracing is an application of Monte Carlo integration to the rendering equation.
Carried by James T. Kajiya.
Molecular Dynamics influenced by Molecular mechanics in Theoretical Chemistry
Molecular mechanics supplied the classical force fields and potential functions that molecular dynamics integrates over time. The hard-sphere and Lennard-Jones potentials used in the first simulations came from this tradition of classical molecular modelling.
Density Functional Theory led to Mean-Field and Nuclear Density Functional Theory in Nuclear Physics
Nuclear density functional theory is a direct adaptation of the computational density functional theory developed for electronic structure, using the same variational principle and self-consistent mean-field solution. The nuclear Skyrme-Hartree-Fock method is the nuclear counterpart of Kohn-Sham DFT.
Carried by Walter Kohn; Pierre Hohenberg.
Density Functional Theory influenced by Hartree–Fock and post-Hartree–Fock wavefunction methods in Computational Chemistry
Hartree-Fock and post-Hartree-Fock wavefunction methods provided the reference calculations and the orbital framework that the Kohn-Sham scheme of 1965 adopted. Kohn and Sham replaced the wavefunction with electron density but kept a single-particle orbital structure derived from Hartree-Fock practice.
Carried by Walter Kohn; Lu Jeu Sham.

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 PhysicsRelated fields
Fields its ideas are linked to
- Physical Simulation 12 links
- Computational Chemistry 5 links
- Theoretical Chemistry 5 links
- Atomic Molecular Physics 5 links
- Numerical Analysis 4 links
- Physical Chemistry 4 links
- Applied Mathematics 3 links
- Biophysics 3 links
- Soft Matter Physics 3 links
- Statistical Mechanics 3 links
- Planetary Science 2 links
- Computational Materials Science 2 links
More in Physics
- Acoustics
- Atomic Molecular Physics
- Biophysics
- Classical Mechanics
- Condensed Matter
- Electromagnetism
- Fluid Dynamics
- Mathematical Physics
- Nonlinear Dynamics And Chaos
- Nuclear Physics
- Optics
- Particle Physics
- Physics
- Plasma Physics
- Quantum Field Theory
- Quantum Information
- Quantum Mechanics
- Relativity
- Soft Matter Physics
- Statistical Mechanics
- Thermodynamics
See where Computational Physics sits on the Map of Human Knowledge