Computational Physics is the application of computational mathematics, numerical algorithms, simulation methods and computer systems to physical problems. It is an interdisciplinary area connecting theoretical physics, experimental physics, applied mathematics and computer science. Computational Physics is used when physical systems are too complicated for exact analytical treatment, when experiments are expensive or inaccessible, or when simulations can complement theory and observation.
Its subjects include classical mechanics, fluid dynamics, electromagnetism, statistical mechanics, quantum mechanics, condensed matter physics, plasma physics, astrophysics, nuclear physics and high-energy physics. Depending on the problem, computational approaches may solve differential equations, sample probability distributions, evolve particle systems, calculate quantum states or simulate interactions on discrete computational lattices. The field is therefore defined by its computational treatment of physical models rather than by a single algorithm or physical subdiscipline.
Classical Computational Physics includes numerical approaches to mechanics, electromagnetism and fluid dynamics. Ordinary differential equations can model trajectories, oscillators and dynamical systems, while partial differential equations describe fields and continua. Computational fluid dynamics solves discretised forms of equations governing fluid motion, often using finite-volume, finite-difference or finite-element techniques. Electromagnetic simulations may solve Maxwell's equations using finite-difference time-domain, finite-element or spectral methods. Particle simulations represent physical objects individually, while continuum methods describe macroscopic fields.
Molecular dynamics applies classical equations of motion to interacting particles and is widely used to study materials, liquids and biomolecular systems. The choice between particle-based and continuum descriptions depends on the physical scale, assumptions and quantities of interest. Numerical stability, conservation laws, boundary conditions and resolution are critical when translating physical equations into computational algorithms.
Statistical and quantum methods create distinctive Computational Physics problems. Monte Carlo methods use random sampling to estimate quantities associated with probability distributions and are particularly important in statistical mechanics, where direct enumeration of microscopic configurations can be impossible. Markov chain Monte Carlo constructs correlated samples from target distributions, while specialised methods such as Metropolis algorithms and cluster algorithms address particular physical systems. Quantum-mechanical calculations may involve solving the Schrödinger equation, computing eigenstates or approximating many-body wavefunctions.
Numerical methods include finite differences, basis expansions, diagonalisation and variational approaches. Density functional theory provides a widely used framework for electronic-structure calculations by reformulating a many-electron problem in terms of electron density under specific theoretical approximations. Quantum Monte Carlo, tensor-network methods and lattice techniques address other classes of quantum problems. Each method balances physical approximation, mathematical complexity and computational expense.
High-performance computing is particularly important in Computational Physics because realistic physical models can require extremely large computational domains or enormous numbers of degrees of freedom. Particle simulations may track millions or billions of particles, while grid-based simulations can involve very large three-dimensional meshes and many time steps. Parallelism can be achieved by distributing spatial domains, particles, matrix operations or independent simulation instances across processors. GPUs are increasingly used for numerical kernels with suitable parallel structure.
Large simulations also create substantial I/O requirements because storing every variable at every time step can produce enormous datasets. Checkpointing allows long simulations to be resumed after interruptions but adds storage and performance costs. Adaptive mesh refinement concentrates computational resolution where physical structures require it, reducing the cost compared with uniformly refining the entire domain. Efficient computational physics therefore depends on the interaction between physical modelling, numerical algorithms, hardware and data management.
Computational Physics frequently addresses multiscale and multiphysics problems. Physical phenomena can occur across length and time scales that differ by many orders of magnitude, making direct simulation of every scale impractical. Reduced models, homogenisation, coarse-graining and adaptive methods can bridge scales by representing small-scale effects through effective parameters or specialised submodels. Multiphysics simulations couple different governing equations, such as fluid flow with heat transfer, electromagnetism with material response, or structural mechanics with fluid dynamics.
Coupling may be tightly integrated within one solver or performed through separate specialised codes. The numerical treatment must preserve appropriate conservation properties and maintain stable interactions between coupled systems. Model coupling also introduces additional uncertainty because errors in one component can propagate into others. Computational Physics therefore often requires careful decomposition of a physical system into models that are sufficiently detailed to answer the scientific question without exceeding practical computational limits.
Verification and validation are essential in Computational Physics. Verification asks whether the numerical implementation solves the mathematical equations as intended, often through analytical test cases, manufactured solutions, convergence studies or conservation checks. Validation asks whether the physical model adequately reproduces observations or experimental measurements under relevant conditions. A simulation can pass verification while failing validation if the governing assumptions are inappropriate.
Conversely, agreement with an experiment may sometimes result from compensating modelling and numerical errors, which is why systematic analysis is important. Boundary conditions, initial conditions, material parameters and numerical resolution all affect outcomes. Sensitivity analysis can identify influential parameters, while uncertainty quantification can determine how uncertain inputs affect predictions. These practices prevent computational results from being treated as automatically equivalent to experimentally established facts and provide a structured way to evaluate the reliability of simulations.
Computational Physics is closely related to other computational disciplines. Computational Mathematics supplies numerical methods for differential equations, linear algebra, optimisation and stochastic computation. Computational Science provides a broader interdisciplinary framework in which Computational Physics is a domain-specific application. Scientific Computing supplies algorithms, software and computing infrastructure used to execute physical simulations. Computational Chemistry overlaps where quantum-mechanical or molecular methods are used to study chemical systems, while Computational Biology overlaps when physical models describe biological processes.
Computer Graphics may visualise physical simulations, but rendering a physical scene is distinct from numerically solving the underlying physical equations. Machine learning is increasingly used for surrogate models, parameter inference and acceleration, but a learned model is not inherently a physical simulation and must be evaluated according to its intended use.
Computational Physics has transformed the practical scope of physical investigation, but it does not eliminate the need for theory or experiment. Simulations inherit the assumptions of their governing models, and increasing computational resolution cannot compensate for a physically inappropriate model. Numerical methods may also introduce dispersion, diffusion, instability or other artefacts, particularly when equations contain shocks, discontinuities or strongly nonlinear behaviour.
Computational cost can limit parameter exploration, while chaotic systems may exhibit strong sensitivity to initial conditions that restrict long-term predictability. Nevertheless, computational approaches allow researchers to investigate extreme temperatures, pressures, densities, timescales and spatial scales that can be difficult to reproduce experimentally. They also permit controlled variation of individual parameters and systematic exploration of theoretical scenarios. Computational Physics is therefore best understood as a complementary mode of physical investigation in which mathematical models become executable experiments subject to numerical and physical scrutiny.
