- Atlas
- Natural Sciences
- Computational Materials Science
Computational Materials Science
Computational materials science is the branch of materials science that studies materials by simulating them on computers rather than, or before, making them in a laboratory.
11 frameworks, from 1928 to today · Materials Science · Natural Sciences
- Electronic structure theory 1928-Present
- Molecular dynamics 1957-Present
- Empirical interatomic potentials 1960-Present
- Density Functional Theory 1985-Present
- Multiscale modeling 1985-Present
- Dislocation dynamics 1990-Present
- Phase-field model 1990-Present
- Integrated Computational Materials Engineering 2008-Present
- Machine learning for materials 2010-Present
- High-throughput computational screening 2011-Present
- Machine-learned interatomic potentials 2015-Present
Overview
Computational materials science is the branch of materials science that studies materials by simulating them on computers rather than, or before, making them in a laboratory. Its practitioners build mathematical models of how atoms, electrons, and microstructures behave, solve those models numerically, and use the results to explain why a material has the properties it does or to predict a material that has not yet been made1. The stakes are practical: an experiment can take months and cost a great deal, while a well-chosen calculation can screen thousands of candidate compounds in the time it takes to synthesize one. The field's history is a story of successive attempts to bridge the enormous gap between the quantum behaviour of electrons, which can be calculated but only for small systems, and the engineering behaviour of a real component, which involves billions of atoms and many length scales.
The quantum foundation: band theory and the electron problem
The intellectual root of the field is the quantum theory of solids developed in the late 1920s and 1930s. Felix Bloch's theorem (1928) showed that electrons in a periodic crystal lattice have wavefunctions of a particular form, and the resulting band structure—the pattern of allowed electron energies—explained why some materials conduct and others insulate. This was a triumph of theory, but solving the equations for a real material was another matter. The full problem involves every electron interacting with every other electron, an intractable many-body problem2.
The first practical route was the nearly-free-electron and tight-binding approximations, which replaced the full interaction with simplified models. These gave qualitative insight but could not predict, say, the exact lattice constant of a metal. A more systematic approach came from density functional theory (DFT), formulated by Pierre Hohenberg and Walter Kohn in 1964 and given its practical computational form by Kohn and Lu Jeu Sham in 1965. DFT's central claim is that all ground-state properties of a system of interacting electrons are determined by the electron density alone—a function of three spatial coordinates rather than the vastly more complicated many-electron wavefunction2. The catch was that the exact energy functional is unknown, so practitioners had to approximate it12. The local density approximation (LDA) and later the generalized gradient approximation (GGA) made DFT computable, and by the 1980s and 1990s it became the workhorse of electronic structure theory in materials science. A concrete early demonstration was the calculation of the structural properties of silicon and other semiconductors, which matched experiment well enough to convince the community that first-principles calculation could be predictive. The term first-principles or ab initio means the calculation uses only the atomic numbers and positions as input, with no empirical fitting to experimental data.
Molecular dynamics: atoms in motion
DFT gives static ground-state properties, but materials at finite temperature vibrate, diffuse, and melt. Molecular dynamics (MD) addresses this by integrating Newton's equations of motion for every atom in a system over small time steps, producing a trajectory from which thermodynamic and transport properties can be extracted. The method dates to the late 1950s, when Berni Alder and Thomas Wainwright simulated hard-sphere systems, and was applied to realistic materials in the 1960s and 1970s. The limitation was the interatomic potential: the force between atoms had to be supplied1. Early empirical interatomic potentials—such as the Lennard-Jones potential for noble gases or the embedded-atom method developed by Murray Daw and Mike Baskes in 1984 for metals—were fitted to experimental data and worked well for specific classes of material but could not be transferred reliably to new systems. The tension between the accuracy of first-principles methods and the reach of empirical ones became a defining theme of the field3.
The multiscale problem and its solutions
By the 1980s and 1990s, practitioners recognized that no single method could span the range from electrons to engineering components. Multiscale modeling emerged as a framework for linking methods at different length and time scales: quantum calculations feed interatomic potentials, which feed molecular dynamics, which feed continuum models of microstructure1. Each link required its own approximations, and the field developed a vocabulary of hierarchical (one-way) and concurrent (simultaneous) coupling1.
Two microstructure-scale methods became central. Phase-field modeling, developed from the 1970s onward and applied widely to materials in the 1990s, represents the boundary between phases or grains as a smooth, diffuse interface rather than a sharp one. A single continuous field variable describes which phase is present, and its evolution is governed by equations that minimize free energy14. This made it possible to simulate complex microstructural evolution—dendritic solidification, grain growth, spinodal decomposition—in two and three dimensions. Dislocation dynamics tackles plastic deformation directly: rather than simulating every atom, it tracks the motion and interaction of dislocation lines, the line defects whose movement carries plastic strain in crystals1. Discrete dislocation dynamics simulations, developed from the 1990s, could model the collective behaviour of thousands of dislocations and connect to continuum plasticity.
Integrated Computational Materials Engineering
The idea that these tools could be combined into a design workflow rather than used piecemeal crystallized in the 2000s as Integrated Computational Materials Engineering (ICME). The term was formalized in a 2008 report by the U.S. National Research Council, which argued that computational tools should be linked across scales to accelerate the design of materials and components. ICME's promise was to let an engineer specify a performance target and work backward through processing, microstructure, and properties to a material composition—the inverse of the traditional trial-and-error approach. The concept drew on earlier integrated design efforts in the aerospace and automotive industries and became a organizing principle for funding and research programs in the 2010s.
High-throughput screening and the data revolution
A parallel development was high-throughput computational screening, which uses automated workflows to calculate properties for thousands of candidate compounds. The approach was enabled by the growth of computing power and by databases such as the Materials Project, launched in 2011 by Gerbrand Ceder and Kristin Persson, which made DFT results for tens of thousands of materials publicly available. The logic is simple: if a calculation is cheap enough, run it on every plausible composition and filter for the ones that meet a target. This has been most successful in areas like battery electrode discovery and thermoelectric materials, where the relevant properties can be computed from first principles with reasonable accuracy. The limitation is that DFT's accuracy depends on the approximation used, and many properties of engineering interest—fracture toughness, fatigue life—remain beyond direct first-principles prediction.
Machine learning enters the toolkit
The most recent turning point is the use of machine learning to bridge the accuracy-reach gap3. Machine-learned interatomic potentials train a flexible statistical model on a large set of quantum calculations, then use that model to run molecular dynamics at a fraction of the cost. Early work in the 2010s used neural networks and Gaussian process regression; by the 2020s, models such as those based on graph neural networks could reproduce DFT energies and forces with sufficient accuracy for many systems while scaling to millions of atoms. The same data-driven logic has spread to property prediction, where models trained on databases can estimate quantities that are expensive or impossible to compute directly. The field now debates how to make these models transferable, how to quantify their uncertainty, and how to integrate them with physical constraints so that they extrapolate sensibly beyond their training data.
The present
Today computational materials science is a mature but still rapidly changing field. DFT remains the backbone for electronic structure and thermodynamic stability, with ongoing work on better exchange-correlation functionals and on methods that go beyond them, such as dynamical mean-field theory for strongly correlated electrons. Molecular dynamics with machine-learned potentials is increasingly the default for finite-temperature simulation. Phase-field and dislocation dynamics remain the standard tools at the microstructure scale, and ICME frameworks attempt to link them. High-throughput databases and machine learning have changed the economics of materials discovery, making it possible to screen candidates computationally before any synthesis is attempted. The open questions are substantial: how to predict properties that depend on rare events or long time scales, how to make multiscale coupling rigorous rather than ad hoc, how to validate computational predictions against experiment at scale, and how to build models that are both accurate and interpretable enough to guide human intuition. The field's history is one of steadily expanding the range of materials phenomena that can be calculated rather than measured—and of learning, at each step, where the calculations can be trusted and where they cannot.
References
- Computational materials science (introduction, Introduction, Density functional theory, Molecular dynamics, Multi-scale simulation, Concurrent multi-scale, Hierarchical multi-scale, Model development, Phase field, Dislocation dynamics), Wikipedia
- Density functional theory (Origins), Wikipedia
- Machine-learned interatomic potential, Wikipedia
- Phase-field model, Wikipedia
The frameworks of Computational Materials Science
Each framework is a way of working the field has used: when it held, and the question it answers.
- 1928-Present
Electronic structure theory
How do electrons in a crystal arrange their energies?
- 1957-Present
- 1960-Present
Empirical interatomic potentials
What force law acts between atoms in a material?
- 1985-Present
Density Functional Theory
Can electron density alone give all ground-state properties?
- 1985-Present
Multiscale modeling
How can models at different length and time scales be linked?
- 1990-Present
Dislocation dynamics
How do moving dislocation lines produce plastic deformation?
- 1990-Present
Phase-field model
How do phase boundaries and microstructures evolve over time?
- 2008-Present
Integrated Computational Materials Engineering
How can computation drive engineering design of materials?
- 2010-Present
Machine learning for materials
Can statistical models predict material properties from data?
- 2011-Present
High-throughput computational screening
Which of thousands of candidate compounds are worth making?
- 2015-Present
Machine-learned interatomic potentials
Can learned models replace fitted force laws between atoms?
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.
Density Functional Theory grew out of Electronic structure theory
Density functional theory was formulated by Hohenberg and Kohn (1964) and Kohn and Sham (1965) as a reformulation of the many-electron problem that band theory had posed but could not solve. It became the practical workhorse of electronic structure theory from the 1980s onward.
Carried by Pierre Hohenberg; Walter Kohn; Lu Jeu Sham.
Machine-learned interatomic potentials grew out of Empirical interatomic potentials
Machine-learned interatomic potentials are the data-driven successor to empirical potentials, replacing fitted analytic forms with flexible statistical models trained on quantum data. They address the same accuracy-transferability problem that empirical potentials posed.
High-throughput computational screening influenced by Density Functional Theory
High-throughput screening automated DFT calculations across thousands of compounds, and databases such as the Materials Project (2011) made the results public. The screening logic depends on DFT being cheap and accurate enough to run at scale.
Carried by Gerbrand Ceder; Kristin Persson; Materials Project (2011).
Machine-learned interatomic potentials influenced by Density Functional Theory
Machine-learned interatomic potentials are trained on large sets of DFT energies and forces, then used to run molecular dynamics at a fraction of the cost. DFT supplies the reference data that makes the learned potentials accurate.
Integrated Computational Materials Engineering grew out of Multiscale modeling
Integrated Computational Materials Engineering grew out of multiscale modeling, turning hierarchical coupling into an engineering design workflow. The 2008 National Research Council report formalized the idea that linked computational tools should span processing, microstructure and properties.
Carried by National Research Council; Integrated Computational Materials Engineering (2008).
Empirical interatomic potentials influenced by Electronic structure theory
Electronic structure calculations supplied the cohesive and defect energetics that empirical potentials such as the embedded-atom method (Daw and Baskes, 1984) were designed to reproduce. The potentials were fitted to and validated against first-principles and experimental data rooted in band theory.
Carried by Murray Daw; Mike Baskes.
Multiscale modeling influenced by Molecular dynamics
Molecular dynamics supplied the atomistic tier of multiscale hierarchies, in which MD trajectories feed continuum and microstructure models. The hierarchical and concurrent coupling schemes of multiscale modeling were built around MD as the atomistic engine.
Machine-learned interatomic potentials influenced by Molecular dynamics
Machine-learned potentials are used to drive molecular dynamics, inheriting MD's integrators, ensembles and analysis tools. The learned model substitutes for the empirical force law inside the same simulation framework.
Beyond the field
Where Computational Materials Science's ideas came from in other fields, and what they led to there: the strongest 12 of 62 links.
Electronic structure theory led to First-Principles Electronic-Structure Prediction in Electronic And Magnetic Materials
First-principles electronic-structure prediction is a branch of electronic structure theory, applying its quantum-mechanical framework to predict material properties from the Hamiltonian.
Electronic structure theory grew out of Hartree–Fock method in Atomic Molecular Physics
Electronic structure theory is the broader programme built on the Hartree-Fock self-consistent-field method, which became its standard starting approximation for atoms and molecules.
Carried by Douglas Hartree; Vladimir Fock.
Electronic structure theory influenced Periodic DFT and computational materials chemistry in Computational Chemistry
Electronic structure theory, from Bloch's theorem and band theory onward, supplied the conceptual and computational apparatus that periodic DFT implementations of materials chemistry use.
Carried by Felix Bloch.
Molecular dynamics influenced Computational materials science in Materials Science
Molecular dynamics, introduced by Alder and Wainwright in the 1950s, supplied the simulation method that computational materials science uses to model atomic motion, diffusion and mechanical response.
Carried by Berni Alder; Tom Wainwright.
Empirical interatomic potentials influenced Machine-Learned Interatomic Potentials in Theoretical Chemistry
Machine-learned potentials took over the role of empirical interatomic potentials, fitting a flexible functional form to reference data in place of hand-parameterised Lennard-Jones or embedded-atom forms.
Carried by Jörg Behler; Michele Parrinello.
Empirical interatomic potentials influenced Machine-learned interatomic potential in Computational Chemistry
Machine-learned potentials were built as replacements for empirical potentials, taking over their task of cheaply evaluating energies and forces for large atomistic systems. The early neural-network potential work of Behler and Parrinello explicitly framed itself against empirical potentials such as embedded-atom and bond-order schemes.
Carried by Jörg Behler; Michele Parrinello.
Empirical interatomic potentials influenced by Molecular mechanics in Theoretical Chemistry
Molecular mechanics force fields, with their functional forms for bond stretching, bending and nonbonded interactions, were transferred directly into empirical interatomic potentials for solids and molecules.
Dislocation dynamics influenced by Dislocation in Materials Science
The materials-science conception of dislocations as line defects carrying plastic deformation is the object that dislocation dynamics simulates.
Dislocation dynamics influenced by Dislocation Theory and Crystal Plasticity in Solid Mechanics
Dislocation theory and crystal plasticity, from Taylor, Orowan and Polanyi in 1934 onward, supplied the line-defect concepts, glide rules and elastic interactions that dislocation dynamics simulates explicitly.
Carried by Geoffrey Ingram Taylor; Egon Orowan; Michael Polanyi.
Dislocation dynamics influenced by Dislocation Theory in Mechanical Behavior Of Materials
Dislocation theory provided the equations of motion, line tension and interaction laws that dislocation dynamics discretises and integrates numerically.
High-throughput computational screening influenced by Periodic DFT and computational materials chemistry in Computational Chemistry
High-throughput computational screening of materials is built directly on periodic DFT, using standardized density functional calculations as the property engine for screening thousands of compounds. The Materials Project and similar efforts automated periodic DFT workflows to make the screening possible.
Machine-learned interatomic potentials influenced by Periodic DFT and computational materials chemistry in Computational Chemistry
Machine-learned interatomic potentials are trained on periodic DFT energies and forces, using density functional calculations as their reference data. The potentials exist to reproduce DFT accuracy at molecular-mechanics cost.

Computational Materials Science is a planet in Nousfarer
Its city has a tower for each of these 11 frameworks. Land, walk in, and climb one concept at a time.
Land on Computational Materials ScienceRelated fields
Fields its ideas are linked to
- Computational Chemistry 10 links
- Materials Science 7 links
- Mechanical Behavior Of Materials 6 links
- Materials Chemistry 5 links
- Theoretical Chemistry 5 links
- Physical Chemistry 3 links
- Atomic Molecular Physics 3 links
- Transport Phenomena 2 links
- Electronic And Magnetic Materials 2 links
- Materials Thermodynamics And Phase Transformations 2 links
- Computational Physics 2 links
- Condensed Matter 2 links
More in Materials Science
- Biomaterials
- Ceramic Materials
- Composite Materials
- Corrosion And Degradation
- Electronic And Magnetic Materials
- Energy Materials
- Materials Characterization
- Materials Processing
- Materials Science
- Materials Thermodynamics And Phase Transformations
- Mechanical Behavior Of Materials
- Metallurgy
- Nanomaterials
- Polymer Science
- Quantum Materials
- Semiconductor Materials
- Structural Materials
See where Computational Materials Science sits on the Map of Human Knowledge