Computational Chemistry

Computational chemistry is the use of computers to model molecules and chemical processes: to calculate how atoms bond, how a reaction proceeds, how a protein folds around a drug, or how a solid conducts heat.

13 frameworks, from 1930 to today · Chemistry · Natural Sciences

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

Overview

Computational chemistry is the use of computers to model molecules and chemical processes: to calculate how atoms bond, how a reaction proceeds, how a protein folds around a drug, or how a solid conducts heat1. Its practitioners ask the same questions as experimental chemists—what is this substance, why does it react this way, what happens if I change one atom—but answer them with mathematics and machines instead of flasks. What is at stake is prediction: the ability to test a molecule or material before anyone synthesizes it, to explain what no instrument can see, and increasingly to search a vast space of candidates for the one that will work. Its history is a story of borrowing physics, stretching approximations, and waiting for computers to grow powerful enough to catch up with the ideas.

From the chemical bond to quantum chemistry

The intellectual foundation was laid by quantum mechanics in the 1920s1. In 1927, Walter Heitler and Fritz London showed that two hydrogen atoms could be held together by the mathematics of the new theory—the first quantum-mechanical account of a chemical bond12. That same year, the physicists Max Born and J. Robert Oppenheimer argued that atomic nuclei, being far heavier than electrons, move so slowly that one can solve for the electrons with the nuclei held fixed; this Born–Oppenheimer approximation became the organizing assumption of nearly all molecular calculation.

Two rival pictures of bonding then grew up. Valence bond theory, associated above all with Linus Pauling in the 1930s and his 1939 book The Nature of the Chemical Bond, described a bond as the overlap of atomic orbitals, with resonance between structures when no single picture suffices; Pauling's hybrid orbitals explained the shapes of molecules such as methane. Molecular orbital theory, developed by Friedrich Hund, Robert Mulliken, Erich Hückel, and others, instead spread electrons over the whole molecule in orbitals built from atomic ones. Hückel's method (1931) could be worked by hand for flat organic molecules, and it explained why benzene is unusually stable. Valence bond was the dominant language of chemistry textbooks for decades; molecular orbital theory eventually won out in practice because it lent itself to systematic calculation on a machine.

The self-consistent field and its corrections

The first genuinely general computational method was the Hartree–Fock approach, named for Douglas Hartree and Vladimir Fock, who in the late 1920s and 1930s formulated a way to find the best molecular orbitals when each electron moves in the average field of the others. In practice, one guesses the orbitals, computes the field they create, then re-solves for improved orbitals, repeating until input and output agree—a self-consistent field3. In 1951, Clemens Roothaan expressed the equations in a form suitable for linear algebra on digital computers. Hartree–Fock, however, ignores that electrons dodge each other in detail—electron correlation—and so misjudges reaction energies and bond-breaking. The post-Hartree–Fock methods that followed addressed this: configuration interaction, Møller–Plesset perturbation theory (from a 1934 paper by Christian Møller and Milton Plesset), and coupled-cluster theory, developed in the 1960s–70s by Jiří Čížek, Josef Paldus, Rodney Bartlett, and others. These can reach high accuracy but their cost climbs steeply with molecular size.

For molecules too large even for Hartree–Fock, chemists turned to semi-empirical methods, which keep the quantum scaffold but replace the expensive integrals with parameters fitted to experiment14. The lineage runs from the Pariser–Parr–Pople method for π electrons (1950s) and Michael Dewar's MNDO (1970s) to the widely used AM1 (1985). They traded accuracy for reach, and for years they and molecular mechanics carried most practical modeling.

Molecules as springs: force fields and dynamics

A parallel tradition treats atoms as balls and bonds as springs. Molecular mechanics replaces quantum mechanics with a classical energy function—a force field—with terms for bond stretching, bending, and rotation, plus non-bonded attraction and repulsion. The approach goes back to efforts in the 1940s–60s to reproduce molecular shapes and vibrational spectra (for instance the work of John A. Pople and of Norman Allinger, whose MM series of force fields from the 1970s onward was fitted to measured structures and energies). Force fields such as AMBER, CHARMM, and OPLS, developed from the late 1970s, made proteins and polymers tractable.

Molecular dynamics simulation put these force fields in motion: compute the force on every atom, move it a tiny step forward in time by Newton's laws, and repeat millions of times15. This technique had a precursor in the 1950s, when Berni Alder and Thomas Wainwright simulated hard spheres, and in the 1960s Aneesur Rahman simulated liquid argon. In 1977, J. Andrew McCammon, Bruce Gelin, and Martin Karplus ran the first molecular dynamics simulation of a protein, the small bovine pancreatic trypsin inhibitor. Such simulations reveal fluctuations, folding, and how a drug lodges in a binding pocket—things a single static structure cannot show.

Mapping the energy landscape

For a reaction to be predicted, chemists needed not just one structure but the whole potential energy surface—the energy of the molecule as a function of every nuclear position. Chemical rate theory had long linked reaction speed to an energy barrier, especially the transition-state theory developed by Henry Eyring and Michael Polanyi in the 1930s6. Computational methods gave this landscape a usable shape: by the 1970s and 1980s, algorithms could locate energy minima (stable molecules) and saddle points (transition states) on surfaces computed quantum-mechanically. Transition-state theory, applied to computed barriers, let chemists predict which of two reaction pathways is faster and why a catalyst lowers a barrier—the core business of computational mechanistic chemistry.

Density functional theory: the workhorse

The method that reshaped the field came from physics. In the 1960s, Pierre Hohenberg, Walter Kohn, and Lu Jeu Sham showed that all ground-state properties could be obtained from the electron density—a function of only three spatial coordinates rather than the many coordinates of a wavefunction—and gave practical equations to find it. The catch was that one term in those equations, the exchange-correlation functional, has no exact known form and must be approximated7. Early approximations, developed through the 1980s and 1990s, made density functional theory (DFT) accurate enough and cheap enough for everyday chemistry. By the 1990s it had overtaken Hartree–Fock as the default quantum method for molecules, and later John Pople and Walter Kohn shared the 1998 Nobel Prize in Chemistry for their respective contributions to computational methods. DFT's weaknesses—self-interaction error and the uneven reliability of functionals—remain a live research problem.

Multiscale modeling and the drug-discovery turn

No single method spans a whole enzyme. QM/MM (quantum mechanics/molecular mechanics) modeling, whose foundations were laid by Arieh Warshel and Michael Levitt in the 1970s, treats the chemically active region quantum-mechanically and the surrounding protein classically, embedding one in the other8. This hybrid approach made it possible to study reactions inside large biomolecules, and Warshel, Levitt, and Karplus received the 2013 Nobel Prize in Chemistry for multiscale models of complex chemical systems18.

The same decades brought computation into pharmaceutical practice. Structure-based drug design begins with an experimentally determined protein structure, often from X-ray crystallography, and uses computation to propose molecules that fit its binding site9. Molecular docking—scoring how well a candidate molecule nests into a pocket—was pioneered by programs such as Kuntz's DOCK in the 1980s. Landmark successes, such as the design of the HIV protease inhibitor saquinavir and the carbonic anhydrase inhibitor dorzolamide, showed that computation could guide real medicines, though docking's scoring remains approximate and its predictions are best treated as hypotheses.

Two further frontiers matured. Computational materials chemistry, using periodic DFT and related methods to model crystals, surfaces, and defects, became a mainstay of catalysis and battery research—its practitioners, notably Michele Parrinello and Roberto Car with their 1985 Car–Parrinello method, found ways to run quantum dynamics efficiently. Spectroscopic prediction—computing vibrational frequencies, NMR shifts, and electronic spectra to match and interpret experimental data—grew alongside every quantum method, and by the 1990s computed spectra were routinely used to assign structures that experiments alone could not resolve.

The machine-learning era

The newest chapter replaces hand-crafted approximations with statistical learning. Machine-learned interatomic potentials train a flexible model on large sets of quantum-mechanical calculations, then predict energies and forces for new configurations at a fraction of the cost. This allows molecular dynamics of unprecedented scale and accuracy—for example, simulating a complex material with near-quantum fidelity. The approach has matured from early neural-network potentials toward standardized, broadly adopted frameworks such as the Gaussian approximation potential introduced by Albert Bartók and colleagues in 2010, and later universal models trained across many elements. Surrogate models now accelerate drug and materials discovery by predicting properties or guiding searches without running expensive calculations at every step. These methods inherit the field's old tension: they are only as trustworthy as the data and assumptions they are trained on.

Where computation stands now

Today's computational chemist has a layered toolkit. DFT and post-Hartree–Fock wavefunction methods handle electronic structure and reactions; force fields and machine-learned potentials drive dynamics at scales from single molecules to materials; docking and free-energy calculations guide drug design; QM/MM bridges the quantum and classical worlds; and learned models increasingly substitute for slow physics. All of it rests on the lineage traced here—Born and Oppenheimer's fixed nuclei, Hartree's self-consistent field, Pauling's and Mulliken's competing bonding pictures, Eyring's barriers, Kohn's density10321167. The open questions are the field's oldest: how to capture electron correlation reliably at scale, how to make an approximate functional or a learned potential trustworthy outside the cases it was fitted to, and how to keep the chemist's judgment—about mechanism, about what is physically plausible—central as ever-more of the calculation is handed to a model.

References

  1. Computational chemistry (introduction, History, Semi-empirical methods, Molecular dynamics), Wikipedia
  2. Valence bond theory (History), Wikipedia
  3. Hartree–Fock method, Wikipedia
  4. Semi-empirical quantum chemistry method, Wikipedia
  5. Molecular dynamics, Wikipedia
  6. Transition state theory, Wikipedia
  7. Density functional theory (Origins), Wikipedia
  8. QM/MM, Wikipedia
  9. Drug design (Structure-based), Wikipedia
  10. Born–Oppenheimer approximation, Wikipedia
  11. Molecular orbital theory (History), Wikipedia

The frameworks of Computational Chemistry

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

  1. 1930-1975

    Valence bond theory

    How do atomic orbitals overlap to form bonds?

    Fly to its tower in Nousfarer

  2. 1931-Present

    Molecular orbital theory

    How are electrons distributed over a whole molecule?

    Fly to its tower in Nousfarer

  3. 1951-Present
  4. 1953-1995

    Semi-empirical quantum chemistry method

    How can quantum calculations be made fast enough for large molecules?

    Fly to its tower in Nousfarer

  5. 1970-Present

    Molecular mechanics

    Can molecular shape and energy be modeled with classical springs?

    Fly to its tower in Nousfarer

  6. 1970-Present
  7. 1976-Present

    QM/MM

    How can a reacting site be treated quantum-mechanically inside a protein?

    Fly to its tower in Nousfarer

  8. 1977-Present

    Molecular dynamics

    How do molecules move and fluctuate over time?

    Fly to its tower in Nousfarer

  9. 1980-Present

    Spectroscopic prediction

    Can spectra be predicted from molecular structure?

    Fly to its tower in Nousfarer

  10. 1985-Present
  11. 1985-Present
  12. 1990-Present

    Density functional theory

    Can all ground-state properties come from electron density?

    Fly to its tower in Nousfarer

  13. 2010-Present

    Machine-learned interatomic potential

    Can machine learning replace expensive quantum calculations?

    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.

Valence bond theoryMolecular orbital theoryHartree–Fock and post-Hartree–Fock wa…Semi-empirical quantum chemistry meth…Molecular mechanicsPotential energy surface exploration …QM/MMMolecular dynamicsSpectroscopic predictionPeriodic DFT and computational materi…Structure-based drug design and molec…Density functional theoryMachine-learned interatomic potential

Beyond the field

Where Computational Chemistry's ideas came from in other fields, and what they led to there: the strongest 12 of 69 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 Chemistry is a planet in Nousfarer

Its city has a tower for each of these 13 frameworks. Land, walk in, and climb one concept at a time.

Land on Computational Chemistry

Related fields

Fields its ideas are linked to

More in Chemistry

See where Computational Chemistry sits on the Map of Human Knowledge