1. The short version, what to know first, and where this fits in the series

• Molecular dynamics (MD) treats atoms as balls and moves them in very short steps with Newton's equation of motion (force = mass × acceleration). In LAMMPS, a widely used code, the default step is 1 fs (femtosecond, 10⁻¹⁵ s), written as 1.0 fs or 0.001 ps depending on the unit system Sourced[Ref. 9]
• How good the result is depends almost entirely on how the forces between atoms are set (the force field). Running 1 ms (millisecond) takes 10¹² steps of 1 fs Our calculation. Time and size both have clear limits
• MD tells you things that come from atomic motion: how easily atoms diffuse, how heat flows, how a material deforms, how a structure changes. But it only means something once compared with measurement. MD is not "AI inventing materials"; it is a tool for observing atomic motion by computation (commentary)

2. A feel for it: atoms in slow motion

Imagine a pool table filmed in extreme slow motion. Frame by frame you can see which way each ball goes and how far. MD does this with atoms. Put atoms in a box, compute "what force acts on each atom right now", move each atom a tiny bit with that force, and compute the forces again. Repeat this over and over and you build a "movie" of atomic motion.

In this movie the position of every atom is known at every time. From it you can calculate quantities linked to material properties: how far atoms wandered (diffusion), how the arrangement of atoms changed, and how much force was needed to stretch the box.

View original SVG, values and sources

Original SVG reference: Concept diagram (vector drawing). One MD step. (1) Find the forces from the current positions; (2) get accelerations from Newton's equation F = m a; (3) update velocities and positions over a very short time Δt; (4) record and return to step 1.

The rule for advancing the equation of motion a little at a time is called an integrator. The most standard is the Verlet method, which derives from the method used in Verlet's 1967 paper; the default method in LAMMPS is its velocity form (velocity Verlet) Sourced[Refs. 7, 8]. The idea of MD itself is old. Alder and Wainwright (1957) studied the solid-fluid transition of a hard-sphere system, Rahman (1964) computed the motion of atoms in liquid argon, and McCammon, Gelin and Karplus (1977) computed the motion of a protein (BPTI) Sourced[Refs. 1, 2, 3].

3. Where the force comes from: an energy slope

So how is "the force on each atom right now" decided? The idea is the same as a ball on a slope. A ball feels a force down the slope. An atom feels a force down an energy (potential energy) slope. In other words, the force is the slope of the energy with its sign reversed. The function that gives the energy for any arrangement of atoms is called the potential.

The simplest example is the force between two atoms. They push each other strongly when too close and pull weakly when apart, and in between lies a "balance" distance where the energy is lowest. A simple formula for this shape is the Lennard-Jones potential (the formula originates with Jones, 1924) Sourced[Ref. 12].

• Close together: at r = σ the energy is U = 0, and closer still it rises steeply. The force pushes the atoms apart and grows large (24ε/σ at r = σ) Our calculation
• The balance distance: at r ≈ 1.12σ (2^(1/6) × σ) the energy is at its lowest, −ε, and the force is zero Our calculation
• Far apart: the attraction weakens. At r = 1.5σ the force is −1.16 (pulling together); at r = 2.5σ, U is −0.016ε, about 1.6% of the well depth Our calculation. MD often drops the force from such distant atoms (a cutoff) to save time. Our small calculation below also cuts off at 2.5σ

4. The time step: why about one femtosecond?

How short should the "very short time Δt" in Fig. 2 be? A larger value covers more time for the same effort. But if it is too large, the calculation goes wrong.

The reason is that atoms vibrate very fast. For example, a typical textbook value for the C–H stretch is about 3000 cm⁻¹. The period is 1/(speed of light × wavenumber), about 11 fs Our calculation. You need roughly ten frames per period to follow such a vibration. A step of about 1 fs is matched to these fast vibrations. The LAMMPS defaults are 1.0 fs in real units and 0.001 ps in metal units Sourced[Ref. 9].

• A fine step: with Δt = T₀/50 (T₀ is the vibration period), the energy error stayed within 0.22% over 200 periods Our calculation. Halving the step cuts the error to about a quarter (0.55×10⁻³ → 0.14×10⁻³) Our calculation
• A coarse step: with Δt = T₀/8 the error reached 13%, and for Δt = T₀/6 or more the calculation broke down Our calculation
• Differences between methods: velocity Verlet keeps swinging within a fixed band and does not grow with time, even when the error is large. By contrast, the most naive update (plain Euler) saw the energy shoot up within a few periods even at the same Δt = T₀/50 Our calculation

Fixing the lengths of bonds that contain hydrogen (SHAKE and similar) removes the fastest vibrations from the calculation, which can allow a longer step Sourced[Ref. 18]. The price is that those vibrations can then no longer be represented (commentary).

5. The role and types of force fields

The quality of an MD result is decided by how the forces are set. The way of setting them (the formulas for the energy between atoms, plus their constants) is called a force field. With the same MD software, a different force field gives a completely different result.

A document published by the European standards body CEN in 2018 (CEN Workshop Agreement 17284) lays out this relationship. The equation of motion is a "physics equation" that is common to all materials. A force field is a "materials relation", specific to a material, valid only for the class of materials it was formulated for Sourced[Ref. 34]. The basic attitude is therefore that a force field cannot be trusted outside the conditions it was built for: combinations of atoms, temperature, pressure, structure (commentary). Note that this document is not an official standard but a consensus document from a workshop [Ref. 34].

Check the original papers for the details of each formula. This article organises them within the range of each paper's title and content.

There are also public places to help you choose a force field. The NIST Interatomic Potentials Repository provides potentials (force fields), related files and evaluation tools for judging their quality and applicability Sourced[Ref. 30]. OpenKIM is a curated repository of interatomic potentials, including machine-learning ones Sourced[Ref. 31]. A paper on choosing and using force fields also exists [Ref. 29].

6. Holding temperature and pressure, and the small-box trick

Real experiments are done "at room temperature" and "at atmospheric pressure". In MD too, you often want to hold the temperature or pressure of the box constant. The combination of what is held constant is called an ensemble. Holding particle number N, volume V and energy E gives NVE (an isolated system); holding N, V and temperature T gives NVT; holding N, pressure P and T gives NPT (commentary) [Ref. 6].

• Holding temperature: a thermostat couples the system to a "heat bath" and adjusts atomic velocities. Representative methods are those of Nosé (1984) and Hoover (1985) Sourced[Refs. 19, 20]
• Holding pressure: a barostat changes the box size a little at a time (commentary) [Ref. 6]
• Repeating a small box: the number of atoms we can compute is thousands to millions, a far smaller lump than a real material. So the opposite faces of the box are joined and the same box is treated as repeated forever (periodic boundary conditions), which removes the effect of surfaces. But if the box is small, the result depends on the box size (commentary) [Ref. 6]

7. Limits in time and size

The biggest weakness of MD is that the time and size it can reach are limited. With 1 fs steps, reaching 1 ns (nanosecond) takes 10⁶ steps, 1 μs (microsecond) takes 10⁹, 1 ms takes 10¹², and 1 s takes 10¹⁵ Our calculation. The same goes for size: 1 mm³ of iron contains about 8.5×10¹⁹ atoms Our calculation.

• A record in time: a special-purpose machine reaches far longer times. Shaw et al. (2010) report an all-atom MD run (classical force field) of the protein BPTI lasting 1 millisecond on a special-purpose machine Sourced[Ref. 21]. With 1 fs steps that corresponds to about 10¹² steps (our conversion) Our calculation
• A record in size: Germann and Kadau (2008) report MD of one trillion atoms (Lennard-Jones particles) on BlueGene/L (212,992 processors) Sourced[Ref. 22]. But one trillion atoms of iron fill a cube only about 2.3 μm on a side Our calculation
• The gap to times that matter in practice: phenomena that proceed over seconds to years, such as diffusion, creep, corrosion and fatigue, cannot be reached directly with ordinary MD (commentary). Very rare events (an atomic hop, nucleation) may never occur even once during the simulation

To get past this wall there are methods that accelerate rare events. For example, hyperdynamics (Voter, 1997) speeds up the progress of infrequent events, and metadynamics (Laio and Parrinello, 2002) "fills" energy valleys so that the system escapes more easily and the shape of the free energy can be studied Sourced[Refs. 23, 24]. But the choices made by the user, such as which variables to accelerate, affect the result (commentary).

8. What can MD tell us?

What MD produces is the time course of the positions and velocities of all atoms. From it you can calculate the quantities below. Here we look at the difference between a liquid and a solid with a simple calculation of our own (256 particles).

• How atoms are arranged (radial distribution function): it shows how many other atoms are found at distance r from a given atom. In the solid the peaks are tall and the order extends far (tallest peak about 5.9). In the liquid only the nearby peaks remain and the curve approaches 1 farther out (tallest peak about 3.0) Our calculation
• How easily atoms move (mean squared displacement and diffusion coefficient): the average of the squared distance moved. In the liquid it grows in proportion to time. In three dimensions, the textbook relation is "diffusion coefficient = slope ÷ 6" [Ref. 6]. In our simple calculation this gave about 0.026 for the liquid (reduced units). In the solid it hardly grows (about 0.011 after 5000 steps) Our calculation. These are rough values from a short run of a small system, with no error analysis

The Bulatov et al. example is a case where a phenomenon predicted by calculation was confirmed by experiment. Many MD results have no direct comparison with experiment and remain computational predictions (commentary).

9. How a study proceeds, and where to check

An MD study is not just running a calculation; it involves a lot of checking. LAMMPS (Thompson et al., 2022) and GROMACS (Abraham et al., 2015) are described in papers as widely used simulation software Sourced[Refs. 10, 11]. With the software you can compute, but whether the result can be trusted depends on the checks made by the user.

• A guide for beginners: a document with a checklist, summarising what people starting molecular simulation should know before and during their first simulations, is publicly available (Braun et al.) Sourced[Ref. 32]
• Checking the statistics: a guideline holds that assessing uncertainty and sampling quality (whether the run was long enough) is essential in molecular simulation (Grossfield et al.). It states explicitly that it does not cover systematic errors such as an inaccurate force field Sourced[Ref. 33]. In other words, small error bars do not mean the force field is right
• Recording the calculation: the CEN document gives a template (MODA) for documenting a simulation, with items such as the use case, model, time step and software Sourced[Ref. 34]

10. A materials engineer's view

11. Limits, what we could not confirm, and a summary

What we could not confirm

• Individual statements in the textbook by Allen and Tildesley (when the classical approximation becomes rough, ensembles, periodic boundary conditions, the relation for the diffusion coefficient) were not checked against the text of the book in the work for this article. They are written as general accounts; please check the original where it matters
• The wavenumber of the C–H bond (about 3000 cm⁻¹) is a textbook rule of thumb, and we did not check individual literature. The period is a conversion from this value
• The results of the simple MD (peak heights of g(r), the diffusion coefficient and so on) are rough values from a short run of a small system; we did not compare them with literature values, evaluate the error or check the dependence on box size
• We did not confirm, even in general terms, which force field has what accuracy for real materials (your alloys and polymers)
• We could not find an international standard that defines MD force fields or procedures

12. How to read the yardsticks and standards

The yardsticks for MD are, first, whether the calculation is numerically sound (energy conservation, whether equilibrium was reached); second, statistical reliability (whether the run was long enough, whether repeats agree); third, validity of the force field (does it reproduce known measurements); and finally, whether it can be compared with measurement in the same quantity. Checked on 22 October 2026.

The organisation of the table is Our calculation (our arrangement). Numbers, titles and status were checked on official pages and in the original papers (checked on 22 October 2026). The CWA text is publicly available, and we confirmed its title, scope and term definitions. We did not confirm any applicable ASTM standard within the scope of this article.

Reading notes (for materials engineers)

• "Small error bars" and "correct" are different things: statistical error shrinks if you run longer, but the systematic error from a wrong force field remains [Ref. 33]
• Rankings of public force fields do not carry over to your material: each force field is tuned to reproduce the properties of its target. First test it against simple measured values for your own material
• Keep computational predictions apart from results confirmed by experiment: many MD results are computational predictions with no direct comparison with experiment. A case confirmed by experiment (Bulatov et al.) states this explicitly [Ref. 26]
• Are standards and guidance recommended or required? The guidance and the CWA cited here are recommendations, a terminology and a documentation format, not standards with legal force [Ref. 34]

13. Glossary

14. References

1. Alder B. J., Wainwright T. E. "Phase Transition for a Hard Sphere System", J. Chem. Phys. 27, 1208–1209 (1957), DOI 10.1063/1.1743957 — doi.org/10.1063/1.1743957
2. Rahman A. "Correlations in the Motion of Atoms in Liquid Argon", Phys. Rev. 136, A405–A411 (1964), DOI 10.1103/PhysRev.136.A405 — doi.org/10.1103/PhysRev.136.A405
3. McCammon J. A., Gelin B. R., Karplus M. "Dynamics of folded proteins", Nature 267, 585–590 (1977), DOI 10.1038/267585a0 — doi.org/10.1038/267585a0
4. Karplus M., McCammon J. A. "Molecular dynamics simulations of biomolecules", Nat. Struct. Biol. 9, 646–652 (2002), DOI 10.1038/nsb0902-646 — doi.org/10.1038/nsb0902-646
5. The Royal Swedish Academy of Sciences "The Nobel Prize in Chemistry 2013, press release (9 October 2013)" — www.nobelprize.org/prizes/chemistry/2013/press-release/
6. Allen M. P., Tildesley D. J. "Computer Simulation of Liquids, 2nd edition", Oxford University Press (2017), DOI 10.1093/oso/9780198803195.001.0001 — doi.org/10.1093/oso/9780198803195.001.0001
7. Verlet L. "Computer "Experiments" on Classical Fluids. I. Thermodynamical Properties of Lennard-Jones Molecules", Phys. Rev. 159, 98–103 (1967), DOI 10.1103/PhysRev.159.98 — doi.org/10.1103/PhysRev.159.98
8. LAMMPS documentation "run_style command (verlet = velocity form of the Stoermer-Verlet algorithm)" — docs.lammps.org/run_style.html
9. LAMMPS documentation "timestep command (default 0.001 ps for metal units, 1.0 fs for real units)" — docs.lammps.org/timestep.html
10. Thompson A. P. et al. "LAMMPS - a flexible simulation tool for particle-based materials modeling at the atomic, meso, and continuum scales", Comput. Phys. Commun. 271, 108171 (2022), DOI 10.1016/j.cpc.2021.108171 — doi.org/10.1016/j.cpc.2021.108171
11. Abraham M. J. et al. "GROMACS: High performance molecular simulations through multi-level parallelism from laptops to supercomputers", SoftwareX 1–2, 19–25 (2015), DOI 10.1016/j.softx.2015.06.001 — doi.org/10.1016/j.softx.2015.06.001
12. Jones J. E. "On the determination of molecular fields. II. From the equation of state of a gas", Proc. R. Soc. Lond. A 106, 463–477 (1924), DOI 10.1098/rspa.1924.0082 — doi.org/10.1098/rspa.1924.0082
13. Daw M. S., Baskes M. I. "Embedded-atom method: Derivation and application to impurities, surfaces, and other defects in metals", Phys. Rev. B 29, 6443–6453 (1984), DOI 10.1103/PhysRevB.29.6443 — doi.org/10.1103/PhysRevB.29.6443
14. Stillinger F. H., Weber T. A. "Computer simulation of local order in condensed phases of silicon", Phys. Rev. B 31, 5262–5271 (1985), DOI 10.1103/PhysRevB.31.5262 — doi.org/10.1103/PhysRevB.31.5262
15. Tersoff J. "New empirical approach for the structure and energy of covalent systems", Phys. Rev. B 37, 6991–7000 (1988), DOI 10.1103/PhysRevB.37.6991 — doi.org/10.1103/PhysRevB.37.6991
16. van Duin A. C. T. et al. "ReaxFF: A Reactive Force Field for Hydrocarbons", J. Phys. Chem. A 105, 9396–9409 (2001), DOI 10.1021/jp004368u — doi.org/10.1021/jp004368u
17. Car R., Parrinello M. "Unified Approach for Molecular Dynamics and Density-Functional Theory", Phys. Rev. Lett. 55, 2471–2474 (1985), DOI 10.1103/PhysRevLett.55.2471 — doi.org/10.1103/PhysRevLett.55.2471
18. Ryckaert J.-P., Ciccotti G., Berendsen H. J. C. "Numerical integration of the cartesian equations of motion of a system with constraints: molecular dynamics of n-alkanes", J. Comput. Phys. 23, 327–341 (1977), DOI 10.1016/0021-9991(77)90098-5 — doi.org/10.1016/0021-9991(77)90098-5
19. Nosé S. "A unified formulation of the constant temperature molecular dynamics methods", J. Chem. Phys. 81, 511–519 (1984), DOI 10.1063/1.447334 — doi.org/10.1063/1.447334
20. Hoover W. G. "Canonical dynamics: Equilibrium phase-space distributions", Phys. Rev. A 31, 1695–1697 (1985), DOI 10.1103/PhysRevA.31.1695 — doi.org/10.1103/PhysRevA.31.1695
21. Shaw D. E. et al. "Atomic-Level Characterization of the Structural Dynamics of Proteins", Science 330, 341–346 (2010), DOI 10.1126/science.1187409 — doi.org/10.1126/science.1187409
22. Germann T. C., Kadau K. "Trillion-atom molecular dynamics becomes a reality", Int. J. Mod. Phys. C 19, 1315–1319 (2008), DOI 10.1142/S0129183108012911 — doi.org/10.1142/S0129183108012911
23. Voter A. F. "Hyperdynamics: Accelerated Molecular Dynamics of Infrequent Events", Phys. Rev. Lett. 78, 3908–3911 (1997), DOI 10.1103/PhysRevLett.78.3908 — doi.org/10.1103/PhysRevLett.78.3908
24. Laio A., Parrinello M. "Escaping free-energy minima", Proc. Natl. Acad. Sci. USA 99, 12562–12566 (2002), DOI 10.1073/pnas.202427399 — doi.org/10.1073/pnas.202427399
25. Schelling P. K., Phillpot S. R., Keblinski P. "Comparison of atomic-level simulation methods for computing thermal conductivity", Phys. Rev. B 65, 144306 (2002), DOI 10.1103/PhysRevB.65.144306 — doi.org/10.1103/PhysRevB.65.144306
26. Bulatov V. V. et al. "Dislocation multi-junctions and strain hardening", Nature 440, 1174–1178 (2006), DOI 10.1038/nature04658 — doi.org/10.1038/nature04658
27. He X., Zhu Y., Mo Y. "Origin of fast ion diffusion in super-ionic conductors", Nat. Commun. 8, 15893 (2017), DOI 10.1038/ncomms15893 — doi.org/10.1038/ncomms15893
28. Kremer K., Grest G. S. "Dynamics of entangled linear polymer melts: A molecular-dynamics simulation", J. Chem. Phys. 92, 5057–5086 (1990), DOI 10.1063/1.458541 — doi.org/10.1063/1.458541
29. Becker C. A., Tavazza F., Trautt Z. T., Buarque de Macedo R. A. "Considerations for choosing and using force fields and interatomic potentials in materials science and engineering", Curr. Opin. Solid State Mater. Sci. 17, 277–283 (2013), DOI 10.1016/j.cossms.2013.10.001 — doi.org/10.1016/j.cossms.2013.10.001
30. NIST (National Institute of Standards and Technology) "Interatomic Potentials Repository" — www.ctcms.nist.gov/potentials/
31. OpenKIM "Knowledgebase of Interatomic Models (OpenKIM)" — openkim.org/
32. Braun E. et al. "Best Practices for Foundations in Molecular Simulations [Article v1.0]", Living J. Comput. Mol. Sci. 1, 5957, DOI 10.33011/livecoms.1.1.5957 — doi.org/10.33011/livecoms.1.1.5957
33. Grossfield A. et al. "Best Practices for Quantification of Uncertainty and Sampling Quality in Molecular Simulations [Article v1.0]", Living J. Comput. Mol. Sci. 1, 5067, DOI 10.33011/livecoms.1.1.5067 — doi.org/10.33011/livecoms.1.1.5067
34. CEN (European Committee for Standardization) "CEN Workshop Agreement CWA 17284:2018, Materials modelling - Terminology, classification and metadata" (2018) — www.cencenelec.eu/media/CEN-CENELEC/CWAs/RI/cwa17284_2018.pdf

15. Claim-to-source audit

Last updated: 22 October 2026. Sources are limited to primary material (original papers, official software documentation, and official NIST, OpenKIM, CEN and Nobel Prize pages). Numerical examples are our own calculations, not measured values of any specific substance. We could not confirm individual statements in the Allen and Tildesley textbook, the C–H stretching wavenumber, comparisons of the simple MD with literature values, the accuracy of force fields for real materials, or any international standard that defines MD force fields or procedures, so we do not assert them. Figs. 2 to 8 are illustrative concept diagrams or drawings that include our calculations; Figs. 1 and 9 are AI-generated images (only the scene to be drawn is specified).