---
title: "Defence and conflict · Simulations · Simulation Hypothesis"
url: https://simulation-hypothesis.com/simulations/defence-conflict
description: "War games, combat models, military trainers and weapons physics: the origin of many of the methods in this taxonomy."
license: https://creativecommons.org/licenses/by/4.0/
credit: "Ross Allan Dahl, Simulation Hypothesis (simulation-hypothesis.com)"
---

[Simulations](https://simulation-hypothesis.com/simulations)domain 16 of 18

# Defence and conflict

War games, combat models, military trainers and weapons physics: the origin of many of the methods in this taxonomy.

Militaries were among the earliest and biggest users of simulation, and much of the toolkit here started with them. Monte Carlo was developed for weapons physics, operations research for wartime logistics, and flight simulators for pilot training. War gaming is older still. Officers have trained on tabletop games since the early nineteenth century. Defence ministries, universities and research institutes now use the same methods to study strategy, escalation and the human cost of war.

## War games and analysis

### Professional war games

Officers and analysts playing opposing sides of a scenario, move by move, with umpires deciding outcomes. They are used to test plans, not to predict results.

Question it answers: Where does the plan break when a thinking opponent pushes back?

Leaves out: A skilled umpire's judgement replaces missing physics, to the advantage or detriment of validity.

Runs on: no computer. umpires, maps and people; computers only assist scoring Since: 1812 Kriegsspiel, from the elder Reisswitz’s war-game table (1811–12); his son’s rules of 1824 were taken up by Prussian officers [source](https://en.wikipedia.org/wiki/Kriegsspiel) In the simulations tree: [Kriegsspiel](https://simulation-hypothesis.com/trees/simulations?n=soc-kriegs)

[Rules and states](https://simulation-hypothesis.com/simulations/methods#m-rules)floor: Rules

### Political–military crisis games

Teams playing governments through an escalating international crisis, run by universities and research institutes as well as ministries.

Question it answers: How do crises escalate, and what lets leaders step back?

Leaves out: Players usually behave more rationally than real leaders under real pressure.

Runs on: about 10^9 FLOPS rough people role-playing governments; little or no computing Since: 1958 by 1958, early political-military games run by defence research institutes

[Rules and states](https://simulation-hypothesis.com/simulations/methods#m-rules)[Game theory](https://simulation-hypothesis.com/simulations/methods#m-game-theory)floor: Rules

### Historical war games

Past battles and campaigns refought under their original conditions, used by historians and military academies.

Question it answers: Was the outcome likely, or could it easily have gone the other way?

Leaves out: Refighting a battle under known conditions still cannot test truly novel decisions.

Runs on: about 10^11 FLOPS rough a hobbyist or academic desktop, seconds to minutes per run Since: 1913 H. G. Wells published Little Wars, rules for refighting battles with toy soldiers

[Rules and states](https://simulation-hypothesis.com/simulations/methods#m-rules)[Monte Carlo](https://simulation-hypothesis.com/simulations/methods#m-monte-carlo)floor: Bodies

### Combat and campaign models

Forces worn down against each other over time, from Lanchester’s attrition equations to large statistical campaign models.

Question it answers: How does the balance of forces change the course and cost of a campaign?

Leaves out: Real combat effectiveness depends on morale and leadership that the equations do not capture.

Runs on: about 10^12 FLOPS rough a desktop, seconds per run of the equations Since: 1916 Frederick Lanchester published his combat attrition equations

[Differential equations](https://simulation-hypothesis.com/simulations/methods#m-equations)[Monte Carlo](https://simulation-hypothesis.com/simulations/methods#m-monte-carlo)floor: Populations

### Agent-based combat models

Individual units with their own sensors, orders and decisions, interacting on simulated terrain.

Question it answers: How much do communication, morale and local decisions change the outcome?

Leaves out: Simple decision rules for each unit rarely capture real soldiers' judgement.

Runs on: about 10^14 FLOPS rough a research cluster, minutes per battle run Since: 1997 by 1997, early agent-based combat models built for military research institutes

[Agent-based](https://simulation-hypothesis.com/simulations/methods#m-agent-based)floor: Bodies

### Military logistics

Fuel, food, ammunition, spare parts and transport moving through a theatre of operations.

Question it answers: Can the force be supplied, and where will it run short first?

Leaves out: Real-world friction, weather and breakdowns are only partly represented.

Runs on: about 10^13 FLOPS rough a logistics command's server, minutes per scenario Since: 1955 by 1955, early operations-research studies of military supply chains

[Discrete event](https://simulation-hypothesis.com/simulations/methods#m-discrete-event)[System dynamics](https://simulation-hypothesis.com/simulations/methods#m-system-dynamics)floor: Bodies

### Deterrence and arms races

Rival states building and cutting arms in response to each other, as feedback equations or repeated games.

Question it answers: When is an arms race stable, and when does it spiral?

Leaves out: Assumes rational, simplified state behaviour. Real politics is messier.

Runs on: about 10^11 FLOPS rough a desktop, seconds to run the equations or games Since: 1939 Lewis Fry Richardson's mathematical model of arms races

[Game theory](https://simulation-hypothesis.com/simulations/methods#m-game-theory)[Differential equations](https://simulation-hypothesis.com/simulations/methods#m-equations)floor: Populations

### Models of civil conflict

Social science models of how grievance, legitimacy and policing interact to produce unrest or civil war.

Question it answers: Why does violence break out in some places and not in others?

Leaves out: Grievance and legitimacy are hard to quantify, so results are sensitive to assumptions.

Runs on: about 10^13 FLOPS rough a research cluster, minutes per run Since: 2002 Joshua Epstein's agent-based models of civil violence

[Agent-based](https://simulation-hypothesis.com/simulations/methods#m-agent-based)[Monte Carlo](https://simulation-hypothesis.com/simulations/methods#m-monte-carlo)floor: Bodies

### Language-model wargame agents

Language models play national leaders in a turn-based crisis game, choosing diplomatic and military actions.

Question it answers: How do model agents escalate or de-escalate in a simulated crisis?

Leaves out: Model choices depend on prompts and do not match real leaders.

Runs on: about 10^13 FLOPS rough Many model calls per game Since: 2024 Escalation risks from language models in military and diplomatic decision-making (2024) [source](https://arxiv.org/abs/2401.03408)

[Agent-based](https://simulation-hypothesis.com/simulations/methods#m-agent-based)[Learned emulators](https://simulation-hypothesis.com/simulations/methods#m-learned)[Game theory](https://simulation-hypothesis.com/simulations/methods#m-game-theory)floor: Rules

### Swarm combat simulation

Many simple unmanned aircraft or boats with local rules attack or defend, and the simulation tests how a swarm fares against a protected target.

Question it answers: Can a swarm overwhelm a defended target?

Leaves out: Communications, jamming and the rules of engagement are simplified.

Runs on: about 10^9 FLOPS rough thousands of agents on a workstation. Since: 2002 Lin and colleagues, simulation of a swarm of unmanned combat air vehicles. [source](https://doi.org/10.21236/ada403633)

[Agent-based](https://simulation-hypothesis.com/simulations/methods#m-agent-based)[Particles and N-body](https://simulation-hypothesis.com/simulations/methods#m-particles)floor: Populations

### Computer virus and malware epidemics

Computers are nodes that infect their neighbours with a program, with antivirus cures, so that the spread follows epidemic models.

Question it answers: How fast does malware spread, and what patching rate stops it?

Leaves out: Attacker adaptation and real network topologies are simplified.

Runs on: about 10^5 FLOPS rough epidemic equations or network simulations on a laptop. Since: 1991 Kephart and White, directed-graph epidemiological models of computer viruses. [source](https://doi.org/10.1109/risp.1991.130801)

[Network models](https://simulation-hypothesis.com/simulations/methods#m-network)[Differential equations](https://simulation-hypothesis.com/simulations/methods#m-equations)floor: Populations

### Search and rescue drift planning

Thousands of possible drift paths of a lost object or person are simulated with currents and winds to find where to search.

Question it answers: Where is the most likely place to search first?

Leaves out: Currents and wind are uncertain and the object may behave differently from the model.

Runs on: about 10^8 FLOPS rough Monte Carlo drift of many particles on a workstation. Since: 1980 Richardson and Discenza, the Coast Guard computer-assisted search planning system. [source](https://doi.org/10.1002/nav.3800270413)

[Monte Carlo](https://simulation-hypothesis.com/simulations/methods#m-monte-carlo)[Particles and N-body](https://simulation-hypothesis.com/simulations/methods#m-particles)floor: Fields

### Naval salvo exchange models

Missile salvos between ships are modelled by simple equations of offensive power and defensive shot-down capacity, to see which side survives.

Question it answers: Which fleet design wins a missile exchange?

Leaves out: Treats ships as identical units and ignores sensors and tactics.

Runs on: about 10^3 FLOPS rough closed-form equations on a laptop. Since: 1995 Hughes, a salvo model of warships in missile combat used to evaluate their staying power. [source](https://doi.org/10.1002/1520-6750%28199503%2942:2<267::aid-nav3220420209>3.0.co;2-y)

[Differential equations](https://simulation-hypothesis.com/simulations/methods#m-equations)floor: Rules

### Anti-submarine search theory

The chance of finding a hidden submarine is worked out from the paths of searchers, the target's speed and the range at which it can be detected.

Question it answers: Where and how should ships and aircraft search to find a submarine soonest?

Leaves out: Detection range depends on ocean conditions and noise that are not known well, and the target may evade.

Runs on: about 10^6 FLOPS rough seconds to minutes on a laptop. Since: 1956 Koopman's theory of search, kinematic bases, in Operations Research. [source](https://doi.org/10.1287/opre.4.3.324)

[Differential equations](https://simulation-hypothesis.com/simulations/methods#m-equations)[Monte Carlo](https://simulation-hypothesis.com/simulations/methods#m-monte-carlo)[Stochastic processes](https://simulation-hypothesis.com/simulations/methods#m-stochastic)floor: Populations

## Military training

### Military flight and mission simulators

Networked cockpits in which crews fly missions together against computer-generated opponents.

Question it answers: Can the crews work together under pressure in a mission too costly or risky to rehearse for real?

Leaves out: Computer-generated adversaries fight predictably compared to real opponents.

Runs on: about 10^13 FLOPS rough networked simulator computers, real time Since: 1985 by 1985, SIMNET, a DARPA programme that networked combat simulators

[Real time](https://simulation-hypothesis.com/simulations/methods#m-real-time)[Agent-based](https://simulation-hypothesis.com/simulations/methods#m-agent-based)floor: Bodies

### Vehicle and crew trainers

Replica tank turrets, ship operations rooms, submarine control rooms and helicopter cabs.

Question it answers: Does every crew member know their role when failures occur?

Leaves out: Confined mock-ups cannot fully replicate real vehicle motion and stress.

Runs on: about 10^12 FLOPS rough the simulator's own computer, real time Since: 1954 by 1954, early gunnery trainers for armoured vehicle crews

[Real time](https://simulation-hypothesis.com/simulations/methods#m-real-time)floor: Bodies

### Networked training exercises

Real units, crewed simulators and computer-generated forces joined into one shared exercise across many sites.

Question it answers: Can large formations coordinate across services and countries?

Leaves out: Network delays and simplifications mean not every unit behaves as it would live.

Runs on: about 10^14 FLOPS rough many linked simulator and computer nodes, real time Since: 1983 SIMNET, a 1980s defence programme that first linked simulators over a network for joint exercises

[Real time](https://simulation-hypothesis.com/simulations/methods#m-real-time)[Agent-based](https://simulation-hypothesis.com/simulations/methods#m-agent-based)floor: Bodies

### Marksmanship and use-of-force trainers

Screen- and laser-based ranges with branching scenarios, used by armed forces and police, including practice in de-escalation.

Question it answers: Does the trainee make the right decision, including the choice not to shoot?

Leaves out: Actors on screen react on a script, not with a real person's unpredictability.

Runs on: about 10^12 FLOPS rough a desktop with projection and sensors, real time Since: 1985 by 1985, early branching-scenario trainers for police and military decisions

[Real time](https://simulation-hypothesis.com/simulations/methods#m-real-time)[Rules and states](https://simulation-hypothesis.com/simulations/methods#m-rules)floor: Bodies

### Cyber ranges

Whole simulated computer networks where defenders practise detecting and stopping attacks.

Question it answers: Can the team spot and contain an intrusion in time?

Leaves out: Scripted attacks in a range are less severe than those of a determined real adversary.

Runs on: about 10^13 FLOPS rough a cluster of virtual machines, real time Since: 2009 by 2009, early dedicated ranges for rehearsing cyber attack and defence

[Discrete event](https://simulation-hypothesis.com/simulations/methods#m-discrete-event)[Rules and states](https://simulation-hypothesis.com/simulations/methods#m-rules)floor: Rules

### Combat medic trainers

Manikins that bleed, breathe and respond to battlefield first aid, for practising trauma care under simulated fire.

Question it answers: Can the medic stop the bleeding and keep the casualty alive under pressure?

Leaves out: Real fear and exhaustion in combat are hard to recreate in a training exercise.

Runs on: about 10^10 FLOPS rough an embedded controller inside the manikin, real time Since: 1997 by 1997, early high-fidelity trauma manikins for combat medic training

[Physical analogue](https://simulation-hypothesis.com/simulations/methods#m-analogue)[Differential equations](https://simulation-hypothesis.com/simulations/methods#m-equations)floor: Bodies

### Autonomous cyber-defence gyms

A simulated network where learning attack and defence agents act and are scored.

Question it answers: Can agents learn to defend a network against an attacker?

Leaves out: Network topology and attacks are far simpler than real ones.

Runs on: about 10^11 FLOPS rough Cheap many-episode training Since: 2021 CybORG: a gym for the development of autonomous cyber agents (2021) [source](https://arxiv.org/abs/2108.09118)

[Agent-based](https://simulation-hypothesis.com/simulations/methods#m-agent-based)[Game theory](https://simulation-hypothesis.com/simulations/methods#m-game-theory)floor: Rules

## Weapons physics and protection

### Nuclear stockpile simulation

Since nuclear test bans, ageing warheads are assessed by simulating their physics instead of testing them. This work built some of the world’s largest supercomputers.

Question it answers: Is the stockpile still safe and reliable without explosive testing?

Leaves out: Simulating ageing materials without real tests still leaves some physics uncertain.

Runs on: about 10^18 FLOPS textbook one of the world's largest supercomputers, a large fraction, for weeks Since: 1995 the US Stockpile Stewardship Program begins after the test moratorium

[Continuum fields](https://simulation-hypothesis.com/simulations/methods#m-continuum)[Monte Carlo](https://simulation-hypothesis.com/simulations/methods#m-monte-carlo)floor: Fields

### Ballistics

Projectiles through the barrel, the air and the target, as three separate problems.

Question it answers: Where will the projectile land, and what occurs on impact?

Leaves out: Wind and manufacturing variation still make individual shots harder to predict than the average.

Runs on: about 10^12 FLOPS rough a desktop, seconds per shot simulated Since: 1945 ENIAC, one of the first electronic computers, whose first job computed artillery firing tables

[Differential equations](https://simulation-hypothesis.com/simulations/methods#m-equations)[Continuum fields](https://simulation-hypothesis.com/simulations/methods#m-continuum)floor: Bodies

### Blast and impact protection

Shock waves and impacts hitting buildings, vehicles and body armour, used to design protection.

Question it answers: Will the structure, and the people inside it, survive?

Leaves out: Fine details of fracture and fragmentation are still hard to capture exactly.

Runs on: about 10^15 FLOPS rough a university or defence lab cluster, hours per case Since: 1961 by 1961, early hydrocodes at national laboratories model blast waves

[Continuum fields](https://simulation-hypothesis.com/simulations/methods#m-continuum)floor: Fields

### Radar and sensor performance

Radar, sonar and infrared signals travelling through the environment and bouncing off targets.

Question it answers: What can the sensor detect, at what range, in this weather?

Leaves out: Real weather and terrain clutter are harder to fully capture than the model assumes.

Runs on: about 10^14 FLOPS rough a defence contractor's cluster, minutes per scenario Since: 1963 by 1963, early computer models of radar propagation and clutter

[Continuum fields](https://simulation-hypothesis.com/simulations/methods#m-continuum)[Monte Carlo](https://simulation-hypothesis.com/simulations/methods#m-monte-carlo)floor: Fields

### Directed energy weapon simulation

A laser or high-power microwave beam travelling through the air and heating or disabling a target.

Question it answers: How much power reaches the target, and how long before it has an effect?

Leaves out: Real atmospheric turbulence and dust scatter the beam more unpredictably than the model assumes.

Runs on: about 10^14 FLOPS rough a defence laboratory cluster, hours per scenario Since: 1974 by 1974, early beam propagation models for high-energy laser weapons

[Continuum fields](https://simulation-hypothesis.com/simulations/methods#m-continuum)[Differential equations](https://simulation-hypothesis.com/simulations/methods#m-equations)floor: Fields

### Torpedo and underwater weapon simulation

A torpedo or naval mine's path, guidance and explosion through water, and the sound it makes doing it.

Question it answers: Will it find, reach and disable the target?

Leaves out: Real ocean noise and layered water temperatures confuse guidance more than the clean model assumes.

Runs on: about 10^14 FLOPS rough a naval research cluster, hours per scenario Since: 1960 by 1960, early digital torpedo guidance and hydrodynamics models

[Continuum fields](https://simulation-hypothesis.com/simulations/methods#m-continuum)[Differential equations](https://simulation-hypothesis.com/simulations/methods#m-equations)floor: Fields

### Chemical, biological and radiological hazard plumes

The release of a hazardous agent is carried by wind and turbulence in a puff model to predict the footprint and dose.

Question it answers: Who is exposed after a release and where should they evacuate?

Leaves out: Urban terrain and uncertain wind conditions limit accuracy.

Runs on: about 10^9 FLOPS rough puff models run in seconds to minutes on a workstation. Since: 1996 Sykes and colleagues, SCIPUFF, a generalised dispersion model. [source](https://doi.org/10.1007/978-1-4615-5841-5%5F45)

[Particles and N-body](https://simulation-hypothesis.com/simulations/methods#m-particles)[Continuum fields](https://simulation-hypothesis.com/simulations/methods#m-continuum)[Stochastic processes](https://simulation-hypothesis.com/simulations/methods#m-stochastic)floor: Fields

## Consequences of war

### Climate effects of nuclear war

Climate models given the smoke from burning cities, to trace the cooling and crop failures that would result.

Question it answers: How many people would a nuclear war harm far from where it was fought?

Leaves out: How much smoke actually reaches the stratosphere is still debated.

Runs on: about 10^16 FLOPS rough a national climate-modelling supercomputer, a share of it for weeks Since: 1983 the TTAPS study modelled smoke from burning cities cooling the climate

[Continuum fields](https://simulation-hypothesis.com/simulations/methods#m-continuum)floor: Fields

### Displacement and refugee flows

People fleeing conflict, choosing routes and destinations, used by humanitarian agencies to plan aid.

Question it answers: Where will people go, and where should help be waiting?

Leaves out: People's real decisions about where to flee are harder to predict than the model assumes.

Runs on: about 10^13 FLOPS rough a humanitarian agency's server, minutes per scenario Since: 2015 by 2015, early agent-based models forecast wartime refugee movement

Models: [Human](https://simulation-hypothesis.com/entities/human)

[Agent-based](https://simulation-hypothesis.com/simulations/methods#m-agent-based)[Network models](https://simulation-hypothesis.com/simulations/methods#m-network)floor: Bodies

### Casualty and humanitarian response

Injured people, medical teams, supplies and evacuation routes after a conflict or disaster.

Question it answers: How can the greatest number of lives be saved with the resources available?

Leaves out: Assumes that supplies and transport arrive as planned, which is rarely the case in practice.

Runs on: about 10^12 FLOPS rough a hospital or military planning desktop, minutes per scenario Since: 1996 by 1996, early discrete-event simulations of mass-casualty triage and evacuation

[Discrete event](https://simulation-hypothesis.com/simulations/methods#m-discrete-event)[Agent-based](https://simulation-hypothesis.com/simulations/methods#m-agent-based)floor: Bodies

### Landmine and unexploded ordnance clearance planning

Suspected minefields mapped and scheduled for clearance, balancing safety, cost and the land people need back.

Question it answers: Which fields should be cleared first, and how long will it take?

Leaves out: Real records of where mines were laid are often incomplete or wrong.

Runs on: about 10^10 FLOPS rough a humanitarian demining charity's laptop, minutes per scenario Since: 1994 by 1994, early planning models for humanitarian minefield clearance

[Monte Carlo](https://simulation-hypothesis.com/simulations/methods#m-monte-carlo)[Discrete event](https://simulation-hypothesis.com/simulations/methods#m-discrete-event)floor: Populations

### War economic and reconstruction modelling

A country's economy after conflict, tracking damaged infrastructure, lost workers and the aid and investment needed to rebuild.

Question it answers: How long will recovery take, and where should reconstruction money go first?

Leaves out: Real reconstruction depends on political stability that the economic model cannot foresee.

Runs on: about 10^11 FLOPS rough a development agency's desktop, minutes per scenario Since: 1993 by 1993, early post-conflict reconstruction economic models

[System dynamics](https://simulation-hypothesis.com/simulations/methods#m-system-dynamics)[Differential equations](https://simulation-hypothesis.com/simulations/methods#m-equations)floor: Populations

### Sanctions and economic warfare modelling

A blockaded or sanctioned country's trade, currency and supplies, tracking how much economic pressure builds and who it actually hurts.

Question it answers: How much economic pain will these sanctions cause, and how long can the target hold out?

Leaves out: Real sanctions leak through smuggling and third countries more than the clean model assumes.

Runs on: about 10^11 FLOPS rough a research institute's desktop, minutes per scenario Since: 1985 by 1985, early economic models of trade sanctions and blockades

[System dynamics](https://simulation-hypothesis.com/simulations/methods#m-system-dynamics)[Network models](https://simulation-hypothesis.com/simulations/methods#m-network)floor: Populations

### Biological attack response models

How a disease spreads after a deliberate release, with vaccination, isolation and hospital capacity added, so response plans can be compared.

Question it answers: Is mass vaccination or targeted tracing better after a release?

Leaves out: Assumes the pathogen's properties and the delay before detection.

Runs on: about 10^7 FLOPS rough epidemic simulations on a laptop. Since: 2002 Kaplan, Craft and Wein, emergency response to a smallpox attack: the case for mass vaccination. [source](https://doi.org/10.1073/pnas.162282799)

[Differential equations](https://simulation-hypothesis.com/simulations/methods#m-equations)[Stochastic processes](https://simulation-hypothesis.com/simulations/methods#m-stochastic)floor: Populations

[◂ Before computers](https://simulation-hypothesis.com/simulations/before-computers)[Film, media and art ▸](https://simulation-hypothesis.com/simulations/media-art)