# Simulation Simulation is the practice of running a [[Mathematical_model|model]] forward in time to learn what closed-form analysis cannot tell you: take a [[Dynamical_system|dynamical system]], a [[Queueing_theory|queue]], a market of agents, or a [[Probability_distribution|distribution]] too tangled to integrate, encode its rules, and let a computer generate its trajectories. Since the [[Monte_Carlo_method|Monte Carlo]] experiments of [[John_von_Neumann|von Neumann]] and Ulam (1946–49), it has grown into the third leg of science beside theory and experiment, and the working method of [[Operations_research|operations research]], [[Reliability_engineering|reliability engineering]], [[Weather_forecasting|weather prediction]], and [[Decision_theory|decision analysis]]. The craft divides on two axes — how state changes (continuous versus discrete) and how time advances (stepped versus event-driven) — and it stands or falls on the discipline of verification and validation. ## Four canonical kinds | Family | State and time | Canonical machinery | Root articles | |---|---|---|---| | Continuous | smooth states, stepped time | [[Ordinary_differential_equation|ODE]]/[[Partial_differential_equation|PDE]] solvers, [[Numerical_integration|numerical integration]] | [[Differential_equation]], [[Fluid_dynamics]] | | Discrete-event | jumps at irregular instants | event calendar, [[Queueing_theory|queueing]] logic | [[Operations_research]], [[Finite-state_machine]] | | Monte Carlo | sampled randomness | pseudorandom draws, estimators | [[Monte_Carlo_method]], [[Probability_theory]] | | Agent-based | many interacting rules | local update rules, emergent statistics | [[Agent-based_model]], [[Cellular_automaton]] | Continuous simulation descends from analog computing and the Runge–Kutta integrators (1895–1901); stiff chemistry and circuits needed the implicit methods Gear systematized in 1971. Discrete-event simulation was industrialized by Gordon's GPSS (1961), and Dahl and Nygaard's SIMULA (1962–67) — built to describe systems as interacting processes — gave programming the class and the object as side effects. [[Monte_Carlo_method|Monte Carlo]] began as neutron transport at Los Alamos; the Metropolis algorithm (1953) extended it to sampling [[Statistical_mechanics|equilibrium ensembles]], ancestor of every modern Markov-chain sampler. [[Agent-based_model|Agent-based modeling]] runs from Schelling's segregation checkerboard (1971) through Sugarscape (1996) to [[Evolutionary_game_theory|evolutionary game]] tournaments, with [[Conway's_Game_of_Life|Life]] and the [[Elementary_cellular_automaton|elementary automata]] as its minimal exhibits of [[Emergence|emergence]]. ## Clocks: time-stepped versus event-driven A time-stepped engine advances Δt at a time and recomputes everything; an event-driven engine leaps to the next scheduled state change and skips the dead air between. The choice is economics. Explicit [[Numerical_integration|integrators]] must respect stability bounds — for advection the Courant–Friedrichs–Lewy condition (1928) caps Δt by grid spacing over wave speed — so resolving [[Turbulence|turbulence]] directly costs roughly the cube of the [[Reynolds_number|Reynolds number]], which is why practical [[Fluid_dynamics|CFD]] models the small scales instead. A job shop or hospital ward, by contrast, changes state a few thousand times a day: [[Discrete_time_and_continuous_time|discrete-event]] treatment makes those runs cheap enough to replicate thousands of times for confidence intervals. Hybrid engines co-simulate both regimes — continuous plant, discrete [[Control_theory|controller]] — the standard architecture for [[Model_predictive_control|predictive control]] design, hardware-in-the-loop rigs for [[Fly-by-wire|flight computers]], and [[NASA]]-style crew trainers. Runtime state estimation belongs to the [[Kalman_filter|Kalman filter]]; simulation supplies the model it corrects. ## Randomness you can replay Stochastic simulation needs randomness that is reproducible on demand. Pseudorandom generators — Lehmer's congruential recurrence (1949), the Mersenne Twister (1997) with period 2¹⁹⁹³⁷ − 1 — deliver deterministic streams that pass statistical batteries yet replay exactly from a seed, so a bug or a rare event can be re-run at will. [[John_von_Neumann|Von Neumann]], who used the middle-square method knowing its flaws, warned from the start that arithmetic generators are formally never random; [[Cryptography|cryptographic]] applications accordingly hold theirs to a stricter standard than [[Monte_Carlo_method|Monte Carlo]] does. On top of the stream sit the estimator crafts: variance reduction (importance sampling, antithetic and control variates), rare-event splitting for [[Reliability_engineering|reliability]] studies where failure probabilities sit at 10⁻⁶ and below, and careful [[Statistics|statistical]] treatment of correlated output — a run is one draw from a [[Probability_distribution|distribution over trajectories]], not a datum. ## Verification is mathematics, validation is science Verification asks whether the code solves the equations right: convergence under grid refinement, conservation checks, manufactured solutions, unit tests — questions of [[Correctness_(computer_science)|program correctness]] and [[Numerical_integration|numerical]] error. Validation asks whether they are the right equations: does the [[Mathematical_model|model]], calibrated on one regime, predict held-out data in another? The two are routinely conflated and must not be: a beautifully converged solution of the wrong physics is the most dangerous artifact in engineering. Between white-box theory and black-box fit sits the [[Grey_box_model|grey-box model]]; where runs are expensive, [[Metamodeling|metamodels]] — response surfaces, Gaussian-process emulators, increasingly [[Machine_learning|machine-learned]] surrogates and [[Neural_network_(machine_learning)|neural nets]] — stand in for the simulator, at the price of a second validation problem. Sensitivity analysis apportions output [[Uncertainty|uncertainty]] to inputs; calibration is [[Statistics|statistics]], not curve-worship; and George Box's reminder that all models simplify is the field's standing epistemology. For decisions, the simulation is an argument to be audited under [[Decision_theory|decision theory]], not an oracle — a point [[Systems_analysis|systems analysis]] institutionalized decades ago. ## Where simulation earns its keep The first numerical [[Weather_forecasting|weather forecast]] ran on ENIAC in 1950 (Charney and [[John_von_Neumann|von Neumann]]); its heirs, global [[Atmospheric_model|atmospheric models]], meet the hard ceiling [[Edward_Norton_Lorenz|Lorenz]] found in 1963 — the [[Lorenz_system|sensitive dependence]] that makes ensembles, not single runs, the honest forecast product. [[Reliability_engineering|Reliability]] work couples [[Fault_tree_analysis|fault trees]] and [[Failure_mode_and_effects_analysis|FMEA]] with Monte Carlo over component lifetimes; factories and supply chains live in [[Operations_research|discrete-event]] models; epidemiologists and ecologists run [[Population_model|population]] and [[Ecosystem_model|ecosystem models]]; energy planners couple [[Energy_modeling|energy-system models]] to [[Integrated_assessment_modelling|integrated assessment]]; chip houses simulate every circuit before committing an [[Integrated_circuit|integrated circuit]] to a mask set, and biologists now [[Modelling_biological_systems|model whole cellular subsystems]]. Training is its own branch: from the Link Trainer (1929) to full-motion airline and [[NASA]] crew simulators, rehearsal in silico is cheaper than failure in vivo — the same logic by which the method has become the default instrument of [[Systems_engineering|engineering]] and policy wherever experiments are unaffordable, unethical, or unrepeatable. **On the spine:** [[Monte_Carlo_method]] · [[Agent-based_model]] · [[Mathematical_model]] · [[Dynamical_system]] · [[Operations_research]]. ## Wikipedia : Wikitube **Strict pair:** [Wikipedia](https://en.wikipedia.org/wiki/Simulation) : [Wikitube](https://en.wikitube.io/wiki/Simulation) ## Previous hub tags Hubs: `Systems`. Portals: [[PORTAL_Systems]], [[PORTAL_Monte_Carlo_method]], [[PORTAL_Decision_theory]], [[PORTAL_Reliability_engineering]], [[PORTAL_Control_theory]]. --- *Repopulated 2026-08-12 · redlink fill · 0 deletions.*