Systems Simulation

Systems simulation is the practice of building a working model of a system—a factory floor, a hospital ward, an air-defense network, a supply chain—and running it forward in time to see what happens, when experimenting on the real thing would be too costly, too slow, or too dangerous.

9 frameworks, from 1917 to today · Industrial Engineering · Engineering

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

Overview

Systems simulation is the practice of building a working model of a system—a factory floor, a hospital ward, an air-defense network, a supply chain—and running it forward in time to see what happens, when experimenting on the real thing would be too costly, too slow, or too dangerous1. Its practitioners do not solve a system's equations once and declare an answer; they let a model behave, step by step or event by event, and watch the consequences unfold1. What is at stake is the difference between a design that looks sound on paper and one that survives contact with variability, congestion, and surprise. The history of the field is largely a history of learning how to make such models trustworthy and how to make them fast enough to matter.

From Wartime Mathematics to the First Simulated Systems

The intellectual roots lie in probability and queueing. In 1909 the Danish engineer Agner Krarup Erlang, working for the Copenhagen telephone company, published a treatment of waiting times for telephone calls that became queueing theory: the mathematics of lines, servers, and arrival rates2. Erlang's formulas described, for example, how many operators a switchboard needed so that callers would not wait too long. Queueing theory gave later simulators their vocabulary—arrivals, service times, utilization, waiting lines—but it could only handle tidy cases. Real systems with irregular arrivals, priorities, breakdowns, and interacting stations resisted closed-form solution, and that gap is where simulation grew.

The decisive push came from the Second World War and the Cold War. The Monte Carlo method, developed in the 1940s at Los Alamos by Stanislaw Ulam, John von Neumann, and Nicholas Metropolis, replaced hard integrals with repeated random sampling: to estimate a quantity, draw many random scenarios and average the results. Its name evoked the casino, and its logic—that randomness, repeated often enough, reveals structure—became the philosophical core of stochastic simulation. The same years produced system dynamics. Jay Forrester, an engineer at MIT who had worked on servomechanisms and early digital computing, published Industrial Dynamics in 1961, modeling firms as networks of feedback loops and stocks and flows. His claim was that many business failures came not from bad managers but from the counterintuitive behavior of complex feedback structures. The book Urban Dynamics (1969) and the later World Dynamics (1971) extended the approach to cities and to global resources, and the field's signature tool, the stock-and-flow diagram, taught a generation to think in accumulations and delays.

The Discrete-Event Idea and the General-Purpose Simulator

The 1960s brought the idea that made simulation an engineering discipline rather than a collection of one-off calculations: discrete-event simulation. In such a model, time jumps from one event to the next—a part arrives, a machine finishes, a customer is served—and nothing happens in between3. This is efficient because most of a system's clock time is spent waiting. Geoffrey Gordon, at IBM, created GPSS (General Purpose Simulation System) in the early 1960s; its blocks and transactions let engineers describe a system in a compact notation and let the computer advance the clock4. GPSS became one of the most widely used simulation languages of the era. Around the same time, the British operational researcher K. D. Tocher developed the activity-cycle approach and, with colleagues, the GSP suite, and later SIMULA, designed by Ole-Johan Dahl and Kristen Nygaard in Norway in the 1960s, introduced object-oriented ideas and the concept of the "process" that would shape later simulation languages. In the United States, SLAM (developed by A. Alan B. Pritsker and colleagues, appearing in the 1970s) and SIMSCRIPT offered competing formalisms—event scheduling, process interaction, and activity scanning—each claiming to describe systems more naturally. The competition mattered because each formalism forced modelers to think differently about what a system is.

Continuous simulation developed alongside, for systems whose state changes smoothly rather than in jumps: chemical processes, population dynamics, control systems. Here the model is a set of differential equations integrated numerically over time, and the lineage runs through analog computing and early numerical integration to modern tools. The distinction between discrete-event and continuous modeling—and hybrid models that combine both—remains a basic organizing idea of the field.

Making Models Believable: Verification and Validation

As models proliferated, a hard question surfaced: how do you know the model is right1? The answer crystallized in the 1960s and 1970s into verification and validation. Verification asks whether the model was built correctly—does the code do what the designer intended5? Validation asks whether the right model was built—does it behave like the real system5? The terminology was codified influentially by Robert G. Sargent in a series of papers from the 1970s onward, and the field developed a toolkit: face validation by experts, comparison with historical data, sensitivity analysis, and the "Turing test" of asking whether experts can distinguish model output from real output. This was not bureaucratic housekeeping. A simulation that is fast but wrong is worse than no simulation, and the credibility of the whole enterprise rested on disciplined validation.

Optimization Inside the Simulation

A model that can be run can also be searched6. Simulation-based optimization asks: given a simulator, what input settings produce the best output6? Because simulation output is noisy—two runs of the same configuration give different answers—classical optimization methods fail, and the field developed methods that tolerate noise: ranking and selection procedures, response-surface methodology (pioneered by George Box and K. B. Wilson in the 1950s for physical experiments and adapted to simulation), and later metaheuristics and stochastic approximation. The practical payoff was large: instead of trying a handful of designs by hand, an engineer could let the computer search thousands of configurations, each evaluated by simulation, and return a defensible best.

Agents, Networks, and the Digital Twin

The 1990s and 2000s brought agent-based modeling, in which a system is represented as many autonomous individuals—people, vehicles, firms—each following simple rules, with system-level patterns emerging from their interactions. Its roots reach back to cellular automata and to the social and biological modeling of the 1970s and 1980s, but it matured as computing power grew. It proved especially suited to systems where heterogeneity and local interaction drive outcomes: traffic, epidemics, markets, and crowds. In parallel, simulation moved from the factory floor into services and networks—hospitals, call centers, logistics—and into the design of communication and defense systems, where discrete-event models of message traffic were essential.

The most recent turn is the digital twin: a simulation model kept continuously synchronized with a real operating asset through sensor data, so that the model can be used not just to design the system but to monitor, predict, and optimize it in service7. The idea draws on earlier "virtual" and "mirror" models in manufacturing and aerospace, and it became a named and widely pursued concept in the 2010s, riding on cheap sensing, cloud computing, and the Internet of Things. Its promise—and its open problem—is fidelity over time: a twin is only useful if it stays faithful as the real system ages and changes.

Where the Field Stands

Today the main approaches coexist rather than compete. Discrete-event simulation remains the workhorse for manufacturing, logistics, healthcare, and service operations, with commercial and open-source platforms descended from the GPSS and SLAM lineage. System dynamics persists in strategy, public policy, and sustainability analysis, where feedback and accumulation dominate. Agent-based modeling has a secure place in transportation, epidemiology, and market design. Continuous and hybrid simulation serve engineering and control applications. Simulation-based optimization is embedded in most modern tools, and verification and validation has become a formal part of model credibility assessment, increasingly guided by professional standards. The digital twin has pulled simulation into the operational life of systems, blurring the line between design-time analysis and real-time decision support. The open questions are familiar in new dress: how to validate models of systems that have never existed, how to keep twins faithful as their physical counterparts drift, and how to make simulation fast enough—through parallel computing and surrogate models—to sit inside the decision loop rather than beside it.

References

  1. Simulation, Wikipedia
  2. Queueing theory (introduction, History), Wikipedia
  3. Discrete-event simulation, Wikipedia
  4. GPSS (introduction, History), Wikipedia
  5. Verification and validation, Wikipedia
  6. Simulation-based optimization, Wikipedia
  7. Digital twin, Wikipedia

The frameworks of Systems Simulation

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

  1. 1917-Present

    Queueing Theory

    How long will customers wait in line?

    Fly to its tower in Nousfarer

  2. 1945-Present

    Monte Carlo method

    How can random sampling estimate hard quantities?

    Fly to its tower in Nousfarer

  3. 1955-Present

    Continuous simulation

    How do smoothly changing systems evolve over time?

    Fly to its tower in Nousfarer

  4. 1955-Present

    Simulation-based optimization

    What inputs make a simulated system perform best?

    Fly to its tower in Nousfarer

  5. 1960-Present

    Discrete-event simulation

    How do systems behave when events happen at random times?

    Fly to its tower in Nousfarer

  6. 1961-Present

    System dynamics

    How do feedback loops cause counterintuitive outcomes?

    Fly to its tower in Nousfarer

  7. 1970-Present
  8. 1990-Present

    Agent-based model

    How do simple individual rules create complex group patterns?

    Fly to its tower in Nousfarer

  9. 2010-Present

    Digital twin

    How can a model stay synchronized with a real system?

    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.

Queueing TheoryMonte Carlo methodContinuous simulationSimulation-based optimizationDiscrete-event simulationSystem dynamicsVerification and validationAgent-based modelDigital twin

Beyond the field

Where Systems Simulation's ideas came from in other fields, and what they led to there: the strongest 12 of 63 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.

Systems Simulation is a planet in Nousfarer

Its city has a tower for each of these 9 frameworks. Land, walk in, and climb one concept at a time.

Land on Systems Simulation

Related fields

Fields its ideas are linked to

More in Industrial Engineering

See where Systems Simulation sits on the Map of Human Knowledge