The first big computer simulation modelled a nuclear blast with 12 spheres
During the Manhattan Project, scientists wanted to model how a nuclear detonation unfolds, a problem no neat equation could solve. So they ran a Monte Carlo calculation of just 12 hard spheres, sampling likely scenarios instead of working out every possibility. That trick of sampling rather than enumerating still underlies simulation today.
People often mix up two words. A model is the set of equations describing how a system behaves; a simulation is the act of running a program that solves those equations, usually approximately. Strictly speaking, then, nobody builds a simulation: you build a model or a simulator, and then you run it. Simulations earn their keep where systems are too complex for exact analytical answers, from climate and chemistry to economics and health care.
Scale has exploded since the 12 spheres. In 1997 an American defence exercise tracked 66,239 tanks, trucks and other vehicles across simulated desert terrain around Kuwait, spread over several supercomputers. Researchers have modelled material deformation with a billion atoms and, in 2005, built a 2.64-million-atom model of the ribosome, the cell's protein factory. In 2012 the entire life cycle of the bacterium Mycoplasma genitalium was simulated, and Switzerland's Blue Brain project, launched in May 2005, set out to model a whole human brain down to its molecules.
Different questions need different machinery. Stochastic models roll digital dice to capture chance events such as genetic drift. Discrete event simulations keep a queue of happenings sorted by simulated time, which suits testing computer logic. Continuous simulations solve differential equations step by step and drive flight simulators and circuit models; these once ran on analog computers wired from op-amps, but by the late 1980s digital machines imitated them instead. Agent-based models represent individual cells, trees or shoppers, each following its own rules.
Inputs matter as much as code. An alternating current waveform needs only a few numbers, while weather models swallow terabytes, and rounding errors can compound any uncertainty in measurements. Output changed too: people spot a coming rainstorm far faster on an animated map than in a table of cloud coordinates.
Source: Computer simulation
