Analysis · Concept hubDeep story

Monte Carlo Methods

1946 CE20th-century United States (Ulam and von Neumann)

Concept

Solving hard problems by random simulation — from atomic bombs to modern AI.

Understand it in one breath

"Approximate hard integrals or probabilities with random samples, then analyze the error." Ulam, von Neumann, Metropolis, and colleagues developed the modern method collaboratively at Los Alamos. Pi estimation, option pricing, particle simulation, and some tree searches share sampling ideas but use different algorithms and guarantees.

At a glance

Samples N

Estimate of π

Error

100

~3.08

~0.06

1,000

~3.156

~0.014

10,000

~3.1432

~0.0016

100,000

~3.1416

~0.0001

1,000,000

~3.14159

~10⁻⁵

Throw random points into the square [−1,1]²; the fraction that land inside the circle × 4 ≈ π. Error is inversely proportional to √N — four times the precision requires sixteen times as many samples.

Key formula

E[f(X)]1Ni=1Nf(Xi),Xirand\mathbb{E}[f(X)] \approx \dfrac{1}{N} \sum_{i=1}^{N} f(X_i),\quad X_i \sim \text{rand}

Key moments

1946 CE

Ulam — inspiration from a card game

While recovering from illness, Stanisław Ulam considered estimating the chance of winning solitaire through repeated random trials rather than exhaustive calculation.

1947 CE

von Neumann brings the method to ENIAC

Ulam, von Neumann, and colleagues applied random sampling to neutron-transport calculations. The name Monte Carlo evoked the famous casino and games of chance.

2016 CE

AlphaGo and Monte Carlo tree search

Monte Carlo tree search, combined with neural networks, helped AlphaGo defeat Go champion Lee Sedol and reshaped AI game-playing strategy.

Modern applications

Financial option pricing, particle and molecular simulation, ray tracing in film, and reinforcement-learning simulators.

Beyond MathVoyage

Loading…