A computational stochastic sampling technique that estimates properties (expectations, probabilities, distributions) of a model by repeatedly drawing random (or quasi‑random) samples from specified probability distributions and computing empirical statistics; uncertainty is quantified by sampling variability and estimator error.

Principle

Under the model’s probability measure, Monte Carlo estimators converge to the true quantities by the law of large numbers; estimator precision improves with sample size N approximately proportional to 1/√N, and sampling error is reducible by variance‑reduction techniques or by increasing N.

Demonstration

Illustrative scenario — Estimating an integral: Situation — an integral E[f(X)] has no closed form. Recognition — define a probability distribution for X and an unbiased estimator f(X). Action — draw N independent samples X_i, compute sample mean (1/N)∑f(X_i), and estimate a confidence interval using sample variance. Consequence — for sufficiently large N the sample mean approximates the integral within a predictable sampling error bound.

Misapplication

Interpreting a single Monte Carlo run as definitive or ignoring model misspecification and sampling diagnostics; the semantic error is conflating estimator sampling error and model error or assuming convergence without assessing variance and independence of samples.

Consequence

Monte Carlo delivers flexible, model‑based numerical estimates and uncertainty quantification for problems in high dimension or with complex likelihoods; its limitations are computational cost, slow convergence for some integrands, and sensitivity to RNG quality and model correctness.

Reversal

In very high dimensional integrals or for integrands with large variance, naive Monte Carlo converges impractically slowly; alternatives include variance‑reduction methods, quasi‑Monte Carlo, importance sampling, deterministic integration, or model reparameterization. If the probabilistic model is unspecified or data‑dependent without a sampling model, Monte Carlo is inapplicable.

Boundary

Clearly within — estimating expectations, tail probabilities, Bayesian posterior integrals, option pricing via simulation. Boundary case — moderate to high dimensional integration where variance reduction is required. Clearly outside — problems that admit efficient closed‑form analytic solutions or where deterministic solvers are exact and cheaper.

Semantic Tension

Tradeoff between computational cost and statistical precision: increasing sample size reduces error but increases computation; there is tension between model fidelity (complex realistic models) and the feasibility of achieving low sampling variance.

Synthesis

Monte Carlo is the practical default for numerical expectation and uncertainty estimation when analytic solutions are unavailable and a probabilistic model exists; effective use requires attention to sample size, random number quality, diagnostics, and variance‑reduction strategies.