
The process of determining the values of parameters in a model from observational data.Parameter estimation is the craft of extracting the best values for a model's adjustable constants from noisy, incomplete observations. In physics, every theory has free parameters — the mass of a neutron star, the stiffness of a crystal lattice, the coupling strength between particles — and the data rarely pins them down exactly. The challenge is to find not just a good fit, but a defensible one: how well is each parameter constrained, and which values are ruled out? This discipline sits at the heart of experimental science, turning raw measurements into quantitative claims about the world.
The method of least squares was introduced by Legendre and Gauss in the early 1800s for astronomical orbit fitting, and the likelihood principle was formalized by Fisher in the 1920s. Bayesian parameter estimation has roots in Bayes' theorem (1763) but became computationally practical only in the late 20th century.
The standard approach is to define a likelihood function that scores how probable the observed data is for a given set of parameter values. The best parameters maximize this score, and the curvature of the score near the peak gives error bars. Modern methods add Bayesian priors to encode prior knowledge, and Markov Chain Monte Carlo (MCMC) algorithms to explore high-dimensional parameter spaces when the model is complex. The key is to propagate uncertainties honestly, so that conclusions come with honest confidence intervals.
Used everywhere from gravitational wave astronomy (estimating black hole masses and spins from detector signals) to plasma physics (inferring turbulence parameters from diagnostics) and medical imaging (reconstructing tissue properties from sparse scans).
The first major use of least squares was to predict the orbit of the newly discovered dwarf planet Ceres in 1801, after it was lost behind the Sun — Gauss's fit let astronomers find it again exactly where predicted.






