A measurement concept defining how outcomes are modeled and how state descriptions are updated after an outcome is recorded. It governs outcome probabilities, information extraction, and the disturbance introduced by the measurement interaction. It does not yield reliable inference without adequate calibration, sufficient data, and appropriate estimation procedures. It supports reconstruction and validation of state and process descriptions from experimental statistics. The concept is generally stable, though practical implementations and estimation methods evolve over time.

Principle

Update prior beliefs π(ρ) to posterior π(ρ|data) ∝ Pr(data|ρ) π(ρ) using Bayes’ theorem while enforcing the physical state space. The posterior encodes both data likelihood and prior regularization, enabling principled uncertainty quantification and incorporation of prior knowledge or constraints.

Demonstration

Reconstructing a qubit using a Bures-like prior: start with a chosen prior over the Bloch ball, compute the likelihood from measurement counts, multiply to obtain the posterior, approximate the posterior by Monte Carlo sampling (e.g., Markov Chain Monte Carlo) and report the posterior mean and 95% credible region for the Bloch vector.

Misapplication

Selecting an ill-justified or improper prior that dominates the posterior for available data, neglecting prior sensitivity analysis, or reporting only a MAP estimate without conveying posterior spread. Using naive, low-quality samplers that fail to explore multimodal posteriors leads to misleading inference.

Consequence

Delivers a full probabilistic description of state uncertainty, regularizes ill-posed reconstruction problems via the prior, and yields credible intervals and loss-minimizing point estimates. Drawbacks include computational cost and dependence of results on prior choice for limited data regimes.

Reversal

Maximum likelihood (and frequentist) approaches supply a single point estimate optimized over data alone; Bayesian methods contrast by producing distributions that reflect prior knowledge and finite-sample uncertainty.

Boundary

Applicable when one can specify a prior and perform posterior computation; excludes contexts that strictly require noninformative frequentist guarantees or when computational resources prevent reliable sampling. Care must be taken in high-dimensional state spaces where priors and samplers strongly influence outcomes.

Semantic Tension

Tension between subjective prior incorporation and the desire for objective, data-only estimates; trade-off between computational tractability and fidelity of posterior approximation; and tension between reporting a single convenient point estimate and communicating full distributional uncertainty.

Synthesis

Bayesian tomography frames state reconstruction as probabilistic inference: data updates a prior to produce a posterior over states, providing principled uncertainty quantification and regularization at the cost of explicit prior choices and heavier computation.