The task of determining unknown model parameters, inputs, or system structure from observed outputs or measurements by inverting a forward model; typically an ill-conditioned or ill-posed inference problem requiring regularization and uncertainty quantification.

Principle

An inverse problem inverts a forward operator that maps parameters to observables; because this operator may be non-injective, non-surjective, or sensitive to noise, stable recovery relies on additional information (regularization, priors, constraints) and assessment of nonuniqueness and sensitivity.

Demonstration

Tomographic reconstruction recovers interior density from line integrals of attenuation; seismic inversion infers subsurface velocity profiles from recorded waveforms; estimating a reaction rate constant from time series of concentrations is another inverse problem.

Misapplication

Treating the inverse problem as a straightforward optimization without checking identifiability or quantifying uncertainty—for example, reporting a single best-fit parameter value from noisy, noninformative data without regularization or confidence statements.

Consequence

Correct formulation leads to stable, interpretable parameter estimates together with uncertainty bounds and diagnostics of identifiability; it informs experiment design to resolve ambiguities and provides principled trade-offs between data fit and prior/regularization strength.

Reversal

The forward problem: given parameters and inputs, compute expected outputs deterministically. Unlike the inverse problem, the forward mapping is typically well-posed and stable to compute but not invertible without extra information.

Boundary

Covers problems of parameter or input recovery from measurements where a forward model exists; excludes purely predictive forward simulations, purely data-driven black-box regression without an explicit forward operator, and ill-defined problems lacking a model or measurement mapping.

Semantic Tension

Often conflated with parameter estimation or system identification; the inverse-problem viewpoint emphasizes operator inversion, ill-posedness, and the need for regularization, whereas parameter estimation in statistics may focus on likelihoods and point estimates without addressing nonuniqueness.

Synthesis

An inverse problem is the reconstruction of unknowns from observations by inverting a forward model; because inversion is often unstable or nonunique, it requires regularization, uncertainty quantification, and careful assessment of identifiability.