Tikhonov Regularization Methods for Ill-Posed Linear Systems

Tikhonov regularisation addresses the inherent instability of linear systems in which small perturbations of the input can lead to large variations in the solution. Such ill-posed problems commonly arise in fields ranging from medical imaging and geophysics to signal processing and inverse scattering. The method stabilises the solution by augmenting the usual least-squares objective with a penalty term, typically involving the norm of the solution or its derivatives. This leads to a modified normal equation that balances fidelity to noisy data against smoothness or other prior information encoded by a regularisation operator. Central to the approach is the selection of the regularisation parameter: overly small values risk noise amplification, whereas overly large values produce overly smooth or biased solutions. A variety of parameter-choice rules—such as discrepancy principles, L-curve criteria and cross-validation—have been developed to navigate this trade-off. Extensions to general-form regularisation allow for non-identity penalty operators, facilitating edge preservation or incorporation of physical constraints. Computational advances, including matrix-free implementations, Krylov subspace solvers and function-based frameworks, have rendered Tikhonov regularisation both scalable and theoretically robust for large-scale applications.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

Recent studies have leveraged advanced numerical techniques to extend the applicability of Tikhonov regularisation. A function-based framework embeds regularisation directly in the continuous operator formulation, allowing automatic discretisation and closer alignment with analytic properties. Adaptive cross approximation methods construct low-rank representations of integral operators by selectively sampling pivotal rows and columns, thereby avoiding the assembly of dense matrices and improving solution accuracy under high noise levels. Iterative preconditioning strategies based on the Arnoldi process generate low-dimensional surrogates of severely ill-conditioned systems, within which Tikhonov regularisation is applied to accelerate convergence while maintaining stability. Collectively, these developments address the computational challenges of large-scale inverse problems and demonstrate scalable algorithms that couple rigorous theory with practical performance.

Tikhonov Regularization Methods for Ill-Posed Linear Systems publication trend

The graph below shows the total number of articles in tikhonov regularization methods for ill-posed linear systems across all publications each year (not limited to Nature Index journals).

Technical terms

Ill-posed linear system: A system of equations for which solutions may not exist, be unique or depend continuously on the input data, often signalled by rapidly decaying singular values.

Tikhonov regularisation: A stabilisation technique that solves an ill-posed problem by minimising a penalised least-squares functional combining data fidelity and a weighted norm of the solution.

Regularisation parameter: A scalar weight that balances the influence of the penalty term against the data misfit, controlling the trade-off between stability and accuracy.

Singular value decomposition (SVD): A factorisation of a matrix into orthogonal modes and singular values, used to characterise the degree of ill-posedness and to design filter-based regularisation.

Krylov subspace method: An iterative solver that projects a large linear system onto subspaces generated by successive powers of the matrix applied to the residual, often used to accelerate regularised solutions.

Adaptive cross approximation: A matrix compression technique that constructs a low-rank representation by selecting pivotal rows and columns, reducing computational costs for integral operator discretisations.

Arnoldi process: An algorithm to construct an orthonormal basis of a Krylov subspace, yielding a smaller Hessenberg matrix that approximates the action of a large operator for efficient regularisation.

References

1. Solution of ill-posed problems with Chebfun. Numerical Algorithms (2022).
2. Adaptive cross approximation for Tikhonov regularization in general form. Numerical Algorithms (2022).
3. An Arnoldi-based preconditioner for iterated Tikhonov regularization. Numerical Algorithms (2022).