Tikhonov regularization, named for Andrey Tikhonov, is the most commonly used method of regularization of ill-posed problems. In statistics, the method is known as ridge regression, and, with multiple independent discoveries, it is also variously known as the Tikhonov-Miller method, the Phillips-Twomey method, the constrained linear inversion method, and the method of linear regularization. It is related to the Levenberg-Marquardt algorithm for non-linear least-squares problems.
When the following problem is not well posed (either because of non-existence or non-uniqueness of )
then the standard approach is known as linear least squares and seeks to minimize the residual
where is the Euclidean norm. This may be due to the system being overdetermined or underdetermined ( may be ill-conditioned or singular). In the latter case this is no better than the original problem. In order to give preference to a particular solution with desirable properties, the regularization term is included in this minimization:
for some suitably chosen Tikhonov matrix, . In many cases, this matrix is chosen as the identity matrix , giving preference to solutions with smaller norms. In other cases, highpass operators (e.g., a difference operator or a weighted Fourier operator) may be used to enforce smoothness if the underlying vector is believed to be mostly continuous. This regularization improves the conditioning of the problem, thus enabling a numerical solution. An explicit solution, denoted by , is given by:
The effect of regularization may be varied via the scale of matrix . For , when = 0 this reduces to the unregularized least squares solution provided that (ATA)−1 exists.
Contents[hide] |
Although at first the choice of the solution to this regularized problem may look artificial, and indeed the matrix seems rather arbitrary, the process can be justified from a Bayesian point of view. Note that for an ill-posed problem one must necessarily introduce some additional assumptions in order to get a stable solution. Statistically we might assume that a priori we know that is a random variable with a multivariate normal distribution. For simplicity we take the mean to be zero and assume that each component is independent with standard deviation . Our data are also subject to errors, and we take the errors in to be also independent with zero mean and standard deviation . Under these assumptions the Tikhonov-regularized solution is the most probable solution given the data and the a priori distribution of , according to Bayes' theorem. The Tikhonov matrix is then for Tikhonov factor .
If the assumption of normality is replaced by assumptions of homoskedasticity and uncorrelatedness of errors, and if one still assumes zero mean, then the Gauss-Markov theorem entails that the solution is the minimal unbiased estimate.
For general multivariate normal distributions for and the data error, one can apply a transformation of the variables to reduce to the case above. Equivalently, one can seek an to minimize
where we have used to stand for the weighted norm (cf. the Mahalanobis distance). In the Bayesian interpretation is the inverse covariance matrix of , is the expected value of , and is the inverse covariance matrix of . The Tikhonov matrix is then given as a factorization of the matrix (e.g. the Cholesky factorization), and is considered a whitening filter.
This generalized problem can be solved explicitly using the formula
Typically discrete linear ill-conditioned problems result from discretization of integral equations, and one can formulate a Tikhonov regularization in the original infinite dimensional context. In the above we can interpret as a compact operator on Hilbert spaces, and and as elements in the domain and range of . The operator is then a self-adjoint bounded invertible operator.
With , this least squares solution can be analyzed in a special way via the singular value decomposition. Given the singular value decomposition of A
with singular values , the Tikhonov regularized solution can be expressed as
where has diagonal values
and is zero elsewhere. This demonstrates the effect of the Tikhonov parameter on the condition number of the regularized problem. For the generalized case a similar representation can be derived using a generalized singular value decomposition.
Finally, it is related to the Wiener filter:
where the Wiener weights are and is the rank of .
The optimal regularization parameter is usually unknown and often in practical problems is determined by an ad hoc method. A possible approach relies on the Bayesian interpretation described above. Other approaches include the discrepancy principle, cross-validation, L-curve method, restricted maximum likelihood and unbiased predictive risk estimator. Grace Wahba proved that the optimal parameter, in the sense of leave-one-out cross-validation minimizes:
where is the residual sum of squares and is the effective number degree of freedom.
Using the previous SVD decomposition, we can simplify the above expression:
and
The probabilistic formulation of an inverse problem introduces (when all uncertainties are Gaussian) a covariance matrix representing the a priori uncertainties on the model parameters, and a covariance matrix representing the uncertainties on the observed parameters (see, for instance, Tarantola, 2004 [1]). In the special case when these two matrices are diagonal and isotropic, and , and, in this case, the equations of inverse theory reduce to the equations above, with .
Tikhonov regularization has been invented independently in many different contexts. It became widely known from its application to integral equations from the work of Tychonoff and David L. Phillips. Some authors use the term Tikhonov-Phillips regularization. The finite dimensional case was expounded by Arthur E. Hoerl, who took a statistical approach, and by Manus Foster, who interpreted this method as a Wiener-Kolmogorov filter. Following Hoerl, it is known in the statistical literature as ridge regression.