Abstract
We discuss, in the context of inverse linear problems in Hilbert space, the notion of the associated infinite-dimensional Krylov subspace and we produce necessary and sufficient conditions for the Krylov-solvability of a given inverse problem, together with a series of model examples and numerical experiments.
Similar content being viewed by others
References
Abramowitz, M., Stegun, I.A.: Handbook of Mathematical Functions with Formulas, Graphs, and Mathematical Tables. National Bureau of Standards Applied Mathematics Series, For Sale by the Superintendent of Documents, vol. 55. U.S. Government Printing Office, Washington, DC (1964)
Brown, P.N., Walker, H.F.: GMRES on (nearly) singular systems. SIAM J. Matrix Anal. Appl. 18, 37–51 (1997)
Caruso, N.A.: On Krylov Methods in Infinite-Dimensional Hilbert Space. Ph.D. thesis, SISSA Trieste (2019)
Caruso, N.A., Michelangeli, A.: Convergence of the conjugate gradient method with unbounded operators. arXiv:1908.10110
Caruso, N.A., Michelangeli, A., Novati, P.: Truncation and convergence issues for bounded linear inverse problems in Hilbert space (2018). arXiv:1811.08195
Cipra, B.A.: The best of the 20th century: editors name top 10 algorithms. SIAM News 33, 1–2 (2005)
Daniel, J.W.: The conjugate gradient method for linear and nonlinear operator equations. SIAM J. Numer. Anal. 4, 10–26 (1967)
Dongarra, J., Sullivan, F.: The top 10 algorithms (guest editors’ intruduction). Comput. Sci. Eng. 2, 22–23 (2000)
Engl, H.W., Hanke, M., Neubauer, A.: Regularization of Inverse Problems. Mathematics and Its Applications, vol. 375. Kluwer Academic Publishers Group, Dordrecht (1996)
Ern, A., Guermond, J.-L.: Theory and Practice of Finite Elements. Applied Mathematical Sciences, vol. 159. Springer, New York (2004)
Freund, R.W., Hochbruck, M.: On the use of two QMR algorithms for solving singular systems and applications in Markov chain modeling. Numer. Linear Algebra Appl. 1, 403–420 (1994)
Gasparo, M.G., Papini, A., Pasquali, A.: Some properties of GMRES in Hilbert spaces. Numer. Funct. Anal. Optim. 29, 1276–1285 (2008)
Gazzola, S., Hansen, P.C., Nagy, J.G.: IR Tools: a MATLAB package of iterative regularization methods and large-scale test problems. Numer. Algorithms 81, 773–811 (2018)
Gehér, L.: Cyclic vectors of a cyclic operator span the space. Proc. Am. Math. Soc. 33, 109–110 (1972)
Halmos, P.R.: A Hilbert Space Problem Book, Graduate Texts in Mathematics, Encyclopedia of Mathematics and Its Applications, vol. 19, 2nd edn, p. 17. Springer, New York (1982)
Hanke, M.: Conjugate Gradient Type Methods for Ill-Posed Problems. Pitman Research Notes in Mathematics Series, vol. 327. Longman Scientific & Technical, Harlow (1995)
Hansen, P.C.: Regularization Tools—A Matlab Package for Analysis and Solution of Discrete Ill-Posed Problems. http://www.imm.dtu.dk/~pcha/Regutools/. Accessed Aug 2019
Herzog, R., Sachs, E.: Superlinear convergence of Krylov subspace methods for self-adjoint problems in Hilbert space. SIAM J. Numer. Anal. 53, 1304–1324 (2015)
Kammerer, W.J., Nashed, M.Z.: On the convergence of the conjugate gradient method for singular linear operator equations. SIAM J. Numer. Anal. 9, 165–181 (1972)
Karush, W.: Convergence of a method of solving linear problems. Proc. Am. Math. Soc. 3, 839–851 (1952)
Liesen, J., Strakoš, Z.E.: Krylov Subspace Methods, Numerical Mathematics and Scientific Computation. Principles and Analysis. Oxford University Press, Oxford (2013)
Nemirovskiy, A.S., Polyak, B.T.: Iterative methods for solving linear ill-posed problems under precise information. I. Izv. Akad. Nauk SSSR Tekhn. Kibernet. 203, 13–25 (1984)
Nevanlinna, O.: Convergence of Iterations for Linear Equations, Lectures in Mathematics ETH Zürich. Birkhäuser Verlag, Basel (1993)
Novati, P.: A convergence result for some Krylov–Tikhonov methods in Hilbert spaces. Numer. Funct. Anal. Optim. 39, 655–666 (2018)
Quarteroni, A.: Numerical Models for Differential Problems, MS&A Modeling, Simulation and Applications, vol. 16, 3rd edn. Springer, Cham (2017)
Riesz, F., Sz.-Nagy, B.: Functional Analysis. Frederick Ungar Publishing Co., New York (1955) (Translated by Leo F. Boron)
Rudin, W.: Real and Complex Analysis, 3rd edn. McGraw-Hill Book Co., New York (1987)
Saad, Y.: Krylov subspace methods for solving large unsymmetric linear systems. Math. Comput. 37, 105–126 (1981)
Saad, Y.: Iterative Methods for Sparse Linear Systems, 2nd edn. Society for Industrial and Applied Mathematics, Philadelphia (2003)
Schmüdgen, K.: Unbounded Self-adjoint Operators on Hilbert Space. Graduate Texts in Mathematics, vol. 265. Springer, Dordrecht (2012)
Winther, R.: Some superlinear convergence results for the conjugate gradient method. SIAM J. Numer. Anal. 17, 14–17 (1980)
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix A. Some prototypical example operators
Appendix A. Some prototypical example operators
Let us review in this Appendix certain operators in Hilbert space that were useful in the course of our discussion, both as a source of examples or counter-examples, and as a playground to understand certain mechanisms typical of the infinite dimensionality.
1.1 A.1. The multiplication operator on \(\ell ^2({\mathbb {N}})\)
Let us denote with \((e_n)_{n\in \mathbb {N}}\) the canonical orthonormal basis of \(\ell ^2({\mathbb {N}})\). For a given bounded sequence \(a\equiv (a_n)_{n\in \mathbb {N}}\) in \({\mathbb {C}}\), the multiplication by a is the operator \(M^{(a)}:\ell ^2({\mathbb {N}})\rightarrow \ell ^2({\mathbb {N}})\) defined by \(M^{(a)}e_n=a_n e_n\) \(\forall n\in {\mathbb {N}}\) and then extended by linearity and density, in other words the operator given by the series
(that converges strongly in the operator sense).
\(M^{(a)}\) is bounded with norm \(\Vert M^{(a)}\Vert _{\mathrm {op}}=\sup _n|a_n|\) and spectrum \(\sigma (M^{(a)})\) given by the closure in \({\mathbb {C}}\) of the set \(\{a_1,a_2,a_3\dots \}\). Its adjoint is the multiplication by \(a^*\). Thus, \(M^{(a)}\) is normal. \(M^{(a)}\) is self-adjoint whenever a is real and it is compact if \(\lim _{n\rightarrow \infty }a_n=0\).
1.2 A.2. The right-shift operator on \(\ell ^2({\mathbb {N}})\)
The operator \(R:\ell ^2({\mathbb {N}})\rightarrow \ell ^2({\mathbb {N}})\) defined by \(Re_n=e_{n+1}\) \(\forall n\in {\mathbb {N}}\) and then extended by linearity and density, in other words the operator given by the series
(that converges strongly in the operator sense), is called the right-shift operator.
R is an isometry (i.e., it is norm-preserving) with closed range \(\mathrm {ran} R=\{e_1\}^\perp \). In particular, it is bounded with \(\Vert R\Vert _{\mathrm {op}}=1\), yet not compact, it is injective, and invertible on its range, with bounded inverse
The adjoint of R on \({\mathcal {H}}\) is the so-called left-shift operator, namely the everywhere defined and bounded operator \(L:{\mathcal {H}}\rightarrow {\mathcal {H}}\) defined by the (strongly convergent, in the operator sense) series
Thus, L inverts R on \(\mathrm {ran} R\), i.e., \(LR=\mathbb {1}\), yet \(RL=\mathbb {1}-|e_1\rangle \langle e_1|\). One has \(\ker R^*=\mathrm {span}\{e_1\}\).
R and L have the same spectrum \(\sigma (R)=\sigma (L)=\{z\in {\mathbb {C}}\,|\,|z|\leqslant 1\}\), but R has no eigenvalue, whereas the eigenvalue of L form the open unit ball \(\{z\in {\mathbb {C}}\,|\,|z|< 1\}\).
1.3 A.3. The compact (weighted) right-shift operator on \(\ell ^2({\mathbb {N}})\)
This is the operator \({\mathcal {R}}:\ell ^2({\mathbb {N}})\rightarrow \ell ^2({\mathbb {N}})\) defined by the operator-norm convergent series
where \(\sigma \equiv (\sigma _n)_{n\in \mathbb {N}}\) is a given bounded sequence with \(0<\sigma _{n+1}<\sigma _n\) \(\forall n\in {\mathbb {N}}\) and \(\lim _{n\rightarrow \infty }\sigma _n=0\). Thus, \({\mathcal {R}}e_n=\sigma _n e_{n+1}\).
\({\mathcal {R}}\) is injective and compact, and (A.5) is its singular value decomposition, with norm \(\Vert {\mathcal {R}}\Vert _{\mathrm {op}}=\sigma _1\), \(\overline{\mathrm {ran}\,{\mathcal {R}}}=\{e_1\}^\perp \), and adjoint
Thus, \({\mathcal {L}}{\mathcal {R}}=M^{(\sigma ^2)}\), the operator of multiplication by \((\sigma _n^2)_{n\in {\mathbb {N}}}\), whereas \({\mathcal {R}}{\mathcal {L}}=M^{(\sigma ^2)}-\sigma _1^2|e_1\rangle \langle e_1|\).
1.4 A.4. The compact (weighted) right-shift operator on \(\ell ^2({\mathbb {Z}})\)
This is the operator \({\mathcal {R}}:\ell ^2({\mathbb {Z}})\rightarrow \ell ^2({\mathbb {Z}})\) defined by the operator-norm convergent series
where \(\sigma \equiv (\sigma _n)_{n\in \mathbb {N}_0}\) is a given bounded sequence with \(0<\sigma _{n+1}<\sigma _n\) \(\forall n\in {\mathbb {N}}_0\) and \(\lim _{n\rightarrow \infty }\sigma _n=0\). Thus, \({\mathcal {R}}e_n=\sigma _{|n|} e_{n+1}\).
\({\mathcal {R}}\) is injective and compact, with \(\mathrm {ran}\,{\mathcal {R}}\) dense in \({\mathcal {H}}\) and norm \(\Vert {\mathcal {R}}\Vert _{\mathrm {op}}=\sigma _0\). (A.7) gives the singular value decomposition. The adjoint of \({\mathcal {R}}\) is
Thus, \({\mathcal {L}}{\mathcal {R}}=M^{(\sigma ^2)}={\mathcal {R}}{\mathcal {L}}\).
The ‘inverse of \({\mathcal {R}}\) on its range’ is the densely defined, surjective, unbounded operator \({\mathcal {R}}^{-1}:\mathrm {ran}\,{\mathcal {R}}\rightarrow {\mathcal {H}}\) acting as
as a series that converges on \(\mathrm {ran}\,{\mathcal {R}}\) in the strong operator sense.
1.5 A.5. The Volterra operator on \(L^2[0,1]\)
This is the operator \(V:L^2[0,1]\rightarrow L^2[0,1]\) defined by
V is compact and injective with spectrum \(\sigma (V)=\{0\}\) (thus, the spectral point 0 is not an eigenvalue) and norm \(\Vert V\Vert _{\mathrm {op}}=\frac{2}{\pi }\). It’s adjoint \(V^*\) acts as
therefore \(V+V^*\) is the rank-one orthogonal projection
onto the function \({\mathbf {1}}(x)=1\).
The singular value decomposition of V is
where both \((\varphi _n)_{n\in {\mathbb {N}}_0}\) and \((\psi _n)_{n\in {\mathbb {N}}_0}\) are orthonormal bases of \(L^2[0,1]\).
Thus, \(\mathrm {ran} V\) is dense, but strictly contained in \({\mathcal {H}}\): for example, \({\mathbf {1}}\notin \mathrm {ran}V\). (Observe, though, that the dense subspace of the polynomials on [0, 1] is mapped by V onto the non-dense \(\mathrm {span}\{x,x^2,x^3,\dots \}\).)
In fact, V is invertible on its range, but does not have (everywhere defined) bounded inverse; yet \(V-z\mathbb {1}\) does, for any \(z\in {\mathbb {C}}{\setminus }\{0\}\) (recall that \(\sigma (V)=\{0\}\)), and
The explicit action of the powers of V is
1.6 A.6. The multiplication operator over \(\Omega \subset {\mathbb {C}}\) in \(L^2(\Omega )\)
This is the operator \(M:L^2(\Omega )\rightarrow L^2(\Omega )\), \(f\mapsto zf\), where \(\Omega \) is a bounded open region in \({\mathbb {C}}\). \(M_z\) is a normal bounded bijection with norm \(\Vert M_z\Vert _{\mathrm {op}}=\sup _{z\in \Omega }|z|\), spectrum \(\sigma (M_z)={\overline{\Omega }}\), and adjoint given by \(M_z^*f={\overline{z}}f\).
Rights and permissions
About this article
Cite this article
Caruso, N., Michelangeli, A. & Novati, P. On Krylov solutions to infinite-dimensional inverse linear problems. Calcolo 56, 32 (2019). https://doi.org/10.1007/s10092-019-0330-7
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s10092-019-0330-7
Keywords
- Inverse linear problems
- Infinite-dimensional Hilbert space
- Ill-posed problems
- Orthonormal basis discretisation
- Bounded linear operators
- Self-adjoint operators
- Krylov subspaces
- Cyclic operators
- Krylov solution
- GMRES
- CG
- LSQR