ebook img

Stable Approximate Evaluation of Unbounded Operators PDF

133 Pages·2007·2.195 MB·English
Save to my drive
Quick download
Download
Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.

Preview Stable Approximate Evaluation of Unbounded Operators

Lecture Notes in Mathematics (cid:49)(cid:56)(cid:57)(cid:52) Editors: J.-M.Morel,Cachan F.Takens,Groningen B.Teissier,Paris Charles W. Groetsch Stable Approximate Evaluation of Unbounded Operators (cid:65)(cid:66)(cid:67) Author CharlesW.Groetsch TheTraubertChair SchoolofScienceandMathematics TheCitadel Charleston,SC(cid:50)(cid:57)(cid:52)(cid:48)(cid:57) USA e-mail:[email protected] LibraryofCongressControlNumber:2006931917 MathematicsSubjectClassification((cid:50)(cid:48)(cid:48)(cid:48)(cid:41)(cid:58)Primary:(cid:52)(cid:55)(cid:65)(cid:53)(cid:50)(cid:44)(cid:54)(cid:53)(cid:74)(cid:50)(cid:48)(cid:59) Secondary:(cid:52)(cid:55)(cid:65)(cid:53)(cid:56)(cid:44)(cid:54)(cid:53)(cid:74)(cid:50)(cid:50) ISSNprintedition:(cid:48)(cid:48)(cid:55)(cid:53)(cid:45)(cid:56)(cid:52)(cid:51)(cid:52) ISSNelectronicedition:(cid:49)(cid:54)(cid:49)(cid:55)(cid:45)(cid:57)(cid:54)(cid:57)(cid:50) ISBN-10 (cid:51)(cid:45)(cid:53)(cid:52)(cid:48)(cid:45)(cid:51)(cid:57)(cid:57)(cid:52)(cid:50)(cid:45)(cid:57)SpringerBerlinHeidelbergNewYork ISBN-13 (cid:57)(cid:55)(cid:56)(cid:45)(cid:51)(cid:45)(cid:53)(cid:52)(cid:48)(cid:45)(cid:51)(cid:57)(cid:57)(cid:52)(cid:50)(cid:45)(cid:49)SpringerBerlinHeidelbergNewYork DOI(cid:49)(cid:48)(cid:46)(cid:49)(cid:48)(cid:48)(cid:55)/(cid:51)(cid:45)(cid:53)(cid:52)(cid:48)(cid:45)(cid:51)(cid:57)(cid:57)(cid:52)(cid:50)(cid:45)(cid:57) Thisworkissubjecttocopyright.Allrightsarereserved,whetherthewholeorpartofthematerialis concerned,specificallytherightsoftranslation,reprinting,reuseofillustrations,recitation,broadcasting, reproductiononmicrofilmorinanyotherway,andstorageindatabanks.Duplicationofthispublication orpartsthereofispermittedonlyundertheprovisionsoftheGermanCopyrightLawofSeptember(cid:57), (cid:49)(cid:57)(cid:54)(cid:53),initscurrentversion,andpermissionforusemustalwaysbeobtainedfromSpringer.Violationsare liableforprosecutionundertheGermanCopyrightLaw. SpringerisapartofSpringerScience+BusinessMedia springer.com (cid:176)c Springer-VerlagBerlinHeidelberg(cid:50)(cid:48)(cid:48)(cid:55) Theuseofgeneraldescriptivenames,registerednames,trademarks,etc.inthispublicationdoesnotimply, evenintheabsenceofaspecificstatement,thatsuchnamesareexemptfromtherelevantprotectivelaws andregulationsandthereforefreeforgeneraluse. TypesettingbytheauthorandSPiusingaSpringerLATEXpackage Coverdesign:WMXDesignGmbH,Heidelberg Printedonacid-freepaper SPIN:(cid:49)(cid:49)(cid:56)(cid:53)(cid:54)(cid:55)(cid:57)(cid:53) VA(cid:52)(cid:49)(cid:47)(cid:51)(cid:49)(cid:48)(cid:48)/SPi (cid:53)(cid:52)(cid:51)(cid:50)(cid:49)(cid:48) In Memory of Joaquin Bustoz, Jr. 1939-2003 Preface This monograph is a study of an aspect of operator approximation theory that emerges from the theory of linear inverse problems in the mathematical sciences.Suchinverseproblemsareoftenmodeledbyoperatorequationsofthe firstkindinvolvingacompactlinearoperatordefinedonaHilbertspace.The conventional solution operator for the inverse problem is the Moore-Penrose generalizedinverseofthemodeloperator.Exceptintheunusualcasewhenthe modeloperatorhasfiniterank,theMoore-Penroseinverseisadenselydefined, closed,unboundedoperator.Thereforeboundedperturbationsinthedatacan be amplified without bound by the solution operator. Indeed, it is a common experience of those who have dealt with such inverse problems to find that low amplitude noise in the data expresses itself as high amplitude oscillations in the computed solution. The successful treatment of these inverse problems therefore requires two ingredients: an approximation of the Moore-Penrose inversecoupledwithastabilizationtechniquetodampenspuriousoscillations in the approximate solution. Hereweconsiderstabilizedevaluationofanunboundedoperatorasaprob- leminitsownrightinoperatorapproximationtheory.Bystabilizedevaluation we mean that the value of an unbounded operator at some vector in its do- mainisapproximatedbyapplyingboundedlinearoperatorstoanapproximate data vector that is not necessarily in the domain of the original unbounded operator. Questions of convergence and orders of approximation will be of foremost concern. A unifying thread for the discussion is a classical theorem of von Neumann on certain bounded “resolvents” of closed densely defined unbounded operators. This result is the bridge that allows passage from the unbounded operator to a class of approximating bounded operators. When von Neumann’s theorem is combined with the spectral theorem for bounded self-adjoint operators a general scheme for stabilized evaluation of the un- bounded operator results. Particular cases of the general scheme, notably the Tikhonov-Morozov method and its variants, are studied in some detail and finite-dimensional realizations are dealt with in the final chapter. VIII Preface The key idea of von Neumann’s proof involves regarding the graph of the operator as a subspace embedded in a product Hilbert space (in this sense the proof is truly Cartesian). We also show that this notion, combined with von Neumann’s alternating projection theorem, applied in the product space, can be used to give an alternate non-spectral proof of one of the best known operator stabilization methods. Ourintentisforthemonographtobereasonablyself-contained.Webegin withafairlyinformalintroductorychapterinwhichanumberofmodelinverse problems leading to the evaluation of unbounded operators are introduced. The next chapter fills in background material from functional analysis and operator theory that is helpful in the sequel. We hope that this approach will make the monograph a useful source of collateral reading for students in graduate courses in functional analysis and related courses in analysis and applied mathematics. Much of the work that is reported here was originally carried out in col- laboration with my friends Otmar Scherzer of the University of Innsbruck, Austria and Martin Hanke-Bourgeois of the University of Mainz, Germany. WithbothOtmarandMartinIhadthehappyexperienceofopenandfriendly collaborations in which my benefits exceeded my contributions. While writing these notes I learned of the tragic death of my earliest colleague and coauthor, Joaquin Bustoz, Jr., of Arizona State University. Besideshismanyresearchpapersonsummabilitytheoryandspecialfunctions, Joaquin’s important andlastingcontributions tothemathematical education of disadvantaged youth made his passing a great loss to the profession, to say nothing of the personal sense of loss felt by his friends and colleagues. This monograph is fondly dedicated to Joaquin’s memory. I am grateful to the Charles Phelps Taft Research Center for providing support, in the form of a Faculty Fellowship, during the preparation of this work. Seabrook Island, S.C. Charles Groetsch May, 2006 Contents 1 Some Problems Leading to Unbounded Operators ......... 1 1.1 Introduction ............................................ 1 1.2 Some Unbounded Operators .............................. 2 1.2.1 Differentiation and Multiplication ................... 3 1.2.2 Dirichlet-to-Neumann Map ......................... 4 1.2.3 Potential Theory .................................. 5 1.2.4 Moment Theory................................... 8 1.3 Unbounded Operators and the Heat Equation............... 9 1.3.1 Time Reversal .................................... 10 1.3.2 Source Identification ............................... 11 1.3.3 Diffusivity from Temperature History ................ 12 1.3.4 The Steady State.................................. 13 1.3.5 Surface Phenomena................................ 13 1.4 Overview............................................... 16 1.5 Notes .................................................. 17 2 Hilbert Space Background ................................. 19 2.1 Hilbert Space ........................................... 19 2.2 Weak Convergence....................................... 22 2.3 Bounded Linear Operators................................ 24 2.4 Unbounded Operators.................................... 30 2.5 Pseudo-inversion ........................................ 41 2.6 Optimization ........................................... 43 2.7 Notes .................................................. 51 3 A General Approach to Stabilization ...................... 53 3.1 A General Method....................................... 53 3.2 Some Cases............................................. 58 3.2.1 The Tikhonov-Morozov Method ..................... 58 3.2.2 The Iterated Tikhonov-Morozov Method ............. 63 3.2.3 An Interpolation-Based Method ..................... 65 X Contents 3.2.4 A Method Suggested by Dynamical Systems .......... 70 3.3 Notes .................................................. 73 4 The Tikhonov-Morozov Method ........................... 77 4.1 The Tikhonov-Morozov Method ........................... 77 4.2 The Discrepancy Criterion................................ 81 4.3 Iterated Tikhonov-Morozov Method ....................... 86 4.4 The Nonstationary Method ............................... 89 4.5 Notes .................................................. 97 5 Finite-Dimensional Approximations........................101 5.1 Stabilization by Projection ...............................101 5.2 Finite Elements .........................................111 5.2.1 A Finite Element Discrepancy Criterion..............116 5.3 Notes ..................................................119 References.....................................................121 Index..........................................................125 1 Some Problems Leading to Unbounded Operators His majesty then turned to me, and requested me to explain the reason why such great effects should proceed from so small a cause ... The Adventures of Hajji Baba of Isfahan J. Morier 1.1 Introduction Unbounded operators can transform arbitrarily small vectors into arbitrar- ily large vectors – a phenomenon known as instability. Stabilization methods strive to approximate a value of an unbounded operator by applying a family of bounded operators to rough approximate data that do not necessarily lie within the domain of the unbounded operator. This monograph is a math- ematical study of stabilization techniques for the evaluation of unbounded operatorsactingbetweenHilbertspaces.Thegoalofthestudyistocontribute to the theoretical basis of a wide-ranging methodology for inverse problems, datasmoothingandimageanalysis.Myintentisentirelytheoretical.Idonot propose to delve into the details of these applications, but rather my aim is to use the application areas as motivators for the study of specific topics in the theory of operator approximations. Stabilization problems inevitably arise in the solution of inverse problems that are phrased in infinite-dimensional function spaces. The direct versions of these problems typically involve a highly smoothing operator and conse- quentlytheinversionprocessisusuallyhighlyill-posed.Thisill-posedproblem can be viewed on a theoretical level (and often on a quite practical level) as evaluatinganunboundedoperatoronsomedataspace(therangeofthedirect operator). The main difficulty arises when inaccuracies (arising in the physi- calsetting frommeasurementerrors)carrytheobservabledata outsideofthe theoretical data space. The unbounded nature of the solution operator can thengiverisetospectacularinstabilitiesintheapproximatesolution.General stabilization techniques, the subject of this monograph, involve insertion of 2 1 Some Problems Leading to Unbounded Operators the given data into the data space of the solution operator and subsequent approximate evaluation of the solution operator. Of course these approxima- tions must be carried out very carefully in order to preserve stability. A key point in this process is the choice of parameters defining the approximate evaluation in such a way as to control the instabilities. It is hoped that this monograph which explores these ideas, and is expressed in rigorous operator theory, will be interesting to mathematicians and will also be potentially useful to those in the scientific community who confront unstable ill-posed problems in their work. Our intent is also to pro- vide a source for collateral study for students in courses in applied functional analysis and related courses. With this in mind the next chapter provides brief details of some of the operator theory that will be used in the sequel. Before proceeding to this background material we give a semi-formal intro- duction to some problems from application areas that lead to the evaluation of unbounded operators. We remind the reader that a linear operator T : X → Y from a normed linear space X into a normed linear space Y is called bounded if the number (cid:2)Tx(cid:2) (cid:2)T(cid:2)=sup x(cid:1)=0 (cid:2)x(cid:2) is finite (unless further specification is necessary, (cid:2)·(cid:2) will always denote the norminanappropriatespace).Forlinearoperatorstheconceptsofcontinuity atapoint,uniformcontinuityandboundednesscoincide.IfT isanunbounded linear operator, then there is a sequence of unit vectors {z } in the domain n of T satisfying (cid:2)Tz (cid:2)→∞ as n→∞. For this reason we frequently say that n for an unbounded linear operator T the mapping z → Tz is unstable. That is, for linear operators we treat the terms stable, bounded and continuous as synonymous. 1.2 Some Unbounded Operators In elementary calculus the differentiation process is regarded as basic and benign. But differentiation is Janus-faced and its darker side is all too fa- miliar to scientists who must differentiate functions that are given empir- ically. This sinister face of differentiation is instability. In operator terms this instability means that the linear operator defined by differentiation is unbounded on func√tion spaces with conventional norms. For example, the functions φ (t) = 2sinnπt are unit vectors in the space L2[0,1], yet the n L2-norm of the derivatives, (cid:2)φ(cid:2) (cid:2)=πn, is unbounded. This means, for exam- n ple, that the differentiation operator defined in the subspace of differentiable functions in L2[0,1], when considered as a operator taking values in L2[0,1], is unbounded.

See more

The list of books you might like

Most books are stored in the elastic cloud where traffic is expensive. For this reason, we have a limit on daily download.