Steinberg, Florian (2017)
Computational Complexity Theory for Advanced Function Spaces in Analysis.
Technische Universität Darmstadt
Dissertation, Erstveröffentlichung
Kurzbeschreibung (Abstract)
This PhD thesis presents progress in the search for a mathematical rigorous framework for efficient numerics of partial differential equations based on a realistic machine model. While the computability theory of continuous structures is well developed and still an active field of research, in most settings it remains unclear what computations should be considered feasible. This problem is tackled within the framework of second-order representations. The name as well as the notion of polynomial-time computability of the framework is inherited from the complexity theory of functionals on the Baire space, also called second-order complexity theory. As model of computation we use oracle Turing machines. Second-order representations offer a great deal of freedom in developing models of computation and a supply of well investigated structures. In particular, there exists an established second-order representation of the set of continuous functions on the unit interval. This representation has been classified as the weakest representation such that evaluation is possible in polynomial time. As a first step we specify the weakest representation of the integrable functions such that integration is polynomial-time computable. This representation turns out to be discontinuous and therefore not suitable. We go on to explore the general restrictions of bounded-time computations on metric spaces within the framework of second-order representations. The notion of metric entropy, originally introduced by Kolmogorov, is used to classify those compact metric spaces that allow for a short representation such that the metric is computable efficiently. To be able to apply the results to spaces of integrable functions we investigate quantitative versions of classification results of their compact subsets called the Arzelà-Ascoli Theorem and the Fréchet-Kolmogorov Theorem. We use the above to propose a family of representations for Lp-spaces. These are shown to be computably equivalent to the standard representations of the same spaces as metric spaces. Furthermore, we prove that the norm can be computed in exponential time. This is also the case for the supremum norm on the continuous functions on the unit interval. A similar family of representations is presented for Sobolev spaces. These representations are investigated in some detail. Several operators on Sobolev spaces are proven to be polynomial-time computable with respect to these representations. For technical reasons only the one-dimensional case is discussed. Finally, we present a result that classifies the computational complexity of the solution operator of the Dirichlet Problem for Poisson’s Equation on the unit disk as that of integrating a continuous function. As a tool for the comparison, polynomial-time Weihrauch reductions are used.
Typ des Eintrags: | Dissertation | ||||
---|---|---|---|---|---|
Erschienen: | 2017 | ||||
Autor(en): | Steinberg, Florian | ||||
Art des Eintrags: | Erstveröffentlichung | ||||
Titel: | Computational Complexity Theory for Advanced Function Spaces in Analysis | ||||
Sprache: | Englisch | ||||
Referenten: | Ziegler, Prof. Dr. Martin ; Kohlenbach, Prof. Dr. Ulrich ; Kawamura, Prof. Dr. Akitoshi | ||||
Publikationsjahr: | 1 Februar 2017 | ||||
Ort: | Darmstadt | ||||
Datum der mündlichen Prüfung: | 2 November 2016 | ||||
URL / URN: | http://tuprints.ulb.tu-darmstadt.de/6096 | ||||
Kurzbeschreibung (Abstract): | This PhD thesis presents progress in the search for a mathematical rigorous framework for efficient numerics of partial differential equations based on a realistic machine model. While the computability theory of continuous structures is well developed and still an active field of research, in most settings it remains unclear what computations should be considered feasible. This problem is tackled within the framework of second-order representations. The name as well as the notion of polynomial-time computability of the framework is inherited from the complexity theory of functionals on the Baire space, also called second-order complexity theory. As model of computation we use oracle Turing machines. Second-order representations offer a great deal of freedom in developing models of computation and a supply of well investigated structures. In particular, there exists an established second-order representation of the set of continuous functions on the unit interval. This representation has been classified as the weakest representation such that evaluation is possible in polynomial time. As a first step we specify the weakest representation of the integrable functions such that integration is polynomial-time computable. This representation turns out to be discontinuous and therefore not suitable. We go on to explore the general restrictions of bounded-time computations on metric spaces within the framework of second-order representations. The notion of metric entropy, originally introduced by Kolmogorov, is used to classify those compact metric spaces that allow for a short representation such that the metric is computable efficiently. To be able to apply the results to spaces of integrable functions we investigate quantitative versions of classification results of their compact subsets called the Arzelà-Ascoli Theorem and the Fréchet-Kolmogorov Theorem. We use the above to propose a family of representations for Lp-spaces. These are shown to be computably equivalent to the standard representations of the same spaces as metric spaces. Furthermore, we prove that the norm can be computed in exponential time. This is also the case for the supremum norm on the continuous functions on the unit interval. A similar family of representations is presented for Sobolev spaces. These representations are investigated in some detail. Several operators on Sobolev spaces are proven to be polynomial-time computable with respect to these representations. For technical reasons only the one-dimensional case is discussed. Finally, we present a result that classifies the computational complexity of the solution operator of the Dirichlet Problem for Poisson’s Equation on the unit disk as that of integrating a continuous function. As a tool for the comparison, polynomial-time Weihrauch reductions are used. |
||||
Alternatives oder übersetztes Abstract: |
|
||||
URN: | urn:nbn:de:tuda-tuprints-60964 | ||||
Sachgruppe der Dewey Dezimalklassifikatin (DDC): | 500 Naturwissenschaften und Mathematik > 510 Mathematik | ||||
Fachbereich(e)/-gebiet(e): | 04 Fachbereich Mathematik > Logik 04 Fachbereich Mathematik |
||||
Hinterlegungsdatum: | 07 Mai 2017 19:55 | ||||
Letzte Änderung: | 07 Mai 2017 19:55 | ||||
PPN: | |||||
Referenten: | Ziegler, Prof. Dr. Martin ; Kohlenbach, Prof. Dr. Ulrich ; Kawamura, Prof. Dr. Akitoshi | ||||
Datum der mündlichen Prüfung / Verteidigung / mdl. Prüfung: | 2 November 2016 | ||||
Export: | |||||
Suche nach Titel in: | TUfind oder in Google |
Frage zum Eintrag |
Optionen (nur für Redakteure)
Redaktionelle Details anzeigen |