Habrich, Oliver André (2024)
Parameter Identification in Cahn-Hilliard Systems.
Technische Universität Darmstadt
doi: 10.26083/tuprints-00027849
Dissertation, Erstveröffentlichung, Verlagsversion
Kurzbeschreibung (Abstract)
The Cahn-Hilliard equation is a mathematical model used to study phase separation processes within physics, chemistry or biology. Due to its phenomenological flavour, the model parameters are not known in real-world applications, and a calibration is needed to derive a quantitative agreement with data obtained from experiments. In this thesis, we address the problem of identifying three model parameters within the Cahn-Hilliard equation, i.e. the interface parameter, the double well potential function, and the mobility function from spatially resolved measurements of the phase fraction. We derive identifiability results and establish a linear and a non-linear approach to solve the parameter identification problems numerically. In the first part of this work, we identify an inherent non-uniqueness of the inverse problem, leading to the exclusion of the interface parameter in the following considerations. We establish the identifiability of the mobility and the potential up to certain scaling invariances under realistic observability conditions. In the second part, we consider an equation error approach to solve the identification problems. Therefore, measurements are directly inserted into the Cahn-Hilliard equation, leading to linear operator equations in Hilbert spaces with perturbed operators. We use Tikhonov regularisation to derive stable approximations for the solutions of the ill-posed problems and show that this approach is well-posed. Numerical experiments demonstrate the feasibility of the method. The equation error method requires high assumptions on the regularity of the data. We address this issue in the third part of our investigations by considering an output least squares approach. This leads to non-linear inverse problems in Hilbert spaces. Again, Tikhonov regularisation is employed to derive stable approximations for the solution. We show the well-posedness and continuity properties of the non-linear forward operator and establish the existence of solutions to the Tikhonov minimisation problem. A Gauss-Newton iteration is applied to solve the resulting minimisation problem. We show the differentiability of the forward operator and derive a representation for the adjoint operator of the derivative. The results regarding the output least squares approach are established by considering auxiliary variational problems. The existence of unique solutions to those problems is derived by Galerkin approximation and energy estimates. Afterwards, we discuss the discretisation of this approach using a Petrov-Galerkin method and present numerical results. In the final part of this work, we consider more complex models and present numerical tests, which indicate that the output least squares approach can also be applied to those problems.
Typ des Eintrags: | Dissertation | ||||
---|---|---|---|---|---|
Erschienen: | 2024 | ||||
Autor(en): | Habrich, Oliver André | ||||
Art des Eintrags: | Erstveröffentlichung | ||||
Titel: | Parameter Identification in Cahn-Hilliard Systems | ||||
Sprache: | Englisch | ||||
Referenten: | Egger, Prof. Dr. Herbert ; Lukácová-Medvidová, Prof. Dr. Mária | ||||
Publikationsjahr: | 6 August 2024 | ||||
Ort: | Darmstadt | ||||
Kollation: | 130 Seiten | ||||
Datum der mündlichen Prüfung: | 19 April 2024 | ||||
DOI: | 10.26083/tuprints-00027849 | ||||
URL / URN: | https://tuprints.ulb.tu-darmstadt.de/27849 | ||||
Kurzbeschreibung (Abstract): | The Cahn-Hilliard equation is a mathematical model used to study phase separation processes within physics, chemistry or biology. Due to its phenomenological flavour, the model parameters are not known in real-world applications, and a calibration is needed to derive a quantitative agreement with data obtained from experiments. In this thesis, we address the problem of identifying three model parameters within the Cahn-Hilliard equation, i.e. the interface parameter, the double well potential function, and the mobility function from spatially resolved measurements of the phase fraction. We derive identifiability results and establish a linear and a non-linear approach to solve the parameter identification problems numerically. In the first part of this work, we identify an inherent non-uniqueness of the inverse problem, leading to the exclusion of the interface parameter in the following considerations. We establish the identifiability of the mobility and the potential up to certain scaling invariances under realistic observability conditions. In the second part, we consider an equation error approach to solve the identification problems. Therefore, measurements are directly inserted into the Cahn-Hilliard equation, leading to linear operator equations in Hilbert spaces with perturbed operators. We use Tikhonov regularisation to derive stable approximations for the solutions of the ill-posed problems and show that this approach is well-posed. Numerical experiments demonstrate the feasibility of the method. The equation error method requires high assumptions on the regularity of the data. We address this issue in the third part of our investigations by considering an output least squares approach. This leads to non-linear inverse problems in Hilbert spaces. Again, Tikhonov regularisation is employed to derive stable approximations for the solution. We show the well-posedness and continuity properties of the non-linear forward operator and establish the existence of solutions to the Tikhonov minimisation problem. A Gauss-Newton iteration is applied to solve the resulting minimisation problem. We show the differentiability of the forward operator and derive a representation for the adjoint operator of the derivative. The results regarding the output least squares approach are established by considering auxiliary variational problems. The existence of unique solutions to those problems is derived by Galerkin approximation and energy estimates. Afterwards, we discuss the discretisation of this approach using a Petrov-Galerkin method and present numerical results. In the final part of this work, we consider more complex models and present numerical tests, which indicate that the output least squares approach can also be applied to those problems. |
||||
Alternatives oder übersetztes Abstract: |
|
||||
Status: | Verlagsversion | ||||
URN: | urn:nbn:de:tuda-tuprints-278491 | ||||
Sachgruppe der Dewey Dezimalklassifikatin (DDC): | 500 Naturwissenschaften und Mathematik > 510 Mathematik | ||||
Fachbereich(e)/-gebiet(e): | 04 Fachbereich Mathematik 04 Fachbereich Mathematik > Numerik und wissenschaftliches Rechnen |
||||
Hinterlegungsdatum: | 06 Aug 2024 12:20 | ||||
Letzte Änderung: | 07 Aug 2024 09:58 | ||||
PPN: | |||||
Referenten: | Egger, Prof. Dr. Herbert ; Lukácová-Medvidová, Prof. Dr. Mária | ||||
Datum der mündlichen Prüfung / Verteidigung / mdl. Prüfung: | 19 April 2024 | ||||
Export: | |||||
Suche nach Titel in: | TUfind oder in Google |
Frage zum Eintrag |
Optionen (nur für Redakteure)
Redaktionelle Details anzeigen |