In this thesis, we consider two models from physics, which are characterized by the interplay of thermodynamical and fluid mechanical phenomena: demixing (spinodal decomposition) and Rayleigh--Bénard convection. In both models, we investigate the dependencies of certain intrinsic quantities on the system parameters. The first model describes a thermodynamically driven demixing process of a binary viscous fluid. During the evolution, the two components of the mixture separate into two domains of the different equilibrium volume fractions. One observes a clear tendency: Larger domains grow at the expense of smaller ones, and thus, the average domain sizes increases --- a phenomenon called coarsening. It turns out that two mechanisms are relevant for the coarsening process. At an early stage of the evolution, material transport is essentially mediated by diffusion; at a later stage, when the typical domain size exceeds a certain value, due to the viscosity of the mixture, a fluid flow sets in and becomes the relevant transport mechanism. In both regimes, the growth rates of the typical domain size obey certain power laws. In this thesis, we rigorously establish one-sided bounds on these growth rates via a priori estimates. The second model, Rayleigh--Bénard convection, describes the behavior of a fluid between two rigid horizontal plates that is heated from below and cooled from above. There are two competing heat transfer mechanisms in the system: On the one hand, thermodynamics favors a state in which temperature variations are locally minimized. Thus, in our model, the thermodynamical equilibrium state is realized by a temperature with a linearly decreasing profile, corresponding to pure conduction. On the other hand, due to differences in the densities of hot and cold fluid parcels, buoyancy forces act on the fluid. This results in an upward motion of hot parcels and a downward motion of cold parcels. We study the dependence of the average upward heat flux, measured in the so-called Nusselt number, on the temperature forcing encoded by the container height. It turns out that the efficiency of the heat transport is independent of the height of the container, and thus, the Nusselt number is a constant function of height. Using a priori estimates, we prove an upper bound on the Nusselt number that displays this dependency --- up to logarithmic errors. Further investigations on the flow pattern in Rayleigh--Bénard convection show a clear separation of length scales: Along the horizontal top and bottom plates one observes thin boundary layers in which heat is essentially conducted, whereas the large bulk is characterized by a convective heat flow. We give first rigorous results in favor of linear temperature profiles in the boundary layers, which indicate that heat is indeed essentially conducted close to the boundaries.

Dendritic solidification has been the subject of continuous research, also because of its high importance in metal production. The challenge of predicting macroscopic material properties due to complex solidification processes is complicated by the multiple physical scales and phenomena involved. Practical modeling approaches are still subject to significant limitations due to remaining gaps in the systematic understanding of dendritic microstructure formation. The present work investigates some of these problems at the microscopic level of interfacial morphology using phase-field simulations. The employed phase-field models are implemented within a finite-element framework, allowing efficient and scalable computations on high-performance computing facilities. Particular emphasis is placed on the evolution and interaction of dendrite sidebranches in the broader context of dendrite fragmentation, varying and dynamical solidification conditions.

In this thesis, we consider two models from physics, which are characterized by the interplay of thermodynamical and fluid mechanical phenomena: demixing (spinodal decomposition) and Rayleigh--Bénard convection. In both models, we investigate the dependencies of certain intrinsic quantities on the system parameters. The first model describes a thermodynamically driven demixing process of a binary viscous fluid. During the evolution, the two components of the mixture separate into two domains of the different equilibrium volume fractions. One observes a clear tendency: Larger domains grow at the expense of smaller ones, and thus, the average domain sizes increases --- a phenomenon called coarsening. It turns out that two mechanisms are relevant for the coarsening process. At an early stage of the evolution, material transport is essentially mediated by diffusion; at a later stage, when the typical domain size exceeds a certain value, due to the viscosity of the mixture, a fluid flow sets in and becomes the relevant transport mechanism. In both regimes, the growth rates of the typical domain size obey certain power laws. In this thesis, we rigorously establish one-sided bounds on these growth rates via a priori estimates. The second model, Rayleigh--Bénard convection, describes the behavior of a fluid between two rigid horizontal plates that is heated from below and cooled from above. There are two competing heat transfer mechanisms in the system: On the one hand, thermodynamics favors a state in which temperature variations are locally minimized. Thus, in our model, the thermodynamical equilibrium state is realized by a temperature with a linearly decreasing profile, corresponding to pure conduction. On the other hand, due to differences in the densities of hot and cold fluid parcels, buoyancy forces act on the fluid. This results in an upward motion of hot parcels and a downward motion of cold parcels. We study the dependence of the average upward heat flux, measured in the so-called Nusselt number, on the temperature forcing encoded by the container height. It turns out that the efficiency of the heat transport is independent of the height of the container, and thus, the Nusselt number is a constant function of height. Using a priori estimates, we prove an upper bound on the Nusselt number that displays this dependency --- up to logarithmic errors. Further investigations on the flow pattern in Rayleigh--Bénard convection show a clear separation of length scales: Along the horizontal top and bottom plates one observes thin boundary layers in which heat is essentially conducted, whereas the large bulk is characterized by a convective heat flow. We give first rigorous results in favor of linear temperature profiles in the boundary layers, which indicate that heat is indeed essentially conducted close to the boundaries.:1 Introduction 2 Coarsening rates in binary viscous fluids 2.1 Background from physics 2.2 Background from mathematics 2.3 The model 2.4 The gradient flow structure 2.5 Heuristics 2.6 Numerical simulations 2.7 Main results 2.8 Preliminaries 2.9 Proof of upper bounds on coarsening rates 2.10 Appendix: Well-posedness and regularity of solutions 3 Scaling of the Nusselt number 3.1 Background from physics 3.2 The model and the Nusselt number 3.3 Heuristics 3.4 Main results 3.5 Scaling law in the linear regime 3.6 Preliminaries and review 3.7 Upper bound using the background field method 3.8 Upper bound using the maximum principle 3.9 Appendix: Some elementary estimates 4 The laminar boundary layer 4.1 Background, model, and motivation 4.2 Main results 4.3 Preparation: Bounds on the velocity field 4.4 On the energy distribution 4.5 Bounds on the second order derivatives of the temperature field 4.6 Bounds on the third order derivatives of the temperature field

Diese Arbeit beschäftigt sich mit Coarse-Graining (dt. ``Vergröberung", ``Zusammenfassung von Zuständen") für Gradientensysteme und Markov-Prozesse. Coarse-Graining ist ein etabliertes Verfahren in der Mathematik und in den Naturwissenschaften und hat das Ziel, die Komplexität eines physikalischen Systems zu reduzieren und effektive Modelle herzuleiten. Die mathematischen Probleme in dieser Arbeit stammen aus der Theorie der Systeme interagierender Teilchen. Hierbei werden zwei Ziele verfolgt: Erstens, Coarse-Graining mathematisch rigoros zu beweisen, zweitens, mathematisch äquivalente Beschreibungen für die effektiven Modelle zu formulieren. Die ersten drei Teile der Arbeit befassen sich mit dem Grenzwert schneller Reaktionen für Reaktionssysteme und Reaktions-Diffusions-Systeme. Um effektive Modelle herzuleiten, werden nicht nur die zugehörigen Reaktionsratengleichungen betrachtet, sondern auch die zugrunde liegende Gradientenstruktur. Für Gradientensysteme wurde in den letzten Jahren eine strukturelle Konvergenz, die sogenannte ``EDP-Konvergenz", entwickelt. Dieses Coarse-Graining-Verfahren wird in der vorliegenden Arbeit auf folgende Systeme mit langsamen und schnellen Reaktionen angewandt: lineare Reaktionssysteme (bzw. Markov-Prozesse auf endlichem Zustandsraum), nichtlineare Reaktionssysteme, die das Massenwirkungsgesetz erfüllen, und lineare Reaktions-Diffusions-Systeme. Für den Grenzwert schneller Reaktionen wird eine mathematisch rigorose und strukturerhaltende Vergröberung auf dem Level des Gradientensystems inform von EDP-Konvergenz bewiesen. Im vierten Teil wird der Zusammenhang zwischen Gleichungen mit Gedächtnis und Markov-Prozessen untersucht. Für Gleichungen mit Gedächtnisintegralen wird explizit ein größer Markov-Prozess konstruiert, der die Gleichung mit Gedächtnis als Teilsystem enthält. Der letzte Teil beschäftigt sich mit verschieden Diskretisierungen für den Fokker-Planck-Operator. Dazu werden numerische und analytische Eigenschaften untersucht. / This thesis deals with coarse-graining for gradient systems and Markov processes. Coarse-graining is a well-established tool in mathematical and natural sciences for reducing the complexity of a physical system and for deriving effective models. The mathematical problems in this work originate from interacting particle systems. The aim is twofold: first, providing mathematically rigorous results for physical coarse-graining, and secondly, formulating mathematically equivalent descriptions for the effective models. The first three parts of the thesis deal with fast-reaction limits for reaction systems and reaction-diffusion systems. Instead of deriving effective models by solely investigating the associated reaction-rate equation, we derive effective models using the underlying gradient structure of the evolution equation. For gradient systems a structural convergence, the so-called ``EDP-convergence", has been derived in recent years. In this thesis, this coarse-graining procedure has been applied to the following systems with slow and fast reactions: linear reaction systems (or Markov process on finite state space), nonlinear reaction systems of mass-action type, and linear reaction-diffusion systems. For the fast-reaction limit, we perform rigorous and structural coarse-graining on the level of the gradient system by proving EDP-convergence. In the fourth part, the connection between memory equations and Markov processes is investigated. Considering linear memory equations, which can be motivated from spatial homogenization, we explicitly construct a larger Markov process that includes the memory equation as a subsystem. The last part deals with different discretization schemes for the Fokker–Planck operator and investigates their analytical and numerical properties.