Document Type


Date of Award

Summer 2017

Degree Name

Doctor of Philosophy in Mathematical Sciences - (Ph.D.)


Mathematical Sciences

First Advisor

Catalin C. Turc

Second Advisor

Michael Siegel

Third Advisor

Eliza Zoi-Heleni Michalopoulou

Fourth Advisor

Peter G. Petropoulos

Fifth Advisor

Francisco-Javier Sayas


In this dissertation, an in-depth comparison between boundary integral equation solvers and Domain Decomposition Methods (DDM) for frequency domain Helmholtz transmission problems in composite two-dimensional media is presented. Composite media are characterized by piece-wise constant material properties (i.e., index of refraction) and thus, they exhibit interfaces of material discontinuity and multiple junctions. Whenever possible to use, boundary integral methods for solution of Helmholtz boundary value problems are computationally advantageous. Indeed, in addition to the dimensional reduction and straightforward enforcement of the radiation conditions that these methods enjoy, they do not suffer from the pollution effect present in volumetric discretization. The reformulation of Helmholtz transmission problems in composite media in terms of boundary integral equations via multi-traces constitutes one of the recent success stories in the boundary integral equation community. Multi-trace formulations (MTF) incorporate local Dirichlet and Neumann traces on subdomains within Green’s identities and use restriction and extension by zero operators to enforce the intradomain continuity of the fields and fluxes. Through usage of subdomain Calderon projectors, the transmission problem is cast into a linear system form whose unknowns are local Dirichlet and Neumann traces (two such traces per interface of material discontinuity) and whose operator matrix consists of diagonal block boundary integral operators associated with the subdomains and extension/projections off diagonal blocks. This particular form of the matrix operator associated with MTF is amenable to operator preconditioning via Calderon projectors.

DDM rely on subdomain solutions that are matched via transmission conditions on the subdomain interfaces that are equivalent to the physical continuity of fields and traces. By choosing the appropriate transmission conditions, the convergence of DDM for frequency domain scattering problems can be accelerated. Traditionally, the intradomain transmission conditions were chosen to be the classical outgoing Robin/impedance boundary conditions. When the ensuing DDM linear system is solved via Krylov subspace methods, the convergence of DDM with classical Robin transmission conditions is slow and adversely affected by the number of subdomains. Heuristically, this behavior is explained by the fact that Robin boundary conditions are first order approximations of transparent boundary conditions, and thus there is significant information that is reflected back into a given subdomain from adjacent subdomains. Clearly, using more sophisticated transparent boundary conditions facilitates the information exchange between subdomains. For instance, Dirichlet-to-Neumann (DtN) operators of adjacent domains or suitable approximations of these can be used in the form of generalized Robin boundary conditions to increase the rate of the convergence of iterative solvers of DDM linear systems. The approximations of DtN operators that are expressed in terms of Helmholtz hypersingular operators (e.g., the normal derivative of the double layer operator) are used in this dissertation. The incorporation of these in a DDM framework is subtle, and an effective method is proposed to blend these transmission operators in the presence of multiple junctions. Conceptually, the information exchange between subdomains is realized through certain Robin-to-Robin (RtR) operators, which how to compute robustly via integral equation formulations is shown.

All of the Helmholtz boundary integral operators that feature in Calderon’s calculus are discretized via Nystr¨om methods that rely on sigmoid transforms, trigonometric interpolation, and singular kernel splitting. Sigmoid transforms are means to polynomially accumulate discretization points toward corners without compromising the discretization density in smooth boundary portions. A wide variety of numerical results is presented in this dissertation that illustrate the merits of each of the two approaches (MTF and DDM) for the solution of transmission problems in composite domains.

Included in

Mathematics Commons