## Doctoral Dissertations

Off-campus UMass Amherst users: To download campus access dissertations, please use the following link to log into our proxy server with your UMass Amherst user name and password.

Non-UMass Amherst users: Please talk to your librarian about requesting this dissertation through interlibrary loan.

Dissertations that have an embargo placed on them will not be available to anyone until the embargo expires.

#### Document Type

Open Access Dissertation

#### Degree Name

Doctor of Engineering (DEng)

#### Degree Program

Electrical and Computer Engineering

2016

#### Month Degree Awarded

May

Marinos N. Vouvakis

#### Subject Categories

Electromagnetics and Photonics

#### Abstract

Numerical modeling of electromagnetic (EM) phenomenon has proved to become an effective and efficient tool in design and optimization of modern electronic devices, integrated circuits (IC) and RF systems. However the generality, efficiency and reliability/resilience of the computational EM solver is often criticised due to the fact that the underlying characteristics of the simulated problems are usually different, which makes the development of a general, ''black-box'' EM solver to be a difficult task.

In this work, we aim to propose a reliable/resilient, scalable and efficient finite elements based domain decomposition method (FE-DDM) as a general CEM solver to tackle such ultimate CEM problems to some extent. We recognize the rank deficiency property of the Dirichlet-to-Neumann (DtN) operators involved in the previously proposed FETI-2$\lambda$ DDM formulation and apply such principle to improve the computational efficiency and robustness of FETI-2$\lambda$ DDM. Specifically, the rank deficient DtN operator is computed by a randomized computation method that was originally proposed to approximate matrix singular value decomposition (SVD). Numerical results show a up to 35\% run-time and 75% memory saving of the DtN operators computation can be achieved on a realistic example. Later, such rank deficiency principle is incorporated into a new global DDM preconditioner (W-FETI) that is inspired by the matrix Woodbury identity. Numerical study of the eigenspectrum shows the validity of the proposed W-FETI global preconditioner. Several industrial-scaled examples show significant iterative convergence advantage of W-FETI that uses 35%-80% matrix-vector-products (MxVs) than state-of-the-art DDM solvers.

COinS