22.10.2014 Views

CERFACS CERFACS Scientific Activity Report Jan. 2010 – Dec. 2011

CERFACS CERFACS Scientific Activity Report Jan. 2010 – Dec. 2011

CERFACS CERFACS Scientific Activity Report Jan. 2010 – Dec. 2011

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

ELECTROMAGNETISM AND ACOUSTICS TEAM<br />

2.6 The interior transmission eigenvalue problem<br />

The Linear Sampling Method fails to image a penetrable scatterer for some exceptional frequencies, for<br />

which the LSM operator is not injective. These frequencies can be characterized as eigenvalues of a nonstandard<br />

problem set inside the scatterer, referred in the literature as the interior transmission problem.<br />

These eigenvalues are directly related to the constitutive properties of the scatterer and they could be used<br />

to deduce from multi-frequency data some knowledge on this constitution.<br />

The study of transmission eigenvalues is closely linked to the study of the interior transmission problem<br />

which has been a subject of great interest in scattering theory for the past few years. This is due to the<br />

fact that transmission eigenvalues can give information on the properties of an obstacle, for instance on the<br />

index of refraction or if it contains a cavity or a crack. The interior transmission problem is defined by :<br />

curlcurlE − k 2 NE = 0 in D (2.1)<br />

curlcurlE 0 − k 2 E 0 = 0 in D (2.2)<br />

ν × E − ν × E 0 = 0 on ∂D (2.3)<br />

ν × curlE − ν × curlE 0 = 0 on ∂D (2.4)<br />

and transmission eigenvalues are values of k for which the interior transmission problem has a nontrivial<br />

solution. Although theoretical results about existence of transmission eigenvalues and the fact that they<br />

form a discrete set has been proven in many articles, a few papers consider the computation of transmission<br />

eigenvalues for general geometry and even less in electromagnetics.<br />

The method we shall use here is based on the CESC code developped by the <strong>CERFACS</strong> which combines<br />

integral equations and finite elements. It first consists in expressing the solutions (E,E 0 ) of the previous<br />

interior transmission problem with integral equations. ( Then the ) boundary conditions (2.3) and (2.4) lead<br />

J<br />

to solve a system of the form Z k X = 0 where X = with J = −ν × curlE and M = ν × E.<br />

M<br />

Transmission eigenvalues are values of k for which 0 is an engenvalue of Z k . The main difficulty is that<br />

the operator Z k is compact and therefore 0 is an accumulation point of its eigenvalues. As a consequence,<br />

the real eigenvalue 0 is “lost” numerically in the accumulation region. To get around this difficulty, we<br />

use a preconditioner B k to shift the accumulation to 1 and we solve the generalized eigenvalue problem<br />

Z k X = λB k X. We shall discuss proper choices of the operator B k .<br />

Another way to compute the transmission eigenvalues is to use far field data and the Linear Sampling<br />

Method. The Linear Sampling Method is based on solving an ill-posed far field equation using Tikhonov<br />

regularization. This method is usually used to determinate the shape of an obstacle. However, it can be<br />

shown that when k is a transmission eigenvalue for a sampling point z ∈ D the norm of the regularized<br />

solution to the far field equation cannot be bounded as the parameter tends to zero. Thus this property<br />

provides us another method to find transmission eigenvalues as the location of the peaks on the plot of the<br />

regularized solution’s norm against k. Figures 2.11 and 2.12 demonstrate the concordance of the results<br />

provided by both methods.<br />

<strong>CERFACS</strong> ACTIVITY REPORT 51

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!