# Contents of /release/4.0/doc/inversion/Forward2DMTTEMode.tex

Revision 5380 - (show annotations)
Tue Dec 16 00:12:58 2014 UTC (4 years, 4 months ago) by jduplessis
File MIME type: application/x-tex
File size: 8248 byte(s)
doco changes and example cleanup

 1 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 2 % Copyright (c) 2003-2014 by University of Queensland 3 4 % 5 % Primary Business: Queensland, Australia 6 % Licensed under the Open Software License version 3.0 7 8 % 9 % Development until 2012 by Earth Systems Science Computational Center (ESSCC) 10 % Development 2012-2013 by School of Earth Sciences 11 % Development from 2014 by Centre for Geoscience Computing (GeoComp) 12 % 13 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% 14 15 16 \section{MT inversion: 2D TE Mode}\label{sec:forward 2DMT TEMode} 17 In this section we present the forward model for the magnetotelluric (MT)\index{magnetotelluric}\index{MT}inversion case~\cite{ChaveJones2012a} 18 where we particularly looking into the two-dimensional case of the electrical field polarization\index{electrical field polarization}. 19 For this case the horizontal electrical field $E_x$\index{electrical field} and the magnetic field $H_y$\index{magnetic field} component 20 for a given frequency $\omega$ are measured and the horizontal impedance $Z_{xy}$\index{impedance} is recorded as the ratio of 21 electrical and magentical field: 22 \begin{equation}\label{ref:2DMTTE:EQU:1} 23 E_x = Z_{xy} \cdot H_y 24 \end{equation} 25 It is common practice to record the impedance using the apparent resistivity $\rho_a$\index{resistivity!apparent} and the 26 phase $\phi$ as 27 \begin{equation}\label{ref:2DMTTE:EQU:2} 28 Z_{xy} = (\rho_a \cdot \omega \cdot \mu) ^{\frac{1}{2}} \cdot e^{\mathbf{i} \phi} 29 \end{equation} 30 where $\mu$ is the permeability (e.g. $mu = 4 \pi \cdot 10^{-7} \frac{Vs}{Am}$). Notice that the impedance is independent of the scale of the $E_x$ and $E_y$. 31 The electrical field $E_x$ as function of depth $z$ and horizontal coordinate $y$ is given as the solution of the 32 PDE 33 \begin{equation}\label{ref:2DMTTE:EQU:3} 34 - E_{x,kk} + \mathbf{i} \omega \mu \sigma \cdot E_x = 0 35 \end{equation} 36 where $\mathbf{i}$ is the complex unit, and $k=0$ and $k=1$ correspond to the $y$ and $z$ direction. 37 $\sigma$ is the unknown conductivity to be calculated through the inversion. The domain 38 of the PDE is comprised of the subsurface region in which the conductivity is to be calculated 39 and a sufficient high airlayer in which the conductivity is assumed to be zero. 40 41 42 It is assumed that $E_x$ takes the value $1$ at the top of the domain $\Gamma_0$ representing an incoming 43 electro-magnetic wave front. On the other faces of the domain homogeneous Neuman conditions 44 are set to model the assumption that the electrical field is constant away from the domain. 45 \begin{equation}\label{ref:2DMTTE:EQU:4} 46 H_y = - \frac{1}{\mathbf{i} \omega \mu} E_{x,z} 47 \end{equation} 48 49 The impedance $Z_{xy}^{(s)}$ is measured at certain points ${\mathbf x}^{(s)}$. The defect 50 of the measurements for the prediction $E_x$ is given as 51 \begin{equation}\label{ref:2DMTTE:EQU:5} 52 J^{MT}(\sigma) = \frac{1}{2}\sum_{s} \int_{\Omega} 53 w^{(s)} \cdot \| E_x - Z_{xy}^{(s)} \cdot H_y \|^2 d{\mathbf x} 54 \end{equation} 55 where $w^{(s)}$ is a spatially dependent weighting faction. Here we use the Gaussian profile 56 \begin{equation}\label{ref:2DMTTE:EQU:6} 57 w^{(s)}({\mathbf x}) = \frac{w^{(s)}_0}{2\pi (\eta^{(s)})^2} \cdot e^{-\frac{\|{\mathbf x} - {\mathbf x}^{(s)} \|^2}{2(\eta^{(s)})^2}} 58 \end{equation} 59 where $\eta^{(s)}$ is measuring the spatial validity 60 and $w^{(s)}_0$ is confidence of the measurement $s$ (e.g. $w^{(s)}_0$ is set to the inverse of the measurement error. ) 61 62 \subsection{Usage} 63 64 \begin{classdesc}{MT2DModelTEMode}{domain, omega, x, Z_XY, eta, 65 \optional{, w0=1} 66 \optional{, mu=4E-7 * PI } 67 \optional{, coordinates=\None} 68 \optional{, fixAtBottom=False} 69 \optional{, tol=1e-8} 70 } 71 \end{classdesc} 72 73 \subsection{Gradient Calculation} 74 75 For the implemtation we set 76 \begin{equation}\label{ref:2DMTTE:EQU:100} 77 E_x = u_0 + \mathbf{i} \cdot u_1 78 \end{equation} 79 which translates teh forward model~\ref{ref:2DMTTE:EQU:3} 80 \begin{align}\label{ref:2DMTTE:EQU:103} 81 - u_{0,kk} - \omega \mu \sigma u_1 = 0 \\ 82 - u_{1,kk} + \omega \mu \sigma u_0 = 0 83 \end{align} 84 In weak form this takes the form 85 \begin{equation}\label{ref:2DMTTE:EQU:104} 86 \int_{\Omega} 87 \left( 88 v_{0,k}u_{0,k} 89 + v_{1,k}u_{1,k} 90 + \omega \mu \sigma \cdot ( v_1 u_0 - v_0 u_1) \right) dx =0 91 \end{equation} 92 for all test function $v_i$ with $v_i$ zero on $\Gamma_0$. 93 Using the complex data 94 \begin{equation}\label{ref:2DMTTE:EQU:105} 95 d^{(s)} = - \frac{1}{\mathbf{i} \omega \mu} \cdot Z_{xy}^{(s)} = d^{(s)}_0 + \mathbf{i} \cdot d_1^{(s)} 96 \end{equation} 97 the cost function~\ref{ref:2DMTTE:EQU:6} takes the form 98 \begin{equation}\label{ref:2DMTTE:EQU:106} 99 J^{MT}(\sigma)= 100 \frac{1}{2}\sum_{s} \int_{\Omega} w^{(s)} \cdot \left( (u_0- u_{0,1} \cdot d_0^{(s)} + u_{1,1} \cdot d_1^{(s)} )^2 101 + ( u_1- u_{0,1} \cdot d_1^{(s)} -u_{1,1} \cdot d_0^{(s)} )^2 \right) dx 102 \end{equation} 103 104 We need to calculate the gardient 105 $\frac{\partial J^{MT}}{\partial \sigma}$ of the cost function $J^{{MT}}$ with 106 respect to the susceptibility $\sigma$. We follow the concept as outlined in section~\ref{chapter:ref:inversion cost function:gradient}. 107 108 If $\Gamma_{\sigma}$ denotes the region of the domain where the susceptibility is 109 known and for a given direction $\hat{\sigma}$ with $\hat{\sigma}=0$ on $\Gamma_{\sigma}$ one has 110 \begin{align}\label{ref:2DMTTE:EQU:201aa} 111 \int_{\Omega} \frac{\partial J^{MT}}{\partial \sigma} \cdot \hat{\sigma} \; dx & = & 112 \sum_{s} \int_{\Omega} w^{(s)} \cdot \left( u_0 - u_{0,1} \cdot d_0^{(s)} + u_{1,1} \cdot d_1^{(s)} \right ) \left( \hat{u}_0 - \hat{u}_{0,1} \cdot d_0^{(s)} + \hat{u}_{1,1}\cdot d_1^{(s)} \right ) \\ 113 & + & w^{(s)} \cdot \left(u_1- u_{0,1} \cdot d_1^{(s)} -u_{1,1} \cdot d_0^{(s)} \right) \left(\hat{u}_1- \hat{u}_{0,1} \cdot d_1^{(s)} -\hat{u}_{1,1} \cdot d_0^{(s)} \right) \; dx 114 \end{align} 115 with 116 \begin{equation}\label{ref:2DMTTE:EQU:201A} 117 \int_{\Omega} 118 \left( 119 q_{0,k}\hat{u}_{0,k} 120 + q_{1,k}\hat{u}_{1,k} 121 + \omega \mu \sigma \cdot ( q_1 \hat{u}_0 - q_0 \hat{u}_1) 122 + \omega \mu \hat{\sigma} \cdot ( q_1 u_0 - q_0 u_1) \right) dx =0 123 \end{equation} 124 for all $q_i$ with $q_i=0$ on $\Gamma_{0}$. This equation is obtained from equation~(\ref{ref:2DMTTE:EQU:104}). 125 With 126 \begin{align}\label{ref:2DMTTE:EQU:202b} 127 Y_0 = & \sum_{s} w^{(s)} \cdot \left( u_0 - u_{0,1} \cdot d_0^{(s)} + u_{1,1} \cdot d_1^{(s)} \right) \\ 128 Y_1 = & \sum_{s} w^{(s)} \cdot \left( u_1 - u_{0,1} \cdot d_1^{(s)} - u_{1,1} \cdot d_0^{(s)} \right) \\ 129 X_{01} = & \sum_{s} w^{(s)} \cdot \left( u_{0,1} \cdot ( ( d_0^{(s)})^2 + (d_1^{(s)})^2) - u_0 \cdot d_0^{(s)} - u_1 \cdot d_1^{(s)} \right) \\ 130 X_{11} = & \sum_{s} w^{(s)} \cdot \left( u_{1,1} \cdot ( ( d_0^{(s)})^2 + (d_1^{(s)})^2) + u_0 \cdot d_1^{(s)} - u_1 \cdot d_0^{(s)} \right) 131 \end{align} 132 and $X_{00}=X_{10}=0$ we can write Equation~(\ref{ref:2DMTTE:EQU:201aa}) as 133 \begin{equation}\label{ref:2DMTTE:EQU:202c} 134 \int_{\Omega} \frac{\partial J^{MT}}{\partial \sigma} \cdot \hat{\sigma} \; dx = 135 \int_{\Omega} Y_i \hat{u}_i + X_{ij} \hat{u}_{i,j} \; dx 136 \end{equation} 137 We then set adjoint function $u_i^*$ with $u_i^*=0$ on $\Gamma_{0}$ as the solution of the variational problem 138 \begin{equation}\label{ref:2DMTTE:EQU:202d} 139 \int_{\Omega} 140 \left( 141 u^*_{0,k} v_{0,k} 142 + u^*_{1,k} v_{1,k} 143 + \omega \mu \sigma \cdot ( u^*_1 v_0 - u^*_0 v_1) \right) dx = \int_{\Omega} Y_i v_i + X_{ij} v_{i,j} \; dx 144 \end{equation} 145 for all $v_i$ with $v_i=0$ on $\Gamma_{0}$. Setting $q=u^*$ in Equation~(\ref{ref:2DMTTE:EQU:201A}) and 146 $v_i=\hat{u}_i$ in Equation~(\ref{ref:2DMTTE:EQU:202d}) Equation~(\ref{ref:2DMTTE:EQU:201A}) gives 147 \begin{equation}\label{ref:2DMTTE:EQU:290} 148 \int_{\Omega} \omega \mu \hat{\sigma} \cdot ( u^*_0 u_1 - u^*_1 u_0) \; dx 149 = \int_{\Omega} Y_i \hat{u}_i + X_{ij} \hat{u}_{i,j} \; dx 150 \end{equation} 151 This gives 152 \begin{equation}\label{ref:2DMTTE:EQU:300} 153 \int_{\Omega} \frac{\partial J^{MT}}{\partial \sigma} \cdot \hat{\sigma} \; dx = 154 \int_{\Omega} \omega \mu \hat{\sigma} \cdot ( u^*_0 u_1 - u^*_1 u_0) \; dx 155 \end{equation} 156 or 157 \begin{equation}\label{ref:2DMTTE:EQU:301} 158 \frac{\partial J^{MT}}{\partial \sigma} = \omega \mu \cdot ( u^*_0 u_1 - u^*_1 u_0) 159 \end{equation} 160