An Efficient EM Modeling Scheme for Large 3-D Models-A Magnetotelluric Case Study

We present an efficient scheme for computing 3-D magnetotelluric (MT) forward responses. The scheme is especially valuable for large models resulting from fine discretization or the large survey area. The proposed approach overcomes the iterative solvers' slow convergence that occurs in large modeling problems due to a sizeable ill-conditioned system matrix that needs to be solved. Primarily, the slow convergence arises due to the grid stretching that is necessary to apply the boundary conditions (BCs). Our approach partly removes the grid stretching, thus improving the computational efficiency. In this scheme, a model is represented using two different meshes. One is a coarse mesh with grid stretching, and another is a fine mesh of the desired discretization excluding grid stretching. Using the electric field computed for the coarse mesh, a radiation boundary (RB) vector is calculated at the outer boundary of the fine mesh and is used to compute the necessary BCs along with an initial guess to be utilized by the iterative solver for the fine mesh. The RB vector can be computed at any arbitrarily shaped interface, thus allowing more flexibility in the shape of the fine mesh boundary. It is a significant advantage compared to the traditional finite difference (FD)-based algorithms where the boundaries must be same as the cuboid surfaces. Through different resistivity models, both synthetic and real, we demonstrate that the proposed approach improves the computational efficiency without compromising the accuracy of the solution while providing more flexibility in the shape of the fine mesh.