Steepest-Descent Method: This chapter introduces the optimization method known as steepest descent (SD), in which the solution is found by searching iteratively along the negative gradient-g direction, the path of steepest descent. Relative to the Newton method for large problems, SD is inexpensive computationally because the Hessian inverse is not needed, but it can suffer from slow convergence with ill-conditioned Hessians because it does not take into account information about curvature. Nevertheless, experience shows that iterative SD with preconditioning and regularization can be quite useful when combined with multiscale methods for solving large seismic inverse problems.
Figures & Tables
This book describes the theory and practice of inverting seismic data for the subsurface rock properties of the earth. The primary application is for inverting reflection and/or transmission data from engineering or exploration surveys, but the methods described also can be used for earthquake studies. I have written this book with the hope that it will be largely comprehensible to scientists and advanced students in engineering, earth sciences, and physics. It is desirable that the reader has some familiarity with certain aspects of numerical computation, such as finite-difference solutions to partial differential equations, numerical linear algebra, and the basic physics of wave propagation (e.g., Snell’s law and ray tracing). For those not familiar with the terminology and methods of seismic exploration, a brief introduction is provided in the Appendix of Chapter 1. Computational labs are provided for most of the chapters, and some field data labs are given as well. Matlab and Fortran labs at the end of some chapters are used to deepen the reader’s understanding of the concepts and their implementation. Such exercises are introduced early and geophysical applications are presented in every chapter. For the non-geophysicist, geophysical concepts are introduced with intuitive arguments, and their description by rigorous theory is deferred to later chapters.