Basic operators and adjoints

Basic operators and adjoints

Jon Claerbout

A great many of the calculations we do in science and engineering are really matrix multiplication in disguise. The first goal of this chapter is to unmask the disguise by showing many examples. Second, we see how the adjoint operator (matrix transpose) back projects information from data to the underlying model.

Geophysical modeling calculations generally use linear operators that predict data from models. Our usual task is to find the inverse of these calculations; i.e., to find models (or make images) from the data. Logically, the adjoint is the first step and a part of all subsequent steps in this inversion process. Surprisingly, in practice the adjoint sometimes does a better job than the inverse! This is because the adjoint operator tolerates imperfections in the data and does not demand that the data provide full information.

Using the methods of this chapter, you will find that once you grasp the relationship between operators in general and their adjoints, you can obtain the adjoint just as soon as you have learned how to code the modeling operator.

If you will permit me a poet's license with words, I will offer you the following table of operators and their adjoints:

	matrix multiply		conjugate-transpose matrix multiply
	convolve		crosscorrelate
	truncate		zero pad
	replicate, scatter, spray		sum or stack
	spray into neighborhoods		sum within bins
	derivative (slope)		negative derivative
	causal integration		anticausal integration
	add functions		do integrals
	assignment statements		added terms
	plane-wave superposition		slant stack / beam form
	superpose curves		sum along a curve
	stretch		squeeze
	scalar field gradient		negative of vector field divergence
	upward continue		downward continue
	diffraction modeling		imaging by migration
	hyperbola modeling		CDP stacking
	ray tracing		tomography

The left column above is often called ``modeling,'' and the adjoint operators on the right are often used in ``data processing.''

When the adjoint operator is not an adequate approximation to the inverse, then you apply the techniques of fitting and optimization explained in Chapter . These techniques require iterative use of the modeling operator and its adjoint.

The adjoint operator is sometimes called the ``back projection'' operator because information propagated in one direction (earth to data) is projected backward (data to earth model). Using complex-valued operators, the transpose and complex conjugate go together; and in Fourier analysis, taking the complex conjugate of $\exp(i\omega t)$ reverses the sense of time. With more poetic license, I say that adjoint operators undo the time and phase shifts of modeling operators. The inverse operator does this too, but it also divides out the color. For example, when linear interpolation is done, then high frequencies are smoothed out, so inverse interpolation must restore them. You can imagine the possibilities for noise amplification. That is why adjoints are safer than inverses. But nature determines in each application what is the best operator to use, and whether to stop after the adjoint, to go the whole way to the inverse, or to stop partway.

The operators and adjoints above transform vectors to other vectors. They also transform data planes to model planes, volumes, etc. A mathematical operator transforms an ``abstract vector'' which might be packed full of volumes of information like television signals (time series) can pack together a movie, a sequence of frames. We can always think of the operator as being a matrix but the matrix can be truly huge (and nearly empty). When the vectors transformed by the matrices are large like geophysical data set sizes then the matrix sizes are ``large squared,'' far too big for computers. Thus although we can always think of an operator as a matrix, in practice, we handle an operator differently. Each practical application requires the practitioner to prepare two computer programs. One performs the matrix multiply $\bold y =\bold A \bold x$ and another multiplys by the transpose $\tilde{\bold x} =\bold A' \bold y$ (without ever having the matrix itself in memory). It is always easy to transpose a matrix. It is less easy to take a computer program that does $\bold y =\bold A \bold x$ and convert it to another to do $\tilde{\bold x} =\bold A' \bold y$. In this chapter are many examples of increasing complexity. At the end of the chapter we will see a test for any program pair to see whether the operators $\bold A$ and $\bold A'$ are mutually adjoint as they should be. Doing the job correctly (coding adjoints without making approximations) will reward us later when we tackle model and image estimation problems.

Basic operators and adjoints