Next: CODING THE TOMOGRAPHY
Up: Claerbout: Reflection tomography
Previous: INTRODUCTION
The value of extending the data should not be judged
on results shown above.
First, the extension methodology is new and there was no time
to tune the parameters.
Another unfamiliar new problem is that logarithms have an additive constant
that I was unsure of how to optimize.
My motivation for the extension was caused by observing steady growth
of side boundary effects (midpoint axis) as the number of iterations
in the tomography problem increased.
Keep in mind the background that
this data set is points so assuming we seek
a model space of the same number of points,
the solver theoretically requires 18,000 iterations.
My experimental work was limited to 5-25 iterations.
A question in my mind was and is,
how close are these solutions to the final limit?
I hypothesize that truncation of a data set could severely
limit the rate of convergence during the first 5-25 iterations
and that extending the data would accelerate convergence.
Time did not allow adequate testing of this interesting hypothesis.
I also saw that the solution
deteriorates in an interesting manner at high iterations.
Damping can be installed in the tomography,
but I had no time to experiment or think of the theory.
I wondered how the above problems would be affected by
extending the data, which seemed to help, but which
seems to have been less important than other variables,
particularly, the iteration count.
An overarching doubt is whether the underlying absorption model
warrants any more effort, since I believe that focusing
is a more likely model.
Next: CODING THE TOMOGRAPHY
Up: Claerbout: Reflection tomography
Previous: INTRODUCTION
Stanford Exploration Project
11/16/1997