15.11.2014 Views

principles and applications of microearthquake networks

principles and applications of microearthquake networks

principles and applications of microearthquake networks

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

112 5. Inversion <strong>and</strong> Optimization<br />

number <strong>of</strong> unknowns is infinite. But because geophysical data are limited,<br />

the number <strong>of</strong> observations is finite. In other words, we write a finite<br />

number <strong>of</strong> equations corresponding to our observations, but our unknown<br />

vector x is <strong>of</strong> infinite dimension. In order to obtain a unique answer, the<br />

solution chosen is the one which maximizes or minimizes a subsidiary<br />

integral. In actual practice, we minimize a sum <strong>of</strong> squares to produce a<br />

smooth solution. A typical mathematical formulation may look like<br />

(5.32)<br />

where the top block A represents the underdetermined constraint equations<br />

with the observed data vector b, the vector x contains the unknowns,<br />

<strong>and</strong> the bottom block is a b<strong>and</strong> matrix which specifies that some filtered<br />

version <strong>of</strong>x should vanish. As pointed out by Claerbout (1976, p. 120), the<br />

choice <strong>of</strong> a filter is highly subjective, <strong>and</strong> the solution is <strong>of</strong>ten very sensitive<br />

to the filter chosen. The Backus-Gilbert inversion is intended for<br />

analysis <strong>of</strong> systems in which the unknowns are functions, i.e., they are<br />

infinite-dimensional, as opposed to, say, hypocenter parameters in the<br />

earthquake location problem, which are four-dimensional. For an application<br />

<strong>of</strong> the Backus-Gilbert inversion to travel time data, readers may<br />

refer, for example, to Chou <strong>and</strong> Booker (1979).<br />

5.1.3.<br />

Analysis <strong>of</strong> an Overdetermined System<br />

Most scientists are modest in their data modeling <strong>and</strong> would have more<br />

observations than unknowns in their model. On the surface it may seem<br />

very safe, <strong>and</strong> one should not expect difficulties in obtaining a reasonable<br />

solution for an overdetermined system. However, an overdetermined system<br />

may in fact be underdetermined because some <strong>of</strong> the equations may<br />

be superfluous <strong>and</strong> do not add anything new to the system. For example, if<br />

we use only first P-arrivals to locate an earthquake which is coniiderably<br />

outside a <strong>microearthquake</strong> network, the problem Is underdetermined no<br />

matter how many observations we have. In other words, in many physical<br />

situations we do not have sufficient information to solve our problem<br />

uniquely. Unfortunately, many scientists overlook this difficulty. Therefore,<br />

it is instructive to quote the general principle given by Lanczos<br />

(1961, p. 132) that “a lack <strong>of</strong> information cannot be remedied by any<br />

mathematical trickery. ”<br />

Usually a given overdetermined system is mathematically incompatible,<br />

i.e., some equations are contradictory because <strong>of</strong> errors in the observations.<br />

This means that we cannot make all the components <strong>of</strong> the re-

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!