Next: Data regularization as an
Up: Fundamentals of data regularization
Previous: Statistical estimation
In order to understand the structure of the matrices
and
, we need to make some assumptions about the
relationship between the true model
and the data
. A natural assumption is that if the model were known
exactly, the observed data would be related to it by a forward
interpolation operator
as follows:
|  |
(4) |
where
is an additive observational noise. For simplicity,
we can assume that the noise is uncorrelated and normally distributed
around zero:
|  |
(5) |
where
is an identity matrix of the data size, and
is a scalar. Assuming that there is no linear correlation
between the noise and the model, we arrive at the following
expressions for the second moment matrices in
formula (
):
|  |
(6) |
| ![\begin{displaymath}
\bold{C}_{md} = E\left[\bold{m}\,
\left(\bold{m}^T\,\bold{L}^T + \bold{n}^T\right)\right] =
\bold{C}_{m}\,\bold{L}^T\;.\end{displaymath}](img29.gif) |
(7) |
Substituting equations (
) and (
) into
(
), we finally obtain the following specialized form of
the Gauss-Markoff formula:
|  |
(8) |
Assuming that
is invertible, we can also rewrite
equation (
) in a mathematically equivalent form
|  |
(9) |
The equivalence of formulas (
) and (
)
follows from the simple matrix equality
|  |
(10) |
It is important to note an important difference between
equations (
) and (
): The inverted matrix
has data dimensions in the first case, and model dimensions in the
second case. I discuss the practical significance of this distinction
in Chapter
.
In order to simplify the model estimation problem further, we can
introduce a local differential operator
. A model
complies with the operator
if the residual after we apply
this operator
is uncorrelated and
normally distributed. This means that
| ![\begin{displaymath}
E\left[\bold{D}\,\bold{m}\,\bold{m}^T\,\bold{D}^T\right] =
\bold{D}\,\bold{C}_m\,\bold{D}^T = \sigma_m^2\,\bold{I}\;,\end{displaymath}](img36.gif) |
(11) |
where the identity matrix
has the model size. Furthermore,
assuming that
is invertible, we can represent
as follows:
|  |
(12) |
Substituting formula (
) into (
) and
(
), we can finally represent the model estimate in the
following equivalent forms:
|  |
(13) |
| (14) |
where
and
.
The first simplification step has now been accomplished. By
introducing additional assumptions, we have approximated the
covariance matrices
and
with the forward
interpolation operator
and the differential operator
. Both
and
act locally on the model.
Therefore, they are sparse, efficiently computed operators. Different
examples of operators
,
, and
are
discussed later in this dissertation. In the next section, I proceed
to the second simplification step.
Next: Data regularization as an
Up: Fundamentals of data regularization
Previous: Statistical estimation
Stanford Exploration Project
12/28/2000