Next: Kernels for
Up: Initial configurations and kernel
Previous: Truncated equations
  Contents
For
we have the truncated equation
 |
(688) |
Normalizing the exponential of the solution gives
 |
(689) |
or
 |
(690) |
Notice that normalizing
according to Eq. (689)
after each iteration
the truncated equation (688)
is equivalent to a one-step iteration with uniform
=
according to
 |
(691) |
where only
is missing from the
nontruncated equation (683),
because the additional
-independent term
becomes inessential if
is normalized afterwards.
Lets us consider as example the choice
=
for uniform initial
corresponding to a normalized
and
=
(e.g., a differential operator).
Uniform
means uniform
,
assuming that
exists
and, according to Eq. (137),
=
for
=
.
Thus, the Hessian (161) at
is found as
 |
(692) |
which can be invertible due to the presence of the second term.
Another possibility is to start with an
approximate empirical log-density, defined as
 |
(693) |
with
given in Eq. (239).
Analogously to Eq. (686),
the empirical log-density may for example also be smoothed
and correctly normalized again,
resulting in an initial guess,
 |
(694) |
Similarly, one may let a kernel
,
or its normalized version
defined below in Eq. (698),
act on
first and then take the logarithm
 |
(695) |
Because already
is typically
nonzero it is most times not necessary to work here
with
.
Like in the next section
may be also be replaced by
as defined in Eq. (238).
Next: Kernels for
Up: Initial configurations and kernel
Previous: Truncated equations
  Contents
Joerg_Lemm
2001-01-21