Up: Expansions for nearly Gaussian
In Sect. 5, the relation (30) for the n-th derivative of a
composite function
is used
for the derivation of the Edgeworth asymptotic expansion.
Here a simplified derivation of Eq. (30) is given.
Petrov (1972, 1987) suggests a proof
by induction. We note that this derivation is more transparent
if one simply considers the Taylor expansion for
expressed
in terms of the Taylor expansions of f and g - see Bourbaki
(1958). We have
|  |
(A1) |
and
|  |
(A2) |
Truncating the expansions at some n and m we find that
|  |
|
| |
| (A3) |
| |
On the other hand, we can write down the Taylor series for the
composite function,
|  |
|
| (A4) |
Now using the polynomial theorem,
|  |
|
| |
| (A5) |
where summation extends over all sets of non-negative integers
satisfying r=k1+k2+ ... +ks,
and comparing the terms
with equal s in
(A4) and (A4), we obtain
|  |
|
| (A6) |
This is the relation (30) which we sought.
Here the set
consists of non-negative solutions
of the Diophantine equation
|  |
(A7) |
and r=k1+k2+ ... +ks.
Up: Expansions for nearly Gaussian
Copyright The European Southern Observatory (ESO)