Convolutions of Complex Exponential Functions
Frank Massey
Notation. The variable t is assumed to be restricted to nonnegative real values.
Prime and * denote differentiation and convolution with respect to t, e.g. f (t) = f/ t and
f(t) * g(t) = . Let F1(t;) = e-t and Fn(t;1,...,n) = Fn-1(t;1,...,n-1) * F1(t;n) and
En(t;1,...,n) = 1n Fn(t;1,...,n). Thus Fn(t;1,...,n) = e-1t * * e-nt is the
convolution of exponential functions e-jt. If we write Fn(t) or En(t) it is assumed that the
parameters are 1,...,n. En(t) and Fn(t) are symmetric functions of the j since
convolution is commutative and associative. Fn(t) = c1e-1t ++ cne-nt with
cj = [(1-j)(j-1-j)(j+1-j)(n-j)]-1 for distinct j while Fn(t;,...,) = tn-1e-t/(n-1)! for
equal j; see Ross [2, p. 272, 284].
In [1] we gave approximations to Fn(t) in the case of real j. Here we extend one
of the small time approximations to the case where the j may be complex, but occur in
complex conjugate pairs; see Propositions 3 and 4 below. We begin with a formula for
the convolution of two exponential functions with a pair of complex conjugate
parameters.
Proposition 1.
(1) 1 * f(t) =
(2) 1 * e-t =
(3) e-t * e-t =
(4) eit * e-it =
(5) e(-+it * e(--it =
1
Proof. (1) follows immediately from the definition of convolution. (2) follows
from (1). To prove (3) we use the following convolution identity
(6) [etf1(t)]*[etf2(t)]**[etfn(t)] = et [f1(t)*f2(t)**fn(t)]
which is proved in Theorem 1a of Section 2. If we use this and (2) we get
e-t * e-t = e-t [ 1 * e-(-)t ] = e-t [ ] =
To prove (4) we use (3) to get
eit * e-it = =
Finally (5) follows from (6) and (4). //
The next proposition shows that one can bound the convolution of a pair of
exponential functions with complex conjugate parameters in terms of the convolution of a
pair of exponential functions whose parameter is the real part of the complex parameters.
Note that te-t = e-t * e-t.
Proposition 2. If t > 0 and > 0 and is real then
(7) | sin t | < t
(8) cos t > 1 -
(9) sin t > t -
(10) < te-t
(11) > te-t [ 1 - ]
Proof. One has sin t = . Therefore | sin t | . Since | cos(s) | < 1 except for a
finite number of points on 0 s t, one obtains (7). Similarly cos t = 1 - . Using (7) to
estimate sin(s) one obtains (8). Combining (8) with sin t = gives (9). Finally (10) and
(11) follow from (7) and (9). //
2
The following Proposition shows that Fn(t;1,...,n) for the case where complex 's
occur in conjugate pairs can be bounded by Fn(t;1,...,n) where the j are the real parts
of the j.
Proposition 3. Suppose j = j + j where the j and j are real. Suppose j is
the complex conjugate of j-1 for j = 1, 2, , m and j is real for j = 2m + 1, , n. Let
m = 1 + 3 + + . Then
(12) | Fn(t;1,...,n) | Fn(t;1,...,n) for t 0
(13) Fn(t;1,...,n) Fn(t;1,...,n) [ 1 - ] for 0 t .
Proof. We prove this by induction on m and n. In the case m = 0 there is nothing
to prove since all the are real. Consider the case m = 1 and n = 2. By (5) one has
F2(t;1,2) = e(-+it * e(-+it = e(-+it * e(--it =
Using (10) and (11) and the fact that F2(t;1,2) = F2(t;1,1) = te-t one has
(14) | F2(t;1,2) | F2(t;1,2)
(15) F2(t;1,2) F2(t;1,2) [ 1 - ]
which proves (12) and (13) in the case m = 1 and n = 2.
Now suppose (12) and (13) are true for m = k and n = 2k, i.e.
(16) | F2k(t;1,...,2k) | F2k(t;1,...,2k) for t 0
(17) F2k(t;1,...,2k) F2k(t;1,...,2k) [ 1 - ] for 0 t .
We shall show (12) and (13) are true for m = k+1 and n = 2k+2. We have
(18) F2k+2(t;1,...,2k+2) = F2k(t;1,...,2k) * F2(t;2k+1,2k+2)
Therefore
| F2k+2(t;1,...,2k+2) | | F2k(t;1,...,2k) | * | F2(t;2k+1,2k+2) |
Using (14) and (16) gives
3
| F2k+2(t;1,...,2k+2) | F2k(t;1,...,2k) * F2(t;2k+1,2k+2) = F2k+2(t;1,...,2k+2)
which proves (12) for m = k+1 and n = 2k+2.
Now suppose 0 t . Then (17) holds and also
(19) F2(t;2k+1,2k+2) F2(t;2k+1,2k+2) [ 1 - ]
Note that the right sides of (17) and (19) are positive for t in the range being considered.
Using (17), (18) and (19) one obtains
F2k+2(t;1,...,2k+2)
[F2k(t;1,...,2k) ( 1 - )] * [F2(t;2k+1,2k+2) ( 1 - )]
F2k(t;1,...,2k) * F2(t;2k+1,2k+2) - [ t2F2k(t;1,...,2k) ] * F2k(t;1,...,2k) -
F2k(t;1,...,2k) * [ t2F2(t;2k+1,2k+2) ]
Since t2 is an increasing function, one has [t2f(t)]*g(t) t2 [f(t)*g(t)] if f(t) and g(t) are
non-negative. Therefore
F2k+2(t;1,...,2k+2)
F2k(t;1,...,2k) * F2(t;2k+1,2k+2) - [F2k(t;1,...,2k)*F2k(t;1,...,2k)] -
[F2k(t;1,...,2k)*F2(t;2k+1,2k+2)]
= F2k(t;1,...,2k) * F2(t;2k+1,2k+2)( 1 - )]
= F2k+2(t;1,...,2k+2) ( 1 - )]
which proves (13) for m = k+1 and n = 2k+2. Therefore (12) and (13) hold in the case
n = 2m.
Now we prove (12) and (13) hold for n > 2m. One has
(20) Fn(t;1,...,n) = F2m(t;1,...,2m) * Fn-m(t;2m+1,...,n)
= F2m(t;1,...,2m) * Fn-m(t;2m+1,...,n)
Therefore
| Fn(t;1,...,n) | | F2m(t;1,...,2m) | * Fn-m(t;2m+1,...,n)
F2m(t;1,...,2m) * Fn-m(t;2m+1,...,n)
4
= Fn(t;1,...,n)
where we have used the fact that (12) holds for n = 2m. So (12) holds for general n. In
order to prove (13) suppose 0 t . Since (13) holds when n = 2m one has
F2m(t;1,...,2m) F2m(t;1,...,2m) [ 1 - ]
and the right side is non-negative. Combining this with (20) gives
Fn(t;1,...,n) F2m(t;1,...,2m) * Fn-m(t;2m+1,...,n)
- [ t2F2m(t;1,...,2m) ] * Fn-m(t;2m+1,...,n)
Again using [t2f(t)]*g(t) t2 [f(t)*g(t)] for non-negative f(t) and g(t) one obtains
Fn(t;1,...,n) F2m(t;1,...,2m) * Fn-m(t;2m+1,...,n) [ 1 - ]
= Fn(t;1,...,n) [ 1 - ]
which proves (13) in general.
If we combine Proposition 3 with Theorem 4a in Section 2 one obtains the
following proposition which implies Fn(t) can be approximated by with a relative error
that is bounded by a constant times t2 for small t.
Proposition 4. Suppose j = j + j where the j and j are real. Suppose j is
the complex conjugate of j-1 for j = 1, 2, , m and j is real for j = 2m + 1, , n. Let
m be as in Proposition 3, * = min{1,...,n}, be the mean of 1,,n, be the variance
and . Then
| Fn(t;1,...,n) | (1 + A(t)) for t 0
Fn(t;1,...,n) [ 1 - ] for 0 t .
References.
[1] Massey, F. J. and J. Prentis. "Approximations for Radioactive Decay Series."
2007.
5
[2] Ross, Sheldon M. Introduction to Probability Models, eighth edition. Academic
Press, San Diego, 2003.
