You are on page 1of 7

Approximate Inversion of the

Laplace Transform

where
has th

Alexander H-D. Cheng, Paston Sidauruk, University of Delaware


Younane Abousleiman, University of 0 klahoma

ln[3]:=

F[

In[4]:=

Nl

The Laplace transform is often applied to linear partial differential equations to eliminate the time
dimension. The analytical solutions thus obtained need to be inverted to the time domain. This

OLlt[4]=

paper provides Matbematica functions for approximate inversion of the Laplace transform.

The e:

The Laplace transform


F(s) =

F(s) c

Smf(t)e%t
0

is often used in the solution of linear partial differential equations to eliminate the time dimension. The resulting systems
become easier to solve by analytical techniques [Detournay
and Cheng 19881, or numerical methods such as finite element or boundary element methods [Cheng and Ou 1989;
Cheng and Detournay
19881. The solutions thus obtained
need to be inverted to the time domain. The exact inversion
is normally difficult to carry out, so approximate
inversion
techniques are used.
There are many approximate
Laplace inversion
algorithms. (For comprehensive
bibliographies,
see [Piessens
197.5; Piessens and Dang 19761.) With issues of computational efficiency and accuracy in mind, a handful of techniques have been tested and compared [Cost 1964; Davies
and Martin 1979; Narayanan
and Beskos 19821. Based on
recommendations
in the literature and our own experience in
engineering applications,
we present here the Mathematics
package NLaplaceInversion .m, which implements five inversion
algorithms, one in each of the five categories classified by
Davies and Martin [ 19791.
It is important
to emphasize that the inverse of the

tions. In particular, functions of periodic nature, of spike


shapes, and whose Laplace transforms contain poles (singularities), are normally difficult to handle. With these limitations in mind, we present below the five approximate
Laplace inversion algorithms that are implemented
in our
package.

Paston Sidauruk holds a Masters degree in civil engineering from the University of
Delaware and is currently a Ph.D. candidate. He is on leave from the Hydrology
Section of the Atomic Energy Agency in Jakarta, Indonesia.
Younane Abousleiman holds a Ph.D. from the University of Delaware. He is currently a Senior Research Scientist at the Rock Mechanics Research Center at the
University of Oklahoma.

76 THE MA THEMA TICA JOURNAL 0 1994 Miller Freeman Publications

ln[6]:=

f1

Out[6]=

BY

expec
error
xq

ln[B]:=

yi7,cnF(*)

f(t) ==

n=l

where the coefficient


c,

(-1)

n+N/2

c, is given by

min( n, N/2)

x
k=(n+1)/2

/P/2(2@!
(N/2 - A$!Ic!(k - l)! (n - A$!(2k - n)!

The number of terms N in the series must be even.


The inverse function NLInvSteh can be defined as
In[l]:=

csteh[n_, i,] = (-l)^(i + n/2) Sum[ k^(n/2) (2k)! /


((n/2-k)!

k!

(k-l)! (i-k)! (2k-i)! >,

{k, Floor[(i+l)/2],Min[i, n/2] 1 I;


Alexander H-D. Cheng is a professor of civil engineering at the University of
Delaware. His research interests include groundwater, poroelasticity, boundary
element methods, and chaotic dynamics.

fI

The Stehfest method [Stehfest 19701 is a popular numerical


inversion techn ique used in groundwater flow and petroleum
reservoir engineering applications.
The inversion formula is
based on computing a sample of the time function using a
delta-convergent
series. The resultant series approximation
takes the following form:

Laplace transform is not stable under reasonable perturbations [Bellman et al. 19661. This can be seen by considering

the function f(t) = sin at, a > 0, and its Laplace transform
F(s) = a/( a2 + s2). While F(s) can be made arbitrarily small
by taking a sufficiently large, the inverse transform f(t) oscillates between -1 and 1. It is therefore impossible to devise a
universal algorithm that performs well for all types of func-

ln[5]:=

ln[7]:=

Stehfest Method

ln[z]:=

NLInvSteh[F_, s_, t,, n-1 :=


Log[2]/t Sum[ csteh[n,i] F /. s -> i

{i, 1, n) 1 // N
For example,

the function

Log[2l/t,

Th
aboul
event
asym
accur
diver1
range
Fig

(2) ( !?S
for tl
tions
that
enhai

2Kdm
F(s) =s

(1)

where K0 is the modified Bessel function


has the exact Laplace inversion

of the second kind,

&f(t)= -E&-l/t)
F(s) can be numerically

(2)

inverted for the solution

at t = 1:

ln[3]:=
F[s_] = (2/s) BesselK[O, 2 Sqrt[s]];
ln[4]:=
Out[4]=

NLInvSteh[F[s],s,

I, 61

Papoulis

Legendre

Polynomial

Method

0.223359

The exact solution

The Papoulis method [Papoulis


sion of f(t) into the exponential

is

ln[5]:=

f [t,] = - ExpIntegralEi[-l/t ] // N;

ln[6]:=

f [I]

Out[6]=

of the approximation
in the small time range is quite amazing.
Solutions with behavior similar to that of (2) often arise in
groundwater
and reservoir engineering applications. In fact,
expression (2) is known as the well function in groundwater literature. It corresponds to a sink in a confined, compressible aquifer. For complex aquifers, such as a multiplelayered aquifer system, the solution is available only in the
Laplace transform domain [Cheng and Morohunfola
19931.
The Stehfest method is quite handy to invert the solution to
the time domain.

19.571 is based on the expanseries:

f(t) =$$?2&~~)
n=O

where Pn(x) is the Legendre polynomial,


the recursion formula

0.219384

an is computed

from

By increasing the number of terms in the series, the error is


expected to diminish. The following plot shows the relative
error versus the number of terms N in log scale:
tn[7]:=
<< GraphicsGraphics'
h-(8]:=

and (i)n is the Pochhammer

LogListPlotC

ci)n

Table[{n, Abs[ (f[l] - NLInvSteh[F[s],s, 1, n]>/f[l] I),


{n, 6, 35, 2)

Prolog -> AbsolutePointSize[3]1

ln[g]:=

apap[F_, s_, k_, r-1 := apap[F, s, k, r] =


( (r F /. s -> (2k + 1) r > -

O.l}

f
1,

1.5

(2 Pochhammer[k + l/2, m + 11) apap[F, s, m, r],

,-.

10

Sum[ Pochhammer[k - m + I, m] /

. .
l

1.

forn>O

10.

FIGURE

. . . (j+n-1)

The coefficient T is an arbitrary number defining the exponential series. A default value of r = 0.2 is selected for this
package based on a recommendation
in [Davies and Martin
l979], but users can select their own T values. The inverse
function NLInvPap is then provided as:

Relative Error

0.00001

for n=O

= j(j+l)

1 // 1,

AxesLabel -> {"N", "Relative Error"},

0.001

symbol:

{m, 0, k - 1) 1 > 2 Pochhammer[k + l/2, k + 11 /

20

2.5

30

Pochhammer[i, k]

,N

Convergence of the Stehfest method.

In[lO]:=

apap[F_, s_, 0, r_] := r F /. s -> r

In[ll]:=

NLInvPap[F_, s_, t_, n_, r-:0.2] :=


Sum[ apap[F, s, k, r] LegendreP[2k, Exp[- r t]],

The convergence to the exact solution is excellent until


about N = 20. The numerical solution then deteriorates and
eventually blows up. This type of behavior is similar to an
asymptotic expansion series, in that it is highly efficient and
accurate when a modest number of terms are used, but
diverges as N -+ 00. For engineering purposes, we suggest the
range 6 i N I 20.
Figure 2 shows the comparison between the exact result
(2) (solid line) and the numerical inversion (long dashed line)
for the range 0.1 i t 2 100, with N = 14. These two solutions practically overlap in the figure. We note in particular
that Figure 2 is presented
in logarithmic
scale, which
enhances the detail of the small-time solution. The accuracy

{k, 0, n)

1 // N

In Figure 2, the dotted line shows the Papoulis inversion


the function (1) using N = 20.
ln[lZ]:=

of

LOgLOgPlOt[
{NLInvSteh[F[s],s, t, 141, NLInvPap[F[s], s, t, 201,

flm

{t, 0.1, 1001,

PlotRange -> ((0.1,

IOO},{O.OOOOOl, loo)),

PlotPoints -> 60,


PlotStyle -> Dashing /Q C(O.03, 0.03}, (0.005, O.Ol),
{I, 0%
AxesLabel -> {"t", "f(t)"} 1

VOLUME 4, ISSUE 2

77

FIGURE

2.

Comparison

of Stehfest

results with exact solution (solid line),

(dashed line) and Papoulis (dotted

line) inversion

f(t) = -Ei(-1 /t).

We observe that it is inferior to the Stehfest result at both


small and large times.
Figure 3 shows the inversion
for the Laplace transform
pair

FIGURE 4. Comparison of Stehfest


results with exact solution (solid line)

(dashed line) and Papoulis (dotted

line) inversion

The approximating
function,
with period T = 4 t,,, , is represented by its Fourier Series. The coefficients
of the Laplace
transform
are equated to samples of the transform
F(s):

f(t) =

${y +

+ y)cos(y)]

f)e[F(a

-&

(3a )

= sin t

(3b)

W) =
f(t)

Figure 4 displays the results for the Stehfest


inversion.
They are both unsatisfactory.

-Im[F(a

where

(4)

+ y)sin(F)]/

i = 11-1. Th e parameter
a

=a

5.

a can be chosen

as

-- 1nE

2T

where E is the error tolerance,


for
E = 1 x lo-* has been assigned,
and
leading
pole of the function
F(s).
poles, the default value a = 0 has
code implements
the Durbin method:

Next,

We note
s=l.
-I

k=l

using the Stehfest and the Papoulis methods.


The results are
excellent in both cases. It is not surprising
that the Papoulis
method works well for this problem,
since its bases are exponential functions.
For many diffusion problems,
the solution
is expected
to contain multiple components
of exponential
decay. The exponential
series based inversion techniques
are
highly efficient in those applications.
Next, we test the sine function:

FIGURE

terms. The e

f(t) = sin

which the default value


a is the real part of the
For functions
without
been set. The following

NLInvDurt

the dot
demons
and the
Despi
that any
tally in\
sion. Tfi
ema tica
method
The
Durbin
19641. r

and the Papoulis


adurb[tmax_, alpha_, tol_]

:= adurb[tmax,

alpha, toll

alpha - Log[tol] / (8 tmax) // N


Durbin-Crump

Method

cdurb[F_, s_, tmax_, alpha_, tol_, k-1 :=

The Durbin method


[Durbin 19761 approximates
the time
function f(t) by a periodic function
of period T. The user is
asked to provide
the parameter
t,,, , specifying
that the
inverse tranform
will only be used in the interval [0,4 t,,,].

F /. s -> adurb[tmax, alpha, toll + I Pi k / (4 tmax) // N


cdurb[F_, s_, tmax_, alpha_, tol_,

01 :=

F/2 /. s -> adurb[tmax, alpha, toll // N


NLInvDurb[F_, s_, t_, j_, tmax_, alpha_:O, to1_:0.00000001] :=
NLInvDurb[F, s, t, j, tmax, alpha, toll =

f(t)

Exp[ adurb[tmax, alpha, toll t I / (4 tmax) *

1
0.5

Sum[ Re[ cdurb[F, s, tmax, alpha, tol, k] I


Cos[k Pit /(4 tmax)] -

0.1
0.05

Im[ cdurb[F, s, tmax, alpha, tol, k]

Sin[k Pit /(4 tmax)],

0.01
0.005

{k, 0,

j) 1 // N

0.001

3. Comparison
of Stehfest (dashed line) and Papoulis (dotted
results with exact solution (solid line), f(t) = te-

FIGURE

78

THE MA THEMA TICA JOURNAL 0

1994 Miller Freeman Publications

line) inversion

The Durbin
method
has the same deficiency
as a Fourier
series: it converges slowly. Figure 5 shows the convergence
of
the Durbin inversion
of the function
(3a) to the exact solution with 20 and 50 terms. The SO-term result practically
coincides with the exact solution.

where t
quantit
note he
cedure j

epsilon1

epsilon1
epsilo

( epsi
eP

epsilon1

epsilon[

NLInvD

NLInvCrm

epsilo

The op
Crump

For the case of the sine function (3), the Crump method
with N = 20 produces results comparable with the Durbin
method using N = 50. However, the need to store a large
number of li(m) quantities in memory to avoid re-computation
makes the Crump algorithm

memory-inefficient.

f(t)

150:t

12s.
100:
75 -

FIGURE 5. Convergence of the Durbin inversion with 20 (dashed line) and 50 (dotted line)
terms. The exact solution is f(t) = sin t (solid line).

SO2s :

Next, the test is extended

to the Laplace transform

pair:
1

(5a )

We note in particular that the function (5a) has a pole at


s = 1. The proper inversion
requires the use of a = 1 in
NLInvDurb. The result using N = 100 is shown in Figure 6 as
the dotted line. The result is reasonable.
Although
not
demonstrated
in the figure, we note that both the Stehfest
and the Papoulis method badly fail in this situation.
Despite the reasonable approximation,
it is recommended
that any poles in a function should be removed and analytically inverted prior to the execution of an approximate inversion. The proper tool for extracting the poles is again Muthematica. In those cases, even the Stehfest and the Papoulis
method can perform a proper inversion.
The Crump method [Crump 19761 is an accelerated
Durbin method that uses the epsilon-algorithm
[MacDonald
19641. The algorithm is given as

St

FIGURE 6. Comparison of Durbin (dotted line) and Weeks (dashed line) inversion results
with the exact solution f(t) = et (solid line).

Weeks

Method

The Weeks method [Weeks 19661 is classified as a method


that uses a bilinear transformation
of s [Davies and Martin
19791. The approximate inversion formula is

where Lk are the Laguerre


are

polynomials,

and the parameters

with a the leading pole of the function F(s), and H the Heaviside unit step function. The coefficients ak are
where $7) = 0 and &brn)is the mth partial sum of (4). The
(O) hence represents the extrapolated
result. We
quantity EN
note here that N must be an even number. The Crump procedure is implemented as:

N
a()

N+l

h(0 j )
j=o
N

epsilon[F_, s_, t_, tmax_, alpha_, Ml_, i_, m-1 :=

al,

epsilon[F, s, t, tmax, alpha, tol, i, ml =

2, m + II + 1 /
( epsilon[F, s, t, tmax, alpha, tol, i - 1, m + 11 epsilon[ F, s, t, tmax, alpha, tol, i - 1, ml >

2
N+l

epsilon[F, s, t, tmax, alpha, tol, i -

(+----

alpha_, toI_, 0, m-1=

NLInvDurb[F, s, t, m, tmax, alpha, toll

h(O) -

NLInvCrmp[F_, s_, t_, n_, tmax_, alpha_:O, tol_:O.OOOOOOO~l :=


epsilon[F, s, t, 0.2 tmax, alpha, tol, n, 01

n 2j+l
2 N+l

)I

-cot
$Im
Fc+& )II
cot :

[i

The optimal value for T is fixed at 0.8 t,,,


Crump [1976].

N#O

j=o

in which

epsilon[F_, s_, t_, tmax_, alpha_, tol_, -1, m-1 := 0


epsilon[F_, s_, t_, tmax_,

h(8,) cos(ke,),

, as suggested by

(6

The code for NLInvWks is given as:

VOLUME 4, ISSUE 2

79

talk,

tmax_] := tn[n, tmax] = tmax / n

NLInvPsn
SumC

cuks[tmax_, alpha_] := cuks[tmax, alpha] =

0.8

If[ (alpha + 1 / tmax) >= 0, alpha + 1 / tmax, 0 ]

0.6

thetab_, j-1

:= (Pi/2) (2 j + I) / (n + 1)

0.4

h[F_, s_, n_, tmax_,alpha_, j-1 :=


h[F, s, n, tmax, alpha, j] =
l/(2 tn[n, tmax]) (Re[F] - CotI theta[n, j]/2

1 Im[F]) /.

s -> cuks[tmax, alpha] + I /


(2 tn[n, tmax]) Cot[ thetarn, j]/2 I // N
FIGURE 7. Comparison of Weeks (dotted line) and Piessens (dashed
results with the exact solution f(t) = Jo(t)(solid line).

auks[F_, s_, n_, tmax_, alpha_, k_] :=


2/(n + I) Sum[ h[F, s, n, tmax, alpha, j]
Cod

k thetaIn, jl

I, {j, 0, n) 1

auks[F_, s_, n_, tmax_, alpha_, 01

in which

:=

l/(n + 1) SumI h[F, s, n, tmax, alpha, j],

{j, 0, n)l

NLInvWks[F_, s_, t_, n_, tmax,, alpha_:01 :=

Exp[( cuks[tmax, alpha] - 1 / (2 tn[n, tmaxl) > t]

ai =

(2i + j3 - 3)(2i + /3 - 2)
i+/L--2

bi =

(2i + j3 - 3)(2 - /5)


(i+B-2)(2i+/3-4)

Ci

(2i + /3 - 2)(i - 1)
(i+j3--2)(2i+/3--4)

Sum[ auks[F, s, n, tmax, alpha, k]


LaguerreL[ k, t / tn[n, tmax]

line) inversion

1, {k, 0, n) 1 // N

The Weeks method received one of the highest ratings in


[Davies and Martin 19791 for its overall accuracy tested
against an assorted group of functions. Indeed, using merely
10 terms (and a = l), the method inverts (5a) quite accurately. The result is presented in dashed line in Figure 6,
which virtually coincides with the exact solution.
Another tough test is provided by the Bessel function J.
For the pair:

The weights uti are given by

wi

(N-l)!

(-y-1

T(N + B - I)Nx

[2N+B_2]

pN-l(l/Xi)

where I is the Gamma function.


The complete code is the following:
apsns[i_, beta_] := apsns[i, beta] =

f(t) = Jo(t)
the inverted result (dotted line) is displayed
solution in Figure 7.
Piessens

Gaussian

Quadrature

VW
against the exact

Method

where B is a positive real constant such that #F(s) is analytic


and has no branch point at infinity. The abscissas Q,
i = 1, . . . ) N, are the zeros of the orthogonal
polynomials
&(x). The polynomials Pi(z) are related to the generalized
Bessel polynomials
and are orthogonal with respect to the
weight sp. They can be evaluated from the recursive relation:

(2 i + beta - 3) (2 - beta) /

(2 i + beta - 2) (i - 1) /
( (i + beta - 2) (2 i + beta - 4) >
p[i_, beta_, x-1 :=
(apsndi, beta] x + bpsndi, beta]) p[i - I, beta,

/3x - 1

Pi(X)

(U$ + bJ Pi-1 (x) + ci P&x),

1994 Miller Freeman Publications

FIGURE
(dotted lini

p[l,

beta_,

x,] := beta x - I

proot[n_, beta_] := proot[n, beta] =


NSolveE p[n, beta, l/x] == 0, x ]
xi[n_, beta_, i,] := x /. proot[n, beta] [[i]]
ui[n_, beta_, i,] := ui[n, beta, i] =

- l)(n-I)!/

(n Gamma[n + beta - 11 xi[n, beta, ij-2 >

i2 2

x] +

cpsns[i, beta] p[i - 2, beta, x ]

(-l>^(n

THE MA THEMA TICA JOURNAL 0

bpsns[i_, beta_] := bpsns[i, beta] =

p[O, beta,, x-1 := I

i=l

80

(2 i + beta - 3) (2 i + beta - 2) / (i + beta - 2)

cpsns[i_, beta_] := cpsns[i, beta] =

fo=&wixiF(y

PI(X)

Due
point 2
branch
The prc
ure 8 s
default
in daslusing I\
essenti;
the fun

( (i + beta - 2) (2 i + beta - 4) >

The Piessens method seeks to evaluate the Bromwich integral


by numerical quadrature
[Piessens 1971; 19731. The time
function is expressed as a Gaussian quadrature sum:

PO(X) =

We not
To tc
using 1
line, WI
the opti
ber of t
represe
Nth or
sBF(s)
Maclau
with sl(
to the f
recomn
As tl
pan-:

( (2 n + beta - 2) /

p[n - 1, beta,

l/xi[n,

beta, i] ] )-2 // N

WLInvPsns[F_, s_, t_, n_, beta_:11 := (l/t)


Sum[ ui[n, beta,

Rd

i.] xi[n, beta, i]^beta

F /. s -> (xi[n, beta,

ilk), {i, 1, d 1 1 // N

We note that the default value p = 1 has been set.


To test the algorithm, the inversion of (7a) is performed
using 12 terms. The result is presented in Figure 7 in dashed
line, which overlaps the exact solution. In fact, this result is
the optimal one. The solution rapidly deteriorates if the number of terms is increased. The probable cause is the inefficient
representation
of (7a) by a polynomial
series in l/s. (The
Nth order Gaussian quadrature
gives the exact result of
#F(s) if it is a polynomial
in l/s of degree 2N - 1.) The
Maclaurin series expansion of (7a) gives an alternating series
with slowly dying coefficients, which might have contributed
to the failure of the method. For engineering purposes, it is
recommended that N 2 10 be used for all functions.
As the final example, we choose the following transform
pair:
F( s=-e
)

1
21S

-l/s

f(t) -

LcosJa
?r
4-t

Conclusion

We have provided five approximate


inversion algorithms
(Stehfest, Papoulis, Durbin-Crump,
Weeks, Piessens), one
each in the five categories presented by Davies and Martin
[ 19791. Based on a set of assorted test functions, some algorithms were found to be more accurate, while others are
more efficient. The performance is dependent on the function
type. It is difficult to recommend just one inversion method.
The user should have some idea of the behavior of the function in order to choose the proper method. When in doubt,
multiple algorithms should be used and their results compared. If used with care, the approximate Laplace inversion
methods can broaden our range of techniques for finding
(semi-) analytical solutions of partial differential equations.
References

Bellman, R., R.E. Kalaba, and J.A. Lockett. 1966. Numerical


inversion of the Laplace transform: Applications to biology, economics, engineering and physics. American Else-

vier, New York.


Cheng, A.H-D., and E. Detournay. 1988. A direct boundary
ht. J.
element method for plane strain poroelasticity.
Num. Anal. Meth. Geomech.

Due to the presence of the factor l/G, there is a branch


point at infinity. However, the function s1i2 F(s) has no
branch point and can be represented as a polynomial in l/s.
The proper inversion should be performed with /3 = l/2. Figure 8 shows the inversion by the Piessens method with the
default (but wrong) b value in dotted line, and with /3 = l/2
in dashed line (indistinguishable
from the exact solution),
using N = 10. It is clear that for the quadrature method it is
essential to learn about the existence of any branch point of
the function before carrying out the inversion.
f(t)

12:551-572.

Cheng, A.H-D., and O.K. Morohunfola.


1993. Multilayered
leaky aquifer systems: I. Pumping well solution. Water
Resour. Res. 29:2787-2800.

Cheng, A.H-D., and K. Ou. 1989. An efficient Laplace transform solution for multiaquifer systems. Water Resour. Res.
25:742-748.

Cost, T.L. 1964. Approximate


Laplace transform inversion
in viscoelastic stress analysis. AIAA Journal 2:2157-2166.
Crump, K.S. 1976. Numerical inversion of Laplace transforms using a Fourier series approximation.
J. Assoc.
Computing

Machinery

23~89-96.

Davies, B., and B. Martin. 1979. Numerical


Laplace transform: a survey and comparison
J. Camp. Phys. 33:1-32.
Detournay,
response

E., and A.H-D.


Cheng. 1988.
of a borehole in a non-hydrostatic

ht. ]. Rock Me&.

Min. Sci. Geomech.

Durbin, F. 1974. Numerical


an efficient improvement
FIGURE 8. Compar ison of the Piessens inversion using B = 0.5 (dashed line) and /3 = 1
cos N
(solid line).
(dotted line) with the exact solution f(t) = (1 /fi)

Computer].

inversion of
of methods.
Poroelastic
stress field.

Abstr. 25:171-l

82.

inversion of Laplace transforms:


to Dubner and Abates method.

17:371-376.

MacDonald, J.R. 1964. Accelerated convergence, divergence,


iteration, extrapolation,
and curve fitting. J. Appl. Phys.
10:3034-3041.
Narayanan,
G.V., and D.E. Beskos. 1982. Numerical operational methods for the time-dependent
linear problems.
ht. J. Numer. Meth. Engng. 18:1829-l 854.
Papoulis, A. 1957. A new method of inversion
transform. Quart. Appl. Math. 14:405-414.

of the Laplace

VOLUME4,lSSUEZ 81

Piessens, R. 1971. Gaussian quadrature


formulas for the
numerical
integration
of Bromwichs integral and the
inversion of the Laplace transform. J. Eng. Math. 51-9.

Or
Prl
0

Piessens,
R. 1973. Gaussian
quadrature
formulas
Bromwichs integral. Comm. ACM 16:486-487.

for

Piessens, R. 1975. A bibliography on numerical inversion of


the Laplace transform and applications. J. Camp. Appl.

Bernt

Math. 1:ll.S126.

Piessens, R., and N.D.P. Dang. 1976. A bibliography


on
numerical inversion of the Laplace transform and applications: a supplement. J. Comp. Appl. Math. 2:22.5-228.
Stehfest, H. 1970. Numerical
inversion of Laplace
forms. Comm. ACM 13: 47-49 and 624.

trans-

Weeks, W.T. 1966. Numerical inversion of Laplace transforms using Laguerre functions. J. ACM. 13:419-426.

The FOI
waves c
by its al
analyze
Oppenl

WQI,

phase s
determi
[1989]
text.
The I
graphic
tudes a
charact
Oppen!
[1982].
charactl
these in
of amp1
image r
The 1
charact
book in
input st
In[l]:=

Sh

Bernt C.
Californk
devices.

82 THE MA THEMA TICA JOURNAL 0 1994 Miller Freeman Publications

You might also like