You are on page 1of 6

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/241339763

Laplace transform inversion for late-time behavior of groundwater flow


problems

Article  in  Water Resources Research · October 2003


DOI: 10.1029/2003WR002246

CITATIONS READS
7 29

2 authors:

Simon Mathias R. W. Zimmerman


Durham University Imperial College London
99 PUBLICATIONS   1,529 CITATIONS    214 PUBLICATIONS   8,594 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

SAIGUP View project

CONTAIN - The impact of hydrocarbon depletion on the treatment of caprocks within performance assessment for CO2 injection schemes View project

All content following this page was uploaded by R. W. Zimmerman on 08 July 2016.

The user has requested enhancement of the downloaded file.


WATER RESOURCES RESEARCH, VOL. 39, NO. 10, 1283, doi:10.1029/2003WR002246, 2003

Laplace transform inversion for late-time behavior


of groundwater flow problems
Simon A. Mathias
Department of Civil and Environmental Engineering, Imperial College London, London, UK

Robert W. Zimmerman
Department of Earth Science and Engineering, Imperial College London, London, UK

Received 10 February 2003; revised 10 June 2003; accepted 22 July 2003; published 15 October 2003.

[1] Laplace transforms are widely used in solving groundwater flow problems. In the
groundwater literature, it has frequently been asserted or assumed that the late-time
behavior is governed by the behavior of the Laplace transform for small values of the
Laplace variable. Careful reading of authoritative monographs on Laplace transforms
shows that this correspondence does not generally hold. In this article the proper
asymptotic formula is reviewed and is applied to the problem of water influx into a slab-
like matrix block in a dual-porosity medium. In doing so, we clear up a long-standing
discrepancy between the fracture/matrix transfer coefficient that has been calculated by
Laplace methods and the coefficient which has been found by applying Fourier methods in
the time domain. INDEX TERMS: 1829 Hydrology: Groundwater hydrology; 1869 Hydrology:
Stochastic processes; 1894 Hydrology: Instruments and techniques; KEYWORDS: asymptotics, dual-porosity,
fracture-matrix, fractured rocks, groundwater flow, Laplace transforms
Citation: Mathias, S. A., and R. W. Zimmerman, Laplace transform inversion for late-time behavior of groundwater flow problems,
Water Resour. Res., 39(10), 1283, doi:10.1029/2003WR002246, 2003.

1. Introduction [4] Unfortunately, these two theorems yield only the


[2 ] Laplace transforms are widely used in solving limiting values of h in the time domain, but say nothing
groundwater flow problems. In many cases, an inversion about the rate at which these values are approached [Chen
from the Laplace domain back to the time domain that is and Stone, 1993]. However, because of the structure of
valid for all times cannot be obtained in closed form. these two formulae, which in some sense show an inverse
Appeal is therefore often made to various asymptotic results relationship between the two variables t and s, it is very
that are valid at large or small times. often asserted, particularly in the groundwater literature,
[3] Let the hydraulic head be denoted by h(r, t), where r that the asymptotic behavior of h(t) at large times is
generically denotes the position and t is the time. The governed by the behavior of ^ h(s) for very small values of
Laplace transform of the head is given in the usual way by s. Chen and Stone [1993] examined the problem of a well of
finite radius in a laterally infinite confined reservoir, flow-
Z1 ing under conditions of constant drawdown in the well, and
^hðr; sÞ ¼ hðr; tÞest dt; ð1Þ showed that application of the concept of ‘‘large t corre-
sponds to small s’’ gives erroneous results in that case.
0
[5] Our contention is that a similar error has often been
made with regard to the problem of calculating the shape
where s (sometimes denoted by p or l) is the Laplace
factors/transfer coefficients in dual-porosity media. The aim
variable. In the sequel, we will suppress the r dependence
of this article is to present the correct theorem regarding the
and simply write h(t) and ^ h(s). If h(t) is bounded for all t,
asymptotic behavior of the head at large times and to use
then the limiting values of the head, as t approaches 0 or 1,
this result to clarify an existing discrepancy in the literature
if these limits exist, can be found from the following
on dual-porosity media.
Tauberian theorems [Doetsch, 1961, p. 207; Sneddon,
1972, pp. 185– 187; Chen and Stone, 1993]:
2. Transfer Coefficient for a Slab-Like
hð0Þ ¼ lim s^hðsÞ; ð2Þ Matrix Block
s!1
[6] A very clear illustration of the use of the ‘‘large t-
small s’’ correspondence can be found in the calculation
hð1Þ ¼ lim s^hðsÞ: ð3Þ made by Gerke and van Genuchten [1993] of the water
s!0
transfer coefficient for a slab-like matrix block that is
bounded by two parallel, infinite fractures. We now briefly
review their analysis, using the original notation.
Copyright 2003 by the American Geophysical Union. [7] In most formulations of a dual-porosity model,
0043-1397/03/2003WR002246$09.00 referred to as ‘‘first-order’’ models [Gerke and van
TNN 1-1
TNN 1-2 MATHIAS AND ZIMMERMAN: TECHNICAL NOTE

Genuchten, 1993; Young and Ball, 1997] or ‘‘Warren-Root- matrix interface. Using the expansion tanh x = x  (x3/3) +
type’’ models [Zimmerman et al., 1993], the water flux . . . for small x, we find that the first two terms of the
between the fractures and matrix blocks is assumed to be Laurent expansion of equation (8) about s = 0 are
proportional to the difference between the head in the
fracture, hf, and the mean head in the matrix block, hm. In    
the notation of Gerke and van Genuchten [1993], denoted hf hm;i  hf a2 1  wf Cm
^
hm;exact ðsÞ  þ : ð10Þ
hereafter by GvG, the fracture/matrix flux is governed by s 3Ka
the following first-order differential equation:
  dhm   Equating (7) and (10) immediately leads to the erroneous
w ¼ 1  w f C m ¼ aw hf ðtÞ  hm ðtÞ ; ð4Þ conclusion that the water transfer coefficient should be
dt
given by
where w is the rate of water transfer between the fracture
and matrix, wf is the fracture porosity, Cm is the derivative 3 12
aw ¼ Ka ¼ Ka : ð11Þ
of the water content with respect to the head, and aw is the a2 ð2aÞ2
‘‘water transfer coefficient.’’ The Laplace transform of
equation (4) is (The latter form is presented to facilitate comparison with
papers that work in terms of the width of the matrix block,
  h i h i
rather than the half-width.) The implicit, but incorrect,
1  wf Cm s^hm ðsÞ  hm;i ¼ aw ^hf ðsÞ  ^hm ðsÞ ; ð5Þ
assumption of this derivation is that the behavior of ^ h(s) at
small values of s will correspond to the behavior of h(t) at
where hm,i is the initial head in the matrix. (Note that GvG large values of t, so that the first-order transfer model will
were considering unsaturated flow and needed to linearize asymptotically agree with the exact Richards equation
the problem to make it amenable to Laplace transform model for the matrix block, at large times.
methods. However, any issues related to this linearization, [10] However, as illustrated concretely by Chen and Stone
such as the proper definition of ‘‘averaged’’ parameters such [1993], there is in general no justification for the somewhat
as Cm, are not at all relevant to the present discussion. imprecise identification of the large time behavior of the
Indeed, essentially the same equations would apply in a hydraulic head with the behavior of its Laplace transform at
saturated flow problem or solute transport problem, in small values of s. Rather, the relevant asymptotic formula
which justification of the linearization would not be of for large times is given by Carslaw and Jaeger [1949,
major concern.) p. 280], which we state now, with only one minor change in
[8] GvG consider the canonical problem in which the notation: replacing their Laplace variable l with the more
matrix is initially at a uniform head hm,i, after which, at t = common s. Let s0 be the singularity of ^f (s) having the largest
0, the head in the two fractures that bound the matrix block real part. Assume that ^f (s) can be expanded in the neigh-
is changed to some value hf and held constant at that value. borhood of s0 as follows:
In this case the function ^ hf (s) appearing in equation (5)
takes the form ^hf (s) = hf /s, and GvG solve for
  X
1 X
1

aw hf 1  wf Cm hm;i
^f ðsÞ ¼ an ðs  s0 Þn1 þðs  s0 Þb1 bn ðs  s0 Þn ; 0 < b < 1:
^hm ðsÞ ¼    þ   : ð6Þ n¼0 n¼0
s aw þ 1  wf Cm s aw þ 1  wf Cm s
ð12Þ

To find the approximate behavior for small values of s, they The first series in equation (12) is a Laurent-type expansion,
expand ^hm(s) in a Laurent series about s = 0 (i.e., a1s1 + whereas the second series is of a more general type and
a0 + a1s + . . .), keeping only the first two terms: contains nonintegral powers. (The coefficient a 0 in
   equation (12), which is the residue of ^f (s) at s0, is usually
^hm ðsÞ  hf þ hm;i  hf 1  wf Cm : ð7Þ denoted by a1 in the complex variable literature, but we
s aw will adhere to the notation of Carslaw and Jaeger.) The
asymptotic expansion of f (t) for large values of t is then
[9] The exact solution for the mean head in the matrix given by [see also Doetsch, 1961, p. 213]
block can be found by solving the linearized Richards ( )
equation within the matrix block. The solution, in the sin pb X
1
Laplace domain, is given by equation (17) of GvG: f ðt Þ es0 t a0 þ ð1Þn bn ðb þ nÞtbn ; ð13Þ
p n¼0

^hm;exact ðsÞ ¼ hf  hm;i tanh x hm;i
þ ; ð8Þ where  is the gamma function. It is clear from this formula
s x s
that the key value of s is the singularity having the largest
real part, which may or may not occur at the origin.
where x is defined by
[11] If this formula were directly applied to the function
   1=2
^
hm(s), given by either the first-order model (6), or the exact
x ¼ a2 1  wf Cm s=Ka ; ð9Þ expression (8), we would see that the singularity of the
largest real part was indeed located at the origin and that the
a is the half-width of the matrix block, and Ka is the fractional power terms do not appear in either case. In each
effective hydraulic conductivity of the matrix at the fracture/ case we have s0 = 0 and a0 = hf, and the asymptotic
MATHIAS AND ZIMMERMAN: TECHNICAL NOTE TNN 1-3

formula (13) would yield, for both the exact and the first- which shows that the apparent singularity at s = 0 was in
order models, fact removable.
[14] To find the dominant singularity of equation (18), we
hm ðt Þ hf as t ! 1: ð14Þ recall that tanh = sinh x/cosh x, and hence the singularities
of (18) should occur at the zeroes of cosh x, which are at
This result is of course correct, albeit trivial. Note that the xn = (pi/2) + npi, where n is any integer [Levinson and
expansion about s = 0 gives no information about the rate at Redheffer, 1970, p. 418]. Recalling from equation (9) that
which this limit is approached. x = [a2(1  wf)Cms/Ka]1/2, it follows that the singularities of
[12] In order to use formula (13) to gain nontrivial (18) are located at
asymptotic information about the large-time behavior of
hm(t), we can work with the head difference, hm(t) = hm(t) Ka x2n K p2 9Ka p2
sn ¼   ¼  a  ;   ; etc:
 hf (t). In terms of the head difference, the first-order a2 1  wf Cm 4a2 1  wf Cm 4a2 1  wf Cm
model, equation (6), takes the form ð21Þ
   
hf 1  wf Cm hm;i  hf
^
hm ðsÞ ¼ ^hm ðsÞ  ¼   [15] The singularities of this function are all on the
s aw þ 1  wf Cm s negative real axis. The one having the largest real part is
hm;i  hf the first one on the right-hand side of (21), specifically
¼    : ð15Þ
s þ aw = 1  wf Cm
K p2
s0 ¼  a  : ð22Þ
Comparison of equation (15) with (12) reveals that 4a2 1  wf Cm
 
a0 ¼ hm;i  hf ; s0 ¼ aw = 1  wf Cm ; ð16Þ Equating this to the value of s0 given in (16) by the first-
order model yields
and all other terms are absent. Hence equation (13) shows
that the asymptotic behavior of the first-order model, for p2 p2
large times, is given by aw ¼ Ka ¼ Ka ; ð23Þ
4a2
ð2aÞ2
 
hm ðt Þ hm;i  hf eaw t=ð1wf ÞCm : ð17Þ which is consistent with the value found by Dykhuizen
[1990] and Zimmerman et al. [1993], who analyzed the
This is of course a well-known exact result (for the first- problem using Fourier methods in the time domain.
order model), but our point here is to show that it is also [16] Alternatively, s0 could also have been found by
fully consistent with the asymptotic formula (13). appealing to the following series representation of the tanh
[13] We now apply the same asymptotic analysis to the function [Bromwich and MacRobert, 1949, p. 296]:
exact model, equation (8). Written in terms of the head
difference, (8) takes the form X
1
2x
tanh x ¼ 2 2 2
: ð24Þ
   n¼0 x þ ½n þ ð1=2Þ p
hf hm;i  hf tanh x
^hm;exact ðsÞ ¼ ^hm;exact ðsÞ  ¼ 1 : ð18Þ
s s x Again recalling that x = [a2(1  wf)Cms/Ka]1/2, and
temporarily writing this as x = [s/b]1/2, equation (17) can
At first glance it may appear that ^hm,exact(s), as given by be written as
equation (18), has a singularity of some sort at s = 0.
However, s = 0 corresponds to x = 0, and again using the  " #
hm;i  hf X
1
2b
expansion tanh x = x  (x3/3) + . . . for small x, we find that ^
hm;exact ðsÞ ¼ 1 ; : ð25Þ
near s = 0, s n¼0 s þ b½n þ ð1=2Þ 2 p2

 "     # Although not a Laurent series, equation (25) clearly shows


^ hm;i  hf x  x3 =3 þ x5 =5  . . .
hm;exact ðsÞ  1 the location of the poles of ^ hm,exact, which agree with
s x
  2 those given in equation (21). (Again, the apparent

hm;i  hf x x4 singularity at s = 0 is removable.) The residue of
  þ ... : ð19Þ ^hm,exact(s) at s0, denoted in our present notation by a0, is
s 3 5
readily found by the standard method [Levinson and
Using equation (9) to express x in terms of s, we see that Redheffer, 1970, p. 191]:
near s = 0,
   
^  hm;i  hf 2b 8 hm;i  hf
 "    2 # a0 ¼ lim ðs  s0 Þhm;exact ðsÞ ¼ ¼ :
hm;i  hf a2 1  wf Cm s a4 1  wf ðCm Þ2 s2
s!s0 s0 p2
^
hm;exact ðsÞ   þ ...
s 3Ka 5ðKa Þ2 ð26Þ
"     #
  a2 1  wf Cm a4 1  wf 2 ðCm Þ2 s
 hm;i  hf  þ ... ; The same result could also be obtained directly from
3Ka 5ðKa Þ2
equation (18), by recalling (9) and expanding cosh x about
(20) s = s0, although the calculation is somewhat lengthier.
TNN 1-4 MATHIAS AND ZIMMERMAN: TECHNICAL NOTE

[ 17 ] Choosing the water transfer coefficient as per


equation (23) will therefore ensure that the time constants
that govern the rates at which the head difference between
the fracture and the matrix block decays to zero will be
identical for the exact model and the first-order model. The
numerical constants a0 that appear in the asymptotic expan-
sion (13) will be slightly different, as seen by comparing
equations (16) and (26). As there is only one adjustable
parameter in the first-order model, it is not reasonable to
expect that both s0 and a0 can be made to agree. However, as
shown in the next section, the choice of s0 that emerges from
the asymptotic expansion (13) leads to closer agreement with
the exact solution than does the value given by equation (11).

3. Discussion
[18] The exact solution for the mean head in a slab-like
matrix block subjected to constant-head boundary condi-
tions on both faces is given by [Crank, 1956, p. 45;
Dykhuizen, 1990] Figure 1. Normalized mean head in a slab-like matrix
" # block of width 2a, with constant-head boundary conditions,
hm ðtÞ  hf 8 X
1
1 ð2n þ 1Þ2 p2 Ka t as a function of dimensionless time, according to the exact
¼ 2 2
exp   : ð27Þ
hm;i  hf p n¼0 ð2n þ 1Þ ð2aÞ2 1  wf Cm solution (27), the first-order model with aw = 12Ka/(2a)2
(equation (29)), and the first-order model with aw = p2Ka/
The first term in the full solution, which is dominant at large (2a)2 (equation (30)).
times, is identical to the asymptotic term obtained from
equation (13), as can be seen by inserting (22) and (26) exact solution more closely than does the aw = 12Ka/(2a)2
into (13): first-order model. Hence either interpretation of the above
" #
hm ðt Þ  hf 8 p2 Ka t developments leads to a preference for equation (23)
 2 exp   : ð28Þ over (11). Note that there is no choice of the parameter aw that
hm;i  hf p ð2aÞ2 1  wf Cm
that will cause the first-order model to agree with the exact
The first-order model, with the transfer coefficient given by model at early
pffi times, as the exact model shows that the head
equation (11), yields varies as t , whereas the first-order model predicts a linear
variation of head with time. This is a fundamental limitation
" # of the first-order model [Zimmerman et al., 1993; Young and
hm ðtÞ  hf 12Ka t
¼ exp   : ð29Þ Ball, 1997], but is not the subject of the present article.
hm;i  hf ð2aÞ2 1  wf Cm [21] We believe that the above discussion explains the
discrepancy, often noted in the literature, between the transfer
coefficients that have been calculated using Laplace domain
[19] The analysis that makes use of equations (12) methods by making use of the supposed ‘‘small s-large t’’
and (13) clearly leads to a more accurate late-time asymp- correspondence [Barker, 1985; Gerke and van Genuchten,
totic expression than does the analysis that utilizes an 1993; Landereau et al., 2001], and those found by working in
expansion of ^hm(s) around s = 0. However, it could be the time domain [Zimmerman et al., 1993, Lim and Aziz,
claimed that this is not the appropriate comparison to make 1995]. Equations (12) and (13), applied above to diffusion
in the present context. One could view the Laplace space into slab-like blocks, can also be applied to other geometries
manipulations discussed above as being mainly a device to such as spheres or cylinders [van Genuchten and Dalton,
find the best value of the water transfer coefficient to use in 1986; Zimmerman et al., 1993]. In each case, the apparent
the first-order model, rather than a method for deriving a discrepancy between the predictions of the Laplace domain
solution to the specific problem of a matrix block with approach and the time domain approach would disappear
constant-head boundary conditions. In this case, we should (S. A. Mathias, Ph.D. dissertation in preparation, 2003).
consider the implications of using equation (23) for the
transfer coefficient, as opposed to (11). Figure 1 shows the [22] Acknowledgments. Simon Mathias thanks the Engineering and
normalized head difference between the fracture and matrix Physical Sciences Research Council (EPSRC) of the UK for financial support,
and thanks Adrian Butler of Imperial College for helpful discussions.
block as a function of dimensionless time, for the exact
solution given by equation (27), the first-order model with References
aw given by (11), and the first-order model with aw given
Barker, J. A., Block-geometry functions characterizing transport in densely
by (22). In this latter case the first-order model yields fissured media, J. Hydrol., 77, 263 – 279, 1985.
" #
hm ðtÞ  hf p2 Ka t Bromwich, T. J. I., and T. M. MacRobert, Theory of Infinite Series, Mac-
¼ exp 2   : ð30Þ millan, Old Tappan, N. J., 1949.
hm;i  hf ð2aÞ 1  wf Cm Carslaw, H. S., and J. C. Jaeger, Operational Methods in Applied Mathe-
matics, 2nd ed., Oxford Univ. Press, New York, 1949.
Chen, C.-S., and W. D. Stone, Asymptotic calculation of Laplace inverse in
[20] Visual inspection of Figure 1 indicates that at late analytical solutions of groundwater problems, Water Resour. Res., 29(1),
times the first-order model using aw = p2Ka/(2a)2 mimics the 207 – 209, 1993.
MATHIAS AND ZIMMERMAN: TECHNICAL NOTE TNN 1-5

Crank, J., Mathematics of Diffusion, Oxford Univ. Press, New York, 1956. van Genuchten, M. T., and F. N. Dalton, Models for simulating salt move-
Doetsch, G., Guide to the Applications of Laplace Transforms, Van Nos- ment in aggregated field soils, Geoderma, 38, 165 – 183, 1986.
trand Reinhold, New York, 1961. Young, D. F., and W. P. Ball, Effects of column conditions on the first-order
Dykhuizen, R. C., A new coupling term for dual-porosity models, Water rate modeling of nonequilibrium solute breakthrough: Cylindrical macro-
Resour. Res., 26(2), 351 – 356, 1990. pores versus spherical media, Water Resour. Res., 33(5), 1149 – 1156,
Gerke, H. H., and M. T. van Genuchten, Evaluation of a first-order water 1997.
transfer term for variably saturated dual-porosity flow models, Water Zimmerman, R. W., G. Chen, T. Hadgu, and G. S. Bodvarsson, Water
Resour. Res., 29(4), 1225 – 1238, 1993. Resour. Res., 29(7), 2127 – 2137, 1993.
Landereau, P., B. Noetinger, and M. Quintard, Quasi-steady two-equation
models for diffusive transport in fractured porous media: large-scale
properties for densely fractured systems, Adv. Water Resour., 24(8),
863 – 878, 2001. 

Levinson, N., and R. M. Redheffer, Complex Variables, Holden-Day, Boca S. A. Mathias, Department of Civil and Environmental Engineering,
Raton, Fla., 1970. Imperial College London, London SW7 2AZ, UK. (simon.mathias@
Lim, K. T., and K. Aziz, Matrix-fracture transfer shape factors for dual- imperial.ac.uk)
porosity simulators, J. Petrol. Sci. Eng., 13(3), 169 – 178, 1995. R. W. Zimmerman, Department of Earth Science and Engineering,
Sneddon, I. N., The Use of Integral Transforms, McGraw-Hill, New York, Imperial College London, London SW7 2AZ, UK. (r.w.zimmerman@
1972. imperial.ac.uk)

View publication stats

You might also like