Abstract
For transport in statistically homogeneous random velocity fields with properties that are routinely assumed in stochastic groundwater models, the one‐particle dispersion (i.e., second central moment of the ensemble average concentration for a point source) is a “memory‐free” quantity independent of initial conditions. Nonergodic behavior of large initial plumes, as manifest in deviations of actual solute dispersion from one‐particle dispersion, is associated with a “memory term” consisting of correlations between initial positions and displacements of solute molecules. Reliable numerical experiments show that increasing the source dimensions has two opposite effects: it reduces the uncertainty related to the randomness of center of mass, but, at the same time, it yields large memory terms. The memory effects increase with the source dimension and depend on its shape and orientation. Large narrow sources oriented transverse to the mean flow direction yield ergodic behavior with respect to the one‐particle dispersion of the longitudinal dispersion and nonergodic behavior of the transverse dispersion, whereas for large longitudinal sources, the longitudinal dispersion behaves nonergodically, and the transverse dispersion behaves ergodically. Such memory effects are significantly large over hundreds of heterogeneity scales and should therefore be considered in practical applications, for instance, calibration of model parameters, forecasting, and identification of the contaminant source.
Authors
N. Suciu
Institute of Applied Mathematics, Friedrich-Alexander University of Erlangen-Nuremberg, Erlangen, Germany
C. Vamoş
Tiberiu Popoviciu Institute of Numerical Analysis, Romanian Academy, Cluj-Napoca, Romania
H. Vereecken
ICG-IV, Research Center Julich, Julich, Germany
K. Sabelfeld
Weierstrass Institute for Applied Analysis and Stochastics, Berlin,Germany
Institute of Computational Mathematics and Mathematical Geophysics,
Siberian Branch of Russian Academy of Sciences, Novosibirsk, Russia
P. Knabner
Institute of Applied Mathematics, Friedrich-Alexander University of Erlangen-Nuremberg, Erlangen, Germany
Keywords
Cite this paper as:
N. Suciu, C. Vamoş, H. Vereecken, K. Sabelfeld, P. Knabner, Memory effects induced by dependence on initial conditions and ergodicity of transport in heterogeneous media, Water Resour. Res., 44 (2008), W08501, doi: 10.1029/2007WR006740
About this paper
Print ISSN
Not available yet.
Online ISSN
Not available yet.
Google Scholar Profile
google scholar link
[1] Chile`s, J. P., and P. Delfiner (1999), Geostatistics: Modeling Spatial Uncertainty, John Wiley, New York.
[2] Dagan, G. (1990), Transport in heterogeneous porous formations: Spatial moments, ergodicity and effective dispersion, Water Resour. Res., 26, 1281 – 1290.
[3] Dentz, M., and J. Carrera (2007), Mixing and spreading in stratified flow, Phys. Fluids, 19, 017107, doi:10.1063/1.2427089.
[4] Dentz, M., H. Kinzelbach, S. Attinger, and W. Kinzelbach (2000), Temporal behavior of a solute cloud in a heterogeneous porous medium: 2. Spatially extended injection, Water Resour. Res., 36, 3605 – 3614.
[5] Eberhard, J., N. Suciu, and C. Vamos¸ (2007), On the self-averaging of dispersion for transport in quasi-periodic random media, J. Phys. A Math. Theor., 40, 597 – 610, doi:10.1088/1751-8113/40/4/002.
[6] Fiori, A., and I. Jancovic´ (2005), Can we determine the transverse macrodispersivity by using the method of moments?, Adv. Water Resour., 28, 589 – 599, doi:10.1016/j.advwaters.2004.09.909.
[7] Kitanidis, P. K. (1988), Prediction by the method of moments of transport in a heterogeneous formation, J. Hydrol., 102, 453 – 473.
[8] Le Doussal, P., and J. Machta (1989), Annealed versus quenched diffusion coefficient in random media, Phys. Rev. B, 40(12), 9427 – 9430.
[9] Morales-Casique, E., S. P. Neuman, and A. Gaudagnini (2006), Nonlocal and localized analyses of nonreactive solute transport in bounded randomly heterogeneous porous media: Theoretical framework, Adv. Water Resour., 29, 1238 – 1255, doi:10.1016/j.advwatres.2005.10.002. Naff, R. L., D. F.
[10] Haley, and E. A. Sudicky (1998), High-resolution Monte Carlo simulation of flow and conservative transport in heterogeneous porous media: 2. Transport results, Water Resour. Res., 34, 679 – 697.
[11] Phythian, R., and W. D. Curtis (1978), The effective long-time diffusivity for a passive scalar in a Gaussian model fluid flow, J. Fluid Mech., 89(2), 241 – 260.
[12] Radu, F. A., I. S. Pop, and P. Knabner (2004), Order of convergence estimates for an Euler implicit, mixed finite element discretization of Richards’ equation, SIAM J. Numer. Anal., 42(4), 1452 – 1478, doi:10.1137/S0036142902405229
[13] Radu, F. A., I. S. Pop, and S. Attinger (2008), Analysis of an Euler implicit –mixed finite element scheme for reactive solute transport in porous media, CASA Rep. 08-06, Tech. Univ. Eindhoven, Eindhoven, Netherlands.
[14] Sposito, G. (2001), Topological groundwater hydrodynamics, Adv. Water Resour., 24, 793 – 801.
[15] Sposito, G., and G. Dagan (1994), Predicting solute plume evolution in heterogeneous porous formations, Water Resour. Res., 30, 585 – 589.
[16] Suciu, N., and C. Vamos¸ (2007), Comment on ‘‘Nonstationary flow and nonergodic transport in random porous media’’ by G. Darvini and P. Salandin, Water Resour. Res., 43, W12601, doi:10.1029/ 2007WR005946.
[17] Suciu, N., C. Vamos¸, J. Vanderborght, H. Hardelauf, and H. Vereecken (2006a), Numerical investigations on ergodicity of solute transport in heterogeneous aquifers, Water Resour. Res., 42, W04409, doi:10.1029/ 2005WR004546.
[18] Suciu, N., C. Vamos¸, and J. Eberhard (2006b), Evaluation of the first-order approximations for transport in heterogeneous media, Water Resour. Res., 42, W11504, doi:10.1029/2005WR004714.
[19] Suciu, N., C. Vamos¸, and K. Sabelfeld (2007a), Ergodic simulations of diffusion in random velocity fields, in Monte Carlo and Quasi-Monte Carlo Methods 2006, edited by A. Keller, S. Heinrich, and H. Niederreiter, pp. 659 – 668, Springer, Heidelberg, Germany.
[20] Suciu, N., C. Vamos¸, K. Sabelfeld, and C. Andronache (2007b), Memory effects and ergodicity for diffusion in spatially correlated velocity fields, Proc. Appl. Math. Mech., 7, 2010015, doi:10.1002/pamm.20070057.
[21] Suciu, N., C. Vamos¸, H. Vereecken, K. Sabelfeld, and P. Knabner (2008), Lagrangian stationarity and memory effects for dispersion in ergodic velocity fields, paper presented at General Assembly 2008, Eur. Geosci. Union, Vienna, 13 – 18 April.
[22] Vamos, C., N. Suciu, and H. Vereecken (2003), Generalized random walk algorithm for the numerical modeling of complex diffusion processes, J. Comput. Phys., 186(2), 527 – 544, doi:10.1016/S0021-9991(03)00073-1.
[23] van Kampen, N. G. (1981), Stochastic Processes in Physics and Chemistry, North-Holland, Amsterdam. Yaglom, A. M. (1987), Correlation Theory of Stationary and Related Random Functions, vol. 1, Basic Results, Springer, New York.
[24] Yaglom, A. M. (1987), Correlation Theory of Stationary and Related Random Functions, vol. 1, Basic Results, Springer, New York.
[25] Zirbel, C. L. (2001), Lagrangian observations of homogeneous random environments, Adv. Appl. Prob., 33, 810 – 835.
Memory effects induced by dependence on initial
conditions and ergodicity of transport in
heterogeneous media
N. Suciu,
1
C. Vamos¸,
2
H. Vereecken,
3
K. Sabelfeld,
4,5
and P. Knabner
1
Received 6 December 2007; revised 20 May 2008; accepted 3 June 2008; published 9 August 2008.
[1] For transport in statistically homogeneous random velocity fields with properties that
are routinely assumed in stochastic groundwater models, the one-particle dispersion (i.e.,
second central moment of the ensemble average concentration for a point source) is a
‘‘memory-free’’ quantity independent of initial conditions. Nonergodic behavior of large
initial plumes, as manifest in deviations of actual solute dispersion from one-particle
dispersion, is associated with a ‘‘memory term’’ consisting of correlations between initial
positions and displacements of solute molecules. Reliable numerical experiments show
that increasing the source dimensions has two opposite effects: it reduces the uncertainty
related to the randomness of center of mass, but, at the same time, it yields large memory
terms. The memory effects increase with the source dimension and depend on its shape
and orientation. Large narrow sources oriented transverse to the mean flow direction yield
ergodic behavior with respect to the one-particle dispersion of the longitudinal dispersion
and nonergodic behavior of the transverse dispersion, whereas for large longitudinal
sources, the longitudinal dispersion behaves nonergodically, and the transverse dispersion
behaves ergodically. Such memory effects are significantly large over hundreds of
heterogeneity scales and should therefore be considered in practical applications, for
instance, calibration of model parameters, forecasting, and identification of the
contaminant source.
Citation: Suciu, N., C. Vamos¸, H. Vereecken, K. Sabelfeld, and P. Knabner (2008), Memory effects induced by dependence on initial
conditions and ergodicity of transport in heterogeneous media, Water Resour. Res., 44, W08501, doi:10.1029/2007WR006740.
1. Introduction
[2] The transport in heterogeneous media is conveniently
characterized by the second central spatial moment tensor of
the concentration field s
lm
(l, m = 1, 2, 3). As well as
providing a measure for the spatial extension of the solute
plume, the dependence on time t of the second moment is
commonly used to investigate whether the transport is
diffusive, i.e., s
lm
t [Sposito and Dagan, 1994]. Since it
can be estimated, by either analytical approximations or
numerical simulations, without solving the transport equa-
tions [Suciu et al., 2006b; Eberhard et al., 2007] this
quantity is particularly useful in investigations on pre-
asymptotic transport regime, for which generally there are
no close form solutions [see, e.g., Morales-Casique et al.,
2006].
[3] A frequently used approach considers a local disper-
sion process and models the heterogeneity of the velocity at
larger scales by a space random function. Let X
l
be the
l component of the trajectory of a solute molecule. To
simplify matters, we consider only the diagonal components
ll of the various second moments. For a fixed realization of
the velocity, the second moment s
ll
of the actual concentra-
tion is given by the dispersion of the molecules at a given
time
s
ll
¼
X
l
X
l
DX0
2
DX0
: ð1Þ
The subscripts D and X
0
in (1) denote respectively the
average over the realizations of the local dispersion and the
space average with respect to the initial distribution of
molecules.
[4] To emphasize the role of initial conditions, we con-
sider displacements
e
X
l
= X
l
X
0l
relative to initial positions
X
0l
which introduced in (1) yield
s
ll
¼ S
ll
0 ð Þþ
e
X
l
e
X
l
DX0
2
DX0
þ m
ll
; ð2Þ
where S
ll
(0) = h[X
0l
hX
0l
i
X
0
]
2
i
X
0
is the deterministic initial
second moment. The last term in (2),
m
ll
¼ 2
X
0l
X
0l
X0
e
X
l
D
X0
; ð3Þ
1
Institute of Applied Mathematics, Friedrich-Alexander University of
Erlangen-Nuremberg, Erlangen, Germany.
2
Tiberiu Popoviciu Institute of Numerical Analysis, Romanian Academy,
Cluj-Napoca, Romania.
3
ICG-IV, Research Center Ju¨lich, Ju¨lich, Germany.
4
Weierstrass Institute for Applied Analysis and Stochastics, Berlin,
Germany.
5
Institute of Computational Mathematics and Mathematical Geophysics,
Siberian Branch of Russian Academy of Sciences, Novosibirsk, Russia.
Copyright 2008 by the American Geophysical Union.
0043-1397/08/2007WR006740
W08501
WATER RESOURCES RESEARCH, VOL. 44, W08501, doi:10.1029/2007WR006740, 2008
1 of 6
describes spatial correlations between relative displace-
ments on trajectories and initial positions and, therefore, it
has been called ‘‘memory term’’ [Suciu and Vamos¸ , 2007;
Suciu et al., 2007b]. The implications of such terms for
predicting solute dispersion were first investigated by
Sposito and Dagan [1994], in the context of deterministic
advective transport. A decade later, Fiori and Jancovic´
[2005] simulated the time behavior of m
ll
, for advective
transport as well, and concluded that a representative
transverse dispersivity cannot be inferred from experiments
done for large transverse sources. Recent numerical
investigations which considered both advection and local
dispersion [Suciu et al., 2006a] also indicated that the
transverse dispersion for large transverse sources signifi-
cantly differs from theoretical ‘‘ergodic’’ results. This paper
aims at highlighting relationships between initial conditions,
memory terms, and the nonergodic behavior of the pre-
asymptotic dispersion. Our study is based on ‘‘global
random walk’’ (GRW) numerical simulations [Vamos¸ et al.,
2003], for different initial conditions, which complete
previous ones presented by Suciu et al. [2006a].
2. Second Moments and Memory Terms
[5] Another equivalent expression of dispersion (1),
which now highlights the randomness of the center of
mass, is
s
ll
¼ s
ll
r
ll
ð4Þ
s
ll
¼
X
l
X
l
DX0 V
2
DX0
ð5Þ
r
ll
¼
X
l
DX0
X
l
DX0V
2
; ð6Þ
where hX
l
i
DX
0
V
is the average over the ensemble of
realizations of the random velocity field (hereafter
indicated by a subscript V) of the center of mass. The
ensemble average of (4) is the well known identity
[Kitanidis, 1988; Le Doussal and Machta, 1989; Naff et al.,
1998; Suciu et al., 2006a] which relates the expected second
moment S
ll
= hs
ll
i
V
to the second moment of the mean
concentration S
ll
= hs
ll
i
V
and the variance of the center of
mass R
ll
= hr
ll
i
V
:
S
ll
¼ S
ll
R
ll
: ð7Þ
[6] Assuming all necessary joint measurability conditions
which allow permutations of averages [Zirbel, 2001]
leads to
S
ll
¼ S
ll
0 ð ÞþhX
ll
i
X0
þ M
ll
þ Q
ll
; ð8Þ
where X
ll
= h[
e
X
l
h
e
X
l
i
DV
]
2
i
DV
is the one-particle
dispersion (defined by averaging with respect to D and V
for a fixed initial position), M
ll
= hm
ll
i
V
is the mean memory
term, and Q
ll
= h½h
e
X
l
i
DV
h
e
X
l
i
DX
0
V
2
i
X
0
is the spatial
variance (computed by averages over X
0
) of the one-particle
center of mass h
e
X
l
i
DV
[Suciu et al., 2007b; Suciu and
Vamos¸ , 2007].
[7] The terms of (8) depend, via the trajectory equation,
on Lagrangian velocity field V
L
(X
0
, t)= V(X(t; X
0
)), which
consists of observations of the random Eulerian velocity
field V at random locations on the trajectory. If the
Lagrangian field is statistically homogeneous the one-
particle center of mass h
e
X
l
i
DV
and dispersion X
ll
are
independent of X
0
. Then M
ll
and Q
ll
vanish and (7)–(8)
takes on the simpler form [see, e.g., Dagan, 1990]
S
ll
¼ S
ll
0 ð Þþ X
ll
R
ll
: ð9Þ
[8] The homogeneity of V
L
holds under conditions rou-
tinely assumed in stochastic modeling which essentially
consist of statistical homogeneity of the Eulerian field,
continuity of the velocity samples, and existence of unique
solutions of trajectory equations. The equality of Lagrangian
and Eulerian mean velocities requires stronger conditions.
Depending on whether dispersion processes are modeled as
spatially correlated noises or as Brownian motions, these are
the flow’s incompressibility or both incompressibility and
sample continuity of first derivatives of V [Zirbel, 2001].
Implications of velocity properties for a transport model
based on Itoˆ equation, explicit expressions for dispersion,
and first-order approximations are presented at length in the
preprint by N. Suciu et al. (Dependence on initial conditions,
memory effects, and ergodicity of transport in heterogeneous
media, 2008, Institute of Applied Mathematics, Friedrich-
Alexander University Erlangen-Nuremberg, available at
http://www.am.uni-erlangen.de/de/preprints2000.html), on
which this paper is based.
[9] The one-particle quantities entering (8) are ensemble
averages of the first two moments of plumes starting from
point sources located at x
0
which can be calculated from the
ensemble average of the transition probability defined by
[see, e.g., van Kampen, 1981] p(x, t|x
0
)= hd[x X(t; x
0
)]i
D
.
Since the ensemble average hpi
V
depends on velocity statis-
tics through the moments of the Lagrangian velocity V
L
, it
is translation invariant for homogeneous V
L
[Suciu et al.,
2008] (see also the preprint by Suciu et al., 2008). This
property is helpful in renormalized series expansions for
dispersion coefficients [Phythian and Curtis, 1978] and has
been used by Dentz et al. [2000] to show the independence
of the ensemble coefficients
1
2
dS
ll
/dt from the initial con-
centration distribution.
[10] Homogeneous random fields with finite correlation
range are ergodic: space averages of the velocity are
unbiased estimators strongly convergent to the ensemble
mean velocity [Chile`s and Delfiner, 1999]. There is also a
numerical evidence that space averages with respect to the
location of the source of the concentration moments
resulted from simulations of diffusion in Gaussian fields
converge to their ensemble averages [Suciu et al., 2007a].
For plumes with initial dimensions larger than the hetero-
geneity scale [Dagan, 1990] one can therefore assume that,
according to (6), r
ll
R
ll
0. When ergodic behavior
prevails, it follows from (9) that the one-particle dispersion
X
ll
provides an idealized description of the expected second
moment,
S
ll
S
ll
0 ð Þþ X
ll
: ð10Þ
2 of 6
W08501 SUCIU ET AL.: TECHNICAL NOTE W08501
19447973, 2008, 8, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2007WR006740 by Cochrane Romania, Wiley Online Library on [28/08/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
[11] A similar ergodic argument suggests that the sec-
ond term of (2) behaves like X
ll
for large initial plumes.
But even if (10) can approximate the expectation, it might
not be accurate for the actual dispersion. Indeed, (3)
shows that for large sources the memory terms can be
important. So, at best, the actual dispersion can be
approximated as
s
ll
S
ll
0 ð Þþ X
ll
þ m
ll
: ð11Þ
[12] The relevance of the stochastic description for the
dispersion observable in a real case; that is, the ergodicity in
a broad sense, can be assessed quantitatively by an ‘‘ergo-
dicity range’’ defined as root mean square deviation of
observable quantities from stochastic model predictions
[Suciu et al., 2006a]. The numerical experiment presented
in the following shows that for large initial plumes the
ergodicity range of s
ll
S
ll
(0) with respect to X
ll
can be
estimated, according to (11), by standard deviations of
memory terms m
ll
.
3. Memory Effects and Ergodicity
[13] We considered an isotropic two-dimensional aquifer
system, characterized by log-hydraulic conductivity with
small variance equal to 0.1 and exponentially decaying
isotropic correlation with correlation length l = 1 m.
First-order approximations for incompressible Darcy veloc-
ity fields were generated numerically. The task was
achieved with the Kraichnan algorithm by using 6400
periodic modes, which guarantees accurate simulations of
transport in Gaussian fields over thousands of heterogeneity
scales [Eberhard et al., 2007]. For fixed mean flow velocity
U = 1 m/d and isotropic local dispersion with constant
coefficient D = 0.01 m
2
/d, the Pe´clet number got a typical
value Pe =Ul/D = 100. In every velocity realization, 10
10
particles, that were initially uniformly distributed in rect-
angular domains L
1
l L
2
l or released from the origin of
the computational grid, were tracked simultaneously with
the GRW algorithm (for details on the implementation of the
numerical method see Suciu et al. [2006a, Appendix A]). For
all cases investigated here we simulated 1024 realizations,
which rendered the statistical oscillations of the estimated
means and standard deviations of the dispersion terms (4),
(5), and (6) smaller than half the local dispersion Dt
uniformly in time (see the preprint by Suciu et al., 2008).
[14] Figure 1 shows that the dispersion of the center of
mass R
ll
decays monotonically with the increasing dimen-
sions of the source, irrespective of its shape and orientation.
The standard deviations SD[r
ll
] were found to be of the
same order of magnitude. They are significantly larger than
R
ll
only for small sources. The results for L 50 provide a
numerical support for the ergodic argument which suggests
that for large plumes R
ll
0. Hence, if relation (9) holds,
then the expectation of the second moment of the actual
concentration field can be approximated with relation (10)
by the sum between the deterministic initial second moment
S
ll
(0) and the one-particle dispersion X
ll
.
[15] Numerical results for the second moment of the
ensemble averaged concentration S
ll
(not presented here)
show that for slab sources perpendicular to l-axis X
ll
can be
accurately estimated by S
ll
S
ll
(0), in keeping with theoret-
ical predictions for transport in statistically homogeneous
velocity fields (compare (7) and (9)). For sources with large
extensions on l-axis we found differences between the two
quantities (of the order of a few local dispersions 2Dt for
L = 100) which indicate non-vanishing mean memory terms
M
ll
. Both situations described above (for transverse slab
sources only) and the ‘‘ergodic’’ dispersion coefficients
X
ll
/(2Dt) (derived from point source simulations) were pre-
viously presented by Suciu et al. [2006a, Figures 11 and 13].
The non-zero values of M
ll
can be explained by small
Figure 1. Variance of the plume center of mass R
ll
and corresponding standard deviation SD(r
ll
) (thin
lines) for different shapes and extensions of the initial plume.
W08501 SUCIU ET AL.: TECHNICAL NOTE
3 of 6
W08501
19447973, 2008, 8, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2007WR006740 by Cochrane Romania, Wiley Online Library on [28/08/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
variations with the initial position of the simulated mean
Lagrangian velocity [Suciu and Vamos¸, 2007, Figure 2].
[16] Similar GRW simulations for smoother velocity
fields and larger ensembles of realizations led to values
M
22
smaller than Dt [Suciu et al., 2007b]. Preliminary test
simulations show that increasing the number of periodic
modes in the Kraichnan routine from 64 to 6400, for
transverse sources with L = 100 and a fixed number of
256 realizations, causes the increase of M
22
from values
smaller than 2Dt to about 5Dt [Suciu et al., 2008]. Since
larger number of modes yield closer approximations for the
random velocity fields with exponential correlations, con-
sidered in this numerical setup, the occurrence of non-zero
mean memory terms can be associated with the lack of
smoothness of the samples of such random fields [Yaglom,
1987]. Simulations based on the mixed finite element
method for both water flow [Radu et al., 2004] and solute
transport [Radu et al., 2008] will be compared in a forth-
coming paper with the present simulations, which are based
on first-order approximations of the velocity field. A
conclusion is yet premature and further work is needed to
clarify whether the mean memory terms reflect irregularities
of the velocity model or whether they are finite size effects
inherent in numerical simulations which always reproduce
the nominal values of the velocity statistics with some finite
precision.
[17] The mean memory terms are expected to be quanti-
tatively important for transport in inhomogeneous velocity
fields, when relation (9) is no longer verified and has to be
replaced by (7) and (8) [Suciu and Vamos¸ , 2007]. Here we
investigate memory effects on dispersion in single realiza-
tions of a homogeneous random velocity field. Neverthe-
less, because single realization quantities such as s
ll
and m
ll
are random variables, we cannot give up using ensemble
averages. Like in statistical inference problems [Yaglom,
1987], we need estimations by distances in the mean square
sense (ergodicity ranges), computed by averaging over
statistical ensembles. Since in this case the mean memory
terms are 1 or 2 orders of magnitude smaller than the large
standard deviations of the actual dispersion (see Figure 2),
they can be neglected. On the other hand, their standard
deviation will be shown to be relevant for the quantitative
evaluation of the memory effects presented in the following.
[18] Memory effects on single-realization dispersion are
demonstrated by the strong influence of the shape and
dimension of the source on the ergodicity range h of s
ll
S
ll
(0) with respect to the memory-free dispersion X
ll
pre-
sented in Figure 2. For large slab sources perpendicular to
l-axis h decreases with L, whereas for large extensions of
the source on the l-axis h strongly increases to values that at
early times are 1 or 2 orders of magnitude larger than for a
point source. The strongest increase is found in case of
longitudinal dispersion for longitudinal slabs.
[19] The ergodicity range h[s
ll
S
ll
(0)] = h(s
ll
S
ll
(0)
X
ll
)
2
i
V
1/2
was computed from the deviation of the mean
S
ll
S
ll
(0) X
ll
= h
1
and the standard deviation SD(s
ll
)= h
2
via h =[h
1
2
+ h
2
2
]
1/2
[Suciu et al., 2006a]. The longitudinal
one-particle dispersion X
11
was estimated by S
11
S
11
(0) in
case of transverse slab l 100l and the transverse one,
X
22
, by S
22
S
22
(0) in case of longitudinal slab 100l l.
Given the smallness of h in these cases, we approximate
s
ll
S
ll
(0) X
ll
. From (3) and the Cauchy-Schwartz
inequality, m
ll
2
(t) 4S
ll
(0)hh
e
X
l
i
D
2
i
X
0
, we also find that the
memory terms for slabs perpendicular to l-axis can be
neglected as compared with those for slabs oriented along
l. In these conditions, the second term of (2) estimates X
ll
and
the approximation (11) can be adopted. Since SD(s
ll
) and
h[s
ll
S
ll
(0)] practically coincide for L 10, regardless the
shape and the orientation of the source, it follows that h
1
0;
that is, the actual dispersion is an (almost) unbiased estima-
Figure 2. Ergodicity range h with respect to the memory-free dispersion and standard deviation SD(s
ll
)
of the actual dispersion (thin lines) for different shapes and extensions of the initial plume.
4 of 6
W08501 SUCIU ET AL.: TECHNICAL NOTE W08501
19447973, 2008, 8, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2007WR006740 by Cochrane Romania, Wiley Online Library on [28/08/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
tor of X
ll
. Then the ergodicity range is given by h
2
= SD(s
ll
),
which according to (11) is just the standard deviation SD(m
ll
)
of the memory terms.
4. Discussion and Conclusions
[20] Persistent influences of the shape and dimension of
the source on the second moments of the plume reported in
this paper are certainly related to memory terms consisting
of correlations between initial positions and displacements
of solute molecules. Such correlations naturally occur as
components of the dispersion whenever the dependence of
the trajectory on the initial position is considered explicitly,
as previously shown by Sposito and Dagan [1994]. For a
given direction, the memory terms increase with the in-
crease of the source dimension in the same direction and are
responsible for nonergodic behavior of the actual dispersion
with respect to the memory-free one-particle dispersion. The
uncertainty caused by the randomness of the center of mass,
instead, is reduced by increasing the dimension of the
source and is less sensitive to its shape and orientation.
[21] The analysis of the GRW simulations reveals that for
a typical situation of contaminant transport in aquifers the
memory terms can be tens to hundreds of times larger than
the one-particle dispersion. Their observed decay after some
hundreds of heterogeneity scales indicates that the displace-
ments averaged over local dispersion, h
e
X
l
i
D
, gradually
decorrelates from the initial position and (3) goes to zero.
This behavior is somehow expected for non-vanishing
local dispersion and ergodic velocity fields. The situation
could be different for advective transport in Darcy velocity
fields. Sposito [2001] has shown that the trajectories of
deterministic Darcy flows are generally confined on in-
variant subsets of the flow domain. It is therefore possible
that solute molecules, when driven by Darcy flows and in
absence of local dispersion, never lose the memory of
initial position and memory terms persist indefinitely, as
suggested by numerical investigations of Fiori and Jancovic´
[2005].
[22] Another extreme situation is when the relative dis-
placement
e
X
l
is independent of initial position X
0l
and the
memory term m
ll
defined by (3) vanishes. This happens, for
instance, when
e
X
l
is the superposition of a diffusion process
and a uniform movement with constant velocity, as in case
of confined stratified flows through a single fracture in
geological media [see, e.g., Dentz and Carrera, 2007].
Then, the transverse dimension of the source governs the
interplay between the local dispersion and the coherent
cross-section velocity profile. The significant relation for
the longitudinal dispersion is obtained from (1) by adding
and subtracting hh
e
X
1
i
D
2
i
X
0
, where h
e
X
1
i
D
is the center of
mass of a solute plume originating from a point source,
s
11
¼ S
11
0 ð Þþ
e
X
1
e
X
1
D
2
D
X0
þ
e
X
1
D
e
X
1
i
DX0
2
X0
:
ð12Þ
The last term in (12) is a spatial variance of the point-source
center of mass which carries the memory of initial
conditions: when it becomes negligible, the reduced
dispersion s
11
S
11
(0) behaves as a superposition of
point-source dispersions, hh[
e
X
1
h
e
X
1
i
D
]
2
i
D
i
X
0
.
[23] Memory terms of form (3) always vanish for point
sources, even if the relative displacements depend on initial
positions. But if the only available information is the
concentration distribution at a post-injection stage, which
has to be taken as an initial condition, memory terms cannot
be disregarded until the dispersion has reached a linear time
behavior [Sposito and Dagan, 1994]. Explicit expressions
of memory terms (3), derived from trajectory equations, are
then given by time integrals of correlations of the Lagrangian
velocity at pre- and post-injection times, which account for
the non-linear behavior of dispersion in pre-asymptotic
regime (see the preprint by Suciu et al., 2008).
[24] The dependence of the memory effects on the source
extension and anisotropy can be relevant for the calibration
of the model, predictions, and procedures for identification
of the source of contamination from available measurement
data. For example, it is known that for small sources the
best fit of measured second moments and theoretical
memory-free dispersion in velocity fields with finite corre-
lation range, X
ll
, can underestimate the variance and the
correlation length of the hydraulic conductivity [Suciu et al.,
2006b]. As shown in Figure 2, erroneous estimations will
also be obtained for sources with large extension on the
l direction. Instead, for large narrow sources perpendicular to
l, the ergodic behavior of the actual dispersion with respect
to X
ll
can be used in practice to improve the parameter
identification from field experiments.
[25] Acknowledgments. The research reported in this paper was
supported by Deutsche Forschungsgemeinschaft grant SU 415/1-2, Project
JICG41 at Ju¨lich Supercomputing Centre, Romanian Ministry of Education
and Research grant 2-CEx06-11-96, NATO Collaborative Linkage grant
ESP.NR.CLG 981426, and RFBR grant 06-01-00498.
References
Chile`s, J. P., and P. Delfiner (1999), Geostatistics: Modeling Spatial Un-
certainty, John Wiley, New York.
Dagan, G. (1990), Transport in heterogeneous porous formations: Spatial
moments, ergodicity and effective dispersion, Water Resour. Res., 26,
1281 – 1290.
Dentz, M., and J. Carrera (2007), Mixing and spreading in stratified flow,
Phys. Fluids, 19, 017107, doi:10.1063/1.2427089.
Dentz, M., H. Kinzelbach, S. Attinger, and W. Kinzelbach (2000), Temporal
behavior of a solute cloud in a heterogeneous porous medium: 2. Spatially
extended injection, Water Resour. Res., 36, 3605 – 3614.
Eberhard, J., N. Suciu, and C. Vamos¸ (2007), On the self-averaging of
dispersion for transport in quasi-periodic random media, J. Phys. A Math.
Theor., 40, 597 – 610, doi:10.1088/1751-8113/40/4/002.
Fiori, A., and I. Jancovic´ (2005), Can we determine the transverse macro-
dispersivity by using the method of moments?, Adv. Water Resour., 28,
589 – 599, doi:10.1016/j.advwaters.2004.09.909.
Kitanidis, P. K. (1988), Prediction by the method of moments of transport in
a heterogeneous formation, J. Hydrol., 102, 453 – 473.
Le Doussal, P., and J. Machta (1989), Annealed versus quenched diffusion
coefficient in random media, Phys. Rev. B, 40(12), 9427 – 9430.
Morales-Casique, E., S. P. Neuman, and A. Gaudagnini (2006), Nonlocal
and localized analyses of nonreactive solute transport in bounded ran-
domly heterogeneous porous media: Theoretical framework, Adv. Water
Resour., 29, 1238 – 1255, doi:10.1016/j.advwatres.2005.10.002.
Naff, R. L., D. F. Haley, and E. A. Sudicky (1998), High-resolution
Monte Carlo simulation of flow and conservative transport in hetero-
geneous porous media: 2. Transport results, Water Resour. Res., 34,
679 – 697.
Phythian, R., and W. D. Curtis (1978), The effective long-time diffusivity
for a passive scalar in a Gaussian model fluid flow, J. Fluid Mech., 89(2),
241 – 260.
Radu, F. A., I. S. Pop, and P. Knabner (2004), Order of convergence
estimates for an Euler implicit, mixed finite element discretization of
Richards’ equation, SIAM J. Numer. Anal. , 42(4), 1452–1478,
doi:10.1137/S0036142902405229.
W08501 SUCIU ET AL.: TECHNICAL NOTE
5 of 6
W08501
19447973, 2008, 8, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2007WR006740 by Cochrane Romania, Wiley Online Library on [28/08/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Radu, F. A., I. S. Pop, and S. Attinger (2008), Analysis of an Euler
implicit – mixed finite element scheme for reactive solute transport in
porous media, CASA Rep. 08-06, Tech. Univ. Eindhoven, Eindhoven,
Netherlands.
Sposito, G. (2001), Topological groundwater hydrodynamics, Adv. Water
Resour., 24, 793 – 801.
Sposito, G., and G. Dagan (1994), Predicting solute plume evolution in
heterogeneous porous formations, Water Resour. Res., 30, 585 – 589.
Suciu, N., and C. Vamos¸ (2007), Comment on ‘‘Nonstationary flow
and nonergodic transport in random porous media’’ by G. Darvini
and P. Salandin, Water Resour. Res., 43, W12601, doi:10.1029/
2007WR005946.
Suciu, N., C. Vamos¸, J. Vanderborght, H. Hardelauf, and H. Vereecken
(2006a), Numerical investigations on ergodicity of solute transport in
heterogeneous aquifers, Water Resour. Res., 42, W04409, doi:10.1029/
2005WR004546.
Suciu, N., C. Vamos¸, and J. Eberhard (2006b), Evaluation of the first-order
approximations for transport in heterogeneous media, Water Resour. Res.,
42, W11504, doi:10.1029/2005WR004714.
Suciu, N., C. Vamos¸, and K. Sabelfeld (2007a), Ergodic simulations of diffu-
sion in random velocity fields, in Monte Carlo and Quasi-Monte Carlo
Methods 2006, edited by A. Keller, S. Heinrich, and H. Niederreiter,
pp. 659– 668, Springer, Heidelberg, Germany.
Suciu, N., C. Vamos¸, K. Sabelfeld, and C. Andronache (2007b), Memory
effects and ergodicity for diffusion in spatially correlated velocity fields,
Proc. Appl. Math. Mech., 7, 2010015, doi:10.1002/pamm.20070057.
Suciu, N., C. Vamos¸, H. Vereecken, K. Sabelfeld, and P. Knabner (2008),
Lagrangian stationarity and memory effects for dispersion in ergodic
velocity fields, paper presented at General Assembly 2008, Eur. Geosci.
Union, Vienna, 13 – 18 April.
Vamos¸, C., N. Suciu, and H. Vereecken (2003), Generalized random walk
algorithm for the numerical modeling of complex diffusion processes,
J. Comput. Phys., 186(2), 527 – 544, doi:10.1016/S0021-9991(03)00073-1.
van Kampen, N. G. (1981), Stochastic Processes in Physics and Chemistry,
North-Holland, Amsterdam.
Yaglom, A. M. (1987), Correlation Theory of Stationary and Related
Random Functions, vol. 1, Basic Results, Springer, New York.
Zirbel, C. L. (2001), Lagrangian observations of homogeneous random
environments, Adv. Appl. Prob., 33, 810 – 835.
P. Knabner and N. Suciu, Institute of Applied Mathematics,
Friedrich-Alexander University of Erlangen-Nuremberg, Martensstrasse 3,
D-91058 Erlangen, Germany. (suciu@am.uni-erlangen.de)
K. Sabelfeld, Weierstrass Institute for Applied Analysis and Stochastics,
Mohrenstrasse 39, D-10117 Berlin, Germany.
C. Vamos¸, Tiberiu Popoviciu Institute of Numerical Analysis, Romanian
Academy, P. O. Box 68-1, 400320 Cluj-Napoca, Romania.
H. Vereecken, ICG-IV, Research Center Ju¨lich, D-52425 Ju¨lich,
Germany.
6 of 6
W08501 SUCIU ET AL.: TECHNICAL NOTE W08501
19447973, 2008, 8, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2007WR006740 by Cochrane Romania, Wiley Online Library on [28/08/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License