Nonlinear Multigrid Methods for

Numerical Solution of the Variably

Saturated Flow Equation in Two Space
Dimensions

Gheorghe Juncu, Aurelian Nicola &

Constantin Popa

Transport in Porous Media

ISSN 0169-3913
Volume 91
Number 1

Transp Porous Med (2012) 91:35-47

DOI 10.1007/s11242-011-9831-9

1 23
Your article is protected by copyright and
all rights are held exclusively by Springer
Science+Business Media B.V.. This e-offprint
is for personal use only and shall not be self-
archived in electronic repositories. If you
wish to self-archive your work, please use the
accepted author’s version for posting to your
own website or your institution’s repository.
You may further deposit the accepted author’s
version on a funder’s repository at a funder’s
request, provided it is not made publicly
available until 12 months after publication.

1 23
 Author's personal copy
Transp Porous Med (2012) 91:35–47
DOI 10.1007/s11242-011-9831-9

Nonlinear Multigrid Methods for Numerical Solution

of the Variably Saturated Flow Equation in Two Space
Dimensions

Gheorghe Juncu · Aurelian Nicola · Constantin Popa

Received: 21 December 2010 / Accepted: 23 August 2011 / Published online: 7 September 2011
© Springer Science+Business Media B.V. 2011

Abstract The need of accurate and efficient numerical schemes to solve Richards’ equa-
tion is well recognized. This study is carried out to examine the numerical performances
of the nonlinear multigrid method for numerical solving of the two-dimensional Richards’
equation modeling water flow in variably saturated porous media. The numerical approach
is based on an implicit, second-order accurate time discretization combined with a vertex
centered finite volume method for spatial discretization. The test problems simulate infiltra-
tion of water in 2D saturated–unsaturated soils with hydraulic properties described by van
Genuchten–Mualem models. The numerical results obtained are compared with those
provided by the modified Picard–preconditioned conjugated gradient (Krylov subspace)
approach.

Keywords Saturated–unsaturated porous media · Richards’ equation · Finite volume ·

Nonlinear multigrid algorithm

1 Introduction

Richards’ equation (RE) is the commonly accepted mathematical model for water flow in
variably saturated porous media (Bear and Bachmat 1991; Smith et al. 2002). RE models the
flow in porous media of a wetting fluid (water) in the presence of a non-wetting fluid (air)
assumed to be at constant pressure. From the mathematical point of view, RE is a highly

G. Juncu (B)
Politehnica University Bucharest, Catedra Inginerie Chimica, Polizu 1, Bucharest 011061, Romania
e-mail: juncugh@netscape.net; juncu@easynet.ro

A. Nicola · C. Popa
Department of Mathematics, Ovidius University, Bl. Mamaia 124, Constanta 900527, Romania
e-mail: anicola@univ-ovidius.ro
C. Popa
e-mail: cpopa@univ-ovidius.ro

123
 Author's personal copy
36 G. Juncu et al.

nonlinear, possibly degenerate, parabolic differential equation. For these reasons, numerical
approximations are widely used to obtain the solution to the RE.
There is a wide range of numerical techniques employed in a variety of ways for the
numerical solution of RE. For the case of saturated–unsaturated flow, we mention the con-
tributions of Forsyth et al. (1995), Pan and Wierenga (1995), Diersch and Perrochet (1999),
Berganaschi and Putti (1999), Wiliams et al. (2000), Lima-Vivancos and Voller (2004), Bause
and Knabner (2004), Hao et al. (2005), Marinoschi (2005), Basombrio et al. (2006), McBride
et al. (2006), Pei et al. (2006), Kees et al. (2008), Bevilacqua et al. (2009), Casulli and Zanolli
(2010), Radu et al. (2010), and Wu (2010). For 2D and 3D problems, the standard numerical
approach consists of:

– space discretization by finite volume, finite element (Galerkin), mixed finite element,
finite differences;
– method of lines for time discretization;
– solution of the discrete nonlinear problems using Newton/Picard method;
– solution of the linear problems by iterative linear solvers (preconditioned conjugated
gradient, preconditioned Krylov subspace methods, algebraic multigrid, and so on).

A review of the software packages developed can be viewed in Simunek et al. (2008).
Multigrid (MG) is one of the most powerful iterative methods to solve numerically non-
linear partial differential equations. Two strategies can be used (Hackbusch 1985): (1) the
so-called nonlinear MG (the linearization employed is only local) and (2) linear MG (geomet-
ric MG or algebraic MG) as linear solver of the Newton iteration. The experience accumulated
until now in the numerical solving of RE with nonlinear MG is rather limited. McKeon and
Chu (1987) and Kees et al. (2006) used nonlinear MG for solving steady-state variably sat-
urated flow models. Calvo and Lisbona (2001) and Juncu et al. (2010) solved the transient
water infiltration into unsaturated soils using the dimensionless water content as unknown for
RE. This formulation of RE is not recommended when part of the domain becomes saturated.
The present study represents an effort to employ the nonlinear MG method in the numerical
solution of the RE for the case of transient saturated–unsaturated flow.
This article is organized as follows. In Sect. 2, we describe the mathematical model of
the test problem. Section 3 presents the numerical algorithm. The numerical experiments are
made and the results obtained are presented in Sect. 4. Finally, some concluding remarks are
briefly mentioned in Sect. 5.

2 Mathematical Model

The test problem describes a traveling wave in a soil (Caputo and Stepanyants 2008). From the
numerical point of view, traveling waves cannot be considered an elementary test problem.
In this study, we are interested in saturated–unsaturated flow. Therefore, we retain the
pressure head-based form of RE,
∂ϕ ∂K
C(ϕ) − ∇· (K (ϕ)∇ϕ) + =0 (1)
∂t ∂z
where ϕ (t, x, z) is the pressure head, K (ϕ) is the hydraulic conductivity, and (x, z) ∈
[0, L x ] × [0, L z ]. C(ϕ) is the specific moisture capacity defined by,
dθ
C(ϕ) =
dϕ

123
 Author's personal copy
Nonlinear Multigrid Methods for Numerical Solution 37

where θ (t, x, z) is the moisture (water) content. The vertical coordinate z is considered
positive downward.
In the numerical experiments made, we considered L x = 1 m and L z = 2 m. The initial
and boundary conditions are:

t = 0, ϕ(0, x, z) = ϕ0 (1 − exp (−80z)) , (2a)

z = 0, ϕ (t, x, 0) = f (x), (2b)
z = L z , ϕ (t, x, L z ) = ϕ0 , (2c)
∂ϕ
x = 0(L x ), = 0. (2d)
∂x
The expressions considered for f (x) are:

φ0 ≤ 0, x < 0.25 or x > 0.75
f (x) = (3a)
0.1, 0.25 ≤ x ≤ 0.75
 
1 (x − 0.5)2
f (x) = √ exp − −χ (3b)
σ 2π 2σ 2
f (x) = constant = 0, 10−3 , 10−2 , 10−1 . (3c)

The numerical values for χ, σ , and φ0 depend on the soil water retention curve. The values
of ϕ0 were selected such that the dimensionless water content, u, satisfies u(ϕ0 ) ∼
= 0.1.
To complete the model for variably saturated flow, the constitutive relationships that
describe the interdependencies of pressure head, hydraulic conductivity, and the water con-
tent need to be specified. The model used within this study to characterize the hydraulic
properties of the soils is the van Genuchten–Mualem model (Van Genuchten 1980),

n m, ϕ < 0
1
θ − θr
u= = (1+|αϕ| ) (4a)
θs − θr 1, ϕ≥0

√   
1 m 2
K (u) = K s u 1 − 1 − u m , ϕ < 0 m/h (4b)
Ks, ϕ≥0

where K s , m/h, is the saturated conductivity, m = 1 − n −1 , α, m−1 , and n are shape param-
eters, θs , m3 /m3 , and θr , m3 /m3 , are the saturated and residual water contents, respectively.
Note that u = (θ − θr )/(θs − θr ) is the dimensionless water content, θ (t, x, z), m3 /m3 , being
the water content.
For the van Genuchten–Mualem model, the specific moisture capacity C(ϕ), is given by,

−(m+1)
|αϕ|n |αϕ|n−1 , ϕ < 0
C(ϕ) = (θs − θr ) αmn 1 + m−1 (4c)
0, ϕ≥0

3 Numerical Method

The spatial derivatives of Eq. 1 were discretized with the finite volume method on a uniform
vertex-centered grid with N x × Nz points (see Fig. 1). The control volumes are defined by the
lines that connect the mid-points between nodes. The time discretization is the fully implicit
second-order accurate finite difference scheme. The discrete approximation obtained is:

123
 Author's personal copy
38 G. Juncu et al.

Fig. 1 Discretization of the computational domain

  3ϕ k+1 − 4ϕ k + ϕ k−1   
i, j i, j i, j
C ϕi,k+1
j h 2
− K k+1
i, j+1/2 ϕ k+1
i, j+1 − ϕ k+1
i, j
 2
t    
k+1 k+1 k+1 k+1 k+1 k+1 k+1
− K i, j−1/2 ϕi, j − ϕi, j−1 − K i+1/2, j ϕi+1, j − ϕi, j − K i−1/2, j
   
× ϕi,k+1 k+1
j − ϕi−1, j + h K i,k+1 k+1
j+1/2 − K i, j−1/2 = 0 (5)

where h is the grid step size. In order to simplify the presentation, we have considered in (5)
only the case when the time step is constant,
t = constant. The generalization to the case
variable time step is straightforward. The inter-nodal values of the hydraulic conductivity
were calculated as arithmetic averages of the grid point values. We do not insist here on the
problem of inter-nodal values calculation (Szymkiewicz 2009) because we are interested in
numerical solutions on very fine grids.
The numerical strategy used to solve (5) belongs to the class named implicit discretization
with explicit linearization (Haverkamp et al. 1977). The nonlinear iteration applied to Eq. 5
works as follows: if we know the values of ϕ after (l − 1) iterations, the values corresponding
to the l-th iteration are computed by applying one iteration sweep to the equation
  3ϕ k+1,l − ϕ k + ϕ k−1
i, j i, j i, j
C ϕi,k+1,l−1
j h2
  2
t   
− K i,k+1,l−1 k+1,l
j+1/2 ϕi, j+1 − ϕi, j
k+1,l
− K i,k+1,l−1 k+1,l
j−1/2 ϕi, j − ϕi,k+1,l
j−1
    
k+1,l−1 k+1,l k+1,l k+1,l−1 k+1,l k+1,l
− K i+1/2, j ϕi+1, j − ϕi, j − K i−1/2, j ϕi, j − ϕi−1, j
 
+ h K i,k+1,l−1
j+1/2 − K k+1,l−1
i, j−1/2 =0 (6)

Two nonlinear iterations are used in this study: nonlinear MG and modified Picard (Lehmann
and Ackerer 1998). The nonlinear MG algorithm used is the classical Full Approximation
Storage (FAS) algorithm (Brandt 1977), suitable for both linear and nonlinear problems.
The algorithm was presented in detail in Juncu et al. (2010). We considered that it is not
necessary to repeat here its description in detail. However, the following important problem

123
 Author's personal copy
Nonlinear Multigrid Methods for Numerical Solution 39

 
should be mentioned: the values of C ϕi,k+1,l−1
j should not be changed during one MG
cycle (iteration); these values are calculated once, on the finest grid, at the beginning of the
MG cycle; on the coarse grids, these values were used. It should be also mentioned that in
all numerical experiments, the coarser grid has N x × Nz = 17 × 33 points (the spatial step
size was equal to h = 2−4 ).
When ϕ (t, x, z) = 0, RE becomes a degenerate equation, in the sense that locally it
changes type from parabolic (unsaturated zone) to elliptic (saturated zone). For numerical
solving of degenerate nonlinear PDEs with nonlinear MG, Brandt (1977) recommends the
using of line or alternating line Gauss–Seidel methods as smoothing algorithms. In this study,
the alternating line Gauss–Seidel method, presented in detail in Juncu et al. (2010), was used
as smoothing algorithm.
The linear solvers used in the modified Picard iteration are CGS (Sonneveld 1989),
BI-CGSTAB (Van der Vorst 1992), and GMRES (Saad and Schultz 1986). The software
for CGS was written by the present authors. It was successfully used to solve difficult prob-
lems in Juncu and Popa (2000); Juncu et al. (2006); Juncu (2007). The software packages
used for BI-CGSTAB and GMRES are those developed by M.A. Botchev (BICGSTAB (ell),
ell = 1 in this study, and GMRERS) and available free on internet from www.staf.science.uu.
nl/$\sim$vorst102/software.html. Left preconditioning using the normalization technique
(Weiss 1996) or the incomplete Cholesky factorization that preserves the sparsity structure
of the original matrix, IC (1,1), (Meijerink and van der Vorst 1981) was tested.
The stopping criterion for the nonlinear iterations is:

resi 2
< 10−6 (7)
res0 2

where resi is the residual of (5) after i iterations and 2 is the discrete Euclidean norm. The
same stopping criterion was used for the linear solvers.
The time step was variable. The minimum value of the time step was
t = 10−6 h. This
value was used especially at the start of the time integration. The maximum value of the
time step was
t ≈ 10−3 h. The time step has been controlled by the heuristic time stepping
strategy (D’Haese et al. 2007). We stopped the numerical simulations when the wave reaches
the depth z, 1.4 m < z < 1.5 m (in order to avoid the distortion of the wave).

4 Numerical Results

The numerical performances of the numerical algorithms are monitored by the average num-
ber of nonlinear iterations per time step, ITERn, and the average efficiency per time step, τ .
The efficiency was calculated as (Hackbusch 1985):

W
τ= (8)
|ln ρ|

where W (the work) is the number of arithmetic operations per grid point and iteration step
and ρ is the average reduction factor. The average reduction factor was calculated with the
relation,
 1
resi 2 i
 1
ρ= ∼
= 10−6 ITERn (9)
res0 2

123
 Author's personal copy
40 G. Juncu et al.

Table 1 Average ITERn values

Nx × Nz Number of levels for MG iteration ITERn

Nonlinear MG Modified Picard

65 × 129 3 5.27 5.35

129 × 257 4 5.94 6.26
257 × 513 5 6.86 7.49

Table 2 Average ITERl values

Nx × Nz ITERl

PCGS PBI-CGSTAB (1) PGMRESR

65 × 129 17.83–29.37 15.25–22.28 26.71–42.12

129 × 257 36.10–57.21 32.84–44.61 45.42–68.60
257 × 513 55.69–84.72 51.25–65.41 63.39–92.89

Table 3 Average values for the efficiency, τ

Nx × Nz Number of levels for MG iteration τ

Nonlinear MG Modified Picard

65 × 129 3 125.12 322.46–463.73

129 × 257 4 145.32 792.33–1069.68
257 × 513 5 171.31 1467.04–1866.21

The average reduction factor is one of the quantities used to measure the convergence rate
of an algorithm. The connection between ITERn and the average reduction factor is obvious.
The efficiency can be viewed as an indicator for the computational time.
The work, W , for the modified Picard iteration depends on the convergence rate of the
inner iteration (the preconditioned conjugate gradient algorithm). For this reason, we consid-
ered it necessary to present data about the convergence rate of the preconditioned conjugate
gradient algorithms. The quantity used to express the convergence rate of CGS, BI-CGSTAB
(1), and GMRERS is the number of iterations per nonlinear (Picard) iteration, ITERl. It must
be mentioned that for GMRERS, the truncation parameter was set equal to 5 and the version
parameter was set equal to 1.
The values of ITERn and ITERl are presented in Tables 1 and 2, respectively. The val-
ues of ITERl presented in Table 2 were obtained for the algorithms preconditioned with IC
(1,1). The preconditioning with the normalization technique (Weiss 1996) does not improve
the convergence rate of the algorithms while the preconditioning with IC (1,1) improves it
significantly. The values of τ are presented in Table 3. The soils employed in the numerical
experiments are presented in Table 4. Some of the solutions computed in this study are pre-
sented graphically in Figs. 2, 3, 4, and 5. It should be mentioned that the numerical solutions
provided by different algorithms are identical.
For the soils V (sandy loam) and VI (clay) from Table 4, soils with n < 1.50, both non-
linear iterations used in this study fail. The convergence criterion, relation (7), is not verified
in a prescribed maximum number, ITERn_MAX = 50, of nonlinear iterations.

123
 Author's personal copy
Nonlinear Multigrid Methods for Numerical Solution 41

Table 4 Parameters values

No. n α (m−1 ) K s (m/h) θs θr Reference

I 4.26 5.478 0.21 0.301 0.093 Tocci et al. (1998)

II 3.0 3.50 0.208 0.38 0.044 Durner et al. (2008)
III 2.0 3.35 0.332083 0.368 0.102 Tocci et al. (1998)
IV 1.6745 0.58 0.012825 0.473 0.058 Durner et al. (2008)
V 1.40 3.0 0.0167 0.40 0.05 Durner et al. (2008)
VI 1.1845 1.99 0.007479 0.5169 0.1045 Durner et al. (2008)

Fig. 2 Pressure head profiles for water infiltration in soil I: f (x) from (3a) with φ0 = −0.1; a t = 0.5 h;
b t = 1.0 h

123
 Author's personal copy
42 G. Juncu et al.

Fig. 3 Pressure head profiles for water infiltration in soil II: f (x) from (3b) with σ = 0.50 and χ = 0.6978;
a t = 0.75 h, b t = 1.50 h

It is well known that, for the Mualem–van Genuchten model, the hydraulic conductiv-
ity K is not Lipschitz continuous, much less differentiable, at ϕ = 0 if 1 < n < 2. To
solve this problem, different solutions were proposed (see for example, Ippisch et al. 2006;
Schaap and van Genuchten 2006). A different approach to this problem, based on an integral
transformation, can be viewed in Ciutureanu and Marinoschi (2008) and Marinoschi (2010).
However, until now, a standard solution did not impose itself. For this reason, we did not
insist too much on this problem.
The data presented in Tables 1, 2, and 3 were obtained for soils I–IV from Table 4.
Concerning the computation of these data, the following statements should be made:

123
 Author's personal copy
Nonlinear Multigrid Methods for Numerical Solution 43

Fig. 4 Pressure head profiles for water infiltration in soil III: f (x) from (3b) with σ = 0.50 and χ = 0.6978;
a t = 0.4 h, b t = 1.0 h

– the soil type and the expressions for f (x) [see relations (3)] do not influence signifi-
cantly the values of ITERn; the data presented in Table 1 are average values calculated
for all the expressions of f (x) and the soils I–IV depicted in Table 4; the confidence inter-
val for these average values is approximately ±15%; for the modified Picard iteration,
ITERn does not depend on the linear solver and the preconditioning used;
– when two algorithms solved the same problem on the same grid in a different number of
time steps or an algorithm solved the same problem on two different grids in a different
number of time steps, the necessary adjustments in the values of ITERn were made;
– the expressions for f (x) do not influence significantly the values of ITERl but the soil
type influences ITERl; for this reason, in Table 2, two values for ITERl are presented; the

123
 Author's personal copy
44 G. Juncu et al.

Fig. 5 Pressure head profiles for water infiltration in soil IV: f (x) from (3c) with f (x) = 0.1 = constant;
a t = 10.0 h; b t = 26.0 h

minimum and the maximum value, respectively; the minimum values were obtained for
soils I, II, and IV from Table 4 (the value depicted in Table 2 is an average value calculated
for these three soils); the maximum value was obtained for soil III from Table 4;
– under these conditions, for the modified Picard iteration, two values of τ are presented
in Table 3; the minimum value was calculated for soils I, II, IV, and preconditioned
BI-CGSTAB (1) [PBI-CGSTAB (1)] or preconditioned CGS (PCGS) as linear solver
(Table 2 shows that for soils I, II, and IV, PCGS and PBI-CGSTAB (1) perform better
than the preconditioned GMRESR (PGMRESR); the efficiencies of PCGS and PBI-
CGSTAB (1) are approximately equal); the maximum value was calculated for soil III

123
 Author's personal copy
Nonlinear Multigrid Methods for Numerical Solution 45

and PBI-CGSTAB (1) as linear solver (Table 2 shows that for soil III, PBI-CGSTAB (1)
performs better than PCGS and PGMRESR).

Table 1 shows that the convergence rates of the two nonlinear algorithms used in this study
are comparable. For a given grid, the values of ITERn do not differ significantly. Also, the
mesh behavior of the nonlinear MG and the modified Picard iteration is similar. The increase
in the number of grid points from 65 × 129 to 257 × 513 increases ITERn only with 30%
(for nonlinear MG)–40% (for modified Picard).
Table 3 shows that nonlinear MG outperforms very clearly the modified Picard iteration.
The differences in τ increase considerably with the increase in the number of grid points.
Only for the grid with 65 × 129 points the efficiency of the modified Picard iteration may be
considered comparable to that of nonlinear MG. In all the other cases, the nonlinear MG is
considerably more efficient than the modified Picard iteration.
Figures 2, 3, 4, and 5 illustrate pressure head profiles during the infiltration process. Two
times were selected in order to put into evidence the evolution of the saturated region. In
Figs. 2, 3, and 4, due to the boundary conditions (2a) and (2b), the evolution in time of the
saturated region depends on both spatial coordinates x and z. The wetting fronts have also a
2D shape. The profiles presented in Fig. 5 are 1D. In this case, the boundary condition (3c)
was used. Figures 2, 3, 4, and 5 also show that the rate of advance of the water front depends
strongly on the properties of the soil.

5 Conclusions

In this study, we have analyzed the numerical performances of the nonlinear MG algorithm for
solving RE in two space dimensions in the unsaturated–saturated regime. Different types of
soils were tested. The van Genuchten–Mualem model was used to characterize the hydraulic
properties of the soils. The numerical results obtained with the nonlinear MG algorithm are
compared with those provided by the modified Picard–preconditioned conjugated gradient
(Krylov subspace) method.
The numerical experiments presented in the previous section can be summarized as fol-
lows:

– both nonlinear iterative methods (nonlinear MG and modified Picard) do not converge
for soils with n < 1.5;
– for soils with n > 1.5, when the methods converge, the convergence rate does not depend
significantly on the type of the soil;
– the convergence rate of the two algorithms is comparable; the mesh behavior of the
convergence rate of MG is slightly superior to that of the modified Picard method;
– the mesh behavior of the efficiency of the nonlinear MG algorithm is excellent (as in
many other areas of computational fluid dynamics); on very fine grids (257 × 513 grid
points) the nonlinear MG is approximately ten times faster than the modified Picard
iteration (we refer to the computational time).

The numerical performances of the nonlinear MG method for solving transient saturated–
unsaturated flow in porous media are analyzed for the first time in this study. For this reason,
we think that it is premature to draw a definitive conclusion concerning the numerical abili-
ties of the nonlinear MG algorithm. The case of multi-layered soils should also be taken into
consideration and this will be a challenge for the future.

123
 Author's personal copy
46 G. Juncu et al.

References

Basombrio, F.G., Guarracino, L., Vénere, M.J.: A non-iterative algorithm based on Richardson’s extrapolation.
Application to groundwater flow modeling. Int. J. Numer. Methods Eng. 65, 1088–1112 (2006)
Bause, M., Knabner, P.: Computation of variably saturated subsurface flow by adaptive mixed hybrid finite
element methods. Adv. Water Resour. 27, 565–581 (2004)
Bear, J., Bachmat, Y.: Introduction to Modelling of Transport Phenomena in Porous Media. Kluwer, Dordr-
echt (1991)
Berganaschi, I., Putti, M.: Mixed finite elements and Newton-type linearizations for the solution of Richards’
equation. Int. J. Numer. Methods Eng. 45, 1025–1046 (1999)
Bevilacqua, I., Canone, D., Ferraris, S.: Acceleration techniques for the iterative resolution of the Richards
equation by the finite volume method. Int. J. Numer. Methods Biomed. Eng. (2009). doi:10.1002/cnm.
1364
Brandt, A.: Multi-level adaptive solutions to boundary-value problems. Math. Comput. 31, 333–390 (1977)
Calvo, P., Lisbona, F.: A finite volume-multigrid method for flow simulation on stratified porous media on
curvilinear co-ordinate systems. Int. J. Numer. Methods Fluids 37, 375–397 (2001)
Caputo, J.G., Stepanyants, Y.A.: Front solutions on Richards’ equation. Transp. Porous Media 74, 1–20 (2008)
Casulli, V., Zanolli, P.: A nested Newton-type algorithm for finite volume methods solving Richards’ equation
in mixed form. SIAM J. Sci. Comput. 32, 2255–2273 (2010)
Ciutureanu, C., Marinoschi, G.: Convergence of the finite difference scheme for a fast diffusion equation in
porous media. Numer. Funct. Anal. Optim. 29, 1034–1063 (2008)
D’Haese, C.M.F., Putti, M., Paniconi, C., Verhoest, N.E.C.: Assessment of adaptive and heuristic time stepping
for variably saturated flow. Int. J. Numer. Methods Fluids 53, 1173–1179 (2007)
Diersch, H.J.G., Perrochet, P.: On the primary variable switching technique for simulating unsaturated–
saturated flows. Adv. Water Resour. 23, 271–301 (1999)
Durner, W., Jansen, U., Iden, S.C.: Effective hydraulic properties of layered soils at the lysimeter scale deter-
mined by inverse modeling. Eur. J. Soil Sci. 59, 114–124 (2008)
Forsyth, P.A., Wu, Y.S., Priess, K.: Robust numerical methods for saturated–unsaturated flow with dry initial
conditions in heterogeneous media. Adv. Water Resour. 20, 1–14 (1995)
Hackbusch, W.: Multi-Grid Methods and Applications. Springer, Berlin (1985)
Hao, X., Zhang, R., Kravchenko, A.: A mass-conservative switching method for simulating saturated–
unsaturated flow. J. Hydrol. 311, 254–265 (2005)
Haverkamp, R., Vauclin, M., Touma, J., Wierenga, P.J., Vachaud, G.: A comparison of numerical simulation
models for one-dimensional infiltration. Soil Sci. Soc. Am. J. 41, 285–294 (1977)
Ippisch, O., Vogel, H.-J., Bastian, P.: Validity limits for the van Genuchten–Mualem model and implications
for parameter estimation and numerical simulation. Adv. Water Resour. 29, 1780–1789 (2006)
Juncu, Gh., Popa, C.: Numerical experiments with preconditioning by Gram matrix approximation for non-
linear elliptic equations. Math. Comput. Simul. 52, 53–71 (2000)
Juncu, Gh., Mosekilde, M., Popa, C.: Numerical experiments with MG continuation algorithms. Appl. Numer.
Math. 56, 844–861 (2006)
Juncu, Gh.: Multiplicity analysis of a nonisothermal finite cylindrical catalyst pellet. Int. J. Heat Mass
Transf. 50, 2038–2050 (2007)
Juncu, Gh., Nicola, A., Popa, C.: Nonlinear multigrid methods for numerical solution of the unsaturated flow
equation in two space dimensions. Transp. Porous Media 83, 637–652 (2010)
Kees, C.E., Farthing, M.W., Howington, S.E., Jenkins, E.W., Kelley, C.T.: Nonlinear Multilevel Iterative
Methods for Multiscale Models of Air / Water Flow in Porous Media. In: Proceedings of Computational
Methods in Water Resources XVI, Paper No. 148 (2006)
Kees, C.E., Farthing, M.W., Dawson, C.N.: Locally conservative, stabilized finite element methods for variably
saturated flow. Comput. Meth. Appl. Mech. Eng. 197, 4610–4625 (2008)
Lehmann, F., Ackerer, P.H.: Comparison of iterative methods for improved solution of the fluid flow equation
in partially saturated porous media. Transp. Porous Media 31, 275–292 (1998)
Lima-Vivancos, V., Voller, V.R.: Two numerical methods for modeling variably saturated flow in layered
media. Vadose Zone J. 3, 1031–1037 (2004)
Marinoschi, G.: A free boundary problem describing the saturated–unsaturated flow in a porous medium. II.
Existence of the free boundary in the 3D case. Abstr. Appl. Anal. 2005, 813–854 (2005)
Marinoschi, G.: Well posedness of a time-difference scheme for a degenerate fast diffusion prob-
lem. DCDS 13, 435–454 (2010)
McBride, D., Cross, M., Croft, N., Bennett, C., Gebhardt, J.: Computational modeling of variably saturated flow
in porous media with complex three-dimensional geometries. Int. J. Numer. Methods Fluids 50, 1085–
1117 (2006)

123
 Author's personal copy
Nonlinear Multigrid Methods for Numerical Solution 47

McKeon, T.J., Chu, W.-S.: A multigrid model for steady flow in partially saturated porous media. Water
Resour. Res. 23, 542–550 (1987)
Meijerink, J., van der Vorst, H.: Guidelines for the usage of incomplete decompositions in solving sets of
linear equations as they occur in practical problems. J. Comput. Phys. 44, 134–155 (1981)
Pan, L., Wierenga, P.J.: A transformed pressure head-based approach to solve Richards equation for variably
saturated soils. Water Resour. Res. 31, 925–931 (1995)
Pei, Y., Wang, J., Tian, Z., Yu, J.: Analysis of interfacial error in saturated–unsaturated flow models. Adv.
Water Resour. 29, 515–524 (2006)
Radu, F.A., Pop, I.S., Attinger, S.: Analysis of an Euler implicit-mixed finite element scheme for reactive
solute transport in porous media. Numer. Meth. Partial Diff. Eq. 26, 320–344 (2010)
Saad, Y., Schultz, M.H.: GMRES: a generalized minimal residual algorithm for solving nonsymmetric linear
systems. SIAM J. Sci. Statist. Comput. 7, 856–869 (1986)
Schaap, M.G., van Genuchten, M.T.: A modified Mualem-van Genuchten formulation for improved description
of the hydraulic conductivity near saturation. Vadose Zone J. 5, 27–34 (2006)
Simunek, J., van Genuchten, M.T., Sejna, M.: Development and applications of the HYDRUS and STANMOD
software packages and related codes. Vadose Zone J. 7, 587–600 (2008)
Smith, R.E., Smetten, K.J., Broadbridge, P., Woolhiser, D.A.: Infiltration Theory for Hydrologic Applica-
tions. AGU, Washington (2002)
Sonneveld, P.: CGS, a fast Lanczos-type solver for nonsymmetric linear systems. SIAM J. Sci. Statist. Comput.
10, 36–52 (1989)
Szymkiewicz, A.: Approximation of internodal conductivities in numerical simulation of one-dimensional
infiltration, drainage, and capillary rise in unsaturated soils. Water Resour. Res. 45, W10403 (2009)
Tocci, M.D., Kelley, C.T., Miller, C.T., Kees, C.E.: Inexact Newton methods and the method of lines for
solving Richards’ equation in two space dimension. Comput. Geosci. 2, 291–309 (1998)
Van der Vorst, H.A.: BI-CGSTAB: a fast and smoothly converging variant of BI-CG for the solution of
nonsymmetric linear systems. SIAM J. Sci. Statist. Comput. 13, 631–644 (1992)
Van Genuchten, M.T.: A closed form equation for predicting the hydraulic conductivity of unsaturated
soils. Soil Sci. Soc. Am. J. 44, 892–898 (1980)
Weiss, R.: Parameter-free iterative linear solvers. Akademie, Berlin (1996)
Wiliams, G.A., Miller, C.T., Kelley, C.T.: Transformation approaches for simulating flow in variably saturated
porous media. Water Resour. Res. 36, 923–934 (2000)
Wu, M.: A finite-element algorithm for modeling variably saturated flows. J. Hydrol. 394, 315–323 (2010)

123