Conservative iterative methods for implicit

arXiv:2106.10088v1 [math.NA] 18 Jun 2021

discretizations of conservation laws

Philipp Birken1 , Viktor Linders1

1
Centre for mathematical sciences, Lund University, Lund, Sweden.
email: [email protected]
[email protected]

Abstract
Conservation properties of iterative methods applied to implicit finite
volume discretizations of nonlinear conservation laws are analyzed. It
is shown that any consistent multistep or Runge-Kutta method is glob-
ally conservative. Further, it is shown that Newton’s method, Krylov
subspace methods and pseudo-time iterations are globally conserva-
tive while the Jacobi and Gauss-Seidel methods are not in general. If
pseudo-time iterations using an explicit Runge-Kutta method are ap-
plied to a locally conservative discretization, then the resulting scheme
is also locally conservative. However, the corresponding numerical flux
can be inconsistent with the conservation law. We prove an extension
of the Lax-Wendroff theorem, which reveals that numerical solutions
based on these methods converge to weak solutions of a modified con-
servation law where the flux function is multiplied by a particular con-
stant. This constant depends on the choice of Runge-Kutta method
but is independent of both the conservation law and the discretiza-
tion. Consistency is maintained by ensuring that this constant equals
unity and a strategy for achieving this is presented. Experiments show
that this strategy improves the convergence rate of the pseudo-time
iterations.

Keywords: Iterative methods, Conservation laws, Conservative numerical

methods, Pseudo-time iterations, Lax-Wendroff theorem

1 Introduction
Conservation laws arise ubiquitously in the modelling of physical phenomena
and their discretizations have been the subject of intense study; see e.g. [17,

1
Chapter 1] and [18, Chapter 1]. They derive their name from the fact
that they describe the conservation of some quantities of interest over time.
In computational fluid dynamics (CFD), these quantities are usually mass,
momentum and energy. For convenience we always refer to the conserved
quantity as the ”mass” in this paper.
Many numerical schemes have been designed to mimic mass conserva-
tion. Throughout, we refer to such schemes as globally conservative. As
special cases, locally conservative schemes dictate that local variations in
the mass propagate from one computational cell to neighboring ones with-
out ”skipping” any cells (typically referred to as ”conservative schemes” in
the literature). In particular, finite volume methods are designed in part
upon this principle, although many other schemes possess the same prop-
erty [13]. A major result on the topic is the Lax-Wendroff theorem, see
e.g. [16], which provides sufficient conditions for numerical solutions of con-
vergent and locally conservative discretizations to converge to weak solutions
of the corresponding conservation law.
For stiff problems, e.g. the simulation of wall bounded viscous flows, [23],
implicit time stepping methods are used. Such discretizations result in sys-
tems of linear or nonlinear equations. In the context of fluid flow simulations,
these systems are sparse and very large. Their solutions are approximated
using iterative methods, see e.g. [1, 2, 4–8, 10–12] and the references therein.
It is natural to ask whether such approximate solutions satisfy the con-
servation properties upon which the discretizations are based, and in partic-
ular, whether a Lax-Wendroff type result is available. It is typically claimed
that this is not the case, e.g. in [15], where a version of a Newton-Jacobi
iteration was noted to violate global conservation. In this paper we set out
to answer this question in greater detail for some well-known (families of)
iterative methods. We consider conservative implicit space-time discretiza-
tions whose solutions in each time step are approximated by a fixed number
of iterations with different iterative methods.
After introducing relevant concepts and definitions in section 2, we in-
vestigate the global conservation of iterative methods in section 3. It is
shown that Newton’s method, Krylov subspace methods and pseudo-time
iterations using Runge-Kutta (RK) methods are globally conservative if the
initial guess has correct mass. However, the Jacobi and Gauss-Seidel meth-
ods in general are not. In section 4 we focus on pseudo-time iterations using
explicit RK methods. We show that such methods are locally conservative.
By fixing the number of iterations and considering the limit of infinitesimal
space-time increments, an extension of the Lax-Wendroff theorem is given.
It turns out that the numerical solution in general converges to a weak solu-
tion of a modified conservation law for which the flux function is multiplied
by a particular constant. An expression for this modification constant is
given that only depends on the choice of Runge-Kutta method and the se-
lected pseudo-time steps. A technique for ensuring that the constant equals

2
one is presented, thereby ensuring consistency of the resulting scheme. Ex-
periments indicate that the new technique improvess the convergence rate
of the pseudo-time iterations. Numerical tests corroborate these findings
and further suggest that the results hold for systems of conservation laws in
multiple dimensions. Conclusions are drawn in section 5.
Throughout the paper we separate quantities that belong to a spatial
discretization from those that do not. Any vector living on a grid with
m cells is denoted by a lower case bold letter, e.g. u ∈ Rm . Similarly,
any matrix operating on such a vector is represented by a bold upper case
letter, e.g. A ∈ Rm×m . In Section 4, we consider s-stage Runge-Kutta
methods. Vector quantities spanning these stages are denoted by bold and
underlined lower case letters, e.g. b ∈ Rs . Matrices operating on the stages
are represented by bold and underlined upper case letters e.g. A ∈ Rs×s .

2 Conservation and conservative discretizations

In this paper we consider numerical methods for conservation laws, meaning
partial differential equations of the form

ut + ∇ · f (u) = 0, t ∈ (t0 , te ], x ∈ Ω ⊂ Rd ,
(1)
u(x, t0 ) = u0 (x),
possibly with additional boundary conditions. Here, the kth component uk
of the vector u ∈ Rq represents the concentration Rof a quantity, i.e. the total
amount of that quantity in a domain is given by Ω uk dx. The conservation
law states that this amount is only changed by flow across the boundary of
the domain. Assuming that the net flow across the boundary is zero, we
thus have for all t ∈ (t0 , te ] that
Z Z
u(x, t)dx = u0 (x)dx. (2)
Ω Ω

In this paper we focus on scalar conservation laws in 1D, i.e. the case
when d = q = 1 in (1). Numerical experiments in section 4 suggest that the
results of the paper can be generalized to systems in multiple dimensions,
however this task is deferred for future work. Further, we restrict our anal-
ysis to the Cauchy problem for (1), or the problem with periodic boundary
conditions. Thus, the conservation law of interest becomes
ut + fx = 0, t ∈ (t0 , te ], x ∈ Ω,
(3)
u(x, t0 ) = u0 (x),

where either Ω = (−∞, ∞) or Ω = (a, b] and u(a) = u(b). Which case is

considered will be clear from context.

3
 A very successful line of research in computational fluid dynamics is to
construct numerical methods that respect (2) on a discrete level. Herein,
we predominantly consider finite volume methods that utilize the implicit
Euler method as temporal discretization. Let the computational grid be
given by (xi , tn ) = (i∆x, n∆t) with n ≥ 0. This grid may be either infinite
or finite and periodic in space, i.e. xi+m = xi for some positive integer m.
We consider discretizations on the form
un+1 − uni 1 ˆ 
i
+ fi+ 1 − fˆi− 1 = 0. (4)
∆t ∆x 2 2

Here, uni approximates the solution u(x, tn ) of (3) in the computational cell
Ωi . The notation fˆi+ 1 (u) is used as a placeholder for a generic numerical
2
flux of the form fˆ(un+1 , . . . , un+1 ), where p, q ∈ N determine the bandwidth
i−p i+q
of the stencil. Multiplying (4) by ∆x and summing over all i reveals that
X X
∆x un+1
i = ∆x uni . (5)
i i

Comparing this with (2) shows that the finite volume scheme discretely
mimics conservation of mass.

2.1 Local conservation

It is the telescoping nature of the spatial terms in (4) that leads to the
preceeding result. In the finite volume community, this property is typically
referred to simply as conservation. However, to distinguish it from other
mass conservative schemes, we will henceforth refer to this property as local
conservation:
Definition 1. A discretization of (1) that can be written in the form (4) is
said to be locally conservative.
Local conservation plays a central role in the Lax-Wendroff theorem,
which considers the Cauchy problem for (3). First, the notion of consistency
must be specified:

Definition 2. A numerical flux fˆ is consistent with f if it is Lipschitz

continuous in each argument and if fˆ(u, . . . , u) = f (u).
A numerical flux function is inconsistent if it fails to satisfy Definition
2. We will see examples of this in Section 4.
The Lax-Wendroff theorem applies to the Cauchy problem for (3) and
considers locally conservative discretizations with consistent numerical flux.
If the numerical solution of such a scheme converges to a function u in the

4
limit of vanishing ∆x and ∆t, the theorem provides sufficient conditions for
u to be a weak solution of the conservation law (1) [17, Chapter 12]. More
precisely, consider a sequence of grids (∆xℓ , ∆tℓ ) such that ∆xℓ , ∆tℓ → 0
as ℓ → ∞. Let Uℓ (x, t) denote the piecewise constant function that takes
the solution value uni in (xi , xi+1 ] × (tn−1 , tn ] on the ℓth grid. We make the
following assumptions:
Assumption 1.
1. There is a function u(x, t) such that over every bounded set Ω = [a, b]×
[0, T ] in x-t space,

kUℓ (x, t) − u(x, t)k1,Ω → 0 as ℓ → ∞.

2. For each T there is a constant R > 0 such that

T V (Uℓ (·, t)) < R for all 0 ≤ t ≤ T, ℓ = 1, 2, . . .

The first of these assumptions asserts that the numerical solution is

convergent in the L1 -norm with limit u. The second one states that the
total variation of the numerical solution remains bounded independently of
the grid. The Lax-Wendroff theorem may now be stated as follows:
Theorem 1 (Lax-Wendroff). Consider the locally conservative discretiza-
tion (4), suppose that the numerical flux fˆ is consistent and that Assumption
1 is satisfied. Then, u(x, t) is a weak solution of (1).
The theorem and its assumptions require that the linear or nonlinear
systems arising in (4) are solved exactly. However, in practice these systems
will be solved approximately using iterative methods. The natural question
that now arises is: Do iterative solvers maintain the convergence to weak
solutions?

2.2 Global conservation

To answer this question we must establish if iterative methods preserve
local conservation. However, local conservation is a special case of the more
general notion of global conservation. Let us therefore consider a finite and
periodic grid. Discretizing (3) in space on a mesh with cells Ωi of volume
|Ωi |, we arrive at an initial value problem

ut = f̂ (u), t ∈ (t0 , te ],
(6)
u(t0 ) = u0 ,

with u(t), f̂ ∈ Rm .

5
Definition 3. A discretization f̂ = (fˆ1 (u), . . . , fˆm (u))⊤ is globally conser-
vative if
Xm
|Ωi |fˆi (u) = 0.
i=1

Global conservation implies that (2) is fulfilled on a semidiscrete level

by (6):
Xm m
X
|Ωi |ui (t) = |Ωi |ui (t0 ), ∀t ∈ [t0 , te ].
i=1 i=1

Suppose next that a time integration method is used to solve (6).

Definition 4. A time integration method applied to the conservative semidis-
cretization (6) is globally conservative if
m
X m
X
|Ωi |un+1
i = |Ωi |uni
i=1 i=1

holds for each time step n = 0, 1, . . .

Note that the conservation property (5) derived for the locally conserva-
tive finite volume discretization (4) is a special case of Definition 4, where
Ωi = ∆x for every i. Global conservation is thus necessary for local conser-
vation, which in turn is necessary for the Lax-Wendroff Theorem.
In the CFD community, one hears from time to time the claim that im-
plicit methods are not conservative. However, this is not true: All consistent
linear multistep methods are globally conservative. To see this, consider a
generic s-step linear multistep method
s
X s
X
n+j
aj u = ∆t bj f̂ (un+j ), (7)
j=0 j=0

where aj and bj are given method dependent coefficients. Suppose that

Pm n+j P
i=1 |Ωi |ui = m 0
i=1 |Ωi |ui for j = 0, . . . , s − 1. Multiplying (7) by |Ωi |
and summing over all cells gives
m
X s−1
X m
X s
X m
X
|Ωi |as uin+s =− aj |Ωi |uin+j + ∆t bj |Ωi |fˆi (un+j )
i=1 j=0 i=1 j=0
|i=1 {z }
=0
m
X s−1
X
=− |Ωi |u0i aj .
i=1 j=0

6
Consistent multistep methods satisfy as = 1 and a0 + · · · + as−1 = −1.
Hence, global conservation is ensured.
Similarly, all Runge-Kutta (RK) methods are globally conservative. Con-
sider an s-stage RK method
s
X
n+1 n
u = u + ∆t bj k j ,
j=1
s
! (8)
X
n
kj = f̂ u + ∆t aj,l kl , j = 1, . . . , s,
l=1

where aj,l and bj are given by the method. Since f̂ is globally conservative
it follows from Definition 3 that the mass of kj is zero and therefore that
m
X m
X s
X m
X m
X
|Ωi |un+1
i = |Ωi |uni + ∆t bj |Ωi |kji = |Ωi |uni .
i=1 i=1 j=1 i=1 i=1

Hence, RK methods are globally conservative.

Exponential integrators have been proven to be conservative as well [9],
essentially due to the following useful lemma:
′ ∂ f̂
Lemma 2. For any y ∈ Rm the Jacobian f̂ (u) = ∂u ∈ Rm×m of the
globally conservative discretization f̂ (u) satisfies
m
X  ′ 
|Ωi | f̂ y = 0. (9)
i
i=1

Proof. See [27].

In the remainder we also need the following result:
′
Lemma 3. Let f̂ (u) ∈ Rm×m be the Jacobian of a conservative discretiza-
tion f̂ (u) and let α be any scalar. The solution x of a linear system of the
form  
′
I + αf̂ x = b, (10)

satisfies
m
X m
X
|Ωi |xi = |Ωi |bi .
i=1 i=1

7
Proof. Multiplying the ith element of (10) by |Ωi | and summing over all i
gives
m
X m
X    m
X m
X  ′ 
′
|Ωi |bi = |Ωi | I + αf̂ x = |Ωi |xi + α |Ωi | f̂ x .
i i
i=1 i=1 i=1 i=1

By Lemma 2 the final term on the right-hand side is zero.

3 Globally conservative iterative solvers

We begin by asking: Do iterative solvers respect global conservation? Since
global conservation is necessary for local conservation, which in turn is used
in Theorem 1, a negative answer will immediately exclude the possibility
to extend the Lax-Wendroff theorem to incorporate a particular iterative
method.
Discretizing (6) in time using the implicit Euler method results in a
nonlinear equation system of the form

u − αf̂ (u) = un . (11)

We concern ourselves with a set of iterates u(k) with limit un+1 as k → ∞.

Definition 5. An iterative method with iterates u(k) is said to be globally
conservative if
m
X m
X
(k+1)
|Ωi |ui = |Ωi |uni , k = 0, 1, . . .
i=1 i=1

In the following subsections we investigate the conservation properties

of some of the most common iterative methods.

3.1 Newton’s Method

Newton’s method finds an approximate solution to (11) by solving the se-
quence of linear systems
 ′
   
I − αf̂ u(k) ∆u = un − u(k) + αf̂ u(k) ,
(12)
u(k+1) = u(k) + ∆u,

for k = 0, 1, . . . Since f̂ is conservative it has zero mass by Definition 3.

If u(k) and un have the same mass, the entire right-hand side of the first
equation in (12) has zero mass. Hence, by Lemma 3, the increment ∆u

8
also has zero mass. Thus, the mass of u(k+1) is the same as that of u(k) ,
which by assumption is the same as that of un . Hence, Newton’s method is
globally conservative.
In large scale practical applications, each linear system in (12) is solved
approximately using other iterative methods. In the following, we consider
a generic linear system
(I − αA)u = b, (13)
that may either represent a Newton iteration or a linear discretization of a
conservation law. In line with Lemma 2, it is assumed that the mass of Ay
is zero for any vector y ∈ Rm .

3.2 Stationary linear methods

Any iterative method for solving (13) that can be written in the form

u(k+1) = M u(k) + N −1 un (14)

for some matrices M and N , is termed a stationary linear method. Well

known examples include the Richardson, Jacobi and Gauss-Seidel iterations.

3.2.1 Richardson iteration

For the Richardson iteration, M = I − θ(I − αA) and N −1 = θI, where
θ ∈ R is a fixed parameter. With some minor rearrangements, the iteration
can thus be expressed as

u(k+1) = u(k) + θ(b − (I − αA)u(k) ).

Multiplying the ith element by |Ωi | and summing over i results in

m
X m
X m
X m
X
(k+1) (k) (k)
|Ωi |ui = |Ωi |ui +θ |Ωi |(bi − ui ) + αθ |Ωi |(Au(k) )i .
i=1 i=1 i=1 i=1

By assumption, the matrix-vector product Au(k) has zero mass, hence the
final term vanishes. Further, if u(k) has the same mass as b, then the
second term on the right-hand side vanishes. By induction it follows that
the Richardson iteration is globally conservative.

3.2.2 Jacobi iteration

Consider again the linear system (13). With the matrix decomposition I −
αA = I − α(D + L + U ), where D, L and U are the diagonal, lower and
upper triangular parts of A respectively, the Jacobi iteration is obtained
by setting M = α(I − αD)−1 (L + U ) and N −1 = (I − αD)−1 . Inserting

9
this into (14), multiplying by I − αD, then adding and subtracting αDu(k) ,
gives after some rearrangement

u(k+1) = b + αD(u(k+1) − u(k) ) + αAu(k) .

Multiplying the ith element by |Ωi | and summing over i results in

m
X m
X m
X m
X
(k+1) (k+1) (k)
|Ωi |ui = |Ωi |bi + α |Ωi |ai,i (ui − ui ) + α |Ωi |(Au(k) )i .
i=1 i=1 i=1 i=1
(15)
The final term vanishes by the assumption. However, the second term on
the right-hand side is in general non-zero and thus introduces a conservation
error
m
X (k+1) (k)
α |Ωi |ai,i (ui − ui ).
i=1
The Jacobi method is therefore not globally conservative in general.
However, consider the special case when the diagonal D of A is a scalar
multiple of the identity matrix, say D = aI. Then ai,i = a for i = 1, . . . , m.
If u(k) has the same mass as b we find, after rearrangement of (15), that
m
X m
X
(k+1)
(1 − αa) |Ωi |ui = (1 − αa) |Ωi |bi .
i=1 i=1

Thus, u(k+1) has correct mass, at least as long as αa 6= 1. Hence, the Jacobi
iteration is conservative in this special case.

3.2.3 Gauss-Seidel iteration

The Gauss-Seidel iteration is defined by M = α(I − α(D + L))−1 U and
N −1 = (I −α(D +L))−1 . Inserting this into (14), multiplying by I −α(D +
L), then adding and subtracting αU u(k+1) gives

u(k+1) = b − αU (u(k+1) − u(k) ) + αAu(k+1) .

Following the same procedure as previously, we multiply the ith element by
|Ωi | and sum over i to obtain
m
X m
X m
X m
X
(k+1) (k+1) (k)
|Ωi |ui = |Ωi |bi + α |Ωi | ai,j (uj − uj )
i=1 i=1 i=1 j=i+1
m
X
+α |Ωi |(Au(k+1) )i .
i=1

10
By assumption, the final term vanishes. However, the second term on the
right-hand side is in general non-zero. Thus, the Gauss-Seidel method is not
globally conservative and the conservation error is given by
m
X m
X (k+1) (k)
α |Ωi | ai,j (uj − uj ).
i=1 j=i+1

However, note that for problems where A is lower triangular the Gauss-
Seidel iteration is indeed conservative. In fact, for such problems the method
is by definition exact.

3.3 Krylov subspace methods

Given an initial guess u(0) for the solution of the linear system (13), the kth
iteration of a Krylov subspace method belongs to the space
n o
Kk = u(0) + span r (0) , (I − αA)r (0) , . . . , (I − αA)k−1 r (0)
n o
= u(0) + span r (0) , Ar (0) , . . . , Ak−1 r (0) ,

where r(0) = b − (I − αA)u(0) is the initial residual. Examples of Krylov

subspace methods are GMRES, CG and BICGSTAB [24, Chapters 6–7].
We may express the kth iteration u(k) as
k−1
X
u(k) = u(0) + cl Al r (0) ,
l=0

for some coefficients cl , l = 0, . . . , k − 1 that depend on, and are chosen by

the method. It follows that
m
X m
X m
X k−1
X m
X
(k) (0) (0)
|Ωi |ui = |Ωi |ui + c0 |Ωi |ri + cl |Ωi |(Al r (0) )i .
i=1 i=1 i=1 l=1 i=1

The final term vanishes by assumption. Note that the mass of r (0) is zero if
the mass of u(0) equals that of b. Thus, the second term on the right-hand
side also vanishes. All Krylov subspace methods are therefore conservative
if the initial guess is chosen to have the same mass as the right-hand side.
Typically a preconditioner P −1 will pre- or post-multiply A in Krylov
subspace methods. We will not delve into this subject here but merely
observe that if y and P −1 y have the same mass for any choice of y, then
the above analysis applies without modification.

11
3.4 Multigrid methods
Multigrid methods combine two methods, namely a smoother and a coarse
grid correction (CGC). The idea is to separate the residual into high and
low frequency components. As a smoother, an iterative method is applied,
designed to effectively damp the high frequency components.
The role of the CGC is to remove the low frequency components of
the residual. The residual is mapped to a coarser grid using a restriction
operator, R. Smoothing is then applied and a prolongation operator P
is used to reconstruct the residual on the fine grid. Finally, the fine grid
solution is updated by adding the correction to the previous iterate. The
procedure can be applied to a hirearchy of grids, thereby resulting in a
multigrid method. If we consider only two grid levels and assume that the
system is solved exactly on the coarser one, then the coarse grid correction
for the linear problem (13) has the form

u(k+1) = (I − P (I − αA)−1
ℓ−1 R(I − αA)ℓ )u
(k)
+ P (I − αA)−1
ℓ−1 Rb. (16)

Here, the notation (·)ℓ and (·)ℓ−1 denote matrices operating on the fine and
the coarse grid respectively.
The easiest way to make a multigrid method globally conservative is
to choose its components to be globally conservative. The smoother can
be any of the globally conservative iterations discussed so far. By Lemma
3, the inverted matrices in (16) are mass conserving. Thus, it remains to
choose globally conservative restriction and prolongation operators R and
P . This is achieved using agglomeration. The restriction is performed
by agglomerating a number of neighboring cells by constructing a volume
weighted average of fine grid values. Conversely, the prolongation is done
by injecting coarse grid values at multiple fine grid points. In fact, this is
the standard choice of multigrid method in CFD since it results in faster
convergence than other alternatives. For details, see [3].

3.5 Pseudo-time iterations

Pseudo-time iterations are obtained by adding a pseudo-time derivative term
to the algebraic equations (11) (or (13)),
∂u
+ u − αf̂ (u) = un , u(0) = u0 . (17)
∂τ
Here, u0 is initial data that must be provided, e.g. u0 = un . The idea is
that the system eventually should reach a steady state as τ → ∞ where the
pseudo-time derivative vanishes [14], resuting in a solution to the original
nonlinear system (11). Any time integration method can in principle be
applied to the initial value problem (17). Explicit RK methods are glob-
ally conservative by construction. Implicit methods result in new systems

12
of equations whose solutions are once again approximated using iterative
methods [5]. Whether the resulting approximation is conservative depends
on the choice of method. Any one of the conservative methods discussed so
far can be applied in principle. We will return to pseudo-time iterations in
Section 4, where more details are provided.

3.6 Numerical validation

To validate the results of the preceding section we consider two simple ex-
periments. The first is the linear advection equation with periodic boundary
conditions,
ut + ux = 0, x ∈ (−1.5, 1.5], t ∈ (0, 6],
u(x, 0) = exp (−50x2 ), (18)
u(−1.5, t) = u(1.5, t).
We discretize using the finite volume scheme (4) with a central numerical
flux fˆi+ 1 = (un+1
i+1 − ui
n+1
)/2. This results in a linear system of the form
2

(13) with α = −∆t/∆x and A = Tridiag − 12 , 0, 21 . Here, Tridiag(·) refers
to a tridiagonal matrix with periodic wrap-around. Note that the diagonal
of A is constant. We therefore expect the Jacobi method to be conservative
for this particular discretization.
The second experiment is Burgers’ equation with periodic boundary con-
ditions,  2
u
ut + = 0, x ∈ (−1.5, 1.5], t ∈ (0, 0.6],
2 x
(19)
u(x, 0) = exp (−x2 ),
u(−1.5, t) = u(1.5, t).
Again, we use the finite volume scheme (4), however with the upwind flux

2
fˆi+ 1 = un+1
i /2. The result is a nonlinear system in the form (11) with
2

ˆ
α = −∆t and fi (un+1 ) = (un+1 )2 − (un+1 2
i i−1 ) /2∆x. The Jacobian of this
discretization is lower triangular except for a single element in the top right
corner. Thus, we expect the Gauss-Seidel method to be nearly conservative
if the grid is sufficiently fine.
Burgers’ equation (19) is solved using Newton’s method with initial guess
un . The resulting linear systems are either solved exactly or using the
following iterative methods with 0 as initial guess:
(R) The Richardson iteration using θ = 0.5.
(J) The Jacobi method.
(GS) The Gauss-Seidel method.

13
(GM) GMRES without restarts or preconditioning.
(CGC) A two-level coarse grid correction.
(H) Pseudo-time iterations using Heun’s method with ∆τ = 0.5.
As prolongation and restriction operators for the CGC, agglomeration gives
 
1 1
1 1 1  ⊤
R=   , P = 2R .
2 ..
.

The same methods are used directly on the advection problem (18) with un
as initial guess.
First we consider very coarse discretizations with ∆x = ∆t = 0.5 for both
problems. A single iteration is used with each method and the mass error
and residual is computed after one time step. The results are shown in Table
1. Here, any error smaller than 10−15 is denoted as zero. The Richardson
iteration, GMRES, CGC and the pseudo-time iterations are conservative for
both problems. Similarly, the exact Newton method (Exact) is conservative.
As expected, the Jacobi method is conservative for the advection problem
but not for Burgers’ equation. Gauss-Seidel is not conservative for any of
the two problems.

Table 1: Mass errors and residuals after one time step with ∆t = ∆x = 0.5
for the advection and Burgers’ equations using a single iteration. Computed
values smaller than 10−15 are denoted by 0.

Advection equation (Exact) (R) (J) (GS) (GM) (CGC) (H)

Mass error 0 0 −0.094 0 0 0
Residual 0.331 0.433 0.256 0.327 0.162 0.287

Burgers’ equation (Exact) (R) (J) (GS) (GM) (CGC) (H)

Mass error 0 0 0.031 −0.034 0 0 0
Residual 0.897 0.895 0.901 0.862 0.898 0.924 0.878

Next, we repeat the experiment with well resolved discretizations. For

the advection problem we choose ∆t = ∆x = 0.006 and for Burgers’ equation
∆t = ∆x = 0.003. In the latter case, two Newton iterations are used per
time step. For each of the other methods, 5 iterations are used except for
CGC where a single iteration is performed.
The total mass is computed in each time step and compared with the
mass of the initial data. Fig. 1 shows the results. The main difference to

14
 10 -6
0.01 20

0 15

-0.01 10

5
-0.02

0
-0.03
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6

(a) Linear advection. (b) Burgers’ equation.

Figure 1: Mass error for iterative methods applied to the advection equation
(18) and Burgers’ equation (19).

the previous test is that Gauss-Seidel has a non-detectible mass error for
Burger’s equation. This is in line with our expectations since the Jacobian
only has a single nonzero element above the main diagonal.

4 Local conservation of pseudo-time iterations

Having covered global conservation in the previous section, we now switch
focus to local conservation. The Jacobi and Gauss-Seidel methods cannot
be locally conservative since they are not globally conservative in general.
For the other methods considered so far, the question remains open.
In what follows, we restrict our attention to the Cauchy problem for (3).
We focus on pseudo-time iterations, show that these methods are locally
conservative and prove an extension of the Lax-Wendroff theorem. Noting
that the Richardson iteration is equivalent to pseudo-time iterations with
explicit Euler, this method is also covered.

4.1 Pseudo-time iterations in locally conservative form

We once again consider the finite volume method (4) where the implicit
Euler method is used as time discretization. In order to apply pseudo-time
iterations to this scheme, we introduce a pseudo-time derivative,
∂ui
+ gi (u) = 0, ui (0) = u0i , i = . . . , −1, 0, 1, . . .
∂τ
where the nonlinear function gi is given by
ui − uni 1 ˆ 
gi (u) = + fi+ 1 (u) − fˆi− 1 (u) . (20)
∆t ∆x 2 2

15
Several different methods are available for iterating in pseudo-time [5, 28].
Herein, we use an explicit s-stage Runge-Kutta method (ERK). Let (A, b, c)
denote the coefficient matrix and vectors of the ERK method. We denote the
(k) (k+1)
kth pseudo-time iterate by ui . The subsequent iterate ui is computed
(k)
from ui as
Xs  
(k+1) (k) (k)
ui = ui − ∆τk bj gi U j , (21)
j=1
(k)
where the stage vectors U j , j = 1, . . . , s have elements

j−1
X  
(k) (k)
Ujι = u(k)
ι − ∆τk aj,l gι U l , ι = i − p, . . . , i + q. (22)
l=1

As previously, p and q determine the bandwidth of the finite volume stencil.

In the remainder we always use a fixed number N of iterations. A step
(N ) (0)
in physical time is taken by setting un+1
i = ui . Throughout, ui = uni is
chosen as initial iterate.
Recall that the stability function φ(z) of an RK method (A, b, c) is given
by (see e.g. [31, Chapter IV.3])

φ(z) = 1 + zb⊤ (I − zA)−1 1, (23)

where I is the s × s identity matrix. The stability region of the RK method

is defined as the subset of the complex plane for which |φ(z)| < 1. The
proof of the following lemma is rather lengthy and is therefore deferred to
Appendix A.
Lemma 4. For each pseudo-time iteration k = 0, . . . , N − 1, let µk =
(0)
∆τk /∆t and set ui = uni . Then, for any N ≥ 1, the elements of the
(N )
pseudo-time iterate ui satisfy the relation
(N )  
ui − uni 1 (N ) (N )
+ ĥi+ 1 − ĥi− 1 = 0, i = . . . , −1, 0, 1, . . . (24)
∆t ∆x 2 2

(N )
where the flux ĥi+ 1 is given by
2

N −1 N −1
!
(N )
X Y (k)
ĥi+ 1 = µk b⊤ (I + µk A)−1 φ(−µl ) f̂ i+ 1 (25)
2 2
k=0 l=k+1

(k)
    ⊤
(k) (k)
and f̂ i+ 1 = fˆi+ 1 U 1 , . . . , fˆi+ 1 U s .
2 2 2

16
Remark 1. The product in (25) is empty when k = N − 1. To handle this
case we use the convention
N
Y −1
φ(−µl ) = 1.
l=N

It follows immediately from Lemma 4 that the locally conservative nature

of the discretization (4) is preserved by the pseudo-time iterations, albeit
(N )
with the new numerical flux ĥi+ 1 in place of fˆi+ 1 . The question with which
2 2
(N )
flux ĥi+ 1 is consistent, is answered in the following Theorem:
2

Theorem 5. Let N be given and let µk = ∆τk /∆t for k = 0, . . . , N − 1.

(0)
Set ui = uni and terminate the pseudo-time iteration after N steps. Then
the resulting scheme can be written in the locally conservative form
 
un+1
i − uni 1 (N ) (N )
+ ĥi+ 1 − ĥi− 1 = 0, i = . . . , −1, 0, 1, . . . (26)
∆t ∆x 2 2

(N )
where ĥi+ 1 , given by (25), is a numerical flux consistent with the flux
2
c(µ0 , . . . , µN −1 )f , with
N
Y −1
c(µ0 , . . . , µN −1 ) = 1 − φ(−µl ). (27)
l=0

Proof. Firstly, the locally conservative form (26) follows from setting un+1 i =
(N )
ui in (24). Secondly we recall that the numerical flux fˆi+ 1 depends on
2
p + q + 1 parameters, e.g. fˆ 1 (w) = fˆ(wi−p , . . . , wi+q ). Inspecting the flux
i+ 2
(N )
ĥi+ 1 we see that it is additionally dependent on uni so that we may write
2
(N )
ĥi+ 1 (w) = ĥN (wi−p , . . . , wi+q ; uni ). To establish the consistency we must
2
therefore show that ĥN (u, . . . , u; u) = cf (u). To do this, we first note from
(20) that gi (u, . . . , u; u) = 0 due to the consistency of fˆi+ 1 . Using the fact
2
that A is lower triangular for any ERK method, it follows from (21) and (22)
(k) (k)
that Ujι = ui = u for every j, k and ι = i − p, . . . , i + q. Consequently,
(k)
the vector f̂ i+ 1 becomes
2

k
 
f̂ (u, . . . , u; u) = fˆ(u, . . . , u), . . . , fˆ(u, . . . , u) = fˆ(u, . . . , u)1 = f (u)1,

17
where the consistency of fˆ has been used in the final equality. Inserting this
(N )
into the flux function ĥi+ 1 in (25) and using (23) gives
2

"N −1 N −1
#
(N )
X Y
⊤ −1
ĥi+ 1 = µk b (I + µk A) 1 φ(−µl ) f (u)
2
k=0 l=k+1
"N −1 N −1
#
X Y
= [1 − φ(−µk )] φ(−µl ) f (u)
k=0 l=k+1
N
" −1
−1 N N −1
#
X Y Y
= φ(−µl ) − φ(−µl ) f (u)
k=0 l=k+1 l=k
" N −1
#
Y
= 1− φ(−µl ) f (u)
l=0
= c(µ0 , . . . , µN −1 )f (u),

where we have utilized the fact that the sum in the third equality is telescop-
(N )
ing. Finally, note that ĥi+ 1 is formed by a linear combination of evaluations
2
of fˆi+ 1 and is therefore Lipschitz continuous.
2

(N )
The fact that ĥi+ 1 is inconsistent with f when c 6= 1 is a manifestation
2
of the error associated with the pseudo-time iterations. This has important
implications on the convergence of the resulting scheme. What follows is an
extension of the Lax-Wendroff theorem (see [17, Chapter 12]) pertaining to
ERK pseudo-time iterations:
Theorem 6. Consider a sequence of grids (∆xℓ , ∆tℓ ) such that ∆xℓ , ∆tℓ →
(0)
0 as ℓ → ∞. Fix N independently of ℓ, set ui = uni and terminate the
pseudo-time iterations after N steps. Let ∆τk,ℓ/∆tℓ = µk,ℓ = µk be constants
independent of ℓ for each k = 0, . . . , N −1. Suppose that the numerical flux fˆ
in (4) is consistent with f and that Assumption 1 is satisfied. Then, u(x, t)
is a weak solution of the conservation law
N
Y −1
ut + c(µ0 , . . . , µN −1 )fx = 0, c(µ0 , . . . , µN −1 ) = 1 − φ(−µl ). (28)
l=0

Proof. We follow the proof of the Lax-Wendroff theorem given in [17, Chap-
ter 12] with changes and details added where necessary. Throughout the
proof we let Uℓ (x, t) denote the piecewise constant function that takes the
solution value uni in (xi , xi+1 ] × (tn−1 , tn ] on the ℓth grid. Similarly, for

18
 (k)
j = 1, . . . , s we let Ujℓ (x, t) be the piecewise constant function that takes
(k)
the value Uji in (xi , xi+1 ] × (tn−1 , tn ].
The discretization (26) can equivalently be expressed as
 
(N ) (N )
∆xℓ [Uℓ (xi , tn+1 ) − Uℓ (xi , tn )] + ∆tℓ ĥi+ 1 − ĥi− 1 = 0. (29)
2 2

Let ϕ ∈ C01 be a compactly supported test function. Multiply (29) by

ϕ(xi , tn ) and sum over all n and i. Due to the compact support of ϕ, these
sums can be extended arbitrarily beyond the bounds of x and t, hence we
obtain
∞ X
X ∞
∆xℓ ϕ(xi , tn ) (Uℓ (xi , tn+1 ) − Uℓ (xi , tn ))
n=0 i=−∞
∞ X
∞   (30)
X (N ) (N )
+ ∆tℓ ϕ(xi , tn ) ĥi+ 1 − ĥi− 1 = 0.
2 2
n=0 i=−∞

At this point we will use summation by parts on the sums in (30). Recall
that for two sequences (ai ) and (bi ), the summation by parts formula can
be expressed as
M
X M
X
aj (bj − bj−1 ) = aM bM − a1 b0 − (aj+1 − aj )bj .
j=1 j=1

Applied to the n-sum in the first term and to the i-sum in the second term
in (30), this results in
"∞ ∞  
X X ϕ(xi , tn ) − ϕ(xi , tn−1 )
∆xℓ ∆tℓ Uℓ (xi , tn )
∆tℓ
n=1 i=−∞
∞ X ∞   #
X ϕ(xi+1 , tn ) − ϕ(xi , tn ) (N )
+ ĥi+ 1 (31)
∆xℓ 2
n=0 i=−∞
∞
X
= −∆xℓ ϕ(xi , 0)Uℓ (xi , 0).
i=−∞

Here we have used the fact that ϕ has compact support in order to eliminate
all boundary terms except the one at n = 0.
We now let ℓ → ∞ and investigate the convergence of the terms arising
in (31). The first and third terms are identical to those in the proof of

19
the Lax-Wendroff theorem [17, Chapter 12]. Thus, it can immediately be
concluded that the first term converges to
Z ∞Z ∞
ϕt u(x, t)dxdt
0 −∞

and the third one to Z ∞

− ϕ(x, 0)u(x, 0)dx.
−∞

It remains to investigate the second term in (31).

(N )
Expanding ĥi+ 1 using (25) yields for the second term in (31)
2

∞ X ∞  
X ϕ(xi+1 , tn ) − ϕ(xi , tn )
∆xℓ ∆tℓ
n=0 i=−∞
∆xℓ
N −1 N −1
!
X Y (k)
⊤ −1
µk b (I + µk A) φ(−µl ) f̂ i+ 1 .
2
k=0 l=k+1

To this expression, add and subtract

∞ X ∞  
X ϕ(xi+1 , tn ) − ϕ(xi , tn )
∆xℓ ∆tℓ
n=0 i=−∞
∆xℓ
N −1 N −1
!
X Y
µk b⊤ (I + µk A)−1 φ(−µl ) f (Uℓ (xi , tn ))1
k=0 l=k+1

in order to obtain
∞ X ∞  
X ϕ(xi+1 , tn ) − ϕ(xi , tn )
∆xℓ ∆tℓ
∆xℓ
n=0 i=−∞
"N −1 N −1
!#
X Y
µk b⊤ (I + µk A)−1 1 φ(−µl ) f (Uℓ (xi , tn ))
k=0 l=k+1
∞ ∞  
X X ϕ(xi+1 , tn ) − ϕ(xi , tn )
+∆xℓ ∆tℓ
∆xℓ
n=0 i=−∞
N −1 N −1
!
X Y  (k) 
⊤ −1
µk b (I + µk A) φ(−µl ) f̂ i+ 1 − f (Uℓ (xi , tn ))1 .
2
k=0 l=k+1

20
The bracketed part of this expression evaluates to c(µ0 , . . . , µN −1 ), as in the
proof of Theorem 5. Thus, the first half of this expression equals
∞ X
∞  
X ϕ(xi+1 , tn ) − ϕ(xi , tn )
c(µ0 , . . . , µN −1 )∆xℓ ∆tℓ f (Uℓ (xi , tn )).
∆xℓ
n=0 i=−∞

Apart from the factor c(µ0 , . . . , µN −1 ), which is independent of ℓ, this term

appears identically in the proof of the Lax-Wendroff theorem [17, Chapter
12] and converges to
Z ∞Z ∞
c(µ0 , . . . , µN −1 ) ϕx f dxdt.
0 −∞

It remains to show that

∞ X ∞  
X ϕ(xi+1 , tn ) − ϕ(xi , tn )
∆xℓ ∆tℓ
∆xℓ
n=0 i=−∞
N −1 N −1
! (32)
X Y  (k) 
µk b⊤ (I + µk A)−1 φ(−µl ) f̂ i+ 1 − f (Uℓ (xi , tn ))1
2
k=0 l=k+1

(k)
vanishes in the limit ℓ → ∞. Since Uℓ and Ujℓ are both constant in
(xi , xi+1 ] × (tn−1 , tn ], we can rewrite (32) as
∞ X
X ∞ Z xi+1 Z tn
ϕx (x, tn )
n=0 i=−∞ xi tn−1
N −1 N −1
!
X Y  (k)

⊤ −1
µk b (I + µk A) φ(−µl ) f̂ i+ 1 (x, t) − f (Uℓ (x, t))1 .
2
k=0 l=k+1

(k)
Here, f̂ i+ 1 (x, t) is placeholder notation for the vector whose jth element is
2

 
(k) (k)
fˆi+ 1 Ujℓ (x − p∆x, t), . . . , Ujℓ (x + q∆x, t) .
2

To establish that this term indeed vanishes, it suffices to show that

N −1 N −1
!
X Y  (k) 
µk b⊤ (I + µk A)−1 φ(−µl ) f̂ i+ 1 (x, t) − f (Uℓ (x, t))1 (33)
2
k=0 l=k+1

21
tends to zero for almost every x. By the Cauchy-Schwarz and triangle
inequalities, (33) is bounded by
N −1 N −1
!
X Y (k)
⊤ −1
kµk b (I + µk A) k |φ(−µl )| f̂ i+ 1 (x, t) − f (Uℓ (x, t))1 ,
2
k=0 l=k+1
(34)
where the Euclidean norm in Rs is used.  The only term in (34) that depends
(k)
on ℓ is f̂ i+ 1 (x, t) − f (Uℓ (x, t))1 and it therefore suffices to show that
2
this term vanishes for almost every x.
Recall that f (Uℓ ) = fˆi+ 1 (Uℓ , . . . , Uℓ ) by consistency. A standard norm
2
inequality gives
(k)
f̂ i+ 1 (x, t) − f (Uℓ (x, t))1
2
√ 
(k) (k)

ˆ
≤ s max fi+ 1 Ujℓ (x − p∆x, t), . . . , Ujℓ (x + q∆x, t)
1≤j≤s 2

−fˆi+ 1 (Uℓ (x, t), . . . , Uℓ (x, t))

2
√ n 
(k) (k)

≤ s max fˆi+ 1 Ujℓ (x − p∆x, t), . . . , Ujℓ (x + q∆x, t)
1≤j≤s 2

− fˆi+ 1 (Uℓ (x − p∆x, t), . . . , Uℓ (x + q∆x, t))

2

+ fˆi+ 1 (Uℓ (x − p∆x, t), . . . , Uℓ (x + q∆x, t))

2
o
ˆ
− fi+ 1 (Uℓ (x, t), . . . , Uℓ (x, t)) .
2

Since fˆi+ 1 is Lipschitz continuous in each argument there is a constant L

2
such that the final expression is bounded by
√ (k)
L s max max Ujℓ (x + r∆x, t) − Uℓ (x + r∆x, t)
1≤j≤s p≤r≤q
√
+L s max |Uℓ (x + r∆x, t) − Uℓ (x, t)| .
p≤r≤q

The second of these terms appear in the proof of the Lax-Wendroff theorem
[17, Chapter 12] and vanishes in the limit for almost every x due to the
bounded total variation of Uℓ . Further, from (22) and (21) and the fact
(0) (k)
that ui = uni it follows that for every j = 1, . . . , s and k ≥ 0, Ujι = unι
for each ι = i − p, . . . , i + q in the limit of vanishing ∆τℓ . Consequently,
(k)
Ujℓ (·, t) − Uℓ (·, t) vanishes identically as ℓ → ∞.

22
 A few remarks about Theorem 6 are in place: First, we demand from a
useful iterative method that it converges to the correct solution as N → ∞.
Thus, the pseudo-time steps should be chosen in a way that ensures that
c → 1, or equivalently,
N
Y −1
φ(−µl ) → 0 as N → ∞. (35)
l=0

The simplest way to do this is to choose each pseudo-time step so that

|φ(−µl )| < 1, i.e. to stay within the stability region of the RK method.
Secondly, observe that if φ(−µl ) = 0 for any l, then c = 1 irrespective
of how many further iterations that are carried out. For some RK methods
such a root exists while for others it does not. For instance, the explicit Euler
method has stability function φ(−µ) = 1 − µ, hence µ = 1 is a root of φ. On
the other hand, Heun’s method has stability function φ(−µ) = 1−µ+µ2 /2 >
0 for all µ ∈ R and therefore does not have any real roots. A strategy is thus
to choose a RK method with a root, begin the pseudo-time iterations with
a step that corresponds to this root, then resort to a conventional method
for choosing the remaining pseudo-time steps. The initial iteration will not
change the limit as N → ∞ if the remaining pseudo-time steps are selected
within the stability region.

4.2 Numerical results

Next we validate Theorem 6 by numerically solving a series of linear and
nonlinear conservation laws.

4.2.1 Linear advection

The first setting is the linear advection equation (18). The computational
domain is x ∈ (−1, 1], t ∈ (0, 0.25] and periodic boundary conditions are
used. The upwind flux fˆi+ 1 = un+1i is used for the spatial discretization.
2
The resulting finite volume method becomes

un+1 − uni 1

i
+ un+1
i − un+1
i−1 = 0, i = 1, . . . , m.
∆t ∆x
Throughout the experiments the temporal and spatial increments are chosen
to be equal: ∆t = ∆x.
We validate Theorem 6 by studying the convergence of the numerical
scheme to the solution of the original advection problem (18) as well as to
the solution of the modified version ut + cux = 0. As pseudo-time iteration
we use the explicit Euler method, Heun’s method and the third order strong

23
stability preserving RK method SSPRK3 [26]. Theorem 6 predicts that
these methods respectively will modify the propagation speed by the factor
N
Y −1
c(µ0 , . . . , µN −1 ) = 1 − (1 − µl ),
l=0
N −1  
Y µ2
c(µ0 , . . . , µN −1 ) = 1 − 1 − µl + l ,
2
l=0
N −1  
Y µ2 µ3
c(µ0 , . . . , µN −1 ) = 1 − 1 − µl + l − l .
2 6
l=0

Here, we fix N = 4 and consider two different sequences of pseudo-time

steps, one with constant and one with variable step sizes. The first is given
by µl = 1/20 and the second by µl = 2−l for l = 0, . . . , 3. The corresponding
modification constants c(µ0 , . . . , µ3 ) are shown in Table 2. Note that with
the second sequence, c = 1 for the explicit Euler method. This is due to the
fact that µ0 is a root of the stability polynomial in this case.

Table 2: Modification constant c(µ0 , . . . , µ3 ) for different ERK methods and

choices of pseudo-time steps

Euler Heun SSPRK3

1
µl = 20 0.1855 0.1812 0.1813
µl = 2−l 1 0.7845 0.8616

The advection problem is solved on a sequence of grids with grid spacing

∆x = 2/(40 × 2j ) for j = 1, . . . , 12. The L2 error is calculated with respect
to the exact solution of the original and modified equations. The results
are shown in Fig. 2. When constant pseudo-time steps are used (Fig. 2a),
the lines overlap. None of the methods converge to the solution of the
original problem since the numerical solution is moving with the incorrect
speed. Instead, all three schemes converge to the solution of their respective
modified conservation law. Similar results are seen when variable pseudo-
time steps are used (Fig. 2b).
Solutions of the advection equation (18) propagate from left to right
at unit speed. In this setting it follows from Theorem 6 that the numerical
solution will propagate with the speed c(µ0 , . . . , µN −1 ) in the limit ∆x, ∆t →
0. Thus, if c 6= 1 in each physical time step, the numerical solution will drift
out of phase. However, it should be noted that Theorem 6 is an asymptotic
result and does not necessarily imply that c(µ0 , . . . , µN −1 ) alone captures the
propagation speed error on coarse grids. In fact, we may generally expect a

24
 10 -1 10 0

10 -2 10 -1

10 -3 10 -2

10 -4 10 -3

10 -5 10 -4

10 -6
10 -6 10 -4 10 -2 10 -6 10 -4 10 -2

1
(a) Constant pseudo-time steps, µl = 20 . (b) Variable pseudo-time steps, µl = 2−l .

Figure 2: L2 -errors vs ∆x with respect to the exact solution of the original

and modified advection equations.

contribution to the speed from dispersion errors built into the discretization;
see e.g. [19–22, 29, 30] for details and remedies.
In the next experiment we verify that c depends on N as predicted by
Theorem 6 by measuring the propagation speed c̄ of the numerical solution
while varying N . To measure c̄ we track the x-coordinate of the maximum
of the propagating pulse through time. To get accurate measurements we
extend the computational domain to x ∈ (−1/5, 1/5], t ∈ (0, 6].
For this experiment we fix ∆x = 0.003 and set µl = µ to be fixed for each
l = 0, . . . , N − 1. The measured propagation speed error 1 − c̄ for different
µ are shown in Fig. 3a for Heun’s method and in Fig. 3b for SSPRK3. The
solid lines show the theoretically predicted error,
1 − c(µ0 , . . . , µN −1 ) = φ(−µ)N .
The measured and theoretical speed errors agree well for µ = 0.05 and
µ = 0.2. For µ = 0.5 a discrepancy between theory and measurement is
seen for small errors. This suggests that the dispersion error intrinsic to the
finite volume scheme is starting to dominate the propagation speed error.

4.3 Burgers’ equation

Next we consider a triangular shock wave propagating under the 1D Burgers’
equation with periodic boundary conditions:
 2
u
ut + = 0, x ∈ (0, 1],
2 x
( (36)
x if x ≤ 0.5,
u(x, 0) =
0 otherwise.

25
 10 0 10 0

10 -2
10 -2

10 -4
10 -4
0 5 10 15 20 0 5 10 15 20

(a) Heun’s method. (b) SSPRK3.

Figure 3: Proparagion speed error vs number of iterations per time step for
the linear advection problem.

The exact solution to this problem is given by

( √
x
t+1 if x ≤ 21 t + 1,
u(x, t) =
0 otherwise.

Modifying the conservation law to ut + (cu2 /2)x = 0 while retaining the

same initial condition modifies the exact solution to
(
x 1
√
if x ≤ 2 ct + 1,
u(x, t) = ct+1
0 otherwise.

For this problem we therefore expect that pseudo-time iterations to affect

both the speed and the amplitude of the shock front.
As for the advection problem, we investigate the L2 convergence of the
numerical scheme to the original and modified conservation laws. We once
again use an upwind numerical flux and implicit Euler in time;
!
un+1
i − uni 1 un+1
i
un+1
i−1
+ − = 0, i = 1, . . . , m. (37)
∆t ∆x 2 2

We run the simulation to time t = 0.1 with ∆t = ∆x using the sequence of

grids ∆x = 1/(25 × 2j ) for j = 1, . . . , 15. The explicit Euler method is used
as pseudo-time iterator with N = 12 and µl = 1/4 for l = 0, . . . , 11. The
corresponding modification constant is c ≈ 0.9683. Fig. 4a shows that the
numerical solution converges to the solution of the modified conservation
law as expected.

26
 10 -1 10 -1

10 -2 10 -2

10 -3 10 -3

10 -4 10 -4
10 -6 10 -4 10 -2 10 -6 10 -4 10 -2

(a) Triangular shock. (b) Step function.

Figure 4: L2 -errors vs ∆x with respect to the exact solution of the original

and modified Burgers’ equations for (a) the triangular shock and (b) the
step function.

Fixing ∆x = 0.004 and setting µl = µ = 1/4, we run the simulation to

time t = 1 with different choices of N . Fig. 5 shows the numerical solutions
together with the initial data (dashed line) and the exact solution (dotted
line). The locations of the tips of the shock waves as predicted by Theorem
6 are indicated as crosses in the figure. There is good agreement between
theory and experiment, although the shocks appear slightly smeared due to
the built-in dissipation in the numerical scheme.
Next, we repeat the experiments but change the initial data to a step
function, (
1 if x ≤ 0.24,
u(x, 0) =
0 otherwise.
Since this data is not periodic, we impose the boundary condition u(0, t) = 1.
Theorem 6 does not treat boundary conditions and it is therefore interesting
to see if it still provides useful predictions of the behaviour of the numerical
solution in this setting.
The exact solution is the initial step function travelling to the right. The
shock speed is given by the Rankine-Hugoniot condition as
u2l 2

2 − u2r 1
s= = ,
ul − ur 2
where ul,r denote left and right states of the discontinuity respectively. The
shock speed of the modified conservation law is instead c/2.
We use the exact same grids and pseudo-time iterations as for the trian-
gular shock and measure the L2 error with respect to the exact and modified

27
 0.6

0.5

0.4

0.3

0.2

0.1

-0.1
0 0.2 0.4 0.6 0.8 1

Figure 5: Numerical solutions and predicted shock locations (crosses) for the
solution of Burgers’ equation (19) using different numbers of pseudo-time
iterations, N . The dotted line indicates the exact solution. The dashed line
shows the initial data.

conservation laws. The results are shown in Fig. 4b. Convergence is once
again seen towards the modified equation, thus verifying that Theorem 6
predicts the propagation speed error correctly despite the added boundary
condition.
Next we extend the time domain to t ∈ (0, 1], fix ∆x = ∆t = 4∆τ =
1/100 and vary the number of iterations, N . The computed solutions using
N = 1, N = 3 and N = 12 are shown in Fig. 6 together with the predicted
shock locations (dashed lines). Once again, there is good agreement between
prediction and experiment, although numerical dissipation smears the shock
fronts somewhat.
As mentioned previously, the shock speed error can be eliminated entirely
for the explicit Euler method by noting that µ = 1 is a root of the stability
polynomial 1 − µ. To highlight the effect of this, we introduce two strategies
for choosing the pseudo-time steps:
Strategy 1: Use N = 12 iterations with ∆τ0,...,11 = ∆t/4.
Strategy 2: Use N = 9 iterations with ∆τ0 = ∆t and ∆τ1,...,8 = ∆t/4.

28
 1.2

0.8

0.6

0.4

0.2

-0.2
0 0.2 0.4 0.6 0.8 1

Figure 6: Numerical solutions and predicted shock locations (dashed lines)

for the solution of Burgers’ equation (19) using different numbers of pseudo-
time iterations, N .

The first strategy is the same that we have used in the experiments so far.
The second one ensures that c = 1 by taking a large initial pseudo-time
step. Note that both strategies correspond to integration in pseudo-time to
the same point; τ = 3∆t.
The relative residuals of the pseudo-time iterates in the first physical time
step are shown in Fig. 7 for the two strategies. The large initial pseudo-time
step in Strategy 2 results in a considerably greater residual reduction than
the corresponding iteration using Strategy 1. Interestingly, subsequent
iterations yield faster convergence of the residual using Strategy 2 as seen
by the steeper gradient. This suggests that the incorrect shock speed makes
the dominant and slowest converging contribution to the residual for this
problem when using Strategy 1. We also conclude that the point to which
we march in pseudo-time, here τ = 3∆t, have less of an impact on the
convergence than the choice of pseudo-time steps used to reach this point.

29
 10 0

10 -1

10 -2
0 2 4 6 8 10

Figure 7: Relative residuals of pseudo-time iterates applied to the discretiza-

tion (37) of Burgers’ equation using Strategy 1 and Strategy 2.

4.3.1 The Euler equations

As a second nonlinear problem we consider the 2D compressible Euler equa-
tions,      
ρ ρu ρv
 ρu  2
  +  ρu + p  +  ρuv  = 0,
   
 ρv   ρuv  2
 ρv + p  (38)
ρE t (ρE + p)u x (ρE + p)v y

posed on the domain (x, y) ∈ (−5, 15] × (−5, 5]. Here, ρ, u, v, E and p re-
spectively denote density, horizontal and vertical velocity components, total
energy per unit mass and pressure. The pressure is related to the other
variables through the equation of state

p = (γ − 1)ρe,

where γ = 1.4 and e = E − (u2 + v 2 )/2 is the internal energy density. The
domain is taken to be periodic in both spatial coordinates. The setting is

30
the isentropic vortex problem [25] with initial conditions
 1
 γ−1
1 − ǫ2 (γ − 1)M∞ 2
ρ0 = exp (r) ,
8π 2
ǫy
u0 =1− exp (r/2),
2π
ǫx
v0 = exp (r/2),
2π
ργ0
p0 = 2
,
γM∞

where r = 1 − x2 − y 2 . Here, ǫ = 5 is the circulation and M∞ = 0.5 is the

Mach number. As the solution evolves in time, the initial vortex propagates
in the horizontal direction with unit speed.
As in previous experiments, we use implicit Euler in time, yielding a
finite volume scheme of the form
   
un+1
i − uni 1 ˆn+1 ˆn+1 1 ˆn+1 ˆn+1
+ fi+ 1 ,j − fi− 1 ,j + fi,j+ 1 − fi,j− 1 = 0. (39)
∆t ∆x 2 2 ∆y 2 2

Along the x-coordinate we use a fourth order centered flux

1 7 7 1
fˆi+ 1 ,j = − fi−1,j + fi,j + fi+1,j − fi+2,j
2 12 12 12 12
and similarly along the y-coordinate. With this choice, the resulting problem
violates the assumptions of Theorem 6 in three ways: (i) It is 2D, (ii) it is
a system of equations, and (iii) the scheme is not total variation bounded.
The exact solution of the isentropic vortex problem is given by the initial
data (centred at the origin) translated to the point (x, y) = (T, 0), where
T is the end point of the time domain. However, the exact solution of the
modified conservation law is instead centred at (x, y) = (cT, 0).
The explicit Euler method is again used as pseudo-time iteration. We
study the convergence of the numerical solution to the exact solutions of the
original and modified conservation laws by measuring the L2 error of the
density component. The discretizations are chosen such that ∆x = ∆y =
4∆t on all grids. Two different strategies for choosing the pseudo-time steps
are considered:
Strategy 1: Use N = 9 iterations with ∆τ0,...,8 = 0.2∆t.
Strategy 2: Use N = 5 iterations with ∆τ0 = ∆t and ∆τ1,...,4 = 0.2∆t.
Both strategies integrate in pseudo-time to the point τ = 1.8∆t in each
time step. However, Theorem 6 predicts that Strategy 1 will give a speed

31
Figure 8: Computed density of the isentropic vortex at t = 10 using Strat-
egy 1 and Strategy 2. Dashed lines mark the vortex locations predicted
by Theorem 6.

modification c(µ0 , . . . , µ8 ) ≈ 0.866. On the other hand, Strategy 2 will give

c(µ0 , . . . , µ4 ) = 1, i.e. the correct propagation speed, due to the large initial
pseudo-time step. Fig. 8 shows the numerical solutions at time T = 10 for
the case where ∆x = ∆y = 0.2, ∆t = 0.05. The predicted and observed
vortex locations agree very well.
Fig. 9a shows the convergence of the numerical solutions, measured at
time T = 1. Convergence to the correct solution is observed for Strategy 2
(S2) but not for Strategy 1 (S1). This is expected due to the incorrect lo-
cation of the vortex in the latter case. However, convergence to the solution
of the modified conservation law is seen (S1 mod). Thus, Theorem 6 accu-
rately predicts the behavior of the numerical solution despite the violated
assumptions.
In practical applications, the pseudo-time iterations are terminated when
the residual has decreased beneath some tolerance. It is interestig to see in
what way the convergence is affected by the choice of strategy. Returning
to the setting in Fig. 8, the residual in each pseudo-time iteration for all
200 physical time steps are shown for the two strategies in Fig. 9b. The 200
lines overlap nearly perfectly, suggesting that the residual behaves similarly
in each physical time step. The initial iteration in Strategy 2 evidently has
a large impact on the reduction of the residual that supercedes those of the
other iterations put together. The remaining iterations appear to reduce
the residual by comparable amounts for the two strategies, as seen by the
similar gradients. In contrast to the shock problem considered previously,
this suggests that the convergence rate of the residual is dictated by other

32
 10 -1

10 0

10 -2

10 -3
10 -1

10 -4
10 -2 10 -1 10 0 0 2 4 6 8

(a) L2 -errors. (b) Relative residuals.

Figure 9: (a) L2 -error computed with respect to the exact solution using
Strategy 1 (S1) and Strategy 2 (S2), and with respect to the modified
conservation law (S1 mod). (b) Relative residuals at each iteration for 200
physical time steps.

factors than the propagation speed for this particular problem. Nonetheless,
a correct propagation speed is visibly very beneficial, here with a drop in
relative residual of more than an order of magnitude.

5 Summary and Conclusions

In this paper we have studied conservation properties of a selection of itera-
tive methods applied to 1D scalar conservation laws. The fact that conser-
vation is a design principle behind many numerical schemes motivates such
a study. We have established that Newton’s method, the Richardson itera-
tion, Krylov subspace methods, coarse grid corrections using agglomeration,
as well as ERK pseudo-time iterations preserve the global conservation of a
given scheme if the initial guess has correct mass. However, the Jacobi and
Gauss-Seidel methods do not, unless the linear system being solved possesses
particular properties.
The stronger requirement of local conservation has been investigated for
ERK pseudo-time iterations. We have shown that local conservation is pre-
served for finite volume schemes that employ the implicit Euler method in
time. However, the resulting modified numerical flux may be inconsistent
with the governing conservation law. An extension of the Lax-Wendroff
theorem shows that this inconsistency leads to convergence to a weak solu-
tion of a conservation law modified by a particular constant. We give an
exact expression for the modification constant, which depends only on the
stability function of the ERK method and the pseudo-time steps. Depend-

33
ing on the problem solved, this modification can alter both the propagation
speed and amplitude of the numerical solution if the constant differs from
unity. We present a strategy for ensuring that the constant equals one and
show numerically that the strategy results in faster convergence. Experi-
ments suggest that the results hold even for systems of conservation laws in
multiple dimensions.

References
[1] F. Bassi, A. Ghidoni, and S. Rebay, Optimal Runge–Kutta
smoothers for the p-multigrid discontinuous Galerkin solution of the
1D Euler equations, Journal of Computational Physics, 230 (2011),
pp. 4153–4175.
[2] F. Bassi and S. Rebay, GMRES discontinuous Galerkin solution
of the compressible Navier-Stokes equations, in Discontinuous Galerkin
Methods, Springer, 2000, pp. 197–208.
[3] P. Birken, Numerical methods for the unsteady compressible Navier–
Stokes equations, Habilitation thesis, University of Kassel, Kassel,
(2012).
[4] , Numerical Methods for Unsteady Compressible Flow Problems,
Chapman and Hall/CRC (to appear), 2021.
[5] P. Birken, J. Bull, and A. Jameson, Preconditioned smoothers
for the full approximation scheme for the RANS equations, Journal of
Scientific Computing, 78 (2019), pp. 995–1022.
[6] P. Birken, G. Gassner, M. Haas, and C.-D. Munz, Precondition-
ing for modal discontinuous Galerkin methods for unsteady 3D Navier–
Stokes equations, Journal of Computational Physics, 240 (2013), pp. 20–
35.
[7] P. Birken, G. J. Gassner, and L. M. Versbach, Subcell finite
volume multigrid preconditioning for high-order discontinuous Galerkin
methods, International Journal of Computational Fluid Dynamics, 33
(2019), pp. 353–361.
[8] P. Birken and A. Jameson, On nonlinear preconditioners in
Newton–Krylov methods for unsteady flows, International journal for
numerical methods in fluids, 62 (2010), pp. 565–573.
[9] P. Birken, A. Meister, S. Ortleb, and V. Straub, On stability
and conservation properties of (s)EPIRK integrators in the context of

34
 discretized PDEs, in XVI International Conference on Hyperbolic Prob-
lems: Theory, Numerics, Applications, Springer, 2016, pp. 617–629.
[10] P. Birken, J. D. Tebbens, A. Meister, and M. Tůma, Precon-
ditioner updates applied to CFD model problems, Applied Numerical
Mathematics, 58 (2008), pp. 1628–1641.
[11] D. S. Blom, P. Birken, H. Bijl, F. Kessels, A. Meister, and
A. H. van Zuijlen, A comparison of Rosenbrock and ESDIRK meth-
ods combined with iterative solvers for unsteady compressible flows, Ad-
vances in Computational Mathematics, 42 (2016), pp. 1401–1426.
[12] B. Cockburn, G. Kanschat, and D. Schötzau, A locally con-
servative LDG method for the incompressible Navier–Stokes equations,
Mathematics of Computation, 74 (2005), pp. 1067–1095.
[13] T. C. Fisher and M. H. Carpenter, High-order entropy stable finite
difference schemes for nonlinear conservation laws: Finite domains,
Journal of Computational Physics, 252 (2013), pp. 518–557.
[14] A. Jameson, W. Schmidt, and E. Turkel, Numerical solution of
the Euler equations by finite volume methods using Runge Kutta time
stepping schemes, in 14th fluid and plasma dynamics conference, 1981,
p. 1259.
[15] C. Junqueira-Junior, L. C. Scalabrin, E. Basso, and J. L. F.
Azevedo, Study of conservation on implicit techniques for unstructured
finite volume Navier–Stokes solvers, Journal of Aerospace Technology
and Management, 6 (2014), pp. 267–280.
[16] P. Lax and B. Wendroff, Systems of conservation laws, tech. rep.,
LOS ALAMOS NATIONAL LAB NM, 1959.
[17] R. J. LeVeque, Numerical methods for conservation laws, vol. 3,
Springer, 1992.
[18] , Finite volume methods for hyperbolic problems, vol. 31, Cam-
bridge university press, 2002.
[19] V. Linders, M. H. Carpenter, and J. Nordström, Accurate
solution-adaptive finite difference schemes for coarse and fine grids,
Journal of Computational Physics, 410 (2020), p. 109393.
[20] V. Linders, M. Kupiainen, S. H. Frankel, Y. Delorme, and
J. Nordstrom, Summation-by-parts operators with minimal disper-
sion error for accurate and efficient flow calculations, in 54th AIAA
Aerospace Sciences Meeting, 2016, p. 1329.

35
[21] V. Linders, M. Kupiainen, and J. Nordström, Summation-by-
parts operators with minimal dispersion error for coarse grid flow cal-
culations, Journal of Computational Physics, 340 (2017), pp. 160–176.
[22] V. Linders and J. Nordström, Uniformly best wavenumber approx-
imations by spatial central difference operators, Journal of Computa-
tional Physics, 300 (2015), pp. 695–709.
[23] W. L. Miranker, Numerical methods of boundary layer type for stiff
systems of differential equations, Computing, 11 (1973), pp. 221–234.
[24] Y. Saad, Iterative methods for sparse linear systems, SIAM, 2003.
[25] C.-W. Shu, Essentially non-oscillatory and weighted essentially non-
oscillatory schemes for hyperbolic conservation laws, in Advanced nu-
merical approximation of nonlinear hyperbolic equations, Springer,
1998, pp. 325–432.
[26] C.-W. Shu and S. Osher, Efficient implementation of essentially
non-oscillatory shock-capturing schemes, Journal of computational
physics, 77 (1988), pp. 439–471.
[27] V. Straub, S. Ortleb, P. Birken, and A. Meister, Adopting
(s)EPIRK schemes in a domain-based IMEX setting, in AIP Conference
Proceedings, vol. 1863, AIP Publishing LLC, 2017, p. 410008.
[28] R. C. Swanson, E. Turkel, and C.-C. Rossow, Convergence ac-
celeration of Runge–Kutta schemes for solving the Navier–Stokes equa-
tions, Journal of Computational Physics, 224 (2007), pp. 365–388.
[29] C. K. Tam and J. C. Webb, Dispersion-relation-preserving finite dif-
ference schemes for computational acoustics, Journal of computational
physics, 107 (1993), pp. 262–281.
[30] L. N. Trefethen, Group velocity in finite difference schemes, SIAM
review, 24 (1982), pp. 113–136.
[31] G. Wanner and E. Hairer, Solving ordinary differential equations
II, Springer Berlin Heidelberg, 1996.

A Proof of Lemma 4
The purpose of this appendix is to provide a detailed proof of Lemma 4.

36
Proof. Note first that (21)-(22) can be equivalently written in the form
(k) (k)
Ui = ui 1 − ∆τk Ag i (U (k) ),
(k+1) (k)
i = . . . , −1, 0, 1, . . . , (40)
ui = ui − ∆τk b⊤ g i (U (k) ),
 
(k) (k) ⊤
where 1 = (1, . . . , 1)⊤ ∈ Rs . Here, we use the notation g i (U k ) = gi (U 1 ), . . . , gi (U s ) .
Suppose that for some N ≥ 1 the relation
(N )  
ui − uni 1 (N ) (N )
+ ĥi+ 1 − ĥi− 1 = 0
∆t ∆x 2 2

holds. We will show that it also holds for N + 1. From (20), (24) and (40)
it follows that
" (N ) #
(N ) (N ) U i − uni 1 1  (N ) (N )

U i = ui 1 − ∆τN A + f̂ i+ 1 − f̂ i− 1
∆t ∆x 2 2
 
∆t (N ) (N )
= uni 1 − ĥi+ 1 − ĥi− 1 1
∆x 2 2

(N ) ∆τN  (N ) (N )

− µN AU i + µN A1uni − A f̂ i+ 1 − f̂ i− 1 .
∆x 2 2

(N )
Solving for U i gives
 
(N ) ∆t (N ) (N )
Ui = uni 1 − ĥi+ 1 − ĥi− 1 (I + µN A)−1 1
∆x 2 2

∆t  (N ) (N )

− µN (I + µN A)−1 A f̂ i+ 1 − f̂ i− 1 .
∆x 2 2

Note that (I + µNA)−1 exists since A is lower triangular.

Evaluating g i U (N ) using (20) gives
    
(N ) 1 (N ) (N )
gi U = − ĥi+ 1 − ĥi− 1 (I + µN A)−1 1
∆x 2 2
 (N ) (N )
  (N ) (N )
i
−µN (I + µN A)−1 A f̂ i+ 1 − f̂ i− 1 + f̂ i+ 1 − f̂ i− 1
2 2 2 2
  
1 (N ) (N )
= − ĥi+ 1 − ĥi− 1 (I + µN A)−1 1
∆x 2 2
 (N ) (N )
i
+[I − µN (I + µN A)−1 A] f̂ i+ 1 − f̂ i− 1
2 2
    
1 −1 (N ) (N ) (N ) (N )
= (I + µN A) f̂ i+ 1 − f̂ i− 1 − ĥi+ 1 − ĥi− 1 1 .
∆x 2 2 2 2

37
In the last equality we have used the fact that
I − µN (I + µN A)−1 A = (I + µN A)−1 [(I + µN A) − µN A] = (I + µN A)−1 .
  (41)
(N ) (N +1)
Inserting the above expression for g i U into ui as given in (40)
leads to
 
(N +1) (N )
ui = ui − ∆τN b⊤ g i U (N )
 
n ∆t (N ) (N )
= ui − ĥi+ 1 − ĥi− 1
∆x 2 2
    
∆τN ⊤ −1 (N ) (N ) (N ) (N )
− b (I + µN A) f̂ i+ 1 − f̂ i− 1 − ĥi+ 1 − ĥi− 1 1
∆x 2 2 2 2
 
∆t (N ) (N )
= uni − [1 − µN b⊤ (I + µN A)−1 1] ĥi+ 1 − ĥi− 1
∆x 2 2

∆t  (N ) (N )

− µN b⊤ (I + µN A)−1 f̂ i+ 1 − f̂ i− 1
∆x   2 2

∆t  (N ) (N )

n (N ) (N ) ⊤ −1
= ui − φ(−µN ) ĥi+ 1 − ĥi− 1 + µN b (I + µN A) f̂ i+ 1 − f̂ i− 1 .
∆x 2 2 2 2

Rearranging, using the induction hypothesis and the expression (25) for ĥN
results in
(N +1)
ui − uni
0=
∆t
" N −1 N −1
!
1 X Y  (k) (k)

+ φ(−µN ) µk b⊤ (I + µk A)−1 φ(−µl ) f̂ i+ 1 − f̂ i− 1
∆x 2 2
k=0 l=k+1
 (N ) (N )
i
⊤
+ µN b (I + µN A)−1 f̂ i+ 1 − f̂ i− 1
2 2
(N +1) N N
!
ui − uni 1 X
⊤ −1
Y  (k) (k)

= + µk b (I + µk A) φ(−µl ) f̂ i+ 1 − f̂ i− 1
∆t ∆x 2 2
k=0 l=k+1
(N +1)  
ui − uni 1 (N +1) (N +1)
= + ĥi+ 1 − ĥi− 1 .
∆t ∆x 2 2

It remains to show that the lemma holds when N = 1. To this end,

(0)
recall that ui = uni and note from (20) and (40) that
" (0) #
− u n1  (0) 
(0) U 1 (0)
U i = uni 1 − ∆τ0 A i i
+ f̂ i+ 1 − f̂ i− 1
∆t ∆x 2 2

∆τ0  (0) (0)


(0)
= uni 1 − µ0 AU i + µ0 A1uni − A f̂ i+ 1 − f̂ i− 1 .
∆x 2 2

38
 (0)
Solving for U i gives

∆τ0  (0) (0)


(0)
Ui = uni 1 − (I + µ0 A)−1 A f̂ i+ 1 − f̂ i− 1 .
∆x 2 2

Using (20) it follows that

  1 h  (0) (0)
  (0) (0)
i
g U (0) = −µ0 (I + µ0 A)−1 A f̂ i+ 1 − f̂ i− 1 + f̂ i+ 1 − f̂ i− 1
∆x 2 2 2 2
1  (0) (0)

−1
= [I − µ0 (I + µ0 A) A] f̂ i+ 1 − f̂ i− 1
∆x 2 2
1  (0) (0)

= (I + µ0 A)−1 f̂ i+ 1 − f̂ i− 1 .
∆x 2 2

(1)
Here we have once again used (41) in the final equality. Thus, ui can be
evaluated using (40) as
 (0)   
(1) ∆τ0 ⊤ (0) ∆t (1) (1)
ui = uni − b (I + µ0 A)−1 f̂ i+ 1 − f̂ i− 1 = uni − ĥi+ 1 − ĥi− 1 .
∆x 2 2 ∆x 2 2

Division by ∆t and rearrangement shows that the lemma holds when N = 1.

By induction it holds for all N ≥ 1.

39