## Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

**schemes for Hyperbolic Conservation Laws
**

Marcos Castro

∗

Bruno Costa

†

Wai Sun Don

‡

July 20, 2010

Abstract

In ([10], JCP 227 No. 6, 2008, pp. 3101–3211), the authors have designed a new ﬁfth order

WENO ﬁnite-diﬀerence scheme by adding a higher order smoothness indicator which is obtained

as a simple and inexpensive linear combination of the already existing low order smoothness

indicators. Moreover, this new scheme, dubbed as WENO-Z, has a CPU cost which is equivalent

to the one of the classical WENO-JS ([2],JCP 126, pp. 202–228 (1996)) and signiﬁcantly lower

than that of the mapped WENO-M,([5],JCP 207, pp. 542–567 (2005)), since it involves no

mapping of the nonlinear weights. In this article, we take a closer look at Taylor expansions of

the Lagrangian polynomials of the WENO substencils and the related inherited symmetries of

the classical lower order smoothness indicators to obtain a general formula for the higher order

smoothness indicators that allows the extension of the WENO-Z scheme to all (odd) orders of

accuracy. We further investigate the improved accuracy of the WENO-Z schemes at critical

points of smooth solutions as well as their distinct numerical features as a result of the new sets

of nonlinear weights and we show that regarding the numerical dissipation WENO-Z occupies an

intermediary position between WENO-JS and WENO-M. Some standard numerical experiments

such as the one dimensional Riemann initial values problems for the Euler equations and the

Mach 3 shock density-wave interaction and the two dimensional double-Mach shock reﬂection

problems are presented.

Keywords

Weighted Essentially Non-Oscillatory, WENO-Z, Smoothness Indicators, Nonlinear Weights

AMS

65P30, 77Axx

∗

Department of Mathematics, Hong Kong Baptist University, Hong Kong, China.

E-Mail: wsdon@hkbu.edu.hk

†

Departamento de Matem´atica Aplicada, IM-UFRJ, Caixa Postal 68530, Rio de Janeiro, RJ, C.E.P. 21945-970,

Brazil. E-Mail: bcosta@ufrj.br

‡

Departamento de Matem´atica Aplicada, IM-UFRJ, Caixa Postal 68530, Rio de Janeiro, RJ, C.E.P. 21945-970,

Brazil. E-Mail: marcoscastro@ufrj.br

1

1 Introduction

The weighted essentially non-oscillatory conservative ﬁnite diﬀerence schemes (WENO) [1, 2] are a

popular choice of numerical methods for solving compressible ﬂows modeled by means of hyperbolic

conservation laws in the form

∂u

∂t

+∇· F(u) = 0, (1)

and in the presence of shocks and small scales structures. WENO schemes owe their success to the

use of a dynamic set of stencils where a nonlinear convex combination of lower order polynomials

adapts either to a higher order polynomial approximation at smooth parts of the solution, or to

an upwind spatial discretization that avoids interpolation across discontinuities and provides the

necessary dissipation for shock capturing. It is an evolution of the Essentially Non-Oscillatory

(ENO) schemes, introduced in [7], which choose only the smoothest stencil, instead of forming a

combination of all the stencils available in order to optimize accuracy.

The local computational grids of ENO and WENO schemes are composed of r overlapping substen-

cils of r points, forming a larger stencil with (2r −1) points and yielding a local rate of convergence

that goes from order r, in the case of ENO when only one substencil is used, to order (2r − 1)

when the WENO weighted combination is applied at smooth parts of the solution. The nonlinear

coeﬃcients of WENO’s convex combination, hereafter referred to as nonlinear weights, are based

on lower order local smoothness indicators that measure the sum of the normalized squares of the

scaled L

2

norms of all derivatives of local interpolating polynomials [2]. An essentially zero weight

is assigned to those lower order polynomials whose underlining substencils contain high gradients

and/or shocks, aiming at an essentially non-oscillatory solution close to discontinuities. At smooth

parts of the solution, higher order is achieved through the mimicking of the central upwind scheme

of maximum order, when all smoothness indicators are about the same size. Hence, an eﬃcient and

careful design of these smoothness indicators is a delicate and important issue for WENO schemes.

The ﬁrst set of nonlinear weights of widespread use has been presented in [2], hereafter denoted as

ω

k

, where β

k

are the associated lower order smoothness indicators. The scheme resulting from these

will be hereafter referred as the classical WENO scheme (WENO-JS). In [5] it was pointed out that

these smoothness indicators were the cause of a reduction of the convergence rate at critical points

(points of zero derivatives) of the function. In the same article, a ﬁxing was also proposed in the

form of a mapping on the classical WENO-JS weights, leading to corrected weights that recovered

the formal order of accuracy at critical points. We call the scheme composed by this mapped set

of weights as the mapped WENO scheme (WENO-M). Alternatively, in [10], it was shown that

the incorporation of a global higher order smoothness indicator, which we call τ

2r−1

(in order to

emphasize the utilization of the whole set of (2r −1) points available) into the WENO-JS weights

deﬁnition also improved the convergence at critical points with no need of mapping (in the case of

r = 3). This last scheme has been named the WENO-Z scheme by its authors. However, a direct

comparison of the schemes showed that the improvement on numerical solutions of the WENO-M

over the WENO-JS was not due to the increase of the convergence rate at critical points, as it

was claimed in [5], but to the decreased dissipation of WENO-M when correcting the weights to a

disposition closer to the central upwind scheme. In the same way, the new set of nonlinear weights

of WENO-Z provided even less dissipation than WENO-M, obtaining sharper results among all of

the schemes. Moreover, the mapping procedure of WENO-M incurs in extra computational cost,

while the WENO-Z modiﬁcations are obtained through a simple and inexpensive linear combination

2

of the WENO-JS smoothness indicators β

k

.

In this article, we extend the ﬁfth order WENO-Z scheme introduced in [10] to higher orders of

accuracy by providing a closed-form formula for the generation of the associated global smoothness

indicators τ

2r−1

as linear combinations of the classical lower order local smoothness indicators β

k

.

This was achieved through a thorough investigation of the Taylor expansions of the β

k

that revealed

symmetries that were used to build the τ

2r−1

for all values of r. As we shall see, these symmetries

originate in the Lagrangian interpolation bases of the local substencils due to their symmetrical

dispositions with respect to the global stencil. We also determine the maximum order the higher

order smoothness indicator τ

2r−1

can achieve as a linear combination of the β

k

.

As mentioned above, WENO’s main idea to avoid oscillations is to generate a convex combination

of low order polynomials giving smaller weights to substencils containing discontinuities or high

gradients. In [10] it was pointed out that the diﬀerences of resolutions among the three schemes,

WENO-JS, WENO-M and WENO-Z, were due to their distinct dissipative properties, which in

turn were consequence of the sizes of the weights assigned to the discontinuous substencils. Thus,

the WENO-Z sharper results that were obtained in [10] were due to the assignment of weights to

discontinuous substencils that were larger than the ones of WENO-M. Analogously, in [5], WENO-

M also assigned larger weights than the ones of WENO-JS. In this article, we show that the general

situation is a little bit distinct from what was shown in [5] and [10], for if we take a closer look at

all the parameters involved at the deﬁnition of the weights, we conclude that the weights power

parameter p (see formula (25)) plays a major role in the amount of dissipation a particular WENO

scheme might have, since it aﬀects the weights relative scaling in the presence of discontinuities.

We will also see that the schemes can be classiﬁed in a dissipation scale with WENO-JS being the

most dissipative and the mapped WENO being the less for the same value of p, with WENO-Z

occupying an intermediary position. For higher values of r, this classiﬁcation determines which

scheme keeps the ENO behavior for the values of p currently used. The power parameter p also

has an active role on recovering the formal order of accuracy at critical points of the solution and

we shall see that this is a distinct property of the WENO-Z scheme.

This paper is organized as follows: In Section 2, brief descriptions of the several WENO schemes

are given. The main results of the article will be presented in Section 3, where we derive a general

formula for the higher order smoothness indicators τ

2r−1

of all odd orders WENO-Z scheme. We

also derive a closed-form formula, for arbitrary values of r, for the optimal order that τ

2r−1

can

achieve when expressed as a linear combination of the lower order local smoothness indicators

β

k

, k = 0, . . . , r − 1. The issue of the degradation of the order of convergence at critical points of

smooths solutions is discussed in Section 4. In Section 5, dissipation of the various schemes and the

inﬂuence of the values of the power parameter p are discussed through the numerical simulation of

the linear advection of discontinuous functions, the one dimensional Euler equations with Riemann

initial values problems and the Mach 3 shock density-wave interaction. The two dimensional double-

Mach shock reﬂection problem was also simulated using the high order WENO-Z ﬁnite diﬀerence

scheme. Concluding remarks are given in Section 6.

3

2 Weighted essentially non-oscillatory schemes

In this section we describe all three versions of the (2r−1) order weighted essentially non-oscillatory

conservative ﬁnite diﬀerence scheme with r ≥ 3 when applied to hyperbolic conservation laws as in

(1). Namely, we ﬁrst recall the essentials of the classical WENO scheme as designed in [2], along

with the mapped version introduced in [5] and then the WENO-Z scheme introduced in [10], which

we denote as WENO-JS, WENO-M and WENO-Z, respectively. We will recall the ﬁfth (r = 3) as

an example in the discussion below.

2.1 The Classical WENO-JS scheme

x

i

x

i+1

x

i+2

x

i-1

x

i-2

x

i+1/2

S

2

S

0

S

1

S

5

τ

5

β

0

β

2

β

1

Figure 1: The computational uniform grid x

i

and the 5-points stencil S

5

, composed of three 3-points

substencils S

0

, S

1

, S

2

, used for the ﬁfth-order WENO reconstruction step.

Consider an uniform grid deﬁned by the points x

i

= i∆x, i = 0, . . . , N, which are called cell

centers, with cell boundaries given by x

i+

1

2

= x

i

+

∆x

2

, where ∆x is the uniform grid spacing.

The semi-discretized form of (1), by the method of lines, yields a system of ordinary diﬀerential

equations

du

i

(t)

dt

= −

∂f

∂x

¸

¸

¸

¸

x=x

i

, i = 0, . . . , N, (2)

where u

i

(t) is a numerical approximation to the point value u(x

i

, t).

A conservative ﬁnite-diﬀerence formulation for hyperbolic conservation laws requires high-order

consistent numerical ﬂuxes at the cell boundaries in order to form the ﬂux diﬀerences across the

uniformly-spaced cells. The conservative property of the spatial discretization is obtained by im-

plicitly deﬁning the numerical ﬂux function h(x) as

f(x) =

1

∆x

_

x+

∆x

2

x−

∆x

2

h(ξ)dξ,

4

such that the spatial derivative in (2) is exactly approximated by a conservative ﬁnite diﬀerence

formula at the cell boundaries,

du

i

(t)

dt

=

1

∆x

_

h

i+

1

2

−h

i−

1

2

_

, (3)

where h

i±

1

2

= h(x

i±

1

2

).

High order polynomial interpolations to h

i±

1

2

are computed using known grid values of f, f

i

= f(x

i

).

The classical (2r −1) order WENO scheme uses (2r −1)-points global stencil, which is subdivided

into r substencils {S

0

, S

1

, . . . , S

r−1

} with each substencil containing r grid points. For instance, the

classical ﬁfth-order WENO scheme uses a 5-points stencil, hereafter named S

5

, which is subdivided

into three 3-points substencils {S

0

, S

1

, S

2

}, as shown in Fig. 1. The (2r − 1) degree polynomial

approximation

ˆ

f

i±

1

2

= h

i±

1

2

+O(∆x

2r−1

) is built through the convex combination of the interpolated

values

ˆ

f

k

(x

i±

1

2

), in which f

k

(x) is the r-th degree polynomial below, deﬁned in each one of the

substencils S

k

:

ˆ

f

i±

1

2

=

r−1

k=0

ω

k

ˆ

f

k

(x

i±

1

2

), (4)

where

ˆ

f

k

(x

i+

1

2

) =

ˆ

f

k

i+

1

2

=

r−1

j=0

c

kj

f

i−k+j

, i = 0, . . . , N. (5)

The c

kj

are Lagrangian interpolation coeﬃcients (see [2]), which depend on the left-shift parameter

k = 0, . . . , r −1, but not on the values f

i

.

• In the case of r = 3, it can be shown by Taylor series expansion of (5) that

ˆ

f

k

i±

1

2

= h

i±

1

2

+ A

k

∆x

3

+ O(∆x

4

), (6)

where the values A

k

are independent of ∆x.

The weights ω

k

are deﬁned as

ω

k

=

α

k

r−1

l=0

α

l

, α

k

=

d

k

(β

k

+ ǫ)

p

. (7)

We refer to α

k

as the un-normalized weights. The parameter ǫ is used to avoid the division by

zero in the denominator and power parameter p = 2 is chosen to increase the diﬀerence of scales of

distinct weights at non-smooth parts of the solution. The coeﬃcients {d

0

, d

1

, . . . , d

r−1

} are called

the ideal weights since they generate the (2r −1) order central upwind scheme using the (2r −1)-

points stencil. For example, the coeﬃcients {d

0

=

3

10

, d

1

=

3

5

, d

2

=

1

10

} generate the ﬁfth-order

central upwind scheme for the 5-points stencil S

5

. Ideal weights for higher order WENO schemes

can be found in [2].

The lower order local smoothness indicators β

k

measure the regularity of the (r − 1) th degree

polynomial approximation

ˆ

f

k

(x

i

) at the substencil S

k

and are given by

β

k

=

r−1

l=1

∆x

2l−1

_

x

i+

1

2

x

i−

1

2

_

d

l

dx

l

ˆ

f

k

(x)

_

2

dx. (8)

5

• For example, in the case of r = 3, the expression of the β

k

in terms of the cell averaged values

of f(x), f

i

are given by

β

0

=

13

12

(f

i−2

−2f

i−1

+ f

i

)

2

+

1

4

(f

i−2

−4f

i−1

+ 3f

i

)

2

, (9)

β

1

=

13

12

(f

i−1

−2f

i

+ f

i+1

)

2

+

1

4

(f

i−1

−f

i+1

)

2

, (10)

β

2

=

13

12

(f

i

−2f

i+1

+ f

i+2

)

2

+

1

4

(3f

i

−4f

i+1

+ f

i+2

)

2

, (11)

and their Taylor series expansions at x

i

are

β

0

= f

′

i

2

∆x

2

+

_

13

12

f

′′

i

2

−

2

3

f

′

i

f

′′′

i

_

∆x

4

−

_

13

6

f

′′

i

f

′′′

i

−

1

2

f

′

i

f

′′′′

i

_

∆x

5

+ O(∆x

6

), (12)

β

1

= f

′

i

2

∆x

2

+

_

13

12

f

′′

i

2

+

1

3

f

′

i

f

′′′

i

_

∆x

4

+ O(∆x

6

), (13)

β

2

= f

′

i

2

∆x

2

+

_

13

12

f

′′

i

2

−

2

3

f

′

i

f

′′′

i

_

∆x

4

+

_

13

6

f

′′

i

f

′′′

i

−

1

2

f

′

i

f

′′′′

i

_

∆x

5

+ O(∆x

6

). (14)

The general idea of the weights deﬁnition (7) is that on smooth parts of the solution the smoothness

indicators β

k

are all small and about the same size, generating weights ω

k

that are good approxi-

mations to the ideal weights d

k

. On the other hand, if the substencil S

k

contains a discontinuity, β

k

is O(1) and the corresponding weight ω

k

is small relatively to the other weights. This implies that

the inﬂuence of the polynomial approximation of h

i±

1

2

taken across the discontinuity is diminished

up to the point where the convex combination (4) is essentially non-oscillatory. For instance, in

the case r = 3, Fig. 1 shows the case where substencil S

2

is discontinuous, yielding β

0

and β

1

to be

much smaller than β

2

. By (7), this results on ω

2

being a small number in (4).

The process synthesized by (4)-(5) is called the WENO reconstruction step, for it reconstructs

the values of h(x) at the cell boundaries of the interval I

i

= [x

i−

1

2

, x

i+

1

2

] from its cell averaged

values f(x) in the substencils {S

k

, k = 0, . . . , r − 1}. In [5], truncation error analysis of the ﬁnite

diﬀerence equation (3) led to the following necessary and suﬃcient conditions on the weights ω

k

for the WENO scheme to achieve the formal (2r −1) order of convergence at smooth parts of the

solution:

r−1

k=0

A

k

(ω

+

k

−ω

−

k

) = O(∆x

r

), (15)

ω

±

k

−d

k

= O(∆x

r−1

). (16)

In the case r = 3, we see that a suﬃcient condition for ﬁfth-order convergence is simply given by:

ω

±

k

−d

k

= O(∆x

3

). (17)

It was also found that at ﬁrst order critical points x

c

, points where the ﬁrst derivative of the

function vanishes (f

′

(x

c

) = 0), the rate of convergence degraded to only third order (O(∆x

3

)), a

fact that was hidden by the homogenization of the weights caused by the use of a relatively large

value for ǫ in (7).

Let us now check how the classical WENO-JS nonlinear weights ω

k

in (7) behave with respect

to the restrictions above. In [5], it was shown that if the smoothness indicators β

k

satisfy β

k

=

6

D(1 + O(∆x

q

)), then the weights ω

k

satisfy ω

k

= d

k

+ O(∆x

q

), where D is a nonzero constant

independent of k. Looking at the Taylor series expansions of the smoothness indicators β

k

in

(12)-(14), we see that β

k

= D

_

1 + O(∆x

2

)

_

, implying that ω

k

= d

k

+ O(∆x

2

). This requires that

condition (15) must be satisﬁed as well for the classical WENO to have the expected ﬁfth-order

convergence, which indeed happens and can be easily conﬁrmed with any symbolic calculation

software.

Nevertheless, at critical points this situation becomes more complex depending on the number

of vanishing derivatives of f(x

i

). For instance, if only the ﬁrst derivative vanishes, then β

k

=

D(1 + O(∆x)) and ω

k

= d

k

+ O(∆x), degrading the order of convergence of the scheme to third

order only. If the second derivative also vanishes, then the rate of convergence decreases even more

to second order.

2.2 The Mapped WENO-M scheme

A ﬁx to this deﬁciency of the classical weights ω

k

was proposed in [5]. It consisted of the application

of a mapping function that increased the approximation of ω

k

to the ideal weights d

k

at critical

points to the required third order O(∆x

3

) as in (17). The mapping function g

k

(ω) used in [5] is

deﬁned as

g

k

(ω) =

ω

_

d

k

+ d

2

k

−3d

k

ω + ω

2

_

d

2

k

+ ω (1 −2d

k

)

, (18)

and is a non-decreasing monotone function with the following properties:

1. 0 ≤ g

k

(ω) ≤ 1, g

k

(0) = 0 and g

k

(1) = 1.

2. g

k

(ω) ≈ 0 if ω ≈ 0; g

k

(ω) ≈ 1 if ω ≈ 1.

3. g

k

(d

k

) = d

k

, g

′

k

(d

k

) = g

′′

k

(d

k

) = 0.

4. g

k

(ω) = d

k

+ O

_

∆x

6

_

, if ω = d

k

+ O

_

∆x

2

_

.

Numerical results in [5] conﬁrmed the usefulness of the mapping, since with the modiﬁed weights

the resulting WENO-M scheme recovered the formal ﬁfth-order convergence at critical points of a

smooth solution. Note, however, that if at a critical point the second derivative also vanishes, β

k

=

D(1+O(1)), implying ω

k

= d

k

+O(1) (see equations (12)-(14)) and the mapping is unable to improve

the weights approximation, maintaining the same second order of convergence as the classical

WENO-JS scheme. The downside of using the mapping function is an additional ≈ 20% − 30%

cost of cpu time when compared to the classical WENO-JS scheme.

2.3 The WENO-Z Scheme

The novel idea of the WENO-Z scheme introduced in [10] is the modiﬁcation of the β

k

with

information obtained from a higher order smoothness indicator, which we denote here by τ

2r−1

for

any given order (2r − 1), r ≥ 3. This new smoothness indicator is built using the values of the

numerical solution at the whole (2r − 1) points stencil in the form of a simple linear combination

7

of the β

k

. For instance, as shown in [10], for r = 3, τ

5

is simply deﬁned as the absolute diﬀerence

between β

0

and β

2

at x

i

, namely

τ

5

= |β

0

−β

2

| . (19)

It is straightforward to see from (12)-(14) that the truncation error of τ

5

is

13

3

¸

¸

f

′′

i

f

′′′

i

¸

¸

∆x

5

+ O(∆x

6

), (20)

and that it is a measure of higher derivatives of f, when they exist, and is indeed computed using

the whole 5-points stencil S

5

. The relevant properties of τ

5

to be used in the redeﬁnition of the

WENO weights are:

• If S

5

does not contain discontinuities, then τ

5

= O(∆x

5

) ≪ β

k

, for k = 0, 1, 2;

• if the solution is continuous at some of the S

k

, but discontinuous in the whole S

5

, then

β

k

≪ τ

5

, for those k where the solution is continuous;

• τ

5

≤ max

k

β

k

.

The lower order local smoothness indicators β

Z

k

of the WENO-Z scheme are then deﬁned with the

help of τ

5

as

β

Z

k

=

_

β

k

+ ǫ

β

k

+ τ

5

+ ǫ

_

, k = 0, 1, 2, (21)

and the new normalized nonlinear weights ω

Z

k

and the un-normalized nonlinear weights α

Z

k

becomes

ω

Z

k

=

α

Z

k

2

l=0

α

Z

l

, α

Z

k

=

d

k

β

Z

k

= d

k

_

1 +

τ

5

β

k

+ ǫ

_

, k = 0, 1, 2, (22)

where ǫ is, as usual, a small number used to avoid the division by zero in the denominators of (21)

and (22).

It is straightforward to check from (12)-(14) and the properties of τ

5

that, at smooth parts of the

solution,

τ

5

β

k

+ ǫ

= O(∆x

3

), k = 0, 1, 2, (23)

whenever ǫ << β

k

, and from (22),

ω

Z

k

= d

k

+ O(∆x

3

), k = 0, 1, 2. (24)

Thus, the new weights ω

Z

k

satisfy the suﬃcient condition (17), providing the formal ﬁfth order of

accuracy to the WENO-Z scheme at noncritical points of a smooth solution.

The general deﬁnitions of the normalized and un-normalized nonlinear weights ω

Z

k

and α

Z

k

, respec-

tively, for r ≥ 3 are respectively:

ω

Z

k

=

α

Z

k

r−1

l=0

α

Z

l

, α

Z

k

=

d

k

β

Z

k

= d

k

_

1 +

_

τ

2r−1

β

k

+ ǫ

_

p

_

, k = 0, . . . , r −1, (25)

where p ≥ 1 is the power parameter, used to enhance the ratio between the smoothness indicators

to guarantee convergence at a certain order. Analogously, we need

ω

Z

k

= d

k

+ O(∆x

r

), k = 0, 1, . . . , r −1, (26)

8

to obtain convergence of the WENO-Z at order (2r −1) and this is achieved if

τ

2r−1

β

k

+ ǫ

= O(∆x

r

), k = 0, 1, . . . , r −1. (27)

As we shall see below, the β

k

are always O(∆x

2

), therefore, we need τ

2r−1

= O(∆x

r+2

). In the

next section, we provide a closed-form formula for generating such a τ

2r−1

for all values of r as a

linear combination of the β

k

.

Remark 1 At critical points, the lower order local smoothness indicators β

k

are no longer O(∆x

2

)

and the size of τ

2r−1

also varies in a way that the order of convergence of the ratio in (23)decreases

monotonically from r to 1 as n

cp

, the order of the critical point, goes from 0 to r − 1. The power

parameter p can be used to recover this order. For instance, as shown in [10], for r = 3, at a ﬁrst

order critical point the convergence of the WENO-Z scheme degrades to fourth order if p = 1 in (25)

and regains ﬁfth order when p = 2. This is unique to WENO-Z, changing the value of p in WENO-

JS or in WENO-M does not alter their convergence rate at critical points. Numerical experiments

at Section 4 further illustrate this property and more detailed computations that demonstrate this

exclusive aspect of WENO-Z are shown in [10].

3 The global higher order smoothness indicators

In this section we formulate and prove the necessary theoretical results to obtain a general formula

for the global higher order smoothness indicators τ

2r−1

for all values of r. The proof of the main

theorem explores symmetric structures of the underlying interpolating polynomials deﬁning the β

k

.

This will be done with the help of two lemmas and an auxiliary theorem. We also determine the

existence for all r, although no closed formula is provided, of an improved higher order smoothness

indicator, τ

opt

2r−1

, which has the optimal order among all the linear combinations of the β

k

. We

shall see that the use of τ

opt

2r−1

in place of τ

2r−1

in the weights deﬁnition improves the ability of the

WENO-Z scheme to capture higher order structures in the numerical solution.

3.1 General Formula for τ

2r−1

Theorem 2 Given the order of the WENO reconstruction (2r −1) and the associated lower order

local smoothness indicators β

0

, ..., β

r−1

, the global higher order smoothness indicator τ

2r−1

, deﬁned

as:

τ

2r−1

=

_

|β

0

−β

r−1

| mod (r, 2) = 1

|β

0

−β

1

−β

r−2

+ β

r−1

| mod (r, 2) = 0

, (28)

is of order O(∆x

r+2

).

Before we give the proof of the theorem at the end of this section, we will give some preliminary

results and two lemmas that are necessary for the proof. Without loss of generality, we take i = 0

(x

0

= 0) and denote

d

n

dx

n

f(x

0

) = f

(n)

0

unless stated otherwise. Furthermore, we shall denote the

9

lower order polynomial approximation

ˆ

f

k

(x) of degree (r −1) in the substencil S

k

as p

k

(x):

p

k

(x) =

r−1

j=0

a

k,j

x

j

, (29)

where a

k,j

are the coeﬃcients of the polynomial expansion of f(x) about the point x

0

in the

substencil S

k

.

The n th derivative of p

k

(x), p

(n)

k

(x), is given by

p

(n)

k

(x) =

r−1−n

j=0

b

k,n,j

x

j

, 1 ≤ n ≤ r −1, (30)

where b

k,n,j

=

(j + n)!

j!

a

k,n+j

.

Thus, we may write β

k

as

β

k

=

r−1

n=1

∆x

2n−1

_ 1

2

∆x

−

1

2

∆x

_

p

(n)

k

(x)

_

2

dx

=

r−1

n=1

∆x

2n−1

_ 1

2

∆x

−

1

2

∆x

_

_

r−1−n

j

1

=0

b

k,n,j

1

x

j

1

_

_

_

_

r−1−n

j

2

=0

b

k,n,j

2

x

j

2

_

_

dx

=

r−1

n=1

∆x

2n−1

_ 1

2

∆x

−

1

2

∆x

r−1−n

j

1

=0

r−1−n

j

2

=0

b

k,n,j

1

b

k,n,j

2

x

j

1

+j

2

dx

=

r−1

n=1

r−1−n

j

1

=0

r−1−n

j

2

=0

C

n,j

1

,j

2

a

k,n+j

1

a

k,n+j

2

∆x

j

1

+j

2

+2n

, (31)

where

C

n,j

1

,j

2

=

_

(j

1

+n)!(j

2

+n)!

j

1

!j

2

!

2

−(j

1

+j

2

)

(j

1

+j

2

+1)

mod (j

1

+ j

2

, 2) = 0

0 mod (j

1

+ j

2

, 2) = 1

, j

1

, j

2

= 0, . . . , r −1.

The idea of the proof of Theorem 2 is to rewrite β

k

in (31) as another asymptotic expansion in ∆x

where the coeﬃcients show an anti-symmetric behavior with respect to the substencils index k. We

start with the following two lemmas below, which express the numerical ﬂux h(x) in terms of the

derivatives of the physical ﬂux function f(x) and establish the independence and anti-symmetry

properties of the polynomial coeﬃcients a

k,j

.

Lemma 3 Consider the primitive function h(x), the numerical ﬂux of f(x), as deﬁned in (3), then

h(x) =

∞

δ=0

φ

2δ

f

(2δ)

(x)∆x

2δ

, (32)

where φ

0

= 1 and φ

2δ

, δ = 1, . . . are constants in the expansion.

10

Proof. From (3), and expanding h

±

1

2

= h(x

±

1

2

) about x

0

= 0,

f

′

= h

′

+

∆x

2

4

1

3!

h

(3)

+

∆x

3

16

1

5!

h

(5)

+· · · , (33)

Integrating both sides,

f = h +

∆x

2

4

1

3!

h

(2)

+

∆x

3

16

1

5!

h

(4)

+· · · , (34)

and diﬀerentiating (33),

f

(2)

= h

(2)

+

∆x

2

24

h

(4)

+· · · ,

f

(4)

= h

(4)

+

∆x

2

24

h

(6)

+· · · .

Substituting h

(2)

and h

(4)

in (34) yields, after some algebra,

h = f −

1

24

f

(2)

∆x

2

+

7

5760

f

(4)

∆x

4

+· · · .

We have the constants φ

2j

as shown in the expansion with φ

0

= 1.

The above process can be repeated to replace the higher order derivatives of h(x) by f(x) and to

obtain (32).

Lemma 4 Let ρ

j

=

_

r−j−1

2

_

. If x ∈ [x

−

1

2

, x1

2

], then p

k

(x) in (29) can be written as,

p

k

(x) =

r−1

j=0

a

k,j

x

j

, (35)

where the coeﬃcients a

k,j

are expressed either as:

(a)

a

k,j

=

1

j!

ρ

j

δ=0

φ

2δ

f

(j+2δ)

0

∆x

2δ

+ O(∆x

r−j

), (36)

with the coeﬃcients φ

2δ

as given in lemma 3,

(b) Or alternatively as:

a

k,j

=

∞

l=0

σ

k,j,l

f

(j+l)

0

∆x

l

, (37)

where

σ

k,j,l

= (−1)

l

σ

r−1−k,j,l

, (38)

and, for l ≤ ρ

j

, the above formula is independent of k, that is

σ

k,j,l

=

_

1

j!

φ

l

mod (l, 2) = 0

0 mod (l, 2) = 1

. (39)

11

Proof. p

k

is a (r −1)-th degree polynomial approximation of h:

p

k

(x) = h(x) + O(∆x

r

). (40)

Combining equations (32) and (40) and expanding f and its derivatives in Taylor series around

x

0

= 0, we have

p

k

(x) =

∞

j=0

1

j!

∞

δ=0

φ

2δ

f

(j+2δ)

(x

0

)(x −x

0

)

j

∆x

2δ

+ O(∆x

r

).

One can see that the terms in the second summation are of O(∆x

j+2δ

) since (x −x

0

) = O(∆x).

For j + 2δ > r −1, or l > ρ

j

=

_

r−j−1

2

_

, one has (x −x

0

)

j

∆x

2δ

≤ O(∆x

r

), therefore

p

k

(x) =

r−1

j=0

1

j!

ρ

j

δ=0

φ

2δ

f

(j+2δ)

(x

0

)(x −x

0

)

j

∆x

2δ

+ O(∆x

r

), (41)

with φ

0

= 1.

Hence

p

k

(x) =

r−1

j=0

a

k,j

(x −x

0

)

j

, (42)

with

a

k,j

=

1

j!

ρ

j

δ=0

φ

2δ

f

(j+2δ)

0

∆x

2δ

+ O(∆x

r−j

).

Proof of (b): Our objective is to build (r − 1)-th degree polynomial approximations of h in the

substencils S

k

and S

r−1−k

in order to obtain the symmetry in (38). For that we use the functions

F

k

and G

k

, primitives of h deﬁned as

F

k

(x) =

_

x

x

[k−(r−1)]−

1

2

h(ξ)dξ, and G

k

(x) = −

_

x

[(r−1)−k]+

1

2

x

h(ξ)dξ.

We rewrite these as:

F

k

(x

i+

1

2

) =

i

s=k−(r−1)

_

x

s+

1

2

x

s−

1

2

h(ξ)dξ =

i

s=k−(r−1)

f

s

∆x, i ∈ {k −(r −1), ..., k}, (43)

G

k

(x

i+

1

2

) = −

(r−1)−k

s=i+1

f

s

∆x, i ∈ {−k, ..., (r −1) −k},

where f

s

=

_

x

s+

1

2

x

s−

1

2

h(ξ)dξ.

Let P

F

k

(x) be the only polynomial of degree less than or equal to r that interpolates F

k

(x) in the

r + 1 points x

i+

1

2

, i ∈ {k −r, ..., k}, and similarly, let P

G

k

(x) be the only polynomial of degree less

12

than or equal to r that interpolates G

k

(x) in the r + 1 points x

i+

1

2

, i ∈ {−k − 1, ..., (r − 1) − k}.

One can show that (see [3] Shu)

h(x) = F

′

k

(x) =

_

P

F

k

_

′

(x) + O(∆x

r

), x ∈ [x

[k−(r−1)]−

1

2

, x

k+

1

2

],

h(x) = G

′

k

(x) =

_

P

G

k

_

′

(x) + O(∆x

r

), x ∈ [x

−k−

1

2

, x

(r−1)−k+

1

2

],

and thus

p

k

(x) =

_

P

F

k

_

′

(x), x ∈ [x

[k−(r−1)]−

1

2

, x

k+

1

2

],

p

(r−1)−k

(x) =

_

P

G

k

_

′

(x), x ∈ [x

−k−

1

2

, x

(r−1)−k+

1

2

],

are the approximations we are searching for. We can rewrite P

F

k

(x) and P

G

k

(x) using the following

Lagrangian interpolation formulas:

P

F

k

(x) =

k

i=k−r

F

k

(x

i+

1

2

)L

k,i

(x),

P

G

k

(x) =

(r−1)−k

i=−k−1

G

k

(x

i+

1

2

)L

(r−1)−k,i

(x),

with

L

k,i

(x) =

k

s=k−r

s=i

x −x

s+

1

2

x

i+

1

2

−x

s+

1

2

=

k

s=k−r

s=i

x −(s +

1

2

)∆x

(i −s)∆x

=

k

s=k−r,s=i

_

x −(s +

1

2

)∆x

_

k

s=k−r,s=i

(i −s)∆x

=

r

j=0

c

1

k,i,j

∆x

r−j

x

j

c

2

k,i,j

∆x

r

=

r

j=0

c

k,i,j

∆x

−j

x

j

.

It is also easily seen that

L

k,i

(x) = L

(r−1)−k,−i−1

(−x). (44)

13

Using (43) and (44) , we obtain

P

F

k

(x) =

k

i=k−r

i

s=k−(r−1)

f

s

∆xL

k,i

(x)

=

k

i=k−r

i

s=k−(r−1)

f

s

r

j=0

c

k,i,j

∆x

−(j−1)

x

j

P

G

k

(x) =

(r−1)−k

i=−k−1

(r−1)−k

s=i+1

−f

s

∆xL

(r−1)−k,i

(x) =

(r−1)−k

i=−k−1

(r−1)−k

s=i+1

−f

s

∆xL

k,−i−1

(−x)

=

k+1

i

∗

=k−(r−1)

(r−1)−k

s=−i

∗

+1

−f

s

∆xL

k,i

∗

−1

(−x) =

k

i=k−r

(r−1)−k

s=−i

−f

s

∆xL

k,i

(−x)

=

k

i=k−r

(r−1)−k

s=−i

−f

s

r

j=0

c

k,i,j

∆x

−(j−1)

(−x)

j

=

k

i=k−r

i

s=k−(r−1)

−f

−s

r

j=0

c

k,i,j

∆x

−(j−1)

(−x)

j

.

Moreover, diﬀerentiating the equalities above we obtain

p

k

(x) =

k

i=k−r

i

s=k−(r−1)

r

j=0

f

s

c

k,i,j

∆x

−(j−1)

jx

j−1

,

p

(r−1)−k

(x) =

k

i=k−r

i

s=k−(r−1)

r

j=0

−f

−s

c

k,i,j

∆x

−(j−1)

(−1)

j

jx

j−1

.

Deﬁning Γ

k,i,j

= jc

k,i,j

and reorganizing the indexes, we arrive at the following expressions for the

interpolating polynomials of h in the substencils S

k

and S

r−1−k

:

p

k

(x) =

r−1

j=0

_

_

k

s=k−(r−1)

_

k

i=s

Γ

k,i,j

_

f

s

∆x

j

_

_

x

j

p

(r−1)−k

(x) =

r−1

j=0

_

_

k

s=k−(r−1)

_

k

i=s

Γ

k,i,j

_

f

−s

∆x

j

(−1)

j

_

_

x

j

14

Thus, their respective coeﬃcients are given by

a

k,j

=

k

s=k−(r−1)

_

k

i=s

Γ

k,i,j

_

f

s

∆x

j

=

k

s=k−(r−1)

_

k

i=s

Γ

k,i,j

_

∞

δ=0

1

δ!

f

(δ)

0

s

δ

∆x

δ

∆x

j

=

∞

δ=0

_

_

k

s=k−(r−1)

_

k

i=s

Γ

k,i,j

_

1

δ!

f

(δ)

0

s

δ

_

_

∆x

δ−j

,

a

(r−1)−k,j

=

k

s=k−(r−1)

_

k

i=s

Γ

k,i,j

_

f

−s

∆x

j

(−1)

j

=

k

s=k−(r−1)

_

k

i=s

Γ

k,i,j

_

∞

δ=0

1

δ!

f

(δ)

0

(−1)

δ

s

δ

∆x

δ

∆x

j

(−1)

j

=

∞

δ=0

_

_

k

s=k−(r−1)

_

k

i=s

Γ

k,i,j

_

1

δ!

f

(δ)

0

(−1)

δ−j

s

δ

_

_

∆x

δ−j

.

where above we expanded f

s

in Taylor series over the origin and reorganized the summations in

terms of ∆x. We now deﬁne l = δ −j and note from (36) that δ −j ≥ 0 to arrive at

a

k,j

=

∞

l=0

_

_

k

s=k−(r−1)

_

k

i=s

Γ

k,i,j

_

s

j+l

(j + l)!

f

(j+l)

0

_

_

∆x

l

,

a

(r−1)−k,j

=

∞

l=0

_

_

k

s=k−(r−1)

_

k

i=s

Γ

k,i,j

_

s

j+l

(j + l)!

f

(j+l)

0

(−1)

l

_

_

∆x

l

.

Thus, comparing both expressions we see that

σ

k,j,l

=

k

s=k−(r−1)

k

i=s

Γ

k,i,j

s

j+l

(j + l)!

,

σ

r−1−k,j,l

=

k

s=k−(r−1)

k

i=s

Γ

k,i,j

s

j+l

(j + l)!

(−1)

l

,

satisfy σ

k,j,l

= (−1)

l

σ

r−1−k,j,l

.

Deﬁnition 5 In order to simplify the notation we deﬁne:

M

∗

=

_

M

2

_

, (45)

and

E(k, n, j

1

, j

2

, l

1

, l

2

) = C

n,j

1

,j

2

σ

k,n+j

1

,l

1

σ

k,n+j

2

,l

2

. (46)

15

Using (38), we obtain the anti-symmetry/symmetry condition

E(k, n, j

1

, j

2

, l

1

, l

2

) = (−1)

l

1

+l

2

E(r −1 −k, n, j

1

, j

2

, l

1

, l

2

). (47)

We now state and proof the new asymptotic expansion for the lower order local smoothness indi-

cators β

k

.

Theorem 6 The lower order local smoothness indicators β

k

can be written as

β

k

=

∞

M=2

M

∗

j=1

A

k,M,j

f

(j)

0

f

(M−j)

0

∆x

M

, (48)

with

A

k,M,j

= (−1)

M

A

(r−1)−k,M,j

. (49)

In addition, if j < r and M −j < r, A

k,M,j

is independent of k, that is,

A

0,M,j

= A

1,M,j

= · · · = A

r−1,M,j

.

Proof. Using the deﬁnition of a

k,j

in (37), one has

a

k,n+j

1

a

k,n+j

2

=

∞

l

1

=0

∞

l

2

=0

σ

k,n+j

1

,l

1

σ

k,n+j

2

,l

2

f

(n+j

1

+l

1

)

0

f

(n+j

2

+l

2

)

0

∆x

l

1

+l

2

.

Substituting (46) into (31), we obtain

β

k

=

r−1

n=1

r−1−n

j

1

,j

2

=0

∞

l

1

,l

2

=0

E(k, n, j

1

, j

2

, l

1

, l

2

)f

(n+j

1

+l

1

)

0

f

(n+j

2

+l

2

)

0

∆x

j

1

+j

2

+2n+l

1

+l

2

.

Fixing M

1

= n + j

1

+ l

1

and M

2

= n + j

2

+ l

2

, we can reorganize the sum as

β

k

=

∞

M

1

=1

∞

M

2

=1

Ω

M

1

,M

2

E(k, n, j

1

, j

2

, l

1

, l

2

)f

(M

1

)

0

f

(M

2

)

0

∆x

M

1

+M

2

,

where

Ω

M

1

,M

2

= {(n, j

1

, j

2

, l

1

, l

2

) ∈

r−1

_

n=1

Ψ

n

| n + j

1

+ l

1

= M

1

and n + j

2

+ l

2

= M

2

},

with

Ψ

n

= {1, . . . , r −1} ×{1, . . . , r −1 −n}

2

×N

2

.

Deﬁning M = M

1

+ M

2

, we have

β

k

=

∞

M=2

M−1

j=1

Ω

j,M−j

E(k, n, j

1

, j

2

, l

1

, l

2

)f

(j)

0

f

(M−j)

0

∆x

M

.

16

Since f

(j)

0

f

(M−j)

0

= f

(M−(M−j))

0

f

(M−j)

0

, we can reorganize the sum above as

β

k

=

∞

M=2

M

∗

j=1

Ω

j,M−j

∪ Ω

M−j,j

E(k, n, j

1

, j

2

, l

1

, l

2

)f

(j)

0

f

(M−j)

0

∆x

M

.

By deﬁning

A

k,M,j

=

Ω

j,M−j

∪ Ω

M−j,j

E(k, n, j

1

, j

2

, l

1

, l

2

),

we have

β

k

=

∞

M=2

M

∗

j=1

A

k,M,j

f

(j)

0

f

(M−j)

0

∆x

M

.

Using the property (47) and the fact that j

1

+ j

2

is always even (otherwise C

n,j

1

,j

2

= 0), we have

Ω

j,M−j

∪ Ω

M−j,j

E(k, n, j

1

, j

2

, l

1

, l

2

) =

Ω

j,M−j

∪ Ω

M−j,j

(−1)

l

1

+l

2

E(r −1 −k, n, j

1

, j

2

, l

1

, l

2

)

=

Ω

j,M−j

∪ Ω

M−j,j

(−1)

2n+j

1

+j

2

+l

1

+l

2

E(r −1 −k, n, j

1

, j

2

, l

1

, l

2

)

=

Ω

j,M−j

∪ Ω

M−j,j

(−1)

M

E(r −1 −k, n, j

1

, j

2

, l

1

, l

2

),

yielding

A

k,M,j

= (−1)

M

A

(r−1)−k,M,j

.

We can also see that if j < r and M − j < r, A

k,M,j

is independent of k. Indeed, using (31) and

(36),

β

k

=

r−1

n=1

r−1−n

j

1

,j

2

=0

C

n,j

1

,j

2

_

_

ρ

n+j

1

δ

1

=0

ρ

n+j

2

δ

2

=0

φ

2δ

1

φ

2δ

2

∆x

2δ

1

+2δ

2

(n + j

1

)!(n + j

2

)!

f

(n+j

1

+2δ

1

)

0

f

(n+j

2

+2δ

2

)

0

+

O(∆x

r−max(j

1

,j

2

)

)

_

∆x

j

1

+j

2

+2n

.

The formula above shows that the coeﬃcients associated to f

(n+j

1

+2δ

1

)

0

f

(n+j

2

+2δ

2

)

0

are independent

of k. Therefore the largest possible value for n + j

1

+ 2δ

1

(and for n + j

2

+ 2δ

2

) is

n + j

1

+ 2ρ

n+j

1

= n + j

1

+ 2

_

r −1 −(n + j

1

)

2

_

=

_

r −1 mod (r −1 −(n + j

1

), 2) = 0

r −2 mod (r −1 −(n + j

1

), 2) = 1

.

Thus if j < r and M −j < r, the coeﬃcient A

k,M,j

is independent of k.

Corollary 7 If M ≤ r, A

k,M,j

, j = 1, . . . , M

∗

is independent of the shifting parameter k.

17

We are now ready to prove Theorem 2:

Proof. Using the deﬁnition of β

k

in (48) and the symmetry condition in (49), one has

β

0

=

∞

M=2

M

∗

j=1

A

0,M,j

f

(j)

0

f

(M−j)

0

∆x

M

, β

r−1

=

∞

M=2

M

∗

j=1

A

0,M,j

f

(j)

0

f

(M−j)

0

∆x

M

(−1)

M

,

β

1

=

∞

M=2

M

∗

j=1

A

1,M,j

f

(j)

0

f

(M−j)

0

∆x

M

, β

r−2

=

∞

M=2

M

∗

j=1

A

1,M,j

f

(j)

0

f

(M−j)

0

∆x

M

(−1)

M

.

We seek to ﬁnd a linear combination of the above β

k

in such a way that the sum of all the terms

of O(∆x

k

) is zero for all k ≤ r + 1. In other words, we seek non-trivial constants c

0

, c

1

, c

r−2

, c

r−1

such that the global higher order smoothness indicator

τ

2r−1

= c

0

β

0

+ c

1

β

1

+ c

r−2

β

r−2

+ c

r−1

β

r−1

= O(∆x

r+2

). (50)

First of all, using Corollary 7, all the coeﬃcients β

0

and β

r−1

associated to the terms ∆x

M

, M ≤ r

are equal. Furthermore,

• assuming r is odd, the coeﬃcients of β

0

and β

r−1

associated to the terms ∆x

M

, M = r + 1

(M is an even number) are equal. Hence

τ

2r−1

= |β

0

−β

r−1

| = O(∆x

r+2

).

• assuming r is even, the coeﬃcients associated with the O(∆x

r+1

) (odd order) term of (β

0

+

β

r−1

) and (β

1

+ β

r−2

) are zero. Also, the coeﬃcients of these two sums are equal to

2

M

∗

j=1

A

k,M,j

and of order O(∆x

M

), for each M ≤ r. This means that the coeﬃcients

of diﬀerence of these two sums (β

0

+β

r−1

) −(β

1

+β

r−2

) of order ∆x

M

for all M ≤ r +1 are

all zero, yielding

τ

2r−1

= |β

0

−β

1

−β

r−2

+ β

r−1

| = O(∆x

r+2

).

3.2 The optimal higher order smoothness indicator τ

opt

2r−1

The linear combinations of the β

k

displayed in Theorem 2 are not unique, nor optimal, in the sense

of generating the highest possible order for τ

2r−1

. Even though it is suﬃcient to have τ

2r−1

=

O(∆x

r+2

) to guarantee the formal (2r − 1) order of accuracy of the WENO-Z scheme, the Taylor

expansions of Theorem 6 may yield a higher order. In this section we perform a deeper investigation

in the vector space generated by the linear combinations of the β

k

in order to ﬁnd the global optimal

order smoothness indicator, τ

opt

2r−1

, for each value of r ≥ 3. We state our main result in the following

proposition:

Proposition 8 The greatest lower bound for the order of τ

opt

2r−1

, M

2r−1

, r ≥ 3, is given by the

largest value of m such that

max

_

ρ

+

,

_

r

2

__

+ max

_

ρ

−

,

_

r

2

__

< r,

18

where

ρ

+

= 1 + φ

0

, ρ

−

= φ

1

, φ

k

=

_ _

m−r

2

_ __

m−r

2

_

+ 1

_

mod (m, 2) = k

_

m−r

2

_

2

mod (m, 2) = 1 −k

.

The proposition above yields the following table with the value of M

2r−1

for r up to 20:

r 2r −1 M

2r−1

r 2r −1 M

2r−1

3 5 5 12 23 17

4 7 7 13 25 18

5 9 8 14 27 19

6 11 9 15 29 21

7 13 11 16 31 22

8 15 12 17 33 23

9 17 13 18 35 24

10 19 15 19 37 25

11 21 16 20 39 27

We see from the table that when derived as a linear combination of the lower order local smoothness

indicators β

k

, the order of τ

opt

2r−1

, M

2r−1

is always smaller than the order of the scheme, (2r − 1),

for r > 4, but it is also bigger than the order of the conventional τ

2r−1

, r +2, given by Theorem 2,

for r > 3.

Let us now check by means of a numerical example the claim that the use of τ

opt

2r−1

improves over

τ

2r−1

when capturing high order structures in the solution. For instance, consider the following

test function,

f(x) = x

k

+ exp(lx), x ∈ [−1, 1], (51)

with parameters k = 8 and l = 5. The WENO-Z scheme built with the global high order smoothness

indicator τ

11

= O(∆x

8

) does not resolve well functions with information containing higher order

than O(∆x

7

). On the other hand, when equipped with τ

opt

11

= O(∆x

9

), WENO-Z is able to capture

high order information up to O(∆x

9

). As illustrated in Table I, one can see that the WENO-Z11

scheme using τ

11

is signiﬁcantly less accurate than the one using the τ

opt

11

when computing the

derivative of the test function above.

We would like now to present more clearly the issues involved in the proof of Proposition 8. To

obtain a τ

2r−1

of order m, it is suﬃcient to ﬁnd a nontrivial set of constants {c

0

, . . . , c

r−1

} such

that

r−1

k=0

c

k

β

k

= O(∆x

m

). Rearranging the summation in (48), we want

m−1

M=2

M

∗

j=1

_

r−1

k=0

c

k

A

k,M,j

_

f

(j)

0

f

(M−j)

0

∆x

M

= 0,

or

r−1

k=0

c

k

A

k,M,j

= 0, M = 2, . . . , m−1, j = 1, . . . , M

∗

.

19

WENO-Z11 Higher order τ

11

Optimal order τ

opt

11

∆x E

∞

m E

∞

m

3.12500e-02 1.5e-10 1.6e-10

1.56250e-02 2.9e-13 9.0 8.2e-14 10.9

7.81250e-03 4.8e-16 9.3 4.2e-17 10.9

3.90625e-03 9.7e-19 9.0 2.1e-20 11.0

1.95312e-03 1.4e-21 9.4 1.0e-23 11.0

9.76562e-04 4.1e-24 8.4 5.0e-27 11.0

Table I: The maximum l

∞

error, E

∞

and the order of accuracy m, O(∆x

m

), are shown with

increasing resolution ∆x for the test function f(x) = x

k

+ exp(lx) with k = 8, l = 5. The schemes

are the eleventh order WENO-Z scheme (WENO-Z11) with r = 6, p = 2 and ǫ = 1 × 10

−40

using

the higher order τ

11

and the optimal order τ

opt

11

in the deﬁnitions of the WENO-Z nonlinear weights

ω

Z

k

.

We need to ﬁnd a nontrivial solution of the system of linear equations A

m

c = 0, where c =

(c

0

, . . . , c

r−1

)

T

and

A

m

=

_

_

_

_

_

_

_

_

_

_

_

A

0,2,1

A

1,2,1

· · · A

r−1,2,1

.

.

.

.

.

. · · ·

.

.

.

A

0,r,1

A

1,r,1

· · · A

r−1,r,1

.

.

.

.

.

.

.

.

.

.

.

.

A

0,m−1,⌊

m−1

2

⌋−1

A

1,m−1,⌊

m−1

2

⌋−1

· · · A

r−1,m−1,⌊

m−1

2

⌋−1

A

0,m−1,⌊

m

2

⌋

A

1,m−1,⌊

m−1

2

⌋

· · · A

r−1,m−1,⌊

m−1

2

⌋

_

_

_

_

_

_

_

_

_

_

_

.

Note that the matrix A

m

has r column vectors and some of the row vectors of A can be linearly

dependent. Since the number of columns of A

m

is r, one needs rank(A

m

) < r to ensure that a

non-trivial solution exists.

20

To clarify the notation above we consider r = 5 and look to the Taylor expansions of the lower

order local smoothness indicators β

k

for orders less than ∆x

9

at x = x

0

:

β

0

= f

′

2

0

∆x

2

+

13

12

f

′′

2

0

∆x

4

+

_

−

2

5

f

′

0

f

(5)

0

−

1

360

f

′′

0

f

(4)

0

+

781

720

f

′′′

2

0

_

∆x

6

+

_

2

3

f

′

0

f

(6)

0

−

9

5

f

′′

0

f

(5)

0

_

∆x

7

+

_

−

13

21

f

′

0

f

(7)

0

+

1235

432

f

′′

0

f

(6)

0

−

5467

1440

f

′′′

0

f

(5)

0

+

32803

30240

f

(4)2

0

_

∆x

8

+ O(∆x

9

),

β

1

= f

′

2

0

∆x

2

+

13

12

f

′′

2

0

∆x

4

+

_

1

10

f

′

0

f

(5)

0

−

1

360

f

′′

0

f

(4)

0

+

781

720

f

′′′

2

0

_

∆x

6

+

_

−

1

12

f

′

0

f

(6)

0

+

11

60

f

′′

0

f

(5)

0

_

∆x

7

+

_

1

21

f

′

0

f

(7)

0

−

251

2160

f

′′

0

f

(6)

0

−

781

1440

f

′′′

0

f

(5)

0

+

32803

30240

f

(4)2

0

_

∆x

8

+ O(∆x

9

),

β

2

= f

′

2

0

∆x

2

+

13

12

f

′′

2

0

∆x

4

+

_

−

1

15

f

′

0

f

(5)

0

−

1

360

f

′′

0

f

(4)

0

+

781

720

f

′′′

2

0

_

∆x

6

+

_

−

1

126

f

′

0

f

(7)

0

−

53

2160

f

′′

0

f

(6)

0

+

781

1440

f

′′′

0

f

(5)

0

+

32803

30240

f

(4)2

0

_

∆x

8

+ O(∆x

9

),

β

3

= f

′

2

0

∆x

2

+

13

12

f

′′

2

0

∆x

4

+

_

1

10

f

′

0

f

(5)

0

−

1

360

f

′′

0

f

(4)

0

+

781

720

f

′′′

2

0

_

∆x

6

+

_

1

12

f

′

0

f

(6)

0

−

11

60

f

′′

0

f

(5)

0

_

∆x

7

+

_

1

21

f

′

0

f

(7)

0

−

251

2160

f

′′

0

f

(6)

0

−

781

1440

f

′′′

0

f

(5)

0

+

32803

30240

f

(4)2

0

_

∆x

8

+ O(∆x

9

),

β

4

= f

′

2

0

∆x

2

+

13

12

f

′′

2

0

∆x

4

+

_

−

2

5

f

′

0

f

(5)

0

−

1

360

f

′′

0

f

(4)

0

+

781

720

f

′′′

2

0

_

∆x

6

+

_

−

2

3

f

′

0

f

(6)

0

+

9

5

f

′′

0

f

(5)

0

_

∆x

7

+

_

−

13

21

f

′

0

f

(7)

0

+

1235

432

f

′′

0

f

(6)

0

−

5467

1440

f

′′′

0

f

(5)

0

+

32803

30240

f

(4)2

0

_

∆x

8

+ O(∆x

9

),

The corresponding matrices A

m

(omitting the zero lines) for m varying from 6 to 9 are

A

6

=

_

1 1 1 1 1

13

12

13

12

13

12

13

12

13

12

_

, A

7

=

_

_

_

_

_

_

_

_

1 1 1 1 1

13

12

13

12

13

12

13

12

13

12

−

2

5

1

10

−

1

15

1

10

−

2

5

−

1

360

−

1

360

−

1

360

−

1

360

−

1

360

781

720

781

720

781

720

781

720

781

720

_

_

_

_

_

_

_

_

,

A

8

=

_

_

_

_

_

_

_

_

_

_

_

_

_

1 1 1 1 1

13

12

13

12

13

12

13

12

13

12

−

2

5

1

10

−

1

15

1

10

−

2

5

−

1

360

−

1

360

−

1

360

−

1

360

−

1

360

781

720

781

720

781

720

781

720

781

720

2

3

−

1

12

0

1

12

−

2

3

−

9

5

11

60

0 −

11

60

9

5

_

_

_

_

_

_

_

_

_

_

_

_

_

, A

9

=

_

_

_

_

_

_

_

_

_

_

_

_

_

_

_

_

_

_

_

_

_

_

_

1 1 1 1 1

13

12

13

12

13

12

13

12

13

12

−

2

5

1

10

−

1

15

1

10

−

2

5

−

1

360

−

1

360

−

1

360

−

1

360

−

1

360

781

720

781

720

781

720

781

720

781

720

2

3

−

1

12

0

1

12

−

2

3

−

9

5

11

60

0 −

11

60

9

5

−

13

21

1

21

−

1

126

1

21

−

13

21

1235

432

−

251

2160

−

53

2160

−

251

2160

1235

432

−

5467

1440

−

781

1440

781

1440

−

781

1440

−

5467

1440

32803

30240

32803

30240

32803

30240

32803

30240

32803

30240

_

_

_

_

_

_

_

_

_

_

_

_

_

_

_

_

_

_

_

_

_

_

_

.

The rank of the matrices above are:

rank(A

6

) = 1, rank(A

7

) = 2, rank(A

8

) = 4 < r = 5, rank(A

9

) = 5.

21

Therefore, the maximum order τ

9

can achieve is given by M

9

= 8 since rank(A

9

) ≥ r = 5.

We now adopt the following strategy to prove proposition 8: For a given matrix A

m

with r column

vectors and with row vectors containing all the coeﬃcients of β

k

of order up to and equal to m−1

that does not have full rank, we increase the order by one to m and append non-zero row vectors

that correspond to the coeﬃcients of β

k

of order m to the row vectors of the old matrix A

m

to

form a new matrix A

m+1

. This process will be repeated until the newly formed matrix reaches full

rank. This process will terminate in a ﬁnite number of iterations since only ﬁnite number of row

vectors can be appended as the maximum possible order of τ

2r−1

≤ 2r − 1. These new non-zero

row vectors are represented by the submatrix B

m

as below:

B

m

=

_

_

_

_

_

A

0,m,1

A

1,m,1

· · · A

r−1,m,1

A

0,m,2

A

1,m,2

· · · A

r−1,m,2

.

.

.

.

.

.

.

.

.

.

.

.

A

0,m,⌊

m

2

⌋

A

1,m,⌊

m

2

⌋

· · · A

r−1,m,⌊

m

2

⌋

_

_

_

_

_

.

Remark 9 In the proof of Proposition 8 below, we search for a rule for the increase of the rank

of A

m

when the submatrix B

m

is appended to A

m

. Obviously, the constant vectors of B

m

do

not count, on the other hand, we will assume that all the remaining ones do count, i.e., they are

pairwise linearly independent. Although we do not have a proof for that, this is what we observed

in all the cases we tested. We are computing the worst case scenario for the rank of A

m

, or its

lowest upper bound as stated in the proposition.

Proof. First of all, it is easy to see that all the row vectors of A

r+1

are linearly dependent since

by Corollary 7, A

0,M,j

= . . . = A

r+1,M,j

for M ≤ r and thus rank(A

r+1

) = 1. Let m > r + 1 be

ﬁxed. We want to ﬁnd out how many row vectors of B

m

do not depend on k. These row vectors

will be responsible for the increase of the rank of A

m

. By Theorem 2, the elements that are not

dependent on k take the form of A

k,M,j

, with j < r and m−j < r. Therefore the number of new

row vectors that are not dependent on k is the cardinality of the set

_

j ∈

_

1, . . . ,

_

m

2

__

| j ≥ r or m−j ≥ r

_

,

which is also the rank of B

m

, rank(B

m

) = m − r. it is not always true that rank(A

m+1

) =

rank(A

m

) + rank(B

m

) and to understand the increase of the rank of A

m

, we need to analyze the

matrix A

m

.

• If r is odd, the row vectors take the form

_

A

0,m,j

A

1,m,j

· · · Ar−1

2

,m,j

· · · A

r−2,m,j

A

r−1,m,j

_

.

Using the anti-symmetry/symmetry condition (49), it is equivalent to

_

A

0,m,j

A

1,m,j

· · · Ar−1

2

,m,j

· · · (−1)

m

A

1,m,j

(−1)

m

A

0,m,j

_

.

Considering the vector space formed by all the row vectors that takes the form above, one

can see that each row vector has

_

r

2

_

free variables A

0,m,j

, . . . , Ar−1

2

,m,j

, when m is even, and

_

r

2

_

free variables, when m is odd, for in this case Ar−1

2

,m,j

= −Ar−1

2

,m,j

= 0.

22

• If r is even, there is no central term Ar−1

2

,m,j

and there are always

r

2

free variables, regardless

of the value of m.

Due to the factor (−1)

m

, the l th element of each row vector can be anti-symmetric or symmetric

to the (r − 1 − l) th element. This duality allows us to build two diﬀerent vector spaces, V

+

and

V

−

, deﬁned by

V

+

=

__

a

0

a

1

· · · a

r−1

_

∈ R

r

| a

l

= a

r−1−l

, l = 0, . . . , r −1

_

,

V

−

=

__

a

0

a

1

· · · a

r−1

_

∈ R

r

| a

l

= −a

r−1−l

, l = 0, . . . , r −1

_

,

with dimensions equal to

_

r

2

_

and

_

r

2

_

respectively (note that when r is even the dimensions are

equal).

These vector spaces are in a direct sum; that is, if we intersect A

m

with each one of these vector

spaces, we shall have two new matrices A

+

m

and A

−

m

and

rank(A

m

) = rank(A

+

m

) + rank(A

−

m

).

We will take this moment to observe that rank(A

+

r+1

) = 1 and rank(A

−

r+1

) = 0.

Now we can deﬁne rank(A

m+1

) based on rank(A

m

). Indeed, by appending the row vectors of B

m

,

we can see that if m is even (odd), only the rank of A

+

m+1

(A

−

m+1

) will increase. On the other hand,

this increase is limited by

_

r

2

_

(

_

r

2

_

) for rank(A

+

m+1

) (rank(A

−

m+1

)), because of the dimensions of

the vector spaces. Thus,

rank(A

m+1

) =

_

rank(A

−

m

) + min

_

rank(A

+

m

) + rank(B

m

),

_

r

2

__

mod (m, 2) = 0

rank(A

+

m

) + min

_

rank(A

−

m

) + rank(B

m

),

_

r

2

__

mod (m, 2) = 1

.

From this result we obtain a recurrence relation for rank(A

m

), with initial conditions rank(A

+

r+1

) =

1 and rank(A

−

r+1

) = 0, with the closed form stated in the Proposition.

For a large order (2r −1), it is diﬃcult to ﬁnd the kernel and rank of the matrix A analytically. In

this situation, symbolic computational system such as Maple can be employed. In Maple, the built-

in function RowReduce is used to ﬁnd the kernel c, rank(A) and the leading order term of order

O(∆x

M

2r−1

). Table II gives the parameters r, (2r − 1), rank(A), optimal order M

2r−1

, coeﬃcient

vector c and the leading order for the global optimal order smoothness indicator τ

opt

2r−1

, r = 3, . . . , 9

for the (2r −1) order WENO-Z scheme.

Remark 10 From this point on, we shall replace the deﬁnition of τ

2r−1

in the deﬁnition of the

nonlinear weights ω

Z

k

in the WENO-Z scheme with τ

opt

2r−1

in the rest of the paper unless stated

otherwise.

4 Critical points

In [5] it was shown that the classical ﬁfth-order WENO-JS scheme loses convergence at critical

points due to the inability of its nonlinear weights to distinguish between ﬂat and rough parts of

23

r 2r −1 rank(A) M

2r−1

c Leading Order at x

i

3 5 2 5 (−1, 0, 1) −3f

(1)

i

f

(4)

i

+ 13f

(2)

i

f

(3)

i

4 7 3 7 (−1, −3, 3, 1) 160f

(1)

i

f

(6)

i

− 1040f

(2)

i

f

(5)

i

+3124f

(3)

i

f

(4)

i

5 9 4 8 (1, 2, −6, 2, 1) −5040f

(1)

i

f

(7)

i

+ 27216f

(2)

i

f

(6)

i

−65604f

(3)

i

f

(5)

i

6 11 4 9 (−1, 0, 10, −10, 0, 1) −12096f

(1)

i

f

(8)

i

− 67324f

(2)

i

f

(7)

i

(0, −1, −2, 2, 1, 0) +44352f

(3)

i

f

(6)

i

7 13 6 11 (−1, −36, −135, 0, 135, 36, 1)

8 15 7 12 (1, 35, 99, −135, −135, 99, 35, 1)

9 17 7 13 (−1, 0, 514, 1832, 0, −1832, −514, 0, 1)

(0, −1, −17, −47, 0, 47, 17, 1, 0)

Table II: The parameter r, (2r −1), rank(A), optimal order M

2r−1

, the coeﬃcient vector c and the

leading order terms for the global optimal order smoothness indicator τ

opt

2r−1

, r = 3, . . . , 9 for the

(2r −1) order WENO-Z scheme.

the solution. This is due to the normalization of the nonlinear weights, since on both situations

their relative sizes may show large variations, although only in the latter they are large in absolute

value. In [10], it was shown that the use of the higher order smoothness indicator τ

2r−1

into the

formula for the WENO-Z weights improved the ability of the scheme to detect such situations. This

happens because the use of τ

2r−1

through the ratio

_

τ

2r−1

βk + ǫ

_

p

inserts a measure of relativeness of

the sizes of the lower order smoothness indicators β

k

. Moreover, as it was shown in [10], this also

allows a speed up of the rate of convergence at critical points just by increasing the value of the

power parameter p.

We now perform a numerical experiment to illustrate the behavior of the several schemes in the

presence of smooth solutions containing critical points. The following computations were done

with quadruple precision with 34 signiﬁcance digits to avoid the contamination of roundoﬀ errors

in numerical results with decreasing ∆x. Unless explicitly indicated otherwise, we use the standard

value of ǫ = 10

−40

.

Consider the following test function,

f(x) = x

k

exp(lx), x ∈ [−1, 1], (52)

in which its ﬁrst k −1 derivatives f

(j)

(0) = 0, j = 0, . . . , k −1. That is, this function has a critical

point of order n

cp

= k −1 at x = 0.

We show in Table III the convergence rates for the classical WENO-JS, the mapped WENO-M and

for WENO-Z at a critical point of second order, i.e., k = 3. We used the ﬁxed value of p = 2 for

WENO-JS and WENO-Z, since their rates of convergence do not change when p varies. On the

other hand, increasing the value of p from 2 to 3 allows WENO-Z to recover the formal order of

accuracy of the scheme. Table IV shows that this is indeed a general behavior, that is, increasing

p, increases the rate of convergence of WENO-Z and, particularly, when p = r −1 and n

cp

< r −1,

24

∆x WENO-JS7 WENO-M7 WENO-Z7 (p = 2) WENO-Z7 (p = 3)

1.00000e-01 1.2e-03 3.5e-04 4.1e-04 8.1e-04 0.0

5.00000e-02 3.0e-05 5.3 1.9e-06 7.5 2.8e-06 7.2 3.4e-06 7.9

2.50000e-02 8.6e-07 5.1 8.6e-09 7.8 1.8e-08 7.3 7.7e-09 8.8

1.25000e-02 2.6e-08 5.1 9.8e-11 6.5 1.3e-10 7.1 2.3e-11 8.4

6.25000e-03 8.0e-10 5.0 2.1e-12 5.6 1.0e-12 7.0 8.3e-14 8.1

3.12500e-03 4.0e-11 4.3 3.7e-14 5.8 1.2e-14 6.4 3.4e-16 7.9

1.56250e-03 2.2e-12 4.2 6.0e-16 5.9 1.9e-16 6.0 2.7e-18 7.0

7.81250e-04 1.3e-13 4.1 9.7e-18 6.0 2.9e-18 6.0 2.1e-20 7.0

Table III: Rate of convergence at a second order critical point (n

cp

= 2) for the seventh order

(r = 4) WENO-JS, WENO-M and WENO-Z schemes.

the rate of convergence of WENO-Z can always be recovered to the fullest order. For a detailed

analysis of this accuracy enhancement of WENO-Z when p increases, see [10].

Remark 11 It is shown in [5] that the formal order of the scheme, (2r −1), can also be recovered

with WENO-M by further applications of the mapping, if n

cp

< r −1, however, as it was shown in

[10], the use of the mapping incurs on a signiﬁcant increase of the computational cost.

r 2r −1 n

cp

JS M Z (p = 1) Z (p = 2) Z (p = r −1)

4 7 0 7 7 7 7 7

1 5 7 7 7 7

2 4 6 5 6 7

3 3 3 3 3 3

5 9 0 9 9 9 9 9

1 7 9 9 9 9

2 6 9 8 9 9

3 5 7 6 7 9

4 4 4 4 4 4

6 11 0 11 11 11 11 11

1 9 11 11 11 11

2 8 11 11 11 11

3 7 11 9 11 11

4 6 8 7 8 11

5 5 5 5 5 5

Table IV: Rates of convergence at critical points of increasing order n

cp

for (2r−1) order WENO-JS,

WENO-M and WENO-Z schemes. A ﬁxed constant ǫ = 1 ×10

−40

is used.

Critical points have become a point of discussion since when in [5] it was shown that the standard

value of the parameter ǫ = 10

−6

was in fact hiding the loss of accuracy of WENO-JS. At critical

points, the β

k

have much smaller sizes and such an ǫ dominates their relative variations towards a

central upwind ﬁfth order scheme. When a smaller value for ǫ is used the order of accuracy of the

scheme degrades to third order, since the variation of the β

k

is understood as the indication of a

high gradient.

WENO-M was presented as a ﬁx to this situation, for it used a mapping that corrected the weights

of WENO-JS and recovered the formal order of the scheme even when using very small values of

25

ǫ. The numerical results with WENO-M were superior at shock problems and this was credited to

the improvement of the weights at critical points. Nevertheless, it is well known that at problems

with shocks one cannot expect better order of accuracy than O(1), and this dominates any eventual

increase of convergence order at critical points. Thus, in [10] it was shown that the extra sharpness

obtained at discontinuities by WENO-M was in fact due to its smaller dissipation. A smaller

dissipation is achieved when close to discontinuities the scheme generates a set of nonlinear weights

that provide a more centralized scheme and we will see in the next section that this is the case for

the improved results of WENO-Z and WENO-M over WENO-JS.

Nevertheless, ﬂat and high gradients regions of the solution can be distinguished by the absolute

sizes of the β

k

and a conditional statement could be used to improve the WENO schemes conver-

gence at critical points. However, a conveniently size of ǫ can be chosen in order to play the role of

this conditional. Based on the structure disclosed by their Taylor series in the last section, such an

ǫ could be of size ∆x

2

, since at critical points the sizes of the β

k

will be smaller than that. Table

V below show numerical results for critical points of order r −1, that is, k = r −1 at (52), for the

WENO-Z schemes. Note that the formal order of accuracy is recovered and the same occurs for

WENO-JS and WENO-M (not shown).

∆x WENO-Z5 WENO-Z7 WENO-Z9 WENO-Z11

1.00000e-01 2.3e-05 2.7e-05 2.1e-05 6.3e-05

5.00000e-02 5.8e-07 5.3 2.2e-09 13.6 3.1e-09 12.7 3.5e-09 14.2

2.50000e-02 1.9e-08 4.9 3.6e-11 5.9 8.8e-14 15.1 2.5e-16 23.7

1.25000e-02 6.0e-10 5.0 2.8e-13 7.0 1.8e-16 9.0 1.4e-19 10.8

6.25000e-03 1.9e-11 5.0 2.2e-15 7.0 3.5e-19 9.0 6.8e-23 11.0

3.12500e-03 5.9e-13 5.0 1.7e-17 7.0 6.8e-22 9.0 3.3e-26 11.0

1.56250e-03 1.9e-14 5.0 1.4e-19 7.0 1.3e-24 9.0 1.6e-29 11.0

7.81250e-04 5.8e-16 5.0 1.1e-21 7.0 2.6e-27 9.0 7.9e-33 11.0

Table V: Rates of convergence at a critical point of order n

cp

= r − 1 for (2r − 1) order WENO-Z

schemes with power parameter p = 1 and ǫ = ∆x

2

.

Remark 12 The critical points issue has been the departure point of several investigations on

improved sets of WENO nonlinear weights. However, the literature has yet to provide meaningful

examples with shocks where the distinguished treatment of critical points provides any substantial

improvement to the quality of the numerical solution. In the numerical experiments of next section

we look closer to the issue of the numerical dissipation as a result of qualitative diﬀerences among

the distinct sets of nonlinear weights of the WENO schemes studied in this article. Thus, we will

pay closer attention to the power parameter p and use the satisfactory variable value of ǫ = ∆x

r−1

for all the numerical experiments. This is a compromise value that avoids the original predominance

of a ﬁxed value of ǫ and also sets a lower bound for the ratio of the smoothness indicators.

5 Numerical Results

In this section We compare the sets of nonlinear weights generated by the three WENO schemes

discussed in this article. We will see that the ﬁnal amount of dissipation observed at the numerical

solutions is a result of the sizes of the weights attributed to discontinuous substencils by each

26

scheme. We also show numerical experiments with the one dimensional scalar advection equation

with initial condition consisted of a triangle, a Gaussian, a square and an ellipse functions, the one

dimensional Euler Equations of compressible gas dynamics with Riemann initial condition such

as the Lax problem and the one dimensional Mach 3 shock-density wave interaction. We end

this section showing numerical simulations of the two dimensional Mach 10 double-Mach shock

reﬂection problem using high orders WENO-Z schemes with increasing resolutions. The time

dependent problems are all solved via the third order Runge-Kutta TVD scheme with CFL = 0.45.

5.1 Discontinuities and nonlinear weights

We now compare the diﬀerent sets of normalized nonlinear weights ω

k

generated by WENO-JS,

WENO-Z and WENO-M. We ﬁx ǫ = 1 ×10

−40

in the discussion of this section.

Consider the following discontinuous function

g(x) =

_

−sin(πx) −

x

3

2

, −1 < x < 0

1 −sin(πx) −

x

3

2

, 0 ≤ x ≤ 1

, g(x) = g(x −2), (53)

with a single discontinuity located at x = 0. In ﬁgure 2, the nonlinear weights ω

k

of the ﬁfth

order WENO-JS, WENO-Z and WENO-M schemes with power parameters p = 1 and p = 2 are

displayed.

x = −0.0015 x = −0.005

r = 3, p = 1 WENO-JS5 WENO-Z5 WENO-M5 WENO-JS5 WENO-Z5 WENO-M5

ω

0

1.427e-01 1.426e-01 1.271e-01 9.946e-01 9.919e-01 9.842e-01

ω

1

8.570e-01 8.567e-01 8.713e-01 4.509e-03 6.303e-03 1.190e-02

ω

2

3.060e-04 6.116e-04 1.671e-03 9.064e-04 1.808e-03 3.886e-03

r = 3, p = 2 WENO-JS5 WENO-Z5 WENO-M5 WENO-JS5 WENO-Z5 WENO-M5

ω

0

1.426e-01 1.426e-01 1.272e-01 9.999e-01 9.999e-01 9.999e-01

ω

1

8.574e-01 8.574e-01 8.728e-01 3.425e-06 3.979e-06 3.063e-06

ω

2

2.187e-07 4.376e-07 1.198e-06 2.768e-07 5.554e-07 1.200e-06

Table VI: Nonlinear weights ω

k

, k = 0, 1, 2 of the ﬁfth order WENO-JS, WENO-Z and WENO-M

schemes with power parameters p = 1 and p = 2 at locations x = −0.0015 and x = −0.005. The

number of uniformly spaced grid points used is N = 200.

Analyzing the results shown at Table VI and at Figure 2, for the ﬁfth order WENO schemes, we

see that:

1. At x = −0.015, where the substencils S

0

and S

1

are smooth and only S

2

is nonsmooth:

• WENO-Z assigns a larger weight ω

2

for the nonsmooth substencil S

2

than WENO-JS,

while WENO-M assigns an even larger one as evidenced in the Table.

• Note also that ω

Z

2

= 2ω

2

; a straightforward computation justifying this fact is given in

[10].

27

WENO-JS5 WENO-Z5 WENO-M5

p = 1

-0.03 -0.01 0.01 0.03

10

-8

10

-7

10

-6

10

-5

10

-4

10

-3

10

-2

10

-1

10

0

ω

2

ω

0

ω

1

-0.03 -0.01 0.01 0.03

10

-8

10

-7

10

-6

10

-5

10

-4

10

-3

10

-2

10

-1

10

0

ω

2

ω

0

ω

1

-0.03 -0.01 0.01 0.03

10

-8

10

-7

10

-6

10

-5

10

-4

10

-3

10

-2

10

-1

10

0

ω

2

ω

0

ω

1

p = 2

-0.03 -0.01 0.01 0.03

10

-8

10

-7

10

-6

10

-5

10

-4

10

-3

10

-2

10

-1

10

0

ω

2

ω

0

ω

1

-0.03 -0.01 0.01 0.03

10

-8

10

-7

10

-6

10

-5

10

-4

10

-3

10

-2

10

-1

10

0

ω

2

ω

0

ω

1

-0.03 -0.01 0.01 0.03

10

-8

10

-7

10

-6

10

-5

10

-4

10

-3

10

-2

10

-1

10

0

ω

2

ω

0

ω

1

Figure 2: Nonlinear weights ω

k

, k = 0, 1, 2 of the ﬁfth order (r = 3) WENO-JS, WENO-Z and

WENO-M schemes with power parameters p = 1 and p = 2.

• Nevertheless, not always the weights of WENO-M are larger than the corresponding

ones of WENO-Z and WENO-JS; for instance, ω

0

= ω

Z

0

> ω

M

0

. This occurs because the

increase of the smallest weight is compensated due to the normalization of the weights.

2. At x = −0.005, only the substencils S

0

is smooth, whileS

1

and S

2

both contain the disconti-

nuity:

• Here, ω

2

is the smallest weight since the local lower order polynomial approximation of

g(x) at x = 0 is computed through an extrapolation of the values in S

2

. Note once again

that this smallest weight, ω

2

, is signiﬁcantly larger with WENO-Z than in WENO-JS,

double the value, and also once again is even larger with WENO-M.

We see that WENO-M assigns larger weights than WENO-Z for discontinuous substencils for the

same value of p. Also, in its turn, WENO-Z assigns larger weights to discontinuous substencils

than WENO-JS for the same value of p. Moreover, a greater value of p decreases the absolute sizes

of the weights assigned to discontinuous substencils by all three schemes. Since a bigger weight for

a discontinuous stencil contributes to a more central upwind approximation, we may conclude that

there is a hierarchy of dissipation that puts WENO-M as the least dissipative scheme and WENO-JS

as the most dissipative one, while WENO-Z assumes an intermediary position. Also, increasing the

value of p, decreases the absolute sizes of the weights assigned to discontinuous substencils, making

28

for a more lateral linear combination of the substencils and, therefore, increases dissipation. We

remark that this is a general behavior that does not depend on the order of the scheme; for instance,

see Figure 3 where we show analogous graphs for the nonlinear weights of the seventh order case.

WENO-JS7 WENO-Z7 WENO-M7

p = 1

-0.05 -0.03 -0.01 0.01 0.03 0.05

10

-10

10

-8

10

-6

10

-4

10

-2

10

0

ω

2

ω

3

ω

1

ω

0

-0.05 -0.03 -0.01 0.01 0.03 0.05

10

-10

10

-8

10

-6

10

-4

10

-2

10

0

ω

2

ω

3

ω

1

ω

0

-0.05 -0.03 -0.01 0.01 0.03 0.05

10

-10

10

-8

10

-6

10

-4

10

-2

10

0

ω

2

ω

3

ω

1

ω

0

p = 2

-0.05 -0.03 -0.01 0.01 0.03 0.05

10

-10

10

-8

10

-6

10

-4

10

-2

10

0

ω

2

ω

3

ω

1

ω

0

-0.05 -0.03 -0.01 0.01 0.03 0.05

10

-10

10

-8

10

-6

10

-4

10

-2

10

0

ω

2

ω

3

ω

1

ω

0

-0.05 -0.03 -0.01 0.01 0.03 0.05

10

-10

10

-8

10

-6

10

-4

10

-2

10

0

ω

2

ω

3

ω

1

ω

0

Figure 3: Nonlinear weights ω

k

, k = 0, 1, 2, 3 of the seventh order (r = 4) WENO-JS, WENO-Z

and WENO-M schemes, with power parameters p = 1 and p = 2.

5.2 Linear Advection

Consider the one dimensional linear wave equation,

∂u

∂t

+

∂u

∂x

= 0, x ∈ [0, 1),

with an initial condition given by

u(x, t = 0) =

_

¸

¸

¸

¸

¸

_

¸

¸

¸

¸

¸

_

1

6

[G(x, β, z −δ) + 4G(x, β, z) + G(x, β, z + δ)] , x ∈ [−0.8, −0.6]

1 , x ∈ [−0.4, −0.2]

1 −[10(x −0.1)] , x ∈ [0, 0.2]

1

6

[F(x, α, a −δ) + 4F(x, α, a) + F(x, α, a + δ )] , x ∈ [0.4, 0.6]

0 , else

G(x, β, z) = e

−β(x−z)

2

, F(x, α, a) =

_

max(1 −α

2

(x −a)

2

, 0),

29

where z = −0.7, δ = 0.005, β =

log 2

36δ

2

, a = 0.5 and α = 10. This is the one dimensional scalar

linear advection problem with an initial condition consisting of a smooth Gaussian, a discontinuous

Heavside function, a piecewise linear triangle function and a smooth elliptic function. Periodical

boundary conditions are imposed on the two ends of the domain.

We compute the numerical solutions for all three schemes of order 11 (r = 6) with N = 200

uniformly spaced grid points and ﬁnal time T = 8. The third-order Runge-Kutta TVD scheme

uses an adjusted time step ∆t = CFL × ∆x

2r−1

3

in order to maintain the convergence rate of the

underlying spatial WENO scheme. The results in Figure 4 show the computed solutions and the

absolute point-wise errors for p = 1, 2 and r − 1. Symbols represent the numerical solution while

the lines indicate the absolute point-wise error at each grid point. The black solid line is the exact

solution.

We interpret the behavior of the solution by looking at the error. A monotone error curve away

from the discontinuity implies a non-oscillatory solution; otherwise, one may infer the presence of

oscillations. Also, the oscillations are very small if variations in the error curve are close to the

bottom of the log scale. In summary, one can observe that

• for p = 1, all three schemes do not have enough dissipation to simulate the advection of the

square wave without oscillations.

• for p = 2, WENO-JS and WENO-Z achieve the ENO property for the square wave, however

WENO-M is still oscillatory.

• for p = r − 1, all schemes show no oscillations, and their solutions are very similar in what

regards the sharp approximation of corners.

• increasing the number of grid points will reduce the error away from the discontinuities in

the solution and in its derivatives (not shown).

5.3 One dimensional Euler Equations: The Lax Problem

This same behavior is observed in Figure 5 where we used the ninth order, r = 5, WENO-JS,

WENO-Z and WENO-M schemes, for the numerical simulation of the Lax problem. Once again

we note that WENO-M requires the highest value of p among all three schemes in order to become

non-oscillatory.

5.4 One dimensional Shock-density wave interaction

In the Lax problem, the solution is a piece-wise linear function which does not fully justify the cost

of using the high order reconstruction process of the WENO schemes. The standard one dimensional

shock-density wave interaction is generally preferred, since the solution of this problem consist of a

main shock, a high gradient smooth post-shock region and multiple shocklets that develop in a later

time, all of these requiring high order schemes in order to be eﬃciently and accurately represented.

Details of the setup of this problem can be found in [2].

30

(r = 6, p = 1) (r = 6, p = 2)

x

E

r

r

o

r

E

x

a

c

t

S

o

l

u

t

i

o

n

-1 -0.5 0 0.5 1

10

-9

10

-7

10

-5

10

-3

10

-1

10

1

10

3

0

0.2

0.4

0.6

0.8

1

Exact Solution

WENO-M

WENO-JS

WENO-Z

x

E

r

r

o

r

E

x

a

c

t

S

o

l

u

t

i

o

n

-1 -0.5 0 0.5 1

10

-9

10

-7

10

-5

10

-3

10

-1

10

1

10

3

0

0.2

0.4

0.6

0.8

1

Exact Solution

WENO-M

WENO-JS

WENO-Z

(r = 6, p = r −1)

x

E

r

r

o

r

E

x

a

c

t

S

o

l

u

t

i

o

n

-1 -0.5 0 0.5 1

10

-9

10

-7

10

-5

10

-3

10

-1

10

1

10

3

0

0.2

0.4

0.6

0.8

1

Exact Solution

WENO-M

WENO-JS

WENO-Z

Figure 4: Numerical solutions of the linear advection equation at ﬁnal time t = 8 computed by the

eleventh order (r = 6) WENO-JS, WENO-M and WENO-Z with power parameters p = 1, 2, r −1.

The number of grid points used is N = 200. The symbols indicate the numerical solution at the

grid points. The lines show the absolute point-wise error at each grid point. The solid black line is

the exact solution.

31

(r = 5, p = 1) (r = 5, p = 2)

x

-4 -2 0 2 4

0.2

0.4

0.6

0.8

1

1.2

1.4

ρ

Exact solution

WENO-JS

WENO-M

WENO-Z

x

-4 -2 0 2 4

0.2

0.4

0.6

0.8

1

1.2

1.4

ρ

Exact solution

WENO-JS

WENO-M

WENO-Z

(r = 5, p = r −1)

x

-4 -2 0 2 4

0.2

0.4

0.6

0.8

1

1.2

1.4

ρ

Exact solution

WENO-JS

WENO-M

WENO-Z

Figure 5: Numerical solution of the Lax problem as computed by the ninth order (r = 5) WENO-JS,

WENO-M and WENO-Z schemes with power parameters p = 1, 2, r −1.

32

(r = 3, p = 1) (r = 3, p = 2) (r = 3, p = r −1)

x

R

h

o

-4 -2 0 2 4

1

2

3

4

Exact Solution

WENO-JS

WENO-M

WENO-Z

x

R

h

o

-4 -2 0 2 4

1

2

3

4

Exact Solution

WENO-JS

WENO-M

WENO-Z

x

R

h

o

-4 -2 0 2 4

1

2

3

4

Exact Solution

WENO-JS

WENO-M

WENO-Z

(r = 4, p = 1) (r = 4, p = 2) (r = 4, p = r −1)

x

R

h

o

-4 -2 0 2 4

1

2

3

4

Exact Solution

WENO-JS

WENO-M

WENO-Z

x

R

h

o

-4 -2 0 2 4

1

2

3

4

Exact Solution

WENO-JS

WENO-M

WENO-Z

x

R

h

o

-4 -2 0 2 4

1

2

3

4

Exact Solution

WENO-JS

WENO-M

WENO-Z

(r = 5, p = 1) (r = 5, p = 2) (r = 5, p = r −1)

x

R

h

o

-4 -2 0 2 4

1

2

3

4

Exact Solution

WENO-JS

WENO-M

WENO-Z

x

R

h

o

-4 -2 0 2 4

1

2

3

4

Exact Solution

WENO-JS

WENO-M

WENO-Z

x

R

h

o

-4 -2 0 2 4

1

2

3

4

Exact Solution

WENO-JS

WENO-M

WENO-Z

(r = 6, p = 1) (r = 6, p = 2) (r = 6, p = r −1)

x

R

h

o

-4 -2 0 2 4

1

2

3

4

Exact Solution

WENO-JS

WENO-M

WENO-Z

x

R

h

o

-4 -2 0 2 4

1

2

3

4

Exact Solution

WENO-JS

WENO-M

WENO-Z

x

R

h

o

-4 -2 0 2 4

1

2

3

4

Exact Solution

WENO-JS

WENO-M

WENO-Z

Figure 6: Numerical solution of the shock-density wave interaction as computed by the WENO-JS,

WENO-M and WENO-Z schemes of order 2r −1 = 5, 7, 9, 11 with power parameters p = 1, 2, r −1.

The number of grid points used is N = 200.

33

(r = 6, p = 1) (r = 6, p = 2) (r = 6, p = r −1)

x

R

h

o

-4 -2 0 2 4

1

2

3

4

Exact Solution

WENO-JS

WENO-M

WENO-Z

x

R

h

o

-4 -2 0 2 4

1

2

3

4

Exact Solution

WENO-JS

WENO-M

WENO-Z

x

R

h

o

-4 -2 0 2 4

1

2

3

4

Exact Solution

WENO-JS

WENO-M

WENO-Z

Figure 7: Numerical solution of the shock-density wave interaction as computed by the WENO-JS,

WENO-M and WENO-Z schemes of order 2r −1 = 11 with power parameters p = 1, 2, r −1. The

number of grid points used is N = 250.

Similarly, as in the examples before, one can observe from ﬁgures 6 and 7, using N = 200 and

N = 250 number of uniformly spaced grid points, respectively, that the WENO-JS scheme is

the most dissipative one, WENO-M scheme is the least one, and WENO-Z scheme occupies an

intermediary position for any ﬁxed order (2r −1) and power parameter p = 1, 2, r −1. WENO-M

scheme provides a better resolution of the small scales high frequency structures behind the main

shock with respect to WENO-Z, and the same can be said about WENO-Z with regards to WENO-

JS; and all of this is due to their distinct levels of dissipation, resulting from their distinct treatment

of discontinuous substencils. Nevertheless, notice that all the WENO solutions are very similar once

a suﬃcient number of grid points (N = 250) is used and this is a point against WENO-M, since its

CPU costs, in this one dimensional test case, are approximately 20% to 30% more expensive than

WENO-JS and WENO-Z schemes, which have very similar computational costs (see [10]).

5.5 Double-Mach shock reﬂection problem

Finally, we apply the high order conservative characteristic-wise WENO-Z ﬁnite diﬀerence scheme

to the two dimensional double-Mach shock reﬂection problem [19] where a vertical shock wave

moves horizontally into a wedge that is inclined by some angle. The domain of the problem is

[0, 4] × [0, 1] and the shock moves diagonally at Mach 10, making an angle of 60 degrees with

the horizontal axis. The equations are the two dimensional Euler equations (γ = 1.4) and initial

conditions are given by

Q = (ρ, u, v, P) =

_

¸

_

¸

_

_

8, 8.25 cos

π

6

, −8.25 sin

π

6

, 116.5

_

, x < x

0

+

y

√

3

(1.4, 0, 0, 1.0) , x ≥ x

0

+

y

√

3

,

with x

0

=

1

6

. Boundary conditions at x = 0 are inﬂow, with post-shock values as above, and at

x = 4 we have outﬂow boundary conditions with

∂Q

∂x

= 0. At y = 0, reﬂecting boundary conditions

are applied to the interval [x

0

, 4], in the x-axis, simulating the wedge: (ρ

y

, u

y

, v, P

y

) = (0, 0, 0, 0)

for x

0

≤ x < 4 and y = 0. At the upper boundary, y = 1, the ﬂow has to be imposed such that

there is no interaction with the moving shock. The exact location of the shockwave at y = 1, at

34

instant t, is given by s(t) = x

0

+

(1 + 20t)

√

3

; we set post-shock and pre-shock conditions before and

after this location, given respectively by:

Q|

y=1

= (ρ, u, v, P) |

y=1

=

_ _

8, 8.25 cos

π

6

, −8.25 sin

π

6

, 116.5

_

, 0 ≤ x < s(t)

(1.4, 0, 0, 1.0) , s(t) ≤ x ≤ 4

.

Numerical results of this problem are well documented in the literature (see also ﬁgure 8) and

further details can be found in [19].

The global structure of the solution at time t = 0.2 is in general very similar across diﬀerent

schemes, resolutions and parameters. However, the resolving power of the higher order WENO

schemes can be determined by the number of small vortices that can be captured along the slip line

and the wall jet behind the lower half of the right moving shock by the underlying scheme with a

given resolution.

Figure 8: Density contours of the double-Mach shock reﬂection as computed by the eleventh order

WENO-Z scheme at time t = 0.2.

In Figure 9 we display the region around the double Mach stems in order to observe the numerical

solutions of the WENO-Z scheme of orders 2r − 1 = 5, 9, 11 using three diﬀerent uniform meshes

with resolutions 400 ×100 , 800 ×200 and 1600 ×400 at time t = 0.2. Here, the power parameter

p = r −1 is used. For this problem, p ≤ 2 is unstable for WENO-M (r > 3) and WENO-Z (r > 4)

and this can be attributed to the insuﬃcient dissipation of the schemes. We see from the ﬁgure

that one may obtain similar results by increasing the order of approximation and decreasing the

number of points in the spatial discretization. Additionally, as it was shown in [4], the CPU cost

of increasing the order is smaller than the one of increasing the spatial resolution.

6 Conclusions

We extended the new WENO schemes introduced in [10] to rates of convergence higher than

5 by providing a formula in closed form for the higher order smoothness indicators τ

2r−1

as a

simple linear combination of the lower order smoothness indicators β

k

. Another formula for the

maximum order of convergence that such linear combinations can achieve was also provided. Both

results were obtained from a thorough study of the properties of the Taylor expansions of the lower

order smoothness indicators departing from the symmetric structure of their underlying Lagrangian

interpolating polynomials with respect to the geometrical disposition of the global and the local

WENO stencils. We also discussed the lack of convergence at critical points of smooth solutions and

revisited the ǫ issue of the classical WENO-JS weights showing that for practical matters it may

work as an implicit conditional that distinguishes ﬂat regions from discontinuities. The formation

35

N = 400 ×100

(r = 3) (r = 5) (r = 6)

2.7 2.8 2.9 3 3.1 3.2 3

0

0.1

0.2

0.3

0.4

0.5

2.7 2.8 2.9 3 3.1 3.2 3

0

0.1

0.2

0.3

0.4

0.5

2.7 2.8 2.9 3 3.1 3.2 3

0

0.1

0.2

0.3

0.4

0.5

N = 800 ×200

(r = 3) (r = 5) (r = 6)

2.7 2.8 2.9 3 3.1 3.2 3

0

0.1

0.2

0.3

0.4

0.5

2.7 2.8 2.9 3 3.1 3.2 3

0

0.1

0.2

0.3

0.4

0.5

2.7 2.8 2.9 3 3.1 3.2 3

0

0.1

0.2

0.3

0.4

0.5

N = 1600 ×400

(r = 3) (r = 5) (r = 6)

2.7 2.8 2.9 3 3.1 3.2 3

0

0.1

0.2

0.3

0.4

0.5

2.7 2.8 2.9 3 3.1 3.2 3

0

0.1

0.2

0.3

0.4

0.5

2.7 2.8 2.9 3 3.1 3.2 3

0

0.1

0.2

0.3

0.4

0.5

Figure 9: Density contours of the double-Mach shock reﬂection problem as computed by the high

order WENO-Z scheme of orders 2r − 1 = 5, 9, 11, with resolutions at 400 × 100, 800 × 200 and

1600 ×400 at time t = 0.2. The power parameter p = r −1 is used in these simulations.

36

of the distinct sets of nonlinear weights for the WENO-JS, WENO-Z and WENO-M were analyzed

and numerical arguments were provided to explain their inﬂuence in the distinct numerical features

of each scheme and their changes with the increasing of the power parameter p, we concluded that

regarding dissipation WENO-Z occupies an intermediary position between WENO-JS and WENO-

M. Numerical results with the system of the Euler Equations in 1D and 2D were presented to show

that the new high order WENO-Z schemes perform well at the treatment of numerical solutions

containing both discontinuities and high order smooth structures.

7 Acknowledgments

The ﬁrst and second authors have been supported by CNPq, grants 300315/98-8 and FAPERJ

E-26/111.564/2008. The third author (Don) would like to thank the support provided by the FRG

grant FRG08-09-II-12 from Hong Kong Baptist University and the RGC grant HKBU-092009 from

Hong Kong Research Grants Council. The author would also like to thanks the Departamento de

Matem´atica Aplicada, IM-UFRJ, for hosting his visit during the course of the research.

References

[1] D. Balsara and C. W. Shu, Monotonicity preserving weighted essentially non-oscillatory

schemes with increasingly high order of accuracy, J. Comp. Phys. 160, pp. 405–452 (2000)

[2] G. S. Jiang and C. W. Shu, Eﬃcient Implementation of Weighted ENO Schemes, J. Comp.

Phys. 126, pp. 202–228 (1996)

[3] C. W. Shu, Essentially Non-Oscillatory and Weighted Essentially Non-Oscillatory Schemes

for Hyperbolic Conservation Laws, NASA/CR-97-206253 ICASE Report No. 97-65

[4] J. Shi, Y. T. Zhang and C. W. Shu, Resolution of high order WENO schemes for complicated

ﬂow structures, J. Comp. Phys. 186, pp. 690–696 (2003)

[5] A. K. Henrick, T. D. Aslam and J. M. Powers, Mapped weighted essentially non-oscillatory

schemes: Achieving optimal order near critical points, J. Comp. Phys. 207, pp. 542–567 (2005)

[6] P. L. Roe, Approximated Riemann solvers, parameter vectors, and diﬀerence schemes, J. Com-

put. Phys. 43, pp. 357–372 (1981)

[7] A. Harten, High resolution schemes for hyperbolic conservation laws, J. Comput. Phys. 49,

pp. 357–393 (1983)

[8] C. W. Shu and S. Osher, Eﬃcient Implementation of Essentially Non-oscillatory Shock-

Capturing Schemes, J. Comp. Phys. 77, pp. 439–471 (1988)

[9] C. W. Shu and S. Osher, Eﬃcient Implementation of Essentially Non-oscillatory Shock-

Capturing Schemes, II, J. Comp. Phys. 83, No. 1, pp. 32–78 (1989)

[10] R. Borges, M. Carmona, B. Costa and W. S. Don, An improved weighted essentially non-

oscillatory scheme for hyperbolic conservation laws, J. Comp. Phys. 227 No. 6, pp. 3101–3211

(2008)

37

[11] M. Latini, O. Schilling and W. S. Don, Eﬀects of order of WENO ﬂux reconstruction and spatial

resolution on reshocked two-dimensional Richtmyer-Meshkov instability, J. Comp. Phys. 221

No. 2, pp. 805–836 (2007)

[12] B. Costa and W. S. Don, High Order Hybrid Central-WENO Finite Diﬀerence Methods for

Conservation Laws, J. Comput. Appl. Math. 204 No. 2, pp. 209–218 (2007)

[13] B. Costa and W. S. Don, Multi-Domain Hybrid Spectral-WENO Methods for Hyperbolic Con-

servation Laws, J. Comput. Phys. 224 No. 2, pp. 970–991 (2007)

[14] G. A. Gerolymos, D. Schal and I. Vallet, Very High Order WENO Schemes, J. Comput. Phys.

228, pp. 8481–8524 (2009)

[15] G. Jacobs and W. S. Don, A High Order WENO-Z Finite Diﬀerence Methods Based Particle-

Source-in-Cell method for Particle-Laden Flows with Shock, J. Comput. Phys. 228 No. 5, pp.

1365–1379 (2009)

[16] G. B. Jacobs, W. S. Don and T. Dittmann, High-Order Resolution of Eulerian-Lagrangian

Simulations of Particle Dispersion in the Accelerated Flow behind a Moving Shock, Theoretical

and Computational Fluid Dynamics, accepted, (2010)

[17] J. H. Jung and W. S. Don, Collocation Methods for Hyperbolic Partial Diﬀerential Equations

with Singular Sources, Advances in Applied Mathematics and Mechanics 1 No. 6, pp. 769–780

(2009)

[18] X. Y. Hu, N. A. Adams and G. Iaccarino, On the HLLC Riemann solver for interface inter-

action in compressible multi-ﬂuid ﬂow, J. Comput. Phys. 22, pp. 6572–6589 (2009)

[19] P. Woodward and P. Collela, The numerical simulation of two dimensional ﬂuid ﬂow with

strong shocks, J. Comp. Phys. 54, pp. 115–173 (1984)

38

1

Introduction

The weighted essentially non-oscillatory conservative ﬁnite diﬀerence schemes (WENO) [1, 2] are a popular choice of numerical methods for solving compressible ﬂows modeled by means of hyperbolic conservation laws in the form ∂u + ∇ · F(u) = 0, ∂t (1)

and in the presence of shocks and small scales structures. WENO schemes owe their success to the use of a dynamic set of stencils where a nonlinear convex combination of lower order polynomials adapts either to a higher order polynomial approximation at smooth parts of the solution, or to an upwind spatial discretization that avoids interpolation across discontinuities and provides the necessary dissipation for shock capturing. It is an evolution of the Essentially Non-Oscillatory (ENO) schemes, introduced in [7], which choose only the smoothest stencil, instead of forming a combination of all the stencils available in order to optimize accuracy. The local computational grids of ENO and WENO schemes are composed of r overlapping substencils of r points, forming a larger stencil with (2r − 1) points and yielding a local rate of convergence that goes from order r, in the case of ENO when only one substencil is used, to order (2r − 1) when the WENO weighted combination is applied at smooth parts of the solution. The nonlinear coeﬃcients of WENO’s convex combination, hereafter referred to as nonlinear weights, are based on lower order local smoothness indicators that measure the sum of the normalized squares of the scaled L2 norms of all derivatives of local interpolating polynomials [2]. An essentially zero weight is assigned to those lower order polynomials whose underlining substencils contain high gradients and/or shocks, aiming at an essentially non-oscillatory solution close to discontinuities. At smooth parts of the solution, higher order is achieved through the mimicking of the central upwind scheme of maximum order, when all smoothness indicators are about the same size. Hence, an eﬃcient and careful design of these smoothness indicators is a delicate and important issue for WENO schemes. The ﬁrst set of nonlinear weights of widespread use has been presented in [2], hereafter denoted as ωk , where βk are the associated lower order smoothness indicators. The scheme resulting from these will be hereafter referred as the classical WENO scheme (WENO-JS). In [5] it was pointed out that these smoothness indicators were the cause of a reduction of the convergence rate at critical points (points of zero derivatives) of the function. In the same article, a ﬁxing was also proposed in the form of a mapping on the classical WENO-JS weights, leading to corrected weights that recovered the formal order of accuracy at critical points. We call the scheme composed by this mapped set of weights as the mapped WENO scheme (WENO-M). Alternatively, in [10], it was shown that the incorporation of a global higher order smoothness indicator, which we call τ2r−1 (in order to emphasize the utilization of the whole set of (2r − 1) points available) into the WENO-JS weights deﬁnition also improved the convergence at critical points with no need of mapping (in the case of r = 3). This last scheme has been named the WENO-Z scheme by its authors. However, a direct comparison of the schemes showed that the improvement on numerical solutions of the WENO-M over the WENO-JS was not due to the increase of the convergence rate at critical points, as it was claimed in [5], but to the decreased dissipation of WENO-M when correcting the weights to a disposition closer to the central upwind scheme. In the same way, the new set of nonlinear weights of WENO-Z provided even less dissipation than WENO-M, obtaining sharper results among all of the schemes. Moreover, the mapping procedure of WENO-M incurs in extra computational cost, while the WENO-Z modiﬁcations are obtained through a simple and inexpensive linear combination 2

of the WENO-JS smoothness indicators βk . In this article, we extend the ﬁfth order WENO-Z scheme introduced in [10] to higher orders of accuracy by providing a closed-form formula for the generation of the associated global smoothness indicators τ2r−1 as linear combinations of the classical lower order local smoothness indicators βk . This was achieved through a thorough investigation of the Taylor expansions of the βk that revealed symmetries that were used to build the τ2r−1 for all values of r. As we shall see, these symmetries originate in the Lagrangian interpolation bases of the local substencils due to their symmetrical dispositions with respect to the global stencil. We also determine the maximum order the higher order smoothness indicator τ2r−1 can achieve as a linear combination of the βk . As mentioned above, WENO’s main idea to avoid oscillations is to generate a convex combination of low order polynomials giving smaller weights to substencils containing discontinuities or high gradients. In [10] it was pointed out that the diﬀerences of resolutions among the three schemes, WENO-JS, WENO-M and WENO-Z, were due to their distinct dissipative properties, which in turn were consequence of the sizes of the weights assigned to the discontinuous substencils. Thus, the WENO-Z sharper results that were obtained in [10] were due to the assignment of weights to discontinuous substencils that were larger than the ones of WENO-M. Analogously, in [5], WENOM also assigned larger weights than the ones of WENO-JS. In this article, we show that the general situation is a little bit distinct from what was shown in [5] and [10], for if we take a closer look at all the parameters involved at the deﬁnition of the weights, we conclude that the weights power parameter p (see formula (25)) plays a major role in the amount of dissipation a particular WENO scheme might have, since it aﬀects the weights relative scaling in the presence of discontinuities. We will also see that the schemes can be classiﬁed in a dissipation scale with WENO-JS being the most dissipative and the mapped WENO being the less for the same value of p, with WENO-Z occupying an intermediary position. For higher values of r, this classiﬁcation determines which scheme keeps the ENO behavior for the values of p currently used. The power parameter p also has an active role on recovering the formal order of accuracy at critical points of the solution and we shall see that this is a distinct property of the WENO-Z scheme. This paper is organized as follows: In Section 2, brief descriptions of the several WENO schemes are given. The main results of the article will be presented in Section 3, where we derive a general formula for the higher order smoothness indicators τ2r−1 of all odd orders WENO-Z scheme. We also derive a closed-form formula, for arbitrary values of r, for the optimal order that τ2r−1 can achieve when expressed as a linear combination of the lower order local smoothness indicators βk , k = 0, . . . , r − 1. The issue of the degradation of the order of convergence at critical points of smooths solutions is discussed in Section 4. In Section 5, dissipation of the various schemes and the inﬂuence of the values of the power parameter p are discussed through the numerical simulation of the linear advection of discontinuous functions, the one dimensional Euler equations with Riemann initial values problems and the Mach 3 shock density-wave interaction. The two dimensional doubleMach shock reﬂection problem was also simulated using the high order WENO-Z ﬁnite diﬀerence scheme. Concluding remarks are given in Section 6.

3

. N. 4 . x=xi i = 0. i = 0. . (2) where ui (t) is a numerical approximation to the point value u(xi . used for the ﬁfth-order WENO reconstruction step. . along with the mapped version introduced in [5] and then the WENO-Z scheme introduced in [10]. we ﬁrst recall the essentials of the classical WENO scheme as designed in [2]. by the method of lines. S1 . A conservative ﬁnite-diﬀerence formulation for hyperbolic conservation laws requires high-order consistent numerical ﬂuxes at the cell boundaries in order to form the ﬂux diﬀerences across the uniformly-spaced cells. which we denote as WENO-JS. where ∆x is the uniform grid spacing. . . . S2 . WENO-M and WENO-Z. 2. t).2 Weighted essentially non-oscillatory schemes In this section we describe all three versions of the (2r−1) order weighted essentially non-oscillatory conservative ﬁnite diﬀerence scheme with r ≥ 3 when applied to hyperbolic conservation laws as in (1). . yields a system of ordinary diﬀerential equations ∂f dui (t) =− dt ∂x . . with cell boundaries given by xi+ 1 = xi + ∆x . N . which are called cell centers. composed of three 3-points substencils S0 . The conservative property of the spatial discretization is obtained by implicitly deﬁning the numerical ﬂux function h(x) as f (x) = 1 ∆x x+ ∆x 2 x− ∆x 2 h(ξ)dξ. respectively. Consider an uniform grid deﬁned by the points xi = i∆x. 2 2 The semi-discretized form of (1). Namely. We will recall the ﬁfth (r = 3) as an example in the discussion below.1 The Classical WENO-JS scheme xi-2 S 5 xi-1 xi xi+1/2 xi+1 β0 xi+2 τ5 S1 S2 S0 β1 β2 Figure 1: The computational uniform grid xi and the 5-points stencil S 5 .

which depend on the left-shift parameter k = 0. . Sr−1 } with each substencil containing r grid points. . the coeﬃcients {d0 = 10 . . 2 2 (3) High order polynomial interpolations to hi± 1 are computed using known grid values of f . . dui (t) 1 h 1 − hi− 1 . k=0 r−1 2 (4) where ˆk ˆ f k (xi+ 1 ) = fi+ 1 = 2 2 ckj fi−k+j . S2 }. the classical ﬁfth-order WENO scheme uses a 5-points stencil. as shown in Fig. which is subdivided into three 3-points substencils {S0 . (βk + ǫ)p (7) We refer to αk as the un-normalized weights. but not on the values fi . deﬁned in each one of the substencils Sk : i± 2 r−1 ˆ fi± 1 = 2 ˆ ωk f k (xi± 1 ). • In the case of r = 3. r − 1. 1. . . . S1 . (5) The ckj are Lagrangian interpolation coeﬃcients (see [2]). . N. S1 . dr−1 } are called the ideal weights since they generate the (2r − 1) order central upwind scheme using the (2r − 1)3 1 3 points stencil. The parameter ǫ is used to avoid the division by zero in the denominator and power parameter p = 2 is chosen to increase the diﬀerence of scales of distinct weights at non-smooth parts of the solution. hereafter named S 5 . in which f k (x) is the r-th degree polynomial below. (8) 5 . d1 . Ideal weights for higher order WENO schemes can be found in [2]. j=0 i = 0. . fi = f (xi ). For instance. 2 2 (6) where the values Ak are independent of ∆x. . The (2r − 1) degree polynomial ˆ approximation fi± 1 = hi± 1 +O(∆x2r−1 ) is built through the convex combination of the interpolated 2 2 ˆ values f k (x 1 ). .such that the spatial derivative in (2) is exactly approximated by a conservative ﬁnite diﬀerence formula at the cell boundaries. it can be shown by Taylor series expansion of (5) that ˆk fi± 1 = hi± 1 + Ak ∆x3 + O(∆x4 ). . The weights ωk are deﬁned as ωk = αk . d2 = 10 } generate the ﬁfth-order central upwind scheme for the 5-points stencil S 5 . d1 = 5 . For example. = 2 dt ∆x i+ 2 where hi± 1 = h(xi± 1 ). . 2 The classical (2r − 1) order WENO scheme uses (2r − 1)-points global stencil. . . The coeﬃcients {d0 . which is subdivided into r substencils {S0 . . The lower order local smoothness indicators βk measure the regularity of the (r − 1) th degree ˆ polynomial approximation f k (xi ) at the substencil Sk and are given by r−1 βk = l=1 ∆x2l−1 xi+ 1 xi− 1 2 2 dl ˆk f (x) dxl 2 dx. r−1 l=0 αl αk = dk .

By (7). . For instance. points where the ﬁrst derivative of the function vanishes (f ′ (xc ) = 0). a fact that was hidden by the homogenization of the weights caused by the use of a relatively large value for ǫ in (7). On the other hand. yielding β0 and β1 to be much smaller than β2 . (12) 2 i i (13) 1 ′ ′′′′ ff ∆x5 + O(∆x6 ).• For example. 4 1 (3fi − 4fi+1 + fi+2 )2 . it was shown that if the smoothness indicators βk satisfy βk = 6 . Let us now check how the classical WENO-JS nonlinear weights ωk in (7) behave with respect to the restrictions above. . in the case r = 3. 3 i i 13 ′′ ′′′ 2 ′ ′′′ ff f f − ∆x4 + 3 i i 6 i i 1 ′ ′′′′ ff ∆x5 + O(∆x6 ). the rate of convergence degraded to only third order (O(∆x3 )). the expression of the βk in terms of the cell averaged values of f (x). fi are given by β0 = β1 = β2 = 13 (fi−2 − 2fi−1 + fi )2 + 12 13 (fi−1 − 2fi + fi+1 )2 + 12 13 (fi − 2fi+1 + fi+2 )2 + 12 13 ′′ 2 f − 12 i 13 ′′ 2 f + 12 i 13 ′′ 2 f − 12 i 1 (fi−2 − 4fi−1 + 3fi )2 . . xi+ 1 ] from its cell averaged 2 2 values f (x) in the substencils {Sk . generating weights ωk that are good approximations to the ideal weights dk . This implies that the inﬂuence of the polynomial approximation of hi± 1 taken across the discontinuity is diminished 2 up to the point where the convex combination (4) is essentially non-oscillatory. this results on ω2 being a small number in (4). (15) (16) In the case r = 3. 4 (9) (10) (11) and their Taylor series expansions at xi are β0 = fi′ 2 ∆x2 + β1 = fi′ 2 ∆x2 + β2 = fi′ 2 ∆x2 + 13 ′′ ′′′ 2 ′ ′′′ fi fi ∆x4 − f f − 3 6 i i 1 ′ ′′′ ff ∆x4 + O(∆x6 ). 1 shows the case where substencil S2 is discontinuous. In [5]. for it reconstructs the values of h(x) at the cell boundaries of the interval Ii = [xi− 1 . r − 1}. in the case of r = 3. (14) 2 i i The general idea of the weights deﬁnition (7) is that on smooth parts of the solution the smoothness indicators βk are all small and about the same size. In [5]. truncation error analysis of the ﬁnite diﬀerence equation (3) led to the following necessary and suﬃcient conditions on the weights ωk for the WENO scheme to achieve the formal (2r − 1) order of convergence at smooth parts of the solution: r−1 k=0 − + Ak (ωk − ωk ) = O(∆xr ). 4 1 (fi−1 − fi+1 )2 . . βk is O(1) and the corresponding weight ωk is small relatively to the other weights. we see that a suﬃcient condition for ﬁfth-order convergence is simply given by: ± ωk − dk = O(∆x3 ). if the substencil Sk contains a discontinuity. (17) It was also found that at ﬁrst order critical points xc . The process synthesized by (4)-(5) is called the WENO reconstruction step. ± ωk − dk = O(∆xr−1 ). Fig. k = 0.

Note. The mapping function gk (ω) used in [5] is deﬁned as ω dk + d2 − 3dk ω + ω 2 k . implying ωk = dk +O(1) (see equations (12)-(14)) and the mapping is unable to improve the weights approximation. gk (dk ) = dk . if ω = dk + O ∆x2 . if only the ﬁrst derivative vanishes. If the second derivative also vanishes. that if at a critical point the second derivative also vanishes. then the weights ωk satisfy ωk = dk + O (∆xq ). It consisted of the application of a mapping function that increased the approximation of ωk to the ideal weights dk at critical points to the required third order O(∆x3 ) as in (17). then βk = D (1 + O (∆x)) and ωk = dk + O (∆x). r ≥ 3. gk (ω) = dk + O ∆x6 . Numerical results in [5] conﬁrmed the usefulness of the mapping. (18) gk (ω) = d2 + ω (1 − 2dk ) k and is a non-decreasing monotone function with the following properties: 1. βk = D(1+O(1)). then the rate of convergence decreases even more to second order. gk (ω) ≈ 0 if ω ≈ 0. where D is a nonzero constant independent of k. we see that βk = D 1 + O(∆x2 ) . gk (dk ) = gk (dk ) = 0. 2. 2. This new smoothness indicator is built using the values of the numerical solution at the whole (2r − 1) points stencil in the form of a simple linear combination 7 . implying that ωk = dk + O(∆x2 ). 0 ≤ gk (ω) ≤ 1. which indeed happens and can be easily conﬁrmed with any symbolic calculation software.3 The WENO-Z Scheme The novel idea of the WENO-Z scheme introduced in [10] is the modiﬁcation of the βk with information obtained from a higher order smoothness indicator. however. Nevertheless. ′ ′′ 3.D (1 + O (∆xq )). degrading the order of convergence of the scheme to third order only. which we denote here by τ2r−1 for any given order (2r − 1). gk (ω) ≈ 1 if ω ≈ 1. 4. Looking at the Taylor series expansions of the smoothness indicators βk in (12)-(14). gk (0) = 0 and gk (1) = 1. This requires that condition (15) must be satisﬁed as well for the classical WENO to have the expected ﬁfth-order convergence. at critical points this situation becomes more complex depending on the number of vanishing derivatives of f (xi ). maintaining the same second order of convergence as the classical WENO-JS scheme. 2. since with the modiﬁed weights the resulting WENO-M scheme recovered the formal ﬁfth-order convergence at critical points of a smooth solution. The downside of using the mapping function is an additional ≈ 20% − 30% cost of cpu time when compared to the classical WENO-JS scheme.2 The Mapped WENO-M scheme A ﬁx to this deﬁciency of the classical weights ωk was proposed in [5]. For instance.

and from (22). τ5 is simply deﬁned as the absolute diﬀerence between β0 and β2 at xi . . (24) Z Thus. . (19) It is straightforward to see from (12)-(14) that the truncation error of τ5 is 13 ′′ ′′′ f f ∆x5 + O(∆x6 ). Z The lower order local smoothness indicators βk of the WENO-Z scheme are then deﬁned with the help of τ5 as βk + ǫ Z . at smooth parts of the solution. a small number used to avoid the division by zero in the denominators of (21) and (22). respeck tively. 2. Z The general deﬁnitions of the normalized and un-normalized nonlinear weights ωk and αZ . 1. For instance. k = 0. . 1. 1. It is straightforward to check from (12)-(14) and the properties of τ5 that. 2. . Z ωk = dk + O(∆x3 ). as usual. (25) where p ≥ 1 is the power parameter. 2. for those k where the solution is continuous. r − 1. 1. the new weights ωk satisfy the suﬃcient condition (17). we need Z ωk = dk + O(∆xr ). for r ≥ 3 are respectively: Z ωk = r−1 Z . . (26) 8 . • τ5 ≤ maxk βk . but discontinuous in the whole S 5 . then τ5 = O(∆x5 ) ≪ βk . 2. k = 0. l=0 αl αZ k αZ = k dk = dk 1 + Z βk τ2r−1 βk + ǫ p . The relevant properties of τ5 to be used in the redeﬁnition of the WENO weights are: • If S 5 does not contain discontinuities. . for k = 0. k = 0. used to enhance the ratio between the smoothness indicators to guarantee convergence at a certain order. k = 0. (22) where ǫ is. namely τ5 = |β0 − β2 | . Analogously. r − 1. k = 0. (21) βk = βk + τ5 + ǫ Z and the new normalized nonlinear weights ωk and the un-normalized nonlinear weights αZ becomes k Z ωk = . providing the formal ﬁfth order of accuracy to the WENO-Z scheme at noncritical points of a smooth solution. for r = 3. 2 Z l=0 αl αZ k αZ = k dk τ5 = dk 1 + Z βk βk + ǫ .of the βk . 3 i i (20) and that it is a measure of higher derivatives of f . when they exist. 2. . then βk ≪ τ5 . (23) βk + ǫ whenever ǫ << βk . 1. k = 0. . 1. τ5 = O(∆x3 ). • if the solution is continuous at some of the Sk . and is indeed computed using the whole 5-points stencil S 5 . as shown in [10].

. Remark 1 At critical points. We opt shall see that the use of τ2r−1 in place of τ2r−1 in the weights deﬁnition improves the ability of the WENO-Z scheme to capture higher order structures in the numerical solution. we take i = 0 (n) dn (x0 = 0) and denote dxn f (x0 ) = f0 unless stated otherwise. therefore. The power parameter p can be used to recover this order. The proof of the main theorem explores symmetric structures of the underlying interpolating polynomials deﬁning the βk . the global higher order smoothness indicator τ2r−1 . . .1 General Formula for τ2r−1 Theorem 2 Given the order of the WENO reconstruction (2r − 1) and the associated lower order local smoothness indicators β0 . Before we give the proof of the theorem at the end of this section. we will give some preliminary results and two lemmas that are necessary for the proof. 2) = 0 is of order O(∆xr+2 ).. which has the optimal order among all the linear combinations of the βk . |β0 − β1 − βr−2 + βr−1 | mod (r. we shall denote the 9 . In the next section. For instance. Without loss of generality. the order of the critical point. 1. Furthermore. This is unique to WENO-Z.. at a ﬁrst order critical point the convergence of the WENO-Z scheme degrades to fourth order if p = 1 in (25) and regains ﬁfth order when p = 2. changing the value of p in WENOJS or in WENO-M does not alter their convergence rate at critical points.. 2) = 1 τ2r−1 = (28) . (27) As we shall see below. . τ2r−1 . r − 1. deﬁned as: |β0 − βr−1 | mod (r. as shown in [10]. although no closed formula is provided. 3.to obtain convergence of the WENO-Z at order (2r − 1) and this is achieved if τ2r−1 = O(∆xr ). we provide a closed-form formula for generating such a τ2r−1 for all values of r as a linear combination of the βk . goes from 0 to r − 1. We also determine the existence for all r. . we need τ2r−1 = O(∆xr+2 ). of an improved higher order smoothness opt indicator. This will be done with the help of two lemmas and an auxiliary theorem. Numerical experiments at Section 4 further illustrate this property and more detailed computations that demonstrate this exclusive aspect of WENO-Z are shown in [10]. the βk are always O(∆x2 ). βr−1 . 3 The global higher order smoothness indicators In this section we formulate and prove the necessary theoretical results to obtain a general formula for the global higher order smoothness indicators τ2r−1 for all values of r. for r = 3. βk + ǫ k = 0. the lower order local smoothness indicators βk are no longer O(∆x2 ) and the size of τ2r−1 also varies in a way that the order of convergence of the ratio in (23)decreases monotonically from r to 1 as ncp .

. is given by r−1−n (n) pk (x) (n) = j=0 bk. 2) = 0 mod (j1 + j2 . δ = 1.j1. . pk (x). which express the numerical ﬂux h(x) in terms of the derivatives of the physical ﬂux function f (x) and establish the independence and anti-symmetry properties of the polynomial coeﬃcients ak.j2 = (j1 +n)!(j2 +n)! 2−(j1 +j2 ) j1 !j2 ! (j1 +j2 +1) mod (j1 + j2 . The n th derivative of pk (x). . (30) where bk.j2 ak. j2 = 0.j1 xj1 bk. we may write βk as r−1 βk = n=1 r−1 ∆x 2n−1 1 ∆x 2 − 1 ∆x 2 1 ∆x 2 pk (x) r−1−n j1 =0 (n) 2 dx r−1−n j2 =0 = n=1 r−1 ∆x2n−1 − 1 ∆x 2 1 ∆x 2 r−1−n r−1−n j1 =0 j2 =0 bk. . 0 j1 . .n. 2) = 1 .n. are constants in the expansion.ˆ lower order polynomial approximation f k (x) of degree (r − 1) in the substencil Sk as pk (x): r−1 pk (x) = j=0 ak.j are the coeﬃcients of the polynomial expansion of f (x) about the point x0 in the substencil Sk . 10 .n+j2 ∆xj1 +j2 +2n .j1 bk.j xj .n+j . . We start with the following two lemmas below.n.j = (j + n)! ak. the numerical ﬂux of f (x). Lemma 3 Consider the primitive function h(x).j xj .n.n.j . .j1. j! Thus. as deﬁned in (3). (32) where φ0 = 1 and φ2δ . (29) where ak. then ∞ h(x) = δ=0 φ2δ f (2δ) (x)∆x2δ . where Cn.j2 xj2 dx = n=1 ∆x2n−1 − 1 ∆x 2 bk. 1 ≤ n ≤ r − 1.n+j1 ak. r − 1. The idea of the proof of Theorem 2 is to rewrite βk in (31) as another asymptotic expansion in ∆x where the coeﬃcients show an anti-symmetric behavior with respect to the substencils index k.n.j2 xj1 +j2 dx (31) r−1 r−1−n r−1−n = n=1 j1 =0 j2 =0 Cn.

x 1 ].j = l=0 ∞ σk. h=f− 1 (2) 2 7 (4) 4 f ∆x + f ∆x + · · · . If x ∈ [x− 1 . 2 2 r−1 pk (x) = j=0 ak. The above process can be repeated to replace the higher order derivatives of h(x) by f (x) and to obtain (32). (37) where σk. and expanding h± 1 = h(x± 1 ) about x0 = 0. (b) Or alternatively as: ak. (39) .j are expressed either as: (a) ak.j. 4 3! 16 5! ∆x2 1 (2) ∆x3 1 (4) h + h + ··· .l .j. after some algebra.l = (−1)l σr−1−k.j. 2 2 f ′ = h′ + Integrating both sides. (35) where the coeﬃcients ak. the above formula is independent of k. 2) = 0 mod (l.j xj . (36) with the coeﬃcients φ2δ as given in lemma 3.j. 2) = 1 11 0 . then pk (x) in (29) can be written as. ∆x2 1 (3) ∆x3 1 (5) h + h + ··· . for l ≤ ρj .Proof. 24 ∆x2 (6) h + ··· . 24 5760 We have the constants φ2j as shown in the expansion with φ0 = 1.l = 1 j! φl (38) mod (l. r−j−1 2 Lemma 4 Let ρj = . = h(4) + 24 (33) (34) f (2) = h(2) + f (4) Substituting h(2) and h(4) in (34) yields. that is σk.l f0 (j+l) ∆xl . From (3). 4 3! 16 5! ∆x2 (4) h + ··· . and. f =h+ and diﬀerentiating (33).j 1 = j! ρj φ2δ f0 δ=0 (j+2δ) ∆x2δ + O(∆xr−j ).

. therefore φ2δ f (j+2δ) (x0 )(x − x0 )j ∆x2δ + O(∆xr ). One can see that the terms in the second summation are of O(∆xj+2δ ) since (x − x0 ) = O(∆x).Proof. Proof of (b): Our objective is to build (r − 1)-th degree polynomial approximations of h in the substencils Sk and Sr−1−k in order to obtain the symmetry in (38). one has (x − x0 )j ∆x2δ ≤ O(∆xr ). i ∈ {k − r.. and Gk (x) = − 1 [(r−1)−k]+ 2 h(ξ)dξ. (42) with ak. where fs = xs+ 1 2 xs− 1 2 h(ξ)dξ..j 1 = j! ρj φ2δ f0 δ=0 (j+2δ) ∆x2δ + O(∆xr−j ).. s=i+1 i ∈ {−k. . (40) Combining equations (32) and (40) and expanding f and its derivatives in Taylor series around x0 = 0. or l > ρj = r−1 r−j−1 2 . (41) pk (x) = j=0 1 j! ρj δ=0 with φ0 = 1.. primitives of h deﬁned as x x Fk (x) = x[k−(r−1)]− 1 2 h(ξ)dξ. x We rewrite these as: i Fk (xi+ 1 ) = 2 xs+ 1 2 i h(ξ)dξ = s=k−(r−1) fs ∆x. For j + 2δ > r − 1. and similarly.. (r − 1) − k}. . (43) Gk (xi+ 1 ) = − 2 fs ∆x. let Pk (x) be the only polynomial of degree less 2 12 .. F Let Pk (x) be the only polynomial of degree less than or equal to r that interpolates Fk (x) in the G r + 1 points xi+ 1 . pk is a (r − 1)-th degree polynomial approximation of h: pk (x) = h(x) + O(∆xr ). we have ∞ pk (x) = j=0 1 j! ∞ δ=0 φ2δ f (j+2δ) (x0 )(x − x0 )j ∆x2δ + O(∆xr ). Hence pk (x) = j=0 r−1 ak. . k}. s=k−(r−1) xs− 1 2 (r−1)−k i ∈ {k − (r − 1)... k}. For that we use the functions Fk and Gk .j (x − x0 )j .

x(r−1)−k+ 1 ]. 2 2 x ∈ [x−k− 1 . We can rewrite Pk (x) and Pk (x) using the following Lagrangian interpolation formulas: k F Pk (x) = i=k−r (r−1)−k G Pk (x) = i=−k−1 Fk (xi+ 1 )Lk.. 2 with k Lk. 2 One can show that (see [3] Shu) ′ F h(x) = Fk (x) = Pk ′ (x) + O(∆xr ).i. 2 2 pk (x) = p(r−1)−k (x) = F Pk ′ (x).i (x). x ∈ [x[k−(r−1)]− 1 . G ′ Pk (x). r h(x) = and thus ′ Gk (x) = G ′ Pk (x) x ∈ [x[k−(r−1)]− 1 .. It is also easily seen that Lk. 2 Gk (xi+ 1 )L(r−1)−k.s=i x − (s + 2 )∆x k s=k−r.j = r k 1 s=k−r. (r − 1) − k}. x(r−1)−k+ 1 ]. i ∈ {−k − 1.i (x). + O(∆x ). xk+ 1 ]. xk+ 1 ].i. (44) 13 . 2 2 F G are the approximations we are searching for.j ∆x c2 ∆xr k.j ∆x−j xj .−i−1 (−x)..i (x) = s=k−r s=i xi+ 1 − xs+ 1 2 2 x − xs+ 1 2 k = s=k−r s=i 1 x − (s + 2 )∆x (i − s)∆x r 1 r−j xj j=0 ck.i (x) = L(r−1)−k. .s=i(i − s)∆x = = j=0 ck.than or equal to r that interpolates Gk (x) in the r + 1 points xi+ 1 . 2 2 x ∈ [x−k− 1 .i.

j = jck. j=0 Moreover.i (x) = i=−k−1 s=i+1 k −fs ∆xLk.i. p(r−1)−k (x) = i=k−r s=k−(r−1) j=0 −f−s ck.i∗ −1 (−x) = r i=k−r s=−i = i=k−r k s=−i i −fs ck. diﬀerentiating the equalities above we obtain k i r pk (x) = i=k−r s=k−(r−1) j=0 k i r fs ck.−i−1 (−x) −fs ∆xLk.i.j ∆x−(j−1) (−x)j j=0 r = i=k−r s=k−(r−1) −f−s ck.j ∆x−(j−1) xj (r−1)−k (r−1)−k = i=−k−1 s=i+1 k+1 −fs ∆xL(r−1)−k.j p(r−1)−k (x) = j=0 f−s xj ∆xj (−1)j xj 14 .Using (43) and (44) .i. we arrive at the following expressions for the interpolating polynomials of h in the substencils Sk and Sr−1−k : r−1 pk (x) = j=0 r−1 k s=k−(r−1) k i=s Γk. we obtain k F Pk (x) i = i=k−r s=k−(r−1) k i fs ∆xLk.j ∆x−(j−1) (−1)j jxj−1 .i.i (−x) (r−1)−k (r−1)−k = i∗ =k−(r−1) s=−i∗ +1 k (r−1)−k −fs ∆xLk.j and reorganizing the indexes.j ∆x−(j−1) (−x)j .i.j fs ∆xj k s=k−(r−1) k i=s Γk.i(x) r = i=k−r s=k−(r−1) (r−1)−k (r−1)−k G Pk (x) fs j=0 ck.i.i. Deﬁning Γk.j ∆x−(j−1) jxj−1 .i.i.

j1 . l2 ) = Cn. δ! 0 (−1)j (−1)j ∞ 1 (δ) δ δ δ δ=0 δ! f0 (−1) s ∆x ∆xj k i=s Γk. Γk.j.n+j2.j a(r−1)−k.j ak.j s=k−(r−1) a(r−1)−k.j = f (−1)l ∆xl . Deﬁnition 5 In order to simplify the notation we deﬁne: M∗ = and E(k.l = s=k−(r−1) i=s Γk. comparing both expressions we see that k k σk. δ! 0 Thus. (46) M .i. n.i. their respective coeﬃcients are given by ak.Thus.j.l2 .n+j1.j ∆xj k k = δ=0 Γk.j = (j + l)! 0 l=0 s=k−(r−1) i=s ∞ k k sj+l (j+l) Γk.j fs ∞ 1 (δ) δ δ δ=0 δ! f0 s ∆x ∆xj k s=k−(r−1) k i=s Γk. 2 (45) 15 .j.i.i.l1 σk. j2 .j i=s k i=s Γk.i.i.j = = ∞ k s=k−(r−1) k s=k−(r−1) f−s 1 (δ) δ f s ∆xδ−j . We now deﬁne l = δ − j and note from (36) that δ − j ≥ 0 to arrive at k k ∞ sj+l (j+l) f ∆xl .l .i.j ∆xj k k = δ=0 Γk.j sj+l . (j + l)! sj+l (−1)l .j. (j + l)! σr−1−k.j2 σk.j i=s s=k−(r−1) where above we expanded fs in Taylor series over the origin and reorganized the summations in terms of ∆x.i.j1 .l = s=k−(r−1) i=s k k Γk.i.l = (−1)l σr−1−k. (j + l)! 0 l=0 s=k−(r−1) i=s 1 (δ) f (−1)δ−j sδ ∆xδ−j .j = = ∞ k s=k−(r−1) k i=s Γk.i. l1 .j satisfy σk.

j1 . Ak. . where r−1 ΩM1 . Substituting (46) into (31). Ψn = {1.Using (38). l1 . if j < r and M − j < r. A0. . .M2 = {(n. n. l2 ) ∈ with n=1 Ψn | n + j1 + l1 = M1 and n + j2 + l2 = M2 }.l2 =0 E(k.j is independent of k.M.j . l1 . . l2 )f0 f0 (j) (M −j) ∆xM .j = (−1)M A(r−1)−k. 16 . j2 .n+j1 ak.M. j1 . we obtain the anti-symmetry/symmetry condition E(k. l2 ).l1 σk. j1 . (48) with Ak. Deﬁning M = M1 + M2 . we have ∞ M −1 βk = M =2 j=1 Ωj.n+j1. n. l1 . . j1 . r − 1} × {1.M.M. that is. one has ∞ ∞ (49) ak. Fixing M1 = n + j1 + l1 and M2 = n + j2 + l2 . n.l2 f0 (n+j1 +l1 ) (n+j2 +l2 ) f0 ∆xl1 +l2 .j f0 f0 (j) (M −j) ∆xM . j2 . j1 .j .j = · · · = Ar−1. l2 )f0 (M1 ) (M2 ) f0 ∆xM1 +M2 . j2 .n+j2 = l1 =0 l2 =0 σk. l1 .M.j = A1. r − 1 − n}2 × N2 .M2 E(k.M. . Proof. l2 ) = (−1)l1 +l2 E(r − 1 − k. . Theorem 6 The lower order local smoothness indicators βk can be written as ∞ M∗ βk = M =2 j=1 Ak. l1 . Using the deﬁnition of ak. l2 )f0 (n+j1 +l1 ) (n+j2 +l2 ) f0 ∆xj1 +j2 +2n+l1 +l2 . (47) We now state and proof the new asymptotic expansion for the lower order local smoothness indicators βk . l1 . j2 . . n.M −j E(k. j1 . n.j2 =0 l1 .n+j2. we obtain r−1 r−1−n ∞ βk = n=1 j1 . In addition. j2 .M.j in (37). j2 . we can reorganize the sum as ∞ ∞ βk = M1 =1 M2 =1 ΩM1 .

2) = 0 .j .j f0 f0 M =2 j=1 (j) (M −j) ∆xM . ∞ M∗ we can reorganize the sum above as E(k.M −j ∪ ΩM −j.M. .M. l2 ) = Ωj. using (31) and (36).j = (−1)M A(r−1)−k.j2) ) ∆xj1 +j2 +2n . .j1 . The formula above shows that the coeﬃcients associated to f0 1 1 f0 2 2 are independent of k.j E(k.M −j ∪ ΩM −j.j . j1 .j2 =0 δ1 =0 δ2 =0 O(∆xr−max(j1 .j yielding Ak. n.M −j ∪ ΩM −j. n. l2 ) (−1)2n+j1 +j2 +l1 +l2 E(r − 1 − k.j ∆xM .M. j2 . = Ωj. l1 . n. j2 . Indeed. Ak. l2 )f0 f0 (j) (M −j) βk = M =2 j=1 Ωj.M.j Ωj. j2 . j1 . l1 .j2 = 0). n. l2 ) (−1)M E(r − 1 − k. j2 . the coeﬃcient Ak. Corollary 7 If M ≤ r.M −j ∪ ΩM −j. l2 ).Since f0 f0 (j) (M −j) = f0 (M −(M −j)) (M −j) f0 .M. l2 ).M −j ∪ ΩM −j. j1 . We can also see that if j < r and M − j < r.j is independent of k. j2 .j2 f f0 + βk = (n + j1 )!(n + j2 )! 0 n=1 j1 . j1 . 17 .j is independent of k. j2 . Therefore the largest possible value for n + j1 + 2δ1 (and for n + j2 + 2δ2 ) is n + j1 + 2ρn+j1 = n + j 1 + 2 r − 1 − (n + j1 ) = 2 r−1 r−2 mod (r − 1 − (n + j1 ). . ρn+j1 ρn+j2 r−1 r−1−n φ2δ1 φ2δ2 ∆x2δ1 +2δ2 (n+j1 +2δ1 ) (n+j2+2δ2 ) Cn. .M −j ∪ ΩM −j. l1 . j1 . 2) = 1 (n+j +2δ ) (n+j +2δ ) Thus if j < r and M − j < r. By deﬁning Ak. Ak. l1 . j = 1.M.j1. j1 . we have βk = ∞ M∗ Ak.j (−1)l1 +l2 E(r − 1 − k.j = Ωj. n. we have E(k. n. M ∗ is independent of the shifting parameter k. mod (r − 1 − (n + j1 ).M. Using the property (47) and the fact that j1 + j2 is always even (otherwise Cn.j = Ωj. l1 . l1 .

(50) First of all. We state our main result in the following proposition: opt Proposition 8 The greatest lower bound for the order of τ2r−1 .j f0 f0 (j) (M −j) We seek to ﬁnd a linear combination of the above βk in such a way that the sum of all the terms of O(∆xk ) is zero for all k ≤ r + 1. the coeﬃcients of β0 and βr−1 associated to the terms ∆xM . In this section we perform a deeper investigation in the vector space generated by the linear combinations of the βk in order to ﬁnd the global optimal opt order smoothness indicator. < r. for each M ≤ r. yielding τ2r−1 = |β0 − β1 − βr−2 + βr−1 | = O(∆xr+2 ). Hence τ2r−1 = |β0 − βr−1 | = O(∆xr+2 ).j f0 f0 (j) (M −j) ∆xM (−1)M . cr−2 . max ρ+ .M.2 opt The optimal higher order smoothness indicator τ2r−1 The linear combinations of the βk displayed in Theorem 2 are not unique.M. Using the deﬁnition of βk in (48) and the symmetry condition in (49).j f0 f0 ∆xM . we seek non-trivial constants c0 . τ2r−1 . In other words. 3. in the sense of generating the highest possible order for τ2r−1 .j and of order O(∆xM ). Also.M. nor optimal. the coeﬃcients of these two sums are equal to ∗ 2 M Ak. M =2 j=1 ∞ M∗ β0 = βr−1 = M =2 j=1 ∞ M∗ A0. the Taylor expansions of Theorem 6 may yield a higher order. c1 . for each value of r ≥ 3. r ≥ 3.M. ∆xM (−1)M . • assuming r is odd. M = r + 1 (M is an even number) are equal. the coeﬃcients associated with the O(∆xr+1 ) (odd order) term of (β0 + βr−1 ) and (β1 + βr−2 ) are zero. one has ∞ M∗ (j) (M −j) A0. M ≤ r are equal. is given by the largest value of m such that r r + max ρ− . • assuming r is even. using Corollary 7. Even though it is suﬃcient to have τ2r−1 = O(∆xr+2 ) to guarantee the formal (2r − 1) order of accuracy of the WENO-Z scheme.j f0 f0 ∆xM . Furthermore. M =2 j=1 ∞ M∗ (j) (M −j) A1.We are now ready to prove Theorem 2: Proof. all the coeﬃcients β0 and βr−1 associated to the terms ∆xM . This means that the coeﬃcients j=1 of diﬀerence of these two sums (β0 + βr−1 ) − (β1 + βr−2 ) of order ∆xM for all M ≤ r + 1 are all zero. M2r−1 . β1 = βr−2 = M =2 j=1 A1. cr−1 such that the global higher order smoothness indicator τ2r−1 = c0 β0 + c1 β1 + cr−2 βr−2 + cr−1 βr−1 = O(∆xr+2 ).M. 2 2 18 .

As illustrated in Table I. M ∗ . r + 2. . the order of τ2r−1 . opt Let us now check by means of a numerical example the claim that the use of τ2r−1 improves over τ2r−1 when capturing high order structures in the solution. WENO-Z is able to capture high order information up to O(∆x9 ). For instance. . for r > 3. − φk = m−r 2 m−r 2 2 m−r 2 +1 mod (m. 2) = k . (2r − 1). . . + ρ = φ1 . j = 1. Rearranging the summation in (48). for r > 4. . when equipped with τ11 = O(∆x9 ). mod (m. given by Theorem 2. . The WENO-Z scheme built with the global high order smoothness indicator τ11 = O(∆x8 ) does not resolve well functions with information containing higher order opt than O(∆x7 ). . cr−1 } such that r−1 ck βk = O(∆xm ). . it is suﬃcient to ﬁnd a nontrivial set of constants {c0 . m − 1. f (x) = xk + exp(lx).j = 0. (51) with parameters k = 8 and l = 5. M2r−1 is always smaller than the order of the scheme.M. we want k=0 m−1 M ∗ M =2 j=1 r−1 ck Ak. .j k=0 f0 f0 (j) (M −j) ∆xM = 0. . To obtain a τ2r−1 of order m. x ∈ [−1. On the other hand. 1]. . k=0 M = 2. 19 . one can see that the WENO-Z11 opt scheme using τ11 is signiﬁcantly less accurate than the one using the τ11 when computing the derivative of the test function above.where ρ = 1 + φ0 . .M. 2) = 1 − k The proposition above yields the following table with the value of M2r−1 for r up to 20: r 3 4 5 6 7 8 9 10 11 2r − 1 5 7 9 11 13 15 17 19 21 M2r−1 5 7 8 9 11 12 13 15 16 r 12 13 14 15 16 17 18 19 20 2r − 1 23 25 27 29 31 33 35 37 39 M2r−1 17 18 19 21 22 23 24 25 27 We see from the table that when derived as a linear combination of the lower order local smoothness opt indicators βk . We would like now to present more clearly the issues involved in the proof of Proposition 8. consider the following test function. but it is also bigger than the order of the conventional τ2r−1 . or r−1 ck Ak.

3 9.9 4.WENO-Z11 ∆x 3. .2e-14 10.1 Am = .56250e-02 7. .m−1.1e-20 11.5e-10 2.m−1.⌊ m−1 ⌋−1 2 A0. . A1.2. . . .12500e-02 1.9e-13 4.r. E∞ and the order of accuracy m. .2.0 1. where c = A1.0 Table I: The maximum l∞ error. p = 2 and ǫ = 1 × 10−40 using opt the higher order τ11 and the optimal order τ11 in the deﬁnitions of the WENO-Z nonlinear weights Z ωk . We need to ﬁnd a nontrivial solution (c0 . l = 5. A 0. ··· .m−1.1 .90625e-03 1. ··· Ar−1.0e-23 11.1 . . ··· .m−1.1 ..0 5.2.1 .8e-16 9. Since the number of columns of Am is r. .95312e-03 9.4 8.⌊ m−1 ⌋−1 2 2 · · · Ar−1.m−1. . . A1.0 9.9 2.81250e-03 3.⌊ m−1 ⌋−1 2 Note that the matrix Am has r column vectors and some of the row vectors of A can be linearly dependent.⌊ m−1 ⌋ · · · Ar−1.r. The schemes are the eleventh order WENO-Z scheme (WENO-Z11) with r = 6.4e-21 4. A0.2e-17 10.⌊ m ⌋ 2 of the system of linear equations Am c = 0.7e-19 1. are shown with increasing resolution ∆x for the test function f (x) = xk + exp(lx) with k = 8.r. . one needs rank(Am ) < r to ensure that a non-trivial solution exists. A1. . cr−1 )T and A0.⌊ m−1 ⌋ 2 Ar−1. 20 .0 9.6e-10 8.76562e-04 Higher order E∞ 1. .m−1.4 opt Optimal order τ11 E∞ m 1.1 .1e-24 τ11 m 9. . O(∆xm ). . .0e-27 11.

21 0 0 2160 0 0 1440 0 0 30240 0 ′ 1 ′ (5) 1 ′′ (4) 781 ′′′ 2 13 ′′ ∆x6 + f f + f f02 ∆x2 + f0 2 ∆x4 + − f0 f0 − 12 15 360 0 0 720 0 1 ′ (7) 53 ′′ (6) 781 ′′′ (5) 32803 (4)2 − ∆x8 + O(∆x9 ). f f − f f − f f + f 21 0 0 2160 0 0 1440 0 0 30240 0 ′ 2 ′ (5) 1 ′′ (4) 781 ′′′ 2 2 ′ (6) 9 ′′ (5) 13 ′′ ∆x7 + f f + f ∆x6 + − f0 f0 + f0 f0 f02 ∆x2 + f0 2 ∆x4 + − f0 f0 − 12 5 360 0 0 720 0 3 5 13 ′ (7) 1235 ′′ (6) 5467 ′′′ (5) 32803 (4)2 − f0 f0 + ∆x8 + O(∆x9 ). 13 12 −2 5 1 − 360 781 720 2 A9 = 3 −9 5 − 13 21 1235 432 5467 − 1440 32803 30240 1 1 13 12 1 10 1 − 360 781 720 1 − 12 11 60 1 21 251 − 2160 781 − 1440 32803 30240 1 13 12 1 − 15 1 − 360 781 720 0 0 1 − 126 53 − 2160 781 1440 32803 30240 13 12 1 10 1 − 360 781 720 1 12 11 − 60 1 21 251 − 2160 781 − 1440 32803 30240 . f f − f f + f 21 432 0 0 1440 0 0 30240 0 ′ β1 = β2 = β3 = β4 = The corresponding matrices Am (omitting the zero lines) for m varying from 6 to 9 are 13 12 2 A7 = − 5 1 − 360 781 720 A6 = 1 13 12 1 13 12 1 13 12 1 13 12 1 13 12 1 1 13 12 1 10 1 − 360 781 720 1 13 12 1 − 15 1 − 360 781 720 1 13 12 1 10 1 − 360 781 720 1 13 12 −2 5 1 − 360 781 720 . f f − f f + f 21 432 0 0 1440 0 0 30240 0 ′ 1 ′ (6) 11 ′′ (5) 1 ′′ (4) 781 ′′′ 2 1 ′ (5) 13 ′′ ∆x6 + − f0 f0 + f0 f0 ∆x7 + f0 f0 − f0 f0 + f0 f02 ∆x2 + f0 2 ∆x4 + 12 10 360 720 12 60 1 ′ (7) 251 ′′ (6) 781 ′′′ (5) 32803 (4)2 f f − f f − f f + f ∆x8 + O(∆x9 ). rank(A8 ) = 4 < r = 5. . The rank of the matrices above are: rank(A6 ) = 1. f f − f f + f f + f 126 0 0 2160 0 0 1440 0 0 30240 0 ′ 1 ′′ (4) 781 ′′′ 2 1 ′ (5) 1 ′ (6) 11 ′′ (5) 13 ′′ ∆x7 + f f − f f + f ∆x6 + f f − f0 f0 f02 ∆x2 + f0 2 ∆x4 + 12 10 0 0 360 0 0 720 0 12 0 0 60 1 ′ (7) 251 ′′ (6) 781 ′′′ (5) 32803 (4)2 ∆x8 + O(∆x9 ). 21 rank(A9 ) = 5. 1 1 13 12 2 −5 1 − 360 781 720 2 −3 9 5 13 − 21 1235 432 5467 − 1440 32803 30240 A8 = 1 13 12 −2 5 1 − 360 781 720 2 3 9 −5 1 13 12 1 10 1 − 360 781 720 1 − 12 11 60 1 13 12 1 − 15 1 − 360 781 720 1 13 12 1 10 1 − 360 781 720 1 12 − 11 60 1 13 12 −2 5 1 − 360 781 720 −2 3 9 5 0 0 . rank(A7 ) = 2.To clarify the notation above we consider r = 5 and look to the Taylor expansions of the lower order local smoothness indicators βk for orders less than ∆x9 at x = x0 : β0 = f02 ∆x2 + 2 ′ (6) 9 ′′ (5) 2 ′ (5) 1 ′′ (4) 781 ′′′ 2 13 ′′ 2 4 ∆x6 + ∆x7 + f0 ∆x + − f0 f0 − f0 f0 + f0 f f − f0 f0 12 5 360 720 3 0 0 5 13 ′ (7) 1235 ′′ (6) 5467 ′′′ (5) 32803 (4)2 − f0 f0 + ∆x8 + O(∆x9 ). .

m. . one r can see that each row vector has 2 free variables A0.M.2 .m.m.m. it is easy to see that all the row vectors of Ar+1 are linearly dependent since by Corollary 7.1 A0. These new non-zero row vectors are represented by the submatrix Bm as below: A0.M.m. Remark 9 In the proof of Proposition 8 below.. = Ar+1. .m. the row vectors take the form A0.1 Ar−1. we will assume that all the remaining ones do count.m. the constant vectors of Bm do not count.m. Using the anti-symmetry/symmetry condition (49). it is not always true that rank(Am+1 ) = rank(Am ) + rank(Bm ) and to understand the increase of the rank of Am . Ar−1.j · · · Ar−2. . .m. A0. .j . 2 A0.m. Considering the vector space formed by all the row vectors that takes the form above.m.m. .j = −A r−1 .m. 2 which is also the rank of Bm . Therefore the number of new row vectors that are not dependent on k is the cardinality of the set m | j ≥ r or m − j ≥ r . A r−1 . • If r is odd.⌊ m ⌋ · · · Ar−1.j 2 Ar−1.j (−1)m A0. we need to analyze the matrix Am .Therefore.M.j . we search for a rule for the increase of the rank of Am when the submatrix Bm is appended to Am . for in this case A r−1 . . . Proof. the maximum order τ9 can achieve is given by M9 = 8 since rank(A9 ) ≥ r = 5.j · · · A r−1 .m. this is what we observed in all the cases we tested.j = .j .m.2 . rank(Bm ) = m − r. By Theorem 2. Obviously.j A1.m. . or its lowest upper bound as stated in the proposition. .m. when m is even. on the other hand. j ∈ 1. .m. when m is odd.j · · · A r−1 . We want to ﬁnd out how many row vectors of Bm do not depend on k.m.m.m.. . These row vectors will be responsible for the increase of the rank of Am . we increase the order by one to m and append non-zero row vectors that correspond to the coeﬃcients of βk of order m to the row vectors of the old matrix Am to form a new matrix Am+1 .m.j 2 · · · (−1)m A1. .j . the elements that are not dependent on k take the form of Ak.2 . We now adopt the following strategy to prove proposition 8: For a given matrix Am with r column vectors and with row vectors containing all the coeﬃcients of βk of order up to and equal to m − 1 that does not have full rank. 2 2 22 .1 A1. . with j < r and m − j < r. they are pairwise linearly independent. This process will terminate in a ﬁnite number of iterations since only ﬁnite number of row vectors can be appended as the maximum possible order of τ2r−1 ≤ 2r − 1. First of all. Although we do not have a proof for that. i.j = 0. 2 Bm = A1. Let m > r + 1 be ﬁxed. . .j . We are computing the worst case scenario for the rank of Am .⌊ m ⌋ .j A1. it is equivalent to A0.j for M ≤ r and thus rank(Ar+1 ) = 1. 2 ··· ··· .m. and r 2 2 free variables. This process will be repeated until the newly formed matrix reaches full rank. .⌊ m ⌋ A1.e.

with the closed form stated in the Proposition. . . symbolic computational system such as Maple can be employed. there is no central term A r−1 .j and there are always 2 of the value of m. r 2 free variables. r − 1 . r+1 For a large order (2r − 1). because of the dimensions of m+1 m+1 the vector spaces. rank(Am+1 ) = rank(A− ) + min rank(A+ ) + rank(Bm ). ∈ Rr | al = ar−1−l . l = 0. m+1 m+1 r r this increase is limited by 2 ( 2 ) for rank(A+ ) (rank(A− )). . . . 4 Critical points In [5] it was shown that the classical ﬁfth-order WENO-JS scheme loses convergence at critical points due to the inability of its nonlinear weights to distinguish between ﬂat and rough parts of 23 . we can see that if m is even (odd). Thus. rank(A). optimal order M2r−1 .• If r is even. . In this situation. (2r − 1). From this result we obtain a recurrence relation for rank(Am ). Table II gives the parameters r. This duality allows us to build two diﬀerent vector spaces. that is. by appending the row vectors of Bm . rank(A) and the leading order term of order O(∆xM2r−1 ). it is diﬃcult to ﬁnd the kernel and rank of the matrix A analytically. In Maple. 2) = 0 mod (m. deﬁned by V+ = V− = with dimensions equal to equal). m We will take this moment to observe that rank(A+ ) = 1 and rank(A− ) = 0. . Remark 10 From this point on. V + and V − . . On the other hand. the l th element of each row vector can be anti-symmetric or symmetric to the (r − 1 − l) th element. . with initial conditions rank(A+ ) = r+1 1 and rank(A− ) = 0. only the rank of A+ (A− ) will increase. . the builtin function RowReduce is used to ﬁnd the kernel c. r − 1 . if we intersect Am with each one of these vector spaces. l = 0.m. 9 for the (2r − 1) order WENO-Z scheme. we shall have two new matrices A+ and A− and m m − rank(Am ) = rank(A+ ) + rank(Am ). regardless Due to the factor (−1)m . a0 a1 · · · ar−1 r 2 a0 a1 · · · ar−1 and r 2 ∈ Rr | al = −ar−1−l . 2) = 1 . r = 3. . r+1 r+1 Now we can deﬁne rank(Am+1 ) based on rank(Am ). m m r 2 r 2 mod (m. we shall replace the deﬁnition of τ2r−1 in the deﬁnition of the opt Z nonlinear weights ωk in the WENO-Z scheme with τ2r−1 in the rest of the paper unless stated otherwise. . Indeed. coeﬃcient opt vector c and the leading order for the global optimal order smoothness indicator τ2r−1 . respectively (note that when r is even the dimensions are These vector spaces are in a direct sum. m m rank(A+ ) + min rank(A− ) + rank(Bm ).

(2r − 1). 1832. i. 514. We used the ﬁxed value of p = 2 for WENO-JS and WENO-Z. 10. increases the rate of convergence of WENO-Z and. −1832. 0. 9 for the (2r − 1) order WENO-Z scheme. increasing the value of p from 2 to 3 allows WENO-Z to recover the formal order of accuracy of the scheme. increasing p. . . Moreover. 1) (−1. 135. particularly. 24 . −135. 35. 1]. this function has a critical point of order ncp = k − 1 at x = 0. we use the standard value of ǫ = 10−40 . 99. . k = 3. the solution. −135. −6. 2. 1) (−1. 99. rank(A). 0. it was shown that the use of the higher order smoothness indicator τ2r−1 into the formula for the WENO-Z weights improved the ability of the scheme to detect such situations. 1) (0. 2. since their rates of convergence do not change when p varies. On the other hand. 1. −1. . −10. We now perform a numerical experiment to illustrate the behavior of the several schemes in the presence of smooth solutions containing critical points. This is due to the normalization of the nonlinear weights. 0) (1. optimal order M2r−1 . 0. since on both situations their relative sizes may show large variations. when p = r − 1 and ncp < r − 1. . 17. 0. . 1. 1) (1. This τ2r−1 p happens because the use of τ2r−1 through the ratio inserts a measure of relativeness of βk + ǫ the sizes of the lower order smoothness indicators βk . −1. 2. the coeﬃcient vector c and the opt leading order terms for the global optimal order smoothness indicator τ2r−1 . this also allows a speed up of the rate of convergence at critical points just by increasing the value of the power parameter p.e. r = 3. 0. −2. j = 0. In [10]. 3.r 3 4 5 6 7 8 9 2r − 1 rank(A) M2r−1 5 7 9 2 3 4 4 6 7 7 5 7 8 9 11 12 13 c (−1. Consider the following test function.. 1) (−1. 0. . although only in the latter they are large in absolute value. Table IV shows that this is indeed a general behavior. that is. x ∈ [−1. . as it was shown in [10]. That is. 0. 1) Table II: The parameter r. −36. −17. k − 1. −47. 47. 0) Leading Order at xi −3fi (1) (4) (2) (3) fi + 13fi fi (1) (6) (2) (5) fi − 1040fi fi (3) (4) fi 160fi +3124fi −5040fi (1) fi (7) + 27216fi fi (5) (2) fi (6) −65604fi (3) 11 13 15 17 −12096fi (1) fi (8) − 67324fi (2) fi (7) +44352fi (3) (6) fi (−1. 1) (0. −135. (52) in which its ﬁrst k − 1 derivatives f (j) (0) = 0. The following computations were done with quadruple precision with 34 signiﬁcance digits to avoid the contamination of roundoﬀ errors in numerical results with decreasing ∆x. −514. 35. f (x) = xk exp(lx). We show in Table III the convergence rates for the classical WENO-JS. 0. the mapped WENO-M and for WENO-Z at a critical point of second order. 36. −3. Unless explicitly indicated otherwise.

the use of the mapping incurs on a signiﬁcant increase of the computational cost.0 4.00000e-02 2.8e-08 7. however.3 2.0 WENO-Z7 (p = 2) 4.7e-18 7.0e-16 5.50000e-02 1.0 1.25000e-02 6. WENO-M and WENO-Z schemes.0e-10 5.8e-06 7. A ﬁxed constant ǫ = 1 × 10−40 is used.6e-09 7.2 1. (2r − 1).12500e-03 1.2e-14 6.0 2.0 3.7e-18 6.5 8.9e-06 7.9e-18 6.2e-12 4. When a smaller value for ǫ is used the order of accuracy of the scheme degrades to third order.5e-04 1.1 WENO-M7 3. since the variation of the βk is understood as the indication of a high gradient. r 4 2r − 1 7 ncp 0 1 2 3 0 1 2 3 4 0 1 2 3 4 5 JS 7 5 4 3 9 7 6 5 4 11 9 8 7 6 5 M 7 7 6 3 9 9 9 7 4 11 11 11 11 8 5 Z (p = 1) 7 7 5 3 9 9 8 6 4 11 11 11 9 7 5 Z (p = 2) 7 7 6 3 9 9 9 7 4 11 11 11 11 8 5 Z (p = r − 1) 7 7 7 3 9 9 9 9 4 11 11 11 11 11 5 5 9 6 11 Table IV: Rates of convergence at critical points of increasing order ncp for (2r−1) order WENO-JS.8 6.7e-14 5.1 8.9 9. the βk have much smaller sizes and such an ǫ dominates their relative variations towards a central upwind ﬁfth order scheme.3e-14 8. if ncp < r − 1.0e-05 5.3e-10 7.1e-12 5.0e-11 4. WENO-M was presented as a ﬁx to this situation.8 9.1 2.8e-11 6.81250e-04 WENO-JS7 1.3 1.1e-04 2.7e-09 8. Critical points have become a point of discussion since when in [5] it was shown that the standard value of the parameter ǫ = 10−6 was in fact hiding the loss of accuracy of WENO-JS. as it was shown in [10]. Remark 11 It is shown in [5] that the formal order of the scheme.6e-07 5.0 2.2 1.0e-12 7. At critical points.3e-13 4.6 3.4 8.6e-08 5.3 8.00000e-01 5.4e-06 7.8 2.1 3.1e-20 7. For a detailed analysis of this accuracy enhancement of WENO-Z when p increases. can also be recovered with WENO-M by further applications of the mapping.∆x 1.9 7.9 2. see [10]. for it used a mapping that corrected the weights of WENO-JS and recovered the formal order of the scheme even when using very small values of 25 .3e-11 8.5 2.1e-04 0.2e-03 3.4e-16 7.56250e-03 7.4 1. the rate of convergence of WENO-Z can always be recovered to the fullest order.1 1.9e-16 6.25000e-03 3.0 WENO-Z7 (p = 3) 8.0 Table III: Rate of convergence at a second order critical point (ncp = 2) for the seventh order (r = 4) WENO-JS. WENO-M and WENO-Z schemes.

2 2. Based on the structure disclosed by their Taylor series in the last section.00000e-01 5. Thus. that is.6e-27 9. We will see that the ﬁnal amount of dissipation observed at the numerical solutions is a result of the sizes of the weights attributed to discontinuous substencils by each 26 . Table V below show numerical results for critical points of order r − 1.9 6.1e-21 7. for the WENO-Z schemes.0 5.0 2.0 WENO-Z11 6.50000e-02 1.0 1.8e-07 5.6e-11 5.0 7.5e-16 23.4e-19 10.3e-05 3.25000e-02 6.9e-14 5. Nevertheless. it is well known that at problems with shocks one cannot expect better order of accuracy than O(1).00000e-02 2.25000e-03 3. A smaller dissipation is achieved when close to discontinuities the scheme generates a set of nonlinear weights that provide a more centralized scheme and we will see in the next section that this is the case for the improved results of WENO-Z and WENO-M over WENO-JS.2e-15 7. k = r − 1 at (52).9 2. However.3e-05 5.7e-05 2.0 3.8e-14 15.2e-09 13.0 3. a conveniently size of ǫ can be chosen in order to play the role of this conditional. ﬂat and high gradients regions of the solution can be distinguished by the absolute sizes of the βk and a conditional statement could be used to improve the WENO schemes convergence at critical points.5e-19 9.9e-08 4.9e-11 5. the literature has yet to provide meaningful examples with shocks where the distinguished treatment of critical points provides any substantial improvement to the quality of the numerical solution.0 1. such an ǫ could be of size ∆x2 .6 3.0 1.ǫ.0 1.8e-16 5.0 2. ∆x 1.7 1.8e-16 9.56250e-03 7. The numerical results with WENO-M were superior at shock problems and this was credited to the improvement of the weights at critical points.8e-13 7. and this dominates any eventual increase of convergence order at critical points.1e-09 12. 5 Numerical Results In this section We compare the sets of nonlinear weights generated by the three WENO schemes discussed in this article. However.7e-17 7. Note that the formal order of accuracy is recovered and the same occurs for WENO-JS and WENO-M (not shown). since at critical points the sizes of the βk will be smaller than that. we will pay closer attention to the power parameter p and use the satisfactory variable value of ǫ = ∆xr−1 for all the numerical experiments.9e-13 5.0 Table V: Rates of convergence at a critical point of order ncp = r − 1 for (2r − 1) order WENO-Z schemes with power parameter p = 1 and ǫ = ∆x2 .81250e-04 WENO-Z5 2.8e-22 9. Nevertheless.5e-09 14.8e-23 11.0 1. Thus.7 8.3e-24 9.12500e-03 1.6e-29 11.0 5.1e-05 3. in [10] it was shown that the extra sharpness obtained at discontinuities by WENO-M was in fact due to its smaller dissipation.1 1.0 1. This is a compromise value that avoids the original predominance of a ﬁxed value of ǫ and also sets a lower bound for the ratio of the smoothness indicators.0 6. In the numerical experiments of next section we look closer to the issue of the numerical dissipation as a result of qualitative diﬀerences among the distinct sets of nonlinear weights of the WENO schemes studied in this article. Remark 12 The critical points issue has been the departure point of several investigations on improved sets of WENO nonlinear weights.0 WENO-Z7 2.3e-26 11.0 WENO-Z9 2.4e-19 7.0 1.3 1.9e-33 11.8 6.0e-10 5.

554e-07 WENO-M5 9. 5. WENO-Z and WENO-M.808e-03 WENO-Z5 9. r = 3.999e-01 3.1 Discontinuities and nonlinear weights We now compare the diﬀerent sets of normalized nonlinear weights ωk generated by WENO-JS.886e-03 WENO-M5 9.015. k = 0.728e-01 1. a Gaussian. At x = −0. g(x) = g(x − 2).272e-01 8.45. The number of uniformly spaced grid points used is N = 200.064e-04 WENO-JS5 9.427e-01 8.946e-01 4.979e-06 5. We end this section showing numerical simulations of the two dimensional Mach 10 double-Mach shock reﬂection problem using high orders WENO-Z schemes with increasing resolutions. x3 2 .426e-01 8.671e-03 WENO-M5 1.713e-01 1.200e-06 Table VI: Nonlinear weights ωk .426e-01 8. WENO-Z and WENO-M schemes with power parameters p = 1 and p = 2 are displayed. We ﬁx ǫ = 1 × 10−40 in the discussion of this section. a square and an ellipse functions. WENO-Z and WENO-M schemes with power parameters p = 1 and p = 2 at locations x = −0.919e-01 6. where the substencils S0 and S1 are smooth and only S2 is nonsmooth: • WENO-Z assigns a larger weight ω2 for the nonsmooth substencil S2 than WENO-JS.425e-06 2.0015 WENO-Z5 1.063e-06 1.187e-07 x = −0. Analyzing the results shown at Table VI and at Figure 2.768e-07 x = −0.190e-02 3. 27 . while WENO-M assigns an even larger one as evidenced in the Table.116e-04 WENO-Z5 1. In ﬁgure 2. the nonlinear weights ωk of the ﬁfth order WENO-JS.005 WENO-Z5 9.509e-03 9.426e-01 8.567e-01 6. 2 of the ﬁfth order WENO-JS.570e-01 3.0015 and x = −0.060e-04 WENO-JS5 1.574e-01 4. Z • Note also that ω2 = 2ω2 . (53) with a single discontinuity located at x = 0. for the ﬁfth order WENO schemes. 1.271e-01 8.999e-01 3.999e-01 3. p = 2 ω0 ω1 ω2 WENO-JS5 1. p = 1 ω0 ω1 ω2 r = 3. we see that: 1. Consider the following discontinuous function g(x) = − sin(πx) − x3 2 . a straightforward computation justifying this fact is given in [10].376e-07 WENO-M5 1.303e-03 1.scheme. The time dependent problems are all solved via the third order Runge-Kutta TVD scheme with CFL = 0.842e-01 1.574e-01 2. the one dimensional Euler Equations of compressible gas dynamics with Riemann initial condition such as the Lax problem and the one dimensional Mach 3 shock-density wave interaction.198e-06 WENO-JS5 9. We also show numerical experiments with the one dimensional scalar advection equation with initial condition consisted of a triangle. 1 − sin(πx) − −1 < x < 0 0≤x≤1 .005.

and also once again is even larger with WENO-M. • Nevertheless.WENO-JS5 WENO-Z5 p=1 10 10 10 10 10 10 0 WENO-M5 10 10 10 10 10 10 0 10 10 10 10 0 -1 -1 -1 -2 -2 -2 -3 -3 -3 -4 -4 10-4 10 -5 -5 -5 10-6 10 10 -7 10-6 10-6 ω0 ω1 ω2 -0.01 0. At x = −0.01 0. is signiﬁcantly larger with WENO-Z than in WENO-JS.03 10 10 -7 ω0 ω1 ω2 -0. increasing the value of p. This occurs because the increase of the smallest weight is compensated due to the normalization of the weights. 2. Moreover. Also.01 0.01 0.01 0. We see that WENO-M assigns larger weights than WENO-Z for discontinuous substencils for the same value of p.01 0. in its turn. 2 of the ﬁfth order (r = 3) WENO-JS. making 28 .03 -0. WENO-Z assigns larger weights to discontinuous substencils than WENO-JS for the same value of p.03 -0. ω2 is the smallest weight since the local lower order polynomial approximation of g(x) at x = 0 is computed through an extrapolation of the values in S2 .03 -0.03 -8 -8 -8 p=2 10 10 10 10 10 10 0 10 10 10 10 10 10 0 10 10 10 10 0 -1 -1 -1 -2 -2 -2 -3 -3 -3 -4 -4 10-4 10 -5 -5 -5 10-6 10 10 -7 10-6 10-6 ω0 ω1 ω2 -0. whileS1 and S2 both contain the discontinuity: • Here.03 10 10 -7 ω0 ω1 ω2 -0. ω0 = ω0 > ω0 .03 -8 -8 -8 Figure 2: Nonlinear weights ωk . double the value. Since a bigger weight for a discontinuous stencil contributes to a more central upwind approximation.01 0.03 -0. WENO-Z and WENO-M schemes with power parameters p = 1 and p = 2.01 0. decreases the absolute sizes of the weights assigned to discontinuous substencils.03 -0. not always the weights of WENO-M are larger than the corresponding M Z ones of WENO-Z and WENO-JS.01 0.01 0. k = 0. ω2 .03 10 10 -7 ω0 ω1 ω2 -0.01 0.005. we may conclude that there is a hierarchy of dissipation that puts WENO-M as the least dissipative scheme and WENO-JS as the most dissipative one. Note once again that this smallest weight. 1.01 0.03 -0.03 10 10 -7 ω0 ω1 ω2 -0. for instance. Also. only the substencils S0 is smooth. a greater value of p decreases the absolute sizes of the weights assigned to discontinuous substencils by all three schemes. while WENO-Z assumes an intermediary position.

01 0. z − δ) + 4G(x.2] . 0. β.05 10 -10 10 -10 10 -10 Figure 3: Nonlinear weights ωk . 2 .4.01 0. x ∈ [0. see Figure 3 where we show analogous graphs for the nonlinear weights of the seventh order case. −0.03 -0.03 -0. t = 0) = 1 [F (x.05 -0.03 0. 1).05 10-8 ω0 ω1 ω2 ω3 -0.03 0.01 0. a − δ) + 4F (x.01 0. −0. α.6] .03 -0. x ∈ [−0. for instance. 0. x ∈ [0. WENO-Z and WENO-M schemes.05 -0. α.03 0. with power parameters p = 1 and p = 2. β. β.05 10 -10 10 -10 10 -10 p=2 10 0 10 0 10 0 10 -2 10 -2 10 -2 10-4 10-4 10-4 10-6 10-6 10-6 10-8 ω0 ω1 ω2 ω3 -0.1)] u(x.01 0.05 -0. k = 0. 0).2 Linear Advection Consider the one dimensional linear wave equation.03 -0.for a more lateral linear combination of the substencils and.03 0.01 0. ∂t ∂x x ∈ [0.4. z + δ)] 1 1 − [10(x − 0.03 -0.05 10-8 ω0 ω1 ω2 ω3 -0. a + δ )] 6 0 G(x. 3 of the seventh order (r = 4) WENO-JS.03 -0.05 -0. We remark that this is a general behavior that does not depend on the order of the scheme. a) + F (x. else F (x .05 -0.2] . 1. therefore. z) + G(x. α. 29 . a) = max(1 − α2 (x − a)2 .01 0.03 0.05 10-8 ω0 ω1 ω2 ω3 -0. 5.05 10-8 ω0 ω1 ω2 ω3 -0. WENO-JS7 WENO-Z7 p=1 10 0 WENO-M7 10 0 10 0 10 -2 10 -2 10 -2 10-4 10-4 10-4 10-6 10-6 10-6 10-8 ω0 ω1 ω2 ω3 -0. z) = e−β(x−z) .01 0.03 0.05 -0. 2. α.6] .01 0.8. ∂u ∂u + = 0.01 0. increases dissipation. β. with an initial condition given by 1 6 [G(x. x ∈ [−0.01 0.01 0.

5. The third-order Runge-Kutta TVD scheme 2r−1 uses an adjusted time step ∆t = CFL × ∆x 3 in order to maintain the convergence rate of the underlying spatial WENO scheme. This is the one dimensional scalar 2 linear advection problem with an initial condition consisting of a smooth Gaussian. In summary. a high gradient smooth post-shock region and multiple shocklets that develop in a later time. We interpret the behavior of the solution by looking at the error. all schemes show no oscillations.7. a = 0. WENO-JS.log where z = −0. one can observe that • for p = 1. δ = 0. The standard one dimensional shock-density wave interaction is generally preferred. Details of the setup of this problem can be found in [2].4 One dimensional Shock-density wave interaction In the Lax problem.3 One dimensional Euler Equations: The Lax Problem This same behavior is observed in Figure 5 where we used the ninth order. r = 5. The results in Figure 4 show the computed solutions and the absolute point-wise errors for p = 1. WENO-Z and WENO-M schemes. otherwise. 30 . a piecewise linear triangle function and a smooth elliptic function. since the solution of this problem consist of a main shock. the oscillations are very small if variations in the error curve are close to the bottom of the log scale. Once again we note that WENO-M requires the highest value of p among all three schemes in order to become non-oscillatory. 2 and r − 1. Also. a discontinuous Heavside function. Symbols represent the numerical solution while the lines indicate the absolute point-wise error at each grid point. • for p = 2.5 and α = 10. β = 36δ2 .005. WENO-JS and WENO-Z achieve the ENO property for the square wave. • increasing the number of grid points will reduce the error away from the discontinuities in the solution and in its derivatives (not shown). Periodical boundary conditions are imposed on the two ends of the domain. all three schemes do not have enough dissipation to simulate the advection of the square wave without oscillations. one may infer the presence of oscillations. the solution is a piece-wise linear function which does not fully justify the cost of using the high order reconstruction process of the WENO schemes. all of these requiring high order schemes in order to be eﬃciently and accurately represented. • for p = r − 1. and their solutions are very similar in what regards the sharp approximation of corners. We compute the numerical solutions for all three schemes of order 11 (r = 6) with N = 200 uniformly spaced grid points and ﬁnal time T = 8. The black solid line is the exact solution. 5. however WENO-M is still oscillatory. A monotone error curve away from the discontinuity implies a non-oscillatory solution. for the numerical simulation of the Lax problem.

5 1 x x (r = 6.6 10-3 0.4 10 -5 0.5 0 0. The number of grid points used is N = 200.5 0 0. The symbols indicate the numerical solution at the grid points. The solid black line is the exact solution.8 Exact Solution 10-1 10-1 Error 10-3 0.4 10 -5 Error 0. p = r − 1) 10 3 Exact Solution WENO-JS WENO-M WENO-Z 1 10 1 0.5 1 x Figure 4: Numerical solutions of the linear advection equation at ﬁnal time t = 8 computed by the eleventh order (r = 6) WENO-JS. The lines show the absolute point-wise error at each grid point.8 10-1 Error 0.2 10 -7 0 10 -9 -1 -0. r − 1. p = 1) 10 3 (r = 6.(r = 6. p = 2) Exact Solution WENO-JS WENO-M WENO-Z 10 1 3 Exact Solution WENO-JS WENO-M WENO-Z 1 10 1 10 1 0.2 10 -7 0 10 -9 0 10 -9 -1 -0.4 10 -5 0. WENO-M and WENO-Z with power parameters p = 1.6 10-3 0.2 10 -7 0.5 1 -1 -0.5 0 0. Exact Solution 31 Exact Solution .6 0.8 0. 2.

2 -4 -2 0 2 4 x Figure 5: Numerical solution of the Lax problem as computed by the ninth order (r = 5) WENO-JS.2 -4 -2 0 2 4 x x (r = 5.4 Exact solution WENO-JS WENO-M WENO-Z 1.4 0. p = r − 1) 1.2 1.4 Exact solution WENO-JS WENO-M WENO-Z 1. p = 2) 1. p = 1) 1.4 0.2 -4 -2 0 2 4 0.4 Exact solution WENO-JS WENO-M WENO-Z (r = 5.6 0. 2. r − 1.8 0.(r = 5.4 0.6 0.8 0. WENO-M and WENO-Z schemes with power parameters p = 1.2 1 1 ρ 0. 32 .2 1 ρ 0.6 0.8 ρ 0.

p = 2) Exact Solution WENO-JS WENO-M WENO-Z (r = 5. 2. p = 2) Exact Solution WENO-JS WENO-M WENO-Z (r = 4.(r = 3. 11 with power parameters p = 1. p = 1) Exact Solution WENO-JS WENO-M WENO-Z (r = 4. The number of grid points used is N = 200. WENO-M and WENO-Z schemes of order 2r − 1 = 5. 7. p = 2) Exact Solution WENO-JS WENO-M WENO-Z (r = 3. 33 . p = 2) Exact Solution WENO-JS WENO-M WENO-Z (r = 6. p = r − 1) Exact Solution WENO-JS WENO-M WENO-Z 4 4 4 3 3 3 Rho Rho 2 2 Rho 2 1 1 -4 -2 0 2 4 1 -4 -2 0 2 4 -4 -2 0 2 4 x x x (r = 4. p = r − 1) Exact Solution WENO-JS WENO-M WENO-Z 4 4 4 3 3 3 Rho Rho 2 2 Rho 2 1 1 -4 -2 0 2 4 1 -4 -2 0 2 4 -4 -2 0 2 4 x x x (r = 5. p = r − 1) Exact Solution WENO-JS WENO-M WENO-Z 4 4 4 3 3 3 Rho Rho 2 2 Rho 2 1 1 -4 -2 0 2 4 1 -4 -2 0 2 4 -4 -2 0 2 4 x x x Figure 6: Numerical solution of the shock-density wave interaction as computed by the WENO-JS. 9. p = 1) Exact Solution WENO-JS WENO-M WENO-Z (r = 3. p = r − 1) Exact Solution WENO-JS WENO-M WENO-Z 4 4 4 3 3 3 Rho Rho 2 2 Rho 2 1 1 -4 -2 0 2 4 1 -4 -2 0 2 4 -4 -2 0 2 4 x x x (r = 6. r − 1. p = 1) Exact Solution WENO-JS WENO-M WENO-Z (r = 6. p = 1) Exact Solution WENO-JS WENO-M WENO-Z (r = 5.

making an angle of 60 degrees with the horizontal axis. The equations are the two dimensional Euler equations (γ = 1. 1.(r = 6.5 . WENO-M and WENO-Z schemes of order 2r − 1 = 11 with power parameters p = 1. 2. Py ) = (0. in the x-axis. reﬂecting boundary conditions ∂x are applied to the interval [x0 . p = 2) Exact Solution WENO-JS WENO-M WENO-Z (r = 6.25 sin π . 8. 0. using N = 200 and N = 250 number of uniformly spaced grid points. we apply the high order conservative characteristic-wise WENO-Z ﬁnite diﬀerence scheme to the two dimensional double-Mach shock reﬂection problem [19] where a vertical shock wave moves horizontally into a wedge that is inclined by some angle. simulating the wedge: (ρy . WENO-M scheme is the least one. Boundary conditions at x = 0 are inﬂow. and at x = 4 we have outﬂow boundary conditions with ∂Q = 0. 0. r − 1. the ﬂow has to be imposed such that there is no interaction with the moving shock. uy . one can observe from ﬁgures 6 and 7. P ) = y . 4] × [0. at Finally.25 cos π . 0. with post-shock values as above. which have very similar computational costs (see [10]). respectively. The domain of the problem is [0. r − 1. p = r − 1) Exact Solution WENO-JS WENO-M WENO-Z 4 4 4 3 3 3 Rho Rho 2 2 Rho 2 1 1 -4 -2 0 2 4 1 -4 -2 0 2 4 -4 -2 0 2 4 x x x Figure 7: Numerical solution of the shock-density wave interaction as computed by the WENO-JS. 0) for x0 ≤ x < 4 and y = 0. resulting from their distinct treatment of discontinuous substencils. At the upper boundary. 1] and the shock moves diagonally at Mach 10. in this one dimensional test case. x ≥ x0 + √ 3 34 . 5. p = 1) Exact Solution WENO-JS WENO-M WENO-Z (r = 6. −8. 116. The number of grid points used is N = 250. WENO-M scheme provides a better resolution of the small scales high frequency structures behind the main shock with respect to WENO-Z. and all of this is due to their distinct levels of dissipation. as in the examples before. and WENO-Z scheme occupies an intermediary position for any ﬁxed order (2r − 1) and power parameter p = 1. At y = 0. 2. v. v.4) and initial conditions are given by y 8. 0. that the WENO-JS scheme is the most dissipative one. x < x0 + √ 6 6 3 Q = (ρ. The exact location of the shockwave at y = 1.5 Double-Mach shock reﬂection problem 1 with x0 = 6 .0) . u. 4]. notice that all the WENO solutions are very similar once a suﬃcient number of grid points (N = 250) is used and this is a point against WENO-M. (1. Similarly. are approximately 20% to 30% more expensive than WENO-JS and WENO-Z schemes. Nevertheless. and the same can be said about WENO-Z with regards to WENOJS.4. since its CPU costs. y = 1.

800 × 200 and 1600 × 400 at time t = 0. Here. v. For this problem.5 . 9. given respectively by: instant t. is given by s(t) = x0 + Q|y=1 = (ρ. However.4.2. (1. 8. 0.25 cos π . the power parameter p = r − 1 is used.2 is in general very similar across diﬀerent schemes. In Figure 9 we display the region around the double Mach stems in order to observe the numerical solutions of the WENO-Z scheme of orders 2r − 1 = 5. Figure 8: Density contours of the double-Mach shock reﬂection as computed by the eleventh order WENO-Z scheme at time t = 0. Additionally. We see from the ﬁgure that one may obtain similar results by increasing the order of approximation and decreasing the number of points in the spatial discretization. 6 Conclusions We extended the new WENO schemes introduced in [10] to rates of convergence higher than 5 by providing a formula in closed form for the higher order smoothness indicators τ2r−1 as a simple linear combination of the lower order smoothness indicators βk .25 sin π . 1. Both results were obtained from a thorough study of the properties of the Taylor expansions of the lower order smoothness indicators departing from the symmetric structure of their underlying Lagrangian interpolating polynomials with respect to the geometrical disposition of the global and the local WENO stencils.(1 + 20t) √ . we set post-shock and pre-shock conditions before and 3 after this location. 0 ≤ x < s(t) 6 6 . the resolving power of the higher order WENO schemes can be determined by the number of small vortices that can be captured along the slip line and the wall jet behind the lower half of the right moving shock by the underlying scheme with a given resolution. The formation 35 . The global structure of the solution at time t = 0.0) . the CPU cost of increasing the order is smaller than the one of increasing the spatial resolution. −8. P ) |y=1 = 8. 0. 116. as it was shown in [4]. s(t) ≤ x ≤ 4 Numerical results of this problem are well documented in the literature (see also ﬁgure 8) and further details can be found in [19]. We also discussed the lack of convergence at critical points of smooth solutions and revisited the ǫ issue of the classical WENO-JS weights showing that for practical matters it may work as an implicit conditional that distinguishes ﬂat regions from discontinuities. Another formula for the maximum order of convergence that such linear combinations can achieve was also provided. resolutions and parameters. p ≤ 2 is unstable for WENO-M (r > 3) and WENO-Z (r > 4) and this can be attributed to the insuﬃcient dissipation of the schemes. u.2. 11 using three diﬀerent uniform meshes with resolutions 400 × 100 .

4 0.2. 9.8 2.9 3 3.3 0.9 3 3.3 0.7 2.2 3 0 2.2 3 (r = 3) 0.1 0.5 (r = 6) 0.7 2.1 3. with resolutions at 400 × 100.1 0.1 0 2. 11.9 3 3.4 0.2 0.2 0.7 2.4 0.1 0.1 0 2.9 3 3.2 0.2 3 (r = 3) 0.7 2. 36 .4 0.3 0.9 3 3.8 2.7 2.3 0.4 0.1 3.5 (r = 6) 0.8 2.8 2.1 0.2 0.2 3 0 2.2 3 0 2.4 0.5 N = 400 × 100 (r = 5) 0.1 3.2 0.3 0.8 2.2 3 0 2.2 0.2 0.2 0.1 0.4 0.3 0.9 3 3.5 (r = 6) 0.7 2.1 0.4 0.8 2.1 3.5 N = 1600 × 400 (r = 5) 0.7 2.5 0. The power parameter p = r − 1 is used in these simulations.1 3.1 3.2 3 0 2.3 0.4 0.5 0.9 3 3.(r = 3) 0.8 2.5 N = 800 × 200 (r = 5) 0.8 2.8 2.2 0.1 3.7 2.7 2.5 0.1 3. 800 × 200 and 1600 × 400 at time t = 0.9 3 3.3 0.3 0.2 3 Figure 9: Density contours of the double-Mach shock reﬂection problem as computed by the high order WENO-Z scheme of orders 2r − 1 = 5.1 3.2 3 0 2.9 3 3.1 0 2.

W. 207. pp. pp. W. L. we concluded that regarding dissipation WENO-Z occupies an intermediary position between WENO-JS and WENOM. High resolution schemes for hyperbolic conservation laws. Balsara and C. Eﬃcient Implementation of Essentially Non-oscillatory ShockCapturing Schemes. W. D. Phys. Don. Carmona. W. Comp. Costa and W. 3101–3211 (2008) 37 . Approximated Riemann solvers. M. Osher. Henrick. and diﬀerence schemes. Powers. IM-UFRJ. II. a References [1] D. M. Mapped weighted essentially non-oscillatory schemes: Achieving optimal order near critical points. 83. Eﬃcient Implementation of Essentially Non-oscillatory ShockCapturing Schemes. WENO-Z and WENO-M were analyzed and numerical arguments were provided to explain their inﬂuence in the distinct numerical features of each scheme and their changes with the increasing of the power parameter p. 690–696 (2003) [5] A. pp. Shu and S. Comp. K. Essentially Non-Oscillatory and Weighted Essentially Non-Oscillatory Schemes for Hyperbolic Conservation Laws. J. Comp. 97-65 [4] J. Comp. parameter vectors. Comput. Phys. No.564/2008. J. J. pp. Comp. Harten. 405–452 (2000) [2] G. Resolution of high order WENO schemes for complicated ﬂow structures. Comp. Osher. Zhang and C. 357–372 (1981) [7] A.of the distinct sets of nonlinear weights for the WENO-JS. S. Phys. J. T. 160. Shu. 7 Acknowledgments The ﬁrst and second authors have been supported by CNPq. grants 300315/98-8 and FAPERJ E-26/111. B. 202–228 (1996) [3] C. W. Shi. pp. Phys. pp. S. Numerical results with the system of the Euler Equations in 1D and 2D were presented to show that the new high order WENO-Z schemes perform well at the treatment of numerical solutions containing both discontinuities and high order smooth structures. Comput. Phys. Jiang and C. 32–78 (1989) [10] R. Phys. Shu. J. Phys. 1. Phys. 542–567 (2005) [6] P. 227 No. J. 126. pp. for hosting his visit during the course of the research. J. 357–393 (1983) [8] C. NASA/CR-97-206253 ICASE Report No. The author would also like to thanks the Departamento de Matem´tica Aplicada. Aslam and J. J. 6. J. W. pp. 49. Comp. pp. Y. 43. Monotonicity preserving weighted essentially non-oscillatory schemes with increasingly high order of accuracy. Shu. T. Shu. 186. 77. An improved weighted essentially nonoscillatory scheme for hyperbolic conservation laws. Roe. Shu and S. Phys. Eﬃcient Implementation of Weighted ENO Schemes. 439–471 (1988) [9] C. The third author (Don) would like to thank the support provided by the FRG grant FRG08-09-II-12 from Hong Kong Baptist University and the RGC grant HKBU-092009 from Hong Kong Research Grants Council. Borges.

H. Don. pp. Multi-Domain Hybrid Spectral-WENO Methods for Hyperbolic Conservation Laws. O. accepted. Jacobs and W. 54. Don. Advances in Applied Mathematics and Mechanics 1 No. Phys. Costa and W. S. Comput. Jung and W. 221 No. The numerical simulation of two dimensional ﬂuid ﬂow with strong shocks. Don and T. 6. Eﬀects of order of WENO ﬂux reconstruction and spatial resolution on reshocked two-dimensional Richtmyer-Meshkov instability. 805–836 (2007) [12] B.[11] M. N. Woodward and P. S. 970–991 (2007) [14] G. Latini. 2. Vallet. Comp. pp. J. Phys. J. A. pp. Schal and I. Phys. W. Hu. 209–218 (2007) [13] B. A. J. J. Comput. 204 No. Comput. pp. Don. 224 No. J. 5. 228. On the HLLC Riemann solver for interface interaction in compressible multi-ﬂuid ﬂow. Appl. pp. Comput. Collela. 2. J. Very High Order WENO Schemes. 228 No. pp. Phys. A High Order WENO-Z Finite Diﬀerence Methods Based ParticleSource-in-Cell method for Particle-Laden Flows with Shock. S. Comp. Comput. Costa and W. pp. High-Order Resolution of Eulerian-Lagrangian Simulations of Particle Dispersion in the Accelerated Flow behind a Moving Shock. (2010) [17] J. Don. S. Math. S. 22. B. S. Jacobs. J. Theoretical and Computational Fluid Dynamics. Gerolymos. 6572–6589 (2009) [19] P. Collocation Methods for Hyperbolic Partial Diﬀerential Equations with Singular Sources. 8481–8524 (2009) [15] G. Dittmann. D. Don. Phys. 769–780 (2009) [18] X. Schilling and W. 115–173 (1984) 38 . Y. Iaccarino. 2. 1365–1379 (2009) [16] G. High Order Hybrid Central-WENO Finite Diﬀerence Methods for Conservation Laws. pp. Phys. Adams and G.

- LA Sheet7 12
- 5.Method of Using Leibnitz
- Mathematical Symbols
- A Study of Numerical Methods for Hyperbolic Conservation Laws with Stiff Source Terms
- Full Text
- Polynomial Interpolation and neville's algorithm
- Rank of Matrix
- Main
- NT_503_504
- Linear Algebra - Friedberg - 4th Ed - Some Notions
- econometrics_I definitions , theorems and facts
- BCAnew
- 10.1.1.31.4358
- ga tau ini judulnya apa lah yaa
- Matrices
- Information and Pool Sap2000 Manuals English Problem 6-006
- Numan Al
- NumericalAnalysisHistory 2
- e 174460
- 1-1-2
- [Jean-Marc_Vanden-Broeck]_Gravity-Capillary_Free-S(BookSee.org).pdf
- Belytcho Galerkin Element-free Methods
- New Microsoft Word Document (3)
- Mathematics and Algorithms
- Miii
- Simpsons_rules.pdf
- Me to Do Newton Krylov
- Heun 2 Correction Example
- 2233-mt1-S99s
- The Dynamic Temperature Field of Two Stage UCG

Skip carousel

- UT Dallas Syllabus for cs4334.501.09f taught by Janos Turi (turi)
- Simulation and Modeling MCQ'S
- UT Dallas Syllabus for math6313.501.08f taught by Janos Turi (turi)
- tmp51DE.tmp
- tmp846E
- ANALYSIS OF A TWO DIMENSIONAL RECTANGULAR FIN USING NUMERICAL METHOD AND VALIDATING WITH ANSYS
- UT Dallas Syllabus for cs4334.501 06f taught by Bentley Garrett (btg032000)
- Mathematical Methods for Numerical Analysis and Optimization
- tmpED85.tmp
- UT Dallas Syllabus for phys3330.501 06f taught by Roy Chaney (chaneyr)
- UT Dallas Syllabus for phys3330.501.08f taught by Roy Chaney (chaneyr)
- UT Dallas Syllabus for ee6481.021.06u taught by Cyrus Cantrell (cantrell)
- tmp79F5.tmp
- UT Dallas Syllabus for math6313.501.10s taught by Janos Turi (turi)
- UT Dallas Syllabus for phys3330.501.09f taught by Roy Chaney (chaneyr)
- tmpDC91.tmp
- UT Dallas Syllabus for math4334.501 06f taught by Bentley Garrett (btg032000)
- UT Dallas Syllabus for math6313.501.11f taught by Janos Turi (turi)
- UT Dallas Syllabus for math4334.501 05f taught by Bentley Garrett (btg032000)
- UT Dallas Syllabus for math6318.501.09s taught by Janos Turi (turi)
- tmp4AB5.tmp
- CFD Analysis of Air Flow Modelling in Apple Cold Storage
- UT Dallas Syllabus for math6313.501.10f taught by Janos Turi (turi)

Sign up to vote on this title

UsefulNot usefulClose Dialog## Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

Close Dialog## This title now requires a credit

Use one of your book credits to continue reading from where you left off, or restart the preview.

Loading