Local Discontinuous Galerkin Methods for Partial ... - CiteSeerX

40 downloads 2094 Views 1MB Size Report
STI Help Desk at. (301 ) 621-0390. Write to: NASA STI Help Desk. NASA Center for AeroSpace. Information. 7121 Standard Drive. Hanover, MD 21076-1320 ...
NASA/CR-2002-211959 ICASE

Report

No. 2002-42

Local Discontinuous Differential

Equations

Jue Yan and Chi-Wang Brown

University,

November

2002

Galerkin

with Higher

Shu

Providence,

Methods

Rhode Island

for Partial

Order Derivatives

The

Since

its founding,

NASA

NASA

STI

has been dedicated

Program

Office...

in Profile

CONFERENCE

PUBLICATIONS.

to the advancement of aeronautics and space science. The NASA Scientific and Technical

Collected

papers

from scientific

technical

conferences,

Information (STI) Program Office plays part in helping NASA maintain this important role.

seminars, or other meetings cosponsored by NASA.

a key

The NASA STI Program Office is operated by Langley Research Center, the lead center for NASA's scientific and technical information. The NASA STI Program Office provides access to the NASA STI Database, the

SPECIAL technical, NASA

and

symposia, sponsored

or

PUBLICATION. Scientific, or historical information from

programs,

projects,

and missions,

often concerned with subjects substantial public interest.

having

largest collection of aeronautical and space science STI in the world. The Program Office is also NASA's institutional mechanism for

TECHNICAL TRANSLATION. language translations of foreign

Englishscientific

disseminating development

and technical material NASA's mission.

to

the results of its research and activities. These results are

published by NASA in the NASA STI Report Series, which includes the following report types:

TECHNICAL PUBLICATION. Reports completed research or a major significant phase of research that present the results

of

of NASA programs and include extensive data or theoretical analysis. Includes

pertinent

Specialized services that complement the STI Program Office's diverse offerings include creating custom thesauri, building customized data bases, organizing and publishing research

results..,

even providing

For more information Program

Office,

videos.

about the NASA

STI

see the following:

compilations of significant scientific and technical data and information deemed to be of continuing reference counterpart of peer-reviewed professional papers, stringent limitations length and extent presentations.

TECHNICAL

value. NASA's formal

but having less on manuscript

Access the NASA STI Program Page at http://www.sti.nasa.gov

Email your question

of graphic

Scientific and technical findings that are preliminary or of specialized interest,

Fax your question to the NASA Help Desk at (301) 621-0134

e.g., quick release reports, working papers, and bibliographies that contain minimal annotation. Does not contain

Telephone the NASA (301 ) 621-0390

extensive

Write to:

analysis.

NASA technical contractors

findings

REPORT.

Scientific

by NASA-sponsored

and grantees.

via the lnternet

to

[email protected]

MEMORANDUM.

CONTRACTOR

Home

and

STI Help Desk at

STI Help Desk

NASA Center 7121 Standard Hanover,

STI

for AeroSpace Drive

MD 21076-1320

Information

NASA/CR-2002-211959 ICASE

Report

No.

2002-42

Local Discontinuous Differential Jue

Yan and

Brown

Equations

Chi-Wang

University,

Galerkin

Methods

with Higher

for Partial

Order Derivatives

Shu

Providence,

Rhode

Island

ICASE NASA

Langley

Research

Hampton,

Virginia

Operated

by Universities

Center

Space

Research

Association

Prepared for Langley Research under Contract NAS 1-97046

November

2002

Center

Available

from the following:

NASA Center 7121 Standard Hanover,

for AeroSpace Drive

MD 21076-1320

(301) 621-0390

Information

(CASI)

National

Technical

Information

5285 Port Royal Road Springfield,

VA 22161-2171

(703) 487-4650

Service

(NTIS)

LOCAL

DISCONTINUOUS

GALERKIN

EQUATIONS

WITH

HIGHER

JUE

Abstract.

In this

for solving space

time

local

second

correct

derivatives

and

technique,

with

are shown

originally

negligible

words,

stability,

error

Subject

discontinuous estimate,

derivative cost,

for linear

type

nonlinear

we present

We then

equations

problems.

estimates,

Numerical

in effectively

in-

develop

new

involving

fourth

we present

Preliminary

numer-

on a post-processing on the local discontin-

experiments

doubling

as well as some

equations

new methods

new results

error

and multiple

diffusion

derivatives.

For these

derivatives.

cases,

in one

methods

show that

the rate

nonlinear

this

of convergence

problems,

with

a local

Introduction.

Applied

we review

partial

We consider

A hyperbolic

method,

partial

and Numerical

In this paper

time dependent

dimensions.

derivatives.

Galerkin

differential

equations

with

higher

derivatives,

post-processing

classification.

ods for solving space

computational

third

negative-order

with higher

derivatives

for convection

bi-harmonic

Finally,

with good

order

derivatives.

for general

methods.

to equations

higher

Galerkin

mesh.

Key

1.

L 2 stability

new local discontinuous

involving

fifth

DIFFERENTIAL

_

methods

dependent

and prove

as well for the new higher additional

time

involving

for methods

applied

equations

for the

these

with

Galerkin

equations

to illustrate

designed

methods

works

uniform

methods

fluxes

equations

type

SHU

develop

PARTIAL

DERIVATIVES*

CHI-WANG

and

discontinuous

FOR

ORDER

AND

the existing

for KdV

differential

numerical

uous Galerkin technique

and

YAN f

differential

local

Galerkin partial

interface

ical examples

partial

We review

discontinuous

derivatives,

we review

dependent

dimensions.

volving

paper

METHODS

existing

differential

a sequence

conservation

Mathematics and develop

equations

new local discontinuous

of such partial

with higher differential

order

derivatives

equations

Galerkin

meth-

in one and multiple

with increasingly

higher

order

law d

Ut + E

fi(U)x,

= 0

(1.1)

i_1

is a partial

differential

equation

with

first derivatives.

The convection

d

d

Ut + E

f,(U)x,

where

(aij(U))

derivatives.

is a symmetric,

The general

KdV

type

E

i=1 *Research

supported and

Contract

by

ARO

grant

NAS1-97046

Hampton, VA 23681-2199, and AFOSR ¢Division of Applied Mathematics, SDivision

of Applied

Mathematics,

definite

(aij(U)U_)_:,

= 0

(1.2)

j=l

matrix,

is a partial

differential

equation

with

second

equation

Ut+Efi(U)x,+

NCC1-01035

E

i=l

semi-positive

equation

d

- E

i----1

diffusion

r_(U)

i=l DAAD19-00-1-0405, while

the

second

Egij(ri(U)x,)xj

=0

j=l NSF author

grants was

(1.3)

xi DMS-9804985 in residence

and

ECS-9906606,

ICASE,

NASA

at

NASA Langley

grant Brown

F49620-99-1-0077. University, Providence,

R!

02912.

E-mail:

yjue_cfm.brown.edu

Brown

University,

RI

02912.

E-mail:

shu_cfm.brown.edu

Providence,

Langley

Research

grant Center,

is a partial

differential

equation

with third

derivatives.

d

+ E(ai(Ux,)Ux,x,)x,x,

i=1

we just

differential

present

equation

(1.4)

with

as an example.

fourth

derivatives,

The following

differential

be presented. physical

equation

present

(1.5)

All these

The type of discontinuous finite

element

Runge-Kutta

time

derivatives section

review

paper

containing

applied.

in a series

equation

A naive

containing

higher

derivatives

method

on the

general

but

of papers

could

or higher

state

letters

be more

derivatives

counterparts,

U etc.

general

could

appear

to denote

also

often

in

the solutions

to

solutions.

developed

in this paper, with

using a discontinuous

explicit,

for the

nonlinearly

conservation

[8, 9, 6, 4, 10].

method

laws

(1.1)

We will briefly

[5], other

papers

in that

Galerkin

stable

review

as well as its implementation

paper

derivatives

to approximate

variables

is why the

method

first

local

and

high

order

containing

first

this

method

in

applications,

Springer

Shu [20] developed

In both

nonlinear

examples

were shown

reviewed

in sections

a "local"

volume,

method second

Rebay

we

and the

[11] and [20], suitable of the methods

to illustrate

the

stability

equation

we develop

(1.4) involving

fluxes

the new development

new local discontinuous derivatives,

and

Galerkin

and in section

of interface

of all the auxiliary

Shu

[11], for the by the

These

Yan and

(1.3) containing

were given,

for the linear

con-

successful

Later,

equation

internees

methods.

in later

Galerkin

of all the auxiliary

equations.

KdV type

estimates

with

in [11].

Navier-Stokes

of these

design

work was motivated

at element

to the

equations

the discontinuous

by Cockburn

for the general

as well as error

differential

The local solvability

Their

directly

in the computation

is the correct

method

at

[18, 12].

and local solvability

Galerkin

was developed

and accuracy

fourth

methods

stability

derivatives.

numerical

L 2 stability

paper

of such

be

"non-conforming"

nicely

apply

cannot

discontinuous

method

solution

partial

then

of the solution.

method

behaves

dependent

[1] for the compressible

Galerkin

Galerkin

to the exact

system,

methods

polynomials

This is a typical

which

errors

discontinuous

Galerkin

2.2 and 2.3, to motivate

3 of this

success

containing

of Bassi and

a method

to guarantee

the derivatives

a local discontinuous

derivatives.

type

(1.2)

for the

derivatives.

for time

Galerkin

of piecewise

of the discontinuous

into a first order

be designed

is called

discontinuous

equation

experiments

In section

fluxes must

discontinuous

consists

higher

yield

methods

the equation

introduced

diffusion

which

and has O(1)

Galerkin

A key ingredient

These

could

equation

discontinuous

system.

derivatives,

space,

application

second

is to rewrite

spatial

and careless

variables

bi-harmonic

steady

coupled

to handle

fluxes.

to provable

sixth

we use capital

enough

numerical

third

be more

(1.5)

the nonlinearity

with

we will discuss

of the

order

with the original

The idea of local

numerical

paper

variables

the solution

is not regular

elements.

but is inconsistent

The

again

independent

[3], the survey

higher

This is because

interfaces,

case in finite

vection

could

[12].

the element

heat

spatial

notes

where equations

the numerical

methods

description

to the lecture

For equations directly

In this

[19], were first

et al.

For a detailed

refer the readers

time

their

to denote

discretization

by Cockburn

2.1.

and

in the

the nonlinearity

=o

derivatives, Similar

Galerkin

approximation

(1.4)

i=1

fifth

applications.

and lower case letters

where

+

example.

equations,

and engineering

the PDEs

with

as an

-- 0

d

i=1

we just

equation

equation

c5 +

but

bi-harmonic

i=1

d

is a partial

dependent

d

Ut + Zf_(U)x,

is a partial

The time

cases.

results

which

led

Numerical

will be briefly

sections. methods

for the

time dependent

4 we do the same

thing

for the

partialdifferential equation(1.5)involvingfifth derivatives. Similarmethodscanbedesigned forwellposed partialdifferentialequations involvingevenhigherderivatives.Asbefore,wegiverecipesfor correct interelement

numerical

is independent suitable

fluxes

of the

for the

which

coefficients

so-called

terms

and hence

extremely

local and hence numerical

usual

and

proven)

piecewise

of this post-processing

technique

is sufficiently

and

smooth,

will achieve

the

convergence

of the

post-processing linear

order

L2-norm

post-processing.

with

higher

piecewise

(1.3),

polynomials

computational Concluding

linear

in Galerkin

the same

f'rom Ax k+l before strongly

suggests

of degree

k are used. accuracy

this

step

is applied

accuracy

enhancement

that

the

The

post-processing

in the

technique

thus

in [20] for the

only at the end of the computation

post-processing the order we apply

in this

paper,

for all these (in many

cases, cases

is fully

this

(1.5) where

Ax 2k+_)

for these

equations

norm,

taking

when

advantage

nonlinear

of

problems,

KdV equations.

only locally,

of

for the

PDE

negative-order

nonlinear

is

paper,

for some

and is applied

the

the solution

than

designed

can even be observed method

than

(1.4), and the linear

we have

of 2k + 1 or higher

Galerkin

5 of this

capability

methods

the

bigger

to Ax 2k+1 or higher

LDG

is locally

post-processing

is linear,

in [20] and

equations

mesh

The key ingredient

then

is usually

method

bi-harmonic

post-processing

The

estimate,

In section

Galerkin

are

laws (1.1) (theoretically

experiments.

(which

methods

We will provide

if the

thus if the problem

norm

enhancement

it to the local discontinuous

observed).

methods).

of convergence

that

(numerically

error

especially

to the higher

of Ax 2k+1 rather

conservation

negative-order

time dependent

have an order

To demonstrate

post-processing

error

We observe

This

derivatives

we also apply

has a good

convolution

in numerical

result

methods.

to an accuracy

estimate,

of the negative-order

of the

has been increased

after

method

error

are

Also, these

for linear

cases

to our new local discontinuous

fifth derivatives.

accuracy

this.

the

technique

(1.2)

is a negative-order

methods

coefficients

terms.

of these

is a local

solution

nonlinear

the

small

The stability

and easy for h-p adaptivity.

k are used,

equations

for some

of convergence

KdV like equations

containing

diffusion

accuracies

with

convection

in [7], which

of degree

methods.

hence

those

and accuracy

the

of the

derivatives, i.e.

derivative

the stability

was introduced

convection

in enhancing

order

implementations

to be able to recover polynomials

L 2 stability

problems,

by the first

to verify

proven

and for linear

even useful

of the higher

for parallel

technique

has been

Ax k+l when

efficient

nonlinear

dominated"

are dominated

examples

A post-processing uniform

in front

"convection

derivative

preliminary

lead to provable

As the

the additional

cost is negligible. remarks

summarizing

results

in this paper

and indicating

future

work are included

in section

6. 2.

Review

derivatives. conservation diffusion

laws

which

2.1.

discontinuous

section

(1.1) with

equations

For simplicity indicate

of the In this

(1.2)

first

with

of presentation, results

Laws.

the

second

methods

essential

derivatives,

the

derivatives

local and

for the general

let's

1,...,

N, with

xj+½

- xj_½.

introduce the

center

some

of the cell denoted

We will denote

Ax

= maxj

The

the KdV

Axj.

type

cases

version

=

second

(1.3) the

for the

with

methods.

and

methods

methods

equations

to present

first,

Galerkin

third

convection derivatives.

However,

laws

we will

(1.1) has the form

= O.

1 (xj_½

third for the

cases.

of the conservation

(2.1)

computational

by xj

with

Galerkin

multi-dimensional

The one dimensional

notations.

PDEs

of the discontinuous

discontinuous

Ut + I(U)x First

for

points

we will use one dimensional

are also valid

Conservation

Galerkin

we review

mesh

is given

+ xj+½)

If we multiply

(2.1)

and

by Ij the

=[xj_½,xj+½],

size of each

by an arbitrary

test

for j =

cell by Axj function

V(x),

=

integrate

over the interval

f This

the

UtVdx

is the starting

discontinuous test

Ij,

function

by parts,

+

flr

point

Galerkin

and integrate

for designing

method

we get

f(U(xj+½,

t))V(xj+½)

the discontinuous

Galerkin

[9, 6, 4, 10] can be described

V by piecewise

polynomials

f(U(xj_½,

of degree

t))V(xj_½)

method.

as follows:

at most

O.

The semi-discrete

we replace

k, and

=

denote

both

them

(2.2)

version

of the

the solution

by u and

U and

v.

That

is,

u, v E 1;A_ where IZA_ = {V : V is a polynomial With

this

xj+½,

as both

crucial

choice,

there

ingredient

choice

of discontinuity

f(u)

u and the

for the success

of the numerical at the

(information

is an ambiguity

the solution

in (2.2)

fluxes

to overcome

element

interfaces.

last

k for x E Ij, two

Galerkin

The

method

idea

j = 1,...,N}.

involving exactly

is to treat

by a single

these

the

boundary

values

at

boundary

points.

A

terms

monotone

laws is the

way, to utilize)

this

finite

volume

numerical

correct

ambiguity

by an upwinding

high resolution

valued

(2.3)

at these

for the conservation

say it in a positive

from the successful

j is given

terms

v are discontinuous

(or we could

borrowed

xj+ ½ for each

at most

in the

test function

of discontinuous

from characteristics),

at the interface

of degree

mechanism

schemes.

Thus

flux

]j+½ = f(j+_,u,+l), ^ u+ which

depends

both

at the interface argument continuous f(u,

a non-increasing

with respect

u) = f(u).

exact

on the left limit uj-+½ and on the right limit

xj+ ½. Here monotone

and

is then

xj+½

to each

from

following

attractive

inside

the cell Ij,

utvdx

-

by a nonlinearly

properties,

1. It can be easily determined

a scalar

high order contains

on arbitrary

then

--

Runge-Kutta

flux f(u)

the monotone

f_

1V_-

hand,

of accuracy. for efficient

1 =

solution of its first Lipschitz

in the sense that

flux is replaced the test function The final

by an v at the

semi-discrete

0

(2.5)

time discretizations

discontinuities.

multi-dimensional

triangulations,

the physical

On the other

fj+½Vj+½

strong

for any order allowing

with

function

to be at least

vj+½ and v+3-½ respectively.

+

numerical

f is a non-decreasing It is also assumed

equation,

namely

for the general

designed

the function argument.

[16] for details.

f(u)v_dx

stable

in each cell, thus

2. It can be used

than

see, e.g.

[17] may be used if the solution

u;+½ of the discontinuous

and to be consistent

rather

solver,

that

of its second

argument

Riemann

is taken

discretized

limiters

flux means

function

If (2.1) is a system

or approximate

interfaces scheme

(2.4)

The schemes

[19]. Nonlinear thus obtained

case

(1.1) with

arbitrary

In fact,

the order

of accuracy

TVB

have the

triangulations: can be locally

p adaptivity.

even those

with hanging

nodes,

thus

allowing

for efficient

h adaptivity. 3. It is extremely

local

communicate

only with

for efficient These

schemes

multi-dimensional 1. The Euler

parallel

in data

communications.

the immediate

implementations.

also have the following case

semi-discrete

(1.1) with scheme

and Crank-Nicholson,

arbitrary (2.5),

The

neighbors, See, e.g.

provable

evolution

regardless

of the solution of the order

in each

of accuracy,

cell needs thus

to

allowing

[2].

theoretical

properties,

all of these

are valid

also for the

triangulations: and

certain

have excellent

time

nonlinear

discretization stability

of it, such

properties.

as implicit

One can prove

backward a strong

L2

stabilityanda cellentropyinequalityfor the squareentropy,for the generalnonlinearscalarcase (1.1),for anyordersof accuracy on arbitrarytriangulations in anyspacedimension, withoutthe needfor nonlinearlimiters[14].Noticethat thesestabilityresultsarevalidevenwhenthesolution contains

discontinuities

2. For linear

problems

have a provable could

observe

for both 3. When

2.2. tion

error

and

nonlinear

scalar

with smooth estimate

of order

nonlinear

TVB

norm

Convection

convergence

[17, 9, 4] are

one dimensional

version

lower order

method

for the

that,

convection

the

methods

problems

The one dimensional

- (a(V)Ux)x

can (2.1),

version

be proven

stable

and stable

in the

total

in the L °c norm

of the convection

diffusion

for

equa-

= 0,

Galerkin

- (b(U)Q)x

= f_ b(U)dU.

equation

to solve

= 0,

(2.6)

method

for solving

Q - B(U)_

We can then (2.7),

resulting

(2.6) [11] approximates

formally in the

= 0,

(2.7)

use the same following

discontinuous

scheme:

find

Galerkin

u, q E V_

such

v, w E VA_,

ingredient

However,

there

In [11], criteria

for the is no longer

are given

"alternating

method

to be stable

a upwinding

for these

principle"

fluxes

in designing

_t-_-

] is chosen

all evaluated

as before

in (2.4).

at this interface

if the left value is chosen

point.

Notice The

for the former

is the

mechanism

correct

choice

of the

or characteristics

to guarantee

stability

and

numerical

to guide convergence.

fluxes

the design The best

(the

of these choice

is

the fluxes:

= B(u +) - B(u-).,

and

one

L 2 and L °¢ norms,

i

a crucial

to use the

triangulations

Ax k+l in both

k

J

j

fluxes.

for most

of degree

(1.1).

of the local discontinuous

and B(U)

J

"hats").

In effect,

polynomials

system

for all test functions

Again,

used,

problems

equations.

U, + ](U)_ b(U) = _,

piecewise

form

where a(U) >_ O. The semi-discrete

where

using

of the order

nonlinear

Vt + ](U)x

the following

methods

Ax k+½ in L 2 [15].

cases)

nonlinear

diffusion

these

problems.

limiters

for scalar

multi-dimensional

(1.2) has the

solutions,

(and prove in many

linear

variation

such as shocks.

0 = q-',

B = B(u +)

(2.9)

?.$-

_

that

we did not write

"alternating then

the subscript

principle"

the right

j +

1 for the fluxes

refers to the alternating

value is chosen

for the

latter,

as they

choices

are

of 0 and/3:

as in (2.9).

One could

also choose

0 = q+; with

all the

other

We remark in cell Ij then

fluxes

that

unchanged.

the appearance

q is eliminated

These of the

by using the

b = B(u-)

choices

of fluxes

auxiliary second

variable

equation

guarantee

stability

q is superficial: in (2.8)

and solving

and optimal when

a local

a small

linear

convergence. basis

is chosen

system

if the

localbasisis notorthogonal. Theactualscheme foru

takes

is a big advantage

"mixed

genuinely

global.

that

for the scheme The

the choice

for u after

schemes

dimensional

part

above

properties,

in the

all of these

L 2

and

(and prove

in many

smooth

cases)

convergence The same

the derivative

equations.

f(U),

r(U)

and

The semi-discrete the following

g(U) version

lower order

(2.10),

formally

resulting

both

in the

three

(1.2),

arbitrary

stencil

methods

using

of accuracy

estimate

Ux, even if q can be locally version

polynomials

a strong arbitrary diffusion

of degree

one could

eliminated

in actual

like equation

k

observe

for both

also for the auxiliary

of the KdV

+ (r'(U)g(r(U)x)x):_

on

size of the

L 2 and L _ norms,

can be obtained

backward

goes to zero.

piecewise

Ax k+l in both

Ill]:

as implicit

of the

coefficient

properties theoretical

One can prove

orders

multi

derivative

triangulations

properties.

is independent

the diffusion

first

provable

of it, such

for any

stability

of the order error

in [11].

general

attractive

Ax k in L _. In effect, for most triangulations

linear

variable

q,

calculation.

(1.3)

has the form

(2.10)

= O,

functions.

of the local discontinuous

Galerkin

method

for solving

(2.10)

[20] approximates

system

Ut+(f(U)+r'(U)P),=O, We can then

case

these

method

for the most

all of the

case (1.2) with

when

This

is usually

the most compact

also have the following

stability

The one dimensional

are arbitrary

retain

They

nonlinear

This

alone.

variable

Galerkin

yields

is nonlinear

discretization

solutions,

of order

which

time

scalar

in the limit

Ut + f(U):_ where

certain

dimension.

estimate

approximates

and

principle

(2.6), or in fact

part,

equations.

nonlinear

space

problems.

like

diffusion

auxiliary

local discontinuous

equation (1.2),

have excellent

with

error

and nonlinear

KdV

(2.8),

is also valid

problems

have a provable

equations

whose

for convection

eliminated.

for the multi-dimensional

general

in any

and hence

is termed

q is locally

derivation

scheme

for the

2. For linear

2.3.

second

Crank-Nicholson,

stability

which

diffusion

are valid

triangulations terms

variable

used on convection

semi-discrete

Euler

the scheme

to that

methods",

in (2.9) by the alternating

for the one dimensional

convection

and

that

of fluxes

the auxiliary

for the method

1. The

over the traditional

thus designed

nonlinear

convection

scheme

This is the reason

We also remark

listed

of the

a form similar

use the

in the following

same

P-g(Q)x=O, discontinuous

scheme:

find u,p,

Galerkin

Q-r(U)x=O. method

q E FAx such

that,

for the

(2.11)

convection

for all test

functions

equation

to solve

v, w, z E lgz_,,

(2.12) J

_j Again, "hats"). and

a crucial

qzdx

+ fb

ingredient

It is found

r(u)z,

dx - ,j+_l z-j+,l -}-,j_½ z;_½

for the

out in [20] that

method

to be stable

_- O.

is the

one can take the following

correct simple

choice choice

of the of fluxes

numerical to guarantee

fluxes

(the

stability

convergence:

f=f(u-,u+),

r '=r(u+)-r(u-) "L/+ -- iS--

15=p +, '

t_=_(q-,q+),

P=r(u-)

(2.13)

where

](u-,u

crucial

part

+) is a monotone is still the

flux

"alternating

]=](u-

for f(u), principle"

u+),

_,=

also

basis

work.

is chosen

(2.12)

and

takes

Again,

in cell Ij then

solving

two small

a form similar

alternating locally

to that

principle

both

of them

linear

systems

a compact

The

three

schemes

thus

designed

nonlinear

attractive provable

arbitrary

triangulations

1. The

KdV

the

Thus

_=_(q-,q+),

auxiliary

if tile local alone.

listed

theoretical

÷=r(u

+)

variables

by using

basis

q is superficial:

the

second

is not orthogonal.

We also remark

for the

p and

scheme

that

The

the choice

for u after

the

and

when

third

auxiliary

equations

actual

of fluxes

a local in

scheme

for u

in (2.13)

variables

by the

p and

q are

(1.3),

and

for any orders

the coefficients

of degree

used are

and

certain

time

linear

and

third

for the

in all the

derivatives,

for the

most

equations.

general

retain They

multi

all of the

also have

multi-dimensional

of the nonlinear

method

these

derivative with

order

discretization stability entropy,

case

the

(1.3)

with

terms

for the general

results

tend

Ax k+l in both

backward

nonlinear

scalar

dimension,

are valid

even

without

in the

limit

L2 case the

when

to zero.

solutions,

of order

as implicit

One can prove a strong

in any space

stability

smooth

of it, such

properties.

triangulations

that

these

methods

using

piecewise

Ax k+ ½ in L 2 [20]. In numerical L 2 and L _

norms

polyno-

experiments

for both

one and multiple

In this section,

cases. method

for the

for the bi-harmonic

bi-harmonic

type

equations.

type

(1.4).

We will concentrate

equation

on the

case + (a(U_)U_)x_,

= O,

0 < x < 1

(3.1)

condition g(x,O)=U°(x),

simplicity

valid

square

error estimate

Galerkin

a LDG

for the

problems

Ut + f(U)_:

and periodic

or in fact

on convection

nonlinear

on arbitrary

k have a provable

and analyze

an initial

method of these

[20]. Notice

linear

discontinuous

one dimensional

(2.10),

is nonlinear

all

inequality

of accuracy limiters

convergence

dimensional A local

equation

which

have excellent

of the dispersive

one observes

we present

(2.12),

a cell entropy

2. For one dimensional

3.

for the

properties,

scheme

need for nonlinear

mials

like

(1.3),

above

and Crank-Nicholson,

stability

KdV

[20]:

semi-discrete

Euler

for the

like equations

properties

following

The

sides.

In fact,

'

can be eliminated

stencil

flux for -g(q).

eliminated.

dimensional

with

/Z--

of the

for convection

yields

_=p-,

--

appearance

is a monotone

to take i5 and ÷ from opposite

_/+

the

-_(q-,q+)

r(u+)-r(u-)

'

would

and

boundary

conditions.

only and is not essential:

generalization

to the

multiple

0