Upload
trinhnga
View
215
Download
0
Embed Size (px)
Citation preview
NASA/CR-2000-21 0632 leASE Report No. 2000-49 NASA-CR-2000-21 0632
20010029126
A Technique of Treating Negative Weights in WENO Schemes
ling Shl, Changqing Hu, and Chi-Wang Shu Brown University, Providence, Rhode Island
December 2000 1111111111111 1111 1111111111111111111111111111
NF01367
,
I I I
https://ntrs.nasa.gov/search.jsp?R=20010029126 2018-05-06T15:45:37+00:00Z
The NASA STI Program Office ... in Profile
SInce Its foundIng, NASA has been dedicated to the advancement of aeronautics and space sCience The NASA SCientific and Technical Information (STI) Program Office plays a key part In helping NASA maIntaIn this Important role
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 largest collectIOn of aeronautical and space sCience STI In the world The Program Office IS also NASA's InstitutIOnal mechanism for dissemInatIng the results of ItS research and development activities These results are published by NASA In the NASA STI Report SerIes, which Includes the follOWIng report types
• TECHNICAL PUBLICATION Reports of completed research or a major Significant phase of research that present the results of NASA programs and Include extensIve data or theoretical analYSIS Include,> compilatIOns of Significant sCientific and technical data and infOrmatIOn deemed to be of continuing reference value NASA's counterpart of peer-reviewed formal profeSSIOnal papers, but haVing less strIngent limitatIOns on manuscrIpt length and extent of graphIC presentatIOns
• TECHNICAL MEMORANDUM SCientific and technical findings that are preliminary or of speCialized Interest, e g , qUick release reports, working papers, and bibliographies that contaIn minimal annotatIOn Does not contain extensive analYSIS
• CONTRACTOR REPORT SCientific and technical findings by NASA-sponsored contractors and grantees
• CONFERENCE PUBLICATIONS Collected papers from sCientifiC and technical conferences, symposia, semInars, or other meetIngs sponsored or cosponsored by NASA
• SPECIAL PUBLICATION SCientific, technical, or historIcal information from NASA programs, projects, and miSSions, often concerned With subjects haVing substantial public Interest
• TECHNICAL TRANSLATION EnglIshlanguage translations of foreign sCICntlfic and technical materIal pertinent to NASA's miSSIOn
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 Videos
For more informatIOn about the NASA STI Program Office, see the follOWing
• Access the NASA STI Program Home Page at Izttp://www.sti.nasa.gov
• Email your questIOn via the Internet to help@stJ nasa gov
• Fax your questIOn to the NASA STI Help Desk at (301) 621-0134
• Telephone the NASA 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 -2000-210632 ICASE Report No. 2000-49
11111111111 1111 1~~~~I[i[fl\jrli\1 ~11{il~111 1111 III 111111 3 1176 01452 7502
A Technique of Treating Negative Weights in WENO Schemes
ling STu, Changqing Hu, and Chi-Wang Shu Brown University, Providence, Rhode Island
leASE NASA Langley Research Center Hampton, Virginza
Operated by Universities Space Research Association
National Aeronautics and Space Administration
Langley Research Center Hampton, Virginia 23681-2199
Prepared for Langley Research Center under Contract NAS 1-97046
I I I I I Ii I I
December 2000
AVaIlable from the follOWIng
NASA Center for AeroSpace InformatIOn (CASI) 7121 Standard Dnve Hanover, MD 21076-1320 (30 I) 621-0390
NatIOnal Techmcal InformatIOn ServIce (NTIS) 5285 Port Royal Road Spnngfield, VA 22161-2171 (703) 487-4650
A TECHNIQUE OF TREATING NEGATIVE WEIGHTS IN WENO SCHEMES·
JING SHIt, CHANGQING HUt, AND CHI-WANG SHU§
Abstract. HIgh order accurate weIghted essentIally non-oscIllatory C\vENO) schemes have recently been
developed for fimte dIfference and fimte volume methods both m structured and m unstructured meshes
A key Idea m 'VENO scheme IS a lmear combmatIOn of lower order fluxes or reconstructIOns to obtam a
lllgher order approxImatIOn The combmatIOn coeffiCIents, also called lmear \velghts, are determmed by
local geometry of the mesh and order of accuracy and may become negatIve 'VENO procedures cannot
be applIed dIrectly to obtam a stable scheme If negatIve lInear weIghts are present PrevIOus strategy for
handlmg tillS dIfficulty IS by eIther regroupmg of stencIls or reducmg the order of accuracy to get nd of the
negatIve lmear weIghts In tIus paper we present a sImple and effectIve techmque for handlIng negatIve lmear
weIghts wIthout a need to get nd of them Test cases are shown to Illustrate the stabIlIty and accuracy of
tIus approach
Key words. weIghted essentIally non-oscIllatory, negatIve weIghts, stablhty, lugh order accuracy, shock
calculatIOn
Subject classification. Apphed and Numencall\lathematIcs
1. Introduction. HIgh order accurate weIghted essentIally non-oscIllatory (\vENO) schemes have re
cently been developed to solve a hyperbohc conservatIOn law
Ut + V' f(u) = 0 (1 1)
The first 'VENO scheme was constructed m [18] for a thIrd order fimte \olume verSIOn m one space dImensIOn
In [10], thIrd and fifth order fimte dIfference 'VENO schemes m multI space dImensIOns are constructed, wIth
a general framework for the desIgn of the smoothness mdlcators and nonlmear wClghts Later, second, tlurd
and fourth order fimte volume WENO schemes for 2D general tnangulatIOn have been developed m [4] and
[8] Very hIgh order fimte dIfference WENO schemes (for orders between 7 and 13) have been developed m
[1] Central WENO schemes have been developed m [12], [13] and [14]
WENO schemes are desIgned based on the successful ENO schemes m [7, 23, 24] Both ENO and
'VENO use the Idea of adaptIve stencIls m the reconstructIOn procedure based on the local smoothness
of the numencal solutIOn to automatIcally aclueve hIgh order accuracy and non-oscIllatory property near
dlscontmmtles ENO uses Just one (optImal m some sense) out of many candIdate stencIls when domg the
reconstructIOn, wlule 'VENO uses a convex combmatIOn of all the candIdate stencIls, each bemg assIgned
a nonhnear weIght WhICh depends on the local smoothness of the numencal solutIOn based on that stencIl
'VENO Improves upon ENO m robustness, better smoothness of fluxes, better steady state convergence,
better provable convergence propertIes, and more efficIency For a detaIled reVIew of ENO and 'VENO
schemes, we refer to the lecture notes [21, 22]
'Research supported by ARO grants DAAG55-97-1-0318 and DAAD19-00-1-0405, NSF grants DMS-9804985 and ECS-
9906606, NASA Langley grant NAG-1-2070 and Contract NASl-97046 whIle the thIrd author was In resIdence at ICASE,
NASA Langley Research Center, Hampton, VA 23681-2199, and AFOSR grant F49620-99-1-0077 tDlvlslOn of ApplIed MathematICs, Brown UnIversIty, ProvIdence, RI 02912 E-maIl shl@cfm brown edu tDlvlslOn of ApplIed MathematICs, Brown UnIversIty, ProvIdence, RI 02912 Current address Oracle, 30P6, 500 Oracle
Parkway, Redwood Shores, CA 94065 E-maIl phu@us oracle com §DlvlslOn of ApplIed MathematICs, Brown UnIversIty, ProvIdence, RI 02912 E-maIl shu@cfm brown edu
1
WENO schemes have already been wIdely used m apphcatlOns Some of the examples mcIude dynamIcal
response of a stellar atmosphere to pressure perturbatlOns [3], shock vortex mteractlOns and other gas
dynamIcs problems [5], [6], mcompresslble flow problems [26], HamIlton-JacobI equatlOns [9], magneto
hydrodynamIcs [11], underwater blast-wave focusmg [15], the composIte schemes and shallmv water equatlOns
[16], [17], real gas computatlOns [19], wave propagatlOn usmg Fey's method of transport [20], etc
A key Idea m \VENO schemes IS a hnear combmatlOn of lower order fluxes or reconstructlOns to obtam
a hIgher order approximatlOn The combmatlOn coeffiCIents, also called hnear weIghts, are determUled by
local geometry of the mesh and order of accuracy and may become negatIve \VENO procedures cannot
be apphed dIrectly to obtam a stable scheme If negatIve lmear weIghts are present PrevlOus strategy for
handhng tlns dIfficulty IS by either regroupmg of stenCils (e g III [8]) or reducmg the order of accuracy (e g
m [12]) to get nd of the negatIVe hnear Weights In thIs paper we present a sImple and effectn e techmque
for handhng negative hnear weIghts WIthout a need to get nd of them Test cases WIll be shown to Illustrate
the stablhty and accuracy of thIs approach
We first summanze the general \VENO reconstructlOn procedure, consIstmg of the followmg steps \Ve
assume we have a gIVen cell 6 (whIch could be an mterval mID, a rectangle m a 2D tensor product mesh,
or a tnangle m a 2D unstructured mesh) and a fixed pomt xC wlthm or on one edge of the cell
1 \Ve IdentIfy several stenCIls S), J = 1, ,q, such that 6 belongs to each stenCil \Ve denote b} q
T = U S) the larger stencIl winch contallls all the cells from the q stencIls )=1
2 \Ve have a (relatIvely) lower order reconstructlOn or mterpolatlOn functlOn (usually a polynomial),
denoted by p)(x), associated WIth each of the stenCils S), for J = 1, ,q \Ve also have a (relatively)
hIgher order reconstructlOn or lllterpolatlOn functlOn (agam usually a polynomial), denoted by Q(x),
associated WIth the larger stencIl T 3 \Ve find the combmatlOn coeffiCIents, also called hnear weIghts, denoted by /'1, , /'q, such that
q
Q(xc ) = L /'lPJ(XC
) (1 2) )=1
for all pOSSIble gIven data m the stenCIls These hnear weIghts depend on the mesh geometry, the
pomt xc, and the speCific reconstructlOn or mterpolatlOn reqmrements, but not on the gIven solutlOn
data m the stenCils
4 We compute the smoothness llldicator, denoted by (3), for each stenCIl SJ' wincIl measures how
smooth the functlOn p) (x) IS III the target cell 6 The smaller thIS smoothness mdlcator (3), the
smoother the functlOn p) (x) IS m the target cell In all of the current \VEN 0 schemes we are USlllg
the followmg smoothness mdlcator
(1 3)
for J = 1, ,q, where k IS the degree of the polynomial p)(x) and 161 IS the area of the cell 6 m 2D
This factor IS dIfferent for ID or 3D the purpose of It IS to bnng the smoothness mdlcator mvanant
under spatial scahng
5 \Ve compute the nonlmear weIghts based on the smoothness mdicators
(1 4)
2
where "/J are the lmear weIghts determmed m step 3 above, and c: IS a small number to aVOId the
denommator to become 0 We are usmg c: = 10-6 m all the computatIOns m tIllS paper The final
WENO approXImatIOn or reconstructIOn IS then gIven by
q
R(xG) = 2:= w)p) (xG
) (1 5) ;=1
We remark that all the coefficIents m the above steps whIch depend on the mesh but not on the data of
the numencal solutIOn, should be computed and stored at the begmmng of the code after the generatlOn of
the mesh but before the time evolutlOn starts
\Ve now use a sImple example to Illustrate the steps outlIned above \Ve a')sume we are gIven a umform
mesh I, = (X,-1/2,X,+1/2) and cell averages of a functIOn u(x) m these cells, denoted by u, \Ve would
lIke to find a fifth order \VENO reconstructlOn to the pomt value U(X,+1/2), based on a stencIl of five cells
{II - 2,!,-I,!,,!,+l,!,+2}, WIth the target cell contammg the pomt X'+I/2 chosen as 6. = I, In step 1 above we could have the followmg three stenclls
whIch make up a larger stencIl
In step 2 above we would have three polynomials PJ(x) of degree at most 1\,0, wIth theIr cell averages
agreemg WIth that of the functIOn u m the three cells m each stencIl S) The Illgher order functIOn Q(x) IS a
polynomIal of degree at most four, WIth ItS cell averages agreemg WIth that of the functlOn u m the five cells
m the larger stencIl T The three lower order approxImatIOns to U(X'+I/2), associated 'Hth p)(x), m terms
of the gIven cell averages of u, are gIven by
1 7 11 PI (X,+1/2) = 3"U'-2 - GUI - I + (fu"
1 5 1 P2(X,+l/2) = -GU,-I + GU' + 3"U'+I, (1 G)
1 5 1 P3(X,+l/2) = 3"u, + GU,+1 - GU,+2
Each of them IS a tlurd order approximatlOn to U(X'+l/2) The hIgher order approxImatIOn to U(X'+l/2),
assocIated WIth Q(x), IS gIven by
whIch IS a fifth order approxImatIOn to U(X'+I/2)
In step 3 above we would have
1 "/1 = 10'
3 "/2 = 5'
It can be readIly venfied, usmg (1 G) and (1 7), that
for all possIble gIVen data UJ , J = Z - 2, Z - 1, Z, Z + 1, Z + 2
3
3 "/3 = 10
(1 7)
2r
0 0
-2 2
-4 -4
6 -6
-8 -8
-10 -10
-12 -12
-14 14
"t -18
-20
-22
24 --~ 1
01 -"-~_~~ __ I
o 01 02
16
18
20 ([)
B-e--EH: \ I
22 \ I
-24 __ ~_I_~ __ ---'~~~---'---- __ I
-01 0 01 02 X X
FIG 1 1 ReconstructIOns to U(X.+1/2) Solzd lmes exact functIOn, symbols numerzcal approximations Left fifth order
WENO Right fifth order traditional
In step 4 above we could easIly work out from (1 3) the three smoothness mdicators gIven by
13 ? 1 ?
/31 = 12 (U,-2 - 2U,-1 + u,t + 4" (U,-2 - 4U,-1 + 3u,t ,
13 _ _ _ 2 1 _ _ 2 /32 = 12 (U,-l - 2u, + U,+l) + 4" (U,-l - u,+!) ,
13 ., 1 ., /33 = 12 (u, - 2u,+! + U,+2t + 4" (3u, - 4U,+1 + U1+2t
\Ve notice m particular that the lmear weights 1'1,1'2,1'3 m step 3 above are all posItive In such cases, the
\VENO reconstructIOn procedure outhned above and the scheme based on It work very well In Fig 1 1 we
plot the approximatIOn to u(x) for a dlscontmuous functIOn u(x) = 2x for x :S 0 and u(x) = -20 otherWIse,
by the fifth order \vENO reconstructIOn on the left and by the fifth order traditIOnal reconstructIOn (1 7)
on the nght, WIth a mesh x, = (z - 0 4965)~x With ~x = 002 \Ve can clearly see that \vENO a"Olds the
over and undershoots near the dlscontmUIty
\Ve now look at another simple example where some of the hnear weights m step 3 above ",ould become
negative \Ve have exactly the same settmg as above except now we seek the reconstructIOn not at the cell
boundary but at the cell center x, TIns IS needed by the central schemes With staggered gnds [12] Thus,
step 1 would stay the same as above, step 2 would produce
1 1 23 pdx,) = - 24 U,-2 + 12 U,-l + 24 u"
1 13 1 P2(X,) = - 24 U,-l + 12u, - 24 U,+l, (1 8)
23 1 1 P3(X,) = 24 U, + 12 u,+! - 24 U,+2
Each of them IS a third order reconstructIOn to u(x,) The higher order reconstructIOn to u(xl ), associated
With Q(x), IS gIven by
(1 9)
4
~ I \
0
-2
-4
-6
-8
-10
-12
-14
-16
-18
-20
-22 l
-01 0 X
4
2
0
-4
-6
-10
-12
-14
-16
-18
-20
L~~ __ -' -22 01 02
o
L_~~-"_~_--'--"---_~~L_~ __ ' -01 0 01 02
X
FIG 1 2 Reconstructions to u(x,) Solid lmes exact functIOn, symbols numerical apprOXimatIOns Left fifth order
WEND Right fifth order traditIOnal
whIch IS a fifth order reconstructIOn to u(x,) Step 3 would produce the followmg weIghts
49 12 = 40'
9 13 =--
80
NotIce that two of them are negatIve The smoothness mdicators m step 4 WIll remam the same TIns
tIme, the 'VENO approxImatIOn, shm,n at the left of FIg 1 2, IS less satIsfactory (m fact, even worse than
a tradItIOnal fifth order reconstructIOn show on the nght), because of the negatrve lmear weIghts
\Ve remark that negatIve lInear weIghts do not appear m fimte dIfference \VENO schemes III any "patial
dImensIOns for conservatIon laws for any order of accuracy [10], [1], and they do not appear m one dimenslOIldl
as well as some multl-dlmenslonal fimte volume \VENO schemes for conservatIOn laws Unfortunately, tll('\
do appear III some other cases, such as the central \VENO schemes usmg staggered meshes "e ha\e sePlI
above, Illgh order fimte volume schemes for two dImensIOns descnbed III [8] and m tins paper, and filllte
difference WENO approxImatIOns for second denvatrves
\VhJle on approXImatIOn alone the appearance of negatIve IIllear weights might be annoymg but perhap"
not fatal (Fig 1 2), m sohmg a PDE the result mIght be more senous As an example, m Fig 13 we shm\
the results of usmg a fourth order fimte volume \VENO scheme [8] on a non-umform tnangular mesh shown
at the left, WhICh has negatIve lInear weIghts, for solvmg the two dImensIOnal Burgers equatIOn
(1 10)
m the domam [-2,2] x [-2,2] With an mitral conditIOn uo(x,y) = 03+07sm(}-(x+y)) and penodlc
boundary condItIOns \Ve can see that senous oscIllatIOn appears m the numencal solutIOn once the shock
has developed The oscIllatIOn eventually leads to IllstabilIty and blowmg up of the numencal solutIOn for
thIS example
The mam purpose of thiS paper IS to develop a Simple and effectIve techmque for handlmg negative lInear
WeIghts Without a need to get nd of them Test cases Will be shown to Illustrate the stabIlIty and accuracy
of thiS approach
5
-,
-1
-, -1 0
X
FIG. 1.,3. 2D Burgers' equation. Left: non-uniform triangular' mesh used in the computation. Right: fourth order- WRNO
result at. t = 0.47:3, cjl=O.2, without any special treatment for the negative linear- weights.
2. A splitting technique. We now introduce a splitting technique to treat the negative weights. It
is very simple, involves little additional cost, yet is quite effective. Tlw WENO procedure outlined in the
previous section is only modified in step 5 in the following way:
5' If minb, , ... , rq) 2: 0 proceed as before. Otherwise, we split the linear weights into two parts:
positive and negative. Define
i = 1, ... ,q (2.1)
where we take () = 3 all the numerical tests. We then seale them by
q
a± = Lit; i = 1, ... , q. (2.2) j=J
We now have two split polynomials
q
L rtPj(:l/') (2.3) j=1
which satisfy
-Q-( G) a x. (2.4)
We can then define the nonlinear weights (1.4) for the positive and negative groups r:f separately,
denoted by cv;, based on the same Rmoothness indicator (3j. We will then define the WENO
approximation R±(:rP) separately by (1.5), using w.f, and form the final WENO approximation by
- R-( 0) a. x .
We remark that the key idea of this decomposition is to make sure that every stencil has a significant
representation in both the positive and the negative weight groups. Within each group, the WENO idea of
6
-2 I -4 I
I -6 I -8 I
-2
-10
-12 I
-14 I -16 I
I -18 I -20
-22 , ., , , ·-1' ~"~ -0.1 0 0.1 0.2
X
FIC.2.1. WENO approximations with the splitting treatment for negative linear weights_ Left: appr'oximation to v.(x;).
Right: Burgers equation, solution at t 5/rr2 , cfi=O.2_
redistributing the weights subject to a fixed sum according to the smoothness of the approximation is still
followed as before_
For the simple example of fifth order WENO reconstruction to U(:r:i), the split linear weights corrnspond
ing to (2.1) are, before the scaling,
9
40' _+ _ 49 __ 49 12 - 20' 12 = 40'
_+ _ 9 _ I:) - 80' 1;3
9 40
We notice that, as the most expensive part of the WENO procedure, namely the computation of the
smoothness indicators (1.3), has not changed, the extra cost of this positive/negative weight splitting is very
smalL
However this simple and inexpensive change makes a big difference to the computations_ In Fig. 2.1 we
show the result of the two previous unsatisfactory cases, the fifth order WENO reconstruction to U(:r:i) in
Fig. 1.2 left, and the approximation to the Burgers equation in Fig. 1.3 right, now llsing WENO schemes
with this splitting treatment. We can see dearly that the results are now as good as one would get from
WENO schemes having only positive linear weights.
It is easy to prove that the splitting maintains the accuracy of the approximation in smooth regions_
We will demonstrate this fact in the following sections. We will also demonstrate the effectiveness of this
simple splitting technique through a few selected numerical examples in the next sections. The main WENO
schemes we will consider are fifth order finite volume WENO schemes on Cartesian meshes, and the third
and fourth order finit.e volume \VENO schemes on triangular meshes. In both cases negative linear weights
appear regularly.
The calculations are performed on SUN Ultra workstations and also on the IBM SP parallel eomputer
at TCASCV of Brown University_ The parallel efficiency of the method is excellent (more than 90%)_
3. 2D finite volume WENO schemes on Cartesian meshes.
3.1. The schemes. 'We describe two different ways to construct fifth order finite volume WENO
schemes on Cartesian meshes_ Comparing with finite diffc-m~nce WENO methods [10], finite volume mc"th-
7
ods have the advantage of an applIcabilIty of usmg arbitrary non-umform meshes, at the pnce of mcreased
computatIOnal cost [2]
We define the cell
(3 1)
for z = 1, , Tn, J = 1, , n, \Vhere I,,} needs not be umform or smooth yarymg
The three-pomt Gaussian quadrature rule IS used at each cell edge when evaluatmg the numerical flux
m order to mamtam fifth order accuracy Let (xC, yC) denote one of the Gaussian quadrature pomts at the
cell boundary of I,,) given by r == {x = X,_!,y)_! :S y:S YJ+~} There are two ways to perform a WENO
reconstructIOn at the pomt (xc, yC)
Genuzne 2D The first \VENO reconstructIOn IS genu me 2D fillite volume We can see that there are
totally nme stencils S8,1 (s,t = -1,0,1) Each stencil S8,1 contams 3 x 3 cells centered around I'+8,)+t
On each stencil we can construct a Q2 polynomial (tensor product of second order polynomials III x and y)
satlsfYlllg the cell average conditIOn (I e ItS cell average III each cell IIlslde the stenCIl equal,> to the given 1
yalue) Let T = U S8,t, wIndl contams 5 x 5 cells centered around I,,) On T we can construct a Q4 8,1=-1
polynomial satIsfymg the cell average conditIOn The \VENO reconstructIOn IS then performed accordmg to
the steps outlllled III sectIOns 1 and 2
\Ve \',ould lIke to make the followmg remarks
1 By uSlllg a Lagrange mterpolatlOn basIs, we can eaSily find the ullique hnear "eights
2 Even for a umform mesh, a negative hnear weight appears for the middle Gaussian pomt (xC, yC) = (x,_ 1.., y)) Such appearance of negative lmear weights has also been observed III the central \VENO
2
schemes [12], see the example III sectIOns 1 and 2 before
3 By Taylor expanSIOns, we can prove that the smoothness IIldlcators Yield a umform fifth order
accuracy III smooth regIOns See [10] for the method of proof
Dzmenswn by Dzmenswn The second \VENO reconstructIOn explOits the tensor product nature of the
mterpolatlOn we use This 'VENO procedure IS performed on a dimensIOn by dimensIOn fasluon The \VENO
schemes applIed m [5], [6] belong to tlus class ConSider the pomt (xC,yC) as above First \\e perform a
one dimensIOnal \VENO reconstructIOn m the y directIOn, m order to get the one dimensIOnal cell averages
(m the x directIOn) w(., yC) Then we perform another one dimensIOnal 'VENO reconstructIOn to w III the
x directIOn, to obtam the final reconstructed POlllt value at (xC, yC)
We "ould lIke to make the followmg remarks
1 For a scalar equatIOn, the underlymg lInear reconstructIOns of the above t"o versIOns are eqUivalent
For nonlInear \VENO reconstructIOns they are not eqUivalent Both of them are fifth order accurate
but the actual errors on the same mesh may be different, see Table 3 1 below
2 For systems of conservatIOn laws such as the Euler equatIOns of gas dynamics, both versIOns of the
'VENO reconstructIOn should be performed III local characteristic fields
3 The dimensIOn by dimensIOn versIOn of the \VENO reconstructIOn IS less expensive and reqUires
smaller memory than the genullle two dimensIOnal versIOn The CPU time savmg IS about a factor
of 4 for the Euler equatIOns m our ImplementatIOn The computed results are mostly Similar from
both versIOns
In the followlllg, we "Ill give numencal examples computed by the above WENO schemes Sphttlllg
techmque has been used III all the computatIOns when negatl\e lmear weIghts appear \Ve will show the
8
TABLE 3 1
2D vortex evolutIOn
II II dlm-bv-dlm II genu me FV . N .6.x Loo error order LOO error order
20 G 71E-l 438E-2 52GE-2
40 377E-l 310E-3 459 5 GGE-3 38G
80 2 OlE-l 120E-4 515 39GE-4 422
lGO 1 OOE-l 439E-G 47G 79GE-G 5 G2
320 500E-2 188E-7 453 290E-7 477
results for both smooth and dlscontmuous problems
3.2. 2D vortex evolution. Flrst, '\e check the accuracy of the \VENO schemes constructed above
The two dllnenslOnal vortex evolutlOn problem [21], [8]lS used as a test problem
\Ve solve the Euler equatlOns for compresslble flow m 2D
Ut + J(U)x + g(U)y = 0, (32)
where
U = (p,pu,pv,Ef,
J(U) = (pu,pu 2 +p,puv,u(E+p))T,
g(U) = (pv,puv,pv2 + p,v(E + p))T
Here p IS the denslty, (u, v) IS the veloclty, E IS the total energy, p IS the pressure, related to the total energy
by E = ~ + ~p(u2 + v 2) wlth / = 1 4
The setup of the problem IS the mean flow IS p = 1, p = 1, (u,v) = (1,1) and the computatlOnal
domam IS [0,10] x [0,10] We add, to the mean flow, an Isentroplc vortex (perturbatlOns m (u, v) and the
temperature T = P., no perturbatlOn m the entropy 5 = *) p p
E 2 (ou,ov) = _e05(1-r )(-y,x),
27r 'T = _ h - I)E2 l_r2 u 8 2 e ,
/7r -
where (Y, y) = (x - 5, Y - 5), r2 = £2 + fP, and the vortex strength E = 5
05 = 0
\Ve use non-umform meshes wInch are obtamed by an mdependent random sInftmg of each pomt from a
umform mesh m each dlrectlOn Wlthm 30% of the mesh Slzes The solutlOn IS computed up to t = 2 Table 3 1
shows the L oo errors of p \Ve can see that both the genume t\\O dlmenslOnal fimte volume \VENO scheme
and the dlmenslOn by dlmenSlOn fimte volume \VENO scheme can aclneve the deslred order of accuracy
wInle the genume two dlmenslOnal scheme glves smaller errors for the same mesh
3.3. Oblique shock tubes. The purpose for thls test IS to see the capablhty of the rectangular \VENO
schemes m resolvmg waves that are obhque to the computatlOnal meshes For detmls of the problem, we refer
to [10] The 2D Sod's shock tube problem IS solved where the Imttal Jump makes an angle 0 agamst the x
axlS We take our computatlOnal domam to be [0, G] x [0,1] and the Imttal Jump startmg at (x, y) = (2 25,0)
9
05 c
05
o~~
o
08
06
04
02
0~------~~----~~~2~~~~3-~~~--~4
X
gt..nUltlt. I V dun hy-t.illli
5 6
FIG 3 1 Dbllque Sod's problem Density p Top contour, genume two dlmenswnal WEND, middle contour, dzmenswn
by dlmenswn WEND, bottom cut at the bottom of the computatwnal domam, the solzd lme IS the exact solutwn, the trzangles
are the genume two dlmenswnal WEND results, and the czrcles are the dlmenswn by dlmenswn WEND results
and makmg a () = % angle wIth the x aXIS The solutIOn IS computed up to t = 1 2 on a 9G x 1G umform
mesh In FIg 31 we plot the denSIty contours computed by the above two WENO schemes and the denSIty
cut at the bottom of the computatIOnal domam We can see that both schemes perform equally well m
resolvmg the waves The genu me two dImensIOnal scheme gIves a shghtly better resolutIOn m the contact
dlscontmUIty and the rarefactIOn wave
3.4. A Mach 3 wind tunnel with a step. ThIS model problem IS ongmally from [25] The setup
of the problem IS The wmd tunnel IS 1 length umt wIde and 3 length umts long The step IS 0 2 length
10
, 'I ---------,,-"'''~~''''''--,,-,~.-----~~~==~-, I
°o~--"
FIG 3 2 Forward step problem, flx = fly = to, -to, I!O' 3~O from top to bottom 30 contours from 0 12 to 641, dzmcTlslOTi
by dzmenslOn WEND
umts lugh and IS located 06 length umts from the left-hand end of the tunnel The problem IS InItIalIzed
by a rIght-gOIng Mach 3 flow ReflectIve boundary condItIons are apphed along the wall of the tunnel and
Inflow /outflow boundary condItIOns are applIed at the entrance/exIt The corner of the step IS a sIngular
POInt and we treat It the same ,,,ay as In [25], wInch IS based on the assumptIOn of a nearly steady flow m
the regIOn near the corner \Ve show the denSIty contours at tIme t = 4 In FIg 3 2 Only the results from
the dImensIOn by dImenSIOn \VENO scheme are shown Umform meshes of 6.x = 6.y = 410' lo' I~O' 3~O are
used
3.5. Double Mach reflection. ThIS problem IS also OrIgInally from [25] The computatIOnal domaIn
for thIS problem IS chosen to be [0,4) x [0, 1) The reflectIng wall lIes at the bottom, startIng from x = i ImtIally a rIght-mOVIng Mach 10 shock IS pOSItIOned at x = t, y = 0 and makes a 600 angle WIth the x aXIS
11
075 -
x
FIG 3 3 Double Mach reflectIOn, box = boy = 2!O (top and lower left) and 4~O (middle and lower Tight) Genume two
dimensIOnal WEND Blow-up regIOns at the bottom for details
For the bottom boundary, the exact post-shock conditIOn IS Imposed for the part from x = 0 to x = t and
a reflective boundary conditIOn IS used for the rest At the top boundary, the flow values are set to descnbe
the exact motion of a :t\Iach 10 shock \Ve compute the solutIOn up to t = 0 2 Fig 3 3 and Fig 3 4 show
the equally spaced 30 denSity contours from 1 5 to 22 7 computed by the genu me two dimensIOnal and the
dimenSIOn by dimenSIOn WENO schemes \Ve use ullIform meshes \Hth ~x = ~y = 2!O' 4!O We can see
that the results from both schemes are comparable
4. 2D finite volume WENO schemes on triangular meshes. Both third and fourth order £illite
volume WENO schemes on tnangular meshes have been constructed m [8] The optIOnallmear weights m
such schemes are not umque These are then chosen to aVOid negative weights whenever pOSSible, and If that
fails, a groupmg (of stencIls) techmque IS used m [8], which works fairly well m the third order case With
qUite general tnangulatIOn but can Yield posItive weights for the fourth order case only With faIrly umform
tnangulatIOn In thiS sectIOn, we do not seek posItive hnear wClghts as m [8], but rather use the sphttmg
techmque to treat the negative hnear weights when they appear For scalar equatIOn, the scheme IS stable
12
075 ...
05,-----------------------~
x
" x
" X
FIG 3 4 Double Mach ref/ectlon, Llx = Lly = 2!O (top and lower left) and 4~O (middle and lower rzght) DimensIOn by
dimensIOn WEND Blow-up regions at the bottom for details
m all runs For systems of conservatIOn laws, there are still occasIOnal cases of overshoot and mstablhty, the
reason seems to be related to characteristic decompositIOns and IS still bemg mvestlgated
4.1. Accuracy check for a smooth problem. We solve the 2D Burgers equatIOn (110) wIth the
same mltlal and boundary conditIOns as before usmg the fourth order finIte volume WENO scheme [8) The
solutIOn IS computed up to t = ~ when no shock has appeared The meshes used are 1) unIform meshes
with eqUilateral tnangulatlOn and 2) random tnangulatIOn For the UnIform meshes we do not seek posItive
weights as was done m [8), rather we use the sphttmg technIque to treat the negative lmear weights when
they appear Table 4 1 mdlcates that close to fourth order accuracy can be achieved
4.2. Discontinuous problem 1: Scalar equation in 2D. Havmg shown the stable results \\lth the
sphttmg treatment of negative hnear weIghts for a fourth order finIte volume 'VENO scheme for the Burgers
equatIOn m sectIOn 2, we now test the fourth order \VENO scheme on the Buckley-Leverett problem whose
13
TABLE 4 1
2D Burgers equation accuracy check
III umform mesh III nonumform mesh III ~x LOO error order ~x LOO error order
257E-1 622E-4 267E-1 211E-3
1 29E-1 46IE-5 375 1 26E-1 235E-4 292
643E-2 218E-6 440 632E-2 290E-5 303
32IE-2 138E-7 398 334E-2 26IE-6 378
16IE-2 693E-9 432 166E-2 271E-7 324
808E-3 670E-10 340 744E-3 157E-8 355
FIG 4 1 2D Buckley-Leverett equation the mesh
flux IS non-convex
u2
f(u) = u2 +025(1-u)2' g(u) = 0
WIth the ImtIal data u = 1 when - J ::; x ::; 0 and u = 0 elsewhere The solutlOn IS computed up to
t = 0 4 The exact solutlOn IS a shock-rarefactlOn-contact dlscontmmty mIxture The mesh we use here l~ a
non-umform tnangulatlOn, shown 1Il FIg 4 1 FIg 4 2 shows that the waves have been resolved very well
4.3. Discontinuous problem 2: System of equations in 2D. We consIder the 2D Euler equatlOns
m the domalll [-1,1] x [0,02] The Sod and Lax shock tube mltIal data IS set m the x dlrectlOn and penodlc
boundary condltlOn IS apphed 1Il the y dlrectlOn \Ve use the fourth order fimte volume scheme Oil tnangular
meshes to solve the above problem The mesh we use here IS umform But we do not seek posItive weIghts as
was done m [8], rather we use the sphttmg techmque m sectlOn 2 to treat the negative Imear \\ eIghts when
they appear In fact we set dehberately certam hnear weIghts to be negatIve to test the sphttmg tcchmque
FIg 4 3 shows the numencal results of the Sod and Lax problems
It seems that there are still oSClllatlOns and lIlstablhty for some non-umform tnangular meshes for the
fourth order \VENO schemes apphed to Euler equatlOns As the method works well for the same meshes
wIth a scalar equatlOn, the problem mIght be from the charactenstlc decomposltlOns ThIS IS stIlI under
mvestlgatlOn
5. Concluding remarks. We have devIsed and tested a SImple sphttmg techmque to treat the negatIve
hnear weIghts 1Il WENO schemes ThIS techmque mvolves very httle addltlOnal CPU time and gIves good
results m most numencal tests The only case where It stIlI YIelds osclllatlOns and mstablhty IS when a fourth
order fimte volume \VENO method IS used on some non-umform tnangular meshes for Euler equatlOns, the
reason of whIch, presumably related to charactenstIc decomposltlOns, IS stIll under mvestlgatlOIl
14
08
06
::::l
04
02
01
, '---OS 0 05
X
FIG 42 2D Buckley-Leverett equatIOn at t=o 4, with splzttmg Left the solution surface, Right the cut at y = 0 1 (solid
lme exact solutIOn, symbols numerical solutIOn)
14
12 08
06 c- o.
08
04 06
"'--02 04
, -1 -05
-.
o X
c
0
-"'
,
r;o
0
p
05
FIG 43 Density plot, Left Sad problem, Right Lax problem, with spllttmg Roughly 100 pomts m the x direction
REFERENCES
[1] D BALSARA AND C -\V SHU, Monotomczty preservzng wezghted essentzally non-osczllatory schemes
wzth zncreasmgly hzgh order oj accuracy, Journal of ComputatlOnal PhysIcs, 160 (2000), pp 405-452
[2] J CASPER, C - \V SHU AND H L ATKINS, Comparzson oj two Jormulatzons Jor hzgh-order accurate
essentzally nonosczllatory schemes, AIAA Journal, 32 (1994), pp 1970-1977
[3] L DEL ZANNA, 1\1 VELLI AND P LONDRILLO, Dynamzcal response oj a stellar atmosphere to pressure
perturbatzons numerzcal szmulatzons, Astron Astrophys , 330 (1998), pp L13-L16
[4] 0 FRIEDRICHS, Wezghted essentzally non-osczllatory schemes Jor the znterpolatzon oj mean values on
unstructured grzds, Journal of ComputatlOnal PhysIcs, 144 (1998), pp 194-212
[5] F GRASSO AND S PIROZZOLI, Shock-wave-vortex znteractzons Shock and vortex deJormatzons, and
sound productzon, Theor Comp FIUld Dyn , 13 (2000), pp 421-456
[6] F GRASSO AND S PIROZZOLI, Shock wave-thermal mhomogenezty znteractzons Analyszs and numerzcal
szmulatzons oj sound generatzon, Phys FIUlds, 12 (2000), pp 205-219
15
[7] A HARTEN, B ENGQUIST, S OSHER AND S CHAKRAVARTHY, UnzJormly hzgh order essentzally non
osczllatory schemes, III, Journal of ComputatIOnal PhYSICS, 71 (1987), pp 231-303
[8] C Hu AND C -W SHU, Wezghted Essentzally Non-Osczllatory Schemes on Trzangular Meshes, Journal
of ComputatIOnal PhYSICS, 150 (1999), pp 97-127
[9] G JIANG AND D -P PENG, Wezghted ENO schemes Jor Hamzlton-Jacob! equatlOns, SIAr..I Journal on
SCIentIfic ComputIllg, v21 (2000), pp 2126-2143
[10] G JIANG AND C - \V SlIU, Efficzent zmpiementatzon oj wezghted ENO schemes, Journal of Computa
tIonal PhYSICS, 126 (1996), pp 202-228
[11] G JIANG AND C -C \Vu, A hzgh order WENO jinzte dzfference scheme Jor the equatzons oj zdeal
magnetohydrodynamzcs, Journal of ComputatIOnal PhYSICS, 150 (1999), pp 561-594
[12] D LEVY, G PUPPO AND G Russo, Central WENO schemes Jor hyperbolzc systems oj conservatzon
laws, MathematIcal Modelhng and Numencal AnalysIs (AJ 2 AN), 33 (1999), pp 547-571
[13] D LEVY, G PUPPO AND G Russo, Compact central WENO schemes Jor multzdzmenslOnal conser
vatzon laws, SIAM Journal on SCIentIfic ComputIllg, 22 (2000), pp 656-672
[14] D LEVY, G PUPPO AND G Russo, A thzrd order central WENO scheme Jor 2D conservatzon laws,
Appl Numer r..Iath 33 (2000), pp 415-421
[15] S LIANG AND H CHEN, Numerzcal szmuiatzon oj underwater blast-wave Jocuszng uszng a lugh-order
scheme, AlA A J , 37 (1999), pp 1010-1013
[16] R LISKA AND B \VENDROFF, Composzte schemes Jor conservatzon laws, SIAr..I.J Numer Anal, 35
(1998), pp 2250-2271
[17] R LISKA AND B \VENDROFF, Two-dzmenslOnal shallow water equatzons by composzte schemes, lnt J
Numer r..leth FI, 30 (1999), pp 461-479
[18] X -D LIU, S OSHER AND T CHAN, Wezghted essentzally non-osczllatory scheme5, Journal of Com
putatIOnal PhYSICS, 115 (1994), pp 200-212
[19] P MONTARNAL AND C - \V SIIU, Real gas computatzon usmg an energy reiaxatzon method and lugh
order WENO schemes, Journal of ComputatlOnal PhysIcs, 148 (1999), pp 59-80
[20] S NOELLE, The MoT-ICE a new hzgh-resolutzon wave-propagatzon algorzthm Jor multzdzmenslOnal
systems oj conservatzon laws based on Fey's method oj transport, Journal of ComputatIOnal PhYSICS,
164 (2000), pp 283-334
[21] C - \V SlIU, Essentzally non-osczllatory and wezghted essentzally non-osczllatory schemes Jor hyper
bolzc conservatzon laws, III Advanced Numerzcal Approxzmatzon oj Nonlznear Hyperbolzc EquatlOns,
B Cockburn, C Johnson, C -W Shu and E Tadmor (EdItor A Quarterom), Lecture Notes III
r..lathematIcs, volume 1697, Spnnger, 1998, pp 325-432
[22] C -\V SlIU, Hzgh order ENO and WENO schemes Jor computatzonal fluzd dynamzcs, III Hzgh-Order
Methods Jor Computatzonal Physzcs, T J Barth and H DeconIllck, edItors, Lecture Notes III Com
putatIOnal SCIence and EngIlleenng, volume 9, Spnnger, 1999, pp 439-582
[23] C -\V SHU AND S OSHER, Efficzent zmpiementatzon oj essentzally non-osczllatory shock capturzng
schemes, Journal of ComputatIOnal PhYSICS, 77 (1988), pp 439-471
[24] C - \V SlIU AND S OSHER, Efficzent zmpiementatzon oj essentzally non-osczllatory shock capturzng
schemes, II, Journal of ComputatIOnal PhYSICS, 83 (1989), pp 32-78
[25] P \VOODWARD AND P COLELLA, The numerzcal szmuiatzon oj two-dzmenslOnai fluzd flow WIth strong
shocks, Journal of ComputatIOnal PhYSICS, 54 (1984) pp 115-173
[26] J YANG, S YANG, Y CHEN AND C Hsu, Implzczt wezghted ENO schemes Jor the three-dzmenslOnal
16
REPORT DOCUMENTATION PAGE Form Approved
OMB No 0704-0188
Public reporting burden for this collection of mformatlon IS estimated to average 1 hour per response, mcludlng the time for reviewing instructions searchmg eXisting data SOurces gathering and maintaining the data needed and completing and reviewing the collection of information Send comments regarding thIS burden estimate or any other aspect of this collection of information, mcludlng suggestions for reduCIng this burden to Washington Headquarters Services Directorate for Information Operations and Reports 1215 Jefferson DavIS Highway SUite 1204, Arlington VA 22202 4302 and to the Ofoce of Management and Budget, Paperwork Reduction Project (07040188) Washington DC 20503
1 AGENCY USE ONLY(Leave blank) 2 REPORT DATE 3 REPORT TYPE AND DATES COVERED
December 2000 Contractor Report 4 TITLE AND SUBTITLE 5 FUNDING NUMBERS
A techlllque of treatmg negatIve weIghts m 'VENO schemes C NASl-97046 WU 503-90-52-01
6 AUTHOR(S)
Jmg SIll, Changqmg Hu, and CIll-'Vang Shu
7 PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) 8 PERFORMING ORGANIZATION
ICASE REPORT NUMBER
Mall Stop 132C ICASE Report No 2000-49
NASA Langley Research Center Hampton, VA 23681-2199
9 SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES) 10 SPONSORING/MONITORING
NatIOnal AeronautIcs and Space AdnlllllstratlOn AGENCY REPORT NUMBER
Langley Research Center NASA/CR-2000-210632 Hampton, VA 23681-2199 ICASE Report No 2000-49
11 SUPPLEMENTARY NOTES
Langley Techlllcal l\10111tor Denllls]\1 Bushnell Fmal Report SubmItted to the Journal of ComputatIOnal PhysIcs
12a DISTRIBUTION/AVAILABILITY STATEMENT 12b DISTRIBUTION CODE
U nclasslfied-U nlul1l ted Subject Category 64 DIstrIbutIOn Nonstandard AVaIlabIlIty NASA-CASI (301) 621-0390
13 ABSTRACT (Maximum 200 words)
HIgh order accurate weIghted essentIally non-OSCIllatory ('VENO) schemes have recently been developed for filllte dIfference and filllte volume methods both m structured and m unstructured meshes A key Idea m WENO scheme IS a llllear combmatlOn of lower order fluxes or reconstructIOns to obtam a hIgher order approxlInatlOn The combmatlOn coeffiCIents, also called lmear weIghts, are deternuned by local geometry of the mesh and order of accuracy and may become negatIve 'VENO procedures cannot be applIed dIrectly to obtam a stable scheme If negatIve llllear weIghts are present PrevIous strategy for handllllg thIs dIfficulty IS by eIther regrouplllg of stenCIls or reducmg the order of accuracy to get rId of the negatIve llllear weIghts In tIllS paper \\e present a SImple and effectIve techlllque for handlmg negatIve lmear \\elghts WIthout a need to get rId of them Test cases are shown to Illustrate the stabIlIty and accuracy of tIllS approach
14 SUBJECT TERMS 15 NUMBER OF PAGES weIghted essentIally non-OSCIllatory, negatIve weIghts, stabIlIty, lllgh order accuracy, 22 shock calculatIOn 16 PRICE CODE
A03 17 SECURITY CLASSIFICATION 18 SECURITY CLASSIFICATION 19 SECURITY CLASSIFICATION 20 LIMITATION
OF REPORT OF THIS PAGE OF ABSTRACT OF ABSTRACT UnclassIfied UnclassIfied
NSN 7540-01-280-5500 Standard Form 298.(Rev 2-8.9) Prescrtbed by ANSI Std Z39 18 298 102
mcompresszble Navzer-Stokes equatzons, Journal of ComputatIOnal PhYSICS, 146 (1998), pp 464-487
17
End of Document