idna-tr-88-286 ad-a211 728 - dtic › dtic › tr › fulltext › u2 › a211728.pdf · t )a rvoe...
TRANSCRIPT
,' . IDNA-TR-88-286
AD-A211 728
SOME NEW ALGORITHMS FOR BOUNDARY VALUE PROBLEMSIN SMOOTH PARTICLE HYDRODYNAMICS
P. M. CampbellMission Research Corporation1720 Randolph Road, SEAlbuquerque, NM 87106-4245
6 June 1989
Technical Report
CONTRACT No. DNA 001-88-C-0079
Approved for public release;distribution is unlimited.
THIS WORK WAS SPONSORED BY THE DEFENSE NUCLEAR AGENCYUNDER RDT&E RMC CODE B3220854662 RB RC 00135 25904D.
DDTICQ AUG24 1989
Prepared for UDirectorDefense Nuclear AgencyWashington, DC 20305-1000
z
DISTRIBUTION LIST UPDATE
This mailer is provided to enable DNA to maintain current distribution lists for reports. We would appreciate your providing the requested information.
0 Add the individual listed to your distribution list.
0 Delete the cited organization/ individual.
0 Change of address.
NAME:
ORGANIZATION: - ---- ---------- - -------- -
OLD ADDRESS CURRENT ADDRESS
~ SUBJECT AREA(s) OF INTEREST: 1-w a::: 0 z < w a::: w I 1-:::J . (.)
DNA OR OTHER GOVERNMENT CONTRACT NUMBER: -----------·
CERTIFICATION OF NEED-TO-KNOW BY GOVERNMENT SPONSOR ( if other than DNA):
SPONSORING ORGANIZATION:
CONTRACTING OFFICER OR REPRESENTATIVE: ___ _
SIGNATURE:
DirectorDefense Nuclear AgencyATTN: TITLWashington, DC 20305-1000
DirectorDefense Nuclear AgencyATTN: TITLWashington. DC 20305-1000
'
UNCLASSIFIED ' : - '... . -) ) - . '
. - . -REPORT DOCUMENTATION PAGE
' • ~t'Oil" ;;:c_~: ;v C..l$) .;: (.l-~0111 · ~ ~:s·~.c· vE V.A~< 'IIGS
UNCLASS I FlED Za ;;:( .,~l Tv C • .l.SS•F'C.l- 7.0111 Au T~ORI TY l )15"~18UT ON ' .l.JAILA31L"" OF 'lE~Q;>r
N/ A s ince Unclass ified fo r r elease ; distribution Approve d public ~ :, ::li:C.ASS,i' ·CA T'O N . ::>OWI'.G~Ai::>·I'.G SC;..Ei)ULE i s unlimited.
N/ A since Unclassified 4 ·~ ~>O'lMiNG ORGANIZA T.ON 'IE PORT Nl,;M8ER(S) s MONITO~I'.IG .:lRG.l.NIZA ~ 0111 'lE~CRT 'IlL. 'V1 ac~IS )
NRC/ABQ-R-1100 DNA- TR- 88- 286
6• 'IIAME Oi' >E'IFORMING ORGANIZATION 6b OF=•C: )V'IIIiiO~ 7a 'IIAME OF 'IIIONITO'l ii'.G O'IG.>.N 1ZA~.CN
Hiss ion Research (If aoolicatJI•J Director Corporation Defense Nuclear Agency
~(. ADDRE)S ,Ciry, Star•. and ZIPCIXMJ 7b ADDRESS ;c ;ry, Stat•. 1nd l iP ~0<1•/
1720 Randolph Road, SE Albuquerque , NN 87106-4245 \~ashington, DC 20305-1000
a •. 'IIAME OF =•JND1NG , S?ONSORING 8b O>F·C: SV:IJI80L ·) PIIOC'JRE\1ENT I'.S~'IL. :V.EI\jT QEI\jT'F•CAT'ON ' 11 .. "<18£'1 ORGANIZATION Ctf •ooticatJI•J
RAAE /Wittwe r DNA 001-88- C-0079
9c. ADOIIESS CC•ry. Srar•. and liPCIXMJ ' 0 SOURCE ')F -. ..... D•IIIG '11\..M SE~S
PROGRAM >IIO, ECT "ASK ·:•;O~I< .. N.; : LEV'IENT NO '110 '110 ACC: SS·OIII NO
62715H RB RC DH008713
.. n •~: (lnc/ucH S•cuflry Cl•u•f•car•on)
SONE NEW ALGORITHHS FOR BOUND.\RY VALUE PROBLEHS I N SNOOTH PARTICLE HYDiWDYNANI CS
11 ~~'ISONAL Av " ;..OR(S)
Campbe ll, Phil N. t )a rvoe OF 'IE?ORT 11)b TIME COVERED r4 )A T< OF 'IE?OFH .'fur Mont,. Day) rs >A.Go COUNT
Technical =ROM 880606 TO 890606 890606 46 16 SuPPLEMENTARY 'IIO TA TION
This \vork was sponsored hy the Defense Nuc l ear Agency unde r RDT&E RNC Code BJ220854662 Rl3 RC 00135 25904D .
17 COSA Tl CODES 18 SU8 •ECT r:~MS \COntlflu• on '•••rs• '' necess•ry 4r>d •aenrlfy Oy Oloclc numO•rJ : .e~o G~OUP SU8-G~OUP Computa t ional Physi cs
20 4 ....- ) Fluid Dynamics .
I Hea t Flow , ( 11,- 'j 1,----19 A.oiS":!ACT \Conrlflu• on r•v•rr• ,f ,.c.iury and •MMtfy lly oloclr tfumO.r)
~ A r ede r ivation of the basic equat i ons of smoothed particle hydrodynamics i s g iven tha t removes some o( the intui t'i ve e lement s in ea t·lier work. New algor i thms .1 1· e ob t ai ne d for hydrodynami cs lvith hea t flow tn-:1t-..:include bounda r y t e rms . These a l gorithms pe rmit appl ication of the method to sys t e ms driven rv e xterna l pressures or heat sources . Hydrodynami · work terms a r e J iffer ent in the ne\v algor ithms and these terms may e limina te some difficulties encounte r ed with expressions (ounJ i n the 1 ite r a tor-e . The met hod developed for thermal diffusion y ields the fl uxes direc tly a nd may be ext e nded to a f lux limi ted dif f usion formulation of r adiation t rans port. /( rl,l}. (-~, .. , '
ZO DISTRIBUTIO N ' A \IAILA91LITY Of ABSTRACT 21 .\oST~ACT SECv RirY CLASSIFICATION
0 UNCLASSIFIEO•UNLII\IIITEO QD SAME AS ~PT 0 :>TIC u SERS UNCLASSlFI ED l2a NAME Of RESPONSIBLE INDIVIDUAL 22b. TELEP~ONE (/nc/ucH Ar., CucH) I Zlc OFFICE SYMBOL
Bennie F. Naddox (202) 325-7042 DNA/CST! DD FO"M 147), 84 MAR 8) APR ea1t10n may o• used ""t1l e•naustee1
All Oth•r ..:lot iOns are O!HOI•te. SECURITY CLASSIFICATION OF T~1S PAGE
UNCLASSIFI ED i
•. UNCLASSIFIED SECURITY CLASSIFICATION OF THIS PAGE
ii
SECURITY CLASSIFICATION OF THIS PAGE
U~JCL\SSIFIED
'
PREFACE
The author thanks Robert F. Stellingwerf for introducing him to the SIImethod and for many enlightening discussions. This work was supported under Con-tract DNA 001-88-C-0079.
Accession ForNTIS GRA&I
D ?~.t TAB0
Unannc~anced L
D4i3tribut Ionl/-_____
Avqltlability Code s
'Avail and/or
Dist special
CONVERSION TABLE Convenion fac:ton for U.S. Customary to metric (SI) units of measurement
MULTIPLY -------~BY-----~~- TOGET TOGET <11111141-------- BY4 DIVIDE
ancstrom
atmoapllere tiK'rm31l
bar bam
Brollsh lllerm31 unll tllle rmochemicall
calori" tlllermochemlrall
cal Uhermochemicoll / cm2
curie
decree tangle)
decree F:>llrt'nhell
elect ron ,·olt
e ...
e"'/second
foot
foot-pooJnd -force
~~~lion ll'. S. liqu1dl
1nch
Jer~
joule/kilot~nm (J / I<ct (radiation dose absorbed)
kilotons
kip 11000 lb()
k1p/lnch2 tke1l
ktap
mtcron
md
mile (lnternataon:aiJ
ounce
pound-!o n ·c tlh,. ;H ott'tiuptu!~;)
pound ·for<'<! 111rh
pound ·force I 111ch
pound ·force/foot~ ~
pound-f!'rcc/ anch .. (fl~l)
pound-mass llbm O\ Oinlup01•1
pound-m:ass-£oot2 lmomcnt of i:t~rtta)
rad tr~1.111on dos'! 3bsorbcdl
roentccn
shuc ..... torr tmm Hit. o• Cl
I. ooo ooo X E -10
I 013 2~ X E •2
IOOOOOOX£•2
I . 000 000 X E ·28
I . OS4 3SO X E •3
4 184 000
4. 184 000 X E -2
3 700 00'1 X E • I
I . 74S 329 X E ·2
'• = tt' r. 4S9. 671/1 . a I. 602 19 X E -19
I . 000 000 X E -7
I. 000 000 X E -7
3. o~s ooo x E -1
I. 3SS 818
3 78S 410: X E -3
2. S40 000 X E -2
I 000 000 X E •9
I. 000 000
4 . 183
4 . 448 222 X E •3
6 894 7S7 X E • 3
I . 000 000 X E • 2
I 000 MO X E -~
2. S40 000 X E ·S
I. 609 344 X E • 3
2. 834 9S2 X E ·2
4 . 44~222
I . 129 ~48 X E -I
I 7SI 26~ X E • 2
4. 788 026 X E -2
6 . ~94 7S7
4 . S3S 924 X E ·I
4 214 011 X E · 2
I~Oili46.<E•I
I . 000 000 X E -2
2 S79 760 X E ·4
I 000 '"'" X E - 8
I . 4S9 390 X E • I
I . 333 2 2 X E · I
•The bcc,tuttr•: l tBq) 1:; the Sl un1t of r :tdto:atll\"tty ; 1 8q ':: 1 ~vent/s . ••Tht' Gr:.~· tG)) 1S the 51 un1t o( :.bsorbcd r:.d1ahon .
IV
met.rs tml
kilo paacal tkPal
kilo pascal tkPal
meter2 tm21
joule tJI
joule tJ)
meca joule/ m2 tMJ/m2t
"Cica becquerel (GBql
radian trad)
decree kelvtn (K)
joule tJI
JOUle tJ)
watt tWI
met.r tml
joule tJI
meter3 tm3)
meter tml
Joule tJI
Gray (Gyl
terajoulea
newton IN)
kilo paacal tkPal
newton·lfCand / m2
tN-a / m2)
meter lml
meter tm)
m<!ler tml
k1l01rom ll<cl
newton tNI
newton - meter tN·ml
newton / meter IN/ ml
k1lo pascal tkPal
k1lo pascal tkPal
kiiOflram tkcl
ki IOflnm -mete r2
lkc · m21
kllot~ram/merer3 l"c / m3)
• '(;ray (Gyl
coulo mb/kil01ram ll'/l<cl
oecond tal
kdot~ram l"cl
k1lo pucal tkPol
•
TABLE OF CONTENTS
Sect ion Page
PREFACE iii
CONVERSION TABLE iv
1 INTRODUCTION 1
2 KERNEL ESTIMATES NEAR BOUNDARIES 2
3 MOMENTUM EQUATION WITH BOUNDARY PRESSURE 5
4 EQUIVALENT FINITE DIFFERENCE FORMS 10
5 WORK TERMS IN THE ENERGY EQUATION 15
6 HEAT CONDUCTION TERMS IN THE ENERGYEQUATION 19
7 CRITIQUE OF BROOKSHAW'S METHOD OF CALCU-
LATING THERMAL DIFFUSION 22
8 CRITIQUE OF SOME MANIPULATIONS IN THE
LITERATURE 26
9 EXTENSION OF HIGHER DIMENSIONS 28
10 SUMMARY OF NEW SPH ALGORITHMS 34
11 LIST OF REFERENCES 36
SECTION 1
INTRODUCTION
Although particle methods in hydrodynamics have been remarkably suc-
cessful in many problems, 1- 4 current methods suffer from an inadequate treatmentof boundary conditions. This is particularly evident when one tries to include heattransport in the energy equation.5 -6 Without accurate values of the flux at the bound-aries, it is not possible to calculate the net exchange of energy between a fluid and itsenvironment.
Another class of problems requiring accurate treatment of boundary condi-tions is illustrated by the shock tube problem where the shock is driven by a piston.
In this case the particle method must be able to accommodate an externally-applied (boundary) pressure. The method should also be able to handle conductionheating of the gas by a hot wall or piston; and, finally, boundaries must be treate(laccurately to include radiative heat exchange, in particular, radiative cooling.
In this paper, some new algorithms are developed for smooth particle hy-(Iro(lynlamics in problems where external boundary conditions are imposed. Althouigihthermal and radiative diffusion are included, radiation transport is neglected in thepresernt paper.
U WIII nllllll m l
SECTION 2
KERNEL ESTIMATES NEAR BOUNDARIES
The standard kernel estimate of a function is given by 3
ff,(x) = f(x')W(x - x', h)dx', (I
where the kernel is normalized by
XJim W(x- x',h) dx' 1. (2)h -,o f
In these expressions the domain of definition is 0 < x < X.
As long as h -* 0 one would not expect normalization problems near theboundary, but ini practical applications kernels with finite h are used. One would thenexpect difficulties within x - h of the boundary because the integrals are truncated bythe boundary. For example
lir W(x - x')dx' (Z- X f 2
which is obvious from Fig. 1.
Consider what this does to the kernel estimate of the density. Assume a denseconstant spacing of particles in the vicinity of the boundary X. Far from the wall, thekernel estimate gives
pW) jp (x') W(x - x', h) dx'
Srn W(x- - ,h)
P I
2
x x
2h 2h 2h h -_
Figure 1. A typical smoothing kernel, showing how the
symmetrical form is truncated as the particle ap-proaches a boundary.
where pC is the constant value. However, following Eq. 3, as x -* X, the kernel estirate
becornes
p(X) lim p(x')W(x - x',h)dx'
- m W(X-xh)
1PC (5)
So within -- . of the boundary the interpolation feature of the kernel estiniate giv's
spurious resuls. Also, presumably, one finds
P(X) lirn P(x')V, (x - x')dx'
2
for the case of a uniform pressure in the neighborliood of the boundary as in Fig. 2.
3
-PC
1/2 PC
h X
Figure 2. The falloff in a uniform pressure distribution asgiven by a kernel estimate in the neighborhoodof the boundary.
In summary, we have seen that because of the truncation of the normalizationintegral within -- h of a boundary, interpolation implied by the kernel estimate rTnaynot be accurate. Renormalization is a possibility but does not appear to work. Also.the boundary value method presented below seems to require the apparent failuremnanifested in Eq. 3.
4
SECTION 3
MOMENTUM EQUATION WITH BOUNDARYPRESSURE
All treatments of SPH of which the author is aware require integration byparts when deriving kernel estimates but simply drop the boundary terms. (See Section8 below.) This is presumably because the kernel is supposed to mimic a 6-function
which goes to zero sufficiently far from the particles representing the edge of the system.
However, if we are to consider problems where the particles interact with aboundary condition, such as an externally-applied pressure, we must specifically includethe boundary terms. In this section the two procedures developed by Monaghan forthe momentum equation 2' are modified to include boundary pressure.
In the first case, write the momentum equation as
dv 1 9Pdt p dx
-4( a. (7)
The kernel estimate of this equation is formed by changing the independent variableto x', multiplying by W(x - x')dx' and integrating over the domain 0 < x < X. Oneo)t.ains
dv, fX a !'j( z)x
dt , (' P)
-- (P) f , W-(x - x')d,()
where the subscrip,. . ienotes "kernel estimate." The second integral has been lin-carized by eval,iving (P/p") by its kernel estimate and taking it outside the integralsign. This is justified since W(x -- x') acts like 6(x - x').
5
After integrating by parts, one finds
dv, - [PW ( j ' x (P) aW(x- 'ddt p 10 P a
())lWxo '- ( P) jX paW (xX') dX', (9)P oP
where we have used dW/drzx -OW/ox' assuming a symmetric kernel.
If we now evaluate the integrals by the particle method, replacing
f f(x')W(x - z') dz' fj W (x - . , (10)i Pi
we obtain an expression for the momentum equation with external boundary pressures,
dto -+ p W (x - o)
2 2 ± W- X)
Thus, particle i feels the effect of the boundary pressure if it is within range of theboundary and W(xi - o) ) 0. The boundary pressure enters in a term similar to thatfor any other particle j.
Does this expression conserve momentum? If we multiply by m, and sum.we find
6
d5 re, v, Po m, mI ) + il W(x, - o)
-/P /"\ dm, +~PX Mi X + ilWX
-ZZ j 2~m 2 (9 Xj (12)• j (Pj (Pil 0x
The last term in the equation vanishes because OW/ax, is antisymmetric in i andj. To get a form where P, and Pj appear symmetrically was the main point in themanipulation of P and p in Eq. 7. This insures the vanishing of the last term in Eq. 12which is necessary for conservation.
Now what about the boundary terms? From Eq. 3, we have
1
m,W(x, - o) = I , (13)i2
so that Eq. 12 becomes
d 1E- MiV, -Po+0 Oop0 M 2W (x1 ,-o)
dt 2 P
1P - Px Mi f )W(xt-X). (14)~ ~ P P -, \/-
Here P, and Px are the given boundary values of the external pressure. The otherterms are the interpolated values of the pressure at the boundary given by the internalcalculation. If this calculation is accurate (apart from the factor of 1/2), it must yield
1 ( ;j )0i i
and
7
Px - px 1: m, W(x, - X), (16)i (Pi
giving for Eq. 12
d mivi Po - Px (17)dt
A similar analysis is possible for Monaghan's other conservative form of themomentum equation. First write
dv 1 OP p11 2 O9p/2- 2 (18)dt p ax p ax
Then the kernel estimate of the momentum equation becomes
d = -2 p) 10 ax' W(x - x')dx'
= -2 p1/ -)p12W x
-2 (p) / 2-- -- (X - x')dx' . (19)
Evaluating the integral by the particle method gives
dv, - 2 (F)2 [p 0 '/'W(Xi - o) -pXk'2W(Xi _ X)]
- 2W(Xi Xi) (20)) PiP, Oxi
where, again, we note the fortuitous appearance of the factor of 2 in the bounldaryterms which will compensate the falloff represented by Eq. 3.
8
After multiplying by m and summing, we obtain
d E-nivj = 2P,/ 2 m pm, W(X, - o)
ii P )
- 2 E E mimjP ,' 2 W (x-x) (21)y PiPj 09xi
where because of the symmetry in i and j, the last term is zero; and since
1 p,/2 - i Em, /2 f~ W(X, -o0)2 P 1i
2m, ("p1W(X, - X) , (22)
we get the conservation of momentum,
d-E Mv = P - Px (23)
It is curious that in both of these forms the boundary terms enter in such away that the factor of 1/2 from the normalization anomaly at the boundary is requiredfor a conservative momentum equation. This is not true of all forms, however. Forexample, the first expression for the pressure gradient in Eq. 7 does not produce aboundary term with the factor of 2 present.
9
SECTION 4
EQUIVALENT FINITE DIFFERENCE FORMS
If particles are assumed equally spaced, it is possible to derive the equivalentfinite difference form which the kernel estimate represents.6 It is instructive to look atthese forms, particularly in the neighborhood of the boundary, to get a feeling for theaccuracy of the particle method.
For the purpose of illustration, consider the kernel shown together with itsderivative in Fig. 3,26
1(2 u2 I 0 < lul <
W(x-' (2- u) 3 < u 2 (21)6h( Iul
0 2 < Jul
where u (x - x')/h, and
w = law (25)
Ox h Ou
Now, consider a series of equally spaced equal mass particles far from anyboundary as shown in Fig. 4. The momentum equation in the form of Eq. 11 is
10
W
h 2h
aw/a x
Figure 3. The smoothing kernel given by Eq. 24 and its(continuous) first derivative.
1l
dv, ME +_________ -xj
l rn[( )+( )] (2l ) (l6)
R P2i+1 + il
since OV ' 1/Dx, is zero -xcept for j =I - 1 and j i + I as can be seen from Fig.3.From Eqs. 24-25 we have
-1
19W (x, - Xj) 0 (27)
Oxi
which leads to Eq. 26 above. This simplifies to the central difference formula
dt 2h i+l i-
where we have used rn/p = h for equally spaced particles.
Let us now consider the case where the boundary falls within h to the right ofparticle i. In this case, the j = i + 1 term is missing but the boundary term is present.Specifically, let us assume the boundary X occurs -- 0.6h to the right of particle i whereW(x, - X) - 1/2h. Eq. 11 then becomes
dv[dt P xK (L)x + (M il Px()
M m [(p 2h12 (As (29)
12
i-2 i-I i i+1 i+2
(a)
Figure 4(a). A series of equally-spaced particles far from a boundary.
p ._i Pi P --
2h
i-I i _
(b)
Figure 4(b). Equivalent difference form for the pressure gradient in theneighborhood of the boundary.
13
since W(x1 - X) = 1/2h and the j = i+1 term is missing. Using h = m/p, this reduces
to
dv, (310[(
dt 2h x (0i-1
which is a fairly good approximation. As shown in Fig. 4 it gives a central differenceforin based on Px situated 2h from Pi-1 , although it was assumed that the boundarywas -- 1.6h from xi-1. All finite difference algorithms are uncertain to within Ax = h,so these boundary terms are probably as good as one would expect of most differencemethods.
14
SECTION 5
WORK TERMS IN THE ENERGY EQUATION
We now consider the work terms which appear in the energy equation in thepresence of boundary pressures. In order to insure conservation, this term must beexpressed in a form that is complimentary to the right-hand side of the momentumequation so that the kinetic energy appears correctly.
We write the work terms as
d P Ov
dt p ax
- a + V a a
where c is the internal energy per unit mass. Taking the kernel estimate of Eq. 31yields
dc ~ ~ W~ -xa P x')dx'
+ Vj~ ~ W(x - x')dx' ,(32)
where the second term has been linearized as before. This term is now rewritten usingthe kernel estimate for the momentum equation,
dt, Xa I W (x x)dx - ( P),] f X -a 'V (X - X')(dir' . (33)dt (),1
15
If we multiply by v, we get
vKj~ ~ W(x -x')dx'=-(m,)fX ,W(x - x')dx', (3)
which gives for the energy equation
- () x°w(- x)dx' (37)
Next, integrating by parts we obtain
& , dv,, [ W[ -
dt + Rt - p 0o P ax
-- (PV X)p-a (x - x')dx' (36)
When this expression is evaluated by the particle method, we get the energy
equation expressed in terms of total energy,
di + di -12+ = o)V- [ ) 2 + 2 ),] xW(, - X)
- Vm + (k)] aw(x, -x,) (37)
16
We can verify that Eq. 37 insures conservation if we multiply by 7n, and sum,
d12) 1 (PV" _____-
I I P(iPi Ps
(Pv)x- PxE miP2 S i POi
-F ZPm)m, a VV W(X-~ (38)
The left-hand side is the time rate of change of total energy, internal plus kinetic. Thedouble sum on the right is zero because of the antisymmetry of OW,/Ox,. And asbefore, we have
Po mi -PW - 0Pi Pi 2
Px _m, (Pv W X -1 (Pv)x , (1 (,))
giving
d 7(f1 = (PV) - (Pv)x()
dt 2vi
The energy Fq. 37 is expressed in terms of the total energy. After the parliclesare imoved using the momentum equation, the change in kinetic energy is known. ThenEq. 37 can be solved f - r the internal energy (,. In sone cases, it may be desirable tohave the energy equation expressed in terms of the internal energy alone.
17
The explicit appearance of the kinetic energy term can be eliminated usingthe momentum Eq. 11,
v( v + poW(xi - o)
- ( v ( PXW(X - X)x P
-Zm, + 8W,+ 1-ox,)()j () j Vii P ' wx j
After subtraction, this leads to a form of the energy equation where the kinetic energydoes not appear explicitly,
[V0
dt [VX - (j W(X - X)
- > m, [Vi - d,[ OX, (42)
In summary, we have developed a conservative expression, Eq. 37, for thework done on a fluid by pressure forces that is complementary to the momentum Eq. IIin the presence of externally-applied (boundary) pressures. These two equations insureIhe, conservation of momentum and energy.
18
SECTION 6
HEAT CONDUCTION TERMS IN THE ENERGYEQUATION
Let us now consider the addition of thermal heat conduction terms in theenergy equation. This will be especially important for ICF applications. Neglectingother terms for the moment, the energy equation becomes
dc 1a ( _
dt pax (KaxIlaFp ax(,a
where K is the thermal conductivity, T the temperature and F the heat flux. Weshall approach this equation in the same way as the momentum equation and obtain acompletely parallel formulation.
Following Section 3, we write
d a (aF\ FOpdt ax p : plx
Which has the following kernel estimate
d(k a ( W(x x')dx'
dt Z, a X' \p
( f x ap W(x - x')d x' (.15)
After integration by parts, one obtains
19
-f(x - x')dx'dt - 10 ki a
(F [pW(Xx,
(F) LxpaW(x-x)dx, (1I6)
which in particle form becomes
- [(D + ()] pxW(x, -d--F : + (F]PeW(xi _ )
o 2
- _m (F ) + ( F )] OW(xi ) (47
P i P (x
It is clear that energy conservation follows in the same way as demonstrated above.
Thus, in Eq. 47 we have the energy equation with heat conduction in. aform completely analogous to the momentum equation, Eq. 11. The boundary valuesfor the fluxes, F, and Fx appear in a similar way to the boundary pressures. Andsince experience has proven Eq. 11 to be a successful particle form of the momentumequation, one would expect similar results for Eq. 47.
There remains only to calculate the fluxes for use in Eq. 47. We write F inthe form
F 10T
0- (T) K-T apo- .=- K~ (-8)
which, by a now-familiar procedure, leads to the particle equation,
20
(~i) K, + poW(X, - o)
- [i (T\x ('T)]\xiX
- K m + W(x - X)(49)
S 2 i P2axi
So the temperatures define the fluxes through Eq. 49 and the fluxes determine theenergy transport through Eq. 47.
It is possible to develop an equivalent treatment of heat conduction basedon the product of square root approach used in Eq. 18, but this is unlikely to be anybetter than the method already given. Other approaches have been proposed for heatconduction (which neglect boundary terms, however). The method of Brookshaw isdiscussed in the following section, and a variation of Brookshaw's method by Monaghangives similar results in test calculations.5
21
SECTION 7
CRITIQUE OF BROOKSHAW'S METHOD OFCALCULATING THERMAL DIFFUSION
In a 1985 paper,6 Brookshaw discusses several approaches to adding radiativeheat diffusion to SPII. He describes Lucy's early attempt 7 at solving the energy equation
dS aFdt ax
F 4acT' i9T3Kp ax' (50)
by using the particle method
( T d S ) . = _ m F j ' W x idt ,j P 0x
F = _( 4acT 3\i (T) aW(x -x,) (13Kp )mj- Px (51)X
It is stated that since W(x, - xj) --+ 0 at the boundary of the fluid, T - 0,and one has an insulating boundary condition. The statement is made, "...so there isno reason to assume that the SPH equation will automatically take into account thecorrect boundary condition."
It is easy to see where Lucy and Brookshaw go wrong here. In order to get theparticle equations, Eq. 51, they had to integrate by parts and drop the very boundaryterms they are so concerned about. For example
IT -I -W(x -x')dx'dt j a x'
, x, aW
-FW(x - x')Io - 10 F dx (x - x')dx', (52)
22
which becomes in particle form
(dS) FpT- FW(x,-o)-FxW(x, -X)
-m (F OW x- Xj) (5)P i x
Even though boundary terms are present here, this is not a good form touse, since it does not guarantee conservation. It is better to use the method developedin Section 6. Test calculations showed Eq. 51 was inaccurate near the boundary andalso the method was sensitive to particle positions. Brookshaw presented a differentapproach to calculating heat diffusion based on a difference algorithm that he claimedgives much smoother results than Eq. 51. The method given in Section 6 should beaccurate near the boundaries, but it may be that Brookshaw's approach would giveadded smoothness. For this reason, we shall attempt to derive Brookshaw's algorithmwith boundary terms included.
Write the energy equation as
dc a a) 154)P dt -~ a(K -O9x]
and form the kernel estimate
PKd K-: Jx 'IK 9}IWx -x'dxI, (55)
where curly brackets are used to indicate a special treatment for that particular term.Integrating by parts yields
P, K- w(x - X') + f ( (- X')dx'. (56)
23
Now, Brookshaw appeals to a Taylor series expansion to write a differenceform for the term in curly brackets.
Since
K(x') K(x) - (X- X') OKax
aT I12T =x) x - (x- x')- + (X- x -)'- + ..,(57)
we can write
K - - [K(x) + K(x')] T(x) - T(x')axT OK T (
2K(x) aT _(X - XI) K(x)--a2T + aK aTJ + (58)
The kernel estimate, Eq. 56, becomes
dE, {K cI} W(X -X')]X-~
fX T(x) - T(X') _(+ f tK(x) + K(z')] x' W(x - x')dx' (59)x ax
where Brookshaw only discusses the integral and assumes everything goes to zero atthe boundaries.
It is interesting to see something like 2 k under the integral sigl,. To seehow the difference form gives the intended heat conduction term, substitute Eq. 58into Eq. 59
dc T a a7') 1
'd-- - (X - X) - yK ) + ... W(X - X')
fX~ I _ (x-_ x)~ -a -(9T) + _._(+o 2KI1X.. -(-') K + . W(x - x') dx' .((0)
24
When the integral is performed, term-by-term cancellation gives
K dE- + O(h 2) (61)
which is the result obtained by Brookshaw. Including the boundary terms in theintegration by parts does not change the result given by series expansion.
Brookshaw's scheme with boundary terms can now be written as
df, T(x T( )(x - ))
P" dt [{K(x) + K(x')} T(x)- - 'dt X - X1
+ 1 [K(x) + K(x')] T(x) - T(x') a W(x - x') d' ,(62)oX - X ax ~ 62
which gives in particle form
__ 1 T, -T 1 T,_-Tx_
S K o) [ W(X-)+ K, + KxI - - X W (Xi -x)dt Pi xi - o Xi - X
+ m Y- (KS. + Ki) T" - T7 a W (Xi - xj) .(63)j PiPj xi - xj axi
It is clear that Eq. 63 gives conservation since the summation is antisymmetricin i and j and the necessary factors of 2 are present in the boundary terms. Whetherthis approach to heat conduction is better than that represented by Eqs. 47-49 can onlybe determined by numerical testing. Eq. 63 certainly solves the problem of an insulatingboundary condition, since any particle within 2h of the boundary (which presumablyis on the order of one diffusion length) will exchange energy with the environmenticprcscnted by the temperatures T, and Tx.
25
SECTION 8
CRITIQUE OF SOME MANIPULATIONS IN THELITERATURE
In the foregoing, it has been shown that boundary terms neglected in otherderivations of SPH algorithms2 - 7 are essential for many problems and particularlyfor energy transport problems. The question arises as to how these terms were soconsistently neglected over the last decade.
In Gingold and Monaghan's paper, 4 they approach the particle method asfollows:
Kernel Estimate P,,(x) = f P(x')W(x - x')dx',
Particle Approx. P, = E m, -P 2
The Gradient (= m 'WPxi - x( ) a (64)Ox, P a91
However, if one calculates the pressure gradient directly from the kernel estimate onefinds:
Kernel Esti~mate f~ W W(x - x') dx' ,(X, + / ' W x-x)xIntegration PWaP ( - x')dx'(1P
Particle Approx. 49P PxIV(x, - X) - PoW(x, - o)09x,
+ Zm,( P) aw(x,-x,) (65)j j 26,
26
This shows clearly that one must be careful of taking derivatives throughparticle approximations to kernel estimates. In Monaghan's 1982 paper,3 he mentionsintegration by parts but says he is assuming W, the function, or both go to zero onthe boundary. This is better than what one sees in References 4 and 6, because it ismathematically defined even though unnecessarily restrictive.
The point to be made here is that in dealing with SPH equations, it is saferto carry everything through to final form as an integral equation and only then replacethe integrals with sums over particles.
27
SECTION 9
EXTENSION TO HIGHER DIMENSIONS
Consider first the kernel estimate for the momentum equation, Eq. 7. Invector notation, this becomes
- _f V1 W(iF- /'I,h)d 3 rI
- f)J V'pW(I#F- i'1,h)d3r', (66)
where we assume the kernel is spherically summetric, extendiig over a range R - 2habout the point . The volume of integration V is therefore a sphere of radius I?centered on Fas in Fig. 5. An exception to this is when Fis within range of a boundaryas in Fig. 6. Then integration by parts yields a surface term, where the boundarypressure at the surface influences the particle at point .
Integrating by parts in Eq. 66 gives
dt7 IB (P) W(IF- F'l,h)d§' - f P(~ VW(jF- F, h)d r'
- (w('-F ',h)d§'- (- pVW(IF - F'Kh)d3 r', (6 7)
where the surface integrals are over that area of the boundary intercepted by the non-zero range of W. The particle approximation to the kernel estimate becomes
28
r
0
Figure 5. Range of influence of kernel W(IF- F'I) about thepoint F.
B
(IA
r
FBB 0
Figure 6. Range of influence of W intercepts boundaryplane B over area AS.
29
a<
Figure 7. The area on a plane boundary intercepted by thesphere is a disk of radius a.
d 6 - ) + 4) PBW (lr, - F ,h g
-(4 + (4)] VW(1t - Fjj,h), (68)
where PB and PB are the density and pressure at the boundary averaged over AS,, r13is the normal point at the boundary closest to , and i is the unit vector normal tothe boundary at ASi, all as shown in Fig. 6.
The calculation of ASi is fairly easy if one neglects curvature of the boundary.which is justified in most cases since h is supposed to be small. In three dimensions.
the area intercepted by a sphere of radius R located a distance d = Jr- rBI along thenormal to a plane boundary, as illustrated in Fig. 7, is
AS, = (R -I- )• (69)
In 2D rectangular geometry (x, y) one considers a unit length in z. Hence, the rangeof influence of W is a cylinder of radius R and unit length. In this case, which isillustrated in Fig. 8, the area intercepted by W on a plane boundary is
As, = 2 (R2 - i- frB) /. (70)
30
x
Figure 8. The area on a plane boundary intercepted by the
cylinder of unit length is a rectangle of height 2a.
'his formula does not apply to the situation in 2D cylindrical geometry where, be-cause of the symmetry, a particle is toroidal in shape. There does not appear to be
a simple expression that covers all boundary configurations in this geometry. In theone-dimensional cases discussed earlier, one considers a unit length in each of the two
transverse directions giving AS = 1. Eq. 68 then reduces to the ID expression derivedearlier. The area AS, in these three cases can be written generally as
AS, C(a)(R' - - B12) - ' , (71)
where a =z 1,2,3 is the geometrical order, and C(a) = (1,2, 7r), respectively. It should
be rioted that an exception to Eq. 71 may occur at the corner of intersecting boundaries.
The normalization anomaly discussed in Section 2 is not altered in highergeometries. In this case, one has
lim f W(Ir- 'i, h)d 3 r' (72)F-FB ,- 2 ,(2
and for a uniform distribution of particles in the neighborhood of the boundary,
imW(lB - FI,h) I p, (7:3)
31
Likewise, the boundary value of the pressure or any similar function in the particle approximation would be expressed as
(74)
In order to demonstrate conservation of momentum from Eq. 68, one obtains the total component of momentum in any direction x,
~ d~ A
~m'di ·:t -I
-ps L Fni(Ps~S.. x)W(Iii- fsl, h)
-ps }:m, (~~§. · x) W(lii- fsj,h), • p,
(75)
where the double sum vanishes because of the antisymmetry in i and i. The first term is the particle approximation to one-half of the boundary value of the net force on the system of particles,
1~ .... - L.J(P ~s, · z)s , 2 •
(76)
and by the same arguments presented earlier in Section 3, the second term should be a good approximation to this expression also. Conservation is then expressed as
~ d~ A ~(P"S .... A) L.Jm•y·x = L.J u ,.x B· i t i
(77)
32
By a similar series of steps the work terms in the energy equation of Section 5
are written
de + d ( ' 2 A ,[, (( -i;B1,hdt dt 2VKJ - + (SpW rj 2 Bh
- mi -() + (O) . VW(! - r'I,h), (78)
and conservation of energy follows in a like manner,
d 12 = (PA. ' - (79)
2 i
The heat flux terms in Section 6 are easily generalized to the form.
with the fluxes given by
-- K, + >r (w4) (4)] W ih)(8
(Ti
33
I
SECTION 10
SUMMARY OF NEW SPH ALGORITHMS
A suggested approach to SPH is summarized here that allows boundary valuesfor pressure, temperature and heat flux to be specified. This allows problems to besolved where the fluid is driven by an externally-applied pressure, temperature or both.It also allows the fluid to lose energy to a cooler environment. Although the diffusiontreatment is derived for thermal conduction, radiation which is in equilibrium with thematerial temperature can be handled in the same way.
The momentum equation for particle i is written
di B- + (P)i] PBW(I- rBj,h)ASjdt P
- P) + "
- jZ ( + Mh) VW(ji-F1 1,h) (82)
In this equation PB is the external scalar pressure imposed on the fluid, specified asproblem input. The quantity PB is the boundary value of the density given by
PB = 2 mjW(IF - FBI,h) . (83)
The factor of 2 appears as in Eq. 74 because of the normalization anomaly at theboundary.
From Eq. 82 we see that any particle i within range of the boundary, i.e.,Iri - FBI< 2h, begins to be influenced by the external pressure. This pressure terInenters in a form similar to that of any neighboring particle except W appears in theboundary term rather than VW. This equation is, therefore, different than if onuassumed that a fixed particle represented the boundary with a pressure defined there.
34
The work term which appears in the energy equation is written
dti +dr I -2- F+ 1, "A B(~- h)- dt 2' + (P6)] - ",PBW(Ir,
-( m g + (k)] VW(I - Fj, h). (84)
After the particles are moved using Eq. 82, the change in kinetic energy is known andEq. 84 is used to obtain the change in internal energy. From the new values of p, andci, the temperatures and pressures can be calculated.
If thermal or equilibrium radiation diffusion is present, the following heat fluxterms must be added to the work terms in the energy equation
dt Work Terms - J iPBW(I - FB ,h)
-m, +.VW(JF - F , h). (85)
The heat fluxes F, are defined at each particle position just as the pressures and tem-peratures are. The fluxes are obtained from the temperatures by
B -K, + - A PW(IF - B , h)
-K, m ( + VW(Ii - ?-jh). (86)P i P
It is not clear how best to implement the diffusion equation in the finite dif-ference approximation. Some numerical experimentation may be necessary. However,the above set of SPH equations allows for boundary conditions on P, f and T andconserves energy, momentum, and mass.
35
SECTION 11
LIST OF REFERENCES
1. Stellingwerf, R. F., "Smooth Particle Hydrodynamics in Three Dimensions,"MRC Report MRC/ABQ-N-384 (1988).
2. Monaghan, J. and Lattanzio, J., "A Refined Particle Method for AstrophysicalProblems," Astron. Astrophys 149, 135 (1985).
3. Monaghan, J., "Why Particle Methods Work," Siam J. Sci. Stat. Comput. 3,422 (1982).
4. Gingold, R. and Monaghan J., "Kernel Estimates as a General Particle Methodin Hydrodynamics," J. Comput. Phys. 46, 429 (1982).
5. Monaghon, J. "SPH Meets the Shocks of Noh," Monash University reprint (1987).
6. Brookshaw, L., "A Method of Calculating Radiative Heat Diffusion in ParticleSimulations," Proc. ASA 6, 207 (1985).
7. Lucy, L., Astron. J. 82, 1013 (1977).
36
DISTRIBUTION LIST
DNA-TR-88-286
DEPARTMENT OF DEFENSE U S ARMY STRATEGIC DEFENSE COMMANDATTN: W DAVIES
DEFENSE ADVANCED RSCH PROJ AGENCYATTN: DR MANSFIELD DEPARTMENT OF THE NAVY
DEFENSE INTELLIGENCE AGENCY NAVAL RESEARCH LABORATORYATTN: DC-6 2 CYS ATTN: H GUNSKYATTN: DIR ATTN: H HECKATHORNATTN: DT-1B ATTN: CODE 4183ATTN: RTS-2B ATTN: S OSSAKOWATTN: VP-TPO ATTN: J DAVIS
ATTN: B RIPINDEFENSE NUCLEAR AGENCY ATTN: P RODRIGUEZ
ATTN: G ULLRICH ATTN: J HUBAATTN: NANFATTN: NASF OFFICE OF NAVAL RESEARCHATTN: OPNA ATTN: A TUCKERATTN: R YOHO
3 CYS ATTN: RAAE DEPARTMENT OF THE AIR FORCEATTN: A MARDIGUIANATTN: K SCHWARTZ AIR FORCE CTR FOR STUDIES & ANALYSISATTN: L SCHROCK ATTN: R GRIFFIN
ATTN: M CRAWFORD ATTN: AFCSA/SASCATTN: P FLEMING AIR FORCE SPACE DIVISIONATTN: RAEE ATTN: YA
ATTN: S BERGGREN 2 CYS ATTN: YN4 CYS ATTN: TITL
DEFENSE NUCLEAR AGENCY AIR FORCE TECHNICAL APPLICATIONS CTR
ATTN: TDNM-CF ATTN: TN
2 CYS ATTN: TDTT W SUMMA AIR FORCE WEAPONS LABORATORY
DEFENSE TECHNICAL INFO CENTER ATTN: NTCA
2 CYS ATTN: DTIC/FDAB ATTN: NTNATTN: SUL
STRATEGIC DEFENSE INITIATIVE ORGN AIR UNIVERSITY LIBRARYATTN: C GIESE ATTN: AUL-LSEATTN: K OBRIENATTN: KE SEC OF RSCH, DEVELOPMENT & ACQATTN: COL GUIBERSON ATTN: AF/ROQl
ATTN: W SEIBERLINGATTN: SLKT SPACE DIVISION/YGATTN: SN ATTN: B BIEHNATTN: COL R ROSS
2 CYS ATTN: B KATZ DEPARTMENT OF ENERGY
DEPARTMENT OF THE ARMY EG&G, INCATTN: D WRIGHT
U S ARMY ATMOSPHERIC SCIENCES LABATTN: DR F NILES LAWRENCE LIVERMORE NATIONAL LABATTN: SLCAS-AE-E ATTN: R HAGER
U S ARMY NUCLEAR & CHEMICAL AGENCY LOS ALAMOS NATIONAL LABORATORYATTN: MONA-NU ATTN: D SAPPENFIELD
ATTN: D SIMONSU S ARMY NUCLEAR EFFECTS LABORATORY ATTN: D WINSKE
ATTN: ATAA-PL ATTN: J WOLCOTTATTN: ATAA-TDC ATTN: J ZINNATTN: ATRC-WCC ATTN: R W WHITAKER
U S ARMY STRATEGIC DEFENSE CMD ATTN: T KUNKLE
ATTN: R BRADSHAW SANDIA NATIONAL LABORATORIESATTN: R SMITH ATTN: D HARTLEYATTN: DASD-H-SAV
Dist-1
DNA-TR-88-286 (DL CONTINUED)
SANDIA NATIONAL LABORATORIES HSS, INCATTN: A D THORNBROUGH ATTN: D HANSEN
ATTN: R BACKSTROMATTN: D DAHLGREN INSTITUTE FOR DEFENSE ANALYSES
ATTN: T P WRIGHT ATTN: E BAUER
ATTN: W D BROWN ATTN: H WOLFHARD
ATTN: SPACE PROJECT DIV J S LEE ASSOCIATES INCATTN: TECH LIB 3141 ATTN: DR J LEE
DEPARTMENT OF DEFENSE CONTRACTORS JAYCOR
ATTN: J SPERLINGAEROSPACE CORP
ATTN: A LIGHTY JOHNS HOPKINS UNIVERSITYATTN: C CREWS ATTN: C MENGATTN: C RICE ATTN: J D PHILLIPSATTN: G LIGHT ATTN: R STOKESATTN: M ROLENZ ATTN: T EVANS
ANALYTICAL SYSTEMS ENGINEERING CORP KAMAN SCIENCES CORPATTN: SECURITY ATTN: E CONRAD
ATLANTIC RESEARCH SERVICES CORP ATTN: G DITTBERNER
ATTN: R MCMILLAN KAMAN SCIENCES CORPORATIONATTN: B GAMBILL
ATMOSPHERIC AND ENVIRONMENTAL RSCH INC ATTN: DAIATTN:M KOATTN: DASAC
ATTN: M KO ATTN: R RUTHERFORD
AUSTIN RESEARCH ASSOCIATES ATTN: W MCNAMARA
ATTN: J THOMPSON KAMAN SCIENCES CORPORATION
AUTOMETRIC INCORPORATED ATTN: DASIAC
ATTN: C LUCAS LOCKHEED MISSILES & SPACE CO, INC
BDM INTERNATIONAL INC ATTN: J HENLEY
ATTN: AVITELLO ATTN: J KUMER
ATTN: L JACOBS ATTN: R SEARS
BERKELEY RSCH ASSOCIATES, INC LOCKHEED MISSILES & SPACE CO, INC
ATTN: C PRETTIE ATTN: D KREJCI
ATTN: J WORKMAN M I T LINCOLN LABATTN: S BRECHT ATTN: D TOWLE
BOEING CO ATTN: I KUPIEC
ATTN: G HALL MARTIN MARIETTA DENVER AEROSPACE
CALIFORNIA RESEARCH & TECH, INC ATTN: H VON STRUVE IMI
ATTN: M ROSENBLATT MAXIM TECHNOLOGIES, INC
CHARLES STARK DRAPER LAB, INC ATTN: J SO
ATTN: A TETEWSKI MCDONNELL DOUGLAS CORP
CORNELL UNIVERSITY ATTN: T CRANOR
ATTN: D FARLEY JR MCDONNELL DOUGLAS CORPORATIONATTN: M KELLY ATTN: J GROSSMAN
EOS TECHNOLOGIES, INC ATTN: R HALPRIN
ATTN: B GABBARD METATECH CORPORATIONATTN: W LELEVIER ATTN: R SCHAEFER
GEO CENTERS, INC ATTN: W RADASKY
ATTN: E MARRAM MISSION RESEARCH CORP
GRUMMAN AEROSPACE CORP ATTN: J KENNEALY
ATTN: J DIGLIO ATTN: R ARMSTRONGATTN: R ARMSTRONGATTN: R LARKIN
Dist-2
DNA-TR-88-286 (DL CONTINUED)
ATTN: S BELANGER R & D ASSOCIATESATTN: WWHITE ATTN: CGREIFINGERATTN: W WHITE ATTN: F GILMORE
ATTN: G HOYTMISSION RESEARCH CORP ATTN: M GANTSWEG
ATTN: B R MILNERATTN: C LAUER R & D ASSOCIATESATTN: D ARCHER ATTN: G GANONGATTN: D KNEPPATTN: D LANDMAN RAND CORPATTN: F FAJEN ATTN: C CRAINATTN: F GUIGLIANO ATTN: E BEDROZIANATTN: G MCCARTOR RAND CORPATTN: K COSNER ATTN: B BENNETTATTN: M FIRESTONEATTN: R BIGONI RJO ENTERPRISES/POET FACATTN: R BOGUSCH ATTN: A ALEXANDERATTN: R DANA ATTN: W BURNSATTN: R HENDRICKATTN: R KILB S-CUBEDATTN: S GUTSCHE ATTN: C NEEDHAMATTN: TECH LIBRARY ATTN: T CARNEY
MISSION RESEARCH CORP SCIENCE APPLICATIONS INTL CORP2 CYS ATTN: P CAMPBELL ATTN: C SMITH
ATTN: D HAMLINMITRE CORPORATION ATTN: D SACHS
ATTN: D RAMPTON ATTN: E STRAKERATTN: M R DRESP ATTN: L LINSON
MITRE CORPORATION SCIENCE APPLICATIONS INTL CORPATTN: J WHEELER ATTN: R LEADABRANDATTN: M HORROCKSATTN: R C PESCI SCIENCE APPLICATIONS INTL CORPATTN: W FOSTER ATTN: J COCKAYNE
NICHOLS RESEARCH CORP INC SCIENCE APPLICATIONS INTL CORPATTN: J EOLL ATTN: D TELAGE
ATTN: M CROSSNORTHWEST RESEARCH ASSOC, INC
ATTN: E FREMOUW SRI INTERNATIONALATTN: D MCDANIEL
PACIFIC-SIERRA RESEARCH CORP ATTN: W CHESNUTATTN: E FIELD JR ATTN: W JAYEATTN: F THOMASATTN: H BRODE STEWART RADIANCE LABORATORY
ATTN: R HUPPIPHOTOMETRICS, INC
ATTN: I L KOFSKY TELECOMMUNICATION SCIENCE ASSOCATTN: RBUCKNER
PHYSICAL RESEARCH INCATTN: W.SHIH TELECOMMUNICATION SCIENCE ASSOC, INC
ATTN: D MIDDLESTEADPHYSICAL RESEARCH INC
ATTN: H FITZ TELEDYNE BROWN ENGINEERINGATTN: P LUNN ATTN: J WOLFSBERGER, JR
PHYSICAL RESEARCH, INC TOYON RESEARCH CORPATTN: R DELIBERIS ATTN: J ISEATTN: T STEPHENS
TRW INCPHYSICAL RESEARCH, INC ATTN: DR D GRYBOS
ATTN: J DEVORE ATTN: R PLEBUCHATTN: J THOMPSON ATTN: H CULVERATTN: W SCHLUETER
Dist-3
DNA-TR.88-286 (DI. CONTINUED)
TRW SPACE & DEFENSE SYSTEMSATTN: D MLAYTON
UTAH STATE UNIVERSITYATTN: K BAKERATTN: L JENSEN
VISIDYNE, INCATTN: J CARPENTER
FOREIGN
FOA 2ATTN: B SJOHOLM
FDA 3ATTN: T KARLSSON
Dist-4