ࡱ> RdO) q@Pictures vPowerPoint Document(?FSummaryInformation(  !"#$%&'()*+,-./0123456789:;<=>@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghijklmnopqrstuvwxyz{|}~      !"#$%&'()*+,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_`abcdefghzۛ-߼_E{wꝦْ GmN2$5[Է6fvtA҆HX:gq`BcŅu-m:'[\0W]-[} b\\07:l/81\0[|c-z--D[6I1hl.J&Ia絕Y͒:vp_sյطq-ǵnZruΖ֭u;m5snͅہh(['Z/ ϼ^sɑר[jvi>I;ۇeoK:RD&edX1Ie?PyLdRVD&eʾ2]+1Ig+Xby!2L.eUc"2dX1IMYc"2oe<&2)={eٕ=/J<(Zgi&2)S..MdR&c"2Jh&VG5Ai Ńj;L<(tJăҪxPZwWUAiu{xPZ-p]<Jexk7;L<(6ŃMxPq_qJ#oo HVAmI<({ă oxPq{?*n{Y<[ăxPq{Ń޻L<( ă;K<(VzăxPq{߼}nxPZU6}e:6k[.Ts}nxPZ]xPq-T+Ń>wxPqF}L*n\z>IϽ/V{soU_/V*nJ/*f/*nl_-T AmJAm?[<Ń>xPq*n|GSxPZٍi&2)~~wn\1r>|ϧxP_JKAiF<|չA_,VWOxPZK^96LL{{vM7<;`"2׆Zq0<ƲLdR氛e0rZ1IY12^-ȤO<&Rƙ2vq&L<c*1261I_e?02ޅ9ώȤ̺Е*~2Giny=M|6Z,H1BZȘ [Q^&RƝe?PZV_nYG)Jui);2ut{ʾwGJXDUtJעN]}LE5MdR٨ȷm"2#{MdR(w1It.Y2u+c"27EUeVDb|s-L2PghkD&eT]22={ap9{0ɀR<`,ȤQg{fǢ];HU6j3NdymY6f Voă鷆J߆[^&2)s}:`!ƇeGM$>m"2wqy$Wm9|]Ė*lk.PZM Y6I"V¡J PiuNxe}LdRfPxe}LU$7V=Azyv.O7r12NPZ9 &|e PfFիŃߗ[Di:XԆEs$ ΑLfXjIxPZ&Rc_Jn(m"2S,sOָwyLUZ1I&<&RXr]| V v[C rvu4IC_'?G<(N&Yu9XosD񠴪?*V~iF9>N<(ɽ&Vsor;ă\xPZee]V0T! MH E_ұ܆abOҭfd(xHT*UÊ ¥RqRo=`5>>Bw1m0y6sȋP u?cp)}0=ϠwЩnn]J twvvE{ 4!sN?EAHo ~NBNxH3h>Dศ0Xɴgŋ \ zY+Cf鏞.IUPZQWm/434)NMj7pVKmWo9Ei' T~QA# FͪBfi7oFⷼxVT xa'RSzn}"UݑLaҷ[ p>WA%f#0a&/ 0SymNݙ:%no߲n`]uq5ູ0~ '"Q7>9 u ѓbOO_5}"Z~Xiu1/w-Odv$ZI4Bw=τuBMPGFNAm>-B냠#r7ѵAҭ$r)d߄c[{:Q Iþb!XS~p0w9^4W!o*k}p2+$IH")fkmpA1EAtȋ1Vo6B]w`rLlZ·,K'q6[rS-FWOnSO ^ fn'IVp# AQ%褹ʽ/QMŅކ1z $j'ȇ b@j3Y "2Sry::HWQn$Z#OF$mWse)%fϱ4_%ba yJL:,3ṼP%RnSfSe]۩ɦi +&,4N惝zɖa3 7mzv&TY](N.˾%O8c瀥QY&Qkt@E_f0zy$uR DȾkA1̀Xh&}VgB60vم標B yus~*ײ<61n؞Tr0L.|5x vF s43®CsydxN6Jat5 lVntL/W(;rkˋ"M'o)K64:{prل,e'{A].d \  +628fB8YZ&t>afp.wh.[Mqaws6NJ&9k?^aUqcEOApSeQg =g@ڋi/_0~~ )4KGåQE ¦TٔmPmymUiJXZL" ¦YUmWmx8Сc én2RK#wUm: y%} .8mt3A^P U`EaZBET8:xk}v&JG@OAD=Ttfjӝsֹwgl0UUj7Y(zoQQ!5 Y%tv73+?դ7pu%zҡP‹2Xx5^Rw:tΛM3jWZE4xSZY =Ω"!VT%F*Tl`NX5gU>3u,$fFe4zБ ׵ ^r]*@MZK跲!2XJX߹u%aRR._-b=&amyTHHXuB-SFIuS+zQZ3Uvfņ^' CPh]qo0.aun`OiޓR~7ɜFjk lT * G:"`8⊄|/氞fa%^SmuJb9†&#{SZR}kIuۿzMº FU3<% #~lxᬬ34 O0zNvfCc`P"䬩 co IX]g} 2ke56]캬5AqO.T.6)O#m/V=NV,'m{di1I k#T`3'mmYjdMYjdmTi^ڎK^YjI[mLº5I۲qYj,Yڦ Y]c-_i[󦄙}"OUo0T0\ ju\KH*WGhZ o"3qbRxZylurw{~\H.<$`R #\h@JnՆhcBT%4-m";HeǶnԂtH]7pI}ߛf9ccuQZ7(Gc!_z_:{Hrj/gJ/4IvvgʙOW?"LIgRlkv3 r w12w [3s_s 1s ؔ8"/ QC#ymg6@3h`"ӏWǿwW4o QyE/~̅x2_=G%\)*W d/ΓCŦvT.H K\]k̹eC>~Ceɷ -c++3g(ۖmq%#yd;q3Ǩg>P 1:jo~S*cnso}Gǜ9~и\\9 4;Ax5̹WxR`57emn[3ڳUE߄No6o նD`]Ud =#R]U%@!,sqq>b׳ , 7 ".ʪ/]D+J]udŴ-Y(h4 ~^594y5>h~͟ۦ7avldak43ߠNۚ>+hAC 4Ɛ2|p!]hg {kVų@\/iȐ6G>*g8_tppܠrsUudnd:\٥ycfx`r6sě] V6 :tl<)䅹Pygnb?cz^/l> x/BZ. ߕ;Y#@JvȉlA@3sF۶w~+ROfs: \slgf ӷ{K8a\+ƬaʝFJgtI|!]9l3c n2b6.,pB$vcnVRֻiO {ZBＧ>)tu&g>>0ph+ R86(P{ BKK<K5 ܢs8 ļݘuԼ}[iFKDK Aw_PiWh*>9IA3 0^P6Ʒ1.zi]i]i]C贮FZҺ$Һ^UHHRxR!K3 i]i] /( a|y]]G+ƿ 9s'^_V(tk;.9|ъ.J<xT[ZFf$N7K{ߟI52k3ye]oc3֯$NZ ql>/AkHaY⋖-'W>7'g=K|*fˬ]s{vKG﬿XYHcr+qr_R'%-KCȗ}U XM=ރ%vur`/Fؕ'9O{lۘ^W ѧGphž0cD1fk쉵}bDlYd=Ƭ=Q~=a~Vz1T-UI@AdIbIɆԜnO0VBv6i;i - .Iz'gT=iJo0eY"mS9ݐf"E>gM9 ]%>O|~u5)~"|K lH,a_jdM%Ff5|TAWC_YSĈ2tLAC3_fHAfv-i#oG-F]ֆhd[FZvC#;NGpZP6_Wh;Oj]?Y?:柒3/_VnľO}|۹oe\sL]1IK}ɟ;=w" OwOh3syAqOl9ۈE3X4Ov|n#N6O.Caf[7)݋=iěi?0~ml@F|o0mHh}8jr^WQw=ML3ucnx oVlS;09]_{Xao_3m Gk-o+_R.tKoCoVyߊ1[qlߊ#ۘ3u%F641%Mc )Vȷ)o 6'f8W~? ZU*W¿ ~7 G>7T%Y%%'No|񡉒 ai²wV<zzzFMЛB_z'n}>zn*Ii8Ii8h3%%%@@B z3Vw@ YC*F/Eמ{cٱ=+*ze[W}5ڿX:ݖe aIӟ36al366>ZXyFn}䏾4~SZNGob5>s}'{m[*{h|t}K|ijVj}Tޑݛ?2>v`:"B}Zy` qbRx{Źg{/; %A墆DPy@#jq_ʬ};Zٷ{@<(fd,V]yxPZ=%V?w{Ai%V ŃfN<(sJ 񠴚^/V_wW;B;n 혒\ zM7cC3kOϣۓ~ZM,9n-9n-9.NҷPNI\2j"i?.W(WG N7o.8.8.8_Jח尘whޱKyǻ.M-g/aఛJ֗尛8Ϸg7tzWл^}QmQ2W658?9s98l;g۹WQR>*|nxǢ9y:rξ_/y ٢}-˹#sk"Y[,7.'%sHp*˵/k¬gז&6nyL$8nݯL$8A]rNXv"=²>&(Hp8lwr؞ȧaO @;ve!K=,raM$8yqj5T^.]͵N8Ʋ;\7qPsn ٽ0ߴ̳_2\Ce8w,sHp,sHp๖yL$8BrwI|e}L$81?L2~2~}wwN5rin%NNvQK6{h:9k@z09䬇Fz$9ǑGz"9䬧Jz9琳Y'N gN:rkYo$gu.9\rE䬋YYY'gu99rr֕+YW&g]Kκu#9Fr䬛Yn#gEκu9rV/ڑ#gDډHr#Y{$gGڏ4rY%g29YG"g2rI䬓YE-rY'gC:u9"rֻY&g]JκqrYW"g}r Y7n"g+9_Y? gu/9^rVg}pgJڕ9kr־}YO#g=u(9Pr֑#YYYYYN#gZr~Vr[YYY g}1rYW"g}Er֍Y g}=rY[Y[Y8k6gmY;ଝp8kDrY?Oyr֡CYG$g'g'gO:u9 r~u.9\r'?!g}ar֧Y"g}yr Yn!gNκ5fCpփBpփL.|&AG#gKڗu09`raY/&gu,9Xr䬓Y&g&r֛Y.$gnreYW$g}9r!8zr-!8rI#'MN4"pҚȎogpts9:kO==GoO&0\ae9dd6&w1HpցmTHp/X1ଃYކ?m"aCYvQ&G= _*_G :²-L$8yms<&B6[1గZ1ണ-Hp˴ui N{EW>ǝXnH;ǝF{ N{9D^oY N{ {9\&ven N#GMm|n Sm"aZ1-DîpYWO[6k,sHp-HpyL$8Z1[\pX̕L vaD wX'ԾpVm {sBDݟsuϰ ׫v!n\Ao }޴/9&;dgzgCza#^t$9(rQc^t 9xp<9to:95|אHw#9\o.9"n9]39]39n 9#8#l+V=C 9Zrl6m&gLζm[Eζm9GUx.KUЉ9ۑlG$gI֟?9 rAlCن$gIv)9ۥlƑM&gL6m&9 l7Bv 9lKv/92rel+V=C 9:rulJζm+9rl-l-lXWe8l``Pg;:;;v*9۩lCنGv9(rQl]N6m 9lW&gMζm9brlO9clVH"9+lm%gJζm9nrl9 l}p>8>8>8Xrl&]Mv59Ml7-$g[HLL9ClO=I,9۳l'g=9&rMlC9GlSPOA%galp888ۈmD6*g lp8ulב!gCv;9lG(9ۣlV#g[GGr?M69ۇlMlBpv!8lNPlglglllllX{e8[c fl?~55=H 9ۓlO=G9r l&g39Nrl}BΆg}u>}&g7gk gFEl"poDl߈#p6<+ ΏGlM8[Sζ4gs{+Vn~1lV&g[Mζm59OTEg C]aϭk᝔wRm AwwCit^f;4:/q[A%n{ISt/X^{e m߫{S)PUT ~_D 6ߋ~WBw~w@ly3ϪO_>3>}|gOE>;}w9鳭 yP*~>n9P*~&T/8ǒ>GZzy\3瞥+~~8٧|j:ǚΟAs<\F:\;e)Mo=S2 ,b)IPCRFb;M\ݝfn5O֞=v7NkL+KuŻflgܚl27*ӿ;_;:}5uaѫ~o髓/zo:O[_s=BuBv|lWi7M~[nU.e=!m=J'§NH+_7bd7tqf1#NKE-qvjxS|xS7{GT~J'#r?b,DmW>|Ka 2C'0aNֻxGz6D'3gf턉cƎq*u{]DoݧNY۫ʱ3j/kydCxOڶ{ZF'X&Q2Akr_u+ջ}cR{*;Gx)4L9?. zLeڵ>8)cG_1qi=S{oc/ӻ6,fKϘ8a) S'^ylet{WbRbn@7L;j]%PNG  IHDRyCgAMABO pHYs.#.#x?v$tEXtSoftwareQuickTime 6.5.1 (Mac OS X)K#tIME 44N+ IDATx}}y=+ivlص.6$>%u+1N+LR'|IUםNJՕVWWZ;-.ǷJhk===|̾z5ygϷh ϝag}Χ8̥˓S|>+_70Czw;;{/\ 7p+&.h&.wJ+k[3<}/ܢp,4Gޜڴ|utZ8'~D }Sp[ RD$F0p+N?.ap3bF}=ɳf.6m2Fp<¥]P5>-}tq~~Pk'ݎ;O&9XN`P/ @ ա gN91 ׻uJ`z}̹X^+zٻK=M D!<޾.]bV}gU7&JkD ]>?X6tB ID0[58VGO|ڙ/?2 "$ ȸFytv~|r]o~V͇ą 8`S-DPs:^S24zgO[Ӵ(2l׵{?tȦ^ +cp 4 eDn \= X4 S};l2b(nrF ;O skt^^/ 73Ξ,cqV,T$ܫ~5wDsND n?,xblMG6jf\6<<2e@Gޜ5 YEa#5АMfg,fD8ɔG%WA ]2ny4n6l# .;hˌB=R_vD c{[Gt=ɖY5ֶn (\ÈN$Kf"~jLKavdtFlᄛ(+@ps0'2zl,{n] Zc 73WTL#0b΃4Wպ@0耰"X X} j xANc}+zjJ04,up[=NFP|Bp}/2cH ?c{w} ZnQ&X΁GV+UҡU+L *`wh,4 R> 04;vߴe[ZB %{,{4|6U $r#\ U ,eCg"d! o|JA$׿`̐΁eλUɲAL|oWZTzE5( 3}#'@d4n2voEM<ָ|Aa՝޷%FumB|.6Ld4YY7<>qq; Uk|n*B-%Zp,,ۛpg[GD A1#{|$5 =X|3|#/s*dȷ`oG^Fv}} >C8n&؁GXm1xlgr?;3] X z=w~țSE3đ `]UC/6^("_z |dtq|WW(q%ks87tpY}s@R8u88h`O|(|!y C9 > g Ng/t00.f5''²Ozblvi9_^oWn+%  8H>L%>q`2V\&)'Qz={ j͂.MT_4|=!^0D0?Aۑ $0 ȒrHbŅa堖bs~{_~?] gL@ۖʁfDWdjA.gWqnuo_Oک4⣋[{*j Ë'r 4Ѭ 8d`;oH[3}&*>PwNO},{ܶ/=ѷoxuo88 R3pD}nL3jϬCӀIцߊ{PXA=Q{b*ڱK5w~dG 9toB M4-Bܳ _z PcϰF t[24a201A@79H'@4t5tfTPnҲhGޜ21:p2[tM0B aFM= <>|f[m/rt%h . Xȕ.XMf P٠s"Iņ>xKuB#F1A";?6Ѹ 2~Ͻ*'n_bP FGzpEЁBkQ8JFE58@hN2'db]Mh, /UXoݼ~F)Sm_Wf¨a1x$hHT.L@UE@n@'fS]=Opg7^ekk ]LkFU!'&Pȴ;\lѼ-|Z=E:78%-":B%,$WOV5:>y7o>7wOrת5ZǵѨ-@J=#3}#lF~o&^0mcaDgWr+lG9"-6蟘]8I؅{94lX˷W6?rߪLQh B}+zJyI顖[Ȱz0LA#!I/TEC9Еn bۘ pšUgWvѯbcV${"J467)Mh8i&vADp@ "r 8.R~չGX}?K뗞mK%:܌H%ݰ|~J& c\ K=w!/m\Œ% ;:+HKQ7RFՙө 9u[7KRq[",7씰jA>tagPc3!#f]u({bP%Ƞic&^Gop aɠVTKÃˍ* SX 2FXm;Չp$G‘U ODh .#9&/w"X?vu[2 ZȈ8d|Q4fo '͍\W鯉lCDiP@iڻ;JG!T^hE jKfљ=#lzguiy%)C|Jº ,u%ӮC4 'dE"J04RdZUu7 ;__D Mk:V 7/B"m]JUϑlۼCOxv6?I.%]o\z:eנP~h5Aڹ3M+Ufڰ5O`T~g!ۼ 7Ф? |/(o J= ̓ט)_|ȷ>{ l`D9Һj>AnuLXD8TK-6o?Uy76|Ә5*u+ Jr|td},^O<~wu$ Ga9>1˲ڈaJXρ!$FM"KN Na7cj ]Gh,b|H=T9j)Ɗn !wl_5L֠$`QG ޒ`Y<{43*07Fİش`&qH AB H&wوO˒8%XK}}3h\ċ=1 DM\N ťWX):(7S"JxԤTigyqD`X[21!=qO6H&(q&6"*V"9|]K(k~E4N-זfobq1d㇒FTXVŘoțժw!ʟlD!—^+%?W N㲦{Q]^dg&͐$i9J?ZE"΂Om!|&%.Q X(A&>t>zF><D|Cڋ(^e~`Y"e:ȀK #6j"ʙW]wLTTQQ֓g#z2`ZPjM/c@I0=*pH ~q&EsH~;Fܜ2Aj?0ƿgw_y4DLt1ї 2ݨBQ0=ґۡ/v@QЀZ+wLEQYJ-?d QGh _$ϯ|b ZJxځ|֏+Udst20JE8w~_ݱU/ P˜C~c]r|9`̊q 9A?RFSeB0I~=v4@q{eǬf2$ "5;\4axlֳ{߽8~<(G]Q倩 AD,Tƒ2@LӔ*6"_ob`13G1/,]u eQFXL3 6?p{41UI04ʓEApL*/^wF5 ulPw 2b" `p޷/G6͹tƁBju`0lV'u!.C9vrUA$ow0+[Pe5 X=X[4THld&GO|XR69/5ؽc]#f.͂zӪ<8'(qќ`ќ.g5mCyޜyxΚT1VF iYqM ɳsz/OzΙS+5<;'&կ|ŗ]l_Hi24#Vb>+1:9> ec=S_aP7+k:c}MB#J0jrPߘis͠^W=O66/hu?(ŵW TPa IhLౕ>vjNc/B|th^"pq(@Z|M? XiXҮhR9vèAJYÃA?h\y(02("Wu R+Ϝ&##u7)' AD-3'F_d)vYAW3>6G%&14쥪AYW'zӪDuxsM(<-XXש#?|©&HHZj; YGQQ(2 uV9@'oΫvdQ3,TADDai0-%ݯۄ_JFEdPQQXuus΁6'PVKH%4TJ4cDiFNiK}1<9ť`-[$’A*%14&Ṟ`4Jzk1aɠ"*%F&"{P\9 'ּs|prٽo}}VEdPQPht;LT9s{NO}|/׽E’A2RQQP_c=h g={1sivӶ/UodJjA"J0\)ɢMѻw}^{ac#[h[X,$ Y"JX}gv)иv<3ǾuteCKY}A'hBAS) HXs&:Ud&! g/T:-5Y;L56|vXcS&MˇdwիE;Œ Ȳ@Jx\ LFs,9>sp,0`Uw2V1w:N:a0 F8؆Ч}A`A [DvdZVQlIyy&E]RZgI2{%7O m-}CKqYp-T,r4{i9X;|Fv?yn:Xiw &6j78xOKfMG Դ  ,;=w7wRveV?\ 0-ֵH;t՞fp99rxl>sY89v,?aYnGO j Loxw^$.܍O> {Ӎ06aZl~:R³{*l)\>2Ty d&0VP91| tǜ$qO zsiaLPVL0^MC|+Ƣ[#,GWwy0xQ wf5^bQ%E\57c 'e%4 dӣ͛L6WE`ji[42\ qD¿~$xW$UxX*@ץ_ў&41%]uLd8R͆%Fmxnqx)F#F9Wl|cbsdasX f!n8K_qiďC` 0~$ ybLY HrGM))N^ɃD ])rb/NPPCUX2XK{߭kؚ&XM*Jd`ǟqm|oT1rL[AZln9qćǾ4,c)HbN _ E>p+LSBL@pcBp<< $`\/:y$QX,ԋҀ ʁzmWrvf"J#Ji#np~oߘyf2WHρ[mH2n岈=CO/GL5Q2Ͻ铗!8a1 aJp1i;ݿ55+8R͌ 56+j=1iRTD 綯3%{R`PĢ\,-Re aW'd)t9/TK`L^&4?(|qʨ#p2.q )I)'Ep~3F[4x㕇Lɓg/4h>5Yj*JJZվڕH+wuEfG U&A TrhO ď\hWd|o\[FX1,RC3vRBdHE)* H"\}Zt>\jrP;KnxNOT` jd K!MIDATQE,(Cid JerR-T@QM%V0 N?YTE"63`>R$|J3yJXi8RK *368W}FSur&>'buA=U1BQH$%%b/}\iiSENMn$%(ӍF LEsBqNW>yl:'\iIjDуTOgܒu4+,R2:M ,9F'g("F @qmYYDIi*RpT@hɺtpɋ3>1/fuH2(JpSDٝP*8A&Dւ1LŘksb۩F(ܷ2_[) JH(.XbʼnmQFlO_!%g&fW!bL XMzZF*I9-T )z:\n315mvHoy%dsBAyBR] r %"Vx"T>yԟ@Y<W@Ã#^De︽(h)!7rckSSKѹ3 @H,qU@R`vK AII'G9UN (|ˎ+R|͗`T:©#'*,^T(1 $M7H@+|%R#s-) %%+s0%<'amAT8TiIrXoZeBv!e8ҢrP @i ;33Qa4ױa&hz1G[Z$]ێBEK~!٥}_KB*OLIq."^hg]Hf6na~16 ET ĚțS)uLM連G@a|Ws*;O\QPN6B$ emwğJY41a״zNoV),R]33@q()8 P(SئƴU4Hul{|lg(&(ƴwIr`m;̨FJ}a2 ѨIj~IzP>cPOQBQ`o>yM)2  2W@uiťq;2Hot LP̆q4rhEQv5(?q,SœD^ZD EQ06w:JI dxAB:Nd2' PSa5@ "Rx1kbLs(;?nEt-jБkE6"w1p>JPC4GUN!P A4E"P^Q>v7 _rJ($Ņa4uću[ ˯D9bȂC#MXkq"qIۑZRʄ^Ǣ#ᰁ+M+jA&.k3*< )I[ "#>o@3Qa.,G84+*V(y*W#VuE,TG<9-`OGO" 3 BEHbƇ^tAhWn*i 1i: ư4(`'_߂fv甀J]IdZrY62% @Y(}ZzA.n DXr#;,cP0աsEgNO/H6a}FL_-v.~FfY aZL9R "$Is'|⊂f@%K3!'i ^k2E>rOL >w Mtq[aZ9~L%nyhJIYf3F%%$"DQesP@ToH2%\{QC<d#JyNBf\#_0TQ +kgͣ dXJ% S721 pπQ驏<5aYEQT&yFI Ph,lJܱꀕ3(Ӕʑ|2hꂀe=zFmm2'OP]gf#3Q 0O88eڮruF.'&&[GԂ1 JS-y0MѿS6K3΍?Jr[h$L5pt'ABNpo43 "NT黼CF΃ h1 #n td3R>ₗ%"Hh|3Yʯ,B"P)ntebIs4LA1J|ܤwD@U(d7 BB3Q�R9: 9m: o|-{ϙK*g"bzI.綯+,5aW|0B; b]"c,ho2H8%rhRz.Vd8 j7'H2"G틇:ն"YUJc7[^j6ÐWtT9(Dֈ qxB2in<@;b~^ZCB`,Qa;d&0 F6ya& n8" Ӣ$ B5}&/& 'ٴ` /iVg%} \T[+9}VFE{p'[~/>^%(R`]wȔ~˟ w_3g 27 ٚ +~+׬ KD3Ƒ^Vlі8ЈpXIȼne_%(8轪7)# >%)x[Q2+B@Y9 :?4ٍ ӌLɰvM3{ޮ,쎞Aڲp;C- ZK}ͮD""$I):PiG3_UB49}Z429|wF;FDHgҥ=]%ht8TmR@04?-70Etxd4^otïC|텍͙O] 3=%@[cJ:O(ksV< %EPxDK O^HR^WE!k$Ic-; B3Q`{aSqXc J0jrrJw?iI& \>?5"7}OFd~oR ׯ D!Nj j_JsW PhgI:^*MGBh"9;Aݍ1hj#bM?!X]ɻ\ D3jY2>0]\}іJL.vߋ鼘op^F9~)Y`oL%!7E r"]q؞{!4R,N)CK.4I@*@ r #w~7AWs> K#FWq'ՆT^BC@o j5cѻ?}AHZuGckr-kwL;AlPVM5R~-;'^0AHFsPn8OC_t8Ȟᄇ؁M8ގA p^:1Mt?B#@n!\Ed I7Q3]4 DUQ0"I6IH4aӐd{rو]2;4%j Afq,#fd:K,Q: g*J a5L1Bi ԣ5#/l}Շz,t[M/ . *9aʁ`"\] @HȞT&f^v%Ƣt 3ʁND8r@hi 1 큁?} dЦmg*:u, }·8 pB(YS<(E$}බߗ >G}5"8\KEǰDAD$-qգ4I(bNrh7P_%%cb~.!7hr?NS?E$04lx`yuDel'iζ(^#F( >Ӹ$s%30tl))a-͌ )^֌4oLMlfҴb_oQ xOs 2H8"S&p6 4߸h%H L3*+t5yI\RPmax0(9LXƖ1hW5G |) uEaq R,3\eTpA^+UTpHD2J W:d"3x WIENDB`0T1Aʨy?F ٲd-[&>k$Q aMPxuSAkAnM6ۢ95*{7&&7.4%49Y'sk {z=J㛸Ԛovơ7W`S] ,10OP[NldGl4Knj?pF:Zo~Hm--pO^c[Vd?$`N?5kAd٭V݅B@ q8j<gj>E ׈C|sŴ>1`a_Lk>n5 d, cc/ǝ}O,Yy5zH#hT:5`~'صu<)<>;'BVl=o~[T-w.pj_Z+aڼ_+?o,0jR_3JE9r}3쏙/6R6gD Dĩ$$AGbq('ay=YgH*%˳.oy}'K\=O YAQߐ~x `MPrxڵTkA3$&znPЂE tsU$dO{-(C ⭇-R]/7ID}{A  (!B!GIO]߆b(0 Cf YPQGs <dy{8 JbpяSJ  7u,o4>tݟ򼁂*f]n6Jq>7p=̜e[GM$g\ Ѧh:"#qZ/@|5fhqp#X+weA_JP.<ұM"*K _7]HXLnZmqۻ?ߟիD" J6oMk(uk-Y-=4}MѤdA2kL VV(ӝԢsTZ}:s ';"UIv %K> 0 BaDA(l:1w)<L@BuU&c( _ CNcMF\i-nEnܝU!=nWzì]DF'<y։z '@61(  J Equation Equation.30$Microsoft Equation #Equation Equation.DSMT40*MathType 5.0 Equationz/ 0|DTimesngsRoma0 I07@\PDArialngsRoma0 I07@\P DComic Sans MS0 I07@\P0DTimes New Roman0 I07@\P@DWingdingsRoman0 I07@\PcI9. Z2@  @@``  @n?" dd@  @@`` :2(Dj        B$bOҭf)  B$WGhZ o0) b$MFSWP?H7: $ $ $ $ $ B$d-[&>k$Qr  wawa     A@  A1 8c8c     ?1 d0u0@Ty2 NP'p<'pA)BCD|E||S" ,ffffff<@ ,8  g4vdvdnP`ppp@ uʚ;2Nʚ;<4!d!d uʚ;<4dddd uʚ;<4BdBd uʚ;(h___PPT2001D<4X___PPTMac11@f   hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography ? %CVerification and Validation Using Code-Based Sensitivity Techniques(The LACSI Code-Based Sensitivity Project(% The Problem  |Development of computer models of physical phenomena requires a  reality check. If the computer model output matches the real data, then all is well If the model output does not match, then ??? Underlying mathematics/physics is inadequate to capture the salient features of the phenomena under study. VALIDATION problem. Program implementing the model is incorrect. VERIFICATION problem Together, referred to as V&V (alternative spelling VnV) 8   k 7   $$((,,004488<<&V&V Techniques Validation Obtain  best fit with (some portion of) the data Evaluate if  best is good enough Evaluate tuned simulation on other data, see if fit is adequate Best fits may be obtained with Newton s Method Verification For now, this means testing. If simulation model is a differential equation, then Method of Manufactured Solutions (MMS) Roundoff error estimation: run-time error bounds (Wilkinson)V      'A Common SubproblemNewton s method compute derivatives MMS compute derivatives Runtime error bounds compute derivatives Some additional uses of derivatives (optimal) design Taylor series techniques for DEs Newton-Krylov iterations HKK  ,0GqResearch OverviewPurpose : Efficiently,accurately, and with minimal human intervention compute sensitivities of computer codes.  Sensitivity means ( calculus ) derivative Accomplish this through Automatic Differentiation (AD)" (State of the Art in AD Fortran 77 Adifor 3.0 First to use compiler technology 1995 Wilkinson prize to Adifor 2 Price/Distribution model fits well TAF (TAMC), TAPENADE Fortran 90 Adifor 3.0, TAF, TAPENADE all support a little C,C++ ADIC, ADOL-C, FADBAD e / e  /Roadmap CAD research Software Engineering and Construction Sample Validation) AD ResearchAreas of inquiry Memory usage for adjoint methods Improved derivative methods for simple assignment statements Techniques for advanced language features Array slices, structured data Pointers, dynamic memory allocation Operator overloading Multiple data representations Association-by-name or Association-by-address Activity Analysis for advanced programming languages AD for MPI programsvW.IW  .I $Software Engineering and DevelopmentoDevelop a framework for multi-language AD. Component Model: separating language knowledge from differentiation. Leverage other work (Open64) Develop a Fortran 90 AD tool (with Unit Tests) 80 Unit tests, all pass Ubiksolve, a component of the Truchas system Good Truchas Surrogate (Brian Lally) Linear solvers, so differentiation is easy to check 50% run and verifiedb+b/Fn+b/F  no AD and V & V at Los AlamosRudy Henninger: Mesa 1D, 2D anti-armor codes Caravana lagrangian test code (hydro methods from FLAG code) Truchas 1d (metal casting code) Ralph Nelson: TRAC (reactor safety code)N{{  2Detonation Shock Dynamics (DSD) Curvature Equation2  DSD - better fit of 6 parameters  Laboratory InteractionsVisit summer 2001 -- Gave a talk Visit summer 2002 -- visit w Rudy to work on explosion code SIAM Session on Validation of Metal Flow Simulations, talk on verification of DEs LACSI Symposium session on Verification and Validation talk on verification of DEs$Runtime Error AnalysisDoug Kothe indicated at the LACSI Priorities and Strategies meeting in 2004 that the V&V groups at LANL are concerned about roundoff error Can get estimated linearized forward error analysis by computing/Psx,, ` e(HH(dh  *` ̙33` ` ff3333f` 333MMM` f` f` 3>?" dU@ ,? " Z2@% d  ,  << fMMM n?" dZ(@   @@``PX      ` p>> *" ( mu    `S wawa1 ?O%@   T Click to edit Master title style! !@  Z`W wawa1 ?u    RClick to edit Master text styles Second level Third level Fourth level Fifth level!     S^  6 ,1?0X^  6 ,1?hp   c :A ,LACSI_logo"B  s *޽h ? lb  Blank * aY `(  ` ` <e 8c ?"` i   _#Click to edit Master subtitle style$#r  ` <1?"@ xX ` B0k 8c ?"   T Click to edit Master title style! !p ` c :A ,LACSI_logo"l ` 6 ,1?"Qyl ` 6 ,1?"hB ` s *޽h ? lb80___PPT10.p_B SK( 1giGo )  Z@ Swawa1 ? @   ;Body Text Second Level Third Level Fourth Level Fifth Level     <  Z`SUU1?ga  ~ Page *Z  ###55FFp  01 ?    B  s *޽h ? a(80___PPT10.pZȬ   .(     ZаSUU1?ga  l Page *Z ###55FFB  s *޽h ? a(80___PPT10.p+ _W@d(  dr d S `2`   r d S ? `` i    d < <8c?"0 ` N80P___PPT100(v___PPT9XPR___PPTMac11,$   hnamd` Arial&Monotype Typography  oMike Fagan Rice University Dept. of Computational and Applied Mathematics http://lacsi.rice.edu/review/slides/Rpn ?$  <$PH d 0޽h ? lb80___PPT10.pg + P*(  r  S  O%@   x  c $    H  0޽h ? lb + `*(  r  S  DO%@   x  c $     H  0޽h ? lb + 0$(  r  S ; O%@   r  S <    H  0޽h ? lbV ++ YQp (  l  C X O%@   x  c $Y    2  H^ 8c?p@  RFinite Differences Hand Coding  <Pc 8c?   < 2 R   s *o? Y    Zi wawa 1?"0 < PH@___PPT9" W3 Ways to Compute Derivatives: Finite Differences Hand Coding Automatic Differentiation@H0Z29I0 Z29 @`H  0޽h ?   lb___PPT10e+D=' ϐ= @B + + $(  r  S  O%@   r  S ` _    H  0޽h ? lb + >(  r  S p3 O%@     c $: #    H  0޽h ? lb + $(  r  S e O%@   r  S g     H  0޽h ? lb + $(  r  S  O%@   r  S      H  0޽h ? lb + $(  r  S 5 O%@   r  S p6 ~    H  0޽h ? lb +  (  r  S      `  c $A ??  dB  <D3o  dB  <D3o0 dB  <D3o @@P dB  <D3o P dB  <D3o@@dB  <D3o  <Ћ  z  >   <    ['How could one tune these 6 parameters??((H  0޽h ? lb + '(  r  S ~ O%@     0 0P  {- SNL DAKOTA package drives the optimization process - Gradients provided by AD of DSD solver - ~40 passes improves the fit||^  6A1? t gH  0޽h ? lbu + %(  r  S  O%@   x  c $        f0 wawa 1?" `   F I0 Z2   Z wawa 1?"0 G  Visit summer 2003 -- gave a talk Methods conference 2004 Truchas conference 2004, talk on Adifor90 6 Registered Adifor users at LANL T-10,T-11,EES-5,XMH, CCS-2,X52I0 Z2&H0Z2H  0޽h ? lb + QI(  r  S X O%@   x  c $0Z E    `  c $A  ??`   G  Z_ wawa 1?" @  QFor each statement executed in the program. Approximate dx by machine epsilon * x RH0Z2R8H  0޽h ? lbJ  0(     H1 ?      # lpSwawa1 ? @    H  0޽h ? a(]xXMGڳcoBas- cLxg7ӳDXrB(A*'˹pGB"@,!B.(H$$nqg{zg Dݯ{_zzz?ˡOSv?Gd;H{q\y}.|}~t/F#z+њFҚ&/duo/HrHZ߳! e;L$$ d<(jhyS~ϐ4dɣ?Mܓ3a?  @I|12p8 OO_ | `ׁo|PJw2p  X p\*s'x<0LSj3"q8 Β/GEd4\9qYkQS.U;0C}DJ%ǭ=3}3/I\rjH7Pu$*'!m}?Y_C&bF'ߗ:6C=&B5•%p ;.J=Xcq@{䏰jKtf)ds {bs`0k^HyjcQSKke];GvXy ~>E^%KJ:/p^I˻z+=_Cv}T#IVզϊ#W.Ne2_Ȼ=zbS&qIef]4$Hef4QQH*~_ѱXozM0{(Z?-{n-6gq+ER(exHlf[mm$K}HA/{~^|tND?+){Si)jjmhOϮ̼ \lzv/x2ZM#u_#Rb~&Z\eE]ǛݝWv^7zvonz/GWnTX_HgQugRYY+l3p3Z,Z=ݎW 3ڜjՖd] V3fʅvP_ #v+k[ 걽cX0й&כQu -хNQo!\!w[Av 2Ýfmv .-k 3ǔN4G<@h0{6K(8:纱A˲\sx¶ɚAݿz]6Y$7h-\ v:jh(O/Agk}#62 ;-GOe+)nY!v  To5 LBh :D/TAz}aR쟸' vOSQC)HZF*Zka nLjX(gJ16)%dבd2'3<|_)&gjjf^V/:+KJ9݌KM^kv k#;ٰw{5_Ȟko=8#>/哯xWoG3a BH C0BzHH=-b;{"H R+!C%!G 潙YBPy3of~ͼvry k硁 *ԍ1ł\+j.Eϐ[-_y϶-ѻ_,!81F:R~CS7`urFi Y?A׭(qoOAi~F\?'@_pZכ'[/R{k tE!3 G!  <ӕ@1JFȟ w oPK~ T8ޢט}!wԱvvm챉t1ωHi}"|eW`8_b+End#Y+syNh)e8.R["ܟG;!-1cBAD̢|,t)4+Z+ȥ'\ 1 nIQ*"l.7!ʝSnP&~g:Ϧs0wydz<+aPqL\`fHʛ/1x&1?v ~DkYpx,Higߑ77nh~J[ۛ!*Q$d ^>0V&|ߤIf-f-jb'p?1S9th73*N̒Ù\rddo9֝v9075eit'38),TO㗫m6{g[r6t26y{S(cLT7-F`{TO/2q$< ~j | ?\ >J^Gp~!Q ya~}[{aa1BER__|Fl\bz9Nn'׮kԟ dȝ"8Wv^W E(/yPm H?Яz?z[Z"sZ`j90dxp^RЀ3ÿ lHbP  @AL G@;b `B&VB>1Op D9(D@Ms`e9X+\ ^h#{RmJLNMV?Z4*03(  J Equation Equation.30$Microsoft Equation #DocumentSummaryInformation8Current User# Oh+'0HPh t 'Add title hereej d52@P(/@M~@z]PZ~@c}L ՜.+,0   'On-screen ShowoFk$ TimesArialComic Sans MSTimes New Roman WingdingsBlankDVerification and Validation Using Code-Based Sensitivity Techniques The Problem V&V TechniquesA Common SubproblemResearch OverviewState of the Art in ADRoadmap AD Research%Software Engineering and DevelopmentAD and V & V at Los Alamos3Detonation Shock Dynamics (DSD) Curvature Equation!DSD - better fit of 6 parametersLabor Equation Equation.DSMT40*MathType 5.0 Equationz/ 0|DTimesngsRoma I7@\PDArialngsRoma I7@\P DComic Sans MS I7@\P0DTimes New Roman I7@\P@DWingdingsRoman I7@\PcI9. Z2@  @@``  @n?" dd@  @@`` @Q          B$bOҭf)  B$WGhZ o0) b$MFSWP?H7: $ $ $ $ $ B$d-[&>k$Qr  wawa     A@  A1 8c8c     ?1 d0u0@Ty2 NP'p<'pA)BCD|E||S" ,ffffff<@ ,8  g4vdvdn`ppp@ uʚ;2Nʚ;<4!d!d0 uʚ;<4dddd0 uʚ;<4BdBd0 uʚ;(h___PPT2001D<4X___PPTMac11@f   hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography ? %CVerification and Validation Using Code-Based Sensitivity Techniques(The LACSI Code-Based Sensitivity Project(% The Problem  xDevelopment of computer models of physical phenomena requires a  reality check. If the computer model output matches the real data, then all is well If the model output does not match, then ??? Underlying mathematics/physics is inadequate to capture the salient features of the phenomena under study. VALIDATION problem. Program implementing the model is incorrect. VERIFICATION problem Together, referred to as V&V (alternative spelling VnV) 8   k 7   $$((,,004488<<&V&V Techniques Validation Obtain  best fit with (some portion of) the data Evaluate if  best is good enough Evaluate tuned simulation on other data, see if fit is adequate Best fits may be obtained with Newton s Method Verification For now, this means testing. If simulation model is a differential equation, then Method of Manufactured Solutions (MMS) Roundoff error estimation: run-time error bounds (Wilkinson)V      'A Common SubproblemNewton s method compute derivatives MMS compute derivatives Runtime error bounds compute derivatives Some additional uses of derivatives (optimal) design Taylor series techniques for DEs Newton-Krylov iterations HKK  ,0GqResearch OverviewPurpose : Efficiently,accurately, and with minimal human intervention compute sensitivities of computer codes.  Sensitivity means ( calculus ) derivative Accomplish this through Automatic Differentiation (AD)" (State of the Art in AD Fortran 77 Adifor 3.0 First to use compiler technology 1995 Wilkinson prize to Adifor 2 Price/Distribution model fits well TAF (TAMC), TAPENADE Fortran 90 Adifor 3.0, TAF, TAPENADE all support a little C,C++ ADIC, ADOL-C, FADBAD e / e  /Roadmap CAD research Software Engineering and Construction Sample Validation) AD ResearchAreas of inquiry Memory usage for adjoint methods Improved derivative methods for simple assignment statements Techniques for advanced language features Array slices, structured data Pointers, dynamic memory allocation Operator overloading Multiple data representations Association-by-name or Association-by-address Activity Analysis for advanced programming languages AD for MPI programsvW.IW  .I $Software Engineering and DevelopmentoDevelop a framework for multi-language AD. Component Model: separating language knowledge from differentiation. Leverage other work (Open64) Develop a Fortran 90 AD tool (with Unit Tests) 80 Unit tests, all pass Ubiksolve, a component of the Truchas system Good Truchas Surrogate (Brian Lally) Linear solvers, so differentiation is easy to check 50% run and verifiedb+b/Fn+b/F  no AD and V & V at Los AlamosRudy Henninger: Mesa 1D, 2D anti-armor codes Caravana lagrangian test code (hydro methods from FLAG code) Truchas 1d (metal casting code) Ralph Nelson: TRAC (reactor safety code)N{{  2Detonation Shock Dynamics (DSD) Curvature Equation2  DSD - better fit of 6 parameters  Laboratory InteractionsVisit summer 2001 -- Gave a talk Visit summer 2002 -- visit w Rudy to work on explosion code SIAM Session on Validation of Metal Flow Simulations, talk on verification of DEs LACSI Symposium session on Verification and Validation talk on verification of DEs$Runtime Error AnalysisDoug Kothe indicated at the LACSI Priorities and Strategies meeting in 2004 that the V&V groups at LANL are concerned about roundoff error Can get estimated linearized forward error analysis by computing/+,-./012 3 4 5 6 7PPsx,, ` e(HH(dh T+ _W@d(  dr d S `2`   r d S ? `` i    d < <8c?"0 ` N80P___PPT100(v___PPT9XPR___PPTMac11,$   hnamd` Arial&Monotype Typography  oMike Fagan Rice University Dept. of Computational and Applied Mathematics http://lacsi.rice.edu/review/slides/Rpn ?$  <$PH d 0޽h ? lb___PPT10u.pg+D=' n= @B + + P*(  r  S  O%@   x  c $    H  0޽h ? lb___PPT10e+D=' ̐= @B + + `*(  r  S  DO%@   x  c $     H  0޽h ? lb + 0$(  r  S ; O%@   r  S <    H  0޽h ? lb + $(  r  S  O%@   r  S ` _    H  0޽h ? lb + >(  r  S p3 O%@     c $: #    H  0޽h ? lb + $(  r  S e O%@   r  S g     H  0޽h ? lb  + $(  r  S  O%@   r  S      H  0޽h ? lb  + $(  r  S 5 O%@   r  S p6 ~    H  0޽h ? lb  +  (  r  S      `  c $A ??  dB  <D3o  dB  <D3o0 dB  <D3o @@P dB  <D3o P dB  <D3o@@dB  <D3o  <Ћ  z  >   <    ['How could one tune these 6 parameters??((H  0޽h ? lb  + '(  r  S ~ O%@     0 0P  {- SNL DAKOTA package drives the optimization process - Gradients provided by AD of DSD solver - ~40 passes improves the fit||^  6A1? t gH  0޽h ? lbu  + %(  r  S  O%@   x  c $        f0 wawa 1?" `   F I0 Z2   Z wawa 1?"0 G  Visit summer 2003 -- gave a talk Methods conference 2004 Truchas conference 2004, talk on Adifor90 6 Registered Adifor users at LANL T-10,T-11,EES-5,XMH, CCS-2,X52I0 Z2&H0Z2H  0޽h ? lb + QI(  r  S X O%@   x  c $0Z E    `  c $A  ??`   G  Z_ wawa 1?" @  QFor each statement executed in the program. Approximate dx by machine epsilon * x RH0Z2R8H  0޽h ? lb tl@( KX~X R  3     r  # aY @    H  0޽h ? a(80___PPT10.bqթH tlP( hnY`[Y R  3     r  # PoY @    H  0޽h ? a(80___PPT10.bqթH tl`( nY`[Y R  3     r  # |Y @    H  0޽h ? a(80___PPT10.bqթH tlp( ?@YuY R  3     r  # *V @    H  0޽h ? a(80___PPT10.bqթH tl ( Y iYuY  R   3     r   # 0X @    H   0޽h ? a(80___PPT10.bqթH tl$(  iViV $R $ 3     r $ # `V @    H $ 0޽h ? a(80___PPT10.bqթH tl(( {5v (R ( 3     r ( # V @    H ( 0޽h ? a(80___PPT10.bqթH  tl,( ?? ,R , 3     r , # P W @    H , 0޽h ? a(80___PPT10.bqթH  tl0( VV 0R 0 3     r 0 # иV @    H 0 0޽h ? a(80___PPT10.bqթH tl4(  X`V 4R 4 3     r 4 # KV @    H 4 0޽h ? a(80___PPT10.bqթH tl8( {Vv 8R 8 3     r 8 # 5V @    H 8 0޽h ? a(80___PPT10.bqթH tl<(  <R < 3     r < # 0< @    H < 0޽h ? a(80___PPT10.bqթH tl@( ??sr @R @ 3     r @ # 0V @    H @ 0޽h ? a(80___PPT10.bqթHr܀0aD u3$`Fͼ+YUQMIEA=951-)%7ٰ2(  J Equation Equation.30$Microsoft Equation #Equation Equation.DSMT40*MathType 5.0 Equationz/ 0|DTimesngsRoma I7@\PDArialngsRoma I7@\P DComic Sans MS I7@\P0DTimes New Roman I7@\P@DWingdingsRoman I7@\PcI9. Z2@  @@``  @n?" dd@  @@`` @Q          B$bOҭf)  B$WGhZ o0) b$MFSWP?H7: $ $ $ $ $ B$d-[&>k$Qr  wawa     A@  A1 8c8c     ?1 d0u0@Ty2 NP'p<'pA)BCD|E||S" ,ffffff<@ ,8  g4vdvdn`ppp@ uʚ;2Nʚ;<4!d!d0 uʚ;<4dddd0 uʚ;<4BdBd0 uʚ;(h___PPT2001D<4X___PPTMac11@f   hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography ? %CVerification and Validation Using Code-Based Sensitivity Techniques(The LACSI Code-Based Sensitivity Project(% The Problem  xDevelopment of computer models of physical phenomena requires a  reality check. If the computer model output matches the real data, then all is well If the model output does not match, then ??? Underlying mathematics/physics is inadequate to capture the salient features of the phenomena under study. VALIDATION problem. Program implementing the model is incorrect. VERIFICATION problem Together, referred to as V&V (alternative spelling VnV) 8   k 7   $$((,,004488<<&V&V Techniques Validation Obtain  best fit with (some portion of) the data Evaluate if  best is good enough Evaluate tuned simulation on other data, see if fit is adequate Best fits may be obtained with Newton s Method Verification For now, this means testing. If simulation model is a differential equation, then Method of Manufactured Solutions (MMS) Roundoff error estimation: run-time error bounds (Wilkinson)V      'A Common SubproblemNewton s method compute derivatives MMS compute derivatives Runtime error bounds compute derivatives Some additional uses of derivatives (optimal) design Taylor series techniques for DEs Newton-Krylov iterations HKK  0Research OverviewPurpose : Efficiently,accurately, and with minimal human intervention compute sensitivities of computer codes.  Sensitivity means ( calculus ) derivative Accomplish this through Automatic Differentiation (AD)" (State of the Art in AD Fortran 77 Adifor 3.0 First to use compiler technology 1995 Wilkinson prize to Adifor 2 Price/Distribution model fits well TAF (TAMC), TAPENADE Fortran 90 Adifor 3.0, TAF, TAPENADE all support a little C,C++ ADIC, ADOL-C, FADBAD e / e  /Roadmap CAD research Software Engineering and Construction Sample Validation) AD ResearchAreas of inquiry Memory usage for adjoint methods Improved derivative methods for simple assignment statements Techniques for advanced language features Array slices, structured data Pointers, dynamic memory allocation Operator overloading Multiple data representations Association-by-name or Association-by-address Activity Analysis for advanced programming languages AD for MPI programsvW.IW  .I $Software Engineering and DevelopmentoDevelop a framework for multi-language AD. Component Model: separating language knowledge from differentiation. Leverage other work (Open64) Develop a Fortran 90 AD tool (with Unit Tests) 80 Unit tests, all pass Ubiksolve, a component of the Truchas system Good Truchas Surrogate (Brian Lally) Linear solvers, so differentiation is easy to check 50% run and verifiedb+b/Fn+b/F  no AD and V & V at Los AlamosRudy Henninger: Mesa 1D, 2D anti-armor codes Caravana lagrangian test code (hydro methods from FLAG code) Truchas 1d (metal casting code) Ralph Nelson: TRAC (reactor safety code)N{{  2Detonation Shock Dynamics (DSD) Curvature Equation2  DSD - better fit of 6 parameters  Laboratory InteractionsVisit summer 2001 -- Gave a talk Visit summer 2002 -- visit w Rudy to work on explosion code SIAM Session on Validation of Metal Flow Simulations, talk on verification of DEs LACSI Symposium session on Verification and Validation talk on verification of DEs$Runtime Error AnalysisDoug Kothe indicated at the LACSI Priorities and Strategies meeting in 2004 that the V&V groups at LANL are concerned about roundoff error Can get estimated linearized forward error analysis by computing/+,-./012 3 4 5 6 7PPsx,, ` e(HH(dh  + 0$(  r  S ; O%@   r  S <    H  0޽h ? lbr'(*7ٰ3(  J Equation Equation.30$Microsoft Equation #Equation Equation.DSMT40*MathType 5.0 Equationz/ 0|DTimesngsRoma I7@\PDArialngsRoma I7@\P DComic Sans MS I7@\P0DTimes New Roman I7@\P@DWingdingsRoman I7@\PcI9. Z2@  @@``  @n?" dd@  @@`` @Q          B$bOҭf)  B$WGhZ o0) b$MFSWP?H7: $ $ $ $ $ B$d-[&>k$Qr  wawa     A@  A1 8c8c     ?1 d0u0@Ty2 NP'p<'pA)BCD|E||S" ,ffffff<@ ,8  g4vdvdn`ppp@ uʚ;2Nʚ;<4!d!d0 uʚ;<4dddd0 uʚ;<4BdBd0 uʚ;(h___PPT2001D<4X___PPTMac11@f   hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography ? %CVerification and Validation Using Code-Based Sensitivity Techniques(The LACSI Code-Based Sensitivity Project(% The Problem  xDevelopment of computer models of physical phenomena requires a  reality check. If the computer model output matches the real data, then all is well If the model output does not match, then ??? Underlying mathematics/physics is inadequate to capture the salient features of the phenomena under study. VALIDATION problem. Program implementing the model is incorrect. VERIFICATION problem Together, referred to as V&V (alternative spelling VnV) 8   k 7   $$((,,004488<<&V&V Techniques" Validation Obtain  best fit with (some portion of) the data Evaluate if  best is good enough Evaluate tuned simulation on other data, see if fit is adequate Best fits may be obtained with Newton s Method Verification For now, this means testing. If simulation model is a differential equation, then Method of Manufactured Solutions (MMS) Roundoff error estimation: run-time error bounds (Wilkinson)V      'A Common Subproblem Newton s method compute derivatives MMS compute derivatives Runtime error bounds compute derivatives Some additional uses of derivatives (optimal) design Taylor series techniques for DEs Newton-Krylov iterations HKK  0Research OverviewPurpose : Efficiently,accurately, and with minimal human intervention compute sensitivities of computer codes.  Sensitivity means ( calculus ) derivative Accomplish this through Automatic Differentiation (AD)" (State of the Art in AD Fortran 77 Adifor 3.0 First to use compiler technology 1995 Wilkinson prize to Adifor 2 Price/Distribution model fits well TAF (TAMC), TAPENADE Fortran 90 Adifor 3.0, TAF, TAPENADE all support a little C,C++ ADIC, ADOL-C, FADBAD e / e  /Roadmap CAD research Software Engineering and Construction Sample Validation) AD ResearchAreas of inquiry Memory usage for adjoint methods Improved derivative methods for simple assignment statements Techniques for advanced language features Array slices, structured data Pointers, dynamic memory allocation Operator overloading Multiple data representations Association-by-name or Association-by-address Activity Analysis for advanced programming languages AD for MPI programsvW.IW  .I $Software Engineering and DevelopmentoDevelop a framework for multi-language AD. Component Model: separating language knowledge from differentiation. Leverage other work (Open64) Develop a Fortran 90 AD tool (with Unit Tests) 80 Unit tests, all pass Ubiksolve, a component of the Truchas system Good Truchas Surrogate (Brian Lally) Linear solvers, so differentiation is easy to check 50% run and verifiedb+b/Fn+b/F  no AD and V & V at Los AlamosRudy Henninger: Mesa 1D, 2D anti-armor codes Caravana lagrangian test code (hydro methods from FLAG code) Truchas 1d (metal casting code) Ralph Nelson: TRAC (reactor safety code)N{{  2Detonation Shock Dynamics (DSD) Curvature Equation2  DSD - better fit of 6 parameters  Laboratory InteractionsVisit summer 2001 -- Gave a talk Visit summer 2002 -- visit w Rudy to work on explosion code SIAM Session on Validation of Metal Flow Simulations, talk on verification of DEs LACSI Symposium session on Verification and Validation talk on verification of DEs$Runtime Error AnalysisDoug Kothe indicated at the LACSI Priorities and Strategies meeting in 2004 that the V&V groups at LANL are concerned about roundoff error Can get estimated linearized forward error analysis by computing/+,-./012 3 4 5 6 7PPsx,, ` e(HH(dh  + `*(  r  S  O%@   x  c $     H  0޽h ? lb + 0$(  r  S ; O%@   r  S <    H  0޽h ? lbr*& ^_*a7ٰ2(  J Equation Equation.30$Microsoft Equation #Equation Equation.DSMT40*MathType 5.0 Equationz/ 0|DTimesngsRomaatory InteractionsRuntime Error Analysis  Fonts UsedDesign TemplateEmbedded OLE Servers Slide TitlesSlide Titles_[Fj drsoft Rice University I7@\PDArialngsRoma I7@\P DComic Sans MS I7@\P0DTimes New Roman I7@\P@DWingdingsRoman I7@\PcI9. Z2@  @@``  @n?" dd@  @@`` @Q          B$bOҭf)  B$WGhZ o0) b$MFSWP?H7: $ $ $ $ $ B$d-[&>k$Qr  wawa     A@  A1 8c8c     ?1 d0u0@Ty2 NP'p<'pA)BCD|E||S" ,ffffff<@ ,8  g4vdvdn`ppp@ uʚ;2Nʚ;<4!d!d0 uʚ;<4dddd0 uʚ;<4BdBd0 uʚ;(h___PPT2001D<4X___PPTMac11@f   hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography ? %CVerification and Validation Using Code-Based Sensitivity Techniques(The LACSI Code-Based Sensitivity Project(% The Problem  xDevelopment of computer models of physical phenomena requires a  reality check. If the computer model output matches the real data, then all is well If the model output does not match, then ??? Underlying mathematics/physics is inadequate to capture the salient features of the phenomena under study. VALIDATION problem. Program implementing the model is incorrect. VERIFICATION problem Together, referred to as V&V (alternative spelling VnV) 8   k 7   $$((,,004488<<&V&V Techniques" Validation Obtain  best fit with (some portion of) the data Evaluate if  best is good enough Evaluate tuned simulation on other data, see if fit is adequate Best fits may be obtained with Newton s Method Verification For now, this means testing. If simulation model is a differential equation, then Method of Manufactured Solutions (MMS) Roundoff error estimation: run-time error bounds (Wilkinson)V      'A Common Subproblem Newton s method compute derivatives MMS compute derivatives Runtime error bounds compute derivatives Some additional uses of derivatives (optimal) design Taylor series techniques for DEs Newton-Krylov iterations HKK  0Research OverviewPurpose : Efficiently,accurately, and with minimal human intervention compute sensitivities of computer codes.  Sensitivity means ( calculus ) derivative Accomplish this through Automatic Differentiation (AD)" (State of the Art in AD Fortran 77 Adifor 3.0 First to use compiler technology 1995 Wilkinson prize to Adifor 2 Price/Distribution model fits well TAF (TAMC), TAPENADE Fortran 90 Adifor 3.0, TAF, TAPENADE all support a little C,C++ ADIC, ADOL-C, FADBAD e / e  /RoadmapCAD research Software Engineering and Construction Sample Validation) AD ResearchAreas of inquiry Memory usage for adjoint methods Improved derivative methods for simple assignment statements Techniques for advanced language features Array slices, structured data Pointers, dynamic memory allocation Operator overloading Multiple data representations Association-by-name or Association-by-address Activity Analysis for advanced programming languages AD for MPI programsvW.IW  .I $Software Engineering and DevelopmentoDevelop a framework for multi-language AD. Component Model: separating language knowledge from differentiation. Leverage other work (Open64) Develop a Fortran 90 AD tool (with Unit Tests) 80 Unit tests, all pass Ubiksolve, a component of the Truchas system Good Truchas Surrogate (Brian Lally) Linear solvers, so differentiation is easy to check 50% run and verifiedb+b/Fn+b/F  no AD and V & V at Los AlamosRudy Henninger: Mesa 1D, 2D anti-armor codes Caravana lagrangian test code (hydro methods from FLAG code) Truchas 1d (metal casting code) Ralph Nelson: TRAC (reactor safety code)N{{  2Detonation Shock Dynamics (DSD) Curvature Equation2  DSD - better fit of 6 parameters  Laboratory InteractionsVisit summer 2001 -- Gave a talk Visit summer 2002 -- visit w Rudy to work on explosion code SIAM Session on Validation of Metal Flow Simulations, talk on verification of DEs LACSI Symposium session on Verification and Validation talk on verification of DEs$Runtime Error AnalysisDoug Kothe indicated at the LACSI Priorities and Strategies meeting in 2004 that the V&V groups at LANL are concerned about roundoff error Can get estimated linearized forward error analysis by computing/+,-./012 3 4 5 6 7PPsx,, ` e(HH(dh V ++ YQp (  l  C X O%@   x  c $Y    2  H^ 8c? P  RFinite Differences Hand Coding  <Pc 8c?   < 2 R   s *o? py    Zi wawa 1?"0 < PH@___PPT9" W3 Ways to Compute Derivatives: Finite Differences Hand Coding Automatic Differentiation@H0Z29I0 Z29 @`H  0޽h ?   lb___PPT10e+D=' n= @B + + >(  r  S p3 O%@     c $: #    H  0޽h ? lbr#brax7ٰ2(  J Equation Equation.30$Microsoft Equation #Equation Equation.DSMT40*MathType 5.0 Equationz/ 0|DTimesngsRoma I7@\PDArialngsRoma I7@\P DComic Sans MS I7@\P0DTimes New Roman I7@\P@DWingdingsRoman I7@\PcI9. Z2@  @@``  @n?" dd@  @@`` @Q          B$bOҭf)  B$WGhZ o0) b$MFSWP?H7: $ $ $ $ $ B$d-[&>k$Qr  wawa     A@  A1 8c8c     ?1 d0u0@Ty2 NP'p<'pA)BCD|E||S" ,ffffff<@ ,8  g4vdvdn`ppp@ uʚ;2Nʚ;<4!d!d0 uʚ;<4dddd0 uʚ;<4BdBd0 uʚ;(h___PPT2001D<4X___PPTMac11@f   hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography ? %CVerification and Validation Using Code-Based Sensitivity Techniques(The LACSI Code-Based Sensitivity Project(% The Problem  xDevelopment of computer models of physical phenomena requires a  reality check. If the computer model output matches the real data, then all is well If the model output does not match, then ??? Underlying mathematics/physics is inadequate to capture the salient features of the phenomena under study. VALIDATION problem. Program implementing the model is incorrect. VERIFICATION problem Together, referred to as V&V (alternative spelling VnV) 8   k 7   $$((,,004488<<&V&V Techniques" Validation Obtain  best fit with (some portion of) the data Evaluate if  best is good enough Evaluate tuned simulation on other data, see if fit is adequate Best fits may be obtained with Newton s Method Verification For now, this means testing. If simulation model is a differential equation, then Method of Manufactured Solutions (MMS) Roundoff error estimation: run-time error bounds (Wilkinson)V      'A Common Subproblem Newton s method compute derivatives MMS compute derivatives Runtime error bounds compute derivatives Some additional uses of derivatives (optimal) design Taylor series techniques for DEs Newton-Krylov iterations HKK  0Research OverviewPurpose : Efficiently,accurately, and with minimal human intervention compute sensitivities of computer codes.  Sensitivity means ( calculus ) derivative Accomplish this through Automatic Differentiation (AD)" (State of the Art in AD$ Fortran 77 Adifor 3.0 First to use compiler technology 1995 Wilkinson prize to Adifor 2 Price/Distribution model fits well TAF (TAMC), TAPENADE Fortran 90 Adifor 3.0, TAF, TAPENADE all support a little C,C++ ADIC, ADOL-C, FADBAD e / e  /RoadmapCAD research Software Engineering and Construction Sample Validation) AD ResearchAreas of inquiry Memory usage for adjoint methods Improved derivative methods for simple assignment statements Techniques for advanced language features Array slices, structured data Pointers, dynamic memory allocation Operator overloading Multiple data representations Association-by-name or Association-by-address Activity Analysis for advanced programming languages AD for MPI programsvW.IW  .I $Software Engineering and DevelopmentoDevelop a framework for multi-language AD. Component Model: separating language knowledge from differentiation. Leverage other work (Open64) Develop a Fortran 90 AD tool (with Unit Tests) 80 Unit tests, all pass Ubiksolve, a component of the Truchas system Good Truchas Surrogate (Brian Lally) Linear solvers, so differentiation is easy to check 50% run and verifiedb+b/Fn+b/F  no AD and V & V at Los AlamosRudy Henninger: Mesa 1D, 2D anti-armor codes Caravana lagrangian test code (hydro methods from FLAG code) Truchas 1d (metal casting code) Ralph Nelson: TRAC (reactor safety code)N{{  2Detonation Shock Dynamics (DSD) Curvature Equation2  DSD - better fit of 6 parameters  Laboratory InteractionsVisit summer 2001 -- Gave a talk Visit summer 2002 -- visit w Rudy to work on explosion code SIAM Session on Validation of Metal Flow Simulations, talk on verification of DEs LACSI Symposium session on Verification and Validation talk on verification of DEs$Runtime Error AnalysisDoug Kothe indicated at the LACSI Priorities and Strategies meeting in 2004 that the V&V groups at LANL are concerned about roundoff error Can get estimated linearized forward error analysis by computing/+,-./012 3 4 5 6 7PPsx,, ` e(HH(dh V ++ YQp (  l  C X O%@   x  c $Y    2  H^ 8c? P  RFinite Differences Hand Coding  <Pc 8c?   < 2 R   s *o? py    Zi wawa 1?"0  PH@___PPT9" W3 Ways to Compute Derivatives: Finite Differences Hand Coding Automatic Differentiation@H0Z29I0 Z29 @`H  0޽h ?   lb___PPT10e+D=' n= @B +y + $(  r  S  O%@   r  S ` _    H  0޽h ? lb___PPT10e+D=' ̐= @B +r(7ٰ2(  J Equation Equation.30$Microsoft Equation #Equation Equation.DSMT40*MathType 5.0 Equationz/ 0|DTimesngsRoma I7@\PDArialngsRoma I7@\P DComic Sans MS I7@\P0DTimes New Roman I7@\P@DWingdingsRoman I7@\PcI9. Z2@  @@``  @n?" dd@  @@`` @Q          B$bOҭf)  B$WGhZ o0) b$MFSWP?H7: $ $ $ $ $ B$d-[&>k$Qr  wawa     A@  A1 8c8c     ?1 d0u0@Ty2 NP'p<'pA)BCD|E||S" ,ffffff<@ ,8  g4vdvdn`ppp@ uʚ;2Nʚ;<4!d!d0 uʚ;<4dddd0 uʚ;<4BdBd0 uʚ;(h___PPT2001D<4X___PPTMac11@f   hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography ? %CVerification and Validation Using Code-Based Sensitivity Techniques(The LACSI Code-Based Sensitivity Project(% The Problem  xDevelopment of computer models of physical phenomena requires a  reality check. If the computer model output matches the real data, then all is well If the model output does not match, then ??? Underlying mathematics/physics is inadequate to capture the salient features of the phenomena under study. VALIDATION problem. Program implementing the model is incorrect. VERIFICATION problem Together, referred to as V&V (alternative spelling VnV) 8   k 7   $$((,,004488<<&V&V Techniques" Validation Obtain  best fit with (some portion of) the data Evaluate if  best is good enough Evaluate tuned simulation on other data, see if fit is adequate Best fits may be obtained with Newton s Method Verification For now, this means testing. If simulation model is a differential equation, then Method of Manufactured Solutions (MMS) Roundoff error estimation: run-time error bounds (Wilkinson)V      'A Common Subproblem Newton s method compute derivatives MMS compute derivatives Runtime error bounds compute derivatives Some additional uses of derivatives (optimal) design Taylor series techniques for DEs Newton-Krylov iterations HKK  0Research OverviewPurpose : Efficiently,accurately, and with minimal human intervention compute sensitivities of computer codes.  Sensitivity means ( calculus ) derivative Accomplish this through Automatic Differentiation (AD)" (State of the Art in AD$ Fortran 77 Adifor 3.0 First to use compiler technology 1995 Wilkinson prize to Adifor 2 Price/Distribution model fits well TAF (TAMC), TAPENADE Fortran 90 Adifor 3.0, TAF, TAPENADE all support a little C,C++ ADIC, ADOL-C, FADBAD e / e  /RoadmapCAD research Software Engineering and Construction Sample Validation) AD ResearchAreas of inquiry Memory usage for adjoint methods Improved derivative methods for simple assignment statements Techniques for advanced language features Array slices, structured data Pointers, dynamic memory allocation Operator overloading Multiple data representations Association-by-name or Association-by-address Activity Analysis for advanced programming languages AD for MPI programsvW.IW  .I $Software Engineering and DevelopmentoDevelop a framework for multi-language AD. Component Model: separating language knowledge from differentiation. Leverage other work (Open64) Develop a Fortran 90 AD tool (with Unit Tests) 80 Unit tests, all pass Ubiksolve, a component of the Truchas system Good Truchas Surrogate (Brian Lally) Linear solvers, so differentiation is easy to check 50% run and verifiedb+b/Fn+b/F  no AD and V & V at Los AlamosRudy Henninger: Mesa 1D, 2D anti-armor codes Caravana lagrangian test code (hydro methods from FLAG code) Truchas 1d (metal casting code) Ralph Nelson: TRAC (reactor safety code)N{{  2Detonation Shock Dynamics (DSD) Curvature Equation2  DSD - better fit of 6 parameters  Laboratory InteractionsVisit summer 2001 -- Gave a talk Visit summer 2002 -- visit w Rudy to work on explosion code SIAM Session on Validation of Metal Flow Simulations, talk on verification of DEs LACSI Symposium session on Verification and Validation talk on verification of DEs$Runtime Error AnalysisDoug Kothe indicated at the LACSI Priorities and Strategies meeting in 2004 that the V&V groups at LANL are concerned about roundoff error Can get estimated linearized forward error analysis by computing/+,-./012 3 4 5 6 7PPsx,, ` e(HH(dh r 7ٰ2(  J  Equation Equation.30$Microsoft Equation  #Equation Equation.DSMT40*MathType 5.0 Equationz/ 0|DTimesngsRoman tttDArialngsRoman ttt DComic Sans MSn ttt0DTimes New Roman ttt@DWingdingsRoman tttcI9. Z2@  @@``  @n?" dd@  @@`` HQ         B$bOҭf)  B$WGhZ o0) b$MFSWP?H7: $ $ $ $ $ B$d-[&>k$Qr  wawa     A@  A1 8c8c     ?1 d0u0@Ty2 NP'p<'pA)BCD|E||S" ,ffffff<@ ,8  g4AdAd4t ppp@ uʚ;2Nʚ;<4!d!dP  uʚ;<4ddddP  uʚ;<4BdBdP  uʚ;(___PPTMac11@f   hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography    hnamd` Arial&Monotype Typography h___PPT2001D<4X`d? %CVerification and Validation Using Code-Based Sensitivity Techniques(The LACSI Code-Based Sensitivity Project(% The Problem  xDevelopment of computer models of physical phenomena requires a  reality check. If the computer model output matches the real data, then all is well If the model output does not match, then ??? Underlying mathematics/physics is inadequate to capture the salient features of the phenomena under study. VALIDATION problem. Program implementing the model is incorrect. VERIFICATION problem Together, referred to as V&V (alternative spelling VnV) 8 k 7  &V&V Techniques" Validation Obtain  best fit with (some portion of) the data Evaluate if  best is good enough Evaluate tuned simulation on other data, see if fit is adequate Best fits may be obtained with Newton s Method Verification For now, this means testing. If simulation model is a differential equation, then Method of Manufactured Solutions (MMS) Roundoff error estimation: run-time error bounds (Wilkinson)T    'A Common Subproblem Newton s method compute derivatives MMS compute derivatives Runtime error bounds compute derivatives Some additional uses of derivatives (optimal) design Taylor series techniques for DEs Newton-Krylov iterations HKK  0Research OverviewPurpose : Efficiently,accurately, and with minimal human intervention compute sensitivities of computer codes.  Sensitivity means ( calculus ) derivative Accomplish this through Automatic Differentiation (AD)" (State of the Art in AD$ Fortran 77 Adifor 3.0 First to use compiler technology 1995 Wilkinson prize to Adifor 2 Price/Distribution model fits well TAF (TAMC), TAPENADE Fortran 90 Adifor 3.0, TAF, TAPENADE all support a little C,C++ ADIC, ADOL-C, FADBAD e / e /      RoadmapCAD research Software Engineering and Construction Sample Validation) AD ResearchAreas of inquiry Memory usage for adjoint methods Improved derivative methods for simple assignment statements Techniques for advanced language features Array slices, structured data Pointers, dynamic memory allocation Operator overloading Multiple data representations Association-by-name or Association-by-address Activity Analysis for advanced programming languages AD for MPI programsvW.IW.I $Software Engineering and DevelopmentoDevelop a framework for multi-language AD. Component Model: separating language knowledge from differentiation. Leverage other work (Open64) Develop a Fortran 90 AD tool (with Unit Tests) 80 Unit tests, all pass Ubiksolve, a component of the Truchas system Good Truchas Surrogate (Brian Lally) Linear solvers, so differentiation is easy to check 50% run and verified`+b/Fn+b/Fn  o AD and V & V at Los AlamosRudy Henninger: Mesa 1D, 2D anti-armor codes Caravana lagrangian test code (hydro methods from FLAG code) Truchas 1d (metal casting code) Ralph Nelson: TRAC (reactor safety code)L{{2Detonation Shock Dynamics (DSD) Curvature Equation2  DSD - better fit of 6 parameters  Laboratory InteractionsVisit summer 2001 -- Gave a talk Visit summer 2002 -- visit w Rudy to work on explosion code SIAM Session on Validation of Metal Flow Simulations, talk on verification of DEs LACSI Symposium session on Verification and Validation talk on verification of DEs$Runtime Error AnalysisDoug Kothe indicated at the LACSI Priorities and Strategies meeting in 2004 that the V&V groups at LANL are concerned about roundoff error Can get estimated linearized forward error analysis by computing/+,-./012 3 4 5 6 7Psx,, ` e(HH(dh K VN@d(  dr d S `` д r d S `` i д  d <8c?"0 ` N$P___PPT100(R___PPTMac11,$   hnamd` Arial&Monotype Typography b___PPT9D< Mike Fagan Rice University Dept. of Computational and Applied Mathematics http://lacsi.rice.edu/review/slides/Fagan_LACSI_review.ppt Fn ?<  <$R+H d 0޽h ? lb___PPT10u.pg+D=' n= @B +r ?  CF7