A linear domain decomposition method for partially saturated flow in porous media
|
|
- Tiffany Carr
- 6 years ago
- Views:
Transcription
1 UHasset Computatona Mathematcs Preprnt Seres A near doman decomposton method for partay saturated fow n porous meda Davd Seus, Koondanbha Mtra, Iuu Sorn Pop, Forn Adran Radu and Chrstan Rohde UHasset Computatona Mathematcs Preprnt Nr. UP August 10th, 2017
2 A near doman decomposton method for partay saturated fow n porous meda Davd Seus a,, Koondanbha Mtra b,c, Iuu Sorn Pop c,d, Forn Adran Radu d, Chrstan Rohde a a Insttute of Apped Anayss and Numerca Smuaton, Char of Apped Mathematcs, Pfaffenwadrng 57, Stuttgart, Germany b Department of Mathematcs and Computer Scence, Technsche Unverstet Endhoven, PO Box 513, 5600 MB Endhoven, The Netherands c Facuty of Scences, Hasset Unversty, Campus Depenbeek, Agoraaan Budng D, BE3590 Depenbeek, Begum d Department of Mathematcs, Unversty of Bergen, P. O. Box 7800, N-5020 Bergen, Norway Abstract The Rchards equaton s a nonnear paraboc equaton that s commony used for modeng saturated/unsaturated fow n porous meda. We assume that the medum occupes a bounded Lpschtz doman parttoned nto two dsjont subdomans separated by a fxed nterface Γ. Ths eads to two probems defned on the subdomans whch are couped through condtons expressng fux and pressure contnuty at Γ. After an Euer mpct dscretsaton of the resutng nonnear subprobems a near teratve L-type) doman decomposton scheme s proposed. The convergence of the scheme s proved rgorousy. In the ast part we present numerca resuts that are n ne wth the theoretca fndng, n partcuar the uncondtona convergence of the scheme. We further compare the scheme to other approaches not makng use of a doman decomposton. Namey, we compare to a Newton and a Pcard scheme. We show that the proposed scheme s more stabe than the Newton scheme whe remanng comparabe n computatona tme, even f no paraesaton s beng adopted. Fnay we present a parametrc study that can be used to optmze the proposed scheme. Keywords: Doman decomposton, L-scheme Lnearsaton, Rchards Equaton 1. Introducton Unsaturated fow processes through porous meda appear n a varety of physca stuatons and appcatons. Notabe exampes are so remedaton, enhanced o recovery, CO 2 storage, harvestng of geotherma energy, or the desgn of fters and fue ces. Mathematca modeng and numerca smuaton are essenta for understandng such processes, snce measurements and experments are very dffcut Correspondng author: Ema address: davd.seus@ans.un-stuttgart.de Davd Seus) August 10, 2017
3 f not mpossbe, and hence ony mtedy avaabe. The assocated mathematca and computatona chaenges are manfod. The mathematca modes are usuay couped systems of nonnear parta dfferenta equatons and ordnary ones, nvovng argey varyng physca propertes and parameters, ke porosty, permeabty or so composton. Together wth the arge scae and possbe compexty of the doman, ths poses sgnfcant computatona chaenges, makng the desgn and anayss of robust dscretsaton methods a non-trva task. In ths work we focus on saturated/unsaturated fow of one fud water) n a porous medum e.g. the subsurface) occupyng the doman Ω R d d {1,2,3}). Besdes water, a second phase ar) s present, whch s assumed to be at a constant atmospherc) pressure. Ths stuaton s descrbed by the Rchards equaton, here n pressure formuaton Φ t Sp) [ K µ k ) r Sp) p + z )] = 0, 1) see e.g. [1], orgnay [2, 3]. In the above Φ denotes the porosty, S s the water saturaton, p s the water pressure, k r s the reatve permeabty, K the ntrnsc permeabty and z = ρ w gx 3 s the gravtatona term n drecton of the x 3 -axs. Fnay, g s the gravtatona acceeraton, ρ w the water densty and µ ts vscosty. Wth T > 0 beng a maxma tme, the equaton s defned for the tme t 0,T ) on the bounded Lpschtz doman Ω. Beow we propose a doman decomposton DD) scheme for the numerca souton of 1). To ths am we assume that Ω s parttoned nto two subdomans Ω {1,2}) separated by a Lpschtz-contnuous nterface Γ, see Fg. 1. In other words one has Ω = Ω 1 Ω 2 Γ. The restrcton to two subdomans s made for the ease of presentaton, but the scheme can be extended straghtforwardy to more subdomans. In each Ω {1,2}) we use the physca pressure p as prmary varabe. Furthermore, the permeabty and porosty n each of the subdomans may be dfferent and even dscontnuous, whch s the case of a heterogeneous medum consstng of bock-type heterogenety ke a fractured medum). In vew of ts reevance for manfod appcatons n rea fe, Rchards equaton has been studed extensvey, both anaytcay and numercay, and the dedcated terature s extremey rch. We restrct ourseves here by mentonng [4, 5] for the exstence of weak soutons and [6] for the unqueness. Numerca schemes for the Rchards equaton, or n genera for degenerate paraboc equatons, are anaysed n [7, 8, 9, 10, 11, 12, 13, 14, 15]. Most of the papers are consderng the backward Euer method for the tme dscretsaton n vew of the ow reguarty of the souton, see [4], and to avod restrctons on the tme step sze. Dfferent approaches wth regard to spata dscretsaton have been consdered. Gaerkn fnte eements were used n [8, 16, 17]. Dscontnuous Gaerkn fnte eement schemes for fows through 2
4 heterogeneous) porous meda have been studed n [18, 19]. Fnte voume schemes ncudng mutpont fux approxmaton ones for the Rchards equaton are anaysed n [20, 21, 13], and mxed fnte eements n [7, 22, 10, 11, 12, 15, 14]. Such schemes are ocay mass conservatve. Appyng the Krchhoff transformaton [4] brngs the mathematca mode to a form that smpfes mathematca and numerca anayss, see e.g. [8, 7, 10, 11]. However, the transformed unknown s not drecty reated to a physca quantty ke the pressure, and therefore a postprocessng step s requred after a numerca approxmaton of the souton has been obtaned. Aternatvey, one may deveop numerca schemes for the orgna formuaton and n terms of the physca quanttes. Nevertheess, when provng the convergence rgorousy, one often resorts to a Krchhoff transformed formuaton as ntermedate step. Aternatvey, suffcent reguarty of the souton, e.g. by avodng cases where the medum s competey saturated, or competey dry, has to be assumed. We pont out that n ths work we w not make use of the Krchhoff transformaton, keepng the equaton n a more reevant form for appcatons. If mpct methods are adopted for the tme dscretsaton, the eptc or fuy dscrete) probems obtaned at each tme step are nonnear. For sovng these, dfferent approaches have been proposed. Exampes are the Newton method [23, 24, 25], the Pcard/modfed Pcard method [26, 27], or the Jäger-Kacur method [28, 29]. We refer to [30] for the convergence anayss of such nonnear schemes. Assumng that the nta guess s the souton from the prevous tme step, the convergence of such schemes can ony be guaranteed under severe restrcton for the tme step n terms of the mesh sze. Addtonay, reguarzng the probem s requred, whch prevents the Jacoban from becomng snguar. Such dffcutes do not appear when the L-scheme s beng used, whch s a fxed pont scheme transformng the teraton nto a contracton, [31, 32, 16]. The convergence s merey near but n a better norm H 1 ) and requres no reguarzaton or severe constrant on the tme step. We aso refer to [33] for a combnaton of the Newton method and the L-scheme. Moreover, we menton [12] for the appcaton of the L-scheme to Höder nstead of Lpschtz contnuous nonneartes. Independent of the chosen dscretsaton method and of the nearsaton scheme, doman decomposton DD) methods offer an effcent way to reduce the computatona compexty of the probem, and to perform cacuatons n parae. Ths s n partcuar nterestng whenever domans wth bock type heterogenetes are consdered, as DD schemes aow decoupng the modes defned n dfferent homogeneous subdomans and sovng these numercay n parae. We refer to [34] for a detaed dscusson of near DD methods and to [35] for a genera ntroducton nto the subject. Comprehensve studes of nonnear DD schemes n the fed of fud dynamcs can be found n [36, 37, 38]. For artces strcty reated to porous meda fow modes, we refer to [39, 40] for an overvew of dfferent overappng doman decomposton strateges. Lnear and nonnear addtve Schwartz methods are compared, and the use 3
5 of such methods as near and nonnear precondtoners s dscussed. Regardess of the type of the DD scheme, choosng the optma parameters s a key ssue. Such aspects are anaysed e.g. n [41, 42]. We aso refer to [43] for a DD agorthm for porous meda fow modes, where a-posteror estmates are used to optmze the parameters and the number of teratons. Reca that the Rchards equaton s a nonnear evouton equaton. For sovng ths type of equaton, methods ke pararea [44] and wave-form reaxaton [45, 46] have been proposed. The man deas there are to decompose the probem nto separate probems defned n tme/space-tme domans. DD methods for the Rchards equaton are dscussed n [47, 48]. In these papers the doman s decomposed nto mutpe ayers and the Rchards modes restrcted to adjacent ayers are couped by Robn type boundary condtons. The approach uses nonoverappng doman-decomposton and generases the deas of the method ntroduced n [49] for near eptc probems see aso [50, 51]), eadng to decouped, nonnear probems n the subdomans. Here we consder a near DD scheme for the numerca approxmaton of the tme dscrete probems obtaned after substructurng nto subprobems and performng an Euer mpct tme steppng. A nonoverappng DD scheme referred to henceforth as LDD scheme) nspred by the DD method ntroduced n [49] s defned. The LDD teratons are near, based on an L-type scheme. Ths approach dffers from the one commony used when deang wth nonnear eptc probems n the context of DD. In most cases, the DD teratons ead to nonnear subprobems. For sovng these, teratve methods n each subdoman are apped. In our approach, the nearsaton step s part of the DD teratons, whch reduces the computatona tme. More precsey, the L-scheme dea s combned wth the nonoverappng DD scheme such that the equatons defned n each subdoman aong wth the Robn type coupng condtons on the nterface become near. For the resutng scheme we prove rgorousy the uncondtona convergence, and provde numerca exampes supportng the theoretca fndngs and demonstratng ts effectveness. The paper s structured as foows. In Sec. 2 we present the mathematca mode and ntroduce the DD scheme. Secton 3 contans the anayss of the scheme. Fnay, Sec. 4 provdes numerca experments n two spata dmensons, together wth an anayss of the practca performance of the scheme. Ths ncudes a comprehensve comparson ncudng robustness and effcency) between the proposed DD scheme and standard monothc schemes based on Newton, modfed Pcard as we as the L-scheme. 2. Probem formuaton and teratve scheme 2.1. Probem formuaton Reca that T > 0 and Ω R d s a bounded Lpschtz doman parttoned n two subdomans Ω 1,2, separated by the Lpschtz-contnuous nterface Γ. The boundary of Ω s denoted by Ω and the portons 4
6 Fgure 1: Iustraton of the doman Ω = Ω 1 Ω 2 R d wth fxed nterface Γ. Aso shown are the norma vectors aong the nterface. of Ω that are aso boundares of Ω are denoted by Ω see aso Fg. 1). To ease the presentaton, the two subdomans are assumed to be homogeneous and sotropc,.e. we can have two dfferent reatve permeabtes k r = k r, on each Ω, the ntrnsc permeabtes K = K are scaar and the two porostes Φ = 1,2) are constant. The product K k r, Φ µ n 1) s abbrevated by k henceforth. We sove equaton 1) n Ω together wth nta and homogeneous Drchet boundary condtons. We refer to [47, 52] for more genera condtons, ncudng outfow-type ones. On the two subdomans, the probem transforms nto two subprobems, couped through two condtons at the nterface Γ: the contnuty of the norma fuxes and the contnuty of the pressures. Wth the fuxes F := k S p ) ) p + z ), 1) becomes t S p ) + F = 0 n Ω 0,T ], 2) F 1 n 1 = F 2 n 2 on Γ [0,T ], 3) p 1 = p 2 on Γ 0,T ], 4) p = 0 on Ω 0,T ]. 5) Ths s cosed by the nta condtons p,0) := p,0 n Ω, where p s the water pressure on Ω, = 1,2, and k are gven) scaed reatve permeabty functons, that are assumed to be smooth enough. In the above, n stands for the outer unt norma vector at Ω. 5
7 Sem-dscrete formuaton dscretsaton n tme) For the tme dscretsaton we et N N be a gven and τ := N T be the correspondng tme step. Then p n s the approxmaton of the pressure p at tme t n = nτ. The Euer mpct dscretsaton of 2) 5) reads ) ) S p n S p n 1 + τ F n = 0 n Ω, 6) F n 1 n 1 = F n 2 n 2 on Γ, 7) p n 1 = pn 2 on Γ, 8) p n = 0 on Ω, 9) where F n := k S p n )) p n + z) s the fux at tme step t n. Observe that 7) and 8) are the coupng condtons at the nterface Γ The LDD teratve scheme If p n 1 1, p n 1 ) 2 s known, p n 1, p n ) 2 can be obtaned by sovng the nonnear system 6) 9). To ths end, we defne an teratve scheme that uses Robn type condtons at Γ to decoupe the subprobems n Ω, and nearses the terms due to the saturaton-pressure dependency by addng stabsaton terms that cance each other n the mt see e.g. [33, 31]). Specfcay, assumng that for some N the approxmatons p n, 1 and g 1 are known, we seek p n, 1, ) pn, 2 sovng the probems L p n, L p n, 1 + τ F n, = S p n, 1 Foowng the prevousy ntroduced notaton, F n, ) ) + S p n 1 n Ω, 10) F n, n = g + λ pn, on Γ [0,T ], 11) g := 2λ pn, 1 3 g ) := k S p n, 1 ) ) p n, + z ) denotes the nearsed fux at teraton. By λ 0, ), we denote a free to be chosen) parameter used to weght the nfuence of the pressure on the nterface condtons at Γ. The parameters L > 0 must adhere to some md constrants n order for the scheme to converge, whch w be dscussed ater, but other than that, are arbtrary. The teraton starts wth p n,0 := p n 1, and g 0 := F n 1 n λ p n 1, and ceary, the dfference L p n, L p n, 1 s vanshng n case of convergence. Remark 1. The usage of the terms g and of the parameter λ s motvated by the foowng. Wth the notaton f n := F n n, the transmsson condtons 7)-8) become f1 n = f 2 n and pn 1 = pn 2. For any λ 0, 6
8 these are equvaent to f1 n = f 2 n λ pn 2 ) + λ pn 1, 13) f2 n = f 1 n λ pn 1 ) + λ pn 2. Denotng the terms between brackets by g, one obtans f n 1 = g 1 + λ p n 1, f n 2 = g 2 + λ p n 2, and g 1 = 2λ p n 2 g 2, g 2 = 2λ p n 1 g 1. 14) The condtons n 11)-12) are the nearsed counterparts of 14). Remark 2 dfferent decoupng formuatons). as convex combnatons of the terms g and p, namey The decouped condtons n 7)-8) can be formuated F n, n = 1 λ)g + λ pn, 11 ) 1 λ)g := 2λ pn, λ)g ) The convergence anayss beow can be carred out for ths formuaton wthout any dffcuty. However, the DD scheme usng ths convex formuaton showed a sower convergence n the numerca experments than when 11)-12) was used. Moreover, t s easer to fnd cose to optma parameters for the atter. Such aspects are dscussed n Secton 4. In vew of ths, n what foows we restrct the anayss to the nta formuaton. Before formuatng the man resut we specfy the notaton that w be used beow. Notaton 1. L 2 Ω) s the space of Lebesgue measurabe, square ntegrabe functons over Ω. H 1 Ω) contans functons n L 2 Ω) havng aso weak dervatves n L 2 Ω). H 1 0 Ω) = C 0 Ω)H1, where the competon s wth respect to the standard H 1 norm and C0 Ω) s the space of smooth functons wth compact support n Ω. The defnton for H 1 Ω ) = 1,2) s smar. Wth Γ beng a d 1) dmensona manfod n Ω, H 1 2 Γ) contans the traces of H 1 functons on Γ see e.g. [53, 54, 34]. Gven u H 1 Ω), by ts trace on Γ s denoted by u Γ. Furthermore, the foowng spaces w be used V := { u H 1 Ω ) u Ω 0 }, 15) V := { u 1,u 2 ) V 1 V 2 u 1 Γ u 2 Γ }, 16) H 1/2 00 Γ) = { ν H 1/2 Γ) ν = w Γ for a w H 1 0 Ω) }. 17) Note, that V = H0 1 Ω). H1/2 00 Γ) denotes the dua space of H 1/2 00 Γ)., X w denote the L 2 X) scaar product, wth X beng one of the sets Ω, Ω = 1,2) or Γ. Whenever sef understood, the notaton of the 7
9 doman of ntegraton X w be dropped. Furthermore, stands aso for the duaty parng between, Γ H 1/2 00 Γ) and H 1/2 00 Γ). In what foows we make the foowng Assumptons 1. Wth = 1,2, we assume that a) k : R [0,1] are strcty monotoncay ncreasng and Lpschtz contnuous functons wth Lpschtz constants L k > 0, b) there exsts m R such that 0 < m k 1 S), k 2 S) for a S R, c) S : R R are monotoncay ncreasng and Lpschtz contnuous functons wth Lpschtz constants L S > 0. For ater use we defne L k := max{ L k1,l k2 } and L S := max{l S1,L S2 }. In a smpfed formuaton, the man resut n ths paper s Theorem 1. Assume there exsts a souton par p n 1, pn 2 ) to 6) 9) that addtonay fufs sup p n + z ) L M <. Let L obey L S < 2L for = 1,2 and assume that the tme step τ > 0 s chosen sma enough, so that for both one has τ < 2m 1 Lk 2 M 2 1 ). 18) L S 2L Then the sequence of souton pars { p n, 1, pn, 2 )} 1 of 10) 11) converges to pn 1, pn 2 ). Remark 3. The precse form of Theorem 1 w be formuated n Secton 3, after havng defned a weak souton. The theorem above s gven for the ease of presentaton. 3. Anayss of the scheme. Ths secton gves the convergence proof for the proposed scheme. The startng pont s the Euer mpct dscretsaton n Secton 2. Assumng p1 n 1, p n 1 ) 2 V to be known, a weak formuaton of 6) 9) s gven by Probem 1 Sem-dscrete weak formuaton). Fnd p n 1, pn 2 ) V such that Fn n H 1/2 00 Γ) for = 1,2 and for a ϕ 1,ϕ 2 ) V. S p n ),ϕ τ F n, ϕ + τ F n 3 n,ϕ Γ Γ = S 1 p n 1 1 ),ϕ 1, 19) 8
10 Remark 4. If p n 1, pn 2 ) V s a souton of Probem 1, we have pn 1 Γ = p n 2 Γ by defnton of V. Testng n 19) by an arbtrary ϕ C 0 Ω ) shows that the dstrbuton F n F n Hdv,Ω ) and s reguar and n L 2, yedng S p n ) S p n 1 ) = τ F n a. e. n Ω 20) by the varatona emma. By Lemma III. 1.1 n [53], F n n H 1/2 Ω ) and ntegratng by parts n 19) yeds for a ϕ 1,ϕ 2 ) V. Therefore 0 = F n n,ϕ Γ n H 1/2 00 Γ) snce the trace s a surjectve operator. Γ + F n 3 n,ϕ Γ Γ 21) F n n = F n 3 n 22) Note addtonay that Probem 1 s equvaent to the sem-dscrete Rchards equaton on the whoe doman, namey to fnd p n 1, pn 2 ) V such that S1 p n 1 ),ϕ 1 τ F n 1, ϕ 1 + S2 p n 2 ),ϕ 2 τ F n 2, ϕ 2 for a ϕ 1,ϕ 2 ) V. = S 1 p n 1 1 ),ϕ 1 + S2 p n 1 2 ),ϕ 2, 23) Remark 5. By appyng a Krchhoff transform n each subdoman Ω, Probem 1 can be reformuated as a nonnear transmsson probem. The exstence and unqueness of a souton for such probems has been studed n [55, 56] for the case when Ω 1 s surrounded by Ω 2, and the common boundary s smooth, however. p n 1 2 Now we can gve the weak form of the teratve scheme. Let n N and assume that the par p n 1 ) V s gven. Furthermore, et λ > 0 and L > 0 = 1,2) be fxed parameters and p n,0 := p n 1, as we as g 0 := F n 1 n λ p n 1 Γ. The teratve scheme s defned through Probem 2 L-scheme, weak form). Let N and assume that the approxmatons { p n,k } 1 k=0 and { g k } 1 k=0 are known for = 1,2. Fnd p n, 1, ) pn, 2 V such that L p n, + g,ϕ Γ,ϕ τ F n,, ϕ + τ λ p n, = L p n, 1,ϕ S p n, 1 ) ) S p n 1,ϕ g,ϕ Γ := 2λ p n, 1 3 g 1 3,ϕ 1, 24) Γ 25) 9
11 hods for a ϕ 1,ϕ 2 ) V Intutve justfcaton of the L-scheme We start the anayss by takng a coser ook at the forma mt of the L-scheme teratons n weak form and show that ths s actuay a reformuaton of Probem 1. Lemma 2 Lmt of the L-scheme). Let n N be fxed and assume that the functons p n V and g H 1/2 00 Γ) = 1,2) exst such that S p n ),ϕ ) S p n 1,ϕ τ F n, ϕ + τ λ p n,+g,ϕ g,ϕ Γ = 2λ p n 3 g 3,ϕ hod for a ϕ 1,ϕ 2 ) V. Then the nterface condtons Γ = 0, 26) Γ, 27) p n 1 Γ = p n 2 Γ n H 1/2 00 Γ), 28) F n 1 n 1 = F n 2 n 1 n H 1/2 00 Γ) 29) are satsfed and p n 1, pn 2 ) soves Probem 1. Moreover, g = λ p n Γ + F n n 30) n H 1/2 00 Γ). Conversey, f p n 1, pn 2 ) V s a souton of Probem 1 and g := λ p n Γ + F n n, then p n and g sove the system 26), 27). Remark 6. Lemma 2 states that sovng Probem 1 s equvaent to fndng a souton to 26), 27). Ths reformuaton w be used to show, that the L-scheme converges to a souton of Probem 1 Proof. Wrtng out 27) for = 1,2 and subtractng the resutng equatons yeds p n 1 Γ = p n 2 Γ n the sense of traces. On the other hand, addng up these equatons eads to g 1 + g 2 ) = λp n 1 Γ + p n 2 Γ). Insertng ths nto the sum of the equatons 26) eads to 23), and by equvaence to the sem-dscrete formuaton 19). Moreover, by 20) one has S p n ) S p n 1 ) = τ F n a.e. and therefore ntegratng by parts n 26) gves g = λ p n Γ + F n n n H 1/2 00 Γ). Conversey, f p n 1, pn 2 ) soves Probem 1, then pn 1 Γ = p n 2 Γ and g = λ p n Γ + F n n = λ p n 3 Γ + F n 3 n 3 = 2λ p n 3 Γ g 3 31) s deduced by the fux contnuty 22). Fnay, 26) now foows by ntegratng 20) by parts and usng the defnton of g. 10
12 3.2. Convergence of the scheme The convergence of the L-scheme nvoves two steps: frst, we prove the exstence and unqueness of a souton to Probem 2 defnng the near teratons, and then we prove the convergence of the sequence of such soutons to the expected mt. Lemma 3. Probem 2 has a unque souton. Proof. Ths s a drect consequence of the Lax-Mgram emma. We now prove the convergence resut, whch was announced n Theorem 1. We assume that the souton p n 1 1, p n 1 ) 2 of Probem 1 at tme step n 1) s known and et p n,0 V be arbtrary startng pressures however, a natura choce s p n,0 := p n 1 ). Lemma 3 enabes us to construct a sequence { p n, } V N N 0 of soutons to Probem 2 and prove ts convergence to the souton p n 1, pn 2) of Probem 1 at the subsequent tme step. Theorem 4 Convergence of the DD scheme). Assume there exsts a souton p n 1, pn 2 ) V to Probem 1 s.t. sup p n + z) L M < and et g be as n 30). Let Assumptons 1 hod, λ > 0 and L R be gven wth L S /2 < L for = 1,2. For arbtrary startng pressures p n,0 := v,0 V = 1,2) et { p n, 1, pn, 2 )} be the sequence of soutons of Probem 2 and et { g } N 0 be defned by 25). Assume N 0 further that the tme step τ satsfes τ < Then p n, p n n V and g g n V as for = 1,2. 2m 1 Lk 2 M 2 1 ). 32) L S 2L Remark 7. The essenta boundedness of the pressure gradents can be proven under the addtona assumpton that the functons S are strcty ncreasng and the doman s of cass C 1,α, see e.g. [57, Lemma 2.1]. Proof. We ntroduce the teraton errors e p, := pn pn, L p n,ϕ to 26) and subtract 24) to arrve at [ L e p,,ϕ + τλ e p,,ϕ Γ + τ e g,,ϕ Γ + τ F n k S p n, 1 +k S p n, 1 ) ) ] p n + z) + F n,, ϕ = L e 1 p,,ϕ as we as e g, := gn g, add L p n,ϕ Insertng ϕ := e p, n 33) and notng that L e p, e 1 p,,e p, = L [ e 2 2 p, e 1 2 p, + e p, e 1 2] p,, ) ) p n + z) S p n ) S p n, 1 ),ϕ. 33) 11
13 yeds L 2 [ e 2 p, e e p, e 1 2] + S p n ) S p n, 1 p, = S p n ) S p n, 1 ),e 1 p, e p, τ }{{} =:I 2 p, ),e 1 p, } {{ } =:I 1 e g,,e p, τ k S p n )) k S p n, 1 ) )) p n + z), e p, }{{} τ =:I 3. k S p n, 1 ) ) e p,, e p, }{{} =:I 4 Γ +τλ e p,,e p, Γ 34) We estmate now the terms I 1 I 4 n 34) one by one. By Assumpton 1c), for I 1 we have 1 S p n L ) S p n, 1 ) 2 S p n ) S p n, 1 S ),e 1 p,. 35) I 2 s estmated by I 2 = S p n ) S p n, 1 L e 1 2 p, e p, ),e 1 p, e p, L S p n ) S p n, 1 ) 2. 36) For an ε > 0 to be chosen beow we use Young s nequaty to dea wth I 3, whch can be estmated by I3 = τ k S p n )) k S p n, 1 ) )) p n + z), e p, τ k S p n )) k S p n, 1 τl k M S p n ) S p n, 1 τl k Mε S p n ) S p n, 1 ) )) p n + z) e p, ) e p, ) 2 + τ L k M 4ε e 2 p,, 37) where we used the Lpschtz contnuty of k and the assumpton sup p n + z) L < M. Fnay, by Assumpton 1b) one has τ k S p n, 1 ) ) e p,, e p, for I 4. Usng the estmates 35) 38), 34) becomes [ e 2 p, e 1 2] p, + 1 S p n L ) S p n, 1 S ) 1 S + τl k Mε p n 2L ) S p n, 1 ) 2 + τ L 2 τm e 2 p, 38) ) 2 + τλ e p,,e p, Lk M 4ε ) e m Γ + τ e g,,e p, p, Γ ) 12
14 In order to dea wth the nterface term τ e g,,e p, reca, that denotes the dua parng of H1/2 Γ, Γ 00 Γ) and H 1/2 00 Γ) and the H1/2 00 Γ)-norm smutaneousy. Subtractng 25) from 27),.e. e g, = 2λe 1 p,3 e 1 g,3, we get e p, 2 Γ = 1 4λ 2 e +1 g,3 2 Γ e 2 g, Γ 4λ e ) p,,e g,. 39) Γ Wth b {p,g} we et e b := e b,1,e b,2 ) V 1 V 2 and e b 2 := =1 e b, 2. Smary, on Γ we et e b,e b Γ := 2 =1 e b,,e b, Γ and correspondngy e b 2 Γ = 2 =1 e b, 2 Γ. Summng n 39) over = 1,2 gets e p 2 Γ = 1 4λ 2 e +1 g Dong the same for 34 ) and nsertng 40), eaves us wth L 2 [ e p 2 e 1 2] + + τ 4λ 2 =1 p 2 Γ e g 2 Γ 4λ e p,e ) g. 40) 2 =1 e +1 g 2 Γ e g 2 Γ 1 L S S p n ) S p n, 1 ) 2 ) 2 + τ =1 m L k M 4ε Γ ) e 2 p, ) 1 S + τl k Mε p n 2L ) S p n, 1 ) 2. 41) Now, summng for the teraton ndex = 1,...,r and notcng teescopc sums one gets r 2 =1 =1 + τ L 2 1 L S 1 2L τl k Mε ) S p n ) S p n, 1 ) 2 r 2 =1 =1 m L k M 4ε ) e [ e 0 p 2 e r p 2 ] + τ 4λ 2 p, e 1 g 2 Γ e r+1 g 2 Γ ). 42) Now we choose ε = L k M 2m, hence m L k M 4ε = m 2 > 0 for both. Recang the restrcton on L, 1 LS 1 2L > 0, as we as that by the tme step restrcton L 1 S 2L 1 τ L2 k M 2 2m > 0 for = 1,2, the estmates r 2 =1 =1 τ 1 1 τ L2 k M 2 L S 2L 2m r m e 2 p L 2 2 =1 ) S p n ) S p n, 1 ) 2 L 2 e 0 p 2 + τ 4λ e1 g 2 Γ, 43) e 0 p 2 + τ 4λ e1 g 2 Γ 44) foow for for any r N. Snce the rght hand sdes are ndependent of r, we thereby concude that the seres on the eft are absoutey convergent and therefore S p n ) S p n, 1 ), e p, 0 as. Moreover, 44) mpes e p, 0, ) as we, by the Poncaré nequaty. 13
15 to get To show that e g, 0 n V we subtract agan 24) from 26) and consder test functons ϕ C 0 Ω ) τ F n Fn,, ϕ = L e p,,ϕ + L e 1 p,,ϕ S p n ) S p n, 1 ),ϕ. 45) ) Thus, F n Fn, exsts n L 2 and τ F n Fn, ) = L e p, e 1 p, amost everywhere. Therefore, for any ϕ V one has ) + S p n ) S p n, 1 ) 46) F n ) L Fn,,ϕ e τ p, e 1 p, ϕ 1 + S p n τ p n, 1 ) ϕ. 47) Abbrevatng the eft hand sde of 47) as Ψ n, ) ϕ, 47) means Ψ n, ) ϕ sup L e ϕ V ϕ V τ p, e 1 p, ϕ S p n τ p n, 1 ) 0 ) 48) as a consequence of 44). In other words Ψ n, V 0 as. Startng agan from 33) wthout the added zero term), ths tme however nsertng ϕ V, ntegratng by parts and keepng n mnd 46) one gets e g,,ϕ Γ = λ e p,,ϕ [F Γ + n ] Fn, n,ϕ. 49) Γ We aready know that e p, V 0 as 0 so by the contnuty of the trace operator the frst term on the rght vanshes n the mt. For the ast summand n 49) we use the ntegraton by parts formua to obtan [F n ] Fn, n,ϕ Γ Γ = Ψn, ϕ ) + F n Fn,, ϕ. 50) Whe the frst term on the rght approaches 0, the second can be estmated by k S p n )) p n + z) k S p n, 1 ) ) p n, + z ), ϕ L k M ) S p n S p n, 1) ϕ V + p n, ϕ V, 51) where we used the same reasonng as n 37). Wth ths we et n 50) to obtan [F sup n ] Fn, n,ϕ ϕ V ϕ V =1 Ψ n, Γ V + L k M S p n ) S p n, 1 ) + p n, 0. 52) 14
16 Fnay, usng the above and ettng n 49) gves e g, sup,ϕ Γ 0. ϕ V ϕ V ϕ 0 Ths shows e g, 0 n V for both and concudes the proof. Remark 8. Note that Theorem 4 states that f a souton to the sem-dscrete couped probem exsts, then t s the mt of the teraton scheme. Snce n the convergence proof we use the exstence of a souton to Probem 1, the argument cannot be used to prove exstence. The dffcuty es n the fact that the nonneartes encountered n the dffuson terms are space dependent and may be dscontnuous w.r.t. x over the nterface. 4. Numerca Experments Ths secton s devoted to numerca experments and the mpementaton of the proposed doman decomposton L-scheme. As our formuaton and anayss dd not specase to a partcuar spaca dscretsaton, the numerca mpementaton of the LDD scheme can n prncpa be done wth fnte dfference, fnte eements as we as fnte voume schemes. Snce mass conservaton s an essenta feature of porous meda fow modes, we adopted a ce-centred two pont fux approxmaton varant of a fnte voume scheme to refect ths on the numerca eve. The doman Ω s assumed to be rectanguar and a rectanguar unform mesh was used. Remark 9 dfferent decoupng formuatons revsted). We saw n Remark 2 that another decoupng formuaton s possbe. In fact, ths can be taken a step further. Equatons 11), 12) as we as 11 ), 12 ) can be embedded nto a combned formuaton. For some 0 < η < 1 and M > 0, consder the generased decoupng [ ] F n, n = M 1 η)g + η pn,, 11 ) 1 η)g = 2η pn, η)g ) Observe that the λ-formuaton 11), 12), as we as the convex-combnaton formuaton 11 ), 12 ), are speca cases of ths genera formuaton: In partcuar, M = 1 η) 1 and λ = η1 η) 1 recovers the λ-formuaton, M = 1 and η = λ yeds the convex-combnaton formuaton. Athough 11 ) and 12 ) mght gve even greater parametrc contro over the numercs, n ths paper we adhere to the λ-formuaton because of ts smpcty. Fg. 10 and Fg. 11 show the nfuence of λ and η n both formuatons. 15
17 We start by consderng an anaytcay sovabe exampe. The LDD scheme s tested aganst other frequenty used schemes that do not use a doman decomposton. A of them are defned on the entre doman and the contnuty of norma fux and pressure over Γ s mantaned mpcty. The frst scheme to be compared s a fnte voume mpementaton of the orgna L-scheme on the whoe doman see [16, 31, 33]), henceforth referred to as LFV scheme. Comparson s aso drawn to the modfed Pcard scheme, whch performs better than the Pcard method, see [26]), whch s gven by S p n, 1 ) p n, ) p n, 1 + τ F n, = τ f S p n, 1 ) ) ) S p n 1 on Ω, 53) F n, n 1 = 0 on Γ. 54) Here, the brackets denote the jump over the nterface. Fnay, a comparson wth the quadratcay convergent Newton scheme s made. Wrtng δ p = pn, p n, 1, t reads as foows: S p n, 1) δ p τ [k S p n, 1 ) ) δ p + k S p n, 1 ) ) S p n, 1) δ p p n, + z )] = τ f S p n, 1) ) ) S p n 1 τ k S p n, 1 ) ) p n, 1 + z )) on Ω 55) k S p n, 1 ) ) δ p n 1 + k S p n, 1 ) ) δ p p n, 1 + z ) n 1 = k S p n, 1 ) ) p n, 1 + z ) n 1 on Γ. 56) We refer to [33] for a recent study on nearsatons for Rchards equaton Resuts for a case wth known exact souton To demonstrate the robustness of the proposed scheme, we sove 2) 5) wth both Drchet and Neumann type boundary condtons. In the frst case we dsregard gravty. Specfcay, we consder Ω 1 = 1,0) 0,1), Ω 2 = 0,1) 0,1), and Γ = {0} [0,1]. 57) The reatve permeabtes are k 1 S 1 ) = S1 2 on Ω 1, k 2 S 2 ) = S2 3 on Ω 2 and the saturatons 1 for p < 0, S p) = 1 p) +1 1, = 1,2. 58) 1 for p 0 The boundares and rght hand sdes are chosen to make the exact souton p 1 x,y,t) = 1 1 +t 2 )1 + x 2 + y 2 ), t > 0, x,y) Ω 1, p 2 x,y,t) = 1 1 +t 2 )1 + y 2 ), t > 0, x,y) Ω 2, 16
18 Fgure 2: The doman used n the numerca exampes. The boundary condtons are gven n Tabe 1. The exact souton s aso gven n each subdoman. Ω 1 Ω 2 t = 0 p 1 x,y,0) = x 2 + y 2 ) p 2 x,y,0) = y 2 BCy = 0 y p 1 = 0 y p 2 = 0 y = 1 k 1 S1 p 1 ) ) y p 1 = 2 2+x 2 k 2 S2 p 2 ) ) y p 2 = 1 x = 1 p 1 1,y,t) = 1 1 +t 2 )2 + y 2 ) x = 1 p 2 1,y,t) = 1 1 +t 2 )1 + y 2 ) Tabe 1: Inta and boundary condtons for the exampe wth exact souton. and ths corresponds to the rght hand sdes f 1 x,y,t) = x 2 + y 2 ) 2 t 1 +t 2 ) x 2 + y 2 ), f 2 x,y,t) = 21 y2 ) 1 + y 2 ) 2 2t t 2 ) y 2 ), for t > 0, and x,y) Ω respectvey. The boundary and nta condtons are summed up n Tabe 1. A near systems were soved usng a restarted generased mnmum resdua method gmres) [58]. To boost up speed, sparse trpet format was used n the matrx computaton. The programs are mpemented n ANSI C. For the mpementaton we took the same L n both sub-domans,.e. L := L 1 = L 2. The resuts are shown n Fgures 3 and 4a. Fg. 3 shows the pressure dstrbuton of the exact souton p := χ Ω1 p 1 + χ Ω2 p 2 wth the numerca souton p n, := χ Ω1 p n, 1 + χ Ω 2 p n, 2 potted on top of t. For 17
19 Fgure 3: Comparson between the exact pressure and the numerca pressure provded by the LDD scheme. x = 10 2, t = as we as parameters L = 0.25 and λ = 4, the maxmum reatve error was ess than 0.03%,.e. pn p n, p n L Ω) < The reatve errors of the LDD, LFV and Newton schemes at the md-ne y = 0.5 are potted n Fg. 4a. The LDD scheme preserves the fux contnuty and pressure contnuty at the nterface at every tme step wthout havng to sove for the entre doman. We test ths theory numercay. Fg. 4b shows how dfferent knds of errors behave wthn one tme step. The errors p n, p n, 1 L 2 Ω), pn, p n, 1 L Ω) defned on the doman Ω, as we as p n, L 2 Γ) and F n, n L 2 defned on the nterface Γ, are shown. We observe that the fux and pressure jump tend Γ) to zero whch mpes that fux and pressure contnuty s acheved. Note that the fux at x = 0 from the exact souton s 0. Next, we compare the LDD scheme wth other schemes and study ther dependence on dscretsaton parameters. We compare the Newton scheme, the modfed) Pcard teraton, the aready mentoned LFV scheme and the LDD scheme, nvestgatng the dependence of tme step refnement and space grd refnement separatey. The frst study, shown n Fg. 5, pots og 10 p n, p n, 1 L 2 Ω)) for a schemes, at the fxed tme step correspondng to t = 0.2. As expected, Newton s the fastest and shows a quadratc convergence rate. But at the same tme, t s most susceptbe to change n mesh sze as observed from the sopes of the eft-most curves. The convergence rate of the Pcard teraton s near, faster than both the L-schemes and s stabe wth respect to varaton n mesh sze. The L-schemes aso exhbt near convergence, abet sower than Pcard, and the convergence speed does not vary much wth mesh sze. LFV and LDD schemes have practcay the same convergence rate. Tabe 2 compements the pot n Fg. 5 and sts expermenta average convergence rates, defned as e n,+1 p / e n, p, for a schemes Newton data s not shown for 18
20 a) Comparson between the numerca soutons pro- b) Dfferent errors vs nner teratons for the case wth vded by the LDD, LFV and the Newton schemes. Pot- exact souton. Here t = 0.2, L = 0.25 and λ = 4. ted are the reatve errors p exact p num p exact as functons of x, for y = 0.5 and t = 1. x = 0.1, 0.05, 0.02 as t reaches an error ower than n 3 teratons). Secondy, we study the dependence of the convergence rates on tme step sze for a fxed mesh x Newton Pcard LFV LDD sze x = 0.02). The error characterstcs of a four schemes n Fg. 6 are shown for t = 0.5. In Fg. 6a both, Newton and Pcard, dverge, whereas both L-schemes converge for L = The LFV scheme exhbts some oscatons, the reason beng the dependence of the choce of L on the tme Type Quadratc Lnear Lnear Lnear Tabe 2: The average convergence rate, e n,+1 / e n,, for the step τ. Hgher vaues of τ mght requre hgher dfferent schemes and wth respect to the mesh-sze. vaues of L. Indeed, f we substtute L = 0.5 n the LFV scheme marked as LFV* n the dagram), we see a more robust behavour. Note, that the LDD scheme converges for a τ and s at east as fast as the LFV scheme n a the cases. For smaer vaues of τ the Newton and Pcard teraton converge faster than both L-schemes, as shown n Fgures 6b and 6c. Accordng to the theory, the convergence of the Newton and Pcard schemes s ony guaranteed f the nta guess s cose enough to the exact souton. Therefore, startng the teraton wth the numerca souton at the prevous tme step ths suggests that the tme step shoud be taken sma enough to have a 19
21 Fgure 5: Performance comparson and mesh study for the convergence of the LDD, LFV, Pcard and Newton schemes. Here L = 0.25 and λ = 4. guaranteed convergence see [24, 30, 33]. Contrarwse, L-schemes are free of ths constrant. To ustrate ths behavour, we have nvestgated the convergence of the schemes for a constant nta guess. Specfcay, p n,0 = 5 has been used nstead of p n,0 = p n 1. In ths case, the Newton and Pcard schemes are dvergent whereas both L-schemes st produce a good approxmaton after severa teratons. Ths s dspayed n Fg. 7. A smar behavour w be observed agan whe dscussng a numerca exampe wth reastc parameters. Remark 10. The convergence behavour of the Fgure 7: Error decay for the dfferent schemes for a constant LDD scheme can be optmzed by choosng λ nta guess, p n,0 = 5. Here L = 0.25, λ = 4. propery. In the above comparson λ was chosen dfferenty for every choce of mesh sze. The optmaty of λ s dependent on the mesh and the tme step 20
22 a) t = 0.1 b) t = 0.01 c) t = Fgure 6: Convergence study for the tme-steps τ = 0.1, 0.01, Here, L = 0.25 for the LFV scheme and L = 0.5 for the LFV* scheme. For the LDD scheme one has L = 0.25, λ = 2 n case 6a, L = 0.25, λ = 4 n case 6b, and L = 0.25, λ = 10 n case 6c. 21
23 sze. Wth a good choce of λ, one can make the LDD scheme at east as fast as the LFV scheme. Ths s dscussed n more deta n Secton Resuts for a reastc case wth van Genuchten parameters We demonstrate the appcabty of the LDD scheme for a case wth reastc parameters, ncorporatng aso gravty effects. We consder a van-genuchten-muaem parametrsaton [59] wth the curves k and S S p) = S,r + S,s S,r )Φ p), Φ p) = α p) ˆn ) m, m = 1 1ˆn, k S) = Φ p) 1 m 1 Φ p) 1 ) ) m 2. 59) The specfc parameter vaues are sted n Tabe 3 and are characterstc for partcuar types of materas, st oam G.E. 3 Ω 1 ) and sandstone Ω 2 ). These materas have very dfferent absoute permeabtes κ 1,κ 2, whch makes the numerca cacuatons more chaengng. The dmensona governng equatons and boundary condtons become = 1,2) In ths case F n, = κ µ k r, L p n, + τ F n, = L p n, 1 φ S p n, 1 ) S p n 1 ) ), on Ω, 60) F n, n = g + 2λ pn,, on Γ, 61) p n,, = 0 on Ω. 62) S p n, 1 ) ) p n, ρ g ). Here g = ge x s the gravtatona acceeraton agned wth the postve x-drecton, ρ, µ are the densty and the vscosty of the fud and κ, φ are the absoute permeabty as we as the porosty of the medum. Note that Fg. 2 s rotated by 90 degrees. The probem s nondmensonased by usng the characterstc pressure p := Pa, ength 1.48m and tme s. Ths eads to the nondmensona quanttes p, x, y) and t. After nondmensonasaton, the doman used s agan taken to be Ω 1 = 1,0) 0,1), Ω 2 = 0,1) 0,1). The nta condton used s px,y,0) = 1 63) and boundary condtons are 1 +ty f y < 1 ε)t 1 p 1,y,t) =, ε f y 1 ε)t 1 p1,y,t) = 1, 22
24 Parameter Unt St Loam G.E. 3 Ω 1 ) Sandstone Ω 2 ) Porosty φ ) Water Densty ρ) kg m Water Vscosty µ) Pa s Absoute permeabty κ ) m 2 s Retenton exponent ˆn ) Retenton parameter α ) Pa Irreducbe water saturaton S,r ) Irreducbe ar saturaton 1 S,s ) Tabe 3: The van Genuchten-Muaem parameters n the reastc test case. together wth a no-fow condton at y = 0,1. We take ε > 0 to avod degeneracy. Fg. 8a shows the dfferent errors for ths case and t can be seen that a the errors are decreasng for the LDD scheme. Errors at the nterface and nsde the doman tend to 0, the convergence s sower compared to the case wth exact souton, however. Ths s due to the arge varance of the parameters as we as the hghy nonnear nature of the assocated functons. Because of ths, both Newton and Pcard schemes dverge. The behavour of dfferent schemes for the same set of parameters s shown n Fg. 8b. Observe that for the Newton scheme the startng error as we as the number of teratons requred ncreases steady wth t unt t = 0.94, at whch pont the errors start dvergng. The Pcard scheme becomes dvergent even before t = 0.2. In contrast, both L-schemes reman stabe n ths case Tme Performance Ths secton s devoted to the comparson of tme performance of the schemes. We have seen that L-schemes are more stabe than Newton and Pcard. But f they are convergng, then Newton and Pcard schemes converge faster than the L-schemes. Beow we nvestgate how the schemes compare to one another wth respect to actua computatona tme. We set an error toerance for the schemes that stops the teratons wthn one tme step, after reachng an error ower than 10 6,.e. p n, p n, 1 L 2 Ω) < Ths s to ensure that we get comparatve CPU-cock-tme for dfferent schemes for the same degree of accuracy. We computed the exacty sovabe case on a LINUX server mammoth.wn.tue.n) for a four schemes usng the same set of parameters x = 0.02, τ = 0.001, L = 0.25 and λ = 10). Fgure 9 ustrates the tme-performances of these schemes over the whoe computatona tme doman. Tabe 5 shows how many 23
25 a) Dfferent errors vs nner teratons for the reastc b) Error vs nner teratons for the reastc case. LDD, case at t = 0.2. The parameters are τ = 0.01, x = 0.02, LFV and Newton errors are potted at t = 0.2. Newton L = 0.25 and λ = 10. Ony the LDD scheme s shown n denotes the errors of Newton scheme at t = 0.9. Pcard s ths pot. potted at t = Here, L = 0.5, λ = 10. Fgure 8: Error pots and scheme comparson for the reastc case. nner-teratons are requred on average for dfferent schemes to reach the error crteron at dfferent ponts n tme. Iteraton requrement per tme step ncreases for a schemes as the boundary condtons change more rapdy wth tme. Tabe 5 shows the average tme taken and how many gmres teratons outer and nner) were requred by each scheme to execute one nner teraton. Unsurprsngy, the Newton scheme s st fastest, foowed by Pcard and the LDD scheme. But LDD competes cosey wth Newton and Pcard. Even more surprsng s the fact that the LFV scheme takes consderaby more tme to reach the Fgure 9: Tme performance of the L-DD, L-FV and the Newtondesred accuracy compared to the LDD scheme, FV schemes. despte both havng amost the same convergence rate. The reason becomes apparent from Tabe 5: The LDD scheme requres much ess tme per nner 24
26 Average nner teratons requred Tme-step/Scheme LDD LFV Pcard Newton Avg. tme per ter Avg. GMRES teratons Tabe 5: The average number of nner teratons per tme step requred by the dfferent schemes to reach the stoppng crteron p n, p n, 1 L 2 Ω) < The ast two rows gve the average tme and gmres-teratons per nner teraton. teraton than a other schemes. The LFV scheme has the second fastest average tme per teraton. For the Pcard teraton, the dervatve of the saturaton functon needs to be evauated whch n turn costs more tme than an teraton n the LFV scheme. The Newton scheme s computatonay most expensve per teraton because t cacuates the Jacoban at every teraton. The schemes that do not decoupe the doman requre much more tme and many more gmresteratons Condton number per nner teraton. The reason s that x the doman decomposton schemes nvove smaer L-DD Ω 1 ) matrces and and they have smaer condton numbers. L-DD Ω 2 ) Ths s ustrated by the ast row of Tabe 5. The LDD scheme requres on average 119 gmresteratons L-FV Ω) Tabe 4: The condton number vs mesh sze for the LDD and on Ω 1 and 123 gmres-teratons on Ω 2 and both domans have eements. Compare ths wth Newton, whch takes amost 400 gmresteratons LFV schemes. Here, τ = 0.001, t = 0.2, L = 0.25, λ = 10. The condton numbers are cacuated for the 200 th tme step for the matrces of the frst nner teraton. and deas wth varabes on each gmres-teraton. Ths expans why the LDD scheme takes so much ess tme per nner teraton. Tabe 4 compares the condton numbers of the LDD and the LFV scheme. It shows that the matrces for the LFV scheme are worse condtoned than the ones of the LDD scheme. The atter has two condton numbers, one for each doman. The 2-norm condton numbers were cacuated wth MATLAB s bud n cond) functon. 25
27 Remark 11. The fact that the LDD scheme performance competes cosey wth Newton and Pcard, means that, LDD can potentay be made much faster than even Newton as t s paraesabe. Ths s the key advantage of the LDD scheme aong wth ts goba convergence property Parameter dependence and key features Havng outned the robustness and speed of the proposed LDD scheme we turn to nvestgate some of ts propertes. Two mportant parameters have been ntroduced n the L-DD scheme,.e. L and λ, and apart from a ower bound on L nothng has been specfed about these parameters. Ths means that they can freey be adjusted to gve optma convergence rate. In fact, n ths secton we w see that the convergence rate depends strongy on these parameters. The nfuence of λ a) The decay of the pressure error n terms of λ. b) The decay of the g-error n terms of λ. Fgure 10: The nfuence of λ on the convergence rate. The parameters for the LDD scheme are τ =.01, x = 0.02, L = 0.25 at t = 0.2. Fgure 10 shows the nfuence of the parameter λ on error characterstcs. A the resuts shown are for the case wth exact souton. Fgure 10a focuses on the errors p n, p n, 1 L 2 Ω) on the doman Ω, whe Fg. 10b depcts the L 2 -errors g g 1 L 2 Γ) on the nterface for the same tme step. Ceary, λ has tremendous mpact on the convergence rate. The convergence rate rapdy ncreases wth λ at frst but after a certan pont the convergence rate starts decreasng. Ths trend s notceabe n both pots of Fgure 10. Ths ndcates that there s an optma ambda λ opt for whch the whoe scheme has a fastest convergence rate. The optmaty of λ s actuay a we studed behavour n the doman decomposton 26
28 terature. In [60, 50] t has been shown that λ opt depends at east on mesh sze and sub- doman sze. Later we w show that t aso depends on L and τ n our case. Ths contro over the convergence rate s the reason why the λ-formuaton was chosen over the convex-combnaton formuaton gven n Remark 2. To ustrate ths, Fg. 11 shows the same pots as Fgure 10 but for the convex-combnaton formuaton. In order to dfferentate between pots more easy, we use the combned formuaton 11 ), 12 ) and set M = 1. For η = 0.01 the convex-combnaton formuaton even fas to converge. In a other cases the convergence s consderaby sower. a) The decay of the pressure error n terms of η. b) The decay of the g-error n terms of η. Fgure 11: The nfuence of η on the convergence rate n the convex-combnaton formuaton M = 1 n Remark 2). The parameters for the LDD scheme are τ = 0.01, x = 0.02, L = 0.25 at t = 0.2. The nfuence of L We brefy gve an overvew over the nfuence of L on the convergence rate. Fgure 12 depcts ths for L := L 1 = L 2. For L-schemes t s common to dverge f L s too sma, whch seems to be the case for L = 0.1. On the other hand, the convergence rate decreases sgnfcanty for very arge L, a behavour that s a common trat of L-schemes as we, cf. [10]. It s best to choose L as sma as possbe, yet great enough to ensure convergence of the scheme. Note that L = 0 represents the orgna nonmodfed) Pcard teraton case and Fgure 12 suggests that the orgna Pcard scheme fas for these probems. The dependence of λ opt on L, τ and x In ths ast secton we nvestgate numercay how λ opt depends on the choce of L, τ and x. For a fxed grd n tme and space Tabe 6 sts convergence rates for dfferent λ and L. 27
29 Wth ths tabe we can guess the nterva n whch λ opt es. Wthn ths estmated nterva, Fg. 13 shows how the convergence rate vares wth λ for fxed L. For L = 0.25, x = 0.02 and τ = 0.01 the fastest convergence s acheved for λ = 4 ths s why λ = 4 was chosen for the above comparsons, wherever the specfed L, x, τ set was used). The λ dependence for hgher vaues of L s ess pronounced. For a fxed L, Tabes 7a, 7b show the varance of λ opt wth respect to tme-step and mesh sze respectvey. The shown tabes are of course ony a rough estmate of λ opt. Due to computaton tme, t s a tedous process to fnd a cose to exact vaue of λ opt, Fgure 12: The nfuence of L on the convergence rate, as obtaned for the nner teratons for the 50 th tme step. especay for very sma tme-step szes. In practce the vaues are numercay guessed. The resuts ndcate qute a strong correaton of λ opt wth the tme-step sze, contrasted by a rather mnor correaton wth the mesh sze. L λ = 0.1 λ = 1 λ = 10 λ = 100 λ opt 0.1 dverged dverged dverged dverged ,10) 1 dverged ,10) 5 dverged ,100) Tabe 6: The dependence of the convergence rates on λ and L: the geometrc average of the contracton rates over the frst 20 teratons and for dfferent L,λ) pars s gven n the frst coumns, whereas the ast gves the nterva for λ opt. Here, x = 0.02, t = 0.01, t = 0.2. t Nr ter.? Avg. CR a) λ opt for x = 0.02, L = 0.25 x Nr ter.? Avg. CR b) λ opt for t = 0.01, L = 0.25 Tabe 7: The dependence of λ opt on t and on x. 28
30 5. Concuson We consdered a nonnear paraboc probem appearng as mathematca mode for varaby saturated fow n porous meda. For the numerca souton of the nonnear, tme dscrete probems we proposed a combned scheme LDD) that s based on a fxed pont teraton the L-scheme), and on a doman decomposton scheme nvovng Robn type coupng condtons at the nterface separatng dfferent subdomans. The resut s a scheme featurng the advantages of both approaches: an uncondtona convergence, regardess of tme step and startng pont, as we as a decoupng Fgure 13: Convergence rate vs λ for L = 0.25 and L = 1. of the tme dscrete probems nto subprobems that For L = 0.25, λ opt 4. can be soved n parae. The stabty, robustness and effcency of the method s tested for varous cases and aso compared to Newton and Pcard schemes. The tests ncude stuatons where the atter dverge whereas the proposed scheme s convergng. In summary, the key advantages of the method are: The LDD scheme converges uncondtonay. It can provde accurate resuts even n stuatons where the Pcard or Newton teratons fa. In conjuncton wth a sutabe space dscretsaton, t provdes a decouped, mass conservatve approach. Ths s very usefu n partcuar when deang wth modes defned n meda wth bocktype heterogenetes, where the matera propertes n dfferent bocks may vary sgnfcanty. Though neary convergent, the computatona tme requred by the LDD scheme for achevng a certan accuracy of the approxmaton s comparabe to the tme needed by Newton and Pcard schemes, and much faster than a standard L-scheme apped to the mode n the entre doman. Ths effcency s due to the fact that the scheme needs ess tme per nner teraton than a scheme defned n the entre doman. Moreover LDD s paraesabe, whch gves the possbty of ncreasng ts effcency even further. The convergence rate of LDD schemes depends on the choce of L and λ. Wth the optma choce of parameters, the convergence order can be reduced sgnfcanty. 29
31 Acknowedgements Ths work was supported by Netherands Organsaton for Scentfc Research NWO Vstors Grant , Odysseus programme of the Research Foundaton - Fanders FWO G0G1316N), by UHasset Speca Research Fund project BOF17BL04, the Norwegan Research Counc through the projects NRC CHI) and NRC IMMENS) as we as She-NWO/FOM CSER programme project 14CSER016) and the German Research Foundaton through IRTG 1398 NUPUS project B17). References [1] R. Hemg, Mutphase fow and transport processes n the subsurface: a contrbuton to the modeng of hydrosystems, Sprnger, Bern, [2] L. F. Rchardson, Weather predcton by numerca process, Camebrdge Unversty Press, [3] L. A. Rchards, Capard conducton of quds through porous medums, Physcs 1 5) 1931) [4] H. W. At, S. Luckhaus, Quasnear eptc-paraboc dfferenta equatons, Math. Z ) 1983) [5] H. W. At, S. Luckhaus, A. Vsntn, On nonstatonary fow through porous meda, Anna d Matematca Pura ed Appcata 136 1) 1984) [6] F. Otto, L1-contracton and unqueness for quasnear eptc paraboc equatons, Journa of Dfferenta Equatons 131 1) 1996) [7] T. Arbogast, M. F. Wheeer, A Nonnear Mxed Fnte Eement Method for a Degenerate Paraboc Equaton Arsng n Fow n Porous Meda, SIAM J. Numer. Ana. 33 4) 1996) [8] R. H. Nochetto, C. Verd, Approxmaton of degenerate paraboc probems usng numerca ntegraton, SIAM J. Numer. Ana. 25 4) 1988) [9] I. S. Pop, Error estmates for a tme dscretzaton method for the Rchards equaton, Comput. Geosc. 6 2) 2002) [10] F. A. Radu, I. S. Pop, P. Knabner, Order of Convergence Estmates for an Euer Impct, Mxed Fnte Eement Dscretzaton of Rchards Equaton, SIAM J. Numer. Ana. 42 4) 2004)
A finite difference method for heat equation in the unbounded domain
Internatona Conerence on Advanced ectronc Scence and Technoogy (AST 6) A nte derence method or heat equaton n the unbounded doman a Quan Zheng and Xn Zhao Coege o Scence North Chna nversty o Technoogy
More informationResearch on Complex Networks Control Based on Fuzzy Integral Sliding Theory
Advanced Scence and Technoogy Letters Vo.83 (ISA 205), pp.60-65 http://dx.do.org/0.4257/ast.205.83.2 Research on Compex etworks Contro Based on Fuzzy Integra Sdng Theory Dongsheng Yang, Bngqng L, 2, He
More informationSupplementary Material: Learning Structured Weight Uncertainty in Bayesian Neural Networks
Shengyang Sun, Changyou Chen, Lawrence Carn Suppementary Matera: Learnng Structured Weght Uncertanty n Bayesan Neura Networks Shengyang Sun Changyou Chen Lawrence Carn Tsnghua Unversty Duke Unversty Duke
More informationMARKOV CHAIN AND HIDDEN MARKOV MODEL
MARKOV CHAIN AND HIDDEN MARKOV MODEL JIAN ZHANG JIANZHAN@STAT.PURDUE.EDU Markov chan and hdden Markov mode are probaby the smpest modes whch can be used to mode sequenta data,.e. data sampes whch are not
More informationNeural network-based athletics performance prediction optimization model applied research
Avaabe onne www.jocpr.com Journa of Chemca and Pharmaceutca Research, 04, 6(6):8-5 Research Artce ISSN : 0975-784 CODEN(USA) : JCPRC5 Neura networ-based athetcs performance predcton optmzaton mode apped
More informationLower Bounding Procedures for the Single Allocation Hub Location Problem
Lower Boundng Procedures for the Snge Aocaton Hub Locaton Probem Borzou Rostam 1,2 Chrstoph Buchhem 1,4 Fautät für Mathemat, TU Dortmund, Germany J. Faban Meer 1,3 Uwe Causen 1 Insttute of Transport Logstcs,
More informationON AUTOMATIC CONTINUITY OF DERIVATIONS FOR BANACH ALGEBRAS WITH INVOLUTION
European Journa of Mathematcs and Computer Scence Vo. No. 1, 2017 ON AUTOMATC CONTNUTY OF DERVATONS FOR BANACH ALGEBRAS WTH NVOLUTON Mohamed BELAM & Youssef T DL MATC Laboratory Hassan Unversty MORO CCO
More informationQUARTERLY OF APPLIED MATHEMATICS
QUARTERLY OF APPLIED MATHEMATICS Voume XLI October 983 Number 3 DIAKOPTICS OR TEARING-A MATHEMATICAL APPROACH* By P. W. AITCHISON Unversty of Mantoba Abstract. The method of dakoptcs or tearng was ntroduced
More informationImage Classification Using EM And JE algorithms
Machne earnng project report Fa, 2 Xaojn Sh, jennfer@soe Image Cassfcaton Usng EM And JE agorthms Xaojn Sh Department of Computer Engneerng, Unversty of Caforna, Santa Cruz, CA, 9564 jennfer@soe.ucsc.edu
More informationOn the Power Function of the Likelihood Ratio Test for MANOVA
Journa of Mutvarate Anayss 8, 416 41 (00) do:10.1006/jmva.001.036 On the Power Functon of the Lkehood Rato Test for MANOVA Dua Kumar Bhaumk Unversty of South Aabama and Unversty of Inos at Chcago and Sanat
More informationKey words. corner singularities, energy-corrected finite element methods, optimal convergence rates, pollution effect, re-entrant corners
NESTED NEWTON STRATEGIES FOR ENERGY-CORRECTED FINITE ELEMENT METHODS U. RÜDE1, C. WALUGA 2, AND B. WOHLMUTH 2 Abstract. Energy-corrected fnte eement methods provde an attractve technque to dea wth eptc
More informationChapter 6. Rotations and Tensors
Vector Spaces n Physcs 8/6/5 Chapter 6. Rotatons and ensors here s a speca knd of near transformaton whch s used to transforms coordnates from one set of axes to another set of axes (wth the same orgn).
More informationCyclic Codes BCH Codes
Cycc Codes BCH Codes Gaos Feds GF m A Gaos fed of m eements can be obtaned usng the symbos 0,, á, and the eements beng 0,, á, á, á 3 m,... so that fed F* s cosed under mutpcaton wth m eements. The operator
More informationPredicting Model of Traffic Volume Based on Grey-Markov
Vo. No. Modern Apped Scence Predctng Mode of Traffc Voume Based on Grey-Marov Ynpeng Zhang Zhengzhou Muncpa Engneerng Desgn & Research Insttute Zhengzhou 5005 Chna Abstract Grey-marov forecastng mode of
More informationThe line method combined with spectral chebyshev for space-time fractional diffusion equation
Apped and Computatona Mathematcs 014; 3(6): 330-336 Pubshed onne December 31, 014 (http://www.scencepubshnggroup.com/j/acm) do: 10.1164/j.acm.0140306.17 ISS: 3-5605 (Prnt); ISS: 3-5613 (Onne) The ne method
More informationDevelopment of whole CORe Thermal Hydraulic analysis code CORTH Pan JunJie, Tang QiFen, Chai XiaoMing, Lu Wei, Liu Dong
Deveopment of whoe CORe Therma Hydrauc anayss code CORTH Pan JunJe, Tang QFen, Cha XaoMng, Lu We, Lu Dong cence and technoogy on reactor system desgn technoogy, Nucear Power Insttute of Chna, Chengdu,
More informationMultispectral Remote Sensing Image Classification Algorithm Based on Rough Set Theory
Proceedngs of the 2009 IEEE Internatona Conference on Systems Man and Cybernetcs San Antono TX USA - October 2009 Mutspectra Remote Sensng Image Cassfcaton Agorthm Based on Rough Set Theory Yng Wang Xaoyun
More informationLecture 12: Discrete Laplacian
Lecture 12: Dscrete Laplacan Scrbe: Tanye Lu Our goal s to come up wth a dscrete verson of Laplacan operator for trangulated surfaces, so that we can use t n practce to solve related problems We are mostly
More informationAssociative Memories
Assocatve Memores We consder now modes for unsupervsed earnng probems, caed auto-assocaton probems. Assocaton s the task of mappng patterns to patterns. In an assocatve memory the stmuus of an ncompete
More informationLower bounds for the Crossing Number of the Cartesian Product of a Vertex-transitive Graph with a Cycle
Lower bounds for the Crossng Number of the Cartesan Product of a Vertex-transtve Graph wth a Cyce Junho Won MIT-PRIMES December 4, 013 Abstract. The mnmum number of crossngs for a drawngs of a gven graph
More informationMAXIMUM NORM STABILITY OF DIFFERENCE SCHEMES FOR PARABOLIC EQUATIONS ON OVERSET NONMATCHING SPACE-TIME GRIDS
MATHEMATICS OF COMPUTATION Voume 72 Number 242 Pages 619 656 S 0025-57180201462-X Artce eectroncay pubshed on November 4 2002 MAXIMUM NORM STABILITY OF DIFFERENCE SCHEMES FOR PARABOLIC EQUATIONS ON OVERSET
More informationModule 3 LOSSY IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur
Module 3 LOSSY IMAGE COMPRESSION SYSTEMS Verson ECE IIT, Kharagpur Lesson 6 Theory of Quantzaton Verson ECE IIT, Kharagpur Instructonal Objectves At the end of ths lesson, the students should be able to:
More informationChapter 5. Solution of System of Linear Equations. Module No. 6. Solution of Inconsistent and Ill Conditioned Systems
Numercal Analyss by Dr. Anta Pal Assstant Professor Department of Mathematcs Natonal Insttute of Technology Durgapur Durgapur-713209 emal: anta.bue@gmal.com 1 . Chapter 5 Soluton of System of Lnear Equatons
More informationNUMERICAL DIFFERENTIATION
NUMERICAL DIFFERENTIATION 1 Introducton Dfferentaton s a method to compute the rate at whch a dependent output y changes wth respect to the change n the ndependent nput x. Ths rate of change s called the
More informationQuantum Runge-Lenz Vector and the Hydrogen Atom, the hidden SO(4) symmetry
Quantum Runge-Lenz ector and the Hydrogen Atom, the hdden SO(4) symmetry Pasca Szrftgser and Edgardo S. Cheb-Terrab () Laboratore PhLAM, UMR CNRS 85, Unversté Le, F-59655, France () Mapesoft Let's consder
More informationNON-CENTRAL 7-POINT FORMULA IN THE METHOD OF LINES FOR PARABOLIC AND BURGERS' EQUATIONS
IJRRAS 8 (3 September 011 www.arpapress.com/volumes/vol8issue3/ijrras_8_3_08.pdf NON-CENTRAL 7-POINT FORMULA IN THE METHOD OF LINES FOR PARABOLIC AND BURGERS' EQUATIONS H.O. Bakodah Dept. of Mathematc
More informationCOXREG. Estimation (1)
COXREG Cox (972) frst suggested the modes n whch factors reated to fetme have a mutpcatve effect on the hazard functon. These modes are caed proportona hazards (PH) modes. Under the proportona hazards
More informationNumerical integration in more dimensions part 2. Remo Minero
Numerca ntegraton n more dmensons part Remo Mnero Outne The roe of a mappng functon n mutdmensona ntegraton Gauss approach n more dmensons and quadrature rues Crtca anass of acceptabt of a gven quadrature
More information3. Stress-strain relationships of a composite layer
OM PO I O U P U N I V I Y O F W N ompostes ourse 8-9 Unversty of wente ng. &ech... tress-stran reatonshps of a composte ayer - Laurent Warnet & emo Aerman.. tress-stran reatonshps of a composte ayer Introducton
More informationA General Column Generation Algorithm Applied to System Reliability Optimization Problems
A Genera Coumn Generaton Agorthm Apped to System Reabty Optmzaton Probems Lea Za, Davd W. Cot, Department of Industra and Systems Engneerng, Rutgers Unversty, Pscataway, J 08854, USA Abstract A genera
More informationNote 2. Ling fong Li. 1 Klein Gordon Equation Probablity interpretation Solutions to Klein-Gordon Equation... 2
Note 2 Lng fong L Contents Ken Gordon Equaton. Probabty nterpretaton......................................2 Soutons to Ken-Gordon Equaton............................... 2 2 Drac Equaton 3 2. Probabty nterpretaton.....................................
More informationBoundary Value Problems. Lecture Objectives. Ch. 27
Boundar Vaue Probes Ch. 7 Lecture Obectves o understand the dfference between an nta vaue and boundar vaue ODE o be abe to understand when and how to app the shootng ethod and FD ethod. o understand what
More informationLecture 21: Numerical methods for pricing American type derivatives
Lecture 21: Numercal methods for prcng Amercan type dervatves Xaoguang Wang STAT 598W Aprl 10th, 2014 (STAT 598W) Lecture 21 1 / 26 Outlne 1 Fnte Dfference Method Explct Method Penalty Method (STAT 598W)
More informationDeriving the Dual. Prof. Bennett Math of Data Science 1/13/06
Dervng the Dua Prof. Bennett Math of Data Scence /3/06 Outne Ntty Grtty for SVM Revew Rdge Regresson LS-SVM=KRR Dua Dervaton Bas Issue Summary Ntty Grtty Need Dua of w, b, z w 2 2 mn st. ( x w ) = C z
More informationDistributed Moving Horizon State Estimation of Nonlinear Systems. Jing Zhang
Dstrbuted Movng Horzon State Estmaton of Nonnear Systems by Jng Zhang A thess submtted n parta fufment of the requrements for the degree of Master of Scence n Chemca Engneerng Department of Chemca and
More informationNumerical Heat and Mass Transfer
Master degree n Mechancal Engneerng Numercal Heat and Mass Transfer 06-Fnte-Dfference Method (One-dmensonal, steady state heat conducton) Fausto Arpno f.arpno@uncas.t Introducton Why we use models and
More informationMMA and GCMMA two methods for nonlinear optimization
MMA and GCMMA two methods for nonlnear optmzaton Krster Svanberg Optmzaton and Systems Theory, KTH, Stockholm, Sweden. krlle@math.kth.se Ths note descrbes the algorthms used n the author s 2007 mplementatons
More informationA New Refinement of Jacobi Method for Solution of Linear System Equations AX=b
Int J Contemp Math Scences, Vol 3, 28, no 17, 819-827 A New Refnement of Jacob Method for Soluton of Lnear System Equatons AX=b F Naem Dafchah Department of Mathematcs, Faculty of Scences Unversty of Gulan,
More informationNested case-control and case-cohort studies
Outne: Nested case-contro and case-cohort studes Ørnuf Borgan Department of Mathematcs Unversty of Oso NORBIS course Unversty of Oso 4-8 December 217 1 Radaton and breast cancer data Nested case contro
More informationLectures - Week 4 Matrix norms, Conditioning, Vector Spaces, Linear Independence, Spanning sets and Basis, Null space and Range of a Matrix
Lectures - Week 4 Matrx norms, Condtonng, Vector Spaces, Lnear Independence, Spannng sets and Bass, Null space and Range of a Matrx Matrx Norms Now we turn to assocatng a number to each matrx. We could
More informationA MIN-MAX REGRET ROBUST OPTIMIZATION APPROACH FOR LARGE SCALE FULL FACTORIAL SCENARIO DESIGN OF DATA UNCERTAINTY
A MIN-MAX REGRET ROBST OPTIMIZATION APPROACH FOR ARGE SCAE F FACTORIA SCENARIO DESIGN OF DATA NCERTAINTY Travat Assavapokee Department of Industra Engneerng, nversty of Houston, Houston, Texas 7704-4008,
More informationAppendix B. The Finite Difference Scheme
140 APPENDIXES Appendx B. The Fnte Dfference Scheme In ths appendx we present numercal technques whch are used to approxmate solutons of system 3.1 3.3. A comprehensve treatment of theoretcal and mplementaton
More informationMore metrics on cartesian products
More metrcs on cartesan products If (X, d ) are metrc spaces for 1 n, then n Secton II4 of the lecture notes we defned three metrcs on X whose underlyng topologes are the product topology The purpose of
More informationLecture Notes on Linear Regression
Lecture Notes on Lnear Regresson Feng L fl@sdueducn Shandong Unversty, Chna Lnear Regresson Problem In regresson problem, we am at predct a contnuous target value gven an nput feature vector We assume
More informationResearch Article H Estimates for Discrete-Time Markovian Jump Linear Systems
Mathematca Probems n Engneerng Voume 213 Artce ID 945342 7 pages http://dxdoorg/11155/213/945342 Research Artce H Estmates for Dscrete-Tme Markovan Jump Lnear Systems Marco H Terra 1 Gdson Jesus 2 and
More informationAPPENDIX A Some Linear Algebra
APPENDIX A Some Lnear Algebra The collecton of m, n matrces A.1 Matrces a 1,1,..., a 1,n A = a m,1,..., a m,n wth real elements a,j s denoted by R m,n. If n = 1 then A s called a column vector. Smlarly,
More informationNumerical Investigation of Power Tunability in Two-Section QD Superluminescent Diodes
Numerca Investgaton of Power Tunabty n Two-Secton QD Superumnescent Dodes Matta Rossett Paoo Bardea Ivo Montrosset POLITECNICO DI TORINO DELEN Summary 1. A smpfed mode for QD Super Lumnescent Dodes (SLD)
More information[WAVES] 1. Waves and wave forces. Definition of waves
1. Waves and forces Defnton of s In the smuatons on ong-crested s are consdered. The drecton of these s (μ) s defned as sketched beow n the goba co-ordnate sstem: North West East South The eevaton can
More informationStanford University CS359G: Graph Partitioning and Expanders Handout 4 Luca Trevisan January 13, 2011
Stanford Unversty CS359G: Graph Parttonng and Expanders Handout 4 Luca Trevsan January 3, 0 Lecture 4 In whch we prove the dffcult drecton of Cheeger s nequalty. As n the past lectures, consder an undrected
More informationXin Li Department of Information Systems, College of Business, City University of Hong Kong, Hong Kong, CHINA
RESEARCH ARTICLE MOELING FIXE OS BETTING FOR FUTURE EVENT PREICTION Weyun Chen eartment of Educatona Informaton Technoogy, Facuty of Educaton, East Chna Norma Unversty, Shangha, CHINA {weyun.chen@qq.com}
More informationA generalization of Picard-Lindelof theorem/ the method of characteristics to systems of PDE
A generazaton of Pcard-Lndeof theorem/ the method of characterstcs to systems of PDE Erfan Shachan b,a,1 a Department of Physcs, Unversty of Toronto, 60 St. George St., Toronto ON M5S 1A7, Canada b Department
More informationHomogenization of reaction-diffusion processes in a two-component porous medium with a non-linear flux-condition on the interface
Homogenzaton of reacton-dffuson processes n a two-component porous medum wth a non-lnear flux-condton on the nterface Internatonal Conference on Numercal and Mathematcal Modelng of Flow and Transport n
More informationChapter 12. Ordinary Differential Equation Boundary Value (BV) Problems
Chapter. Ordnar Dfferental Equaton Boundar Value (BV) Problems In ths chapter we wll learn how to solve ODE boundar value problem. BV ODE s usuall gven wth x beng the ndependent space varable. p( x) q(
More informationAndre Schneider P622
Andre Schneder P6 Probem Set #0 March, 00 Srednc 7. Suppose that we have a theory wth Negectng the hgher order terms, show that Souton Knowng β(α and γ m (α we can wrte β(α =b α O(α 3 (. γ m (α =c α O(α
More information3.1 Expectation of Functions of Several Random Variables. )' be a k-dimensional discrete or continuous random vector, with joint PMF p (, E X E X1 E X
Statstcs 1: Probablty Theory II 37 3 EPECTATION OF SEVERAL RANDOM VARIABLES As n Probablty Theory I, the nterest n most stuatons les not on the actual dstrbuton of a random vector, but rather on a number
More informationA particle in a state of uniform motion remain in that state of motion unless acted upon by external force.
The fundamental prncples of classcal mechancs were lad down by Galleo and Newton n the 16th and 17th centures. In 1686, Newton wrote the Prncpa where he gave us three laws of moton, one law of gravty,
More informationarxiv: v1 [physics.comp-ph] 17 Dec 2018
Pressures nsde a nano-porous medum. The case of a snge phase fud arxv:1812.06656v1 [physcs.comp-ph] 17 Dec 2018 Oav Gateand, Dck Bedeaux, and Sgne Kjestrup PoreLab, Department of Chemstry, Norwegan Unversty
More informationSome Comments on Accelerating Convergence of Iterative Sequences Using Direct Inversion of the Iterative Subspace (DIIS)
Some Comments on Acceleratng Convergence of Iteratve Sequences Usng Drect Inverson of the Iteratve Subspace (DIIS) C. Davd Sherrll School of Chemstry and Bochemstry Georga Insttute of Technology May 1998
More informationA DIMENSION-REDUCTION METHOD FOR STOCHASTIC ANALYSIS SECOND-MOMENT ANALYSIS
A DIMESIO-REDUCTIO METHOD FOR STOCHASTIC AALYSIS SECOD-MOMET AALYSIS S. Rahman Department of Mechanca Engneerng and Center for Computer-Aded Desgn The Unversty of Iowa Iowa Cty, IA 52245 June 2003 OUTLIE
More informationErrors for Linear Systems
Errors for Lnear Systems When we solve a lnear system Ax b we often do not know A and b exactly, but have only approxmatons  and ˆb avalable. Then the best thng we can do s to solve ˆx ˆb exactly whch
More informationExample: Suppose we want to build a classifier that recognizes WebPages of graduate students.
Exampe: Suppose we want to bud a cassfer that recognzes WebPages of graduate students. How can we fnd tranng data? We can browse the web and coect a sampe of WebPages of graduate students of varous unverstes.
More informationApproximate merging of a pair of BeÂzier curves
COMPUTER-AIDED DESIGN Computer-Aded Desgn 33 (1) 15±136 www.esever.com/ocate/cad Approxmate mergng of a par of BeÂzer curves Sh-Mn Hu a,b, *, Rou-Feng Tong c, Tao Ju a,b, Ja-Guang Sun a,b a Natona CAD
More informationA Derivative-Free Algorithm for Bound Constrained Optimization
Computatona Optmzaton and Appcatons, 21, 119 142, 2002 c 2002 Kuwer Academc Pubshers. Manufactured n The Netherands. A Dervatve-Free Agorthm for Bound Constraned Optmzaton STEFANO LUCIDI ucd@ds.unroma.t
More informationEXPERIMENT AND THEORISATION: AN APPLICATION OF THE HYDROSTATIC EQUATION AND ARCHIMEDES THEOREM
EXPERIMENT AND THEORISATION: AN APPLICATION OF THE HYDROSTATIC EQUATION AND ARCHIMEDES THEOREM Santos Lucía, Taaa Máro, Departamento de Físca, Unversdade de Avero, Avero, Portuga 1. Introducton Today s
More information3 Basic boundary value problems for analytic function in the upper half plane
3 Basc boundary value problems for analytc functon n the upper half plane 3. Posson representaton formulas for the half plane Let f be an analytc functon of z throughout the half plane Imz > 0, contnuous
More informationExercise Solutions to Real Analysis
xercse Solutons to Real Analyss Note: References refer to H. L. Royden, Real Analyss xersze 1. Gven any set A any ɛ > 0, there s an open set O such that A O m O m A + ɛ. Soluton 1. If m A =, then there
More informationModule 9. Lecture 6. Duality in Assignment Problems
Module 9 1 Lecture 6 Dualty n Assgnment Problems In ths lecture we attempt to answer few other mportant questons posed n earler lecture for (AP) and see how some of them can be explaned through the concept
More informationCOMPARISON OF SOME RELIABILITY CHARACTERISTICS BETWEEN REDUNDANT SYSTEMS REQUIRING SUPPORTING UNITS FOR THEIR OPERATIONS
Avalable onlne at http://sck.org J. Math. Comput. Sc. 3 (3), No., 6-3 ISSN: 97-537 COMPARISON OF SOME RELIABILITY CHARACTERISTICS BETWEEN REDUNDANT SYSTEMS REQUIRING SUPPORTING UNITS FOR THEIR OPERATIONS
More informationJournal of Multivariate Analysis
Journa of Mutvarate Anayss 3 (04) 74 96 Contents sts avaabe at ScenceDrect Journa of Mutvarate Anayss journa homepage: www.esever.com/ocate/jmva Hgh-dmensona sparse MANOVA T. Tony Ca a, Yn Xa b, a Department
More informationStructure and Drive Paul A. Jensen Copyright July 20, 2003
Structure and Drve Paul A. Jensen Copyrght July 20, 2003 A system s made up of several operatons wth flow passng between them. The structure of the system descrbes the flow paths from nputs to outputs.
More informationInner Product. Euclidean Space. Orthonormal Basis. Orthogonal
Inner Product Defnton 1 () A Eucldean space s a fnte-dmensonal vector space over the reals R, wth an nner product,. Defnton 2 (Inner Product) An nner product, on a real vector space X s a symmetrc, blnear,
More informationMonica Purcaru and Nicoleta Aldea. Abstract
FILOMAT (Nš) 16 (22), 7 17 GENERAL CONFORMAL ALMOST SYMPLECTIC N-LINEAR CONNECTIONS IN THE BUNDLE OF ACCELERATIONS Monca Purcaru and Ncoeta Adea Abstract The am of ths paper 1 s to fnd the transformaton
More informationA General Distributed Dual Coordinate Optimization Framework for Regularized Loss Minimization
Journa of Machne Learnng Research 18 17 1-5 Submtted 9/16; Revsed 1/17; Pubshed 1/17 A Genera Dstrbuted Dua Coordnate Optmzaton Framework for Reguarzed Loss Mnmzaton Shun Zheng Insttute for Interdscpnary
More informationPolite Water-filling for Weighted Sum-rate Maximization in MIMO B-MAC Networks under. Multiple Linear Constraints
2011 IEEE Internatona Symposum on Informaton Theory Proceedngs Pote Water-fng for Weghted Sum-rate Maxmzaton n MIMO B-MAC Networks under Mutpe near Constrants An u 1, Youjan u 2, Vncent K. N. au 3, Hage
More informationAdditive Schwarz Method for DG Discretization of Anisotropic Elliptic Problems
Addtve Schwarz Method for DG Dscretzaton of Ansotropc Ellptc Problems Maksymlan Dryja 1, Potr Krzyżanowsk 1, and Marcus Sarks 2 1 Introducton In the paper we consder a second order ellptc problem wth dscontnuous
More informationn-step cycle inequalities: facets for continuous n-mixing set and strong cuts for multi-module capacitated lot-sizing problem
n-step cyce nequates: facets for contnuous n-mxng set and strong cuts for mut-modue capactated ot-szng probem Mansh Bansa and Kavash Kanfar Department of Industra and Systems Engneerng, Texas A&M Unversty,
More informationDelay tomography for large scale networks
Deay tomography for arge scae networks MENG-FU SHIH ALFRED O. HERO III Communcatons and Sgna Processng Laboratory Eectrca Engneerng and Computer Scence Department Unversty of Mchgan, 30 Bea. Ave., Ann
More informationStrain Energy in Linear Elastic Solids
Duke Unverst Department of Cv and Envronmenta Engneerng CEE 41L. Matr Structura Anass Fa, Henr P. Gavn Stran Energ n Lnear Eastc Sods Consder a force, F, apped gradua to a structure. Let D be the resutng
More informationInterference Alignment and Degrees of Freedom Region of Cellular Sigma Channel
2011 IEEE Internatona Symposum on Informaton Theory Proceedngs Interference Agnment and Degrees of Freedom Regon of Ceuar Sgma Channe Huaru Yn 1 Le Ke 2 Zhengdao Wang 2 1 WINLAB Dept of EEIS Unv. of Sc.
More informationTHE WEIGHTED WEAK TYPE INEQUALITY FOR THE STRONG MAXIMAL FUNCTION
THE WEIGHTED WEAK TYPE INEQUALITY FO THE STONG MAXIMAL FUNCTION THEMIS MITSIS Abstract. We prove the natural Fefferman-Sten weak type nequalty for the strong maxmal functon n the plane, under the assumpton
More informationThe Application of BP Neural Network principal component analysis in the Forecasting the Road Traffic Accident
ICTCT Extra Workshop, Bejng Proceedngs The Appcaton of BP Neura Network prncpa component anayss n Forecastng Road Traffc Accdent He Mng, GuoXucheng &LuGuangmng Transportaton Coege of Souast Unversty 07
More informationKernel Methods and SVMs Extension
Kernel Methods and SVMs Extenson The purpose of ths document s to revew materal covered n Machne Learnng 1 Supervsed Learnng regardng support vector machnes (SVMs). Ths document also provdes a general
More informationON A DETERMINATION OF THE INITIAL FUNCTIONS FROM THE OBSERVED VALUES OF THE BOUNDARY FUNCTIONS FOR THE SECOND-ORDER HYPERBOLIC EQUATION
Advanced Mathematcal Models & Applcatons Vol.3, No.3, 2018, pp.215-222 ON A DETERMINATION OF THE INITIAL FUNCTIONS FROM THE OBSERVED VALUES OF THE BOUNDARY FUNCTIONS FOR THE SECOND-ORDER HYPERBOLIC EUATION
More information2E Pattern Recognition Solutions to Introduction to Pattern Recognition, Chapter 2: Bayesian pattern classification
E395 - Pattern Recognton Solutons to Introducton to Pattern Recognton, Chapter : Bayesan pattern classfcaton Preface Ths document s a soluton manual for selected exercses from Introducton to Pattern Recognton
More information1 Matrix representations of canonical matrices
1 Matrx representatons of canoncal matrces 2-d rotaton around the orgn: ( ) cos θ sn θ R 0 = sn θ cos θ 3-d rotaton around the x-axs: R x = 1 0 0 0 cos θ sn θ 0 sn θ cos θ 3-d rotaton around the y-axs:
More informationLOW-DENSITY Parity-Check (LDPC) codes have received
IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 59, NO. 7, JULY 2011 1807 Successve Maxmzaton for Systematc Desgn of Unversay Capacty Approachng Rate-Compatbe Sequences of LDPC Code Ensembes over Bnary-Input
More informationCHAPTER 5 NUMERICAL EVALUATION OF DYNAMIC RESPONSE
CHAPTER 5 NUMERICAL EVALUATION OF DYNAMIC RESPONSE Analytcal soluton s usually not possble when exctaton vares arbtrarly wth tme or f the system s nonlnear. Such problems can be solved by numercal tmesteppng
More informationAppendix B. Criterion of Riemann-Stieltjes Integrability
Appendx B. Crteron of Remann-Steltes Integrablty Ths note s complementary to [R, Ch. 6] and [T, Sec. 3.5]. The man result of ths note s Theorem B.3, whch provdes the necessary and suffcent condtons for
More informationWeek3, Chapter 4. Position and Displacement. Motion in Two Dimensions. Instantaneous Velocity. Average Velocity
Week3, Chapter 4 Moton n Two Dmensons Lecture Quz A partcle confned to moton along the x axs moves wth constant acceleraton from x =.0 m to x = 8.0 m durng a 1-s tme nterval. The velocty of the partcle
More information4 Analysis of Variance (ANOVA) 5 ANOVA. 5.1 Introduction. 5.2 Fixed Effects ANOVA
4 Analyss of Varance (ANOVA) 5 ANOVA 51 Introducton ANOVA ANOVA s a way to estmate and test the means of multple populatons We wll start wth one-way ANOVA If the populatons ncluded n the study are selected
More informationThe Order Relation and Trace Inequalities for. Hermitian Operators
Internatonal Mathematcal Forum, Vol 3, 08, no, 507-57 HIKARI Ltd, wwwm-hkarcom https://doorg/0988/mf088055 The Order Relaton and Trace Inequaltes for Hermtan Operators Y Huang School of Informaton Scence
More informationSPATIAL KINEMATICS OF GEARS IN ABSOLUTE COORDINATES
SATIAL KINEMATICS OF GEARS IN ABSOLUTE COORDINATES Dmtry Vasenko and Roand Kasper Insttute of Mobe Systems (IMS) Otto-von-Guercke-Unversty Magdeburg D-39016, Magdeburg, Germany E-ma: Dmtr.Vasenko@ovgu.de
More informationLecture 5.8 Flux Vector Splitting
Lecture 5.8 Flux Vector Splttng 1 Flux Vector Splttng The vector E n (5.7.) can be rewrtten as E = AU (5.8.1) (wth A as gven n (5.7.4) or (5.7.6) ) whenever, the equaton of state s of the separable form
More informationDECOUPLING THEORY HW2
8.8 DECOUPLIG THEORY HW2 DOGHAO WAG DATE:OCT. 3 207 Problem We shall start by reformulatng the problem. Denote by δ S n the delta functon that s evenly dstrbuted at the n ) dmensonal unt sphere. As a temporal
More informationLECTURE 21 Mohr s Method for Calculation of General Displacements. 1 The Reciprocal Theorem
V. DEMENKO MECHANICS OF MATERIALS 05 LECTURE Mohr s Method for Cacuaton of Genera Dspacements The Recproca Theorem The recproca theorem s one of the genera theorems of strength of materas. It foows drect
More informationLINEAR REGRESSION ANALYSIS. MODULE IX Lecture Multicollinearity
LINEAR REGRESSION ANALYSIS MODULE IX Lecture - 31 Multcollnearty Dr. Shalabh Department of Mathematcs and Statstcs Indan Insttute of Technology Kanpur 6. Rdge regresson The OLSE s the best lnear unbased
More informationCOMBINING SPATIAL COMPONENTS IN SEISMIC DESIGN
Transactons, SMRT- COMBINING SPATIAL COMPONENTS IN SEISMIC DESIGN Mchae O Leary, PhD, PE and Kevn Huberty, PE, SE Nucear Power Technooges Dvson, Sargent & Lundy, Chcago, IL 6060 ABSTRACT Accordng to Reguatory
More informationG : Statistical Mechanics
G25.2651: Statstca Mechancs Notes for Lecture 11 I. PRINCIPLES OF QUANTUM STATISTICAL MECHANICS The probem of quantum statstca mechancs s the quantum mechanca treatment of an N-partce system. Suppose the
More informationAnalysis of Bipartite Graph Codes on the Binary Erasure Channel
Anayss of Bpartte Graph Codes on the Bnary Erasure Channe Arya Mazumdar Department of ECE Unversty of Maryand, Coege Par ema: arya@umdedu Abstract We derve densty evouton equatons for codes on bpartte
More informationGlobal Sensitivity. Tuesday 20 th February, 2018
Global Senstvty Tuesday 2 th February, 28 ) Local Senstvty Most senstvty analyses [] are based on local estmates of senstvty, typcally by expandng the response n a Taylor seres about some specfc values
More information