An introduction to Hybrid High-Order methods Linear elasticity Daniele A. Di Pietro I3M, University of Montpellier A.A. 2014–2015 1 / 56 Outline 1 Continuous setting 2 Discrete problem and stability 3 Error analysis 4 Numerical examples 2 / 56 Outline 1 Continuous setting 2 Discrete problem and stability 3 Error analysis 4 Numerical examples 3 / 56 Continuous setting I Let Ω Ă Rd , d ě 2, be a bounded connected polygonal domain The linear elasticity problem reads: Find u : Ω Ñ Rd s.t. ´∇¨σpuq “ f u“0 in Ω, on BΩ where, for real Lam´e parameters 0 ď λ ď `8 and 0 ă µ ă `8, σpuq “ 2µ∇s u ` λp∇¨uqI d Here, σpuq and ∇s u are, respectively, the stress and strain tensors 4 / 56 Continuous setting II Assume f P L2 pΩqd and set U :“ H01 pΩqd The weak formulation reads: Find u P U s.t. apu, vq “ pf , vq @v P U , with bilinear form apu, vq :“ 2µp∇s u, ∇s vq ` λp∇¨u, ∇¨vq Equivalently, we have the variational formulation " * 1 inf Jen pvq :“ apv, vq ´ pf , vq vPU 2 5 / 56 Continuous setting III Lemma (Korn’s first inequality) For all functions v P H01 pΩqd , it holds that }∇v}2 ď 2}∇s v}2 . Let v P Cc8 pΩqd . We have ż ż |∇s v|2 “ 2 Ω ż ż |∇v|2 ` Ω ∇v¨∇v T “ Ω ż |∇v|2 ` Ω |∇¨v|2 ě }∇v}2 Ω The conclusion follows from the density of Cc8 pΩqd in H01 pΩqd 6 / 56 Continuous setting IV Combining Korn’s and Poincar´e’s inequalities, we obtain }v} À }∇s v}, and well-posedness follows from Lax–Milgram’s Lemma Similar results can be proved for more general boundary conditions The mechanical engineering terminology for boundary conditions is Dirichlet Ð displacement Neumann Ð traction 7 / 56 Quasi-incompressible limit I Lemma (A priori estimate) Let Ω Ă R2 be a bounded convex polygonal domain. Then, there is CΩ ą 0 only depending on Ω s.t. }u}H 2 pΩqd ` }λ∇¨u}H 1 pΩq ď CΩ }f }L2 pΩqd . This result is basically linked to Cattabriga’s regularity for Stokes 8 / 56 Quasi-incompressible limit II Let us introduce the pressure p :“ λ∇¨u We can reformulate the linear elasticity problem as 2µp∇s u, ∇s vq ` p∇¨v, pq “ pf , vq p∇¨u, qq “ λ´1 pp, qq @v P H01 pΩqd , @q P L20 pΩq For λ Ñ `8, u thus solves a Stokes problem since ∇¨u Ñ 0 Numerically, we need to approximate non-trivial solenoidal fields 9 / 56 Outline 1 Continuous setting 2 Discrete problem and stability 3 Error analysis 4 Numerical examples 10 / 56 DOFs and reduction map I •• •• •• •• •• •• •• •••• •• • •• • •• •• •• •• •• •• •• •• •• •• •• •• •• •• •• •• •• •••• •• •• •• •• •• •• •• •• •• •• •• •• k=2 •• k=1 • •• • k=0 Figure: U kT for k P t0, 1, 2u For all k ě 0 and all T P Th , we define the local space of DOFs # U kT :“ Pkd pT qd + ą ˆ Pkd´1 pF qd F PFT The global space is obtained enforcing single-valuedness at interfaces + # U kh :“ ą T PTh Pkd pT qd # + ą ˆ Pkd´1 pF qd F PFh 11 / 56 DOFs and reduction map II For a collection of DOFs v h P U kh , we use the underlined notation ˘ ` v h “ pv T qT PTh , pv F qF PFh For all T P Th , v T P U kT denotes its restriction to U kT s.t. ˘ ` v T “ v T , pv F qF PFT 12 / 56 DOFs and reduction map III We define the local reduction map I kT : H 1 pT qd Ñ U kT s.t. ` ˘ I kT : v ÞÑ πTk v, pπFk vqF PFT Finally, the global reduction map I kh : H 1 pΩqd Ñ U kh is s.t. pI kh vq|T “ I kT pv |T q @T P Th 13 / 56 Rigid body motions Applied to vector fields, the operator ∇s yields strains Consistently, the kernel RMpΩq of ∇s contains rigid-body motions In d “ 3, RMpΩq has dimension 6 and one can prove that ( RMpΩq :“ v P H 1 pΩq3 | Dα, ω P R3 , vpxq “ α ` ω b x , where we recall that i ω b x “ ω1 x1 j ω2 x2 k ω3 “ pω2 x3 ´ x2 ω3 qi ` pω3 x1 ´ ω1 x3 qj ` pω1 x2 ´ ω2 x1 qk x3 The space of planar rigid-body motions has dimension 3 14 / 56 Displacement reconstruction I Let T P Th . The local displacement reconstruction operator pk`1 : U kT Ñ Pk`1 pT qd T d ` ˘ is s.t., for all v T “ v T , pv F qF PFT P U kT and w P Pk`1 pT qd , d p∇s pk`1 T v T , ∇s wqT “ ´pv T , ∇¨∇s wqT ` ÿ pv F , ∇s wnT F qF , F PFT Rigid-body motions are prescribed from v T setting ż ż pk`1 vT “ T T ż ∇ss pk`1 vT “ T vT , T T ÿ ż 1 pnT F bv F ´v F bnT F q F 2 F PF T 15 / 56 Displacement reconstruction II k Lemma (Approximation properties for pk`1 T IT ) There exists C ą 0 independent of hT s.t., for all v P H k`2 pT qd , k`1 k }v ´ pk`1 I kT v}T ` hT }∇pv ´ pT I T vq}T T 1{2 3{2 k`1 k ` hT }v ´ pk`1 I kT v}BT ` hT }∇pv ´ pT I T vq}BT ď Chk`2 }v}H k`2 pT qd . T T 16 / 56 Displacement reconstruction III pT qd , By definition of pTk`1 and of I kT , we have, for all w P Pk`1 d ÿ p∇s pk`1 I kT v, ∇s wqT “ ´pπTk v, ∇¨∇s wqT ` pπFk v, ∇s wnT F qF T F PFT “ ´pv, ∇¨∇s wqT ` ÿ pv, ∇s wnT F qF F PFT Integrating by parts the right-hand side yields p∇s pTk`1 I kT v ´ ∇s v, ∇s wqT “ 0 @w P Pk`1 pT qd d This orthogonality (Euler) condition implies that k }∇s ppk`1 T I T v´vq}T “ inf wPPk`1 pT qd d }∇s pw´vq}T À hk`1 T }v}H k`2 pT qd 17 / 56 Displacement reconstruction IV The closure conditions for the local problem yield for the translations ż ż ż k`1 k k πT v “ v pT I T v “ T T T Similarly, for the rotations we have ż ÿ ż 1 ∇ss ppTk`1 I kT vq “ pnT F b πFk v ´ πFk v b nT F q 2 T F F PFT ż ÿ ż 1 pnT F b v ´ v b nT F q “ “ ∇ss v F 2 T F PF T k Hence, pk`1 T I T v and v have the same rigid-body motions 18 / 56 Displacement reconstruction V Using the local Poincar´e’s and Korn’s inequalities, we have }pTk`1 I kT v ´ v}T ` hT }∇ppk`1 I kT v ´ vq}T À hT }∇s ppk`1 I kT v ´ vq}T T T À hk`1 }v}H k`2 pT qd T Boundary terms can be estimated as for Poisson 19 / 56 Stabilization I Define, for T P Th , the stabilization bilinear form sT as ÿ k k sT puT , v T q :“ pk`1 h´1 pk`1 T v T ´ v F qqF , T uT ´ uF q, πF pp F pπF pp F PFT pk`1 pT qd s.t. with displacement reconstruction p : U kT Ñ Pk`1 T d @v T P U kT , k`1 k k`1 pk`1 p T v T :“ v T ` ppT v T ´ πT pT v T q We next investigate the stability and consistency properties of sT 20 / 56 Stabilization II Lemma (Stabilization) Assume k ě 1. There is η ą 0 independent of h, µ, and λ such that, for all T P Th and all v T P U kT , the following stability property holds: 2 ´1 η}v T }2ε,T ď }∇s pk`1 }v T }2ε,T , T v T }T ` sT pv T , v T q ď η where the discrete strain norm }¨}ε,T is s.t. ÿ 2 }v T }2ε,T :“ }∇s v T }2T ` h´1 F }v F }F . F PFT Moreover, for all v P H k`2 pT qd , we have the approximation property sT pI kT v, I kT vq {2 À hk`1 T }v}H k`2 pT qd . 1 21 / 56 Stabilization III We only detail the proof of stability (boundedness is similar) 2 }v T }2ε,T À }∇s pk`1 T v T }T ` sT pv T , v T q Taking w “ v T in the definition of pk`1 T , we infer that ÿ }∇s v T }2T “ p∇s pk`1 pv T ´ v F , ∇s v T nT F qF T v T , ∇s v T qT ` F PFT 1 2 2 2 2 ď }∇s pk`1 T v T }T ` }∇s v T }T ` NB Ctr |v T |ε,BT 2 As a result, we have 2 2 }∇s v T }2T À }∇s pk`1 T v T }T ` |v T |ε,BT 22 / 56 Stabilization IV Additionally, for all F P FT , ´1{2 hF }v F ´ v T }F ´1{2 p k`1 }v F ´ πFk p v T }F ` hF T ´1{2 p k`1 }πFk pv F ´ p v T q}F ` hF T ´1{2 k`1 p Tk`1 v T q}F ` Ctr h´1 }πFk pv F ´ p v T ´ πTk pk`1 v T }T F }pT T ď hF “ hF “ hF ´1{2 p k`1 }πFk p v T ´ v T }F T ´1{2 }πFk pp pk`1 v T ´ v T q}F T For any function w P H 1 pT qd , writing "ż wRM “ |T |´1 d T * "ż * w ` |T |´1 ∇ w px ´ xT q, ss d T we observe that πTk wRM “ wRM since k ě 1, whence we get }w´πTk w}T “ }pw´wRM q´πTk pw´wRM q}T ď }w´wRM }T À hT }∇s w}T 23 / 56 Stabilization V Applying this inequality to the rightmost term we infer k`1 k k`1 ´2 Ctr h´1 Ctr CK }∇s pk`1 F }pT v T ´ πT pT v T }T ď % T v T }T , so that ´1{2 hF ´1{2 }v F ´ v T }F À hF k`1 pk`1 }πFk pv F ´ p T v T q}F ` }∇s pT v T }T Squaring and summing over F P FT leads to 2 |v T |2ε,BT À sT pv T , v T q ` NB }∇s pk`1 T v T }T The first inequality follows recalling that we had proven 2 2 }∇s v T }2T À }∇s pk`1 T v T }T ` |v T |ε,BT 24 / 56 Stabilization VI Let now v P H k`2 pT qd . For all T P Th and all F P FT we have ´1{2 hF k k }πFk pp pk`1 T I T v ´ πF vq}F ´1{2 k }p pk`1 T I T v ´ v}F ´1{2 k k`1 k k }ppk`1 T I T v ´ vq ´ πT ppT I T v ´ vq}F ´1{2 k ´1 k`1 k }pk`1 T I T v ´ v}F ` Ctr hF }pT I T v ´ v}T ď hF “ hF À hF À hk`1 T }v}H k`2 pT qd Using the approximation properties of pk`1 the conclusion follows T 25 / 56 Divergence reconstruction We define the local local discrete divergence operator DTk : U kT Ñ Pkd pT q ` ˘ s.t., for all v T “ v T , pv F qF PFT P U kT and all q P Pkd pT q, pDTk v T , qqT :“ ´pv T , ∇qqT ` ÿ pv F ¨nT F , qqF F PFT By construction, we have the following commuting diagram: U pT q ∇¨ I kT U kT L2 pT q πTk DTk Pkd pT q 26 / 56 Discrete problem We define the local bilinear form aT on U kT ˆ U kT as ( k`1 aT puT , v T q :“ 2µ p∇s pk`1 T uT , ∇s pT v T qT ` sT puT , v T q ` λpDTk uT , DTk v T q The discrete problem reads: Find uh P U kh,0 s.t. ah puh , v h q :“ ÿ T PTh aT puT , v T q “ ÿ pf , v T qT @v h P U kh,0 T PTh where displacement boundary conditions are enforced setting ! ) ` ˘ U kh,0 :“ v h “ pv T qT PTh , pv F qF PFh P U kh | v F ” 0 @F P Fhb 27 / 56 Outline 1 Continuous setting 2 Discrete problem and stability 3 Error analysis 4 Numerical examples 28 / 56 Energy error analysis I p h , where u p h P U kh,0 is s.t. We bound the error uh ´ u p h “ ppπTk uqT PTh , pπFk uqF PFh q u We measure the error in the energy norm s.t., for all v h P U kh , }v h }2en,h :“ ah pv h , v h q Using stability, this will also yield a bound in the strain norm since p2µηq}v h }2ε,h ď }v h }2en,h 29 / 56 Energy error analysis II Theorem (Convergence) Assume k ě 1 and the additional regularity u P H k`2 pTh qd and ∇¨u P H k`1 pTh q. Then, there exists C ą 0 independent of h, µ, and λ s.t. ˘ ` 1 p h }en,h ď Chk`1 looooooooooooooooooooooooomooooooooooooooooooooooooon p2µq {2 }uh ´ u p2µq}u}H k`2 pTh qd ` λ}∇¨u}H k`1 pTh q . Bpu,kq 30 / 56 Energy error analysis III Under usual regularity for Ω, the above estimate is locking-free For d “ 2 and Ω convex, it holds (cf. [Brenner and Sung, 1992]) Bpu, 0q “ µ}u}H 2 pΩqd ` λ}∇¨u}H 1 pΩq ď Cµ }f }, with Cµ ą 0 depending on Ω and µ but independent of λ More generally, for k ě 1, we need the regularity shift Bpu, kq ď Cµ }f }H k pΩqd 31 / 56 Energy error analysis IV For all v h P U kh,0 , we observe that # }v h }2en,h “ ah pv h , v h q ď sup wh PU k h,0 ah pv h , wh q }wh }ε,h + ˆ }v h }ε,h Thus, using stability, we have that 1 p2µηq {2 }v h }en,h ď sup wh PU k h,0 ,}w h }ε,h “1 ah pv h , wh q 32 / 56 Energy error analysis V p h q and using the discrete problem yields Making above v h “ puh ´ u 1 p h }en,h ď p2µηq {2 }uh ´ u sup v h PU k h,0 ,}w h }ε,h “1 Eh pwh q, with consistency error uh , wh q Eh pwh q :“ lh pwh q ´ ah pp We next bound Eh pwh q for a generic wh P U kh,0 s.t. }wh }ε,h “ 1 33 / 56 Energy error analysis VI Let an element T P Th be fixed q T :“ pk`1 p T and using the definition of pk`1 Setting u T , we infer T u p T , ∇s pk`1 q T , ∇s wT qT ` p∇s pk`1 u wT qT “ p∇s u T T ÿ q T n T F , w F ´ w T qF p∇s u F PFT “ p∇s u, ∇s wT qT ` ÿ q T n T F , w F ´ w T qF p∇s u F PFT Similarly, using the definition of DTk and the commuting property, p T , DTk wT qT “ pπTk p∇¨uq, DTk wT qT pDTk u ÿ “ p∇¨u, ∇¨wT qT ` pπTk p∇¨uq, pwF ´ wT q¨nT F qF F PFT 34 / 56 Energy error analysis VII Replacing the above expressions we obtain aT pp uT , wT q “ pσ, ∇s wT qT ` p2µqsT pp uT , wT q ÿ q T ` λI d πTk p∇¨uq, wF ´ wT qF ` p2µ∇s u F PFT Using f “ ´∇¨σ and integrating by parts element-wise, we infer # + ÿ ÿ lh pwh q “ pσ, ∇s wT qT ´ pσnT F , wT ´ wF qF T PTh F PFT 35 / 56 Energy error analysis VIII We thus rewrite the consistency error Eh pwh q as Eh pwh q “ ÿ ÿ q T qnT F , wF ´ wT qF p∇s pu ´ u p2µq T PTh F PFT looooooooooooooooooooooooooooooomooooooooooooooooooooooooooooooon T1 ÿ πTk p∇¨uqqnT F , wF ` λpp∇¨u ´ ´ wT qF ´ p2µqsh pp uh , wh q F PFT looooooooooooooooooooooooooooooomooooooooooooooooooooooooooooooon looooooooooomooooooooooon T2 T3 36 / 56 Energy error analysis IX k Using the approximation properties of pk`1 T I T , we have |T1 | À p2µqhk`1 }u}H k`2 pTh qd }wh }ε,h Similarly, we have for T2 using the approximation properties of πTk , |T2 | À λhk`1 }∇¨u}H k`1 pTh q }wh }ε,h To estimate T3 , we use the consistency of sh to infer p h q {2 sh pwh , wh q {2 À p2µqhk`1 }u}H k`2 pTh qd }wh }ε,h |T3 | ď p2µqsh pp uh , u 1 1 Using these bounds to estimate Eh pwh q the conclusion follows 37 / 56 L2 -error displacement for the potential I Assumption (Elliptic regularity) For all g P L2 pΩqd , the unique solution of ´∇¨ς “ g in Ω, ς “ 2µ∇s z ` λp∇¨zqI d in Ω, z“0 on BΩ, satisfies the a priori estimate p2µq}z}H 2 pΩqd ` λ}∇¨z}H 1 pΩq ď Cµ }g}. 38 / 56 L2 -error displacement for the potential II Theorem (L2 -error estimate for the displacement) Let eh P Pkd pTh qd be s.t. eh|T :“ uT ´ πTk u @T P Th . Then, elliptic regularity for Ω and provided that u P H k`2 pTh qd and ∇¨u P H k`1 pTh q, it holds with Cą0 depending on Ω, µ, and % but independent of λ and h, }eh } ď Chk`2 Bpu, kq. 39 / 56 L2 -error displacement for the potential III p h P U kh,0 the energy error estimate yields With eh :“ uh ´ u }eh }ε,h ` sh peh , eh q {2 À hk`1 Bpu, kq 1 Consider the auxiliary problem with g “ eh and solution z and ς Integrating by parts element-wise and since eh|T “ eT , we infer that }eh }2 “ ´ ÿ peT , ∇¨ς qT T PTh + # ÿ “ T PTh p∇s eT , ς qT ` ÿ peF ´ eT , ς nT F qF F PFT 40 / 56 L2 -error displacement for the potential IV Let ph :“ ppπTk zqT PTh , pπFk zqF PFh q P U kh,0 z We have ph q “ ah puh , z ph q´ah pp ph q “ lh pp ph q “ Eh pp ah peh , z uh , z z h q´ah pp uh , z zhq Therefore, we can decompose }eh }2 as follows: $ , . ÿ & ÿ p h q ` Eh pp }eh }2 “ p∇s eT , ς qT ` peF ´ eT , ς nT F qF ´ ah peh , z zh q % T PTh F PFT looooooooooooooooooooooooooooooooooooooooooomooooooooooooooooooooooooooooooooooooooooooon looooomooooon T1 T2 41 / 56 L2 -error displacement for the potential V For all T P Th , we have, letting S kT :“ p2µq∇s pk`1 ` λI d DTk , T # ÿ ph q “ ah peh , z + p∇s eT , S kT uT qT T PTh ÿ ` peF ´ eT , S kT uT nT F qF F PFT ph q ` p2µqsh peh , z pT Plugging this expression into T1 we obtain with δ T :“ ς ´ S kT z |T1 | ď }eh }2ε,h ` sh peh , eh q (1{2 +1{2 # ˆ ÿ “ }δ T }2T ` hT }δ T }2BT ‰ ph q ` p2µq2 sT pp zh, z T PTh Therefore, ` ˘ |T1 | À hk`2 Bpu, kq }z}H 2 pΩqd ` λ}∇¨z}H 1 pΩq À hk`2 Bpu, kq}eh } 42 / 56 L2 -error displacement for the potential VI Consider now T2 . Adding pσ, ∇s zq ´ pf , zq “ 0 and since ÿ lh pp zhq “ pf , πTk zqT , T PTh we have the following decomposition: T2 “ ) ÿ ! pT qT ´ λpDTk u p T , DTk z pT q p T , ∇s pk`1 pσ, ∇s zqT ´ p2µqp∇s pk`1 u z T T T PTh loooooooooooooooooooooooooooooooooooooooooooooooooomoooooooooooooooooooooooooooooooooooooooooooooooooon T2,1 ÿ ÿ pT q ` ´ pf , πTk z ´ zqT p2µqsT pp uT , z T PTh T PTh loooooooooooooomoooooooooooooon looooooooooooomooooooooooooon T2,2 T2,3 43 / 56 L2 -error displacement for the potential VII Using orthogonality, we infer # T2,1 “ ÿ ph qqT p2µqp∇s pu ´ pk`1 I kT uq, ∇s pz ´ pk`1 z T T T PTh + p h qT ` λp∇¨u ´ πTk p∇¨uq, ∇¨z ´ DTk z k k Thus, using the approximation properties of pk`1 T I T and πT , ` ˘ |T2,1 | À hk`2 Bpu, kq }z}H 2 pΩqd ` λ}∇¨z}H 1 pΩq Furthermore, consistency of sT yields |T2,2 | À hk`2 Bpu, kq}z}H 2 pΩqd 44 / 56 L2 -error displacement for the potential VIII Finally, since πTk is self-adjoint and since k ě 1, we infer that pf , πTk z ´ zqT “ pπTk f ´ f , zqT “ pπTk f ´ f , z ´ πT1 zqT , whence, |T2,3 | À hk`2 }f }H k pΩqd }z}H 2 pΩqd Using the estimates for T2,1 , T2,2 and T2,3 with elliptic regularity, |T2 | À hk`2 Bpu, kq}eh } The convergence estimates results from the bound for T1 and T2 45 / 56 Outline 1 Continuous setting 2 Discrete problem and stability 3 Error analysis 4 Numerical examples 46 / 56 Numerical validation I We consider the following exact solution: ` ˘ u sinpπx1 q sinpπx2 q ` p2λq´1 x1 , cospπx1 q cospπx2 q ` p2λq´1 x2 The right-hand side is ` ˘ f “ 2π 2 sinpπx1 q sinpπx2 q, 2π 2 cospπx1 q cospπx2 q The solution u has vanishing divergence in the limit λ Ñ `8 47 / 56 Numerical validation II Figure: Meshes for the numerical example 48 / 56 Numerical validation III k“1 k“2 k“3 10´1 10´1 10´2 10´2 10´3 10´3 10´3 1.89 10´4 10 ´4 2.65 3.54 10´5 10´5 2.97 10´6 10´5 3.93 10´7 10´6 10´7 5.17 4.85 10´9 10´7 ´3 ´2 10 10 10´2 3.02 4.22 4.28 4.95 10´8 k“4 10 ´2.2 10 ´2 ´1.8 10 10 ´2.5 10´2 10´1.5 10´3 10´3 10´4 10´4 10´5 3.75 2.91 10´6 10´8 10´6 3.98 10´7 4.97 10 5.99 10´10 10´3 10´2 ´8 10´9 4.85 10´5 2.99 10´7 4.08 10´9 10´2.2 10´2 10´1.8 10´11 5.05 6 5.94 10´2.5 10´2 10´1.5 Figure: Energy (above) and displacement (below) errors vs. h for λ “ 1 49 / 56 Numerical validation IV k“1 k“2 k“3 10´1 k“4 10´1 10´2 10´2 10´3 10 10´5 2.98 10´6 ´4 10´5 3.94 10´7 10´3 10´3 1.93 10´4 10´5 10´6 10´7 5.15 4.79 10´9 10´7 ´3 ´2 10 3.04 4.26 4.3 4.97 10´8 2.69 3.55 10 10 ´2.2 10 ´2 ´1.8 10 10 ´2.5 10´2 10´1.5 10´2 10´3 10´3 10´4 10´4 10´5 2.92 3.86 10´5 2.99 ´7 4.07 10´6 10´6 3.98 10´8 10´7 4.98 10 4.91 10´9 5.98 10´8 10´10 10´3 10´2 10´2.2 10´2 10´1.8 10´11 5.04 5.99 5.57 10´2.5 10´2 10´1.5 Figure: Energy (above) and displacement (below) errors vs. h for λ “ 1000 50 / 56 Numerical validation V 101 101 100 k k k k 100 10 Figure: τass τsol 2 “1 “2 “3 “4 10´1 10 3 4 10 k k k k “1 “2 “3 “4 102 103 104 vs. cardpFh q for the triangular (left) and hexagonal (right) mesh families 51 / 56 Numerical validation VI 10´1 10´1 10´3 10´3 ´0.84 ´1.04 10´5 10´5 ´1.09 ´1.24 10´7 10´9 k k k k “1 “2 “3 “4 10´1 ´1.73 10´7 ´2.29 10´9 100 101 102 10´2 10´2 10´2 ´4 10´4 10 ´1.64 “1 “2 “3 “4 10´1 ´1.99 100 101 102 ´1.17 ´1.3 10´6 k k k k 10´6 ´1.46 10´8 10´10 10´12 ´1.66 k k k k “1 “2 “3 “4 10´1 ´2.19 ´2.77 100 101 10´8 10´10 102 10´12 10´2 k k k k “1 “2 “3 “4 10´1 ´1.97 ´2.32 100 101 102 Figure: Energy (above) and displacement (below) error vs. τtot (s) for the triangular and hexagonal mesh families 52 / 56 44 A F 16 Cook’s membrane test case I 48 Figure: Cook’s membrane test case (µ “ 0.375, λ “ 7.5 ¨ 106 ) 53 / 56 Cook’s membrane test case II Figure: Deformed configuration for the coarsest (cardpTh q “ 22), intermediate (cardpTh q “ 280) and finest (cardpTh q “ 4192) hexagonal meshes, k “ 1. The color represents the magnitude of the displacement field. 54 / 56 Cook’s membrane test case III u qh,2 pxA q 16.6 ´7.25 ´7.3 qh,1 pxA q u Tria1 Kers1 Hexa1 Tria2 Kers2 Hexa2 16.7 Tria1 Kers1 Hexa1 Tria2 Kers2 Hexa2 ´7.35 16.5 ´7.4 102 103 cardpFh q 104 102 103 cardpFh q 104 Figure: Vertical (left) and horizontal (right) displacement at A 55 / 56 References Brenner, S. C. and Sung, L.-Y. (1992). Linear finite element methods for planar linear elasticity. Math. Comp., 59(200):321–338. Di Pietro, D. A. and Ern, A. (2012). Mathematical Aspects of Discontinuous Galerkin Methods, volume 69 of Math´ ematiques & Applications. Springer, Berlin Heidelberg. Di Pietro, D. A. and Ern, A. (2015). A hybrid high-order locking-free method for linear elasticity on general meshes. Comput. Meth. Appl. Mech. Engrg., 283:1–21. 56 / 56
© Copyright 2024