The barotropic Navier Stokes equations
Transcription
The barotropic Navier Stokes equations
Maillages décalés et écoulements compressibles R. Herbin⋆ , with W. Kheriji⋆† , J.-C. Latché† , B. Piar† R. Herbin ⋆ Université de Provence † Institut de Radioprotection et de Sûreté Nucléaire (IRSN) (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 1 / 32 Introduction Aim and strategy Aim of our work ◮ Cadre : simulation numérique d’écoulements compressibles, code ISIS de l’IRSN R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 2 / 32 Introduction Aim and strategy Aim of our work ◮ Cadre : simulation numérique d’écoulements compressibles, code ISIS de l’IRSN ◮ But : programmer des schémas qui sont ◮ stable and précis tout Mach R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 2 / 32 Introduction Aim and strategy Aim of our work ◮ Cadre : simulation numérique d’écoulements compressibles, code ISIS de l’IRSN ◮ But : programmer des schémas qui sont ◮ ◮ stable and précis tout Mach suffisamment découplés pour que la résolution numérique ne soit pas trop difficile. méthodes de corrections de pression. Classiques pour les fluides incompressibles (Chorin 68, Temam 69), voir (Guermond 06) pour une synthèse. Dévelopées aussi pour les fluides compressibles, soit en colocalisé soit en décalé.. R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 2 / 32 Introduction Aim and strategy Aim of our work ◮ Cadre : simulation numérique d’écoulements compressibles, code ISIS de l’IRSN ◮ But : programmer des schémas qui sont ◮ ◮ stable and précis tout Mach suffisamment découplés pour que la résolution numérique ne soit pas trop difficile. méthodes de corrections de pression. Classiques pour les fluides incompressibles (Chorin 68, Temam 69), voir (Guermond 06) pour une synthèse. Dévelopées aussi pour les fluides compressibles, soit en colocalisé soit en décalé.. ◮ On voudrait avoir des preuves théoriques de stabilité et consistance, étayés par des résultat numériques. R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 2 / 32 Introduction Aim and strategy Aim of our work ◮ Cadre : simulation numérique d’écoulements compressibles, code ISIS de l’IRSN ◮ But : programmer des schémas qui sont ◮ ◮ stable and précis tout Mach suffisamment découplés pour que la résolution numérique ne soit pas trop difficile. méthodes de corrections de pression. Classiques pour les fluides incompressibles (Chorin 68, Temam 69), voir (Guermond 06) pour une synthèse. Dévelopées aussi pour les fluides compressibles, soit en colocalisé soit en décalé.. ◮ On voudrait avoir des preuves théoriques de stabilité et consistance, étayés par des résultat numériques. ◮ Synthèse des résultats récents : H. Kheriji Latché 2012, ESAIM Proc. http://www.cmi.univ-mrs.fr/~ herbin R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 2 / 32 Introduction Aim and strategy Aim of our work ◮ Cadre : simulation numérique d’écoulements compressibles, code ISIS de l’IRSN ◮ But : programmer des schémas qui sont ◮ ◮ stable and précis tout Mach suffisamment découplés pour que la résolution numérique ne soit pas trop difficile. méthodes de corrections de pression. Classiques pour les fluides incompressibles (Chorin 68, Temam 69), voir (Guermond 06) pour une synthèse. Dévelopées aussi pour les fluides compressibles, soit en colocalisé soit en décalé.. ◮ On voudrait avoir des preuves théoriques de stabilité et consistance, étayés par des résultat numériques. ◮ Synthèse des résultats récents : H. Kheriji Latché 2012, ESAIM Proc. http://www.cmi.univ-mrs.fr/~ herbin R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 2 / 32 Introduction Aim and strategy Theoretical study of discretization schemes for nonlinear problems General picture: “Discrete non linear analysis” (see e.g. Eymard Gallouet H. 2000) (i) Mesh dependent a priori estimates R. Herbin (LATP) existence of the approximate solution Maillages décalés et écoulements compressibles Paris 6, Février 2012 3 / 32 Introduction Aim and strategy Theoretical study of discretization schemes for nonlinear problems General picture: “Discrete non linear analysis” (see e.g. Eymard Gallouet H. 2000) (i) Mesh dependent a priori estimates existence of the approximate solution (ii) Uniform a priori estimates R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 3 / 32 Introduction Aim and strategy Theoretical study of discretization schemes for nonlinear problems General picture: “Discrete non linear analysis” (see e.g. Eymard Gallouet H. 2000) (i) Mesh dependent a priori estimates existence of the approximate solution (ii) Uniform a priori estimates (iii) Compactness arguments: any sequence of discrete solutions converges (weakly, up to a subsequence) to a limit as the time and space steps tend to zero. R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 3 / 32 Introduction Aim and strategy Theoretical study of discretization schemes for nonlinear problems General picture: “Discrete non linear analysis” (see e.g. Eymard Gallouet H. 2000) (i) Mesh dependent a priori estimates existence of the approximate solution (ii) Uniform a priori estimates (iii) Compactness arguments: any sequence of discrete solutions converges (weakly, up to a subsequence) to a limit as the time and space steps tend to zero. (iv ) By product: a priori estimates may imply some regularity of the limit. R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 3 / 32 Introduction Aim and strategy Theoretical study of discretization schemes for nonlinear problems General picture: “Discrete non linear analysis” (see e.g. Eymard Gallouet H. 2000) (i) Mesh dependent a priori estimates existence of the approximate solution (ii) Uniform a priori estimates (iii) Compactness arguments: any sequence of discrete solutions converges (weakly, up to a subsequence) to a limit as the time and space steps tend to zero. (iv ) By product: a priori estimates may imply some regularity of the limit. (v ) Passage to the limit in (a weak formulation of) the scheme to show that the limit is a solution of the continuous problem. ◮ For the compressible Navier-Stokes or Euler equations, no hope because of step (ii): lack of estimates on the space translates of the unknown. R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 3 / 32 Introduction Aim and strategy Theoretical study of discretization schemes for nonlinear problems General picture: “Discrete non linear analysis” (see e.g. Eymard Gallouet H. 2000) (i) Mesh dependent a priori estimates existence of the approximate solution (ii) Uniform a priori estimates (iii) Compactness arguments: any sequence of discrete solutions converges (weakly, up to a subsequence) to a limit as the time and space steps tend to zero. (iv ) By product: a priori estimates may imply some regularity of the limit. (v ) Passage to the limit in (a weak formulation of) the scheme to show that the limit is a solution of the continuous problem. ◮ For the compressible Navier-Stokes or Euler equations, no hope because of step (ii): lack of estimates on the space translates of the unknown. ◮ except for the barotropic viscous Stokes equations Eymard Gallouet H. Latché 2009-2010 and steady state Navier-Stokes. R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 3 / 32 Introduction Aim and strategy Theoretical results for general compressible flows ◮ Discrete analogues of the continuous estimates: ◮ positivity of the density ◮ non barotropic flows: positivity of the internal energy, decrease of the total energy ◮ barotropic flows: L2 (H1 ) estimate on the velocity R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 4 / 32 Introduction Aim and strategy Theoretical results for general compressible flows ◮ Discrete analogues of the continuous estimates: ◮ positivity of the density ◮ non barotropic flows: positivity of the internal energy, decrease of the total energy ◮ barotropic flows: L2 (H1 ) estimate on the velocity ◮ Existence of a solution to the (implicit) scheme, by a topological degree argument. R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 4 / 32 Introduction Aim and strategy Theoretical results for general compressible flows ◮ Discrete analogues of the continuous estimates: ◮ positivity of the density ◮ non barotropic flows: positivity of the internal energy, decrease of the total energy ◮ barotropic flows: L2 (H1 ) estimate on the velocity ◮ Existence of a solution to the (implicit) scheme, by a topological degree argument. ◮ Consistency: If the approximate solutions converge in strong enough norms, the limit is a weak solution to the continuous problem. R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 4 / 32 Introduction Space discretization Staggered discretization ◮ Positivity of density finite volume scheme on ρ: ρ piecewise constant on the cells. R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 5 / 32 Introduction Space discretization Staggered discretization ◮ ◮ Positivity of density finite volume scheme on ρ: ρ piecewise constant on the cells. Primal mesh : M = { set of control volumes K , L, M... }. R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 5 / 32 Introduction Space discretization Staggered discretization ◮ ◮ ◮ ◮ Positivity of density finite volume scheme on ρ: ρ piecewise constant on the cells. Primal mesh : M = { set of control volumes K , L, M... }. Scalar variables defined at cell centers: (pK )K ∈M , (̺K )K ∈M ,. . . (i ) Velocity components defined at the (or some of the) edges : (vσ )σ∈F (i ) . R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 5 / 32 Introduction Space discretization Staggered discretization ◮ ◮ ◮ ◮ ◮ Positivity of density finite volume scheme on ρ: ρ piecewise constant on the cells. Primal mesh : M = { set of control volumes K , L, M... }. Scalar variables defined at cell centers: (pK )K ∈M , (̺K )K ∈M ,. . . (i ) Velocity components defined at the (or some of the) edges : (vσ )σ∈F (i ) . (i ) Dual mesh(es) : M∗ = (Dσ )σ∈F (i ) . R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 5 / 32 Introduction Space discretization Staggered discretization ◮ ◮ ◮ ◮ ◮ ◮ Positivity of density finite volume scheme on ρ: ρ piecewise constant on the cells. Primal mesh : M = { set of control volumes K , L, M... }. Scalar variables defined at cell centers: (pK )K ∈M , (̺K )K ∈M ,. . . (i ) Velocity components defined at the (or some of the) edges : (vσ )σ∈F (i ) . (i ) Dual mesh(es) : M∗ = (Dσ )σ∈F (i ) . Normal velocity to the face σ denoted by vσ · nσ . |σ | ′ K σ |L σ=K Dσ ǫ= Dσ | Dσ ′ |M K = L Dσ ′ M Figure: Primal and dual meshes for the Rannacher-Turek and Crouzeix-Raviart elements. R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 5 / 32 Introduction Space discretization The MAC mesh : Dσ (or Kiy,j− 1 ) : Dσ (or Kix− 1 ,j ) 2 2 yj+ 3 2 uyi,j+ 1 2 yj+ 1 2 uxi− 1 ,j+1 2 yj+ 1 2 yj− 1 uxi− 3 ,j 2 uxi− 1 ,j 2 uxi+ 1 ,j 2 uyi−1,j− 1 yj− 1 uyi,j− 1 2 uyi+1,j− 1 2 2 2 2 uxi− 1 ,j−1 yj− 3 2 2 xi − 1 2 uyi,j− 3 yj− 3 2 xi − 3 xi + 1 2 2 2 xi − 3 2 xi − 1 2 xi + 1 2 xi + 3 2 The dual mesh for the MAC scheme, x and y -component of the velocity. R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 6 / 32 The barotropic Navier Stokes equations The continuous equations The continuous barotropic Navier Stokes equations ∂t ρ + div(ρ u) = 0, (mass) ∂t (ρ u) + div(ρ u ⊗ u) + ∇p − div(τ (u)) = 0, (momentum) ρ = Ψ(p), or p = φ(ρ) (eos) 2µ divu Id, 3 Ψ, φ : R+ → R+ . τ (u) = µ(∇u + ∇t u) − R. Herbin (LATP) Maillages décalés et écoulements compressibles µ > 0. Paris 6, Février 2012 7 / 32 The barotropic Navier Stokes equations The continuous equations Properties of barotropic Navier Stokes Properties of the continuous equations: ◮ Positive density: for positive BC’s, ρ > 0 R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 8 / 32 The barotropic Navier Stokes equations The continuous equations Properties of barotropic Navier Stokes Properties of the continuous equations: ◮ Positive density: for positive BC’s, ρ > 0 ◮ Kinetic energy balance: 1 1 ∂t ( ρ |u|2 ) + div ( ρ |u|2 ) u + ∇p · u = div τ (u) · u. 2 2 R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 8 / 32 The barotropic Navier Stokes equations The continuous equations Properties of barotropic Navier Stokes Properties of the continuous equations: ◮ Positive density: for positive BC’s, ρ > 0 ◮ Kinetic energy balance: 1 1 ∂t ( ρ |u|2 ) + div ( ρ |u|2 ) u + ∇p · u = div τ (u) · u. 2 2 ◮ Entropy inequality d dt with H(ρ) = ρ R. Herbin (LATP) Z Z 1 ρ |u|2 + H(ρ) dx ≤ 0. 2 Ω φ(ρ) (recall that p = φ(ρ) by EOS). ρ2 Maillages décalés et écoulements compressibles Paris 6, Février 2012 8 / 32 The barotropic Navier Stokes equations The continuous equations Properties of barotropic Navier Stokes Properties of the continuous equations: ◮ Positive density: for positive BC’s, ρ > 0 ◮ Kinetic energy balance: 1 1 ∂t ( ρ |u|2 ) + div ( ρ |u|2 ) u + ∇p · u = div τ (u) · u. 2 2 ◮ Entropy inequality d dt with H(ρ) = ρ ◮ Z Z 1 ρ |u|2 + H(ρ) dx ≤ 0. 2 Ω φ(ρ) (recall that p = φ(ρ) by EOS). ρ2 Recover these estimates at the discrete level... R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 8 / 32 The barotropic Navier Stokes equations The implicit scheme Finite volume discretization of the mass equation ∂t ρ + div(ρ u) = 0, R. Herbin (LATP) Maillages décalés et écoulements compressibles (mass) Paris 6, Février 2012 9 / 32 The barotropic Navier Stokes equations The implicit scheme Finite volume discretization of the mass equation ∂t ρ + div(ρ u) = 0, ◮ Z (mass) K R. Herbin (LATP) (mass) + implicit time discretization Z n+1 Z ρ − ρn (ρn+1 un+1 · nK ) = 0. + δt K ∂K Maillages décalés et écoulements compressibles Paris 6, Février 2012 9 / 32 The barotropic Navier Stokes equations The implicit scheme Finite volume discretization of the mass equation ∂t ρ + div(ρ u) = 0, ◮ ◮ Z (mass) K (mass) + implicit time discretization Z n+1 Z ρ − ρn (ρn+1 un+1 · nK ) = 0. + δt K ∂K discretization of the fluxes: X |K | n+1 FK ,σ = 0, (ρK − ρnK ) + δt σ∈E(K ) ◮ n+1 FK ,σ = |σ| ρ̌n+1 · nK ,σ , numerical flux through σ. σ uσ ◮ ρ̌n+1 upwind approximation of ρn+1 at the face σ with respect to un+1 · nK ,σ . σ σ R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 9 / 32 The barotropic Navier Stokes equations The implicit scheme Finite volume discretization of the mass equation ∂t ρ + div(ρ u) = 0, ◮ ◮ Z (mass) K (mass) + implicit time discretization Z n+1 Z ρ − ρn (ρn+1 un+1 · nK ) = 0. + δt K ∂K discretization of the fluxes: X |K | n+1 FK ,σ = 0, (ρK − ρnK ) + δt σ∈E(K ) ◮ n+1 FK ,σ = |σ| ρ̌n+1 · nK ,σ , numerical flux through σ. σ uσ ◮ ρ̌n+1 upwind approximation of ρn+1 at the face σ with respect to un+1 · nK ,σ . σ σ ◮ Positive density: ρn+1 > 0 if (ρn > 0 and ρ > 0 at inflow boundary). R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 9 / 32 The barotropic Navier Stokes equations The implicit scheme FV-FE discretization of the momentum equation ∂t (ρ u) + div(ρ u ⊗ u) + ∇p − div(τ (u)) = 0 R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 10 / 32 The barotropic Navier Stokes equations The implicit scheme FV-FE discretization of the momentum equation ∂t (ρ u) + div(ρ u ⊗ u) + ∇p − div(τ (u)) = 0 ◮ Z (momentum) + implicit time discretization Dσ Z Dσ ρn+1 un+1 − ρn un+1 + δt Z ∂Dσ (ρn+1 un+1 ⊗ un+1 · nK ) Z + (∇p n+1 − div(τ (un+1 ))) = 0. Dσ R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 10 / 32 The barotropic Navier Stokes equations The implicit scheme FV-FE discretization of the momentum equation ∂t (ρ u) + div(ρ u ⊗ u) + ∇p − div(τ (u)) = 0 ◮ Z (momentum) + implicit time discretization Dσ Z Dσ ρn+1 un+1 − ρn un+1 + δt Z ∂Dσ (ρn+1 un+1 ⊗ un+1 · nK ) Z + (∇p n+1 − div(τ (un+1 ))) = 0. Dσ ◮ Space discretization X |Dσ | n+1 n+1 n+1 n+1 Fσ,ǫ uǫ (ρσ uσ − ρnσ unσ ) + δt ǫ∈E(Dσ ) +|Dσ |(∇p n+1 )σ − |Dσ |(divτ (un+1 ))σ = 0, ◮ X Z p n+1 divϕ(iσ ) dx = |σ| (pLn+1 − pKn+1 ) nK ,σ · e(i ) , M∈M Z X Z µ X = −µ ∇un+1 · ∇ϕ(iσ ) − div un+1 div ϕ(iσ ) . 3 K K |Dσ |(∇p n+1 )σ,i = − M ◮ |Dσ |(divτ (un+1 ))σ,i ◮ R.σ Herbin Maillages décalés et écoulements compressibles i-th(LATP) finite element shape function. ϕ K ∈M (i ) K ∈M Paris 6, Février 2012 10 / 32 The barotropic Navier Stokes equations The implicit scheme Discretization of the convection operator ◮ n+1 Choice of ρn+1 and Fσ,ǫ in σ X |Dσ | n+1 n+1 n+1 n+1 Fσ,ǫ uǫ ? (ρσ uσ − ρnσ unσ ) + δt ǫ∈E(Dσ ) R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 11 / 32 The barotropic Navier Stokes equations The implicit scheme Discretization of the convection operator ◮ ◮ X |Dσ | n+1 n+1 n+1 n+1 Fσ,ǫ uǫ ? (ρσ uσ − ρnσ unσ ) + δt ǫ∈E(Dσ ) Z Z ρn+1 un+1 − ρn un+1 Discretize + (ρn+1 un+1 ⊗ un+1 · nK ) so as to δt Dσ ∂Dσ obtain a discrete kinetic energy balance. n+1 Choice of ρn+1 and Fσ,ǫ in σ R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 11 / 32 The barotropic Navier Stokes equations The implicit scheme Discretization of the convection operator ◮ ◮ ◮ X |Dσ | n+1 n+1 n+1 n+1 Fσ,ǫ uǫ ? (ρσ uσ − ρnσ unσ ) + δt ǫ∈E(Dσ ) Z Z ρn+1 un+1 − ρn un+1 Discretize + (ρn+1 un+1 ⊗ un+1 · nK ) so as to δt Dσ ∂Dσ obtain a discrete kinetic energy balance. Copy the continuous kinetic energy balance: ∂t (ρ u) + div(ρ u ⊗ u) + ∇p − div(τ (u)) =0 · u n+1 Choice of ρn+1 and Fσ,ǫ in σ 1 1 ∂t ( ρ |u|2 ) + div ( ρ |u|2 ) u + ∇p · u − div τ (u) · u 2 2 with some formal algebra... using ∂t ρ + div(ρ u) = 0. R. Herbin (LATP) Maillages décalés et écoulements compressibles =0, Paris 6, Février 2012 11 / 32 The barotropic Navier Stokes equations The implicit scheme Discretization of the convection operator ◮ ◮ ◮ ◮ X |Dσ | n+1 n+1 n+1 n+1 Fσ,ǫ uǫ ? (ρσ uσ − ρnσ unσ ) + δt ǫ∈E(Dσ ) Z Z ρn+1 un+1 − ρn un+1 Discretize + (ρn+1 un+1 ⊗ un+1 · nK ) so as to δt Dσ ∂Dσ obtain a discrete kinetic energy balance. Copy the continuous kinetic energy balance: ∂t (ρ u) + div(ρ u ⊗ u) + ∇p − div(τ (u)) =0 · u n+1 Choice of ρn+1 and Fσ,ǫ in σ 1 1 ∂t ( ρ |u|2 ) + div ( ρ |u|2 ) u + ∇p · u − div τ (u) · u 2 2 with some formal algebra... using ∂t ρ + div(ρ u) = 0. Do the same at the discrete problem ? ♭ Momentum on dual cells, mass on primal cells... R. Herbin (LATP) Maillages décalés et écoulements compressibles =0, Paris 6, Février 2012 11 / 32 The barotropic Navier Stokes equations The implicit scheme Discretization of the convection operator ◮ ◮ ◮ ◮ X |Dσ | n+1 n+1 n+1 n+1 Fσ,ǫ uǫ ? (ρσ uσ − ρnσ unσ ) + δt ǫ∈E(Dσ ) Z Z ρn+1 un+1 − ρn un+1 Discretize + (ρn+1 un+1 ⊗ un+1 · nK ) so as to δt Dσ ∂Dσ obtain a discrete kinetic energy balance. Copy the continuous kinetic energy balance: ∂t (ρ u) + div(ρ u ⊗ u) + ∇p − div(τ (u)) =0 · u n+1 Choice of ρn+1 and Fσ,ǫ in σ 1 1 ∂t ( ρ |u|2 ) + div ( ρ |u|2 ) u + ∇p · u − div τ (u) · u 2 2 with some formal algebra... using ∂t ρ + div(ρ u) = 0. Do the same at the discrete problem ? ♭ Momentum on dual cells, mass on primal cells... ♯ Idea: reconstruct a mass balance on the the dual cells: X |Dσ | n+1 n+1 ∀σ ∈ Eint , (ρσ − ρnσ ) + Fσ,ǫ =0 δt =0, ǫ∈E(Dσ ) R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 11 / 32 The barotropic Navier Stokes equations The implicit scheme Discretization of the convection operator ◮ ◮ ◮ ◮ X |Dσ | n+1 n+1 n+1 n+1 Fσ,ǫ uǫ ? (ρσ uσ − ρnσ unσ ) + δt ǫ∈E(Dσ ) Z Z ρn+1 un+1 − ρn un+1 Discretize + (ρn+1 un+1 ⊗ un+1 · nK ) so as to δt Dσ ∂Dσ obtain a discrete kinetic energy balance. Copy the continuous kinetic energy balance: ∂t (ρ u) + div(ρ u ⊗ u) + ∇p − div(τ (u)) =0 · u n+1 Choice of ρn+1 and Fσ,ǫ in σ 1 1 ∂t ( ρ |u|2 ) + div ( ρ |u|2 ) u + ∇p · u − div τ (u) · u 2 2 with some formal algebra... using ∂t ρ + div(ρ u) = 0. Do the same at the discrete problem ? ♭ Momentum on dual cells, mass on primal cells... ♯ Idea: reconstruct a mass balance on the the dual cells: X |Dσ | n+1 n+1 ∀σ ∈ Eint , (ρσ − ρnσ ) + Fσ,ǫ =0 δt =0, ǫ∈E(Dσ ) obtained from the primal mass balance for 1 ◮ ρσ = |DK ,σ | ρK + |DL,σ | ρL |Dσ | ◮ Fσ,ǫ : linear combination of the primal fluxes (F K ,σ )σ∈E(K ) . R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 11 / 32 The barotropic Navier Stokes equations The implicit scheme Discrete convection operator ◮ Continuous convection operator Z Z ρn+1 un+1 − ρn un+1 C(ρ, u) = + (ρn+1 un+1 ⊗ un+1 · nσ ). δt Dσ ∂Dσ R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 12 / 32 The barotropic Navier Stokes equations The implicit scheme Discrete convection operator ◮ Continuous convection operator Z Z ρn+1 un+1 − ρn un+1 C(ρ, u) = + (ρn+1 un+1 ⊗ un+1 · nσ ). δt Dσ ∂Dσ ◮ Discrete convection operator Cd (ρd , ud ) = X |Dσ | n+1 n+1 n+1 n+1 Fσ,ǫ uǫ (ρσ uσ − ρnσ unσ ) + δt ǫ∈E(Dσ ) ρσ Fσ,ǫ ◮ ◮ uǫ = R. Herbin 1 2 (LATP) same as in discrete dual mass balance up (un+1 + un+1 σ σ′ ) + δ P 1 ǫ=Dσ |Dσ′ 2 n+1 n+1 sign(Fσ,ǫ ) (un+1 σ,i − uσ ′ ,i ) Maillages décalés et écoulements compressibles Paris 6, Février 2012 12 / 32 The barotropic Navier Stokes equations The implicit scheme Discrete convection operator ◮ Continuous convection operator Z Z ρn+1 un+1 − ρn un+1 C(ρ, u) = + (ρn+1 un+1 ⊗ un+1 · nσ ). δt Dσ ∂Dσ ◮ Discrete convection operator Cd (ρd , ud ) = X |Dσ | n+1 n+1 n+1 n+1 Fσ,ǫ uǫ (ρσ uσ − ρnσ unσ ) + δt ǫ∈E(Dσ ) ρσ Fσ,ǫ ◮ ◮ R. Herbin same as in discrete dual mass balance up (un+1 + un+1 σ σ′ ) + δ ( 1 upwind choice = 0 centered choice. uǫ = δ up 1 2 (LATP) P 1 ǫ=Dσ |Dσ′ 2 n+1 n+1 sign(Fσ,ǫ ) (un+1 σ,i − uσ ′ ,i ) with Maillages décalés et écoulements compressibles Paris 6, Février 2012 12 / 32 The barotropic Navier Stokes equations The implicit scheme Continuous and discrete kinetic energy balance ◮ Continuous setting: Multiply continuous momentum by u: ∂t (ρ u) + div(ρ u ⊗ u) + ∇p − div(τ (u)) = 0 R. Herbin (LATP) Maillages décalés et écoulements compressibles ·u Paris 6, Février 2012 13 / 32 The barotropic Navier Stokes equations The implicit scheme Continuous and discrete kinetic energy balance ◮ Continuous setting: Multiply continuous momentum by u: ∂t (ρ u) + div(ρ u ⊗ u) + ∇p − div(τ (u)) = 0 ·u continuous kinetic energy balance: 1 1 ∂t ( ρ |u|2 ) + div ( ρ |u|2 ) u + ∇p · u − div τ (u) · u = 0 2 2 ... with some formal algebra... using ∂t ρ + div(ρ u) = 0. R. Herbin (LATP) Maillages décalés et écoulements compressibles (♠) Paris 6, Février 2012 13 / 32 The barotropic Navier Stokes equations The implicit scheme Continuous and discrete kinetic energy balance ◮ Continuous setting: Multiply continuous momentum by u: ∂t (ρ u) + div(ρ u ⊗ u) + ∇p − div(τ (u)) = 0 ·u continuous kinetic energy balance: 1 1 ∂t ( ρ |u|2 ) + div ( ρ |u|2 ) u + ∇p · u − div τ (u) · u = 0 2 2 ... with some formal algebra... using ∂t ρ + div(ρ u) = 0. ◮ Discrete setting: Similarly, multiply discrete momentum by un+1 σ : |D | X σ n+1 n+1 n+1 n n Fσ,ǫ uǫ (ρn+1 σ uσ − ρ σ uσ ) + δt (♠) ǫ∈E(Dσ ) + |Dσ |(∇p n+1 )σ,i − |Dσ |(divτ (un+1 ))σ = 0 R. Herbin (LATP) Maillages décalés et écoulements compressibles · uσ Paris 6, Février 2012 13 / 32 The barotropic Navier Stokes equations The implicit scheme Continuous and discrete kinetic energy balance ◮ Continuous setting: Multiply continuous momentum by u: ∂t (ρ u) + div(ρ u ⊗ u) + ∇p − div(τ (u)) = 0 ·u continuous kinetic energy balance: 1 1 ∂t ( ρ |u|2 ) + div ( ρ |u|2 ) u + ∇p · u − div τ (u) · u = 0 2 2 ... with some formal algebra... using ∂t ρ + div(ρ u) = 0. ◮ Discrete setting: Similarly, multiply discrete momentum by un+1 σ : |D | X σ n+1 n+1 n+1 n n Fσ,ǫ uǫ (ρn+1 σ uσ − ρ σ uσ ) + δt (♠) ǫ∈E(Dσ ) + |Dσ |(∇p n+1 )σ,i − |Dσ |(divτ (un+1 ))σ = 0 discrete kinetic energy balance: i 1 1 |Dσ | h n+1 n+1 2 ρσ (uσ ) − ρnσ (unσ )2 + 2 δt 2 · uσ n+1 n+1 n+1 Fσ,ǫ uσ uσ ′ ǫ=Dσ |Dσ′ n+1 un+1 = 0 with Rσn+1 ≥ 0) σ +Rσ X |Dσ | n+1 n+1 ... with(LATP) some algebra... Maillages using décalés et écoulements (ρσ −compressibles ρnσ ) + = 0.6, Février 2012 Fσ,ǫ R. Herbin Paris +|Dσ | (∇p n+1 X n+1 )σ,i un+1 ))σ σ −|Dσ |(divτ (u (♠)σ 13 / 32 The barotropic Navier Stokes equations The implicit scheme Continuous entropy inequality ◮ 1 ρ2 |u|2 ; Continuous kinetic energy equation, Ek = 12 ρ |u|2 = 2 ρ ∂t Ek + div Ek u + ∇p · u = div τ (u) · u, R. Herbin (LATP) Maillages décalés et écoulements compressibles (♠) Paris 6, Février 2012 14 / 32 The barotropic Navier Stokes equations The implicit scheme Continuous entropy inequality ◮ ◮ 1 ρ2 |u|2 ; Continuous kinetic energy equation, Ek = 12 ρ |u|2 = 2 ρ ∂t Ek + div Ek u + ∇p · u = div τ (u) · u, (♠) Find an entropy of the form S = Ek + H(ρ), with H satisfying ∂t H(ρ) + div H(ρ) u + p div(u) = 0. R. Herbin (LATP) Maillages décalés et écoulements compressibles (♣) Paris 6, Février 2012 14 / 32 The barotropic Navier Stokes equations The implicit scheme Continuous entropy inequality ◮ ◮ 1 ρ2 |u|2 ; Continuous kinetic energy equation, Ek = 12 ρ |u|2 = 2 ρ ∂t Ek + div Ek u + ∇p · u = div τ (u) · u, (♠) Find an entropy of the form S = Ek + H(ρ), with H satisfying ∂t H(ρ) + div H(ρ) u + p div(u) = 0. (♣) ◮ Then, summing (♠) and (♣) ◮ ∂t S + div (S + p) u − div τ (u) u = −τ (u) : ∇u. Z Z d 1 d S ) dx = [ ρ |u|2 + H(ρ)] dx ≤ 0 Integrating over Ω: dt Ω dt Ω 2 How to get H′ (ρ) ? ◮ R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 14 / 32 The barotropic Navier Stokes equations The implicit scheme Continuous entropy inequality ◮ ◮ ◮ ◮ ◮ 1 ρ2 |u|2 ; Continuous kinetic energy equation, Ek = 12 ρ |u|2 = 2 ρ ∂t Ek + div Ek u + ∇p · u = div τ (u) · u, (♠) Find an entropy of the form S = Ek + H(ρ), with H satisfying ∂t H(ρ) + div H(ρ) u + p div(u) = 0. (♣) Then, summing (♠) and (♣) ∂t S + div (S + p) u − div τ (u) u = −τ (u) : ∇u. Z Z d 1 d S ) dx = [ ρ |u|2 + H(ρ)] dx ≤ 0 Integrating over Ω: dt Ω dt Ω 2 How to get H′ (ρ) ? Multiply mass balance ∂t ρ + div(ρ u) = 0 by H′ ∂t (H(ρ)) + u · ∇(H(ρ)) + H′ (ρ)ρ divu = 0 ∂t (H(ρ) + div(H(ρ)u) + (H′ (ρ)ρ − H(ρ))divu = 0 Z φ(s) yields (♣) for H′ (ρ)ρ − H(ρ) = p = φ(ρ), i.e. H(ρ) = ρ . s2 γ ρ if γ > 1, . For an ideal gas, = H(ρ) = γ − 1 ρ ln(ρ) if γ = 1, R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 14 / 32 The barotropic Navier Stokes equations The implicit scheme Discrete potential inequality ◮ Multiply discrete mass balance by H′ (ρK ) : X |K | n+1 (ρK − ρnK )H′ (ρK ) + FK ,σ H′ (ρK ) = 0 δt σ∈E(K ) ◮ X |K | n+1 n |σ| H(ρn+1 · nK ,σ (H(ρn+1 σ )uσ K ) − H(ρK )) + δt σ∈E(K ) X n+1 uσ · nK ,σ + RKn+1 = 0 + pK (♣)K σ∈E(K ) H convex + upstream choice in the mass balance R. Herbin (LATP) Maillages décalés et écoulements compressibles RKn+1 ≥ 0 ... Paris 6, Février 2012 15 / 32 The barotropic Navier Stokes equations The implicit scheme Discrete entropy inequality ◮ As in the X continuous case, X sum kinetic energy balance and elastic potential (♠)σ + (♣)K balance: σ∈E X σ∈Eint K ∈M h i 1 n+1 2 n n 2 1 |Dσ | ρn+1 + σ (uσ ) − ρσ (uσ ) 2 δt 2 X n+1 n+1 n+1 Fσ,ǫ uσ uσ ′ ǫ=Dσ |Dσ′ +|Dσ | (∇p n+1 )σ un+1 − |Dσ |(divτ (un+1 ))σ un+1 σ σ + X K ∈M i X X |K | n+1 n |σ| H(ρn+1 · nK ,σ + pK un+1 · nK ,σ ≤ 0. (H(ρn+1 σ )uσ σ K ) − H(ρK )) + δt σ∈E(K ) σ∈E(K ) thanks to positive residual upwind term... R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 16 / 32 The barotropic Navier Stokes equations The implicit scheme Discrete entropy inequality ◮ As in the X continuous case, X sum kinetic energy balance and elastic potential (♠)σ + (♣)K balance: K ∈M σ∈E X σ∈Eint h i 1 n+1 2 n n 2 1 |Dσ | ρn+1 + σ (uσ ) − ρσ (uσ ) 2 δt 2 X n+1 n+1 n+1 Fσ,ǫ uσ uσ ′ ǫ=Dσ |Dσ′ +|Dσ | (∇p n+1 )σ un+1 − |Dσ |(divτ (un+1 ))σ un+1 σ σ + X K ∈M ◮ X X |K | n+1 n |σ| H(ρn+1 · nK ,σ + pK un+1 · nK ,σ ≤ 0. (H(ρn+1 σ )uσ σ K ) − H(ρK )) + δt σ∈E(K ) σ∈E(K ) thanks to positiveX residual upwind term... X X Discrete duality: |σ| un+1 · nK ,σ = 0 pK |Dσ | (∇p n+1 )σ un+1 σ σ,i + E ◮ i X K ∈M σ∈E(K ) |Dσ | (divτ (un+1 ))σ un+1 ≤ 0. σ E R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 16 / 32 The barotropic Navier Stokes equations The implicit scheme Discrete entropy inequality ◮ As in the X continuous case, X sum kinetic energy balance and elastic potential (♠)σ + (♣)K balance: K ∈M σ∈E X σ∈Eint h i 1 n+1 2 n n 2 1 |Dσ | ρn+1 + σ (uσ ) − ρσ (uσ ) 2 δt 2 X n+1 n+1 n+1 Fσ,ǫ uσ uσ ′ ǫ=Dσ |Dσ′ +|Dσ | (∇p n+1 )σ un+1 − |Dσ |(divτ (un+1 ))σ un+1 σ σ + X K ∈M ◮ X X |K | n+1 n |σ| H(ρn+1 · nK ,σ + pK un+1 · nK ,σ ≤ 0. (H(ρn+1 σ )uσ σ K ) − H(ρK )) + δt σ∈E(K ) σ∈E(K ) thanks to positiveX residual upwind term... X X Discrete duality: |σ| un+1 · nK ,σ = 0 pK |Dσ | (∇p n+1 )σ un+1 σ σ,i + E ◮ i X K ∈M σ∈E(K ) |Dσ | (divτ (un+1 ))σ un+1 ≤ 0. σ E ◮ Conservativity of the fluxes i X |K | 1 X |Dσ | h n+1 n+1 2 n ρσ (uσ,i ) − ρnσ (unσ,i )2 + (H(ρn+1 K ) − H(ρK )) ≤ 0. 2 δt δt R. Herbini ,E(LATP) Maillages décalés et écoulementsKcompressibles Paris 6, Février 2012 ∈M 16 / 32 The barotropic Navier Stokes equations The implicit scheme Passage to the limit : weak consistency of the scheme ◮ ◮ (M(k) , δt (k) )k∈N sequence of meshes and time steps, such that h(k) and δt (k) tend to zero as k tends to infinity. Assume estimates : kρ(k) kT ,x,BV + kp (k) kT ,x,BV + ku (k) kT ,x,BV ≤ C , ∀k ∈ N, ku (k) kT ,t,BV ≤ C , ∀k ∈ N. ◮ kqkT ,x,BV = N X n=0 R. Herbin (LATP) δt X n |qLn − qK |, kv kT ,x,BV = σ∈E σ=K |L Maillages décalés et écoulements compressibles N X n=0 δt X |vσn ′ − vσn |, ǫ∈Ē ′ σ=Dσ |Dσ Paris 6, Février 2012 17 / 32 The barotropic Navier Stokes equations The implicit scheme Passage to the limit : weak consistency of the scheme ◮ ◮ (M(k) , δt (k) )k∈N sequence of meshes and time steps, such that h(k) and δt (k) tend to zero as k tends to infinity. Assume estimates : kρ(k) kT ,x,BV + kp (k) kT ,x,BV + ku (k) kT ,x,BV ≤ C , ∀k ∈ N, ku (k) kT ,t,BV ≤ C , ∀k ∈ N. ◮ kqkT ,x,BV = N X n=0 ◮ δt X n |qLn − qK |, σ∈E σ=K |L Assume convergence : ρ(k) , p (k) , u (k) 3 Lr (0, T ) × Ω R. Herbin (LATP) kv kT ,x,BV = N X n=0 k∈N δt X |vσn ′ − vσn |, ǫ∈Ē ′ σ=Dσ |Dσ 3 → (ρ̄, p̄, ū) ∈ L∞ (0, T ) × Ω , in Maillages décalés et écoulements compressibles Paris 6, Février 2012 17 / 32 The barotropic Navier Stokes equations The implicit scheme Passage to the limit : weak consistency of the scheme ◮ ◮ (M(k) , δt (k) )k∈N sequence of meshes and time steps, such that h(k) and δt (k) tend to zero as k tends to infinity. Assume estimates : kρ(k) kT ,x,BV + kp (k) kT ,x,BV + ku (k) kT ,x,BV ≤ C , ∀k ∈ N, ku (k) kT ,t,BV ≤ C , ∀k ∈ N. ◮ kqkT ,x,BV = N X n=0 ◮ δt X n |qLn − qK |, kv kT ,x,BV = σ∈E σ=K |L N X δt n=0 Assume convergence : ρ(k) , p (k) , u (k) 3 Lr (0, T ) × Ω k∈N X |vσn ′ − vσn |, ǫ∈Ē ′ σ=Dσ |Dσ 3 → (ρ̄, p̄, ū) ∈ L∞ (0, T ) × Ω , in Then, the limit (ρ̄, p̄, ū) satisfies Z Z TZ h i ρ(x, 0) ϕ(x, 0) dx = 0, ρ ∂t ϕ + ρ u ∂x ϕ dx dt + 0 Z Ω Ω T 0 Z h 2 i ρ u ∂t ϕ + (ρ u + p) ∂x ϕ dx dt + Ω Z ρ(x, 0) u(x, 0) ϕ(x, 0) dx = 0, Ω ρ = Ψ(p). R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 17 / 32 The barotropic Navier Stokes equations The implicit scheme Passage to the limit : weak consistency of the scheme ◮ ◮ (M(k) , δt (k) )k∈N sequence of meshes and time steps, such that h(k) and δt (k) tend to zero as k tends to infinity. Assume estimates : kρ(k) kT ,x,BV + kp (k) kT ,x,BV + ku (k) kT ,x,BV ≤ C , ∀k ∈ N, ku (k) kT ,t,BV ≤ C , ∀k ∈ N. ◮ kqkT ,x,BV = N X n=0 ◮ δt X n |qLn − qK |, kv kT ,x,BV = σ∈E σ=K |L N X δt n=0 Assume convergence : ρ(k) , p (k) , u (k) 3 Lr (0, T ) × Ω k∈N X |vσn ′ − vσn |, ǫ∈Ē ′ σ=Dσ |Dσ 3 → (ρ̄, p̄, ū) ∈ L∞ (0, T ) × Ω , in Then, the limit (ρ̄, p̄, ū) satisfies Z Z TZ h i ρ(x, 0) ϕ(x, 0) dx = 0, ρ ∂t ϕ + ρ u ∂x ϕ dx dt + 0 Z Ω Ω T 0 Z h 2 i ρ u ∂t ϕ + (ρ u + p) ∂x ϕ dx dt + Ω Z ρ(x, 0) u(x, 0) ϕ(x, 0) dx = 0, Ω ρ = Ψ(p). and the entropy condition Z Z TZ S(x, 0)ϕ(x, 0) dx ≥ 0, ∀ϕ ∈ C∞ S∂t ϕ + (S + p) u · ∇ϕ dx dt + c Ω × [0, T ), R+ . 0 Ω Ω R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 17 / 32 The barotropic Navier Stokes equations The implicit scheme Passage to the limit, sketch of proof Proof: ◮ To get the weak formulation: classical technique (see e.g. Eymard Gallouet H. 2000): multiply the scheme by the discretization of a test function, put discrete derivatives on the test functions and pass to the limit thanks to the estimates. R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 18 / 32 The barotropic Navier Stokes equations The implicit scheme Passage to the limit, sketch of proof Proof: ◮ To get the weak formulation: classical technique (see e.g. Eymard Gallouet H. 2000): multiply the scheme by the discretization of a test function, put discrete derivatives on the test functions and pass to the limit thanks to the estimates. ◮ Entropy Multiply (♠)σ by ϕσ , sum on σ, Multiply (♣)K by ϕK , sum on K . X E∈Eint i 1 |Dσ | h n n+1 2 1 n−1 n 2 ρσ (uσ ) − ρσ (uσ ) + δt 2 2 X ǫ=σ|σ′ n n+1 n+1 Fσ,ǫ uσ u ′ σ n+1 , +|Dσ | (∇p)n+1 uσ σ + X K ∈M h i |K | H(ρn+1 ) − H(ρn K) K δt P n+1 n+1 + σ∈E(K ) H(ρσ ) uσ nK ,σ P n+1 n +p n+1 K ,σ σ∈E(K ) uσ K = − ϕn σ n ϕK , , X n+1 n Rσ ϕσ − E∈Eint All terms pass to the limit, except R. Herbin (LATP) P Eint X K ∈M R n+1 n ϕK , K |Dσ |Rσn+1 ϕσ .... ≥ 0. Maillages décalés et écoulements compressibles Paris 6, Février 2012 18 / 32 The barotropic Navier Stokes equations The pressure correction scheme The pressure correction algorithm 1 - Pressure renormalization step h1 i h i 1 div n ∇p̃ n+1 = div ∇p n n n 1/2 ρ (ρ ρ ) (1) 2 - Prediction step ρn ũn+1 − ρn−1 un + div(ρn ũn+1 ⊗ un ) + ∇p̃ n+1 − div(τ (ũn+1 )) = 0. δt (2) 3 - Correction step ρn un+1 − ũn+1 + ∇(p n+1 − p̃ n+1 ) = 0, δt ρn+1 − ρn + div(ρn+1 un+1 ) = 0, δt ρn+1 = Ψ(p n+1 ). Same stability results and passage to the limit as for the implicit scheme. R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 19 / 32 The barotropic Navier Stokes equations Results for the Sod shock tube test Sod shock tube test, 1D 1.2 ρ u = L 1 0 , ρ u = R 0.125 0 , EOS:p = ρ 1 n=250 n=500 n=1000 n=2000 exact 1 n=250 n=500 n=1000 n=2000 exact 0.9 0.8 0.8 pressure velocity 0.7 0.6 0.4 0.6 0.5 0.4 0.2 0.3 0 0.2 -0.2 0.1 -2 -1 0 1 2 3 -2 -1 x 0 1 2 3 x Sod shock tube problem – Centred scheme – MAC discretization Velocity (left) and pressure (right). Exact solution and numerical solution of the problem at t = 1 with CFL=1 (max velocity =1, speed of sound = 1). Remark: residual viscosity added (either physical or by upwinding) for CR and RT discretizations. (either physical or by upwinding) to avoid the odd-even decoupling phenomenon. R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 20 / 32 The barotropic Navier Stokes equations Results for the Sod shock tube test Sod shock tube test: order of convergence error (L1 norm) error (L1 norm) velocity pressure 0.1 0.01 0.01 0.001 0.01 0.001 h 0.01 h Sod shock tube problem – Centred scheme – L1 norm of the error between the numerical solution and the exact solution at t = 1, as a function of the mesh (or time) step, for CFL=1. Velocity (left) and pressure (right). R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 21 / 32 The full Navier-Stokes equations The continuous equations The full compressible Navier–Stokes equations ◮ balance equations ∂t ρ + div(ρ u) = 0, (mass) ∂t (ρ u) + div(ρ u ⊗ u) + ∇p − div(τ (u)) = 0, (momentum) ∂t (ρ E ) + div(ρ E u) + div(p u) + div(q) = div(τ (u) · u), 1 ρ = Ψ(p, e), (or p = φ(ρ, e)), E = |u|2 + e, 2 (total energy) ◮ E : total energy, e: internal energy ◮ q = −λ∇e, heat conduction flux, λ ≥ 0. ◮ (total energy balance) - (♠) (kinetic energy balance) (eos) ∂t (ρe) + div(ρeu) + p div(u) − div(λ∇e) = τ (u) : ∇u. (internal energy) R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 22 / 32 The full Navier-Stokes equations The continuous equations Total energy balance and internal energy balance ◮ Positivity of e ◮ ∂t (ρe) + div(ρeu) + p div(u) − div(λ∇e) = τ (u) : ∇u. (internal energy) ◮ τ (u) : ∇u ≥ 0 ◮ (mass balance) ∂t (ρe) + div(ρeu) = ρ [∂t e + u · ∇e], ◮ (EOS) ⇒ p = 0 when e = 0, ⇒ e(x, t) > 0 for all x ∈ Ω, t ∈ (0, T ). ◮ Incompressible or low Mach number flows: discretization of (internal energy) natural choice. ◮ Using (internal energy): easy discretization which yields ed > 0. Using (total energy) : if u = 0 and ∇q · n = 0 on ∂Ω, then: Z Z (total energy) ∂t (ρ E ) + div(ρ E u) + div(p u) + div(q) = div(τ (u) · u) Ω Ω Z 1 d ρ |u|2 + ρe dx ≤ 0, (global total energy balance) dt Ω 2 ◮ Z Ω R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 23 / 32 The full Navier-Stokes equations Space and time discretization Discrete total energy balance and discrete internal energy balance ? ◮ Idea: Use a “good” discretization for which we can apply the same “algebra” as in the continuous case. Need of a discrete kinetic energy balance equation that is compatible with the discrete internal energy equation to recover a discrete total energy balance. ◮ Possible if: Z R ◮ (u) : ∇u) can be identified to (divτ (u))d · ud . (τ d Ω Ω ◮ RHS of (internal energy)d ≥ 0 so that ed ≥ 0. ◮ OK for FE... a bit of work to build the MAC version of (divτ (u))d and τ (u) : ∇u)d . ◮ (H., Kheriji, Latché, 2011), (Babik, H., Kheriji, Latché, 2011) R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 24 / 32 The full Navier-Stokes equations Space and time discretization The pressure correction algorithm 1 - Pressure renormalization step i h1 i h 1 ∇p n div n ∇p̃ n+1 = div n n 1/2 ρ (ρ ρ ) (3) 2 - Prediction step ρn ũn+1 − ρn−1 un + div(ρn ũn+1 ⊗ un ) + ∇p̃ n+1 − div(τ (ũn+1 )) = 0. δt (4) 3 - Correction step – Solve (simultanuously) the following non linear equations for p n+1 , un+1 and ρn+1 : un+1 − ũn+1 + ∇(p n+1 − p̃ n+1 ) = 0, δt ρn+1 − ρn + div(ρn+1 un+1 ) = 0, δt ρn+1 e n+1 − ρn e n + div(ρn+1 un+1 e n+1 ) +p n+1divun+1−div(λ∇e n+1 ) = τ (un+1 ) :∇un+1 , δt p n+1 = φ(ρn+1 , e n+1 ). ρn Same stability results and passage to the limit as for the implicit scheme. R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 25 / 32 The Euler equations The continuous equations Shocks !! Euler equations: ∂t ̺ + div(̺u) = 0, ∂t (̺u) + div(̺u ⊗ u) + ∇p = 0, ∂t (̺E ) + div (̺E + p)u = 0, p = (γ − 1) ̺e, E = 1 2 |u| + e. 2 (total energy) 6⇔ (internal energy) Lack of consistency in the discrete kinetic equation: for non regular solutions, Rσ 6→ 0 with the mesh size... R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 26 / 32 The Euler equations The continuous equations Shocks !! Euler equations: ∂t ̺ + div(̺u) = 0, ∂t (̺u) + div(̺u ⊗ u) + ∇p = 0, ∂t (̺E ) + div (̺E + p)u = 0, p = (γ − 1) ̺e, E = 1 2 |u| + e. 2 (total energy) 6⇔ (internal energy) Lack of consistency in the discrete kinetic equation: for non regular solutions, Rσ 6→ 0 with the mesh size... How to obtain the correct weak solutions while solving the internal energy balance ? R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 26 / 32 The Euler equations The continuous equations Shocks !! Euler equations: ∂t ̺ + div(̺u) = 0, ∂t (̺u) + div(̺u ⊗ u) + ∇p = 0, ∂t (̺E ) + div (̺E + p)u = 0, p = (γ − 1) ̺e, E = 1 2 |u| + e. 2 (total energy) 6⇔ (internal energy) Lack of consistency in the discrete kinetic equation: for non regular solutions, Rσ 6→ 0 with the mesh size... How to obtain the correct weak solutions while solving the internal energy balance ? Answer: Make the scheme ”consistent” with total energy equation. . . counterbalance Rσ by an additional term SK in the discrete internal equation, such that XX X X δt|SK |ϕK → 0 as mesh size → 0. δt|Rσ |ϕσ + n R. Herbin σ∈E (LATP) n K ∈M Maillages décalés et écoulements compressibles Paris 6, Février 2012 26 / 32 The Euler equations The continuous equations An implicit scheme X n+1 |K | n+1 FK ,σ = 0, (ρK − ρnK ) + δt σ∈E(K ) X |Dσ | n+1 n+1 n+1 n+1 n Fσ,ǫ uǫ,i + |Dσ | (∇p n+1 )σ,i = 0, (ρσ uσ,i − ρnσ uσ,i )+ δt ǫ∈Ē(Dσ ) X n+1 n+1 |K | n+1 n+1 FK ,σ eσ + |K | pKn+1 (div un+1 ) K = SKn+1 , (ρK eK − ρnK eKn ) + δt σ∈E(K ) pKn+1 = (γ − 1)ρn+1 K eKn+1 . Existence, stability ◮ If the solutions exists, and if SK ≥ 0, e ≥ 0. ◮ Kinetic energy conservation + internal energy conservation (provided that SK may be absorbed). ◮ ∃ at least a solution (topological degree argument) (and e ≥ 0. R. Herbin (LATP) Maillages décalés et écoulements compressibles stability estimate Paris 6, Février 2012 27 / 32 The Euler equations The continuous equations Choice of SK ◮ Kinetic energy balance: (∂t (̺Ek ) + div(̺Ek u) + u · ∇p = 0, Ek = 21 |u|2 ) ◮ Multiply the momentum balance equation by uσ and use the mass balance: 1 X |Dσ | (̺σ |uσ |2 − ̺∗σ |u∗σ |2 ) + Fσ ′ uσ uσ ′ + (∇p)σ · uσ = −Rσ , 2δt 2 ′ ǫ=σ|σ with: Rσ = i h X |ǫ| |Dσ | ∗ εǫ (uσ − uσ ′ ) ·uσ . ̺σ |uσ − u∗σ ′ |2 2δt dǫ ′ ǫ=σ|σ ◮ Multiply kinetic egergy balance and internal energy balance by adequate interpolates of test function ϕ XX n (kinetic terms)σ ϕσ + σ∈E X X n (energy terms)K ϕK = K ∈M X X n δt|SK |ϕK + K ∈M The terms at the left hand side tend to what they should... R. Herbin XX (LATP) X X Maillages décalés et écoulements compressibles XX n δt|Rσ |ϕσ . σ∈E Paris 6, Février 2012 28 / 32 The Euler equations The continuous equations Choice of the source term Rσn+1 = − 2 h 1 |Dσ | n ρσ un+1 − unσ,i − σ 2 δt X ǫ=Dσ |Dσ′ i 1 n+1 n+1 |Fσ,ǫ | (un+1 − un+1 σ σ ′ ) uσ . 2 ∀K ∈ M, SKn+1 = 1 2 X |DK ,σ | n 2 ρK un+1 − unσ + σ δt σ∈E(K ) With this choice, XX n R. Herbin σ∈E (LATP) δt|Rσ |ϕσ + X X n X n+1 |Fσ,ǫ | n+1 2 (uσ ′ − un+1 σ ′′ ) . 2 ǫ∩K̄ 6=∅, ′ ǫ=Dσ |Dσ′′ δt|SK |ϕK → 0 as mesh size → 0. K ∈M Maillages décalés et écoulements compressibles Paris 6, Février 2012 29 / 32 The Euler equations Numerical tests The test case Test 5 Toro 2009 Chap 4. ρL 5.99924 left state: uL = 19.5975 460.894 pL ρR 5.99242 right state: uR = −6.19633 46.0950 pR Ω = (−0.5, 0.5). Discontinuity initially located at x = 0. Centered approximation of the velocity at the dual faces in the momentum balance. 40 35 30 25 density (analytical) state: ∗ intermediate 1691.65 p = 8.68977 u∗ for x ∈ (0.028, 0.428) 20 15 10 numerical results: p ∈ (1691.6, 1691.8) u ∈ (8.689, 8.690) for x ∈ (0.032, 0.417) R. Herbin (LATP) 5 with without exact 0 -0.4 -0.2 0 x 0.2 0.4 Test 5 of [ Toro chapter 4] - Density at t = 0.035, n = 2000 cells, with and without corrective source terms, and analytical solution. Maillages décalés et écoulements compressibles Paris 6, Février 2012 30 / 32 The Euler equations Numerical tests 40 40 35 35 30 30 25 25 density density Density fields 20 20 15 15 10 10 5 5 n=500 n=1000 n=2000 n=500 n=1000 n=2000 0 0 -0.4 -0.2 0 0.2 0.4 -0.4 x -0.2 0 0.2 0.4 x Test 5 of Toro, 2009, chapter 4] - Density obtained with various meshes, with (left) and without (right) corrective source terms. δt = h/20 CFL ≈ 1 R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 31 / 32 The Euler equations Numerical tests 1800 1800 1600 1600 1400 1400 1200 1200 1000 1000 pressure pressure Pressure fields 800 600 600 400 200 800 400 200 n=500 n=1000 n=2000 0 n=500 n=1000 n=2000 0 -0.4 -0.2 0 0.2 0.4 -0.4 x -0.2 0 0.2 0.4 x Test 5 of [Toro, 2009, chapter 4] - pressure obtained with various meshes, with (left) and without (right) corrective source terms. δt = h/20 CFL ≈ 1 R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 32 / 32 The Euler equations Numerical tests Conclusion and perspectives ◮ A suitable discretization of the convection terms allows to obtain a (local) kinetic energy conservation identity (for colocated as well as staggered meshes). R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 33 / 32 The Euler equations Numerical tests Conclusion and perspectives ◮ A suitable discretization of the convection terms allows to obtain a (local) kinetic energy conservation identity (for colocated as well as staggered meshes). ◮ The staggered scheme is stable at all Mach numbers, able to compute solutions of inviscid flows. R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 33 / 32 The Euler equations Numerical tests Conclusion and perspectives ◮ A suitable discretization of the convection terms allows to obtain a (local) kinetic energy conservation identity (for colocated as well as staggered meshes). ◮ The staggered scheme is stable at all Mach numbers, able to compute solutions of inviscid flows. ◮ It has been used successfully for two-phase flows . R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 33 / 32 The Euler equations Numerical tests Conclusion and perspectives ◮ A suitable discretization of the convection terms allows to obtain a (local) kinetic energy conservation identity (for colocated as well as staggered meshes). ◮ The staggered scheme is stable at all Mach numbers, able to compute solutions of inviscid flows. ◮ It has been used successfully for two-phase flows . R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 33 / 32 The Euler equations Numerical tests Conclusion and perspectives ◮ A suitable discretization of the convection terms allows to obtain a (local) kinetic energy conservation identity (for colocated as well as staggered meshes). ◮ The staggered scheme is stable at all Mach numbers, able to compute solutions of inviscid flows. ◮ It has been used successfully for two-phase flows . To do: ◮ Recover the entropy inequality for the Euler equations. R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 33 / 32 The Euler equations Numerical tests Conclusion and perspectives ◮ A suitable discretization of the convection terms allows to obtain a (local) kinetic energy conservation identity (for colocated as well as staggered meshes). ◮ The staggered scheme is stable at all Mach numbers, able to compute solutions of inviscid flows. ◮ It has been used successfully for two-phase flows . To do: ◮ Recover the entropy inequality for the Euler equations. ◮ Explicit schemes: first positive results for the barotropic equations. On going tests for Euler (Ph. D. T.T. Nguyen). R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 33 / 32 The Euler equations Numerical tests Conclusion and perspectives ◮ A suitable discretization of the convection terms allows to obtain a (local) kinetic energy conservation identity (for colocated as well as staggered meshes). ◮ The staggered scheme is stable at all Mach numbers, able to compute solutions of inviscid flows. ◮ It has been used successfully for two-phase flows . To do: ◮ Recover the entropy inequality for the Euler equations. ◮ Explicit schemes: first positive results for the barotropic equations. On going tests for Euler (Ph. D. T.T. Nguyen). ◮ Higher order (entropy viscosity by Guermond) R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 33 / 32 The Euler equations Numerical tests Conclusion and perspectives ◮ A suitable discretization of the convection terms allows to obtain a (local) kinetic energy conservation identity (for colocated as well as staggered meshes). ◮ The staggered scheme is stable at all Mach numbers, able to compute solutions of inviscid flows. ◮ It has been used successfully for two-phase flows . To do: ◮ Recover the entropy inequality for the Euler equations. ◮ Explicit schemes: first positive results for the barotropic equations. On going tests for Euler (Ph. D. T.T. Nguyen). ◮ Higher order (entropy viscosity by Guermond) ◮ Pass to the limit in the scheme (for fixed mesh) as the Mach number tends to 0... R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 33 / 32 The Euler equations Numerical tests Conclusion and perspectives ◮ A suitable discretization of the convection terms allows to obtain a (local) kinetic energy conservation identity (for colocated as well as staggered meshes). ◮ The staggered scheme is stable at all Mach numbers, able to compute solutions of inviscid flows. ◮ It has been used successfully for two-phase flows . To do: ◮ Recover the entropy inequality for the Euler equations. ◮ Explicit schemes: first positive results for the barotropic equations. On going tests for Euler (Ph. D. T.T. Nguyen). ◮ Higher order (entropy viscosity by Guermond) ◮ Pass to the limit in the scheme (for fixed mesh) as the Mach number tends to 0... ◮ Continue the discrete non linear analysis as far as possible... R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 33 / 32 The Euler equations Numerical tests Conclusion and perspectives ◮ A suitable discretization of the convection terms allows to obtain a (local) kinetic energy conservation identity (for colocated as well as staggered meshes). ◮ The staggered scheme is stable at all Mach numbers, able to compute solutions of inviscid flows. ◮ It has been used successfully for two-phase flows . To do: ◮ Recover the entropy inequality for the Euler equations. ◮ Explicit schemes: first positive results for the barotropic equations. On going tests for Euler (Ph. D. T.T. Nguyen). ◮ Higher order (entropy viscosity by Guermond) ◮ Pass to the limit in the scheme (for fixed mesh) as the Mach number tends to 0... ◮ Continue the discrete non linear analysis as far as possible... ◮ Reactive flows R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 33 / 32 The Euler equations Numerical tests Conclusion and perspectives ◮ A suitable discretization of the convection terms allows to obtain a (local) kinetic energy conservation identity (for colocated as well as staggered meshes). ◮ The staggered scheme is stable at all Mach numbers, able to compute solutions of inviscid flows. ◮ It has been used successfully for two-phase flows . To do: ◮ Recover the entropy inequality for the Euler equations. ◮ Explicit schemes: first positive results for the barotropic equations. On going tests for Euler (Ph. D. T.T. Nguyen). ◮ Higher order (entropy viscosity by Guermond) ◮ Pass to the limit in the scheme (for fixed mesh) as the Mach number tends to 0... ◮ Continue the discrete non linear analysis as far as possible... ◮ Reactive flows R. Herbin (LATP) Maillages décalés et écoulements compressibles Paris 6, Février 2012 33 / 32