Diffusion models for mixtures using a stiff dissipative hyperbolic

Mar 7, 2018 - Cauchy problem for Maxwell-Stefan's equations [Giovangigli, Bothe, Jüngel &. Stelzer]. ▷ Using .... Wη(W) is a positive definite quadratic form.
1MB taille 2 téléchargements 239 vues
Diffusion models for mixtures using a stiff dissipative hyperbolic formalism Bérénice Grec1 in collaboration with L. Boudin, V. Pavan 1

MAP5 – Université Paris Descartes, France

Workshop on kinetic and fluid Partial Differential Equations March 7th, 2018

Outline of the talk

1

Gaseous mixtures: macroscopic models

2

Stiff dissipative hyperbolic formalism

3

Entropy and equilibrium

4

Local equilibrium approximation

5

Conclusion and prospects

12 :29

Bérénice Grec

Stiff dissipative hyperbolic formalism for diffusion for mixtures

1/16

Diffusion models for mixtures: Fick/Maxwell-Stefan I

Mixture of p ≥ 2 species

I

ρi : mass density of species i

I

Ni = ρi ui : momentum of species i     ρ1 N1  ..   ..  ρ =  . , N =  . 

I

ρp I

Fick law: N = −F (ρ)∇x ρ Maxwell-Stefan equations:

Np

−∇x ρ = S(ρ)N

Mass conservation: ∂t ρ + ∇ · N = 0

Properties of matrices F and S I

F (ρ) and S(ρ) are not invertible (rank p − 1).

I

Cauchy problem for Maxwell-Stefan’s equations [Giovangigli, Bothe, Jüngel & Stelzer]

I

Using Moore-Penrose pseudo-inverse: structural similarity

12 :29

Bérénice Grec

Stiff dissipative hyperbolic formalism for diffusion for mixtures

2/16

Fick vs. Maxwell-Stefan (macroscopic point of view) Formal analogy of the two systems, but Fick and Maxwell-Stefan are not obtained in the same way

Obtention of the Fick law I

Thermodynamics of irreversible processes (entropy decay) [Onsager]

I

Thermodynamical considerations on fluxes, written as linear combinations of potential gradients

I

Stems from mass equations

Obtention of the Maxwell-Stefan equations I

Mechanical considerations on forces (equilibrium of pressure and friction forces)

I

Assumption: different species have different macroscopic velocities on macroscopic time scales

I

Stems from momentum equations

12 :29

Bérénice Grec

Stiff dissipative hyperbolic formalism for diffusion for mixtures

3/16

Fick vs. Maxwell-Stefan (kinetic point of view) Formal analogy of the two systems, but Fick and Maxwell-Stefan are not obtained in the same way

Perturbative method (Fick) I

Based on the Chapman-Enskog expansion [Bardos, Golse, Levermore], [Bisi, Desvillettes]

Moment method (Maxwell-Stefan) I

Based on the ansatz that the distribution functions are at local Maxwellian states [Levermore], [Müller, Ruggieri]

The Maxwell-Stefan equations can be written X ρi ρj (uj − ui ) 1 , − ∇x ρ i = mi Dij j6=i

mi : molecular mass of species i, Dij > 0 symmetric. 12 :29

Bérénice Grec

Stiff dissipative hyperbolic formalism for diffusion for mixtures

4/16

Outline of the talk

1

Gaseous mixtures: macroscopic models

2

Stiff dissipative hyperbolic formalism

3

Entropy and equilibrium

4

Local equilibrium approximation

5

Conclusion and prospects

12 :29

Bérénice Grec

Stiff dissipative hyperbolic formalism for diffusion for mixtures

4/16

Stiff dissipative model for mixtures For any species i with velocity u i , we write mass and momentum conservation  ∂t ρi + ∇x · (ρi u i ) = 0, 1 ∂t (ρi u i ) + ∇x · (ρi u i ⊗ u i + Pi (ρi )) + Ri = 0 ε I I

Ideal gas law for the partial pressure Pi (ρi ) = ρi kB T /mi Relaxation term: friction force exerted by the mixture on species i Ri =

X

aij ρi ρj (u j − u i ) =

j6=i

X

αij (u j − u i ) =

j6=i

p X

αij u j .

j=1

Using the formalism of [Chen, Levermore, Liu, CPAM, ’94] Obtain a reduced system involving the aligned velocity u when ε remains small P  ( ∇x Pj p ∂t ρi + ∇x · (ρi u) = ε∇x · ` , ij j=1 ρj ∂t (ρu) + ∇x · (ρu ⊗ u) + ∇x P = 0, where P = 12 :29

P

j

Pj (ρj ) is the total pressure, and (`ij ) are real constants. Bérénice Grec

Stiff dissipative hyperbolic formalism for diffusion for mixtures

5/16

Stiff dissipative model for mixtures For any species i with velocity u i , we write mass and momentum conservation  ∂t ρi + ∇x · (ρi u i ) = 0, 1 ∂t (ρi u i ) + ∇x · (ρi u i ⊗ u i + Pi (ρi )) + Ri = 0 ε I I

Ideal gas law for the partial pressure Pi (ρi ) = ρi kB T /mi Relaxation term: friction force exerted by the mixture on species i Ri =

X

aij ρi ρj (u j − u i ) =

j6=i

X

αij (u j − u i ) =

j6=i

p X

αij u j .

j=1

Using the formalism of [Chen, Levermore, Liu, CPAM, ’94] This approach has been used in previous papers for other systems: I

[Kawashima], [Yong], [Kawashima,Yong], ...

I

[Giovangigli, Matuszewski], [Giovangigli, Yong], ...

12 :29

Bérénice Grec

Stiff dissipative hyperbolic formalism for diffusion for mixtures

5/16

Vectorial expression of the system 1 ∂t W + ∇x · F(W) + R(W) = 0 ε I

(∗)

Unknown W = [W1 , . . . , Wp , W p+1 , · · · , W 2p ] ∈ (R∗+ )p × Rdp , with Wi = ρi , W p+i = ρi u i

the k-th column of F(W) and R(W) are given by   W p+1 · e k   0p×1   ..   .       p   W · e  X 2p k W p+j        α1j      W   j  , R(W) =  j=1 W ⊗ W Fk (W) =  . p+1 p+1   + P (W )I e   . 1 1 d k   .   W 1 .       .. p  X   W p+j  .       αpj   W 2p ⊗ W 2p W j j=1 + Pp (Wp )Id e k Wp

I

12 :29

Bérénice Grec

Stiff dissipative hyperbolic formalism for diffusion for mixtures

6/16

Aim of the work I

Adapt the formalism by Chen, Levermore and Liu in the gaseous mixture framework

I

Limiting behavior for small ε Derive an approximation of the local equilibrium and its first-order correction

I

I

Build a relevant entropy which ensures...

I

... the hyperbolicity of the local equilibrium approximation...

I

... and the dissipativity of its first-order correction

Analogy with the kinetic theory I

relaxation time ε ←→ mean free path

I

unknown W ←→ distribution function f

I

friction relaxation term R ←→ collision operator Q (with dissipativity and collisional invariant properties)

I

local equilibria ←→ Maxwellian functions

12 :29

Bérénice Grec

Stiff dissipative hyperbolic formalism for diffusion for mixtures

7/16

Outline of the talk

1

Gaseous mixtures: macroscopic models

2

Stiff dissipative hyperbolic formalism

3

Entropy and equilibrium

4

Local equilibrium approximation

5

Conclusion and prospects

12 :29

Bérénice Grec

Stiff dissipative hyperbolic formalism for diffusion for mixtures

7/16

Building the entropy 1 ∂t W + ∇x · F(W) + R(W) = 0 ε

(∗)

Entropy for Equation (∗) The function η defined by 2

1 W p+i kB T η(W) = + Ei (Wi ), with Ei (Wi ) = 2 Wi mi

    Wi − Wi Wi ln Wi0

is a strictly convex entropy for Equation (∗), i.e. I

∇2W η(W)∇W Fk (W) is symmetric for any k

I

∇W η(W) · R(W) ≥ 0

I

∇2W η(W) is a positive definite quadratic form Choice of the internal energy Ei00 (Wi ) = Pi0 (Wi )/Wi Nonnegativity of the matrix A = (αij )

12 :29

Bérénice Grec

Stiff dissipative hyperbolic formalism for diffusion for mixtures

8/16

Conserved quantities The matrix





0p×dp

Ip

Q= 0d×p

(p+d)×(p+dp)    ∈R Id , · · · , Id

satisfies QR(W) = 0(p+d)×1 . It allows to define " w = QW = W1 , · · · , Wp ,

p X

#| W |p+j

∈ Rp+d

j=1

the p + d independent conserved quantities.  | Conversely, to any w = w1 , · · · , wp , w |p+1 , we associate, via the equilibrium function E, the equilibrium Weq = E(w) such that R(Weq ) = 0, where |  p X wp w1 | | E(w) = w1 , · · · , wp , w ,··· , w , with σ(w) = wi . σ(w) p+1 σ(w) p+1 i=1 12 :29

Bérénice Grec

Stiff dissipative hyperbolic formalism for diffusion for mixtures

9/16

Characterization of the local equilibria Equilibrium function  | p X w1 | wp E(w) = w1 , · · · , wp , w p+1 , · · · , w |p+1 , with σ(w) = wi . σ(w) σ(w) i=1 Characterization of the equilibrium I I

I

R(Weq ) = 0 there exists u ∈ Rd such that Weq = [W1 , · · · , Wp , W1 u | , · · · , Wp u | ]| this corresponds to saying that all species velocities are aligned there exists v ∈ Rp+d such that ∇W η(Weq ) = v| Q

Use of the Legendre-Fenchel transform of η I

there exists v ∈ Rp+d such that Weq = ∇V η ∗ (v| Q)

In terms of the physical variables, |

E(w) = [ρ1 , · · · , ρp , ρ1 u | , · · · , ρp u | ] . Define P(w) = ∇w E(w)Q, which is a projection matrix. 12 :29

Bérénice Grec

Stiff dissipative hyperbolic formalism for diffusion for mixtures

10/16

Outline of the talk

1

Gaseous mixtures: macroscopic models

2

Stiff dissipative hyperbolic formalism

3

Entropy and equilibrium

4

Local equilibrium approximation

5

Conclusion and prospects

12 :29

Bérénice Grec

Stiff dissipative hyperbolic formalism for diffusion for mixtures

10/16

Look for an expansion W = M[w] (with w = QW), such that ∂t w + ∇x · QF(M[w]) ' 0. Then 1 1 ∂t W + ∇x · F(W) + R(W) ' (I − ∇w M[w]Q)∇x · F(M[w]) + R(M[w]). ε ε Satisfying equation (∗) leads to cancelling the RHS. Introduce the formal expansion W = M[w] = E(w) + εM(1) [w] + · · · Linearizing, order 1 in ε becomes (I − ∇w E(w)Q)∇x · F(E(w)) + ∇W R(E(w))M(1) [w] = 0 Provided that the inversion of ∇W R(E(w)) is possible in some sense, −1

M(1) [w] = − (∇W R(E(w)))

(I − ∇w E(w)Q)∇x · F(E(w)) | {z } =P(w)

and the equation on w becomes, with f(w) = QF(E(w)) h i ∂t w + ∇x · f(w) + ε∇x · Q∇W F(E(w))M(1) [w] = 0. 12 :29

Bérénice Grec

Stiff dissipative hyperbolic formalism for diffusion for mixtures

11/16

Local equilibrium approximation Formal expansion Wε = E(w) + εM(1) [w] + · · ·

Chen, Levermore and Liu’s computations The first-order correction is given by M(1) [w] = −B [Ip+dp − P(w)] ∇x · F(E(w)), with P(w) = ∇w E(w)Q. The system (∗) becomes ∂t w + ∇x · f(w) = ε∇x · g(w), where fk (w) = QFk (E(w)), gk (w) = Q∇W Fk (E(w))B [Ip+dp − P(w)] ∇x · F(E(w)), where B is the pseudo-inverse of ∇W R(E(w)) such that im B = ker Q. 12 :29

Bérénice Grec

Stiff dissipative hyperbolic formalism for diffusion for mixtures

12/16

Explicit computations Expression of the equilibrium, definition of P(w) = ∇w E(w)Q   0p×1      ∇x P1 (ρ1 ) − ρ1 ∇x P    ρ (Ip+dp − P(w))∇x · F(E(w)) =  ,   ..   .   ρp ∇x Pp (ρp ) − ∇x P ρ P P where ρ = i ρi and P = i Pi (ρi ). Computation of X = B(Ip+dp − P(w))∇x · F(E(w)): I

B pseudo-inverse of ∇W R(E(w)) such that im B = ker Q

I

ker Q =

I

X ∈ im B = ker Q thus Xi = 0 and

I

im(∇W R(E(w))) = ker Q =⇒ (Ip+dp − P(w))∇x · F(E(w)) ∈ im(∇W R(E(w)))

I

∇W R(E(w))X = (Ip+dp − P(w))∇x · F(E(w)) p X ρi X p+j = ∇x Pi (ρi ) − ∇x P αij ρj ρ



0, · · · , 0, X |p+1 , · · · , X |2p

| Pp ,

Pp j=1

i=1

X p+i = 0d×1



X p+j = 0

Pseudo-invert A = (αij )

j=1

12 :29

Bérénice Grec

Stiff dissipative hyperbolic formalism for diffusion for mixtures

13/16

Explicit computations Expression of the equilibrium, definition of P(w) = ∇w E(w)Q   0p×1      ∇x P1 (ρ1 ) − ρ1 ∇x P    ρ (Ip+dp − P(w))∇x · F(E(w)) =  ,   ..   .   ρp ∇x Pp (ρp ) − ∇x P ρ P P where ρ = i ρi and P = i Pi (ρi ). Computation of X = B(Ip+dp − P(w))∇x · F(E(w)): I

B pseudo-inverse of ∇W R(E(w)) such that im B = ker Q

I

ker Q =

I

X ∈ im B = ker Q thus Xi = 0 and

I

im(∇W R(E(w))) = ker Q =⇒ (Ip+dp − P(w))∇x · F(E(w)) ∈ im(∇W R(E(w)))

I

∇W R(E(w))X = (Ip+dp − P(w))∇x · F(E(w)) p X X p+j ρi αij = ∇x Pi (ρi ) − ∇x P ρj ρ



0, · · · , 0, X |p+1 , · · · , X |2p

| Pp ,

Pp j=1

i=1

X p+i = 0d×1



X p+j = 0

Pseudo-invert A = (αij )

j=1

12 :29

Bérénice Grec

Stiff dissipative hyperbolic formalism for diffusion for mixtures

13/16

Explicit computations Expression of the equilibrium, definition of P(w) = ∇w E(w)Q   0p×1      ∇x P1 (ρ1 ) − ρ1 ∇x P    ρ (Ip+dp − P(w))∇x · F(E(w)) =  ,   ..   .   ρp ∇x Pp (ρp ) − ∇x P ρ P P where ρ = i ρi and P = i Pi (ρi ). Computation of X = B(Ip+dp − P(w))∇x · F(E(w)): I

B pseudo-inverse of ∇W R(E(w)) such that im B = ker Q

I

ker Q =

I

X ∈ im B = ker Q thus Xi = 0 and

I

im(∇W R(E(w))) = ker Q =⇒ (Ip+dp − P(w))∇x · F(E(w)) ∈ im(∇W R(E(w)))

I

∇W R(E(w))X = (Ip+dp − P(w))∇x · F(E(w)) p X X p+j ρi αij = ∇x Pi (ρi ) − ∇x P ρj ρ



0, · · · , 0, X |p+1 , · · · , X |2p

| Pp ,

Pp j=1

i=1

X p+i = 0d×1



X p+j = 0

Pseudo-invert A = (αij )

j=1

12 :29

Bérénice Grec

Stiff dissipative hyperbolic formalism for diffusion for mixtures

13/16

Pseudo-inversion of A I

Let 1 = (1, · · · , 1)|

I

ker A = Span 1 and im A = (Span 1)⊥

I

Let r = [ρ1 , · · · , ρp ] , which is not orthogonal to 1

I

Decompositions

|

Rp = (Span 1) ⊕ (Span r)⊥ = (Span 1)⊥ ⊕ (Span r) I

Existence of a unique pseudo-inverse L = (λij )1≤i,j≤p of A with im L = (Span r)⊥ and ker L = Span r

I

L is symmetric

I

Since A = (αij ) = (ρi ρj aij ), we have that L = (λij ) =

12 :29

Bérénice Grec



1 ρi ρj `ij



Stiff dissipative hyperbolic formalism for diffusion for mixtures

14/16

End of the computations p X j=1 I

αij

X p+j = ρj

  ρi ∇x Pi (ρi ) − ∇x P ρ

|

r = [ρ1 , · · · , ρp ] spans ker L, i.e.

P

X p+i = ρi

j

λij ρj = 0

p X

λij ∇x Pj (ρj )

j=1

P Expression of the equilibrium, and the fact that i X p+i = 0 allow to compute gk (w) = Q∇W Fk (E(w))X    p  p X `1j X ∂xk Pj (ρj ) λ1j ∂xk Pj (ρj )   ρ1  j=1    j=1 ρj         .. ..     . . gk (w) =  =    p p   X  X `pj     ∂xk Pj (ρj ) λpj ∂xk Pj (ρj )  ρp   j=1   j=1 ρj 0d×1 12 :29

Bérénice Grec

0d×1 Stiff dissipative hyperbolic formalism for diffusion for mixtures

15/16

End of the computations   p X X p+i ρj = λij ∇x Pj (ρj ) − ∇x P ρi ρ j=1 I

|

r = [ρ1 , · · · , ρp ] spans ker L, i.e.

P

X p+i = ρi

j

λij ρj = 0

p X

λij ∇x Pj (ρj )

j=1

P Expression of the equilibrium, and the fact that i X p+i = 0 allow to compute gk (w) = Q∇W Fk (E(w))X    p  p X `1j X ∂xk Pj (ρj ) λ1j ∂xk Pj (ρj )   ρ1  j=1    j=1 ρj         .. ..     . . gk (w) =  =    p p   X  X `pj     ∂xk Pj (ρj ) λpj ∂xk Pj (ρj )  ρp   j=1   j=1 ρj 0d×1 12 :29

Bérénice Grec

0d×1 Stiff dissipative hyperbolic formalism for diffusion for mixtures

15/16

End of the computations   p X X p+i ρj = λij ∇x Pj (ρj ) − ∇x P ρi ρ j=1 I

|

r = [ρ1 , · · · , ρp ] spans ker L, i.e.

P

X p+i = ρi

j

λij ρj = 0

p X

λij ∇x Pj (ρj )

j=1

P Expression of the equilibrium, and the fact that i X p+i = 0 allow to compute gk (w) = Q∇W Fk (E(w))X    p  p X `1j X λ1j ∂xk Pj (ρj )  ∂xk Pj (ρj )  ρ1  j=1   j=1 ρj          .. ..     . . gk (w) =  =    p p  X  X  `pj     λpj ∂xk Pj (ρj )  ∂xk Pj (ρj ) ρp  j=1   j=1 ρj  0d×1 12 :29

Bérénice Grec

0d×1 Stiff dissipative hyperbolic formalism for diffusion for mixtures

15/16

Outline of the talk

1

Gaseous mixtures: macroscopic models

2

Stiff dissipative hyperbolic formalism

3

Entropy and equilibrium

4

Local equilibrium approximation

5

Conclusion and prospects

12 :29

Bérénice Grec

Stiff dissipative hyperbolic formalism for diffusion for mixtures

15/16

Conclusion and prospects Reduced system involving the bulk velocity u for small ε (

∂t ρi + ∇x · (ρi u) = ε∇x ·

P

∇x Pj p j=1 `ij ρj



,

∂t (ρu) + ∇x · (ρu ⊗ u) + ∇x P = 0. I I

Hyperbolicity & dissipativity Diffusion correction term of Fick’s type (on the equation of mass conservation)

I

No viscosity term on the momentum equation (convective  diffusive fluxes)

I

Maxwell-Stefan can describe a moderate rarefied regime more than Fick

Prospects I

Obtain an explicit form of Fick’s coefficients from the Maxwell-Stefan’s ones

I

Compare the experimental and theoretical relaxation times (experiments being designed at IUSTI)

I

Taking into account the non isothermal effects

12 :29

Bérénice Grec

Stiff dissipative hyperbolic formalism for diffusion for mixtures

16/16

Thank you for your attention!

12 :29

Bérénice Grec

Stiff dissipative hyperbolic formalism for diffusion for mixtures

16/16

Pseudo-inverses Proposition Let A ∈ Rp×p .Let S and T be two subspaces of Rp such that Rp = ker A ⊕ S and Rp = im A ⊕ T . Then there exists a unique matrix B such that: 1

ABA = A,

2

BAB = B,

3

ker B = T and im B = S.

This matrix B is then called the pseudo-inverse of A with prescribed range S and null space T .

Corollary Let Y ∈ im A. Then there exists a unique X ∈ im B such that AX = Y, it is given by X = BY.

Symmetric case Consider a symmetric matrix A ∈ Rp×p , and a subspace N such that Rp = im A ⊕ N. Then the only symmetric pseudo-inverse of A is the one with prescribed range N ⊥ and null space N. 12 :29

Bérénice Grec

Stiff dissipative hyperbolic formalism for diffusion for mixtures

16/16

Legendre-Fenchel transform We introduce the following domain  V = V ∈ Rp+dp | V = ∇W η(W) for some W ∈ (R∗+ )p × Rdp . The Legendre-Fenchel transform η ∗ of η is the convex function satisfying η(W) + η ∗ (V) = V · W. We can compute η ∗ (V) = V · W − η(W) =

p X kB T i=1

mi

Wi0 exp



mi kB T

  1 Vi + V 2p+i . 2

Let φ∗ : Rp+d → R, v 7→ η ∗ (v| Q). Denote by φ the Legendre-Fenchel transform of φ∗ , we compute     p X wi 1 w 2p+1 kB T wi ln − 1 + . φ(w) = mi 2 σ(w) Wi0 i=1 Then, following [Chen, Levermore, Liu] E(w) = ∇V η ∗ (∇w φ(w)| Q). 12 :29

Bérénice Grec

Stiff dissipative hyperbolic formalism for diffusion for mixtures

16/16