Weierstrass Institute for Applied Analysis and Stochastics Numerical Strategies for Electrolyte Simulations J. Fuhrmann Mohrenstraße 39 · 10117 Berlin · www.wias-berlin.de PEM2013 · NTNU Trondheim · 2013-10-03 Coupled flow and ion transport without electroneutrality assumption Relevant in a significant number of applications: PEM nanopores nanofluidic devices ion channels in biomembranes porous electrode microstructures overlimiting current in electrodialysis Search for numerical approach which takes into account recent work on excluded volume models incorporates the knowledge from semiconductor device simulation (unconditionally stable Scharfetter-Gummel scheme) guarantees mass conservation due to discrete divergence condition Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 2 Problem formulation: Nernst-Planck-Poisson system Mixture of N species; species N : neutral solvent; diagonal Onsager coefficients To be closed by constitutive relationship for cα −∇ε0 εr ∇φ = q ∂t cα + ∇ · (cα v + Nα ) = 0 α = 1...N −1 Dα Mα Nα = − cα ∇µα − ∇µN + zα F∇φ RT MN Dα cα (∇µ̃α + zα F∇φ ) =− RT φ cα µα µ̃α = µα − Nα v q zα Mα Dα Mα MN µN V mol/m3 J/mol J/mol mol/(m2 · s) m/s As/m3 1 kg/mol m2 /s α = 1...N −1 α = 1...N −1 electrostatic potential molar density (concentration) chemical potential (α = 1 . . . N ) effective chemical potential (α = 1 . . . N − 1) molar diffusion flux barycentric velocity space charge density charge number molar mass diffusion coefficient deGroot/Mazur 1962; Dreyer/Guhlke/Müller PCCP 2013 Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 3 Problem formulation: Navier-Stokes equations Incompressible flow of mixture: evolution of barycentric velocity v ∂t (ρv) + ∇ · (ρv ⊗ v) − η∆v + ∇p = −q∇φ ∂t ρ + ∇ · (ρv) = 0 N−1 ρ = MN c + ∑ (Mα − MN )cα α=1 v p η ρ c = ∑Nα=1 cα m/s Pa Pas kg/m3 mol/m3 barycentric velocity pressure viscosity density (constant) summary concentration deGroot/Mazur 1962; Dreyer/Guhlke/Müller PCCP 2013 Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 4 Constitutive relationship for µ : dilute solution theory ideal dilute solution ions as point charges µN = 0 µα = µ̃α = µα◦ + RT ln cα c (α = 1 . . . N − 1). Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 5 Constitutive relationship for µ : dilute solution theory ideal dilute solution ions as point charges µN = 0 µα = µ̃α = µα◦ + RT ln cα c (α = 1 . . . N − 1). Unphysically high concentrations in polarization boundary layer! Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 5 Constitutive relationship for µ : excluded volume models Account for energy needed for species to enter already crowded configuration: µN = 0 µα = µ̃α = µα◦ + RT ln 1 cα + ln c 1−Φ (α = 1 . . . N − 1) Φ: Summary volume fraction of the dissolved species: N−1 Φ= ∑ vα cα , α=1 vα : partial molar volume E.g. estimate from spherical packing: 4 vα = 8NA πrα3 . 3 E.g. all molecules share one one lattice (Bikerman model) 1 vα = v = , c cN 1−Φ = c Bikerman, Phil. Mag. 1942; Paunov, J. Coll. Interf. Sci. 1996; Biesheuvel/v.Soestbergen, J. Coll. Interf. Sci. 2007 Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 6 Constitutive relationship for µ : pressure correction due to Dreyer et al. Derivation from nonequilibrium thermodynamics leads to a constitutive relationship which links pressure, concentrations and chemical potential: cα 1 µα = µα◦ + (p − p◦ ) + RT ln c c cα Mα (p − p◦ ) Mα 1 ◦ µ̃α = µα + RT ln + 1− + RT ln c MN c MN 1−Φ (α = 1 . . . N) (α = 1 . . . N − 1) Gibbs Duhem + Thermodynamic equilibrium ⇒ 1 cN 1 (p − p◦ ) = −RT ln = RT ln c c 1−Φ ⇒ Equivalent to Bikerman model (v = 1c ): in thermodynamical equilibrium for equal molar masses Limitations (currently under consideration in Dreyer group): constant summary concentration c no notation yet for different molar volumes Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 7 Dreyer/Guhlke/Müller PCCP 2013 Mechanical equilibrium How to use pressure correction wihtout Navier-Stokes ? ∂t v = 0 ∇v = 0 ⇒ ∇p = −q∇φ Taking the divergence gives as second order equation for p: −∆p = ∇ · q∇φ Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 8 Constitutive relationship for µ : Fermi statistics See semiconductor device theory: cα = cG . G : (approximations of) Fermi-Dirac integrals of different orders: G (η) ≈ F j (η) = µα − µα◦ RT 1 Γ( j + 1) Z ∞ 0 ξj dξ 1 + exp(ξ − η) Special case: Fermi integral of order −1 F−1 (η) = 1 1 + exp(−η) ⇒ µ̃α = µα = µα◦ + RT ln cα 1 + RT ln c 1 − ccα (α = 1 . . . N − 1) Resembles the excluded volume expression restricted to species α . No interaction term between different species Kornyshev 1981, Landstorfer 2011 Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 9 Concentration based formulation µα → −∞ for small concentrations, especially in depletion regions Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 10 Concentration based formulation µα → −∞ for small concentrations, especially in depletion regions Dilute solution (Boltzmann) → classical Nernst Planck NNP α zα F = −Dα ∇cα + cα ∇φ RT Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 10 Concentration based formulation µα → −∞ for small concentrations, especially in depletion regions Dilute solution (Boltzmann) → classical Nernst Planck NNP α zα F = −Dα ∇cα + cα ∇φ RT Excluded volume model (Bikerman if vβ = v = 1c ): N−1 Nα = Dα ∇cα − cα 1− ∑ vβ cβ β =1 ! N−1 F ∑ vβ ∇cβ − cα zα RT ∇φ β =1 Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 10 ! Concentration based formulation µα → −∞ for small concentrations, especially in depletion regions Dilute solution (Boltzmann) → classical Nernst Planck NNP α zα F = −Dα ∇cα + cα ∇φ RT Excluded volume model (Bikerman if vβ = v = 1c ): N−1 Nα = Dα ∇cα − cα 1− ∑ vβ cβ β =1 ! N−1 F ∑ vβ ∇cβ − cα zα RT ∇φ β =1 Pure pressure correction : zα F 1 Nα = −Dα ∇cα + cα ∇p + cα ∇φ cRT RT Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 10 ! Concentration based formulation µα → −∞ for small concentrations, especially in depletion regions Dilute solution (Boltzmann) → classical Nernst Planck NNP α zα F = −Dα ∇cα + cα ∇φ RT Excluded volume model (Bikerman if vβ = v = 1c ): N−1 Nα = Dα ∇cα − cα 1− ! ∑ vβ cβ β =1 N−1 F ∑ vβ ∇cβ − cα zα RT ∇φ β =1 Pure pressure correction : zα F 1 Nα = −Dα ∇cα + cα ∇p + cα ∇φ cRT RT Fermi statistics: F Nα = −Dα dα ∇cα + cα zα ∇φ RT 1 cα G =F−1 = dα = 1 − ccα cG 0 G −1 ccα Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 10 ! Concentration based formulation µα → −∞ for small concentrations, especially in depletion regions Dilute solution (Boltzmann) → classical Nernst Planck NNP α zα F = −Dα ∇cα + cα ∇φ RT Wrong in boundary layer. Excluded volume model (Bikerman if vβ = v = 1c ): N−1 Nα = Dα ∇cα − cα 1− ! ∑ vβ cβ β =1 Pure pressure correction : Complicated flux coupling. N−1 F ∑ vβ ∇cβ − cα zα RT ∇φ β =1 ! Insufficient pressure regularity in Navier-Stokes. zα F 1 Nα = −Dα ∇cα + cα ∇p + cα ∇φ cRT RT Fermi statistics: F Nα = −Dα dα ∇cα + cα zα ∇φ RT 1 cα G =F−1 = dα = 1 − ccα cG 0 G −1 ccα Degenerates for cα → c. Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 10 Activity based formulation Expression of effective chemical potential similar to dilute solution case defines activity : µ̃α = µ̃α◦ + RT ln aα Activity coefficient γα : aα = γα 1 γα cα c ⇒ cα = c aα = cβα aα γα : inverse activity coefficient βα = Nernst-Planck-Poisson transforms to: ∂t (cβα aα ) + ∇ · (cβα aα v + Nα ) = 0 α = 1...N −1 −∇ε0 εr ∆φ = q F Nα = −Dα cβα ∇aα + aα zα ∇φ RT N−1 q = cF ∑ zα βα aα α=1 Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 11 α = 1...N −1 Inverse activity coefficients Dilute solution: βα = 1 Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 12 Inverse activity coefficients Dilute solution: βα = 1 Volume exclusion models βα = β = 1 1 + c ∑N−1 i=1 vi ai vi = 1c = 1 N−1 1 + ∑i=1 ai Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 12 Inverse activity coefficients Dilute solution: βα = 1 Volume exclusion models βα = β = vi = 1c 1 1 + c ∑N−1 i=1 vi ai = 1 N−1 1 + ∑i=1 ai Pure pressure correction model βα = exp p − p◦ cRT Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 12 Inverse activity coefficients Dilute solution: βα = 1 Volume exclusion models βα = β = 1 + c ∑N−1 i=1 vi ai = 1 N−1 1 + ∑i=1 ai Pure pressure correction model βα = exp vi = 1c 1 p − p◦ cRT Full pressure correction model (weighted with volume exclusion): Mα p − p◦ 1− MN βα = exp cRT N−1 1− ∑ vi βi cai i=1 Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 12 ! Mα MN Inverse activity coefficients Dilute solution: βα = 1 Volume exclusion models βα = β = 1 + c ∑N−1 i=1 vi ai p − p◦ cRT Full pressure correction model (weighted with volume exclusion): Mα p − p◦ 1− MN βα = exp cRT 1 N−1 1 + ∑i=1 ai = Pure pressure correction model βα = exp vi = 1c 1 N−1 1− ∑ vi βi cai i=1 Fermi statistics: βα (aα ) = G (ln aα ) aα G =F−1 = 1 1 + aα Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 12 ! Mα MN Inverse activity coefficients Dilute solution: βα = 1 Volume exclusion models No cross-coupling of gradients. βα = β = 1 + c ∑N−1 i=1 vi ai 1 N−1 1 + ∑i=1 ai = Pure pressure correction model No pressure gradient. βα = exp vi = 1c 1 p − p◦ cRT Full pressure correction model (weighted with volume exclusion): Mα p − p◦ 1− MN βα = exp cRT N−1 1− ∑ vi βi cai i=1 No pressure gradient. Fermi statistics: βα (aα ) = G (ln aα ) aα G =F−1 = 1 1 + aα No degeneration. Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 12 ! Mα MN Equilibrium case: Nonlinear Poisson equation Mechanical and thermodynamical equilibrium: Nα = 0: ∇µ̃α = −zα F∇φ ψα : (constant) quasi-Fermi potential Nonlinear Poisson equation: ⇒ µ̃α = zα F(ψα − φ ) N−1 −∇ε0 εr ∇φ = Fc ∑ zα βα aα α=1 aα = exp zα F (ψα − φ ) RT Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 13 Generalizations of Poisson-Boltzmann “classical” Poisson-Boltzmann: N−1 −∇ε0 εr ∇φ = Fc ∑ α=1 zα exp zα F (ψα − φ ) RT Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 14 Generalizations of Poisson-Boltzmann “classical” Poisson-Boltzmann: N−1 −∇ε0 εr ∇φ = Fc ∑ zα exp α=1 zα F (ψα − φ ) RT Excluded volume model (Bikerman): −∇ε0 εr ∇φ = Fc ∑N−1 α=1 zα exp zα F RT (ψα 1 + c ∑N−1 α=1 vα exp −φ) zα F RT (ψα Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 14 −φ) Generalizations of Poisson-Boltzmann “classical” Poisson-Boltzmann: N−1 −∇ε0 εr ∇φ = Fc ∑ zα exp α=1 Excluded volume model (Bikerman): −∇ε0 εr ∇φ = Fc zα F (ψα − φ ) RT ∑N−1 α=1 zα exp zα F RT (ψα 1 + c ∑N−1 α=1 vα exp −φ) zα F RT (ψα −φ) Pure pressure correction (Dreyer et al.): N−1 zα F p − p◦ −∇ε0 εr ∇φ = q = exp − cF ∑ zα exp (ψα − φ ) cRT RT α=1 ∆p = −∇ · q∇φ Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 14 Generalizations of Poisson-Boltzmann “classical” Poisson-Boltzmann: N−1 −∇ε0 εr ∇φ = Fc ∑ zα exp α=1 Excluded volume model (Bikerman): −∇ε0 εr ∇φ = Fc zα F (ψα − φ ) RT ∑N−1 α=1 zα exp zα F RT (ψα 1 + c ∑N−1 α=1 vα exp −φ) zα F RT (ψα −φ) Pure pressure correction (Dreyer et al.): N−1 zα F p − p◦ −∇ε0 εr ∇φ = q = exp − cF ∑ zα exp (ψα − φ ) cRT RT α=1 ∆p = −∇ · q∇φ Poisson-Fermi (G = F−1 ): −φ) . = Fc ∑ zα F (ψα − φ ) α=1 1 + exp RT N−1 −∇ε0 εr ∇φ = zα exp zα F RT (ψα Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 14 Numerics: Voronoi boxes + time grid Definition Voronoï box K around xK ∈ point set P : Set of points x ∈ Ω which are closer to xK than to any other point xL of P . The set of Voronoï boxes is called Voronoï diagram Voronoï diagram and Delaunay triangulation are dual to each other: Vertices of the Voronoï boxes are triangle cirumcenters. Green: Voronoï boxes Black: Delaunay triangles. Voronoï box boundaries are straight line segments orthogonal to the corresponding triangle edges. Similar construction in 3D Time grid in [0, T ]: 0 = t0 < t1 < · · · < tn−1 < tn < · · · < T Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 15 Numerics: Voronoi finite volume method Flux equation: n Zt Z 0= (∂t c + ∇ · N) dx dt t n−1 K n n Zt Z = Zt Z ∂t c dxdt + t n−1 K Z = K n N · n ds t n−1 ∂ K n−1 (c − c Zt ) dxdt + ∑ Z L neighbor of K n−1 ∂ K∩∂ L t Approximation step (NKL : flux projection onto edge xK xL ) |K| n cnK − cn−1 n K + =0 ∑ |∂ K ∩ ∂ L|NKL t n − t n−1 L neighbor of K Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 16 N · nKL ds Numerics: Voronoi finite volume method Flux equation: n Zt Z 0= (∂t c + ∇ · N) dx dt t n−1 K n n Zt Z = Zt Z ∂t c dxdt + t n−1 K Z = Zt ) dxdt + ∑ cnK − cn−1 n K + =0 ∑ |∂ K ∩ ∂ L|NKL t n − t n−1 L neighbor of K Poisson equation (EKL : electric field projection onto edge xK xL ): ∑ L neighbor of K n Z L neighbor of K n−1 ∂ K∩∂ L t Approximation step (NKL : flux projection onto edge xK xL ) |K| n−1 (c − c K n N · n ds t n−1 ∂ K n |∂ K ∩ ∂ L|EKL = |K|qnK Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 16 N · nKL ds Numerics: Edge fluxes Electric field projection: n EKL = εε0 φKn − φLn . xK − xL Flux field projection using ansatz from semiconductor devices ξ zF (B(ξ ) = exp(ξ )−1 : Bernoulli function, Z = RT ): n NKL = β̄KL D(B(Z(φLn − φKn ))anK − B(Z(φKn − φLn ))anL ) Consistent to the thermodynamic equilibrium: For any given constant value of ψ , assuming NKL = 0 we arrive at aK B(Z(φK − φL )) exp(Z(φL − φK )) − 1 = =− aL B(Z(φL − φK )) exp(Z(φK − φL )) − 1 exp(ZφL ) exp(−ZφK ) − exp(−ZφL ) =− · exp(ZφK ) exp(−ZφL ) − exp(−ZφK ) exp(ZφL ) exp(Z(ψ − φK )) = = exp(ZφK ) exp(Z(ψ − φL )) a=,K = a=,L Scharfetter/Gummel, IEEE Trans. Electron. Dev 1969 Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 17 1D, Binary electrolyte, Nonlinear Poisson Ω = [0, L]. Boundary conditions: φ |x=0 = φ0 , φ |x=L = φ∞ = 0, −∇p · n|x=0 = q∇φ · n|x=0 , p|x=L = p∞ = 0. Quasi-Fermi potentials ψα : obtained from given concentration values cα |x=L = cα,∞ << c such that q|x=L = F ∑N−1 α=1 zα cα,∞ = 0. Symmetric 1:1 electrolyte with bulk solution molarity cα,∞ = c∞ for α = 1, 2 Summary concentration c set to the molarity of water c = 55.508 · mol/dm3 . (Double layer) charge: Z L Qdl = Qdl (φ0 ) = q dx 0 Differential (double layer) capacitance: Cdl = Q (φ0 + δ ) − Qdl (φ0 ) dQdl ≈ dl , dφ0 δ δ << φ0 Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 18 Gouy-Chapman (+Stern) differential double layer capacitance Dilute solution with uncorrected Nernst-Planck equation ⇒ “Gouy-Chapman theory” Guoy-Chapman 700 Guoy-Chapman-Stern 0.001M 0.01M 0.1M 1M 600 120 500 100 Cdl/µFcm-2 Cdl/µFcm-2 0.001M 0.01M 0.1M 1M 140 400 300 80 60 200 40 100 20 0 0 0 50 100 150 ∆φ/mV 200 250 300 0 200 400 600 800 1000 ∆φ/mV Numerically obtained differential double layer capacitances curves according to the Gouy-Chapman model (left) and the Gouy-Chapman-Stern model (right; OHP at x = 0.5nm ) for aqueous 1:1 electrolytes at 25◦ C matching the classical results Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 19 Binary 1:1 electrolyte: Bikerman vs Fermi vs Pressure correction Bikerman vs. pressure correction 600 400 Bikerman, 0.001M Bikerman, 0.01M Bikerman, 0.1M Bikerman, 1M Fermi, 0.001M Fermi, 0.01M Fermi, 0.1M Fermi, 1M 500 400 Cdl/µFcm-2 500 Cdl/µFcm-2 Bikerman vs. Fermi 600 Bikerman, 0.001M Bikerman, 0.01M Bikerman, 0.1M Bikerman, 1M PC, 0.001M PC, 0.01M PC, 0.1M PC, 1M 300 300 200 200 100 100 0 0 0 200 400 600 800 1000 0 200 400 ∆φ/mV 600 800 1000 ∆φ/mV Comparison of the differential double layer capacitance curves between the Bikerman model with vα = 1c and the pure pressure correction model (left) resp. the Fermi-Dirac model of index -1 (right). Bikerman and pressure correction are equivalent in equilibrium due to Gibbs-Duhem Bikerman and Fermi (-1) are indistinguisheable as in β = 1+a11 +a2 either both activities are small or due to different charge one is large and the other is small. Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 20 Profiles Concentration profile for c60 c=55.5 50 40 Bikerman Dreyer et al. Fermi Gouy-Chapman Gouy-Chapman-Stern 400 φ/mV c/(mol/dm3) Voltage Profile 500 Bikerman Dreyer et al. Fermi Gouy-Chapman Gouy-Chapman-Stern 30 20 300 200 100 10 0 0 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 x/nm 0.6 0.8 1 x/nm Comparison of negative ion concentration (left) and potential profile (right) for the different models. The Debye length is λD = 3.04nm. Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 21 Spherical coordinates Differential double layer capacitance for spherical electrode 0.01 mol/dm3 binary electrolyte, Bikerman model with excluded volume =1/csolute 600 r=1nm r=2nm r=4nm r=8nm planar 600 580 500 560 540 520 Cdl/µFcm-2 400 500 200 220 240 260 280 300 300 100 80 200 60 40 20 100 0 0 20 40 60 80 100 0 0 200 400 600 800 Δφ/mV Differential double layer capacitances for different electrode radii Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 22 1000 Dynamic calculations Double Layer Charging Current 100 50 - c Bikerman c PC c GC 40 3 c/(mol/dm ) 80 I/(A/cm2) Evolution of Maximum Concentration 60 c=55.5 IBikerman IPC IGC 60 40 30 20 20 10 0 1e-10 1e-08 1e-06 0.0001 0.01 1 0 1e-10 1e-08 1e-06 t/s 0.0001 0.01 t/s Evolution of current (left) and maximum negative ion concentration (right). Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 23 1 Electrolytic diode, binary 1:1 electrolyte, v = 0 εε0 ∂n φ = −σ εε0 ∂n φ = σ c1,2 = cbulk φ =0 p=0 c1,2 = cbulk φ = φbias p=0 If not stated otherwise, homogeneous Neumann boundary conditions: ∂n φ = 0, ∂n p = 0, N1,2 · n = 0, σ = 500µAs/m2 IV curve for fluidic diode 1M solution 7000 6000 Guoy-Chapman Bikerman v=1/c 5000 I/mA 4000 3000 2000 1000 0 -1000 -4 -2 0 2 4 Δφ/V Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 24 Electrolytic diode: distributions (Bikerman model, 0.01M solution,σ = 150µAs/m2 ) φbias = −2V φbias = 2V φ c+ c− Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 25 Electrolytic diode: maximum concentrations + pressure (σ = 150µAs/m2 ) Maximum concentration Grid refinement level 4, 0.1M solution Maximum concentration Grid refinement level 6, 0.1M solution 160 180 140 160 140 Guoy-Chapman 100 Bikerman Dreyer et al. 120 Guoy-Chapman Bikerman 100 Dreyer et al. I/mA I/mA 120 80 80 60 c=55.5 60 c=55.5 40 40 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 -2 -1.5 -1 -0.5 Δφ/V Maximum pressure Grid refinement level 6 465 460 455 p/MPa 0 Δφ/V 450 0.1M 1M 445 440 435 430 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 Δφ/V Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 26 0.5 1 1.5 2 Coupling to Navier-Stokes Stationary distribution of single neutral species in Navier-Stokes velocity field v Mass balance ∇·N = 0 N = −D∇c + cv Fick’s law + convection Problem: coupling to Navier-Stokes while maintaing concentration maximum principle. Stabilized finite elements do not guarantee this property! Voronoi finite volumes with discrete Scharfetter-Gummel (Il’in) flux: v v KL KL cK − B − cL NKL = D B D D vKL : some projection of v onto edge xK xL Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 27 Discrete Minimax Principle Divergence free discrete flux: |∂ L ∩ ∂ K| |σ |vσ = 0 vKL + ∑ |xK − xL | L∈neighbors(K) σ ∈outbound(K) ∑ (DIV0) Fuhrmann/Linke/Langmach APNUM 2011 Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 28 Discrete Minimax Principle Divergence free discrete flux: |∂ L ∩ ∂ K| |σ |vσ = 0 vKL + ∑ |xK − xL | L∈neighbors(K) σ ∈outbound(K) ∑ (DIV0) Lemma: If (DIV0) is valid, for any solution (cK )K∈K : 1. Global discrete minimax principle: 0 ≤ cK ≤ cinlet ∀K ∈ K Fuhrmann/Linke/Langmach APNUM 2011 Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 28 Discrete Minimax Principle Divergence free discrete flux: |∂ L ∩ ∂ K| |σ |vσ = 0 vKL + ∑ |xK − xL | L∈neighbors(K) σ ∈outbound(K) (DIV0) ∑ Lemma: If (DIV0) is valid, for any solution (cK )K∈K : 1. Global discrete minimax principle: 0 ≤ cK ≤ cinlet ∀K ∈ K 2. Local discrete minimax principle: min L∈neighbors(K) cL ≤ cK ≤ max L∈neighbors(K) cL ∀K ∈ K 0 Fuhrmann/Linke/Langmach APNUM 2011 Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 28 Discrete Minimax Principle Divergence free discrete flux: |∂ L ∩ ∂ K| |σ |vσ = 0 vKL + ∑ |xK − xL | L∈neighbors(K) σ ∈outbound(K) (DIV0) ∑ Lemma: If (DIV0) is valid, for any solution (cK )K∈K : 1. Global discrete minimax principle: 0 ≤ cK ≤ cinlet ∀K ∈ K 2. Local discrete minimax principle: min L∈neighbors(K) cL ≤ cK ≤ max L∈neighbors(K) cL ∀K ∈ K 0 3. The system matrix has the M-Property Fuhrmann/Linke/Langmach APNUM 2011 Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 28 Divergence free discrete fluxes How to guarantee this condition when v comes from the solution of a flow problem ? Continuous divergence free velocity field v: 1 R Exact calculation of vKL = |K∩L| K∩L v(s) · (xK − xL )ds: Exact solutions: Hagen - Poiseuille et al Pointwise divergence free finite elements (e.g. Scott Vogelius for Navier-Stokes) Burman/Linke, App. Num. Math 58(2008)11,1704-1719 Postprocessed Crouzeix-Raviart mixed finite elements (A. Linke 2013, submitted) Finite volume scheme for flow problem including discrete divergence free fluxes Finite volume solution for Navier-Stokes (Eymard/Fuhrmann/Linke 2011, submitted) insufficient convergence properties Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 29 Divergence free Scott-Vogelius finite elements div(velocity space) ⊂ pressure space Lowest order Scott Vogelius elements: (Pd , P−(d−1) ) (d:space dimension) Stable on macro triangulations Arnold/Qin, Proc. IMACS 1992 Burman/Linke, App. Num. Math 2008 Linke, PhD Thesis, FU Berlin 2008 Maintain two independent discretizations for transport (FV) and for flow (FE) (FV): For every simplex S, calculate simplicial contributions σKL;S = ∂ K ∩ ∂ L ∩ S to ∂ K ∩ ∂ L (FE): Calculate velocity projections qKL;S from continuous FE velocity field (FV): Assemble qKL from qKL;S Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 30 Cylindrical DEMS flow cell working chamber working electrode Ar flux electrolyte inlet teflon spacer electrolyte outlet reference and counter electrode + counter electrode MS compartment six capillaries six capillaries connecting electrochemical and mass spectrometrical compartments capillary for electrolyte inlet electrolyte inlet reference and counter electrodes mass spectroscope Ar flux teflon spacer electrolyte outlet + counter electrode holes for screws Jusys/Massong/Baltruschat, J. Electrochem. Soc. 1999 Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 31 Numerical solution ansatz Navier-Stokes solution by Taylor-Hood or Scott Vogelius finite elements (implemented in Alberta) Voronoi finite volume solution implemented in pdelib (WIAS) Grid generation by “extrusion” of a 2D Delaunay mesh (created by triangle) in z direction in order to guarantee proper anisotropic alignment at anode Calculation on 61 of the cell geometry, together with stubs of inlet and outlet channels x inlet x capillary Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 32 Comparison to experiment in a 3D cylindrical cell Limiting Current Ilim / mA 1 H=50µm H=75µm H=100µm Measured 0.17u1/3 0.1 1 10 100 3 Flow Rate u /(mm /s) Measured and calculated values for the limiting current using fitted data from channel flow cell experiment for different values of working chamber height. No “simple” asymptotic expression available. Fuhrmann/Linke/Langmach/Baltruschat, Electrochim. Acta 2009 Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 33 Flow regimes 6.5 0.5mm3 /s 6.0 5.5 5.0 4.5 4.0 3.5 3.0 2.5 2.0 1.5 1.0 0.5 10mm3 /s 80mm3 /s Concentration isolevels for different inlet flow rates Fuhrmann/Linke/Langmach/Baltruschat, Electrochim. Acta 2009 Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 34 Scott Vogelius vs. Taylor Hood (coarse grid) Scott Vogelius Taylor Hood Taylor-Hood velocity field does not guarantee discrete divergece condition ⇒ violation of maximum principle Main measured effects concentrated in boundary layer ⇒ tolerable errors with Taylor Hood, if grid is fine enough. Fuhrmann/Linke/Langmach/Baltruschat Electrochim. Acta 2009 Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 35 Outlook Implement Nernst-Planck-Poisson-Navier-Stokes coupling using divergence free FEM ansatzes of A. Linke based on postprocessed Crouzeix-Raviart updated version of Nernst-Planck including different molar volumes nanopore calculation (model air electrode in DEMS cell, PEM) nanofluidic devices, electrodialysis, biomembranes etc. degenerate semiconductors (OLED, nitride) Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 36 Acknowledgement German Federal Ministry of Education and Research (BMBF) Research initiative on energy storage Collaborative project “Perspectives for rechargeable magnesium air batteries” H. Baltruschat (Bonn), Th. Bredow (Bonn), J. Behm (Ulm), M. Wachtler (Ulm) Grant 03EK3027D Numerical Strategies · PEM2013 · NTNU Trondheim · 2013-10-03 · Page 37