Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Simultaneous Space-Time Adaptive Wavelet Collocation Method for Intermittent Problems in Fluid Dynamics Jahrul Alam Department of Mathematics and Statistics McMaster University, Canada and Nicholas Kevlahan Department of Mathematics and Statistics McMaster University, Canada SHARCNET seminar series on HPC: July 21, 2004 Department of Mathematics and Statistics McMaster University, Canada •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Collaborators • O. V. Vasilyev (University of Colorado at Boulder) • D. Goldstein (University of Colorado at Boulder) •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Outline •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Outline • Motivation •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Outline • Motivation • Adaptive wavelet collocation method •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Outline • Motivation • Adaptive wavelet collocation method • Space-time discretization of PDEs •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Outline • Motivation • Adaptive wavelet collocation method • Space-time discretization of PDEs • Results and Discussion •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Outline • Motivation • Adaptive wavelet collocation method • Space-time discretization of PDEs • Results and Discussion • Conclusion and Future direction •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Outline • Motivation • Adaptive wavelet collocation method • Space-time discretization of PDEs • Results and Discussion • Conclusion and Future direction •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Motivation • Basic equations of Fluid Dynamics •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Motivation • Basic equations of Fluid Dynamics – Newtonian, incompressible – Conservation of mass and momentum •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Motivation • Basic equations of Fluid Dynamics – Newtonian, incompressible – Conservation of mass and momentum ∇·u=0 ∂u 1 + u∇ · u = − ∇P + ν∇2u ∂t ρ •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Motivation • Basic equations of Fluid Dynamics – Newtonian, incompressible – Conservation of mass and momentum ∇·u=0 ∂u 1 2 + u∇ · u = −∇P + ∇u ∂t Re • What are the typical values of Re? •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Motivation • Basic equations of Fluid Dynamics – Newtonian, incompressible – Conservation of mass and momentum ∇·u=0 ∂u 1 2 + u∇ · u = −∇P + ∇u ∂t Re • What are the typical values of Re? For turbulent flows Re ∼ 105 − 106. •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Motivation • Basic equations of Fluid Dynamics – Newtonian, incompressible – Conservation of mass and momentum ∇·u=0 ∂u 1 2 + u∇ · u = −∇P + ∇u ∂t Re • What are the typical values of Re? For turbulent flows Re ∼ 105 − 106. • Let us simplify the Navier-Stokes equation ∂u ∂u ∂ 2u +u =ν 2 ∂t ∂x ∂x •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Motivation • Computational problems of turbulence! – DOF ∼ Re3 in space-time domain • Aim of our study – Adapt the # of DOF according to intermittency •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Motivation • Computational problems of turbulence! – DOF ∼ Re3 in space-time domain • Aim of our study – Adapt the # of DOF according to intermittency Classical grid •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Motivation • Computational problems of turbulence! – DOF ∼ Re3 in space-time domain • Aim of our study – Adapt the # of DOF according to intermittency t x Classical grid We propose •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Adaptive wavelet collocation method • What are wavelets? •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Adaptive wavelet collocation method • What are wavelets? A set of basis functions that are localized in space and scale • Represent a function in terms of wavelet basis: u(x) = ∞ X X djk ψkj (x) j=0 k∈Kj •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Adaptive wavelet collocation method • What are wavelets? A set of basis functions that are localized in space and scale • Represent a function in terms of wavelet basis: u(x) = ∞ X X djk ψkj (x) j=0 k∈Kj • Wavelets: – follow intermittency in position and scale – provide automatic grid adaptation •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Adaptive wavelet collocation method • Adaptive wavelet grid: u(x) = ∞ X X djk ψkj (x) j=0 k∈Kj Gj = {xjk ∈ R : xjk = 2−j k, k ∈ Z, j ∈ Z} •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Adaptive wavelet collocation method • Adaptive wavelet grid: u(x) = ∞ X X djk ψkj (x) j=0 k∈Kj Gja = {xjk ∈ R : xjk = 2−j k, k ∈ Z, j ∈ Z, |djk | ≥ } •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Adaptive wavelet collocation method • ALGORITHM: General grid adaptation strategy •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Adaptive wavelet collocation method • ALGORITHM: General grid adaptation strategy Start with an initial grid Solve the PDE Determine the region of intermittency Refine the grid Continue until sequence of grids are converged •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Adaptive wavelet collocation method • ALGORITHM: General grid adaptation strategy Start with an initial grid Solve the PDE Determine the region of intermittency Refine the grid Continue until sequence of grids are converged 6 5 j 4 3 2 1 0 −1 −0.5 0 x 0.5 1 •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Adaptive wavelet collocation method • ALGORITHM: General grid adaptation strategy Start with an initial grid Solve the PDE Determine the region of intermittency Refine the grid Continue until sequence of grids are converged 6 Adjacent zone: 5 j 4 3 k j+1 j j−1 2 1 0 −1 −0.5 0 x 0.5 1 •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Adaptive wavelet collocation method • ALGORITHM: General grid adaptation strategy Start with an initial grid Solve the PDE Determine the region of intermittency Refine the grid Continue until sequence of grids are converged 6 Adjacent zone: 5 j 4 3 k j+1 j j−1 2 1 0 −1 −0.5 0 x 0.5 1 •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Space-time discretization of PDEs • Discretization stencils • Classical approach: sequence of algebraic problem • Present study: single algebraic problem Explicit •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Space-time discretization of PDEs • Discretization stencils • Classical approach: sequence of algebraic problem • Present study: single algebraic problem Explicit Implicit •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Space-time discretization of PDEs • Discretization stencils • Classical approach: sequence of algebraic problem • Present study: single algebraic problem Explicit Implicit Crank-Nicholson •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Results and discussion • Burgers equation •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Results and discussion • Burgers equation ∂u ∂ 2u ∂u +u = ν 2, ∂t ∂x ∂x (x, t) ∈ Ω ⊂ R × (0, tmax], Ω = (−1, 1) u(−1, t) = u(1, t), u(x, 0) = sin(πx) ν = 10−2/π •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Results and discussion • Burgers equation ∂u ∂ 2u ∂u +u = ν 2, ∂t ∂x ∂x (x, t) ∈ Ω ⊂ R × (0, tmax], Ω = (−1, 1) u(−1, t) = u(1, t), u(x, 0) = sin(πx) ν = 10−2/π Initial condition •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Results and discussion • Burgers equation ∂u ∂ 2u ∂u +u = ν 2, ∂t ∂x ∂x (x, t) ∈ Ω ⊂ R × (0, tmax], Ω = (−1, 1) u(−1, t) = u(1, t), u(x, 0) = sin(πx) ν = 10−2/π Grid 0.3 0.25 y 0.2 0.15 0.1 0.05 0 −1 Initial condition −0.8 −0.6 −0.4 −0.2 0 x 0.2 0.4 0.6 0.8 1 Grid •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Results and discussion • Burgers equation ∂u ∂ 2u ∂u +u = ν 2, ∂t ∂x ∂x (x, t) ∈ Ω ⊂ R × (0, tmax], Ω = (−1, 1) u(−1, t) = u(1, t), u(x, 0) = sin(πx) ν = 10−2/π Solution of Burgers equation at fixed time 1 Initial Computed 0.8 0.6 0.4 u 0.2 0 −0.2 −0.4 −0.6 −0.8 −1 −1 −0.8 −0.6 −0.4 −0.2 0 x 0.2 0.4 0.6 0.8 1 Computed solution •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Results and discussion • Burgers equation ∂u ∂ 2u ∂u +u = ν 2, ∂t ∂x ∂x (x, t) ∈ Ω ⊂ R × (0, tmax], Ω = (−1, 1) u(−1, t) = u(1, t), u(x, 0) = sin(πx) ν = 10−2/π Solution of Burgers equation at fixed time 1 Initial Computed 0.8 0.6 0.4 u 0.2 0 −0.2 −0.4 −0.6 −0.8 −1 −1 −0.8 −0.6 −0.4 −0.2 0 x 0.2 0.4 0.6 Computed solution 0.8 1 Adapted grid •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Results and discussion • Burgers equation ∂u ∂ 2u ∂u +u = ν 2, ∂t ∂x ∂x (x, t) ∈ Ω ⊂ R × (0, tmax], Ω = (−1, 1) u(−1, t) = u(1, t), u(x, 0) = sin(πx) ν = 10−2/π Solution surface Adapted grid (Space-time domain) •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Results and discussion • Burgers equation ∂u ∂ 2u ∂u +u = ν 2, ∂t ∂x ∂x (x, t) ∈ Ω ⊂ R × (0, tmax], Ω = (−1, 1) u(−1, t) = u(1, t), u(x, 0) = sin(πx) ν = 10−2/π Grid Grid 0.3 0.25 y 0.2 0.15 0.1 0.05 0 −1 −0.8 −0.6 −0.4 −0.2 0 x 0.2 0.4 0.6 0.8 1 Solution •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Results and discussion • Burgers equation ∂u ∂ 2u ∂u +u = ν 2, ∂t ∂x ∂x x ∈ (−π, π) u(−π, t) = u(π, t) u(x, 0) = sin(x) Compare wavelet solution with a spectral code. •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Results and discussion • Time evolution of error in spatial grid −4 3.5 x 10 3 L∞ error on spatial grid 2.5 2 1.5 1 0.5 0 0 500 1000 1500 Number of time steps (with t∞ = 0.35) 2000 2500 •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Results and discussion • Efficiency of the method E ∼ N ()−(p−m)/n, – →0 Error vs. # of wlt 0 10 with p=4 log E = −1.5log N −2 10 −4 10 −6 10 2 10 3 4 10 5 10 6 10 10 0 E = ||u(x)−uex(x)||∞ 10 with p=6 log E = −2.5log N −2 10 −4 10 −6 10 2 10 3 4 10 5 10 10 5 10 with p=8 log E = −3.5log N 0 10 −5 10 −10 10 2 10 3 10 4 N(ε) 10 5 10 Error vs N () (Space-time domain) •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Results and discussion • Efficiency of the method E ∼ N ()−(p−m)/n, – N () ∼ −n/p, → 0, Error vs. # of wlt 0 10 ε vs. # of wlt 0 10 with p=4 log E = −1.5log N −2 10 →0 with p=4 log(ε) = −2log N −5 10 −4 10 −6 10 −10 2 10 3 4 10 5 10 10 6 10 10 0 3 4 10 5 10 6 10 10 with p=6 log E = −2.5log N −2 10 ε E = ||u(x)−uex(x)||∞ 2 10 10 0 10 with p=6 log(ε) = −3log N −5 10 −4 10 −6 10 −10 2 10 3 4 10 5 10 10 5 2 10 3 4 10 5 10 10 5 10 10 with p=8 log E = −3.5log N 0 10 with p=8 log(ε) = −4log N 0 10 −5 −5 10 10 −10 10 10 −10 2 10 3 10 4 N(ε) 10 Error vs N () 5 10 10 2 10 3 10 4 N(ε) 10 5 10 vs N () (Space-time domain) •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Results and discussion • Efficiency of the method – Define Compression ratio: CJ = N /N J × 100 – Number of collocation points at smallest scale, N J = 4194304 – Number of collocation actual points in computation, N = 22982 120 100 Cj 80 60 40 20 0 0 2 4 6 Iteration 8 10 12 •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Result and discussion: Cont’d... • Moving shock •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Result and discussion: Cont’d... • Moving shock ∂u ∂ 2u ∂u + (u + v) = ν 2, ∂t ∂x ∂x (x, t) ∈ Ω × (0, tmax], Ω = (0, 2) u(0, t) = 1, u(2, t) = −1, ν = 10−2, x − x0 u(x, 0) = − tanh 2ν x0 = 0.5 •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Result and discussion: Cont’d... • Moving shock ∂u ∂ 2u ∂u + (u + v) = ν 2, ∂t ∂x ∂x (x, t) ∈ Ω × (0, tmax], Ω = (0, 2) u(0, t) = 1, u(2, t) = −1, u ↑→ x ν = 10−2, x − x0 u(x, 0) = − tanh 2ν x0 = 0.5 Initial condition •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Result and discussion: Cont’d... • Moving shock ∂u ∂ 2u ∂u + (u + v) = ν 2, ∂t ∂x ∂x (x, t) ∈ Ω × (0, tmax], Ω = (0, 2) u(0, t) = 1, u(2, t) = −1, u ↑→ x ν = 10−2, x − x0 u(x, 0) = − tanh 2ν x0 = 0.5 Solution at t = 1.0 •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Result and discussion: Cont’d... • Moving shock ∂u ∂ 2u ∂u + (u + v) = ν 2, ∂t ∂x ∂x (x, t) ∈ Ω × (0, tmax], Ω = (0, 2) u(0, t) = 1, u(2, t) = −1, u ↑→ x ν = 10−2, Solution at t = 1.0 x − x0 u(x, 0) = − tanh 2ν x0 = 0.5 Adapted grid •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Result and discussion: Cont’d... • Moving shock ∂u ∂ 2u ∂u + (u + v) = ν 2, ∂t ∂x ∂x (x, t) ∈ Ω × (0, tmax], Ω = (0, 2) u(0, t) = 1, u(2, t) = −1, u ↑→ x Solution ν = 10−2, x − x0 u(x, 0) = − tanh 2ν x0 = 0.5 Adapted grid •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Conclusion and Future direction •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Conclusion and Future direction • Conclusion •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Conclusion and Future direction • Conclusion – An adaptive numerical method is developed •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Conclusion and Future direction • Conclusion – An adaptive numerical method is developed – An evolution type boundary condition is proposed •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Conclusion and Future direction • Conclusion – An adaptive numerical method is developed – An evolution type boundary condition is proposed – A multilevel adaptive solver has been developed •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Conclusion and Future direction • Conclusion – An adaptive numerical method is developed – An evolution type boundary condition is proposed – A multilevel adaptive solver has been developed • Future plan •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Conclusion and Future direction • Conclusion – An adaptive numerical method is developed – An evolution type boundary condition is proposed – A multilevel adaptive solver has been developed • Future plan – Solve more complicated problems in Fluid Dynamics •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Conclusion and Future direction • Conclusion – An adaptive numerical method is developed – An evolution type boundary condition is proposed – A multilevel adaptive solver has been developed • Future plan – Solve more complicated problems in Fluid Dynamics ∗ Fluid-structure interaction ∗ Turbulent transport ∗ Passive scalar advection ∗ Nano-technology ∗ Meteorology •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit Thank You •First •Prev •Next •Last •Go Back •Full Screen •Close •Quit