adjoints
play

Adjoints MEC651 denis.sipp@onera.fr Adjoints 1 Outline - - PowerPoint PPT Presentation

Adjoints MEC651 denis.sipp@onera.fr Adjoints 1 Outline - Governing equations - Asymptotic development Order 0 : Base-flow Order 1 : Global modes - Bi-orthogonal basis and adjoint global modes Definition of adjoint


  1. Adjoints MEC651 denis.sipp@onera.fr Adjoints 1

  2. Outline - Governing equations - Asymptotic development οƒ˜ Order πœ— 0 : Base-flow οƒ˜ Order πœ— 1 : Global modes - Bi-orthogonal basis and adjoint global modes οƒ˜ Definition of adjoint global modes οƒ˜ Optimal initial condition οƒ˜ Optimal forcing in stable flow - Adjoint operator οƒ˜ Definition οƒ˜ Adjoint global modes as solutions of adjoint eigen-problem - Adjoint linearized Navier-Stokes operator οƒ˜ Adjoint of linearized advection operator οƒ˜ Adjoint of Stokes operator οƒ˜ Adjoint global modes of cylinder flow MEC651 denis.sipp@onera.fr Adjoints 2

  3. Governing equations Incompressible Navier-Stokes equations: πœ– 𝑒 𝑣 + π‘£πœ– 𝑦 𝑣 + π‘€πœ– 𝑧 𝑣 = βˆ’πœ– 𝑦 π‘ž + πœ‰ πœ– 𝑦𝑦 𝑣 + πœ– 𝑧𝑧 𝑣 + 𝑔 πœ– 𝑒 𝑀 + π‘£πœ– 𝑦 𝑀 + π‘€πœ– 𝑧 𝑀 = βˆ’πœ– 𝑧 π‘ž + πœ‰ πœ– 𝑦𝑦 𝑀 + πœ– 𝑧𝑧 𝑀 + 𝑕 βˆ’πœ– 𝑦 𝑣 βˆ’ πœ– 𝑧 𝑀 = 0 Can be recast into: β„¬πœ– 𝑒 π‘₯+ 1 2 π’ͺ π‘₯, π‘₯ + β„’π‘₯ = 𝑔 where: π‘₯ = 𝑣 π‘ž 𝑔 = 𝑔 0 ℬ = 1 0 0 , 0 π’ͺ π‘₯ 1 , π‘₯ 2 = 𝑣 1 β‹… 𝛼𝑣 2 + 𝑣 2 β‹… 𝛼𝑣 1 0 βˆ’πœ‰Ξ”() 𝛼() β„’ = βˆ’π›Ό β‹… () 0 Boundary conditions: Dirichlet, Neumann, Mixed MEC651 denis.sipp@onera.fr Adjoints 3

  4. Some properties a) π’ͺ π‘₯ 1 , π‘₯ 2 = π’ͺ π‘₯ 2 , π‘₯ 1 πœ— 2 b) 1 1 2 π’ͺ π‘₯ 0 + πœ—πœ€π‘₯, π‘₯ 0 + πœ—πœ€π‘₯ = 2 π’ͺ π‘₯ 0 , π‘₯ 0 + πœ— π’ͺ π‘₯ 0 , πœ€π‘₯ + 2 π’ͺ πœ€π‘₯, πœ€π‘₯ + β‹― Hessian Jacobian =π’ͺ π‘₯0 πœ€π‘₯ π‘₯ 0 πœ€π‘₯ = π’ͺ π‘₯ 0 , πœ€π‘₯ = πœ€π‘£ β‹… 𝛼𝑣 0 + 𝑣 0 β‹… π›Όπœ€π‘£ c) π’ͺ 0 d) ℬπ‘₯ = ℬ 𝑣 π‘ž = 𝑣 0 e) πœ– 𝑒 𝑣 + 𝑣 β‹… 𝛼𝑣 = βˆ’π›Όπ‘ž + πœ‰π›Ό 2 𝑣 β‡’ βˆ’π›Ό 2 π‘ž = 𝛼 β‹… 𝑣 β‹… 𝛼𝑣 , πœ– π‘œ π‘ž = πœ‰π›Ό 2 𝑣 β‹… π‘œ on solid walls . Hence, p is a function of u and should not be considered as a degree of freedom of the flow. βˆ— 𝑣 2 + 𝑀 1 βˆ— 𝑀 2 𝑒𝑦𝑒𝑧 = ∬ (π‘₯ 1 β‹… ℬπ‘₯ 2 )𝑒𝑦𝑒𝑧 so f) Scalar-product: < π‘₯ 1 , π‘₯ 2 > = ∬ 𝑣 1 that < π‘₯, π‘₯ > is the energy. MEC651 denis.sipp@onera.fr Adjoints 4

  5. Outline - Governing equations - Asymptotic development οƒ˜ Order πœ— 0 : Base-flow οƒ˜ Order πœ— 1 : Global modes - Bi-orthogonal basis and adjoint global modes οƒ˜ Definition of adjoint global modes οƒ˜ Optimal initial condition οƒ˜ Optimal forcing in stable flow - Adjoint operator οƒ˜ Definition οƒ˜ Adjoint global modes as solutions of adjoint eigen-problem - Adjoint linearized Navier-Stokes operator οƒ˜ Adjoint of linearized advection operator οƒ˜ Adjoint of Stokes operator οƒ˜ Adjoint global modes of cylinder flow MEC651 denis.sipp@onera.fr Adjoints 5

  6. Asymptotic development Solution: π‘₯ 𝑒 = π‘₯ 0 + πœ—π‘₯ 1 𝑒 + β‹― with Ο΅ β‰ͺ 1 Governing equations: β„¬πœ– 𝑒 π‘₯+ 1 2 π’ͺ π‘₯, π‘₯ + β„’π‘₯ = 𝑔 Introduce solution into governing eq:: β„¬πœ– 𝑒 (π‘₯ 0 +πœ—π‘₯ 1 + β‹― )+ 1 2 π’ͺ π‘₯ 0 + πœ—π‘₯ 1 + β‹― , π‘₯ 0 + πœ—π‘₯ 1 + β‹― + β„’(π‘₯ 0 +πœ—π‘₯ 1 + β‹― ) = 𝑔 1 2 π’ͺ π‘₯ 0 , π‘₯ 0 + β„’π‘₯ 0 = 𝑔 at order 𝑃(1) β„¬πœ– 𝑒 π‘₯ 1 + 1 2 [π’ͺ π‘₯ 1 , π‘₯ 0 + π’ͺ π‘₯ 0 , π‘₯ 1 ] + β„’π‘₯ 1 = 0 at order 𝑃(πœ—) β‡’ π’ͺ π‘₯0 π‘₯ 1 π‘₯ 0 π‘₯ 2 + β„’π‘₯ 2 = βˆ’ 1 2 π’ͺ π‘₯ 1 , π‘₯ 1 at order 𝑃(πœ— 2 ) β„¬πœ– 𝑒 π‘₯ 2 +π’ͺ MEC651 denis.sipp@onera.fr Adjoints 6

  7. Oder πœ— 0 : Base-flow Definition: 𝐺 π‘₯ = 1 2 π’ͺ π‘₯, π‘₯ + β„’π‘₯ βˆ’ 𝑔 π‘₯ 𝑒 = π‘₯ 0 + πœ—π‘₯ 1 (𝑒) + β‹― Non-linear equilibrium point : 1 2 π’ͺ π‘₯ 0 , π‘₯ 0 + β„’π‘₯ 0 = 𝑔 π‘₯ π‘₯ 0 How to compute a base-flow ? Newton iteration: 1 2 π’ͺ π‘₯ 0 + πœ€π‘₯ 0 , π‘₯ 0 + πœ€π‘₯ 0 + β„’(π‘₯ 0 +πœ€π‘₯ 0 ) = 𝑔 Linearization: π’ͺ π‘₯ 0 , πœ€π‘₯ 0 + β„’πœ€π‘₯ 0 = 𝑔 βˆ’ 1 2 π’ͺ π‘₯ 0 , π‘₯ 0 βˆ’ β„’π‘₯ 0 βˆ’1 𝑔 βˆ’ 1 β‡’ πœ€π‘₯ 0 = π’ͺ π‘₯ 0 + β„’ 2 π’ͺ π‘₯ 0 , π‘₯ 0 βˆ’ β„’π‘₯ 0 MEC651 denis.sipp@onera.fr Adjoints 7

  8. Oder πœ— 0 : Base-flow The case of cylinder flow 𝑆𝑓 = 47 Streamwise velocity field of base-flow. MEC651 denis.sipp@onera.fr Adjoints 8

  9. Order πœ— 1 : Global modes Definition π‘₯ 𝑒 = π‘₯ 0 + πœ—π‘₯ 1 (𝑒) + β‹― Linear governing equation: β„¬πœ– 𝑒 π‘₯ 1 + π’ͺ π‘₯ 0 π‘₯ 1 + β„’π‘₯ 1 = 0 Solution π‘₯ 1 under the form: π‘₯ 1 = 𝑓 πœ‡π‘’ π‘₯ + c.c This leads to : πœ‡β„¬π‘₯ + π’ͺ π‘₯ 0 + β„’ π‘₯ = 0 Eigenvalue: πœ‡ = 𝜏 + π‘—πœ• Eigenvector: = π‘₯ π‘₯ r + iw 𝑗 Real solution: π‘₯ 1 = 𝑓 πœ‡π‘’ π‘₯ + c.c = 2𝑓 πœπ‘’ (cos πœ•π‘’ π‘₯ 𝑠 βˆ’ sin πœ•π‘’ π‘₯ 𝑗 ) MEC651 denis.sipp@onera.fr Adjoints 9

  10. Order πœ— 1 : Global modes How to compute global modes ? Eigenvalue problem solved with shift-invert strategy: - Power method, easy to find largest magnitude eigenvalues of 𝐡𝑦 = πœ‡π‘¦ . For this, evaluate 𝐡 π‘œ 𝑦 0 To find eigenvalues of 𝐡 closest to zero, search largest magnitude eigenvalues of - 𝐡 βˆ’1 : 𝐡 βˆ’1 𝑦 = πœ‡ βˆ’1 𝑦 . For this, evaluate 𝐡 βˆ’1 π‘œ 𝑦 0 To find eigenvalues of 𝐡 closest to 𝑑 , search largest magnitude eigenvalues of - 𝐡 βˆ’ 𝑑𝐽 βˆ’1 π‘œ 𝑦 0 𝐡 βˆ’ 𝑑𝐽 βˆ’1 : 𝐡 βˆ’ 𝑑𝐽 βˆ’1 𝑦 = πœ‡ βˆ’ 𝑑 βˆ’1 𝑦 . For this, evaluate - Instead of power-method, use Krylov subspaces -> Arnoldi technique - Cost of algorithm = cost of several complex matrix inversions MEC651 denis.sipp@onera.fr Adjoints 10

  11. Order πœ— 1 : Global modes Case of cylinder flow Real part of cross-stream velocity field Spectrum 𝑆𝑓 = 47 Marginal eigenmode MEC651 denis.sipp@onera.fr Adjoints 11

  12. The Ginzburg-Landau eq. We consider the linear Ginzburg-Landau equation πœ– 𝑒 π‘₯ 1 + β„’π‘₯ 1 = 0 where 𝑦 2 β„’ = π‘‰πœ– 𝑦 βˆ’ 𝜈 𝑦 βˆ’ π›Ώπœ– 𝑦𝑦 , 𝜈 𝑦 = π‘—πœ• 0 + 𝜈 0 βˆ’ 𝜈 2 2 . Here 𝑉, 𝛿, πœ• 0 , 𝜈 0 and 𝜈 2 are positive real constants. The state π‘₯(𝑦, 𝑒) is a complex variable on βˆ’βˆž < 𝑦 < +∞ such that |π‘₯| β†’ 0 as 𝑦 β†’ ∞ . In the following, +∞ π‘₯ 𝑏 , π‘₯ 𝑐 = π‘₯ 𝑏 𝑦 βˆ— π‘₯ 𝑐 𝑦 𝑒𝑦 . βˆ’βˆž 1/ What do the different terms in the Ginzburg Landau equation represent? MEC651 denis.sipp@onera.fr Adjoints 12

  13. The Ginzburg-Landau eq. 1 2𝛿 π‘¦βˆ’ πœ“2𝑦2 𝑉 𝜈 2 πœ“ 4 and πœ‚ = 2/ Show that π‘₯ (𝑦) = πœ‚π‘“ 2 with πœ“ = verifies πœ‡π‘₯ + β„’π‘₯ = 0 . 𝑉2 2𝛿 1 1 𝛿2πœ“2 8 𝜌 4 𝑓 What is the eigenvalue πœ‡ associated to this eigenvector? The constant πœ‚ has been selected so that π‘₯ , π‘₯ = 1 . 3/ Show that the flow is unstable if the constant 𝜈 0 is chosen such that: 𝜈 0 > 𝜈 𝑑 , where 𝑉 2 π›Ώπœˆ 2 𝜈 𝑑 = 4𝛿 + 2 . 2𝛿 π‘¦βˆ’ πœ“2𝑦2 𝑉 πœ‡ π‘œ = iπœ• 0 + 𝜈 0 βˆ’ 𝑉 2 π›Ώπœˆ 2 4𝛿 βˆ’ 2π‘œ + 1 2 , π‘₯ π‘œ = πœ‚ π‘œ 𝐼 π‘œ πœ“π‘¦ 𝑓 2 Nota: are all the eigenvalues/eigenvectors of β„’ , 𝐼 π‘œ being Hermite polynomials. MEC651 denis.sipp@onera.fr Adjoints 13

  14. Outline - Governing equations - Asymptotic development οƒ˜ Order πœ— 0 : Base-flow οƒ˜ Order πœ— 1 : Global modes - Bi-orthogonal basis and adjoint global modes οƒ˜ Definition of adjoint global modes οƒ˜ Optimal initial condition οƒ˜ Optimal forcing in stable flow - Adjoint operator οƒ˜ Definition οƒ˜ Adjoint global modes as solutions of adjoint eigen-problem - Adjoint linearized Navier-Stokes operator οƒ˜ Adjoint of linearized advection operator οƒ˜ Adjoint of Stokes operator οƒ˜ Adjoint global modes of cylinder flow MEC651 denis.sipp@onera.fr Adjoints 14

  15. Bi-orthogonal basis and adjoint global modes (1/3) In finite dimension Global mod odes: 𝐡π‘₯ 𝑗 = πœ‡ 𝑗 π‘₯ 𝑗 The eigenvectors π‘₯ 𝑗 form a basis: π‘₯ = 𝛽 𝑗 w i 𝑗 Definition of adjoint global modes: with <> as a given scalar-product (say βˆ— π‘₯ 2 ), there exists for each 𝛽 𝑗 a unique π‘₯ < π‘₯ 1 , π‘₯ 2 > = π‘₯ 1 𝑗 such that 𝛽 𝑗 =< π‘₯ 𝑗 , π‘₯ > for all π‘₯ . The adjoint global modes are the structures π‘₯ 𝑗 . In the following: π‘₯ 𝑗 , π‘₯ 𝑗 = 1 . Properties: οƒ˜ π‘₯ 𝑙 and w j are bi-orthogonal bases: they verify π‘₯ π‘˜ = < π‘₯ 𝑗 , π‘₯ π‘˜ > w i and so 𝑗 βˆ— 𝑋 = 𝐽 ) < π‘₯ 𝑙 , w j > = πœ€ π‘™π‘˜ (in matrix notations 𝑋 1 1 οƒ˜ Cauchy-Lifschitz: 1 = < π‘₯ 𝑗 , w i > ≀< π‘₯ 𝑗 , π‘₯ 𝑗 > 2 < π‘₯ 𝑗 , π‘₯ 𝑗 > 2 1 1 Hence: < π‘₯ 𝑗 , π‘₯ 𝑗 > 2 β‰₯ 1 and cos angle π‘₯ 𝑗 , π‘₯ 𝑗 = 1 <π‘₯ 𝑗 ,π‘₯ 𝑗 > 2 MEC651 denis.sipp@onera.fr Adjoints 15

  16. Bi-orthogonal basis and adjoint global modes (2/3) In finite dimension π‘₯ = (π‘₯ 1 β‹… π‘₯)π‘₯ 1 + (π‘₯ 2 β‹… π‘₯)π‘₯ 2 Def of π‘₯ 1 : π‘₯ 2 π‘₯ 1 β‹… π‘₯ 1 = 1 π‘₯ 1 β‹… π‘₯ 2 = 0 π‘₯ 2 Def of π‘₯ 2 : π‘₯ 2 β‹… π‘₯ 2 = 1 π‘₯ 2 β‹… π‘₯ 1 = 0 π‘₯ 1 π‘₯ 1 βˆ— 𝑋 = 𝐽 𝑋 = W βˆ—βˆ’1 Method 1 : 𝑋 βˆ’1 β‡’ 𝑋 βˆ’1 = 𝑋 π‘Œ β‡’ π‘Œ βˆ— 𝑋 βˆ— 𝑋 = 𝐽 β‡’ π‘Œ = 𝑋 βˆ— 𝑋 = 𝑋 𝑋 βˆ— 𝑋 Method 2 : 𝑋 Method 3 : adjoint global modes MEC651 denis.sipp@onera.fr Adjoints 16

Recommend


More recommend