User Tools

Site Tools


gtspring2009:gibson:w03

This is an old revision of the document!


A PCRE internal error occured. This might be caused by a faulty plugin

====== W03 cell ====== ===== 2009-02-05 Intro ===== In which I write at length of my long personal quest for periodic enlightenment. Ok, at long last I have starting putting effort into the (alpha, gamma) = (1.14, 2.5) a.k.a. W03 ([[:references|Waleffe Phys Fluids 2003]]) or narrow cell again. This was prompted by a comment by Roman in class a couple weeks ago about relative ease of finding EQBs and POs in HKW ([[:references|Hamilton et al JFM 1995]]) vs W03 cells and a recommendation from Predrag to begin our periodic orbit paper (in development) with the W03 cell, whose state-space structure we understand a little better. Plus, previous results suggested that the two orbits we had for W03 were perturbations of each other (a short one P35.86 and a long one P97.08 that was like the short one plus an extra wiggle). I computed these orbits way back in summer 2007 just before going to India. Yikes! That was the very beginning of my orbit-computing days. The 97.08 orbit was not very well converged (|f^t(u) - u| = 5e-3 compared to 1e-8 for the other), and I didn't have a good enough sense of these computations to know what the problem was and whether I really had an orbit or not. So I went on to hone my orbit-finding chops on easier problems (HKW, where good initial guesses are plentiful). And doing this, in India and more recently, I uncovered several problems in my implementation of trust-region heuristics that might have prevented P97.08 from converging. Anyway back to the present. Over the last couple weeks I have run maybe thirty orbit searches in the WO3 box on initial guesses from moderately long-lived trajectories from perturbations of equilibria, plus the pre-India P35.86 and P97.08 orbits. P35.86 converged more precisely and is now renamed P35.77. PACE is still pounding on P97.08 and has the residual down to 1e-3, after about a hundred Newton-hooksteps. But I have found several other orbits to double precision (for a given discretization). Here are some pics. {{p35p77e1e2.png?220}}{{p35p77e1e3.png?220}}{{p35p77e2e3.png?220}}{{p35p77e1e2e3.png?220}} {{p47p18e1e2.png?220}}{{p47p18e1e3.png?220}}{{p47p18e2e3.png?220}}{{p47p18e1e2e3.png?220}} {{p50p16e1e2.png?220}}{{p50p16e1e3.png?220}}{{p50p16e2e3.png?220}}{{p50p16e1e2e3.png?220}} {{p82p36e1e2.png?220}}{{p82p36e1e3.png?220}}{{p82p36e2e3.png?220}}{{p82p36e1e2e3.png?220}} {{p83p60e1e2.png?220}}{{p83p60e1e3.png?220}}{{p83p60e2e3.png?220}}{{p83p60e1e2e3.png?220}} In the pics: * The black dot is laminar flow * The blue (green) dots are Nagata lower (upper) branch eqbs ([[:references|Nagata JFM 1990]]) * Red dots are EQ4 a.k.a "newbie" equilibrium from our recent JFM papers * Thin (blue, green, red) lines are trajectories in low-d unstable manifolds of (Nagata LB, Nagata UB, EQ4) * The thick lines are the periodic orbits * Projection is same as in most figures in [[:references|Gibson et al. JFM 2008]] And some properties. Will add eigenvalues when they're done cooking. ^ T ^ σ ^ %%<D>%% ^ %%|Λ|>1%% ^ max %%|Λ|%% ^ max Re λ ^ %%|Λ|%%<sub>S</sub> >1 ^ max %%|Λ|%%<sub>S</sub> ^ max Re λ<sub>S</sub> ^ | 35.77 | Lx/2 | 2.78 | 19 | 13.0 | 0.072 | 5 | 2.5 | 0.026 | | 50.16 | | 2.32 | 19 | 26.9 | 0.040 | 5 | 26.9 | 0.040 | | 82.36 | Lx/2 | 1.90 | 5 | 63.7 | 0.050 | 2 | 63.7| 0.050 | | 83.60 | | 2.21 | 8 | 57.6 | 0.049 | 3 | 9.1 | 0.026 | * Λ is the characteristic multiplier of the map u → σ f<sup>T</sup>(u), computed with Arnoldi iteration. * The %%|Λ|>1%% column gives the number of unstable characteristic multipliers. * Re λ = 1/T ln(%%|Λ|%%) is a continuous-time measure of the orbit's instability. * Quantities with S subscripts are restricted to the set of eigenvalues with S = {1, s1, s2, s3} symmetry. See [[:docs:math:symmetry]]. ===== 2009-02-08 Movies of periodic orbits ==== <flashplayer width="720" height="320">file=/movies/gibson/p35p77.flv&image=/movies/gibson/p35p77movie.png&repeat=none</flashplayer> <flashplayer width="720" height="320">file=/movies/gibson/p47p18.flv&image=/movies/gibson/p47p18movie.png&repeat=none</flashplayer> <flashplayer width="720" height="320">file=/movies/gibson/p50p16xz.flv&image=/movies/gibson/p50p16movie.png&repeat=none</flashplayer> <flashplayer width="720" height="320">file=/movies/gibson/p82p36.flv&image=/movies/gibson/p82p36movie.png&repeat=none</flashplayer> <flashplayer width="720" height="320">file=/movies/gibson/p83p60z.flv&image=/movies/gibson/p83p60movie.png&repeat=none</flashplayer> ===== 2009-02-16 Desymmetrization ===== Modding out the 4th-order discrete symmetry: The 3D projections above are onto basis functions {e0,e1,e2,e3} that are symmetric and antisymmetric in the half-cell shift translation symmetries <latex> 1, \tau_x, \tau_z, \tau_{xz}</latex>. {{:gtspring2009:gibson:p47e1e2fund.png?300}} {{:gtspring2009:gibson:p47e1e3fund.png?300}} {{:gtspring2009:gibson:p47e2e3fund.png?300}} ===== 2009-03-17 Poincare section ===== {{:gtspring2009:gibson:eq2poincare1.png?400}} Here's a first shot at a Poincare section through the Nagata upper branch. The unstable complex eigenvalue is λ = 0.03253 +/- 0.1070, giving a period of T=58.72 and a characteristic multiplier over that period of Λ = 6.7545. The orthonormal basis e0,e1,e2,e3 is formed from the complex instability (the plane of e0,e1) and the most weakly stable of the remaining eigenvectors in the S-invariant subspace. I seeded N=16 trajectories along direction in the plane of complex instability <latex> u_n(0) = EQ2 + \epsilon \Lambda^{n/N} \cdot e_0, \quad n=0,...,15 </latex> with ε = 1e-05 and integrated them for 300 time units, chosen since 1e-05 exp(300 Re λ) is about 0.2, roughly the maximum distance observed in the state space. The Poincare section shown is defined by the condition <latex> (u(t) - EQ2, e_{\theta}) = 0 </latex> where <latex> e_{\theta} = e_0 \cos \theta + e_1 \sin \theta </latex> I used a small nonzero value of θ so that trajectories started slightly before the Poincare section, rather than before or after by +/- 1e-14 floating point rounding error, which would have caused a few section points to go missing. Crossings were calculated by quadratic interpolation between successive time steps that pierce the section. As solutions of Navier-Stokes dynamics the interpolation accuracy between steps spaced by dt is the same as the accuracy of the time-stepping algorithm (O(dt^3)). The interpolated fields met the Poincare condition to floating-point accuracy. Prior to checking for section crossings, the field u(t) was translated by half-cell shifts into the canonical first quadrant defined by <latex> $ \begin{align*} (u, e_x) &\geq 0 \\ (u, e_z) &\geq 0 \end{align*} $ </latex> where ex and ez are the x and z antisymmetric basis elements of the UB (EQ2) translation basis of our 2008 JFM paper. Factoring out the 4th-order discrete translation group this way introduces discontinuities in the unstable manifold. Red / blue signify ingoing / outgoing crossings of the section. The open dots are the crossings of the P47p18 periodic orbit. Orientation of the red sections is the same, but the blue sections are reversed. I think this is just an artifact of the projection since it doesn't make sense dynamically. Much is left to be done of course: getting a more completepicture of the discontinuous sections of the unstable manifold, coordinating the unstable eigenvalues of theperiodic orbit with the EQ2 unstable manifold, looking at the same picture for various values of theta, coordinating these results with previous images of the EQ2 unstable manifold. Dustin, I hesitated to do this because I know it's your project, but there was a fair amount of coding involved, so I'm glad I went ahead and did it. There is still tons to do. I will make notes tomorrow on exactly how I did the calculations and provide you with the needed code. ===== 2009-03-18 more slice and dice ===== By the way, here are the eigenvalues for EQ2. ASA symmetry, for example, means antisymmetry in s1 and s3 and symmetry in s2. ^ n ^ Re ^ Im(lambda) ^ symmetry ^ | 1 | 0.05558362 | 0 | AAS | | 2,3 | 0.03252919 | 0.107043 | SSS | | 4,5 | 0.01606001 | 0.03923802 | SAA | | 6,7 | 0.01529245 | 0.2841767 | SAA | | 8 | 0.01060338 | 0 | ASA | | 11,12 | -0.01412172 | 0.05774757 | SSS | | 13 | -0.01818171 | 0 | SAA | | 14,15 | -0.02091932 | 0.1735921 | AAS | | 16,17 | -0.02429506 | 0.1479473 | SSS | | 18,19 | -0.02646826 | 0.3121915 | ASA | | 20,21 | -0.02741331 | 0.147147 | AAS | | 22,23 | -0.02936277 | 0.1387544 | SSS | | 24,25 | -0.0303013 | 0.2516945 | ASA | | 26 | -0.03073016 | 0 | ASA | | 27,28 | -0.03770692 | 0.1986541 | AAS | The 2,3 eigenvalues have a period of 58.7 and a multiplier of 6.75. The next SSS eigenvalue pair 11,12 has a multiplier of exp(-0.0141*58.7) = 0.437 in that period, which is not very strongly contracting. There's roughly a factor of three between the most expanding SSS direction and the next, most weakly stable. ===== 2009-03-18 more on desymmetrization ===== I don't think the double-theta trick that works for desymmetrizing Lorenz will work for S-symmetric plane Couette. This trick factors out a <latex> C^{1/2} : (x,y,z) \rightarrow (-x,-y,z) </latex> symmetry in Lorenz --an orientation-preserving rotational symmetry. In S-symmetric plane Couette we have a fourth order symmetry that equates <latex> (\pm x,y,\pm z)</latex>. (Here I am letting x stand for all the τx-antisymmetric coordinates in the plane Couette state space, etc.) The double-angle trick would collapse the pair (x,y,z) and (-x,y,-z) (in this case we'd apply it in x,z rather than x,y), and also the pair (-x, y, z) and (x,y,-z), but it won't connect (x,y,z) to (-x,y,z) or (x,y,z) to (x,y,-z), which are related by a mirror symmetries. And it's the (x,y,z) to (-x,y,z) connection we are particularly concerned with, since the P47p18 orbit goes back and forth between EQ2 and τx EQ2. ===== 2009-03-20 and more... ===== Also, I don't see how to avoid introducing discontinuities when mapping into a canonical quadrant. In PCF, you have many τx antisymmmetric coordinates. Label them (x1, x2, x3, ...). Same with z. The trouble is these coordinates cross zero independently. So if you define the canoical 1st quadrant to be x1 ≥ 0 and z1 ≥ 0, and you apply τx when x1 goes from positive to negative (etc), the x1 coordinate is continuous, but all the other nonzero x coordinates change sign, too, and thus undergo finite jumps. ===== 2009-03-24 more poincare sections ===== {{eq2poincare_ubefdiff1.png?300}} {{eq2poincare_ubefdiff2.png?300}} Above shows our old friend the Nagata upper branch EQ2 (at the origin) and its S-symmetric 2D unstable manifold in green along with the P47p18 periodic orbit in blue. The 4th-order translation symmetry has been factored out as described above. The projection is onto a 3D basis, of which e0,e1 span the complex instability and e2 is (EQ2 - τx EQ2) with the components parallel to e0,e1 removed. The reason for this choice of basis is that the dynamics in this region seems to be governed by the complex instability and flopping back and forth between EQ2 and τx EQ2. I am game for suggestions of other basis sets. Once the sections are computed (as codim-1 slices orthogonal to %%e(θ) = e0 cos θ + e1 sin θ%%), it takes 2 or 3 hours to compute a low-d projection of the section, make the plots, and post them. Poincare sections, in black, have been computed at several orientations marked by values of θ. Factoring out the discrete translation translation symmetry introduces discontinuities in trajectories as described above. I kept the abrupt line segments marking these discontinuities in the plot, even though they aren't strictly part of the trajectory, because they help you seen how the jumps are connected. {{eq2poincare_ubefdiff_zero.png?300}} {{eq2poincare_ubefdiff_pi4.png?300}} {{eq2poincare_ubefdiff_pi2.png?300}} {{eq2poincare_ubefdiff_3pi4.png?300}} The four plots above are the 2D Poincare sections with horizontal axis at various orientations in the e0,e1 plane and the vertical axis along e2. Each plot corresponds to a marked value of θ in the 3D plots above. The black portions are "ingoing" trajectories, which are on the side of the section where θ is labeled in the 3D plots (for the most part). The green portions are "outgoing" trajectories on the other side. The blue circles mark the crossings of the the P47p18 periodic orbit. These pictures are nice to look at but I'm not sure how much we can take away from them. You can't necessarily interpret an apparent fold in these sections as the classic nonlinear stretch and fold scenario, like in Rossler or Lorenz. It is true that an apparent fold, like the black part of the θ=0 section, shows that the unstable manifold has stopped stretching in the θ=0 direction and has returned back to lower values of (u, eθ). However, we can't attach too much meaning to the vertical axis. That "fold" could be a very gentle curvature in the full space, and the portion of the unstable manifold that appears to be right above EQ2 (the origin) could be way, way far away from it, in directions we've projected out. ===== 2009-03-26 ===== Eigenvalues of EQ2 in the S-symmetric subspace

gtspring2009/gibson/w03.1238085001.txt.gz · Last modified: 2009/03/26 09:30 by gibson