Quaternions and particle dynamics in the Euler fluid equations. J. D. Gibbon. Mathematics Department Imperial College London, London SW7 2AZ

Transcription

1 Quaternions and particle dynamics in the Euler fluid equations J. D. Gibbon Mathematics Department Imperial College London, London SW7 2AZ Maryland CSCAMM Meeting; October 23-27, 2006 J. D. Gibbon, CSCAMM; October

2 Collaborators JDG & Darryl Holm 06: JDG, Holm, Kerr & Roulstone: Nonlinearity 19, , 2006 JDG, Physica D, 166, 17-28, Galanti, JDG & Heritage; Nonlinearity 10, 1675, Galanti, JDG & Kerr, in Turbulence structure & vortex dynamics, (pp 23-34, eds: Hunt & Vassilicos, CUP 2000). J. D. Gibbon, CSCAMM; October

3 Summary of this talk Question: Do the Euler equations possess some subtle geometric structure that guides the direction of vorticity see Peter Constantin, Geometric statistics in turbulence, SIAM Reviews, 36, Quaternions: what are they? 2. Lagrangian particle dynamics: We find explicit equations for the Lagrangian derivatives of an ortho-normal co-ordinate system at each point in space. (JDG/Holm 06) 3. For the 3D-Euler equations; Ertel s Theorem shows how Euler fits naturally into this framework (JDG, Holm, Kerr & Roulstone 2006). 4. Review of work on the direction of Euler vorticity, particularly that of Constantin, Fefferman & Majda 1996; Deng, Hou & Yu 2005/6 & Chae A different direction of vorticity result involving the pressure Hessian. J. D. Gibbon, CSCAMM; October

4 Lord Kelvin (William Thompson) once said: Quaternions came from Hamilton after his best work had been done, & though beautifully ingenious, they have been an unmixed evil to those who have touched them in any way. O Connor, J. J. & Robertson, E. F Sir William Rowan Hamilton, history/mathematicians/hamilton.html Kelvin was wrong because quaternions are now used in the computer animation, avionics & robotics industries to track objects undergoing sequences of tumbling rotations. Visualizing quaternions, by Andrew J. Hanson, MK-Elsevier, Quaternions & rotation Sequences: a Primer with Applications to Orbits, Aerospace & Virtual Reality, J. B. Kuipers, Princeton University Press, J. D. Gibbon, CSCAMM; October

5 What are quaternions? (Hamilton (1843)) Imperial College London Quaternions are constructed from a scalar p & a 3-vector q by forming the tetrad p = [p, q] = pi q σ, q σ = 3 q i σ i i=1 based on the Pauli spin matrices that obey the relations σ i σ j = δ ij ɛ ijk σ k ( ) ( ) ( ) 0 i 0 1 i 0 σ 1 = σ 2 = σ 3 =. i i Thus quaternions obey the multiplication rule p 1 p 2 = [p 1 p 2 q 1 q 2, p 1 q 2 + p 2 q 1 + q 1 q 2 ]. They are associative but obviously non-commutative. J. D. Gibbon, CSCAMM; October

6 Quaternions, Rotations and Cayley-Klein parameters Let ˆp = [p, q] be a unit quaternion with inverse ˆp = [p, q] with p 2 + q 2 = 1, which guarantees ˆp ˆp = [1, 0]. For a pure quaternion r = [0, r] there exists a transformation r r r = ˆp r ˆp = [0, (p 2 q 2 )r + 2p(q r) + 2q(r q)]. Now choose p = ± cos 1 2θ and q = ± ˆn sin 1 2θ, where ˆn is the unit normal to r r = ˆp r ˆp = [0, r cos θ + ( ˆn r) sin θ], where ˆp = ±[cos 1 2θ, ˆn sin 1 2θ]. This represents a rotation by an angle θ of the 3-vector r about its normal ˆn. The elements of the unit quaternion ˆp are the Cayley-Klein parameters from which the Euler angles can be calculated. All terms are quadratic in p and q, and thus allow a double covering (±) (see Whittaker 1945). J. D. Gibbon, CSCAMM; October

7 General Lagrangian evolution equations Imperial College London Consider the general Lagrangian evolution equation for a 3-vector w such that Dw D = a(x, t) = t + u transported by a velocity field u. Define the scalar α a the 3-vector χ a as α a = w 1 (ŵ a), χ a = w 1 (ŵ a). for w = 0. Via the decomposition a = α a w + χ a w, w & ŵ satisfy D w Dŵ = α a w, = χ a ŵ. α a is the growth rate (Constantin 1994) & χ a is the swing rate. The tetrads q a = [α a, χ a ], w = [0, w]. allow us to write this as Dw = q a w. J. D. Gibbon, CSCAMM; October

8 Theorem: (JDG/Holm 06) If a is differentiable in the Lagrangian sense s.t. Da = b(x, t), (i) For for w = 0, q a and q b satisfy the Ricatti equation Dq a + q a q a = q b ; (ii) At each point x there exists an ortho-normal frame (ŵ, ˆχ a, ŵ ˆχ a ) SO(3) whose Lagrangian time derivative is expressed as Dŵ = D ab ŵ, D(ŵ ˆχ a ) = D ab (ŵ ˆχ a ), D ˆχ a = D ab ˆχ a, where the Darboux angular velocity vector D ab is defined as D ab = χ a + c 1 ŵ, c 1 = ŵ ( ˆχ χ a χ b ). a J. D. Gibbon, CSCAMM; October

9 Lagrangian frame dynamics: tracking a particle Imperial College London ˆχ a ŵ ŵ ˆχ a ŵ ŵ ˆχ a ˆχ a t 1 t 2 The dotted line represents a particle ( ) trajectory moving from (x 1, t 1 ) to (x 2, t 2 ). The orientation of the orthonormal unit vectors {ŵ, ˆχ a, (ŵ ˆχ a )} is driven by the Darboux vector D ab = χ a + c 1 χ a ŵ where c 1 = ŵ ( ˆχ a χ b ). Thus we need the quartet of vectors to make this process work {u, w, a, b}. J. D. Gibbon, CSCAMM; October

10 Proof : (i) It is clear that (with q b = [α b, χ b ]) D 2 w = [0, b] = q 2 b w. Compatibility between this and the q-equation means that ( ) Dqa + q a q a q b w = 0, (ii) Now consider the ortho-normal frame (ŵ, ˆχ a, ŵ ˆχ a ) as in the Figure below. The evolution of χ a comes from ŵ. a ˆχa b...ŵ... ˆχ a Dq a + q a q a = q b, and gives Dχ a = 2α a χ a + χ b. J. D. Gibbon, CSCAMM; October

11 b can be expressed in this ortho-normal frame as the linear combination b = w [α b ŵ + c 1 ˆχ a + c 2 (ŵ ˆχ a )], χ b = c 1 (ŵ ˆχ a ) c 2 ˆχ a, where c 1 = ŵ ( ˆχ a χ b ) and c 2 = ( ˆχ a χ b ). From the Ricatti equation for the tetrad q a = [α a, χ a ] (where χ a = χ a ) Dχ a = 2α a χ a + χ b, Dχ a = 2α aχ a c 2, There follows D ˆχ a = c 1 χ 1 D(ŵ ˆχ a (ŵ ˆχ a ), a ) = χ a ŵ c 1 χ 1 a ˆχ a, which, together with Dŵ a = χ a ŵ, can be re-expressed in terms of the Darboux vector D a = χ a + c 1 χ a ŵ. J. D. Gibbon, CSCAMM; October

12 Ertel s Theorem & the 3D Euler equations Imperial College London Dω = ω u = Sω Euler in vorticity format Theorem: (Ertel 1942) If ω satisfies the 3D incompressible Euler equations then any arbitrary differentiable μ satisfies ( ) [ ] D Dμ D (ω μ) = ω =, ω = 0. Proof: Consider ω μ ω i μ,i D (ω i μ,i ) = Dω { ( ) } i Dμ μ,i + ω i u k,i μ, k x i ( ) Dμ = {ω j u i,j μ,i ω i u k,i μ, k } + ω }{{} i x i zero under summation In characteristic (Lie-derivative) form, ω x (t) = ω invariant (Cauchy 1859) and is frozen in. x (0) is a Lagrangian J. D. Gibbon, CSCAMM; October

13 Various references Ertel; Ein Neuer Hydrodynamischer Wirbelsatz, Met. Z. 59, , (1942). Truesdell & Toupin, Classical Field Theories, Encyclopaedia of Physics III/1, ed. S. Flugge, Springer (1960). Ohkitani; Phys. Fluids, A5, 2576, (1993). Kuznetsov & Zakharov; Hamiltonian formalism for nonlinear waves, Physics Uspekhi, 40 (11), (1997). Bauer s thesis 2000 (ETH-Berlin); Gradient entropy vorticity, potential vorticity and its history. Viudez; On the relation between Beltrami s material vorticity and Rossby- Ertel s Potential, J. Atmos. Sci. (2001). J. D. Gibbon, CSCAMM; October

14 Ohkitani s result & the pressure Hessian Imperial College London Define the Hessian matrix of the pressure { 2 } p P = {p,ij } = x i x j then Ohkitani took μ = u i (Phys. Fluids, A5, 2576, 1993). Result: The vortex stretching vector ω u = Sω obeys D(ω u) = D(Sω) ( ) Du = ω = P ω Thus for Euler, via Ertel s Theorem, we have the identification: w ω a ω u = Sω b P ω with a quartet (u, w, a, b) (u, ω, Sω, P ω). J. D. Gibbon, CSCAMM; October

15 Euler: the variables α(x, t) and χ(x, t) Imperial College London ˆω S ˆω ˆχ ˆω ˆχ S ˆω = α ˆω + χ ˆω See JDG, Holm, Kerr & Roulstone (α a ) α = ˆω S ˆω χ = ˆω S ˆω (χ a ) ( α b ) α p = ˆω P ˆω χ p = ˆω P ˆω ( χ b ) q = [α, χ] q b = q p = [α p, χ p ] Dq + q q + q p = 0, constrained by T rp = Δp = u i,j u j,i = 1 2ω 2 T rs 2. J. D. Gibbon, CSCAMM; October

16 Lagrangian frame dynamics: tracking an Euler fluid particle ˆχ ˆω ˆω ˆχ ˆω ˆω ˆχ ˆχ t 1 t 2 The dotted line represents the fluid packet ( ) trajectory moving from (x 1, t 1 ) to (x 2, t 2 ). The orientation of the orthonormal unit vectors {ˆω, ˆχ, (ˆω ˆχ)} is driven by the Darboux vector D = χ + c 1 χ ˆω, c 1 = ˆω ( ˆχ χ p ). Thus the pressure Hessian within c 1 drives the Darboux vector D. J. D. Gibbon, CSCAMM; October

17 The α and χ equations In terms of α and χ, the Ricatti equation for q Dq a + q a q a = q b ; becomes Dα = χ2 α 2 Dχ α p, = 2αχ χ p. (Galanti, JDG & Heritage; Nonlinearity 10, 1675, 1997). Stationary values are α = γ 0, χ = 0, α p = γ0 2 which correspond to Burgers -like vortices. When tubes & sheets bend & tangle then χ 0 and q becomes a full tetrad driven by q p which is coupled back through the elliptic pressure condition. Note: Off-diagonal elements of P change rapidly near intense vortical regions across which χ p and α p change rapidly. J. D. Gibbon, CSCAMM; October

18 Phase plane On Lagrangian trajectories, the α χ equations become α t = χ2 α 2 α p, where C p = ˆχ χ p. χ t = 2αχ + C p. In regions of the α χ phase plane where α p = const, C p = const there are 2 critical points: (α, χ) = (±α 0, χ 0 ) 2α0 2 = α p + [αp 2 + Cp] 2 1/2 The critical point in the LH-half-plane ( α 0, χ 0 ) is an unstable spiral; The critical point in the RH-half-plane is (α 0, χ 0 ) is a stable spiral. J. D. Gibbon, CSCAMM; October

19 The next few slides: remarks on the direction of vorticity in Euler 1. The BKM theorem 2. The work of Constantin, Fefferman & Majda 1996 and Constantin 1994 The work of Deng, Hou & Yu 2005/6 Can our quaternionic Ricatti equation give anything in terms of P? J. D. Gibbon, CSCAMM; October

20 The Beale-Kato-Majda Theorem (CMP 94, 61-6, 1984) Theorem: There exists a global solution of the Euler equations u C([0, ]; H s ) C 1 ([0, ]; H s 1 ) for s 3 if, for every t > 0, t 0 ω(τ) L (Ω) dτ <. The proof is based on u c ω [1 + log H 3 ]. Thus one needs to numerically monitor only t 0 ω(τ) dτ. Corollary: If a singularity is observed in a numerical experiment of the form ω (t t) β then β must lie in the range β 1 for the singularity to be genuine & not an artefact of the numerical calculation. J. D. Gibbon, CSCAMM; October

21 Constantin, Fefferman & Majda; Comm PDEs, 21, , 1996 The image W t of a set W 0 is given by W t = X(t, W 0 ). W 0 is said to be smoothly directed if there exists a length ρ > 0 and a ball 0 < r < 1 2ρ such that: 1. ˆω(, t) has a Lipschitz extn to the ball of radius 4ρ centred at X(q, t) & M = lim sup t T q W 0 t 0 ˆω(, t) 2 L (B 4ρ ) dt <. i.e. the direction of vorticity is well-behaved in the nbhd of a set of trajectories. 2. The condition sup B3r (W t ) ω(x, t) m sup Br (W t ) ω(x, t) holds for all t [0, T ) with m = const > 0 ; i.e. this nbhd captures large & growing vorticity but not so that it overlaps with another similar region & sup B4r (W t ) u(x, t) U(t) := sup x u(x, t) < (Cordoba & Fefferman 2001; for tubes). Theorem: (CFM 1996) Assume that W 0 is smoothly directed as in (i) (ii). Then a time τ > 0 & a constant Γ s.t. for any 0 t 0 < T and 0 t t 0 τ sup B r (W t ) ω(x, t) Γ sup B ρ (W t ) ω(x, t 0 ). J. D. Gibbon, CSCAMM; October

22 The work of Deng, Hou & Yu; Comm PDEs, 31, , 2006 Consider a family of vortex line segments L t in a region of max-vorticity. Denote by L(t) the arc length of L t, ˆn the unit normal & κ the curvature. DHY define Uˆω (t) max (u ˆω) (x, t) (u ˆω) (y, t), x,y L t U n (t) max Lt u ˆn, and M(t) max ( ˆω L (L t ), κ L (L t )). Theorem: (Deng, Hou & Yu 06) : Let A, B (0, 1) with B = 1 A, and C 0 be a positive constant. If 1. Uˆω (t) + U n (t) (T t) A, 2. M(t)L(t) C 0, 3. L(t) (T t) B, then there will be no blow-up up to time T. Also J. Deng, T. Y. Hou & X. Yu; Comm. PDEs, 30, , J. D. Gibbon, CSCAMM; October

23 Using the pressure Hessian Imperial College London (see also Chae: T 0 S ˆω P ˆω dτ < ; Comm. P&A-M., 109, 1 21, 2006). Theorem: (JDG, Holm, Kerr & Roulstone 06) : a global solution of the Euler equations, u C([0, ]; H s ) C 1 ([0, ]; H s 1 ) for s 3 if T 0 χ p L (D) dτ <, with the exception of when ˆω becomes collinear with an e-vec of P at t = T. Proof: With Sŵ 2 = α 2 + χ 2, D Sŵ α Sŵ + α α p + χ χ p. (α 2 + χ 2 ) 1/2 Because D ω / = α ω, our concern is with α 0 D Sŵ α p + χ p. Possible that P ˆω blows up simultaneously as the angle between ˆω and P ˆω 0 thus keeping χ p finite; i.e. t 0 χ p L (D)dτ < but t 0 α p L (D)dτ. J. D. Gibbon, CSCAMM; October

24 Frame dynamics & the Frenet-Serret equations Imperial College London With ŵ as the unit tangent vector, ˆχ as the unit bi-normal and ŵ ˆχ as the unit principal normal, the matrix N can be formed N = ( ŵ T, (ŵ ˆχ) T, ˆχ T), with 0 χ a 0 DN = GN, G = χ a 0 c 1 χ 1 a. 0 c 1 χ 1 a 0 The Frenet-Serret equations for a space-curve are 0 κ 0 dn ds = F N where F = κ 0 τ, 0 τ 0 where κ is the curvature and τ is the torsion. J. D. Gibbon, CSCAMM; October

25 The arc-length derivative d/ds is defined by d ds = ˆω. The evolution of the curvature κ and torsion τ may be obtained from Ertel s theorem expressed as the commutation of operators [ D, ω ] = 0 d [ D α a ds +, d ] = 0. ds This commutation relation immediately gives α a F + DF = dg ds + [G, F ]. Thus Ertel s Theorem gives explicit evolution equations for the curvature κ and torsion τ that lie within the matrix F and relates them to c 1, χ a and α a. J. D. Gibbon, CSCAMM; October

26 Mixing Consider a passive vector line-element δl in a flow transported by an independent velocity field u. For small δl we have the same equations as Euler for ω Dδl D = δl u = t + u Following the analogy with Euler, Ertel s Theorem holds so there is a b-field: ( ) D(δl u) Du = δl Du/ represents any dynamics one wishes to impose on the problem. Thus all the conditions hold for Theorem 1: 1. w = δl 2. a = δl u 3. b = δl ( ) Du a Ricatti equation plus an ortho-normal frame... J. D. Gibbon, CSCAMM; October

27 Ideal MHD Consider a magnetic field B coupled to a fluid (div u = 0 = div B) Du DB = B B p = B u Defining Elsasser variables with ±-material derivatives (two time-clocks) v ± = u ± B ; D ± = t + v± the magnetic field B and v ± satisfy with div v ± = 0 D ± v = p ; D ± B = B v ± Moffatt (1985) suggested that B takes the place of ω in ideal MHD. J. D. Gibbon, CSCAMM; October

28 Ertel s Theorem (proof omitted) for this system is D (B v ± ) = P B. With two time-clocks, we have the correspondence w B a ± B v ± b P B α pb = ˆB P ˆB Define tetrads q ± and q pb as follows χ pb = ˆB P ˆB q ± = [ α ±, χ ±] q pb = [ ] α pb, χ pb. The tetrads q ± satisfy the compatibility relation D q ± + q ± q + q pb = 0 J. D. Gibbon, CSCAMM; October

29 MHD-Lagrangian frame dynamics Imperial College London We have 2 sets of orthonormal vectors opposite Lagrangian time derivatives. ˆB, ( ˆB ˆχ ± ), ˆχ ± acted on by their D ˆB D ( ˆB ˆχ ± ) D ˆχ ± = D ˆB, = D ( ˆB ˆχ ± ), = D ˆχ ± where the pair of Darboux vectors D are defined as D = χ c 1 χ ˆB, c 1 = ˆB [ ˆχ ± (χ p b + α ± χ )]. J. D. Gibbon, CSCAMM; October