Hamiltonian and Non-Hamiltonian Reductions of Charged Particle Dynamics: Diffusion and Self-Organization

Transcription

1 NNP th July 2017 Lawrence University Hamiltonian and Non-Hamiltonian Reductions of Charged Particle Dynamics: Diffusion and Self-Organization N. Sato and Z. Yoshida Graduate School of Frontier Sciences The University of Tokyo

2 INTRODUCTION z θ r Magnetospheres are ubiquitous in the universe. Dipole magnetic fields confine charged particles. Spontaneous confinement (self-organization) is achieved by inward diffusion. 2/38

3 MAGNETOSPHERIC CONFINEMENT OF CHARGED PARTICLES Magnetospheric confinement is a promising source of applications. Advanced fusion reactor based on D- 3 H reaction [1]. Trap for matter and antimatter plasmas [2]. Study of astrophysical systems [3]. The creation of the density gradient seemingly violates the entropy principle. OUTLINE 1. Diffusion in a dipole field: correct expression of diffusion operator in magnetic coordinates. 2. Diffusion in non-integrable magnetic field B B 0. [1] A. Hasegawa et al., A D- 3 H Fusion Reactor Based on a Dipole Magnetic Field, Nuc. Fus. 30, 11 (1990). [2] T. S. Pedersen and A. H. Boozer, Confinement of Nonneutral Plasmas on Magnetic Surfaces, Phys. Rev. Lett. 88, 20 (2002). [3] A. M. Persoon et al., The Plasma Density Distribution in the Inner Region of Saturn s Magnetosphere, J. Geo. Res.: Space Physics 118, (2013). 3/38

4 THE RT-1 MACHINE (I) RT-1 produces a laboratory magnetosphere. A similar device is LDX at MIT. [4] Z. Yoshida et al., Self-organized confinement by magnetic dipole: recent results from RT-1 and theoretical modeling, Plasma Phys. Control. Fus (2012). [5] Z. Yoshida et al., Magnetospheric Vortex Formation: Self-Organized Confinement of Charged Particles, Phys. Rev. Lett. 104, (2010). 4/31

5 THE RT-1 MACHINE (II) 5/31

6 EXPERIMENTAL OBSERVATIONS [6] H. Saitoh et al., Observation of a New High-β and High-Density State of a Magnetospheric Plasma in RT-1, Phys. Plasmas 21, (2014). [7] Y. Kawazura et al., Observation of Particle Acceleration in Laboratory Magnetosphere, Phys. Plasmas 22, (2015). 6/31

7 THE POINT DIPOLE MAGNETIC FIELD B = ψ θ B l ψ FLUX FUNCTION ψ r, z = r 2 r 2 + z 2 3/2 LENGTH OF A FIELD LINE [7] l r, z = 1 2ψ ቈ 1 log ቆ 3 1 rψ 2/3 3 θ [8] N. Sato and Z. Yoshida, A Stochastic Model of Inward Diffusion in Magnetospheric Plasmas, J. Phys. A: Math. Theor (2015). 7/31

8 INWARD DIFFUSION Fickian Diffusion Inward Diffusion Flux-tube ψ It is thought that particle number per flux-tube volume tends to be homogenized [9,10]. [9] A. C. Boxer et al., Turbulent Inward Pinch of Plasma Confined by a Levitated Dipole Magnet, Nature Physics, Vol. 6, (2010). [10] A. Hasegawa, Motion of a Charged Particle and Plasma Equilibrium in a Dipole Magnetic Field, Phys. Scr. T (2005). 8/31

9 THE MECHANISM OF INWARD DIFFUSION In a dipole field the motion of a charged particle is constrained by 3 adiabatic invariants: ψ J μ STRENGTH OF CONSTRAINT Each invariant is preserved on a different time scale: ω d ω b ω c In a dipole plasma the key normal transport mechanism is diffusion by E B drift: v = E B. 9/31

10 DIFFUSION ON A FOLIATED PHASE SPACE Conservation of adiabatic invariants foliates phase space: particle motion is constrained on leaves [11,12]. Electromagnetic fluctuations break the weakest topological constraints. The result is a random process among leaves. A diffusion operator must be formulated on the corresponding invariant measure [8,13]. Entropy should be maximized on the symplectic leaf [14]. [8] N. Sato and Z. Yoshida, A Stochastic Model of Inward Diffusion in Magnetospheric Plasmas, J. Phys. A: Math. Theor (2015). [11] N. Sato, N. Kasaoka, and Z. Yoshida, Thermal Equilibrium of Non-Neutral Plasma in Dipole Magnetic Field, Phys. Plasmas 22 (2015), [12] Z. Yoshida and S. M. Mahajan, Self-Organization in Foliated Phase Space: Construction of a Scale Hierarchy by, Prog. Theor. Exp. Phys (2014), 073J01 [13] N. Sato, Z. Yoshida, and Y. Kawazura, Self Organization and Heating by Inward Diffusion in Magnetospheric Plasmas, Plasma Fus. Res (2016). 10/31 [14] N. Sato and Z. Yoshida, Up-Hill Diffusion, Creation of Density Gradients: Entropy Measure for Systems with Topological Constraints, Phys. Rev. E (2016).

11 INWARD DIFFUSION: A NON-CANONICAL HAMILTONIAN SYSTEM The constraint causing non-canonicality is the conservation of μ: the pair θ c, μ can be reduced: dx dy dz dp x dp y dp z = m 2 dl dv dθ dψ dθ c dμ. Dynamical variables are reduced to four: l, v, θ, ψ. The reduced equations are the guiding center equations: ቊ ሶ v = m 1 μb + eφ l + v v E B k, v = v + v E B + v B + v k. Transforming to the new coordinates: ሶ l = v qφ θ, v ሶ = m 1 μb + eφ l + v q l φ θ, θ ሶ = ψ + q l e 1 μb + φ e 1 mv 2 q l, ψ ሶ = φ θ. 11/31

12 CONSTRUCTION OF CANONICAL COORDINATES The reduced system should be Hamiltonian because the reduced variable μ is a dynamical constant and the phase θ c can be neglected. We look for the nondegenerate closed two-form ω = dλ such that: i X ω = dh 0, with H 0 = μb + m v eφ the Hamiltonian function. Setting q = l ψ : ω = mdv dl + dψ d eθ + qv. Setting: η = eθ + qv, the canonical pairs are l, mv and η, ψ. They describe the motion parallel and perpendicular to the magnetic field respectively. 12/31

13 INVARIANT MEASURE AND APPLICATION OF THE ERGODIC HYPOTHESIS Observe that: Thus, the flow X = dl dv dθ dψ = e 1 dl dv dη dψ. l, ሶ vሶ, θ, ሶ ψሶ is measure preserving in virtue of Liouville s theorem [15]. To formulate a diffusion operator on the symplectic leaf μ, consider two hypothesis: 1. Perturbations with vanishing ensemble average E = Ergodic Hypothesis [16] on the symplectic leaf: E = dφ = m e D 1/2 1/2 1/2 Γ dl + D Γ dθ + D θ Γθ dψ, with dw = Γdt a Wiener process. The E B drift velocity causing diffusion is then written as: 1/2 dx = 1 δφ rb θ dt = D rb dw. [15] J. W. Gibbs, Elementary Principles in Statistical Mechanics, chap. 1 (Scribner s Sons, New York, 1902), pp [16] G. D. Birkhoff, Proof of the Ergodic Theorem, Proc. N. A. S. 17, pp (1931). 13/31

14 FOKKER-PLANCK EQUATION ON THE SYMPLECTIC LEAF (I) dl = v + C l α D ψ + q l q + q ψ + q l log rb dt + qd 1/2 dw dv = μ B m l + γv C v dt + D 1/2 dw D 1/2 v q l dw f t = l + v v + C l α D ψ + q l q + q ψ + q l log rb f + m2 2e 2 D D dθ = μ e ψ + q l B + C θ m e v 2 q l μ B m l + γv C v ψ D f θ μ e ψ + q l B + C θ m e v 2 q l 1 2 α ψ + q l log rb + C ψ f D 1/2 dt D θ dwθ m e qd 1/2 dw dψ = D 1 2 α ψ + q l log rb + C ψ dt + D 1/2 dw 2 θ 2 q2 f m e D 2 v θ qf + D 2 l ψ qf αd l 2 2 v v q 2 2 l f D v ψ v q l f+αd v v q l + ψ q l v q 2 l f 2 f l 2 q2 f D 2 f 2 v D 2 f ψ D 2 f θ θ 2 2 ψ + q l q f D l v v qq l f f dl dψ dθ dv dμ dθ c = fb dx dy dz dv dμ dθ c 14/31

15 CHANGE OF COORDINATES TO THE LABORATORY FRAME Diffusion is controlled by D (strength of fluctuations), α (time-properties of fluctuations), and q = l ψ (non-orthogonality of tangent vectors). Approximating D = D, α = 1/2, q~0, θ = 0, we have: f t = v f l + μ B m l f v D Note: assuming a diffusion operator arbitrarily, such as 2 2 D x f, gives wrong results. The creation of the density gradient is caused by the inhomogeneous Jacobian B linking symplectic leaf to laboratory frame: dl dψ dθ dv = B dx dy dz dv. The density ρ x, y, z in the laboratory frame x, y, z is: x 2 f ψ 2 ρ = B u = B න f dv dμ dθ c, where u is the density on the symplectic leaf. If u becomes flat due to inward diffusion, ρ must become inhomogeneous. 15/31

16 CREATION OF DENSITY GRADIENT Symplectic Leaf Density u Laboratory Frame Density ρ = B u BEFORE z AFTER r 16/31

17 EMERGENCE OF TEMPERATURE ANISOTROPY Laboratory Frame Density ρ Temperature Anisotropy T /T 17/31

18 ENTROPY (I) The system satisfies Boltzmann s H-theorem on the invariant measure [14]: dv IM = dl dψ dθ dv dμ dθ c = B dx dy dz dv dμ dθ c = B dv. Therefore, the proper entropy measure is: Σ = න Ω f log f dv IM. The entropy measure in the laboratory coordinates reads: ሚS = න Ω fb log fb dv = Σ log B. Recalling that t f = div fz, the rate of change in Σ and ሚS are: dσ dt = div Z Φ Ω = σ Φ Ω, d ሚS dt = div Z Φ Ω d dt log B = σ Φ Ω d dt log B. Here σ is the entropy production rate. [14] N. Sato and Z. Yoshida, Up-Hill Diffusion, Creation of Density Gradients: Entropy Measure for Systems with Topological Constraints, Phys. Rev. E (2016). 18/31

19 ENTROPY (II) Pictures from [13]. (a) Time evolution of entropy measures Σ and ሚS during formation of a radiation belt. (b) Time evolution of entropy production σ. This result can be extended to all Poisson operators I (and to any operator endowed with an invariant measure for any choice of H 0 ): the proper entropy measure Σ is defined on dv IM, and the solution f to the FPE satisfies: lim I d log f + βdh 0 = 0 a. e. t Darboux s theorem [14] ensures that a Poisson operator has an integrable kernel I dc i = 0. In regions U M where rank I is constant: lim f = exp βh 0 γ i C i a. e. t [14] N. Sato and Z. Yoshida, Up-Hill Diffusion, Creation of Density Gradients: Entropy Measure for Systems with Topological Constraints, Phys. Rev. E (2016). [17] M. de León and M. R. Rodrigues, Methods of Differential Geometry in Analytical Mechanics, Elsevier, p (1989). 19/31

20 TOPOLOGICAL CONSTRAINTS AND INTEGRABILITY A dipole magnetic field is integrable, i.e. it does not have helicity: B = ζ B B = 0. What happens to diffusion in an arbitrarily complex magnetic field B B 0? [20] Microscopic systems arise in Canonical Hamiltonian form. μ leaf [18] ρ x, y Macroscopic systems result from the reduction of negligible degrees of freedom that impart topological constraints. In non-canonical Hamiltonian systems, selforganization is driven by integrable constraints. The reduction process may destroy the Hamiltonian form: the standard formulation of statistical mechanics does not apply due to absence of phase space. The simplest non-hamiltonian behavior arises in 3 dimensions: V = w H 0. [18] Z. Yoshida, Self-Organization by Topological Constraints: Hierarchy of Foliated Phase Space, Adv. Phys. X 1, pp (2016). 20/31

21 6D phase space: ω 2 = dp i dx i = J c 1 E B DRIFT IS NOT AN HAMILTONIAN SYSTEM Motion of a charged particle in an e-m field m dv dt Charged particle Hamiltonian: H 0 = 1 p qa 2 + qφ 2m equations of motion: i V ω 2 = dh with V = p, ሶ xሶ = q V B + E Canonical Hamiltonian form Reduction m 0 (non canonical Hamiltonian form) 3D phase space: ω 2 = da i dx i = da = B Non inertial Hamiltonian: H 0 = qφ Equations of motion: i V da = dφ with V B + E = 0 and X = x, ሶ y, ሶ zሶ w w = Reduction V 0 V = w H 0 = B B 2 E. B B B 4 0 violation of Jacobi identity. 21/31

22 Consider a Beltrami field w = μ w. TOPOLOGICAL CONSTRAINTS IN 3D Since V = w H 0, the motion is constrained to be orthogonal to w: w V = 0. Euler s equation for the motion of a rigid body with angular momentum x, moment of inertia I = I x, I y, I z and Hamiltonian H 0 = 1 2 This system has a non-integrable constraint since the Jacobi identity is not satisfied: x 2 V = x H 0. I x + y2 I y + z2 I z : This is a non-canonical Hamiltonian system with integrable constraint x V = 0 which can be integrated to the Casimir invariant C = x 2 /2: ሶ C = C V = x V = 0. w w = μw /31

23 CONSTRAINED ORBITS Euler Rigid Body (Hamiltonian) w = x Beltrami (Non-Hamiltonian) w = cos z sin y, sin z, cos y X = w H = w H 0 + Γ 23/31

24 E B DIFFUSION AND CLASSIFICATION OF CONSERVATIVE DYNAMICS Assume charge neutrality H = H 0 + Γ = e φ + Γ = Γ. Then: V = w Γ, ρ t = 1 w ρw. 2 The form of ρ x depends on the geometrical properties of w = B/B 2. Magnetic field with helicity B B 0. Magnetic field without helicity B B = 0. NON-BELTRAMI w = w ρ + b ρ + bρ = 0 BELTRAMI b = 0 ρ = constant POISSON (Hamiltonian) w w = 0 w = λ C ρ λ = f C NULL w = 0 ρ = ρ NON-BELTRAMINESS b = b = w w 24/31

25 NUMERICAL TEST OF 3D DIFFUSION (I) The theory is tested by comparing the numerical integration of the stochastic equation of motion with the analytical solution of the Fokker-Planck equation. (a) (b) (c) Sample size: particles per simulation. (a) Initial uniform particle distribution in 3D cube with periodic boundaries. (b) Noise Γ i as a function of time t. (c) 250 sample paths in x, y plane for w = 1 + cos 2 x z cos x cos y. 25/31

26 w B/B 2 NUMERICAL TEST OF 3D DIFFUSION (II) w 1 x, y B x, y ρ x, y Poisson Operator (Constant Magnetic Field) w = B = z. Analytical Solution ρ = constant. Poisson Operator (Gradient Magnetic Field) w = z cos x cos y. Analytical Solution ρ = constant. 26/31

27 w w = B/B 2 B/B 2 NUMERICAL TEST OF 3D DIFFUSION (III) w 1 x, y B x, y B x, y ρ x, y Poisson Operator (Gradient Magnetic Field) w = λ C = 1 + cos 2 x z cos x cos y. Analytical Solution ρ = λ 1 f C. Beltrami Operator (Beltrami Magnetic Field) w = 2B = cos z + sin z cos z sin z 0 Analytical Solution ρ = constant. 27/31

28 w w = B/B 2 B/B 2 NUMERICAL TEST OF 3D DIFFUSION (IV) w 1 x, y B x, y B x, y ρ x, y Non-Beltrami Operator (Non-Integrable Magnetic Field) b 0 w = 1, sin x + cos y, cos x. Non-Beltrami Operator (Non-Integrable Magnetic Field) (free boundaries) b 0 w = 1 + y sin y cos y 1,0, 2 y sin y cos y 2 sin x sin x 2. 28/31

29 NUMERICAL TEST OF 3D DIFFUSION (V) In the most general case of a non-integrable operator w w 0 with b 0 the density profile is determined by field strength w and non-beltraminess b. To elucidate the role of b, we consider an operator of uniform strength: w 1 = constant. w w = B/B 2 B/B 2 b = b b x, y = w w ρ x, y Non-Beltrami Operator (Non-Integrable Magnetic Field) b 0 w = cos y, cos x, sin y 1 + cos 2 x w = constant 29/31

30 CONCLUSION Dipole fields can efficiently trap charged particles by the mechanism of inward diffusion. The inward diffusion occurs in virtue of the topological constraint imposed by μ. The creation of the density gradient is consistent with the second law of thermodynamics. The density profile resulting from diffusion by a white noise electric field in an arbitrary magnetic field B is determined by its geometric properties. POISSON (Hamiltonian) BELTRAMI w w = 0 w = λ C b = 0 ρ = ρ λ, C ρ = constant In 3D, three classes of w = B/B 2 can be identified: Poisson, Beltrami, and Non-Beltrami. NON- BELTRAMI b 0 ρ = ρ w, b The construction of a diffusion operator requires careful treatment, especially when the system is not endowed with a canonical Hamiltonian structure. The statistical theory developed here can be applied to general conservative systems. Diffusion by E B drift is of special mathematical interest because it involves a non-elliptic partial differential equation. 30/31

31 End