Broadwell Model in a Thin Channel Peter Smereka Collaborators: Andrew Christlieb James Rossmanith Affiliation: University of Michigan Mathematics Department Motivation Example: – Gas at Low Density • Satellites and Solar Winds • Plasma Thrusters • Space Planes – High Density Gases • Flow in a Nano-Tube – Applications: Chemical Sensors Oxford University's Carbon and Nanotech Group Quic kTime™ and a TIFF (Uncompres sed) decompressor ar e needed to see this picture. NASA Starting Point Boltzmann’s Equation: f F f v x f v f t m t collision y=0 Maxwell’s Boundary condition (v>0): f (u,v) (1 ) f (u,v) f M (u,v) v'0 f (u',v') | v'| dv'du' Limiting Behavior with No Walls • Fluid Dynamic Limit: Kn – Large length scales, Kn<<1, highly collisional. L – Solution of Boltzmann equation can be expressed as u u x,t 2 0 f M x,t exp T x,t where is density, u is velocity and T is temperature which are governed by the Navier-Stokes Equations • Free Molecular Flow: – Small length scales, Kn>>1, fluid appears collisionless – In this case, there is no ‘simple’ reduction Flow In a Thin Channel • Mean Free Path Air ~ 70 nm • Nano-Tube Diameter ~ 30 nm • Knudsen Number, Kn ~ O(1) We make the collisionless flow approximation but keep the wall collisions Knudsen Gas • Collisionless Flow • Maxwell’s Boundary Condition on walls f v x f 0 t h Diffusive Behavior • Knudsen Gas has Diffusive Behavior • The depth averaged density, ,under appropriate scaling, satisfies a diffusion equation f v x f 0 t h Average and “ wait long enough’’ Maxwell’s Boundary Condition 2 D 2 t x Diffusive Behavior Diffusion Coefficient: D 2 2 Thin Tube: time scale = 1/h Babovsky (1986) Thin channel : time scale = 1/(h log h) Cercignani (1963), Borgers et.al. (1992), Golse (1998) Discrete Velocity Models Discrete velocity models are very simplified versions of the Boltzmann equation which preserve some features, namely: • H-theorem: Entropy must increase • Kn small-> Chapman-Enskog -> Fluid equations Reference: T. Platkowski and R. Illner (1988) ‘Discrete velocity models of the Boltzmann equation: A survey on the mathematical aspects of the theory.’ SIAM REVIEW, 30(2):213. The Broadwell Model 3 6 velocities • No Long Range Forces 2 • 6 velocities with magnitude = 1 N i N i vi Si Li t x i 6 1 5 4 i 1,2,3,4,5,6 Si and Li are source and losses due to collisions Collisions No Gain or Loss for 1 Gain for 1 from 3-4 collision 3 3 3 1 Loss for 1 from a 1-2 collision 1 2 1 2 1 3 Result: 4 4 1 1 2 S1 L1 c N 3 N 4 N 5 N 6 N1 N 2 3 3 3 BROADWELL MODEL 1 N1 N1 1 2 c c N 3 N 4 N 5 N 6 N1 N 2 3 t x 3 3 1 N 2 N 2 1 2 c c N 3 N 4 N 5 N 6 N1 N 2 3 t x 3 3 1 N 3 N 3 1 2 c c N1 N 2 N 5 N 6 N 3 N 4 3 t y 3 3 1 N 4 N 4 1 2 c c N1 N 2 N 5 N 6 N 3 N 4 3 t y 3 3 1 N 5 N 5 1 2 c c N1 N 2 N 3 N 4 N 5 N 6 3 t z 3 3 1 N 6 N 6 1 2 c c N1 N 2 N 3 N 4 N 5 N 6 3 t z 3 3 Broadwell Model There is large body of work on Broadwell models mainly focusing on the fluid dynamic limit. This is the regime in which inter-particle collisions dominate. • Broadwell (1964): 1D Shock Formation: Kinetic vs. Fluid • Gatignol (1975): H- Theorem + Kinetic theory • Caflisch (1979): Proved validity of 1D fluid-dynamical to Broadwell model up to formation of shocks • Beale (1985): Proved existence of time global solutions to a1D Broadwell model Flow in a Thin Channel Set Up • Use Broadwell Model to Understand Flow in a Thin Channel y h x z d L • Assumptions: – Channel height, h, is small compared to length, L. – Channel depth is infinite – Dominant collisional effect: WALL Broadwell with Boundaries To incorporate wall effects we “rotate’’ the Broadwell model by 45 degrees in the x-y plane. The other velocities are parallel to the wall. y=h N2 N3 N1 N2 N4 N4 N3 N1 y=0 Boundary Conditions : Accommodation Coefficient N4 N2 Inward Flux N1 N3 Specular N1 Diffuse N4 has specular reflections into N1 : N1=(1N4 N4 has diffusive reflection into N1 : N1=(N4)/2 N2 has diffusive reflection into N1 : N1=(N2)/2 At lower wall: N1 1 N4 2 N2 N4 FULL MODEL 1 N1 N1 N1 1 2 2 c c 2c N 3 N 4 N 5 N1N 2 3 t x y 3 3 1 N 2 N 2 N 2 1 2 2 c c 2c N 3 N 4 N 5 N1N 2 3 t x y 3 3 1 N 3 N 3 N 3 1 2 2 c c 2c N1N 2 N 5 N 3 N 4 3 t x y 3 3 1 N 4 N 4 N 4 1 2 2 c c 2c N1N 2 N 5 N 3 N 4 3 t x y 3 3 1 N 5 1 2 2c N1N 2 N 3 N 4 N 52 3 t 3 3 N 1 N N 2 N 4 4 1 2 N 1 N N 2 N 4 2 3 2 y=0 N 1 N N1 N 3 3 2 2 N 1 N N N 1 1 3 4 2 y=h Free Molecular Flow y=h N3 N1 N2 N4 N2 N1 N N c 1 c 1 0 t x y N 2 N N c 2 c 2 0 t x y N 3 N 3 N 3 c c 0 t x y N 4 N 4 N 4 c c 0 t x y N4 N3 N1 y=0 N 1 N 4 N 2 N 4 1 2 N 1 N N N 2 2 4 3 2 y=0 N 1 N 3 N1 N 3 2 2 N 1 N N N 1 1 3 4 2 y=h Depth Average Define: 1 N i (t, x) h h N (t, x, y)dy i 0 Depth Average Equation: N N1 c 1 c N1(t, x,0) N1 (t, x,h) t x h N 2 N 2 c c N 2 (t, x,h) N 2 (t, x,0) t x h N 3 N 3 c c N 3 (t, x,0) N 3 (t,x,h) t x h N 4 N 4 c c N 4 (t, x,h) N 4 (t, x,0) t x h N1 1 N 4 2 N 2 N 4 N 1 N N N 2 2 4 3 2 y=0 N 2 1 N 3 2 N1 N 3 N 1 N N N 1 1 3 4 2 y=h Depth Average Applying the boundary conditions gives: N1 N1 c c (1 )N 4 0 N 2 0 N 4 0 N1 h t x h 2 N 2 N 2 c c (1 )N 3 h N1 h N 3 h N 2 0 t x h 2 N 3 N 3 c c (1 )N 2 0 N 2 0 N 4 0 N 3 h t x h 2 N 4 N 4 c c (1 )N1 h N1 h N 3 h N 4 0 t x h 2 Depth Average Define: (t, x) N1(t, x) N 2 (t, x) N 3 (t, x) N 4 (t, x) m(t, x) c N1(t, x) N 2 (t, x) N 3 (t, x) N 4 (t, x) Adding N1 through N4 gives: m 0 t x Adding cN1 and cN4 then subtracting cN2 and cN3 gives: 2 m c c 2 N1 h N 2 0 N 3 h N 4 0 t x h Thin Channel Approximation Taylor Series: h N1 (t, x,h) N1 (t, x,h) N1 (t, x) O(h 2 ) 2 y Combined with: N1 N1 N1 c c 0 t x y h N1 N1 Gives: N1 (t, x,h) N1(t, x) c 2c t x Thin Channel Approximation This approximation for N1(t, x,h) along with similar approximations for the other boundary terms gives m 2 m 2 c c c m t x 2h t x h We have the system of equations are: m 0 t x Loss of 2 c m 2 c m Momentum t x 2 h To Wall Telegraph Equation These maybe combined to give: 2 h 2 2 2 hc 2 2 t c t x Previous Results • Solutions to Telegraph Equation Converge to Diffusion Equation on a long time scale. (Zauderer: Partial Differential Equations of Applied Mathematics) 2 h 2 2 2 2 hc 2 hc 2 2 2 t c t 2 x x t • So we Expect that Solutions of Broadwell Model Converge to Solutions of Diffusion Equation Limiting Behavior Rescale so that c=h=1 Domain we consider: D (x, y) | (,) [0,1] Define: N(x, y,t) (N1,N 2,N 3,N 4 )T 1 product: u,v u H v dy Define an inner 0 Define: 1=(1,1,1,1)T and 1+/-=(1,-1,-1,1)T (x,t) 1,N(x,,t) and m(x,t) 1 /,N(x,,t) N B(D) if max N N N x dx N L1 H1 () 0y1 2 2 Theorem 1 - Diffusive Behavior Diffusive scaling: X=x/ and T=t/2 Scaled Number Density : M(X,y,T) = N(X,y, 2 T) Define Scaled Density: y (X,T) 1,M (X,,T) Theorem 1: If the initial conditions are N(x,y,0)= Mo(x/ ,y)/,where Mo(x ,y) is in B(D), then as > y(X,T) converges weakly to y(X,T) where 2 yT yXX , with y(X,0) 1, Mo (X,) . 2 Theorem 2 - Hyperbolic Behavior Hyperbolic scaling: X=x/ T=t/ Scaled Number Density: P(X,y,T) = N(X,y, T,=2G/) Define Scaled Density: (X,T) 1,P (X,,T) Theorem 2: If the initial conditions are N(x,y,0)= Mo(x/ ,y)/ in B(D), then as > , (X,T) converges weakly to (X,T) which is a solution of the telegraph T TT XX 2G equation: with initial conditions : (X,0) 1, M o (X,) , m(X,0) 1 , M o (X,) where: T mX 0 Theorem 3 Long-Time Behavior Theorem 3: If N(x, y,0) = No(x, y) in B(D) and Nˆ o (k, y) 4 gˆ n j1 j,n (k) j,n (k, y) where are j,n vector-valued eigenfunctions, then the density has the following asymptotic behavior: 2 x im(xt ) im(x t ) (x,t) exp e c m e o1 Dt 4Dt m D=(2/2 and the c’s are determined initial conditions (continued) Theorem 3 Long-Time Behavior Furthermore, if No=(f(x)/4)1 then (x,0)=f(x) and it follows from the above expressions that x 2 i m(xt ) i m(x t ) ˆ (x,t) exp f e o1 (m )e Dt 4Dt m This shows the convergence in Thm 1 cannot be better than weak Results - Initial Condition = f(x) Results-Initial condition = f(x,y) Effects of Collisions N1 N1 c 2c c N1 h N1 0 t x h 3h h 2 N N N 3 4 5 2N1N 2 dy 0 N 2 N 2 c 2c c N 2 h N 2 0 t x h 3h 2 N N N 3 4 5 2N1N 2 dy N 3 N 3 c 2c c N 3 h N 3 0 t x h 3h h N 4 N 4 c 2c c N 4 h N 4 0 t x h 3h N 5 t 2c 3h h 0 2 N N N 1 2 5 2N 3N 4 dy 0 h 2 N N N 1 2 5 2N 3N 4 dy 0 h 2 N N N N 2N 1 2 3 4 5 dy 0 Depth Averaging c The boundary terms, h Ni h Ni 0 , are treated using the thin channel approximation. Need to approximate the terms h NiN j dy i {1,3,5} and j {2,4,5} h 0 By Taylor expanding one can show h N N dy N N h i j i 2 O( h ) j 0 The approximation is O(h) provided O(h 1 ) Collisional Thin Channel Defining the averaged variables: (t, x) N1 (t, x) N 2 (t, x) N 3 (t, x) N 4 (t, x) 2N 5 (t, x) m(t, x) cN1 (t, x) N 2 (t, x) N 3 (t, x) N 4 (t, x) z(t, x) N5 (t, x) After similar algebra as before we arrive at: m 0 t x 2 c m 2 c m 2z 2 h t x z m 2 2z 6z 2 t 24 c Long time behavior When = O(1) and t = O(1/h) then one has 2 D 2 t 3 x 2 where 2 D ch 2 is the diffusion coefficient in the collisionless case Results Conclusions • We have provided a coarse-grained description for the Broadwell model with and without collisions which is valid over a wide range of time scales. • We expect this model to provide insight for the more realistic case when the gas is modeled by the Boltzmann equation.