Lecture 15 Notes: The Coffee Cup

18.354J Nonlinear Dynamics II: Continuum Systems 15 Lecture 15 The coffee cup Let’s try and apply our knowledge of fluid dynamics to a real observation, to test whether the theory actually works. We shall consider a question you will encounter in the last problem set21 : how long does it take a cup of coffee (or glass of water) to spin down if you start by stirring it vigorously? To proceed, we need a model of a coffee cup. For mathematical simplicity, let’s just take it to be an infinite cylinder occupying r ≤ R. Suppose that at t = 0 the fluid and cylinder are spinning at an angular frequency ω, and then the cylinder is suddenly brought to rest. Assuming constant density, we expect the solution to be cylindrically symmetric p(x, t) = p(r, t) , u(x, t) = uφ (r, t)eφ , (398) with only a component of velocity in the angular direction eφ . This component will only depend on r, not the angular coordinate or the distance along the axis of the cylinder (because the cylinder is assumed to be infinite). We shall just plug the assumed functional form into the Navier-Stokes equations and see what comes out. Let’s put our ansatz p = p(r, t) and u = (0, uφ (r, t), 0), which sat­ isfies V · u = 0, into the cylindrical Navier-Stokes equations. The radial equation for er ­ component becomes u2φ ∂p = . (399a) r ∂r Physically this represents the balance between pressure and centrifugal force. The angular equation to be satisfied by the eφ -component is ∂uφ =ν ∂t ∂ 2 uφ 1 ∂uφ uφ + − 2 r ∂r r ∂r2 , (399b) and the vertical equation is 0= 1 ∂p . ρ ∂z (399c) The last equation of these three is directly satisfied by our solution ansatz, and the first equation can be used to compute p by simple integration over r once we have found uφ . We want to solve these equation (399a) and (399b) with the initial condition uφ (r, 0) = ωr, (400a) and the boundary conditions that uφ (0, t) = 0 , uφ (R, t) = 0 ∀t > 0. (400b) This is done using separation of variables. Since the lhs. of Eq. (399b) features a firstorder time derivative, which usually suggests exponential growth or damping, let’s guess a solution of the form 2 uφ = e−k t F (r). (401) Putting this into the governing equation (399b) gives the ODE −k 2 F = ν F "" + 21 See Acheson, Elementary Fluid Dynamics, pp. 42-46 F" F − 2 r r (402) 0.6 J1'Ξ� 0.4 0.2 0.0 -0.2 0 10 5 15 Ξ Figure 3: The Bessel function of the first order J1 (ξ). This equation looks complicated. However, note that if the factors of r weren’t in this equation we would declare victory. The equation would just be F "" + k 2 /νF = 0, which √ has solutions that are sines and cosines. The general solution would be Asin(k/ νr) + √ Bcos(k/ νr). We would then proceed by requiring that (a) the boundary conditions were satisfied, and (b) the initial conditions were satisfied. We rewrite the above equation as r2 F "" + rF " + and make a change of variable, The equation becomes k2 2 r − 1 F = 0, ν (403) √ ξ = kr/ ν. ξ 2 F "" + ξF " + ξ 2 − 1 F = 0. (404) Even with the factors of ξ included, this problem is not more conceptually difficult, though it does require knowing solutions to the equation. It turns out that the solutions are called Bessel Functions. You should think of them as more complicated versions of sines and cosines. There exists a closed form of the solutions in terms of elementary functions. However, people usually denote the solution to the Eq. (404) as J1 (ξ), named the Bessel function of first order22 . This function is plotted in Fig. 3, satisfies the inner boundary conditions J1 (0) = 0. For more information, see for example the book Elementary Applied Partial Differential Equations, by Haberman (pp. 218-224). Now let’s satisfy the boundary 22 Bessel functions Jα (x) of order α are solutions of ( ) x2 J ff + xJ f + x2 − α J = 0 condition uφ (R, t) = 0. Since we have that √ uφ = AJ1 (ξ) = AJ1 (kr/ ν), (405) √ this implies that AJ1 (kR/ ν) = 0. We can’t have A = 0 since then we would have nothing √ √ left. Thus it must be that J1 (kR/ ν) = 0. In other words, kR/ ν = λn , where λn is the nth zero of J1 (morally, J1 is very much like a sine function, and so has a countably infinite number of zeros.) Our solution is therefore ∞ 2 2 An e−νλn t/R J1 (λn r/R). uφ (r, t) = (406) n=1 To determine the An ’s we require that the initial conditions are satisfied. The initial con­ dition is that uφ (r) = ωr. (407) Again, we now think about what we would do if the above sum had sines and cosines instead of J1 ’s. We would simply multiply by sine and integrate over a wavelength. Here, we do the same thing. We multiply by rJ1 (λm r/R) and integrate from 0 to R. This gives the formula R An R ωr2 J1 (λm r/R)dr. rJ1 (λn r/R)J1 (λm r/R) dr = 0 (408) 0 Using the identities R rJ1 (λn r/R)J1 (λm r/R) dr = 0 and R ωr2 J1 (λm r/R)dr = 0 R2 J2 (λn )2 δnm , 2 ωR3 J2 (λm ), λm (409) (410) we get An = − 2ωR , λn J0 (λn ) (411) where we have used the identity J0 (λn ) = −J2 (λn ). Our final solution is therefore ∞ uφ (r, t) = − n=1 2ωR 2 2 e−νλn t/R J1 (λn r/R). λn J0 (λn ) (412) Okay, so this is the answer. Now lets see how long it should take for the spin down to occur. Each of the terms in the sum is decreasing exponentially in time. The smallest value of λn decreases the slowest. It turns out that this value is λ1 = 3.83. Thus the spin down time should be when the argument of the exponential is of order unity, or t∼ R2 . νλ21 (413) This is our main result, and we should test its various predictions. For example, this says that if we increase the radius of the cylinder by 4, the spin down time increases by a factor of 16. If we increase the kinematic viscosity ν by a factor of 100 (roughly the difference between water and motor oil) then it will take roughly a factor of 100 shorter to spin down. Note that for these predictions to be accurate, one must start with the same angular velocity for each case. In your problem set you are asked to look at the spin down of a coffee cup. From our theory we have a rough estimate of the spin down time, which you can compare with your experiment. Do you get agreement between the two?