WKB Analysis Slowly varying coefficients/media: We consider equations of the form or utt + ω 2 (t)u = f 0 (u, ut ) ∇2 u + k 2 (x̄)u = 0 (1) (2) The first equation describes an oscillator with slowly varying frequency, with a possible weakly nonlinear forcing. The second is the Helmholtz equation in a slowly varying media. Recall, that the Helmholtz equation is the equation for spatial dependence of time-harmonic waves ⇒ utt − c2 uxx = 0 u = eiωt v(x) vxx + k 2 v = 0 If the physical parameters vary slowly in space we get equation (2) above. It appears obvious that there are 2 scales in the problem, so one would naturally look for a multiscale solution. Consider (1), let u = u(t, τ ) = u0 (t, τ ) + u1 (t, τ ), τ = t. Then to leading order we get: u0tt + ω 2 (t)u0 = 0 u0 = A sin ωt + B cos ωt A = A(τ ), B(τ ) Then the next order contribution has secular terms. Instead, we have to use a more complicated multi-scale, let T = f (t) , | {z } fast function of time τ = t } slow time Haberman covers this method, but it is relatively inefficient. Instead, consider the equation (1) again. Let’s make the change of variables s = t, v(s) = u(t) ⇒ 2 vss + ω 2 (s)v = f (v, vs ) Looks like a singular perturbation problem – but s is not a spatial variable, we don’t have boundary conditions in s. Notice that the “s” is a slow scale; reconsidering the equation on the “t” scale we saw that the leading order equation gives us a solution A sin ωt where the frequency appears almost constant on the t scale. On the “s” scale, that solution appears to oscillate very fast 1 t or s t=O(1) s=O(1) Then, we look for solutions of the form v = Re iψ(s)/ , where we can use R ∼ R0 + R1 + . . . ψ ∼ ψ0 + ψ1 + . . . Substituting in the equation for v: # " (ψs )2 iψss 2iRs ψs 2 − R+ R eiψ + Rω 2 (s)eiψ = 0 Rss + 2 p 2 ⇒ 0(1) −ψ0s + ω 2 (s) = 0 ⇒ ψ0s = ± ω 2 (s) (so the solution should really be v = Re iψ/ + c. c.) Then, the next order equation is: 0() : ψ0s ψ1s ⇒ ψ0ss 2Rs =− ψ0s R = 0 ⇒ ψ1 = const 2Rsψ0s + ψ0ss R = 0 ⇒ log ψ0s = −2 log R ⇒ Rs √ Rs √ C2 C1 ω 2 (ŝ)dŝ ω 2 (ŝ)dŝ i −i p + v= p e e 4 4 2 2 ω (s) ω (s) What about weakly nonlinear problems? We can write the solution as iψ v = Re | {z } +v1 v0 Then the 0() equation has the form eiψ (2Rs ψs + ψss R) + 2 v1ss + ω 2 v1 = (g1 (R, Rs )) eiψ/ + n X j=2 Then 2Rs ψs + ψss R = g1 (R, Rs ) X v1 = aj eijψ/ For strongly nonlinear problems one cannot use WKB (in general). What if ω 2 (x) < 0? or if ω 2 (x) = 0 for some x = x0 . Consider equations of the form utt + h(t)u = 0 h(t) > 0 for t < t0 h(t) < 0 for t > t0 2 (gj )eijψ/ For t > t0 Rs √ Rs √ C1 C2 −h(ŝ)dŝ −h(ŝ)dŝ − p u∼ p + e e 4 4 −h(ŝ) −h(ŝ) s = t. For t < t0 Near t ∼ t0 Rs √ Rs √ D1 D2 h(ŝ)ds h(ŝ)dŝ i −i p u∼ p + e e 4 4 h(ŝ) h(ŝ) h(s) ∼ −A(s − s0 ) (s0 = t0 ) Then solve 2 uss − A(s − s0 )u = 0 γ ξ = (s − s0 ) ⇒ Rs − C1 s0 e as ξ → ∞ match to p 4 −h(s) √ γ = 2/3 uξξ − Aξu = 0 u = Ai A−1/3 ξ + Bi A−1/3 ξ −h(ŝ) dŝ Ai decays as ξ → ∞, Bi grows exponentially as ξ → ∞. −1/4 +1/3 ξ 3/2 1 ) Ai ∼ √ A1/3 ξ e−2/3(A 2 π For t > t0 u ∼ C1 √ e 4 +Aγ ξ = − C1 √ e 4 γ A ξ − 2/3 ξ s0 + R s0 2/3 R ξ √Az 0 √ A(ŝ−s0 ) dŝ C1 dz = p e 4 A2/3 ξ (A1/3 2/3 ξ) −2/3 3/2 This gives an equation for C1 (the exponents match) For t < t0 as ξ → −∞ For t < t0 , t ∼ t0 1 Ai ∼ √ p sin π 4 −A1/3 ξ 3/2 π 2 1/3 + −A ξ 3 4 3/2 3/2 D1 D2 i2/3( −A1/3 2/3 ξ ) −i2/3(−A1/3 2/3 ξ ) p u∼ p e e + 4 4 −A2/3 ξ −A2/3 ξ 3