18.303 Midterm Solutions, Fall 2015 November 12, 2015 Problem 1: Adjoint (33 points) ´ As in class, let’s define the inner product as hu, vi = Ω ūv/c, where the division by c(x) will cancel the asymmetrical-looking c factor in Â. Then we have: ˆ ˆ ˛ hu, Âvi = − ū∇2 v = + ∇u · ∇v − ū∇v · n̂da ∂Ω ˛Ω ˆΩ c∇2 uv/c + [v∇ū − ū∇v] · n̂da = − ∂Ω Ω ˛ · = hÂu, vi + [v∇ū − ū∇v] n̂da ∂Ω where in the last line we have used the boundary conditions to cancel the integrand: [v∇ū − ū∇v] · n̂ = α [vū − ūv] = 0. Problem 2: Finite differences (34 points) Since we have a boundary condition u0 (L) = αu(L) that requires us to know the slope at x = L, the most convenient thing is to put x = L halfway between two grid points so that we can use a center-difference approximation. In particular, we set up the grid so that um ≈ u(m∆x) as usual, but (M + 0.5)∆x = L . Then we have the discretized boundary conditions: u0 = 0 uM +1 − uM uM + uM +1 =α , ∆x 2 in which the left-hand side is the 2nd-order accurate center-difference approximation for u0 (L) and the righthand side is the 2nd-order accurate approximation for αu(L). (The 2nd-order accuracy of both of these was shown in class.) We have the M unknowns u1 , . . . , uM , and we can solve the second equation above for uM +1 = 1 + α∆x/2 uM = βuM , 1 − α∆x/2 defining the number β. If we use the usual center-difference approximation for the second derivative, we have −u00m = −um−1 + 2um − um+1 , ∆x2 which for the specific cases of m = 1 and m = M (the first and last matrix rows) gives (from the boundary conditions): 2um − um+1 −u001 = , ∆x2 −um−1 + (2 − β)um −u00M = . ∆x2 1 Hence, we have the matris −1 2 .. . 2 −1 1 A= ∆x2 −1 .. . −1 .. . 2 −1 −1 2−β . Since this matrix is obviously real-symmetric, we have self-adjointness under the usual inner product hu, vi = u∗ v. Problem 3: Green (33 points) We want to solve the nonlinear equation Âu + αu2 = f for some small |α(x)| 1. We begin by writing write u = u0 + u1 + u2 + · · · as a power series in α, where un denotes all the terms proportional to αn . If we plug this in to our equation, we get Âu0 + Âu1 + Âu2 + · · · + αu20 + 2αu0 u1 + αu21 + · · · = f . If we collect terms by power of α, we get the equations: Âu0 = f, Âu1 + αu20 = 0, Âu2 + 2αu0 u1 = 0, and so on. From the first equation, we get u0 = Â−1 f , i.e. ˆ u0 (x) = G0 (x, x0 )f (x0 ) d3 x0 . Ω 1. We want u ≈ u0 + u1 . From the second equation above, we get u1 = −Â−1 αu20 , i.e. ˆ u1 (x) = − G0 (x, x0 )α(x0 )u0 (x0 )2 d3 x0 . Ω Note that if we substitute the u0 integral, we would have two nested integrals. 2. We want u ≈ u0 + u1 + u2 . From the third equation above, we get u2 = −2Â−1 αu0 u1 , i.e. ˆ u2 (x) = −2 G0 (x, x0 )α(x0 )u0 (x0 )u1 (x0 ) d3 x0 . Ω Note that if we substituted the u0 and u1 integrals, we would have a nested integral for u0 and another two nested integrals for u1 . It would be an amusing exercise to devise a Feynman-diagram notation for these integrals. Graphically, the nonlinear term αu2 appears as a source term that arises when two Green’s functions “collide”. 2