In [3]:
using Plots, ComplexPhasePortrait, ApproxFun, SingularIntegralEquations,SpecialFunctions, OscillatoryIntegrals
gr();
Dr Sheehan Olver
s.olver@imperial.ac.uk
Website: https://github.com/dlfivefifty/M3M6LectureNotes
We can now employ the Wiener–Hopf method to solve $$ \lambda u(x) + \int_{0}^\infty K(x-t)u(t) \dt = f(x)\qqfor 0 < x < \infty. $$ For simplicity, take $\lambda = 1$ (note we can divide through).
The Wiener–Hopf method consists of the following steps:
We will demonstrate these four steps using the example $K(x) = \E^{-|x|}$ and $f(x) = \E^{-x}$.
We have $$ \widehat{K}(s) = \int_{-\infty}^0 \E^{t}\E^{-\I s t} \dt + \int_0^\infty \E^{-t} \E^{-\I s t} \dt = {1 \over 1-\I s} - {1 \over -1-\I s} = {2 \over 1+s^2} $$ Because we have exponential decay at $\pm \infty$, $\widehat{K}$ is analytic in a strip. Furthermore, $$ \widehat{f_{\rm R}}(s) = \int_0^\infty \E^{-t} \E^{-\I t s} \dt = -{1 \over -1-\I s} = {\I \over \I-s} $$ As predicted, $\widehat{\fR}(s)$ is analytic in the lower half-plane.
We thus have the RH problem: $$ \underbrace{\Phi_+(s)}_{\widehat{p_{\rm L}}(s)} - \underbrace{g(s)}_{1 + \widehat K(s)} \underbrace{\Phi_-(s)}_{\widehat{u_{\rm R}}(s)} = \underbrace{h(s)}_{-\widehat{\fR}(s)} \qqand \lim\Phi(z) = 0 $$ where $$ p(x) = \int_{-\infty}^\infty K(x-t)u_{\rm R}(t) \dt $$ is unknown, or in our case $$ \Phi_+(s) - {3 + s^2 \over 1 + s^2} \Phi_-(s) = {\I \over s - \I} \qqand \lim\Phi(z) = 0 $$
Here we confirm our calculations are correct for the Fourier transforms:
In [4]:
t = Fun(0.0 .. 40)
f = exp(-t)
fourier(f, -2.0), im/(im-2.0)
Out[4]:
In [5]:
1-(fourier(-exp(-t), -2.0) + fourier(-exp(Fun(-40 .. 0)), -2.0)), (2.0^2 + 3)/(2.0^2+1)
Out[5]:
Last lecture we employed the "guess and check" kernel factorization method to find $$ g(s) = {3 + s^2 \over 1 + s^2} = \underbrace{s + \I \sqrt{3} \over s + \I}_{\kappa_+(s)} \underbrace{s - \I \sqrt{3} \over s-\I }_{\kappa_-(s)^{-1}} $$ That is, we have the solution $$ \kappa(z) = \begin{cases} {z + \I \sqrt{3} \over z + \I} & \Im z > 0 \\ {z - \I \over z - \I \sqrt{3}} & \Im z < 0 \end{cases} $$ Always verify this solution satisfies the right conditions:
Here we confirm that we have the right jump:
In [6]:
κ = z -> imag(z) > 0 ? (z+im*sqrt(3))/(z+im) :
(z-im)/(z-im*sqrt(3))
κ(0.1+eps()im) - κ(0.1-eps()im)*(3+0.1^2)/(1+0.1^2)
Out[6]:
Writing $\Phi(z) = \kappa(z) Y(z)$ we have $$ \Phi_+(s) - g(s) \Phi_-(s) = \kappa_+(s) (Y_+(s) - Y_-(s)) $$ hence we need to solve $$ Y_+(s) - Y_-(s) = {h(s) \over \kappa_+(s)} = {\I \over s- \I} {s + \I \over s+ \I \sqrt{3}} $$ As in last lecture, we found: $$ Y(z) = \begin{cases} {-\I \over 1+\sqrt 3} {1-\sqrt 3 \over z+ \I \sqrt{3}} & \Im z > 0 \\ {-2 \I \over z-\I}{1 \over 1+ \sqrt{3}} & \Im z < 0 \end{cases} $$ Let's double check that $Y$ has the right jump:
In [7]:
Y = z -> imag(z) > 0 ? -im*(1-sqrt(3))/(1+sqrt(3))/(z+im*sqrt(3)) :
-2im/((z-im)*(1+sqrt(3)))
s = 0.1
Y(s+eps()im) - Y(s-eps()im)
Out[7]:
In [8]:
im/(s-im)*(s+im)/(s + im*sqrt(3))
Out[8]:
Thus we get $$ \begin{align*} \Phi(z) &= \kappa(z) \CC_\R\br[{h \over \kappa_+}](z) = \begin{cases} {-\I \over 1+\sqrt 3} {1-\sqrt 3 \over z+\I} & \Im z > 0 \\ {-2 \I \over 1+ \sqrt{3}} {1 \over z-\sqrt{3}\I} & \Im z < 0 \end{cases} \end{align*} $$
We can confirm it satisfies the right RHP:
In [9]:
φ = z -> imag(z) > 0 ? -im*(1-sqrt(3))/(1+sqrt(3))/(z+im) :
-2im/((z-sqrt(3)*im)*(1+sqrt(3)))
g = s -> (s^2+3)/(s^2+1)
s = 0.1
φ(s+eps()im) - φ(s-eps()im)*g(s) - im/(s-im)
Out[9]:
In particular, we have $$ \widehat{u_{\rm R}}(s) = \Phi_-(s) = \kappa_-(s) Y_-(s) = {-2 \I \over (1+\sqrt{3})(s-\I \sqrt 3)} $$ This is analytic in the lower half plane, and for $x > 0$ we can use Jordan's lemma hence use Residue calculus in the upper-half plane: $$ u(x) = {1 \over 2 \pi} \int_{-\infty}^\infty \widehat{u_{\rm R}}(s) \E^{\I s x} \ds = {-\I \over \pi(1+\sqrt{3})} \int_{-\infty}^\infty {1 \over s-\I \sqrt 3} \E^{\I s x} \ds = {2 \over (1+\sqrt{3})} \Res_{z=\I \sqrt 3} {1 \over z-\I \sqrt 3} \E^{\I z x} = {2 \E^{-\sqrt{3}x} \over 1+\sqrt{3}} $$
Did this work? Amazingly, yes (and only took me four tries to get the sign right! 🕺): \begin{align*} u(x) + \int_0^\infty K(t -x) u(t) \dt &= {2 \over 1+\sqrt{3}} \left( \E^{-\sqrt{3}x} + \int_0^x \E^{t-x} \E^{-\sqrt{3}t} \dt + \int_x^\infty \E^{x-t} \E^{-\sqrt{3}t}\dt \right) \\ &={2 \over 1+\sqrt{3}}\left( \E^{-\sqrt{3}x} + \E^{-x} \int_0^x \E^{(1-\sqrt{3})t} \dt + \E^x \int_x^\infty \E^{-(1+\sqrt{3})t}\dt \right) \\ &={2 \over 1+\sqrt{3}}\left( \E^{-\sqrt{3}x} + {1 \over 1-\sqrt 3} (\E^{-\sqrt{3}x} - \E^{-x}) + { \E^{-\sqrt 3 x } \over 1+\sqrt{3}} \right) \\ &=\E^{-x} + {2 \E^{-\sqrt{3}x} \over 1+\sqrt{3}}\left({-2 \over -2} + {1+ \sqrt 3 \over -2} + {1 - \sqrt 3 \over -2} \right) \\ &= \E^{-x} \end{align*}
We can verify it numerically:
In [10]:
t = Fun(0 .. 10)
u = 2exp(-sqrt(3)*t)/(1+sqrt(3))
x = 0.1
u(x) + sum(exp(-abs(t-x))*u) - exp(-x)
Out[10]: