Homework 06
1 Green’s function
Consider the equation
\[ x^{\prime \prime}(t)+2 \gamma x(t)+\omega_0^2 x(t)=f(t) . \]
Using the Green’s function for this equation, which satisfies
\[ G^{\prime \prime}(t)+2 \gamma G^{\prime}(t)+\omega_0^2 G(t)=\delta(t) \]
derive the response \(x(t)\) to a square pulse
\[ f(t)=\left\{\begin{array}{ll} 1 & 0 \leq t \leq 1 \\ 0 & \text { otherwise } \end{array} .\right. \]
Do this by solving for \(\tilde{G}(\omega)\) in the Fourier domain and note
\[ \tilde{x}(\omega)=\tilde{G}(\omega) \tilde{f}(\omega) \]
(Note that when inverting this Fourier transform it is important to treat this as a distribution.) Check that \(x(t)\) is causal and check that its Fourier transform \(\tilde{x}(\omega)\) satisfies the Kramers-Kronig relations.
The differential equation for \(G(t)\) in the Fourier domain is
\[ (-\omega^2 + 2i\gamma\omega + \omega_0^2) \tilde{G}(\omega) = 1 \]
So
\[ \tilde{G}(\omega) = \frac{1}{-\omega^2 + 2i \gamma \omega + \omega_0^2} \]
Its Fourier transform \(\tilde{f}(\omega)\) is given by:
\[ \tilde{f}(\omega) = \int_{-\infty}^{\infty} f(t) e^{-i\omega t} \, dt = \int_{0}^{1} e^{-i\omega t} \, dt \]
\[ \tilde{f}(\omega) = \left. \frac{-1}{i\omega} e^{-i\omega t} \right|_0^1 = \frac{1 - e^{-i\omega}}{i\omega} \]
Now, we use the relation \(\tilde{x}(\omega) = \tilde{G}(\omega) \tilde{f}(\omega)\):
\[ \tilde{x}(\omega) = \frac{1}{-\omega^2 + 2i\gamma\omega + \omega_0^2} \cdot \frac{1 - e^{-i\omega}}{i\omega} \]
To find \(x(t)\), we compute the inverse Fourier transform of \(\tilde{x}(\omega)\).
\[ x(t) = \frac{1}{2\pi} \int_{-\infty}^{\infty} \tilde{x}(\omega) e^{i\omega t} \, d\omega \]
We use contour integration in the complex plane to calculate this.
\(\tilde{x}(\omega)\) has poles at \(p_1 = iγ+\sqrt{\omega_0^2-\gamma^2}, p_2=iγ-\sqrt{\omega_0^2-\gamma^2}\)
The residue at \(p_1\) is
\[ \frac{ e^{-\gamma + (\gamma + \sqrt{\gamma^2 - \omega_0^2}) (1 - t)} (e^\gamma - e^{ \sqrt{\gamma^2 - \omega_0^2}}) } {2 \sqrt{-\gamma^2 + \omega_0^2} (-\gamma +\sqrt{\gamma^2-\omega_0^2})} \]
…
We can see that \(x(t) = 0\) for \(t<0\), which means \(x(t)\) is causal.
2 Different drive function
Solve the equation in problem 1 with the drive
\[ f(t)=\left\{\begin{array}{ll} e^{-t} & t \geq 0 \\ 0 & t<0 \end{array} .\right. \]
The solution can be expressed as the convolution of the Green’s function with the driving force \(f(t)\):
\[ x(t) = \int_{-\infty}^{\infty} G(t - \tau) f(\tau) d\tau \]
The form of \(G(t)\) depends on the values of \(\gamma\) and \(\omega_0\). For simplicity, let’s assume \(\gamma > 0\) and \(\omega_0 > 0\). The exact form of \(G(t)\) depends on whether the system is underdamped, overdamped, or critically damped. Without loss of generality, let’s assume an underdamped system, where \(\gamma^2 < \omega_0^2\), which gives a Green’s function of the form:
\[ G(t) = \Theta(t) \frac{e^{-\gamma t}}{\omega_d} \sin(\omega_d t) \]
where \(\omega_d = \sqrt{\omega_0^2 - \gamma^2}\) and \(\Theta(t)\) is the Heaviside step function.
Now, we compute the convolution integral:
\[ x(t) = \int_{-\infty}^{\infty} G(t - \tau) f(\tau) d\tau \]
Since \(f(\tau) = 0\) for \(\tau < 0\), the integral simplifies to:
\[ x(t) = \int_{0}^{t} G(t - \tau) e^{-\tau} d\tau \]
Substitute the expression for \(G(t - \tau)\) and carry out the integration:
\[ x(t) = \int_{0}^{t} \Theta(t - \tau) \frac{e^{-\gamma (t - \tau)}}{\omega_d} \sin(\omega_d (t - \tau)) e^{-\tau} d\tau \]
3 Green’s function and boundary conditions
Find the Green’s function \(G(x)\) satisfying
\[ \frac{d^2}{d x^2} G(x, y)=\delta(x-y) \]
with the boundary conditions \(G(0, y)=G(1, y)=0\). Show that with this boundary condition,
\[ \phi(x)=\int_0^1 G(x, y) f(y) d y \]
satisfies
\[ \frac{d^2}{d x^2} \phi(x)=f(x) \]
and the boundary conditions \(\phi(0)=\phi(1)=0\).
We consider two cases, \(x < y\) and \(x > y\), because the delta function \(\delta(x-y)\) changes the behavior of the solution at \(x = y\). We define \(G(x, y)\) piecewise for these two cases:
- For \(x < y\), let \(G(x, y) = A(y)x + C(y)\).
- For \(x > y\), let \(G(x, y) = B(y)(1 - x) + D(y)\).
Applying the boundary conditions
- \(G(0, y) = 0\) gives \(C(y) = 0\)
- \(G(1, y) = 0\) gives \(D(y) = 0\)
The function \(G(x, y)\) itself must be continuous at \(x = y\), this gives us:
\[ A(y)y = B(y)(1 - y) \]
The derivative \(\frac{d}{dx} G(x, y)\) should have a discontinuity of 1 at \(x = y\) (this comes from the delta function). Therefore, the jump at \(x = y\) is
\[ A(y) - (-B(y)) = 1 \]
The solutions for \(A(y)\) and \(B(y)\) are:
\[ A(y)=1−y, B(y)=y \]
With these, the Green’s function \(G(x, y)\) for \(x < y\) and \(x > y\) can be fully specified:
- For \(x < y\), \(G(x, y) = A(y)x = (1 - y)x\).
- For \(x > y\), \(G(x, y) = B(y)(1 - x) = y(1 - x)\).
Proving \(\phi(x)\) Satisfies the Given Conditions:
We differentiate \(\phi(x)\) twice with respect to \(x\):
\[ \frac{d^2}{dx^2} \phi(x) = \frac{d^2}{dx^2} \int_0^1 G(x, y) f(y) dy \]
Because \(G(x, y)\) is a Green’s function, its second derivative with respect to \(x\) is \(\delta(x-y)\). Therefore, the integral becomes:
\[ \frac{d^2}{dx^2} \phi(x) = \int_0^1 \delta(x-y) f(y) dy \]
The delta function picks out the value of \(f(y)\) at \(y = x\), so:
\[ \frac{d^2}{dx^2} \phi(x) = f(x) \]
- Since \(G(0, y) = 0\), it follows that \(\phi(0) = \int_0^1 G(0, y) f(y) dy = 0\).
- Similarly, since \(G(1, y) = 0\), it follows that \(\phi(1) = \int_0^1 G(1, y) f(y) dy = 0\).
Therefore, \(\phi(x)\) satisfies the differential equation with the boundary conditions \(\phi(0) = \phi(1) = 0\).
4 Nyquist–Shannon sampling theorem
Suppose \(f(t)\) is band-limited, meaning its Fourier transform satisfies \(\tilde{f}(\omega)=0\) for \(|\omega| \geq 2 \pi \Lambda\). If \(T<1 /(2 \Lambda)\), we showed it is possible to reconstruct \(f(t)\) from the set of sample values \(f(n T)\), where \(n \in \mathbb{Z}\). Give an explicit formula for \(f(t)\) in terms of the sample values.
When \(x(t)\) is a function with a Fourier transform \(X(f)\) :
\[ X(f) \triangleq \int_{-\infty}^{\infty} x(t) e^{-i 2 \pi f t} \mathrm{~d} t \]
the Poisson summation formula indicates that the samples, \(x(n T)\), of \(x(t)\) are sufficient to create a periodic summation of \(X(f)\). The result is:
\[ X_s(f) \triangleq \sum_{k=-\infty}^{\infty} X\left(f-k f_s\right)=\sum_{n=-\infty}^{\infty} T \cdot x(n T) e^{-i 2 \pi n T f} \]
When there is no overlap of the copies (also known as “images”) of \(X(f)\), the \(k=0\) term of Eq. 1 can be recovered by the product:
\[ X(f)=H(f) \cdot X_s(f) \]
where:
\[ H(f) \triangleq \begin{cases}1 & |f|<B \\ 0 & |f|>f_s-B\end{cases} \]
The sampling theorem is proved since \(X(f)\) uniquely determines \(x(t)\). All that remains is to derive the formula for reconstruction. \(H(f)\) need not be precisely defined in the region \(\left[B, f_s-B\right]\) because \(X_s(f)\) is zero in that region. However, the worst case is when \(B=f_s / 2\), the Nyquist frequency. A function that is sufficient for that and all less severe cases is:
\[ H(f)=\operatorname{rect}\left(\frac{f}{f_s}\right)= \begin{cases}1 & |f|<\frac{f_s}{2} \\ 0 & |f|>\frac{f_s}{2}\end{cases} \]
where rect is the rectangular function. Therefore:
\[ \begin{aligned} X(f) & =\operatorname{rect}\left(\frac{f}{f_s}\right) \cdot X_s(f) \\ & =\operatorname{rect}(T f) \cdot \sum_{n=-\infty}^{\infty} T \cdot x(n T) e^{-i 2 \pi n T f} \\ & =\sum_{n=-\infty}^{\infty} x(n T) \cdot \underbrace{T \cdot \operatorname{rect}(T f) \cdot e^{-i 2 \pi n T f}}_{\mathcal{F}\left\{\operatorname{sinc}\left(\frac{t-n T}{T}\right)\right\}} \end{aligned} \]
The inverse transform of both sides produces the Whittaker-Shannon interpolation formula:
\[ x(t)=\sum_{n=-\infty}^{\infty} x(n T) \cdot \operatorname{sinc}\left(\frac{t-n T}{T}\right) \]
5 Jacobi theta function
Consider the Jacobi theta function
\[ \vartheta_3(\tau)=\sum_{n=-\infty}^{\infty} e^{i \pi \tau n^2} \]
Show by Poisson summation that
\[ \vartheta_3(-1 / \tau)=\sqrt{i \tau} \vartheta_3(\tau) \]
Consider the function \(f(x) = e^{i \pi \tau x^2}\). The Fourier transform of \(f(x)\) is given by:
\[ \hat{f}(\xi) = \int_{-\infty}^{\infty} e^{i \pi \tau x^2} e^{-2 \pi i x \xi} dx \]
Applying the Poisson summation formula, we have:
\[ \sum_{n=-\infty}^{\infty} e^{i \pi \tau n^2} = \sum_{m=-\infty}^{\infty} \hat{f}(m) \]
Substituting \(\hat{f}(m)\) we get:
\[ \vartheta_3(\tau) = \sqrt{\frac{i}{\tau}} \sum_{m=-\infty}^{\infty} e^{-\pi i m^2 / \tau} \]
Change of Variable: Let \(\tau' = -1/\tau\). Then \(e^{-\pi i m^2 / \tau} = e^{i \pi \tau' m^2}\). Therefore, the sum becomes:
\[ \vartheta_3(\tau) = \sqrt{\frac{i}{\tau}} \sum_{m=-\infty}^{\infty} e^{i \pi \tau' m^2} \]
Recognizing the sum as \(\vartheta_3(\tau')\) with \(\tau' = -1/\tau\), we get:
\[ \vartheta_3(\tau) = \sqrt{\frac{i}{\tau}} \vartheta_3\left(-\frac{1}{\tau}\right) \]
or equivalently:
\[ \vartheta_3\left(-\frac{1}{\tau}\right) = \sqrt{i\tau} \vartheta_3(\tau) \]
6 Jacobi theta function by integral
Show another way by considering the integral
\[ I_N(\tau)=\int_{\gamma_N} \frac{e^{-\pi i z^2 \tau}}{e^{2 \pi i z}-1} d z \]
where \(\gamma_N\) is a rectangular contour with vertices \(\pm(N+1 / 2) \pm i\). Compute this by residues and take the limit \(N \rightarrow \infty\) to show it equals \(\vartheta_3(\tau)\). Then argue we can rewrite the contour integral as
\[ \vartheta_3(\tau)=\int_{-\infty-i \epsilon}^{\infty-i \epsilon} \frac{e^{-\pi i z^2 \tau}}{e^{2 \pi i z}-1} d z-\int_{-\infty+i \epsilon}^{\infty+i \epsilon} \frac{e^{-\pi i z^2 \tau}}{e^{2 \pi i z}-1} d z \]
and do geometric series expansions and direct Gaussian integration to show this is also \(\frac{1}{\sqrt{i \tau}} \vartheta_3(-1 / \tau)\).
Compute by Residues: Inside the contour \(\gamma_N\), the poles of the integrand occur at integers. The residue at each pole \(n\) (an integer) is given by evaluating the limit as \(z \to n\) of \((z - n)\) times the integrand. This leads to residues of the form \(e^{-\pi i n^2 \tau}\).
Summing the residues for all poles within the contour, we get:
\[ \sum_{n=-N}^N e^{-\pi i n^2 \tau} \]
This is a finite sum approximation of the theta function \(\vartheta_3(\tau)\). As \(N \to \infty\), this sum becomes \(\vartheta_3(\tau)\). So, we have:
\[ \lim_{N \to \infty} I_N(\tau) = \vartheta_3(\tau) \]
The contour integral can be rewritten as:
\[ \vartheta_3(\tau) = \int_{-\infty-i\epsilon}^{\infty-i\epsilon} \frac{e^{-\pi i z^2 \tau}}{e^{2 \pi i z} - 1} dz - \int_{-\infty+i\epsilon}^{\infty+i\epsilon} \frac{e^{-\pi i z^2 \tau}}{e^{2 \pi i z} - 1} dz \]
Here, \(\epsilon\) is a small positive number. The integrand can be expanded using the geometric series for \(e^{2 \pi i z}\). This series converges absolutely for \(|\text{Im}(z)| > 0\). After the series expansion, the integrals involve terms of the form \(e^{-\pi i z^2 \tau}\) which are Gaussian integrals. These integrals can be evaluated directly, and due to the symmetry of the Gaussian, they will involve a factor of \(\frac{1}{\sqrt{i \tau}}\). By summing these integrals, we obtain the identity:
\[ \vartheta_3(\tau) = \frac{1}{\sqrt{i \tau}} \vartheta_3\left(-\frac{1}{\tau}\right) \]