1

I wish to obtain the eigenfunctions/eigenvalues of $\nabla^2 \phi(x,y) +\lambda \phi(x,y) = 0$ with boundary conditions $\phi(x,0) = \phi(x,\pi) = \phi(0,y) = \phi(\pi,y) = 0$. To do this, I consider the method of Separation of Variables.

Ansatz: $\phi = X(x)Y(y)$. Then substituting appropriately I obtain: $$ \frac{X''}{X} = -\frac{(Y'' + \lambda Y)}{Y} = -\mu $$ which yields the two equations $$\begin{cases}X'' + \mu X &= 0 \\ Y'' + (\lambda - \mu)Y &=0\end{cases}$$ and corresponding boundary conditions: $$X(x)Y(0) = X(x)Y(\pi) = X(0)Y(y) = X(\pi)Y(y) = 0$$

Questions:

  1. In the equation for $X$ the variable $-\mu$ is called the eigenvalue of the $X$ equation because we can write $DX = -\mu X$ where $D := \partial_x^2$ and so $-\mu$ is an eigenvalue of this equation. For the $Y$ equation the eigenvalue is $-(\lambda - \mu )$. If I find the values for $\lambda, \mu$ then...what does this mean for the original problem. Is the product of the eigenfunctions of the respective equations again an eigenfunction for the original problem?
  2. I have issue actually solving this system of equations with the boundary conditions. From my observation it does not seem possible to solve for the four coefficients the result from solving the equations. See below:

Solving the $X$ equation I obtain: $X = c_1 \cos(\sqrt(\mu)x)+c_2\sin(\sqrt{\mu}x)$

Solving the $Y$ equation I obtain $Y = c_3\cos(\sqrt{(\lambda - \mu)}y) + c_4\sin(\sqrt{(\lambda-\mu)}y)$


I obtain $c_1 = c_3 = 0$. But now I have the equations $$\begin{cases}c_2\sin(\sqrt{\mu}x)c_4\sin(\sqrt{(\lambda-\mu)}\pi) &=0 \\ c_2\sin(\sqrt{\mu}\pi)c_4 \sin(\sqrt{(\lambda-\mu)}y) &= 0\end{cases}$$

hirotaFan
  • 301
  • 1
    If $X(x)Y(0) = 0$ for every $x$, clearly you must have $Y(0) = 0$. Can you go from here? – Danilo Gregorin Afonso Mar 30 '21 at 19:26
  • @DaniloGregorinAfonso Is that because we considering only nontrivial $X(x)$ and $Y(y)$? Otherwise if from $X(x) = 0$ then the ansatz is $0$ which is the trivial solution and is uninteresting. Therefore, we consider $Y(0) = 0 \implies c_3 = 0$.

    Correct?

    – hirotaFan Mar 30 '21 at 19:33
  • Yes, that's correct. – Andrew McMillan Mar 30 '21 at 19:37
  • My question $(1)$ regarding the interpretation of eigenfunction/eigenvalue of the original PDE remains however. I will work on obtaining the coefficients $c_1,c_2, c_3, c_4$ now. Indeed, by the same argument we obtain $c_1 = 0$. – hirotaFan Mar 30 '21 at 19:40
  • It is odd that your solution for $Y$ has no square roots in it... – Eric Towers Mar 30 '21 at 20:55
  • @EricTowers Good catch! I for some reason just didn't write the square-root sign and then continued the error throughout my work. – hirotaFan Mar 30 '21 at 21:06

1 Answers1

1

This method of jamming your separated variables together is not a good way to go. Instead, you should recognize, since each of these holds for all $(x,y) \in [0,\pi]^2$, \begin{align*} X(x)Y(0) &= 0 &&\implies& Y(0) &= 0 \text{,} \\ X(x)Y(\pi) &= 0 &&\implies& Y(\pi) &= 0 \text{,} \\ X(0)Y(y) &= 0 &&\implies& X(0) &= 0 \text{, and} \\ X(\pi)Y(y) &= 0 &&\implies& X(\pi) &= 0 \text{.} \end{align*} That is, you've separated the variables, which is predicated on being able to separate the boundary/initial conditions...


(This answer is written for the version of the two conditions at the end of the Question as it was when the answer was written. I hope that some square roots have appeared since then.)

Nevertheless, we can still answer your question.

Let's look at your first equation, which is to hold for all $x \in [0,\pi]$: $$ c_2 \sin(\sqrt{\mu}x) c_4 \sin((\lambda + \mu)\pi) = 0 \text{.} $$ A product is zero if one (or more) of its multiplicands is zero. This gives four possibilities:

  • $c_2 = 0$,
  • $\sin(\sqrt{\mu}x) = 0$, which requires $\mu = 0$ (for any other choice of $\mu$, we can choose an $x$ giving a nonzero left-hand side; recall that $\lambda$ and $\mu$ are constants, so do not depend on $x$ or $y$),
  • $c_4 = 0$, or
  • $\sin((\lambda + \mu)\pi) = 0$, requiring $\lambda + \mu \in \Bbb{Z}$.

Your second equation gives the options

  • $c_2 = 0$,
  • $\sqrt{\mu} \in \Bbb{Z}$,
  • $c_4 = 0$, or
  • $\lambda + \mu = 0$.

The intersection of these two lists of options (the conjuction of these two disjunctions followed by some simplification) gives the options

  • $\mu = \lambda = 0$,
  • $\sqrt{\mu}, \lambda + \mu \in \Bbb{Z}$, that is, $\mu$ is the square of an integer and $\lambda$ is an integer,
  • $c_2 = 0$,
  • $c_4 = 0$,
  • $\lambda = -\mu$, or
  • $\mu = 0$.

We now know that every solution (compatible with separated variables) satisfies one or more of those six constraints.

Eric Towers
  • 67,037
  • Going further, I have noted that $\lambda - \mu \neq 0$ and $\mu \neq 0$ due to the given boundary conditions. From this, I obtain the system $\sin(\sqrt{\mu}\pi) = \sin((\lambda + \mu)\pi) = 0$ yielding the families for $\mu$ and $\lambda - \mu$. – hirotaFan Mar 30 '21 at 21:13
  • Question: The families for $\mu$ and $\lambda - \mu$ are: $\begin{cases}\mu &= (\frac{1}{2}+2k_1)^2 \ (\lambda - \mu) &= (\frac{1}{2} + 2k_2)^2\end{cases}$ where $k_1,k_2 \in \mathbb{Z}$. The product of our eigenfunctions, due to our ansatz, are eigenfunctions for the original problem, right? And so the same for the eigenvalues? So then:

    $(\frac{1}{2} + 2k_1)^2 \cdot (\frac{1}{2}+2k_2)^2$ are the eigenvalues of our original problem?

    – hirotaFan Mar 30 '21 at 21:43
  • @hirotaFan : I don't see an update to your Question with corrected equations, but I have on objection to your comments for the version of the equations that is there: $\lambda = \mu = 0$ is certainly a solution satisfying the boundary conditions, but is incompatible with the conditions in your comments. Also, the question establishes that eigenvalues are values of $\lambda$. There may be multiple eigenvectors (differing in choice of $\mu$, or differing in choice of subset of conditions following from the boundary conditions) with the same eigenvalue. – Eric Towers Mar 30 '21 at 22:16
  • Then I have a fundamental misunderstanding on how eigenfunctions/eigenvalues work for Separation of Variables. I am going to post a new question as this will become an extended discussion. – hirotaFan Mar 30 '21 at 22:23
  • @hirotaFan : There are two layers of eigensystem here. The first layer is the eigensystem of the Laplacian -- (negative) $\lambda$ is given for the eigenvalues of this operator and solutions are eigenfunctions of this operator. When you apply separation of variables, now you have two functions, so two operators and a new eigensystem for these two functions. The eigenvalues/-functions of the two systems are (1) distinct and (2) very, very related. – Eric Towers Mar 30 '21 at 22:28
  • How exactly are the eigenfunctions/eigenvalues of the (2) system related to the eigenvalues/eigenfunctions of the original system (1) ? By the ansatz, the solution to the original system (1) is the product of the eigenfunctions of system (2). And so shouldn't the eigenvalues of the original system (1) also be the product of the eigenvalues of (2) ? – hirotaFan Mar 30 '21 at 22:37
  • 1
    @hirotaFan : In system (2), the eigenvalues are $(\lambda, \mu)$ pairs and the eigenfunctions are $(X, Y)$ pairs. The given equation's system can only depend on $\lambda$ -- the given equation is the eigenvalue equation for the Laplacian operator with eigenvalue $\lambda$. So in system (1), the eigenvalues are $\lambda$s and the eigenfunctions are $\phi$s. So, from $((\lambda, \mu), (X,Y))$ in system (2), we obtain $(\lambda, \phi) = (\lambda, XY)$ in system (1). – Eric Towers Mar 30 '21 at 22:42