Differential Equations › Section H · Boundary value problems and Fourier series › Topic H.2
Eigenvalues and eigenfunctions
A homogeneous boundary value problem usually has only the zero solution, but for certain special values of a parameter, nonzero solutions appear. Those values and solutions describe the notes a string can play, the loads at which a column buckles, and the building blocks of every Fourier series.
A parameter that decides everything
In "Boundary value problems", \(y'' + 9y = 0\) with \(y(0) = y(\pi) = 0\) had nonzero solutions, while \(y'' + 5y = 0\) with the same conditions did not. So put a parameter in the equation,
and ask: for which \(\lambda\) does the homogeneous BVP have a nonzero solution? This mirrors linear algebra, where \(A\mathbf{x} = \lambda\mathbf{x}\) has nonzero solutions only for special \(\lambda\). There, \(A\) acts on vectors; here, the operation \(y \mapsto -y''\) acts on functions that satisfy the boundary conditions.
Eigenvalues and eigenfunctions defined
A number \(\lambda\) is an eigenvalue of the BVP if the problem has a nonzero solution for that \(\lambda\); each such nonzero solution is an eigenfunction for \(\lambda\). The trivial solution \(y = 0\) never counts. Because the problem is homogeneous and linear, any nonzero constant multiple of an eigenfunction is again an eigenfunction, so we list one representative such as \(\sin 2x\) and understand that \(5\sin 2x\) is just as good. Finding all eigenvalues means proving that every other \(\lambda\) gives only the trivial solution.
Three cases for lambda
The characteristic equation \(r^2 + \lambda = 0\) has roots \(r = \pm\sqrt{-\lambda}\), so the form of the solution depends on the sign of \(\lambda\), and each sign must be checked. If \(\lambda > 0\), write \(\lambda = \mu^2\) with \(\mu > 0\); then \(y = c_1\cos\mu x + c_2\sin\mu x\). If \(\lambda = 0\), the equation is \(y'' = 0\) and \(y = c_1 + c_2x\). If \(\lambda < 0\), write \(\lambda = -\mu^2\); then \(y = c_1\cosh\mu x + c_2\sinh\mu x\). The hyperbolic form is the same as using \(e^{\mu x}\) and \(e^{-\mu x}\), but it makes conditions at \(x = 0\) easier, since \(\cosh 0 = 1\) and \(\sinh 0 = 0\).
Worked example: both ends fixed
Find the eigenvalues of \(y'' + \lambda y = 0\), \(y(0) = 0\), \(y(L) = 0\). 1. \(\lambda < 0\): \(y(0) = 0\) gives \(c_1 = 0\), then \(c_2\sinh\mu L = 0\); since \(\sinh\mu L > 0\), \(c_2 = 0\). No eigenvalues. 2. \(\lambda = 0\): \(c_1 = 0\) and \(c_2L = 0\), so again only \(y = 0\). 3. \(\lambda > 0\): \(c_1 = 0\) and \(c_2\sin\mu L = 0\). To keep \(c_2 \ne 0\) we need \(\sin\mu L = 0\), so \(\mu L = n\pi\). 4. Result:
The value \(n = 0\) is excluded because \(\sin 0 = 0\) is the trivial solution.
Worked example: one end fixed, one end free
Now take \(y(0) = 0\) and \(y'(L) = 0\). 1. \(\lambda < 0\): \(c_1 = 0\), then \(y'(L) = c_2\mu\cosh\mu L = 0\), and \(\cosh\) is never zero, so \(c_2 = 0\). 2. \(\lambda = 0\): \(c_1 = 0\) and \(y' = c_2 = 0\). 3. \(\lambda > 0\): \(c_1 = 0\) and \(c_2\mu\cos\mu L = 0\), so \(\cos\mu L = 0\) and \(\mu L\) is an odd multiple of \(\frac{\pi}{2}\). 4. Result:
for \(n = 1, 2, 3, \dots\) With \(L = 1\), the first two eigenvalues are \(\frac{\pi^2}{4} \approx 2.47\) and \(\frac{9\pi^2}{4} \approx 22.2\).
Both ends insulated
With slope conditions \(y'(0) = 0\) and \(y'(L) = 0\), zero becomes an eigenvalue. For \(\lambda = 0\), \(y = c_1 + c_2x\) has \(y' = c_2\), so \(c_2 = 0\) while \(c_1\) stays free: the constant \(y = 1\) is an eigenfunction. For \(\lambda > 0\), \(y'(0) = 0\) removes the sine, and \(y'(L) = 0\) needs \(\sin\mu L = 0\). For \(\lambda < 0\), only the trivial solution survives. Together: \(\lambda_n = (\frac{n\pi}{L})^2\) with \(y_n = \cos\frac{n\pi x}{L}\) for \(n = 0, 1, 2, \dots\), where \(n = 0\) gives the constant. Physically, an insulated rod can settle at any uniform temperature.
Periodic boundary conditions
On \(-L \le x \le L\), the periodic conditions \(y(-L) = y(L)\) and \(y'(-L) = y'(L)\) say the solution joins up smoothly with itself, as on a ring. For \(\lambda = \mu^2 > 0\), the two conditions become \(2c_2\sin\mu L = 0\) and \(2c_1\mu\sin\mu L = 0\). When \(\sin\mu L = 0\), both constants are free. So \(\lambda_n = (\frac{n\pi}{L})^2\) for \(n \ge 1\) has two independent eigenfunctions, \(\cos\frac{n\pi x}{L}\) and \(\sin\frac{n\pi x}{L}\). The case \(\lambda = 0\) gives the constant 1, and \(\lambda < 0\) gives nothing. This list of sines, cosines and a constant is exactly what a full Fourier series is built from.
Sturm-Liouville problems
Each example fits a wider pattern. A regular Sturm-Liouville problem has the form
on \([a, b]\), with continuous coefficients, \(p > 0\) and \(w > 0\), and one separate condition at each end that may combine \(y\) and \(y'\). The theory guarantees infinitely many real eigenvalues \(\lambda_1 < \lambda_2 < \cdots\) that grow without bound, each with one independent eigenfunction, and eigenfunctions that are orthogonal with weight \(w\). Periodic conditions link the two ends, so they are not of this kind, which is why they allowed two eigenfunctions per eigenvalue. Negative eigenvalues can occur: \(y'' + \lambda y = 0\) with \(y'(0) = y(0)\) and \(y'(1) = y(1)\) has \(\lambda = -1\), with eigenfunction \(e^x\). Checking all three cases is never optional.
Applied example: a buckling column
A slender, straight, elastic column of length \(L\) with pinned ends carries a compressive load \(P\) along its axis. For small sideways deflections \(y(x)\), beam theory gives \(EIy'' + Py = 0\) with \(y(0) = y(L) = 0\), where \(EI\) measures stiffness. This is the fixed-ends problem with \(\lambda = \frac{P}{EI}\), so a bent shape exists only when \(\frac{P}{EI} = (\frac{n\pi}{L})^2\). The smallest is Euler's critical load
with buckled shape \(\sin\frac{\pi x}{L}\). Doubling the length cuts the critical load to a quarter. A string fixed at both ends obeys the same eigenvalue problem, which is why its modes are sines, as "The vibrating string" will show.
Common mistakes to avoid
First, checking only \(\lambda > 0\); the cases \(\lambda = 0\) and \(\lambda < 0\) must be ruled in or out. Second, listing \(n = 0\) for fixed ends, where it gives \(y = 0\), or forgetting it for insulated ends, where it gives the constant eigenfunction. Third, writing \(\sqrt{\lambda}\) as a real number when \(\lambda\) is negative. Fourth, treating \(\sinh\mu L = 0\) as possible for \(\mu > 0\); it is not. Fifth, reporting \(\lambda_n = \frac{n\pi}{L}\) instead of its square.
Key takeaways
An eigenvalue is a value of \(\lambda\) for which a homogeneous BVP has a nonzero solution, its eigenfunction. For \(y'' + \lambda y = 0\), check \(\lambda > 0\), \(\lambda = 0\) and \(\lambda < 0\) separately. Fixed ends give \((\frac{n\pi}{L})^2\) with sines for \(n \ge 1\); insulated ends give the same values with cosines for \(n \ge 0\); periodic conditions give both, plus the constant; one fixed and one free end give \((\frac{(2n-1)\pi}{2L})^2\). Euler's buckling load \(\frac{\pi^2EI}{L^2}\) is the smallest eigenvalue in physical form.
Flashcards
Flashcard
A value of λ for which the homogeneous BVP has a nonzero solution
Test your understanding
Question 1 of 5
For y″ + λy = 0 with y(0) = 0 and y(4) = 0, what is the smallest eigenvalue?
Question 2 of 5
For y″ + λy = 0 with y′(0) = 0 and y′(π) = 0, is λ = 0 an eigenvalue?
Question 3 of 5
For y″ + λy = 0 with y(0) = 0 and y′(π) = 0, which value is an eigenvalue?
Question 4 of 5
A pinned column's length is doubled while E and I stay the same. What happens to its Euler critical load?
Question 5 of 5
Which statement about the eigenfunctions of a homogeneous BVP is true?
