Boundary Value Problems

A boundary value problem gives conditions at two different points, such as y(0) = 0 and y(L) = 0, instead of at one starting point. Unlike an initial value problem, it can have one solution, no solution or infinitely many. When the equation has a parameter λ, the values of λ that allow nonzero solutions are its eigenvalues.

Before this lesson: Second-Order Homogeneous Linear Equations, Complex Roots of the Characteristic Equation and Existence and Uniqueness Theorem

Initial value problems vs. boundary value problems

An initial value problem fixes y and y′ at one point, like launching a ball from a known spot at a known velocity. The existence and uniqueness theorem then promises exactly one solution.

A boundary value problem fixes y (or y′) at two ends of an interval, like a string tied down at both ends. Now there is no guarantee. You still find the general solution first, but the two conditions can agree, clash, or leave a constant undecided.

How to solve boundary value problems

  1. Find the general solution, such as y = c₁ cos kx + c₂ sin kx.
  2. Substitute each boundary condition to get two equations in c₁ and c₂.
  3. Solve that system. A single solution means the problem has exactly one solution.
  4. If the conditions contradict each other (such as 0 = 2), there is no solution. If one equation reads 0 = 0, a constant can take any value and there are infinitely many solutions.

Eigenvalue problems

Many boundary value problems contain a parameter λ, as in y″ + λy = 0 with y(0) = 0 and y(L) = 0. The zero function y = 0 always solves it. The values of λ that also allow a nonzero solution are the eigenvalues, and those nonzero solutions are the eigenfunctions.

This matches eigenvalues and eigenvectors in linear algebra: the equation −y″ = λy says the operator −d²/dx² turns y into a multiple of itself. To find the eigenvalues, test three cases: λ < 0, λ = 0 and λ > 0.

y'' + \lambda y = 0,\quad y(0) = 0,\quad y(L) = 0 \quad\Longrightarrow\quad \lambda_n = \left(\frac{n\pi}{L}\right)^2,\quad y_n = \sin\frac{n\pi x}{L},\quad n = 1, 2, 3, \ldots

The table lists the three eigenvalue problems for y″ + λy = 0 on 0 ≤ x ≤ L that come up most often. Each eigenfunction can be multiplied by any nonzero constant.

Boundary conditionsEigenvaluesEigenfunctions
y(0) = 0,\ y(L) = 0\lambda_n = (n\pi/L)^2,\ n = 1, 2, 3, \ldots\sin(n\pi x/L)
y'(0) = 0,\ y'(L) = 0\lambda_n = (n\pi/L)^2,\ n = 0, 1, 2, \ldots\cos(n\pi x/L); y_0 = 1
y(0) = 0,\ y'(L) = 0\lambda_n = \left(\frac{(2n-1)\pi}{2L}\right)^2,\ n = 1, 2, 3, \ldots\sin\left(\frac{(2n-1)\pi x}{2L}\right)

Boundary value problem examples

Example 1: A boundary value problem with one solution

Solve y″ + 4y = 0 with y(0) = 1 and y(π/4) = 3.

r^2 + 4 = 0 \Rightarrow r = \pm 2i

Write the characteristic equation and solve it.

y = c_1\cos 2x + c_2\sin 2x

Use the complex roots form with α = 0 and β = 2.

y(0) = c_1\cos 0 + c_2\sin 0 = c_1

Substitute x = 0.

c_1 = 1

Use y(0) = 1.

y\left(\frac{\pi}{4}\right) = c_1\cos\frac{\pi}{2} + c_2\sin\frac{\pi}{2} = c_2

Substitute x = π/4; cos(π/2) = 0 and sin(π/2) = 1 (trig values of special angles).

c_2 = 3

Use y(π/4) = 3.

Answer: y = cos 2x + 3 sin 2x. The two conditions fixed both constants, so this is the only solution.

Example 2: No solution or infinitely many solutions

Keep y″ + 4y = 0 and y(0) = 1, but move the second condition to x = π. Solve the problem with (a) y(π) = 2 and (b) y(π) = 1.

y = c_1\cos 2x + c_2\sin 2x,\quad c_1 = 1

Same general solution, and y(0) = 1 still gives c₁ = 1.

y(\pi) = c_1\cos 2\pi + c_2\sin 2\pi = c_1

Substitute x = π; cos 2π = 1 and sin 2π = 0.

\text{(a)}\quad c_1 = 2

Use y(π) = 2.

1 = 2

Both conditions fix c₁, and they disagree.

\text{(b)}\quad c_1 = 1

Use y(π) = 1.

1 = 1

This agrees with c₁ = 1, and c₂ never appears.

Answer: (a) has no solution. (b) has infinitely many solutions, y = cos 2x + c₂ sin 2x for any c₂. The reason is that sin 2x is 0 at both x = 0 and x = π, so any multiple of it can be added without changing either boundary value.

Example 3: Eigenvalues with y(0) = 0 and y(2) = 0

Find the eigenvalues and eigenfunctions of y″ + λy = 0 with y(0) = 0 and y(2) = 0.

\lambda = -\mu^2 \lt 0:\quad y = c_1\cosh\mu x + c_2\sinh\mu x

Case 1: the roots are r = ±μ; write the solution with hyperbolic functions.

y(0) = c_1 = 0

cosh 0 = 1 and sinh 0 = 0.

y(2) = c_2\sinh 2\mu = 0 \Rightarrow c_2 = 0

sinh 2μ ≠ 0 when μ > 0, so only y = 0 works.

\lambda = 0:\quad y = c_1 + c_2x

Case 2: y″ = 0, so y is linear.

y(0) = c_1 = 0,\quad y(2) = 2c_2 = 0

Both constants are 0, so λ = 0 is not an eigenvalue.

\lambda = \mu^2 \gt 0:\quad y = c_1\cos\mu x + c_2\sin\mu x

Case 3: the roots are r = ±μi.

y(0) = c_1 = 0

Substitute x = 0.

y(2) = c_2\sin 2\mu = 0

Substitute x = 2; take c₂ ≠ 0 to get a nonzero solution.

2\mu = n\pi \Rightarrow \mu = \frac{n\pi}{2},\quad n = 1, 2, 3, \ldots

Sine is 0 at whole multiples of π (solving trig equations); μ > 0, so n ≥ 1.

\lambda_n = \mu^2 = \frac{n^2\pi^2}{4}

Square μ.

Answer: λₙ = n²π²/4 with eigenfunctions yₙ = sin(nπx/2), for n = 1, 2, 3, …. The first three eigenvalues are π²/4 ≈ 2.467, π² ≈ 9.870 and 9π²/4 ≈ 22.207. Each eigenfunction fits n half-waves between x = 0 and x = 2, as the graph shows.

Graph for x from 0 to 2 of the first three eigenfunctions. A solid blue curve y1 = sin(pi x/2) rises from 0 to a peak of 1 at x = 1 and returns to 0 at x = 2. A dashed orange curve y2 = sin(pi x) makes one hump up and one down, crossing 0 at x = 1. A dotted green curve y3 = sin(3 pi x/2) makes three half-waves, crossing 0 at x = 2/3 and x = 4/3. All three are 0 at both ends.
Example 3: y₁ = sin(πx/2) (solid blue), y₂ = sin(πx) (dashed orange) and y₃ = sin(3πx/2) (dotted green) all meet the boundary conditions y(0) = 0 and y(2) = 0.

Example 4: Eigenvalues with y′(0) = 0 and y′(π) = 0

Find the eigenvalues and eigenfunctions of y″ + λy = 0 with y′(0) = 0 and y′(π) = 0.

\lambda = -\mu^2:\quad y' = c_1\mu\sinh\mu x + c_2\mu\cosh\mu x

Case 1: differentiate y = c₁ cosh μx + c₂ sinh μx.

y'(0) = c_2\mu = 0 \Rightarrow c_2 = 0

μ > 0.

y'(\pi) = c_1\mu\sinh\mu\pi = 0 \Rightarrow c_1 = 0

sinh μπ ≠ 0, so there is no negative eigenvalue.

\lambda = 0:\quad y = c_1 + c_2x,\quad y' = c_2

Case 2: y is linear.

y'(0) = c_2 = 0

Both conditions just say c₂ = 0, and c₁ can be any number.

\lambda_0 = 0,\quad y_0 = 1

A nonzero constant works, so 0 is an eigenvalue.

\lambda = \mu^2:\quad y' = -c_1\mu\sin\mu x + c_2\mu\cos\mu x

Case 3: differentiate y = c₁ cos μx + c₂ sin μx.

y'(0) = c_2\mu = 0 \Rightarrow c_2 = 0

μ > 0.

y'(\pi) = -c_1\mu\sin\mu\pi = 0

Take c₁ ≠ 0.

\mu\pi = n\pi \Rightarrow \mu = n,\quad \lambda_n = n^2

Sine is 0 at whole multiples of π; square μ.

Answer: λₙ = n² with eigenfunctions yₙ = cos nx, for n = 0, 1, 2, …. With conditions on y′ (called Neumann conditions), λ = 0 is an eigenvalue and its eigenfunction is the constant 1.

Example 5: Buckling of a column (word problem)

A thin vertical rod 1.5 meters long is pinned at both ends and pressed from above by a load P (in newtons). Its sideways bend y(x) satisfies EIy″ + Py = 0 with y(0) = 0 and y(1.5) = 0, where the stiffness is EI = 450 N·m². The rod stays straight unless the problem has a nonzero solution. Find the smallest load that makes the rod buckle.

y'' + \frac{P}{EI}y = 0

Divide by EI; this is y″ + λy = 0 with λ = P/EI.

\lambda_n = \left(\frac{n\pi}{1.5}\right)^2

Use the eigenvalues for y(0) = 0 and y(L) = 0 with L = 1.5.

\lambda_1 = \frac{\pi^2}{2.25}

The smallest load uses n = 1.

P_1 = EI\,\lambda_1 = \frac{450\pi^2}{2.25}

Solve λ = P/EI for P.

P_1 = 200\pi^2 \approx 1973.9

Divide 450 by 2.25.

Answer: the rod first buckles at about 1974 N, and it bends into the shape y = A sin(πx/1.5), one smooth bow. This critical load is called the Euler buckling load. Any smaller load leaves only the solution y = 0, so the rod stays straight.

Common mistakes

MistakeExample of the errorFix
Expecting one solution, as for an initial value problemAssuming Example 2(a) has an answerSolve for c₁ and c₂; a contradiction means no solution.
Skipping the λ < 0 and λ = 0 casesChecking only λ > 0 in Example 4Test all three cases; λ = 0 is an eigenvalue in Example 4.
Counting n = 0 when it gives the zero functionListing λ₀ = 0 with sin(0) in Example 3sin 0 = 0 is not an eigenfunction, so start at n = 1.
Taking c₂ = 0 in the last stepWriting c₂ sin 2μ = 0, so c₂ = 0That gives only y = 0. Set sin 2μ = 0 instead.
Forgetting to square μWriting λₙ = nπ/2 in Example 3λ = μ², so λₙ = n²π²/4.
Using y instead of y′ in Neumann conditionsSubstituting y(0) = 0 when the condition is y′(0) = 0Differentiate the general solution first.

Related ideas

Shooting method

A computer often solves a boundary value problem by turning it into initial value problems. It guesses y′(0), solves forward with a method such as the Runge-Kutta method, and adjusts the guess until the solution hits the right value at the far end.

Fourier series and partial differential equations

The eigenfunctions sin(nπx/L) and cos(nπx/L) are the building blocks of the next lesson, Fourier Series. Separating variables in the heat and wave equations leads straight to the eigenvalue problems in this lesson.

Practice problems

  1. Solve y″ + 9y = 0 with y(0) = 2 and y(π/6) = 5.
  2. Solve y″ + y = 0 with y(0) = 1 and y(π) = −1.
  3. Solve y″ + y = 0 with y(0) = 1 and y(π) = 1.
  4. Solve y″ − y = 0 with y(0) = 0 and y(1) = 2.
  5. Solve y″ = 6x with y(0) = 1 and y(1) = 4.
  6. Find the eigenvalues and eigenfunctions of y″ + λy = 0 with y(0) = 0 and y(5) = 0.
  7. Show that λ = 0 is not an eigenvalue of y″ + λy = 0 with y(0) = 0 and y(L) = 0.
  8. Find the eigenvalues and eigenfunctions of y″ + λy = 0 with y(0) = 0 and y′(π) = 0.
  9. Is λ = 4 an eigenvalue of y″ + λy = 0 with y′(0) = 0 and y′(π) = 0? If so, give an eigenfunction.
  10. A pinned column 3 meters long has EI = 900 N·m² and satisfies EIy″ + Py = 0, y(0) = 0, y(3) = 0. Find its buckling load.
Show answers

1.

y = c_1\cos 3x + c_2\sin 3x

The characteristic roots are r = ±3i.

y(0) = c_1 = 2

Substitute x = 0.

y\left(\frac{\pi}{6}\right) = c_1\cos\frac{\pi}{2} + c_2\sin\frac{\pi}{2} = c_2 = 5

Substitute x = π/6.

Answer: y = 2 cos 3x + 5 sin 3x

2.

y = c_1\cos x + c_2\sin x,\quad y(0) = c_1 = 1

The characteristic roots are r = ±i.

y(\pi) = -c_1 = -1

cos π = −1 and sin π = 0.

c_1 = 1

Divide by −1; this agrees with y(0) = 1, and c₂ can be any number.

Answer: infinitely many solutions, y = cos x + c₂ sin x

3.

y(0) = c_1 = 1

Same general solution as Problem 2.

y(\pi) = -c_1 = 1 \Rightarrow c_1 = -1

Substitute x = π.

Answer: c₁ cannot be both 1 and −1, so there is no solution

4.

y = c_1\cosh x + c_2\sinh x

The characteristic roots are r = ±1.

y(0) = c_1 = 0

Substitute x = 0.

y(1) = c_2\sinh 1 = 2 \Rightarrow c_2 = \frac{2}{\sinh 1} \approx 1.7018

Substitute x = 1 and divide by sinh 1 ≈ 1.1752.

Answer: y = 2 sinh x / sinh 1

5.

y' = 3x^2 + c_1

Integrate y″ = 6x (antiderivatives).

y = x^3 + c_1x + c_2

Integrate again.

y(0) = c_2 = 1

Substitute x = 0.

y(1) = 1 + c_1 + 1 = 4 \Rightarrow c_1 = 2

Substitute x = 1 and subtract 2.

Answer: y = x³ + 2x + 1

6.

\lambda_n = \left(\frac{n\pi}{5}\right)^2 = \frac{n^2\pi^2}{25}

Use the y(0) = y(L) = 0 row of the table with L = 5.

Answer: λₙ = n²π²/25 and yₙ = sin(nπx/5), n = 1, 2, 3, …

7.

y = c_1 + c_2x

With λ = 0 the equation is y″ = 0.

y(0) = c_1 = 0

Substitute x = 0.

y(L) = c_2L = 0 \Rightarrow c_2 = 0

L ≠ 0.

Answer: only y = 0 is left, so λ = 0 is not an eigenvalue

8.

\lambda = \mu^2:\quad y(0) = c_1 = 0,\quad y' = c_2\mu\cos\mu x

Case λ > 0: start from y = c₁ cos μx + c₂ sin μx.

y'(\pi) = c_2\mu\cos\mu\pi = 0

Take c₂ ≠ 0.

\mu\pi = \frac{(2n-1)\pi}{2} \Rightarrow \mu = \frac{2n-1}{2}

Cosine is 0 at odd multiples of π/2.

\lambda_n = \frac{(2n-1)^2}{4}

Square μ. The cases λ ≤ 0 give only y = 0, as in Example 3.

Answer: λₙ = (2n − 1)²/4 and yₙ = sin((2n − 1)x/2), n = 1, 2, 3, …

9.

\lambda_n = n^2 \Rightarrow 4 = 2^2

From Example 4, n = 2.

y = \cos 2x,\quad y' = -2\sin 2x

Check the conditions with the derivative.

y'(0) = 0,\quad y'(\pi) = -2\sin 2\pi = 0

Both conditions hold.

Answer: yes; y = cos 2x is an eigenfunction

10.

\lambda_1 = \left(\frac{\pi}{3}\right)^2 = \frac{\pi^2}{9}

Smallest eigenvalue with L = 3.

P_1 = 900 \cdot \frac{\pi^2}{9} = 100\pi^2 \approx 986.96

P = EI·λ₁.

Answer: about 987 N

Boundary value problems FAQ

What is a boundary value problem?

It is a differential equation together with conditions at two or more different points, usually the ends of an interval. A typical example is y″ + 4y = 0 with y(0) = 1 and y(π/4) = 3.

What is the difference between an initial value problem and a boundary value problem?

An initial value problem gives y and y′ at the same point, and a nice linear equation then has exactly one solution. A boundary value problem splits the conditions between two points, so it can have one solution, none or infinitely many.

What are eigenvalues and eigenfunctions of a boundary value problem?

For a problem such as y″ + λy = 0 with y(0) = y(L) = 0, the eigenvalues are the values of λ that allow a nonzero solution. Those nonzero solutions, here sin(nπx/L), are the eigenfunctions.

Can a boundary value problem have no solution?

Yes. If the boundary conditions force a constant to take two different values, as in y″ + 4y = 0 with y(0) = 1 and y(π) = 2, no function satisfies both.

Summary

  • Boundary value problems give conditions at two points, so they can have one, none or infinitely many solutions.
  • Solve by finding the general solution and then solving for c₁ and c₂.
  • An eigenvalue is a λ that allows a nonzero solution; that solution is an eigenfunction.
  • Always check the cases λ < 0, λ = 0 and λ > 0.
  • For y(0) = y(L) = 0, the eigenvalues are (nπ/L)² with eigenfunctions sin(nπx/L).

Previous lesson: Method of Frobenius

Next lesson: Fourier Series

Back to the Differential Equations Index