The Euler–Lagrange equation finds stationary curves, but geometry supplies a useful warning: even a smooth stationary surface spanning two fixed circles need not have least area. We study the restricted class of surfaces formed by revolving a positive graph about the horizontal axis. This restriction matters: no conclusion about arbitrary spanning surfaces follows automatically.
Area and the variational equation
Definition 11.1.1
(Admissible rotational graphs and the reduced area)
Fix . The admissible class consists of the functions with on and . Revolving the graph about the -axis produces a surface spanning two coaxial circles of radius at distance , with area
The functional is called the reduced area.
Indeed, the parametrization has area element . The factor does not change minimizers, so we work with . For any test with , the perturbed graph remains positive for all sufficiently small , because has a positive minimum on the compact interval. Thus a local minimum in the topology must have vanishing first variation and satisfy
The Lagrangian has no explicit dependence, so along a stationary curve the Beltrami identity gives
for some constant , which is positive because . The next result identifies every stationary graph.
Proposition 11.1.2
(Stationary rotational graphs are catenaries)
A positive function satisfies the Euler–Lagrange equation of if and only if
In particular every stationary graph is strictly convex.
Proof
(Reduction to and integration)
Expanding the derivative,
Multiplying the Euler–Lagrange equation by gives , that is,
Since , this forces , so a stationary graph is strictly convex.
Now let be stationary and let be its Beltrami constant, so . Put ; then , and . Differentiating gives , and the reduced equation becomes . As , we get , so for some . Hence and .
Conversely, for we have and , so at every point, including the horizontal tangent .
Proof X-ray
What the first integral does and does not prove
The proof uses two equations, and they play different roles. The Beltrami relation alone is too weak: differentiating it yields , which holds at a point either when or when the Euler–Lagrange equation holds there. A positive constant satisfies the first integral with constant , yet it is not stationary, because the reduced equation would read . The proof therefore never divides by . It uses the first integral only to write , and uses the reduced Euler equation to obtain , where the division is by , which never vanishes. The converse direction substitutes the catenary back into the Euler equation itself. Positivity of enters twice: it makes positive and it gives the convexity sign. Nothing in the argument says that a catenary minimizes ; that requires a comparison, studied below.
Equal rings and the number of candidates
The equal endpoint conditions imply , hence , and since this gives . Put . The endpoint condition becomes
Proposition 11.1.3
(Counting symmetric stationary catenoids)
Let be the unique solution of , and let . In the admissible class, the number of stationary graphs is if , exactly if , and exactly if .
Proof
(Monotonicity of on each side of )
By the proposition above and , stationary graphs correspond bijectively to solutions of , where , through . We have , which has the sign of . Since for , and , the function has exactly one zero , characterized by . Thus strictly decreases on and strictly increases on , and as and as . Each value above the minimum is attained once on each side, the minimum once, and smaller values never. Finally gives .
This counts smooth stationary positive graphs in our class; it does not prove that a minimizer exists or that either candidate minimizes.
Worked example 11.1.4
(Both candidates for , )
Here , so there are two stationary graphs. One root is , giving and . The other root lies in ; solving numerically gives , so and .
For we have , hence
With , this is . With it is about . The shallow candidate (large ) has the smaller reduced area. This comparison of two stationary graphs still says nothing about competitors that are not stationary.
Worked example 11.1.5
(Rings too far apart: no stationary graph)
Take and . Then , so no admissible graph is stationary. Directly: a symmetric catenary through the rings would need for some , but for all . Since a -local minimizer in the admissible class would be stationary, the class has no such minimizer at all. The infimum of is not attained by a positive graph in this configuration.
Worked example 11.1.6
(The borderline ratio)
If exactly, the two candidates merge into a single one with , that is . For this means and . Check: . At this ratio the endpoint equation has a double root, and , so a small decrease of makes the stationary graph disappear.
Quick check
(Why must a stationary graph with have ?)
Use that is even and strictly increasing on .
Solution · Answer
The endpoint conditions give . Since is even and injective on , this means . Squaring gives , and forces .
A stationary surface that loses a comparison
Counterexample mode
Stationary does not mean least area
A tempting claim is that a smooth stationary graph meeting the boundary conditions must minimize in the admissible class. Take and . The stationary graph is admissible, and so is the constant graph . Their reduced areas are
So the stationary catenoid has strictly greater area than a permitted cylinder and is not a global minimum in this graph class. The cylinder is not stationary: substituting a positive constant in the reduced Euler equation gives . A competitor can disprove minimality without being a minimizer. The correct statement is only the necessary one: a minimizer, if it exists in the class, is stationary. Minimality needs a separate comparison or second-variation argument.
The previous worked example shows that the competitor beating the deep catenoid is not the end of the story either: the shallow catenoid with has smaller reduced area than both.
Common mistake
(Using the first integral as the equation)
The Beltrami relation is satisfied by every positive constant , which is not stationary. Solutions of the first integral must be checked in the Euler–Lagrange equation, or derived from it, as in the proof above.
Common mistake
(Reading the candidate count as an existence theorem)
Having two stationary graphs when does not show that either one minimizes, and having none when only shows that no positive graph minimizes. Counting solutions of is a statement about stationary points in the chosen class, not about minimizers or about arbitrary spanning surfaces.
Practice and solutions
Exercise 11.1.7
(For and , verify that is stationary and compare its reduced area with that of the constant graph.)
Foundation
Solution(3)
- 1
Check the boundary values and the equation
Here , and . Directly .
- 2
Compute the catenoid value
Because: .
- 3
Compare with the cylinder
, so the catenoid wins this comparison. This comparison alone does not prove global minimality.
Exercise 11.1.8
(Explain why the constant graph is not a local minimum of for any in the admissible class.)
CoreProof
Solution(3)
- 1
Choose an inward perturbation
Take a nonnegative, nonzero smooth with and set . For small , is positive and admissible.
- 2
Differentiate at
Key step
Because: and when .
- 3
Conclude
∎The reduced area strictly decreases for all sufficiently small positive , so the cylinder is not a local minimum.
Quick check
(Could a stationary graph in the admissible class have an interior maximum?)
Look at the sign of from the reduced equation.
Solution · Answer
No. At an interior maximum we would have , but every stationary graph satisfies . In the admissible class , so a stationary graph is decreasing on , increasing on , and has its only minimum, the neck, at .