Why is a circle the most economical way to enclose area? Fourier series turn the answer into a comparison of frequencies: area weights a frequency by its signed index, whereas energy weights it by the square of that index. After taking absolute values, the nonzero frequencies with are exactly .
Length, signed area and the sharp inequality
Definition 1.8.1
(Length and signed area of a closed curve)
We work with a closed regular curve , with matching endpoints and derivatives and . Write and
The quantity is signed area with multiplicity. For a positively oriented Jordan curve, Green's theorem identifies it with the ordinary enclosed area. For a self-intersecting curve, lobes can cancel; no simplicity assumption is needed in the analytic inequality below.
Theorem 1.8.2
(Sharp length–area inequality)
Every closed regular curve above satisfies
Equality holds exactly when its image is a circle traversed once, in either direction, with any regular orientation-preserving reparametrization; once means the parametrization is injective on . Here follows from regularity.
Proof
(Make the speed constant, then compare modes)
Let . Since the speed is continuous and positive, is a increasing bijection from to with inverse. Define the new parameter and the reparametrized curve . Its derivative has matching endpoints, and it still has length and signed area , but now has constant speed . Rename as and as .
Write the Fourier expansion in . Parseval applied to and gives
The second identity uses the inner product of with ; its series is absolutely convergent by Cauchy–Schwarz. Constant speed gives . Thus
because for integral . This is (1).
For equality, only can occur. Equality in the first inequality additionally forces all nonzero-frequency contributions to have the same sign, so either or . Since the speed is nonzero, the remaining coefficient is nonzero. Consequently or , which traverses a circle once. Conversely a once-traversed circle has and . Reparametrization does not change these quantities.
Proof X-ray
Where the sharp constant comes from
Regularity is used twice: to reparametrize by arc length with a inverse, and at the end to rule out the zero coefficient in the equality case. The key move is the constant-speed parametrization. For any parametrization, Cauchy–Schwarz gives only , an upper bound on length by energy, which points the wrong way for proving (1). At constant speed this Cauchy–Schwarz step is an equality, so becomes exactly and the frequency comparison can be applied to the length itself.
The comparison then consists of two separate inequalities, and the equality analysis reads them one at a time. The first, , is an equality only when all nonzero frequencies present have the same sign; this is where orientation enters. The second, , is an equality only at ; this is where the circle appears. Together they leave one first frequency and a translation.
A quantitative view of the deficit
Suppose the curve has been parametrized at constant speed and orient it so that . Then equations (2) give the exact identity
Every weight is nonnegative. The terms vanish: translations and positively oriented circles cost nothing. The term has weight , and every other nonzero weight is at least . Therefore, with ,
This is an absolute stability statement in this particular parametrization: a small deficit makes the curve close in to a translated, positively oriented circle, possibly of radius zero. It does not by itself give a pointwise distance bound or a stability statement for an arbitrary parametrization. If , reverse the orientation before using (3)–(4).
Examples
Worked example 1.8.3
(An ellipse)
For with , direct integration gives . Its length generally has no elementary formula, but (1) yields , with equality only when . The displayed parametrization has nonconstant speed unless , so its energy cannot be replaced by .
Worked example 1.8.4
(A multiply covered circle)
For with and a positive integer , and . Hence : tracing the same circle repeatedly makes grow like while grows only like . Equality requires .
Worked example 1.8.5
(Two-sided length bounds from two modes)
Let with . Then , so the curve is regular, but its speed is not constant. By (2), which holds for any parametrization,
The theorem gives , and Cauchy–Schwarz gives . Hence
For the bounds are about and , and numerical integration of gives . The gap is the cost of the second mode alone; at both bounds collapse to the circle's length .
Counterexample mode
Signed area is not always the size of the enclosed region
The regular closed curve has signed area zero: the two integrals and vanish by orthogonality. Its derivative never vanishes, since implies . The two lobes are oppositely oriented and cancel. Thus does not mean that the curve has zero length or no lobes.
Common mistake
(Reading an energy gap as a non-circle)
For a general parametrization, measures both shape and uneven speed. The unit circle traversed once as has , although equality holds in (1). Only after reparametrizing at constant speed does the gap become the geometric deficit in (3).
Common mistake
(Skipping the orientation step in the stability bound)
The bound (4) concerns closeness to a positively oriented circle and assumes . For the isoperimetric deficit is zero, yet , since . Applying (3) with gives , which is not the isoperimetric deficit. Reverse the orientation first; the reversed curve is then exactly a positively oriented circle.
Practice with solutions
Quick check
(Let . Compute and .)
Solution · Answer
Orthogonality gives and . Thus . These formulas hold regardless of speed; replacing by requires constant speed.
Quick check
(For , find , , and the deficit in (1).)
Solution · Answer
The radius is , so and . Hence .
Quick check
(Find the length–area ratio for .)
Solution · Answer
Here and , so . Its image is a circle, but the parametrized curve traverses it twice and does not give equality.
Quick check
(For a constant-speed curve with , show .)
Solution · Answer
In (3), outside . Parseval yields (4), whose right-hand side is exactly .
Quick check
(Consider , . Compute , and .)
Solution · Answer
The phase increases from to , so the curve traverses the unit circle once: , . But , giving . Equality in (1) is geometric and does not force every parametrization to have constant speed. The proof uses the constant-speed reparametrization, so there is no contradiction with here.