Read planar area/Stokes, arclength and the controlled corner-smoothing chapter. All periodic scalar functions used in the differential inequality are real ; plane curves in the main equality theorem are regular closed with matching endpoint derivatives. We prove the approximation facts needed below rather than assume an unproved Fourier expansion theorem.
Motivation
A closed curve of fixed length cannot enclose arbitrarily large area. A circle is the natural candidate for equality, but a sharp proof must explain why every other shape loses area and why equality really forces a circle.
The analytic input is a periodic inequality: a mean-zero function cannot be large while its derivative stays too small. Its best constant is set by the first sine and cosine modes. We prove this first for finite trigonometric polynomials, then approximate functions and their derivatives by the same positive kernels. Thus infinite series convergence is not an extra hidden premise.
Definitions
For a function of period , put ; the function has zero mean if . The periodic Wirtinger inequality will compare the squared integral of such a function with that of its derivative.
Definition 1.4.1
(Positive kernel averages)
For the -periodic kernel is
For continuous -periodic , its kernel average is
The exponential notation merely abbreviates finite sine/cosine sums.
Definition 1.4.2
(Signed area integral)
For an oriented regular closed plane curve , the signed area integral is
It is independent of orientation-preserving parametrization and of translations, since the integral of over a closed curve is . For a positively oriented simple curve, the proved planar Stokes formula identifies with the actual area of its bounded region. For self-intersecting curves it is an algebraic area, not the sum of all lobe areas.
Theorem/proposition
Theorem 1.4.3
(Positive finite kernels approximate function and derivative)
is nonnegative and has integral on a period, and is a finite trigonometric polynomial converging uniformly to every continuous -periodic . If is , then converges uniformly to as well. Means and absent sine/cosine coefficients are preserved.
Theorem 1.4.4
(Sharp periodic Wirtinger inequality)
For every real -periodic mean-zero ,
Equality holds exactly for , including the zero function.
Theorem 1.4.5
(The sharp isoperimetric inequality)
A regular closed plane curve of length satisfies . Equality holds exactly for a circle traversed once, with either orientation. In particular a simple positively oriented curve encloses area
with equality exactly for a circle. The inequality also extends to the finite noncuspidal piecewise regular Jordan boundaries treated by controlled smoothing; the main equality proof below uses the smooth hypotheses.
Proof sketch or proof idea
Proof
(Positivity and a tail bound for the kernel)
Expanding the finite squared geometric sum counts pairs with difference , proving the displayed cosine expansion. Hence , and its constant term gives integral . The geometric-sum identity also gives
away from multiples of , with the continuous value at . Thus for it is at most .
Let . Uniform continuity gives for , uniformly in . Using positivity, total mass after normalization, and the preceding outside bound,
First fix for , then let grow. This proves uniform convergence. The finite expansion shows is a trigonometric polynomial whose coefficients are the original sine/cosine coefficients multiplied by . In particular the mean and any missing modes stay so. For , differentiation under the finite integral gives as the same convolution of . Applying the identical argument to continuous proves derivative convergence. No assertion about arbitrary raw Fourier partial sums was used.
With uniform approximation of both and available, Wirtinger's inequality reduces to finite sums.
Proof
(Finite sums, a limit, and the first-mode gap)
For a mean-zero trigonometric polynomial on with coefficients , orthogonality of sine and cosine gives
These are finite sums, verified by elementary product-to-sum integration. Thus . Approximate a mean-zero by ; uniform convergence of and lets their squared integrals pass to the limit. This proves Wirtinger for period , and the substitution gives the stated constant for general .
For equality, still use period . Let be the first-mode projection of , with and , and put . Then has no constant or first modes. Its finite approximants therefore satisfy , so the limit gives . Orthogonality gives . Integration by parts, using periodic endpoint values, gives , and . Consequently
Equality forces by continuity; first modes plainly attain equality. Rescaling completes the equality characterization.
Concept lensAnalytic
The sharp constant is an eigenvalue
On nonzero mean-zero -periodic functions, Wirtinger's theorem says that the Rayleigh quotient
has minimum value , attained exactly by the first modes, which are the solutions of . Thus is the smallest positive eigenvalue of on -periodic functions; the mean-zero condition removes the constants, whose eigenvalue is . The finite-sum proof is this operator in diagonal form: it multiplies the -th mode by .
The isoperimetric inequality applies this scalar result to both coordinates of the curve.
Proof
(Two estimates on the centered position curve)
Arclength-parametrize on and translate its mean to . Apply the scalar inequality to both coordinates:
Let be the counterclockwise quarter-turn. Then . Integral Cauchy–Schwarz gives
For completeness, its equality condition follows by minimizing the nonnegative quadratic over constant real . Equality means that quadratic is for its minimizing ; continuity then gives pointwise. The nonconstant regular curve has , so the minimizing coefficient is defined.
If equality holds in the area bound, both inequalities above are equalities. Since and , the relation forces . Solving this constant rotation equation makes a circle about the translated origin, with radius from unit speed. Since it is traversed once. Conversely such a circle attains equality. For a positive simple boundary Stokes makes , giving the geometric statement with its full rigidity.
For a piecewise boundary, use the earlier simple regular smoothing . Positions converge and derivatives converge uniformly off short junction intervals, with all derivatives uniformly bounded. Thus , by separating those intervals from their complement. The reverse triangle inequality then gives convergence of lengths. The proved boundary-tube estimate gives convergence of enclosed areas. Apply the smooth inequality and pass to the limit. Uniform position convergence alone would not justify the length step.
Proof X-ray
Two inequalities, and what equality in each forces
The proof chains two estimates, and the hypotheses are used at definite points. Centering the curve is what makes each coordinate mean-zero, so Wirtinger applies; translation is harmless because is translation invariant. Arclength makes , turning Wirtinger into . The key move is writing , which turns area into an inner product to which Cauchy–Schwarz applies. Rigidity comes from the two equality cases separately: Wirtinger's forces only first modes, which still allows ellipses, while Cauchy–Schwarz forces , and only the two together give , a circle of radius traversed once.
Worked examples
Worked example 1.4.6
(Repeated circles give a strict algebraic-area bound)
For a radius- circle traversed positive times, and . Hence . Equality holds for and is strict for every integer . A circle as an image is not enough: the parametrized traversal matters. Reversal changes the sign of but not or the absolute bound.
Worked example 1.4.7
(An ellipse has a strict perimeter lower bound)
For with , the determinant integral gives area . The sharp inequality gives , without evaluating the elliptic perimeter integral. If the curve is not a circle, so the inequality is strict. This uses the proved equality theorem rather than an informal comparison of drawings. Translation or rotation does not affect either quantity.
Worked example 1.4.8
(Rectangles stay below the circle optimum)
A rectangle of sides is a noncuspidal piecewise regular Jordan boundary, so the smoothing extension applies. Its and satisfy
Thus even a square is strictly below the circle optimum. At fixed rectangular perimeter, shows the square maximizes rectangle area, but that restricted optimization does not prove the global isoperimetric theorem among all curves.
Worked example 1.4.9
(A small deficit controls the higher position modes)
Center a unit-speed curve and let be the first sine/cosine projection of both its coordinates, with . Set and . The strengthened first-mode-gap argument, rescaled to period and added for both coordinates, gives
Cauchy–Schwarz also gives , hence
The equality in the middle uses ; no geometric approximation was substituted. Therefore
The comparison vectors are explicitly and , the actual coordinate projections of the centered curve, not fitted parameters. The first projection is an ellipse or a degenerate one; the estimate alone does not assert a sharp distance to a circle. For a uniform bound, note that has no first modes, so
and . Also has mean , and
Consequently . This nonsharp estimate also bounds the Hausdorff distance between the two images by the same parameter correspondence. When , higher modes vanish, and the Cauchy–Schwarz equality condition further forces the ellipse to be a circle. As a check, the dimensions agree: the integral has length cubed, as does times the area deficit.
Common mistakes
Counterexample mode
Wirtinger fails without zero mean
Claim: every -periodic satisfies . Take . Then , while gives . The claimed inequality would read , which is false. The constant mode has positive squared integral and zero derivative, so it must be removed. Translating a closed curve to mean is legitimate because translation leaves its signed area integral unchanged.
Common mistake
(Overstating what kernel averages prove)
Positive finite-kernel averages converge uniformly here. This does not assert that all Fourier partial sums converge uniformly for arbitrary continuous functions. The derivative conclusion also uses regularity.
Common mistake
(Reading signed area as total area)
For self-intersections, signed area can cancel. The bound proved for is not automatically a bound for a sum of unsigned overlapping lobe areas. Nor does boundary convergence alone control perimeter.
Summary
Finite positive kernels supply exactly the function/derivative approximation needed for sharp Wirtinger, including its equality case. Applying it to the centered position curve and then using integral Cauchy–Schwarz yields the sharp area–length inequality. Equality forces a constant rotation equation and hence a once-traversed circle; controlled smoothing extends the inequality to the stated corner boundaries.
Exercises
Quick check
(Derive the kernel's unit normalized mass from its finite cosine expansion. Where does positivity enter uniform approximation?)
Solution · Answer
Every nonconstant cosine has integral over a period, leaving from the constant . Dividing by gives mass . Nonnegativity lets the error integral be bounded by the local uniform-continuity error times that mass, while the outside kernel estimate controls the tail; signed cancellation would not supply the same estimate.
Quick check
(For , compute the ratio of derivative and function squared integrals. Why can it not attain the sharp Wirtinger equality?)
Solution · Answer
The function squared integral is , and the derivative squared integral is . Their ratio is , four times the first-mode value . This is a second mode; only first sine/cosine combinations attain equality.
Quick check
(For , show and explain why this does not say its two lobes have zero total ordinary area.)
Solution · Answer
Its determinant integrand is . Replacing by changes its sign, so the full integral cancels. The two lobes are traversed with opposite orientations and have positive unsigned areas. Algebraic cancellation is precisely what records.
Quick check
(Check how length, area and the isoperimetric ratio change under scaling , .)
Solution · Answer
Length multiplies by and area by . Therefore is unchanged. A circle attains at every scale, confirming both the dimensional form and the sharp constant. Orientation reversal changes signed area separately and is not a positive scaling.
Quick check
(Explain why the uniformly convergent arcs , , do not have controlled lengths, and contrast this with positive smoothing.)
Solution · Answer
The positions converge uniformly to the straight segment, but the derivative's vertical component is . Length is at least , which grows without bound. Our smoothing proof instead has uniformly bounded velocities, convergence off short junction intervals and consequent derivative convergence. Those extra facts, not position convergence alone, justify its perimeter limit.