Curves here are regular closed plane curves with matching first and second derivatives at the chosen endpoints; simple means injective except for those identified endpoints. Read signed planar curvature, circle angle lifting and Stokes/frame curvature. Jordan separation and its local inside/outside interpretation are named topological inputs only for identifying the positive boundary orientation. Piecewise smooth corners and general triangulation existence are not proved in this chapter.
Motivation
A tangent vector to a closed curve returns to its initial direction, but the angle may have increased by several full turns. Curvature records its infinitesimal turning; the turning number records the integer that remains after one complete traversal.
For a simple closed curve that integer is forced to be or . This is stronger than saying that the tangent comes back: an immersed figure-eight comes back with turning number , while a circle traversed twice has turning number . We will prove the simple-curve restriction through all its secant directions at once, rather than importing the turning theorem as an unexplained geometric fact.
Definitions
For a regular parametrization , let and let be counterclockwise rotation by in the oriented plane.
Definition 1.3.1
(Signed curvature)
The signed curvature is defined by , where is arclength. Equivalently,
Definition 1.3.2
(Turning number)
Let be a continuous angle lift of on the parameter interval , so . The turning number is
Angle lifting on an interval was proved using the circle covering. Two lifts differ by a constant multiple of , so is well defined.
Definition 1.3.3
(Regular homotopy and positive traversal)
A regular homotopy is a family of closed curves whose position and parameter derivative vary continuously in , with that derivative nowhere zero. Simplicity is not required during the homotopy. For a simple curve, the traversal is positive if its bounded Jordan region is on the left.
Jordan separation supplies that region; at a lowest point of the curve, its local interior is above the horizontal supporting line, so positive traversal points right there.
Theorem/proposition
Theorem 1.3.4
(Curvature integral and regular-homotopy invariance)
Every regular closed plane curve satisfies
Turning number is unchanged by regular homotopies, invariant under increasing regular reparametrization, and changes sign under reversal or a reflection of the plane.
Theorem 1.3.5
(The smooth plane turning theorem)
A simple regular closed plane curve has or . Its total signed curvature is for positive boundary orientation and for the reverse orientation.
Theorem 1.3.6
(Total absolute curvature and its equality case)
Every regular closed plane curve has . For a simple curve this is at least . With positive orientation, equality holds exactly when everywhere; with negative orientation exactly when everywhere. No convexity characterization is assumed in this statement.
Theorem 1.3.7
(Smooth one-chart disk Gauss–Bonnet)
Let a compact region of an oriented smooth surface lie in one positive coordinate patch, with a simple regular boundary and its induced positive orientation. Then
This version follows from the proved smooth turning theorem and Stokes. It does not require an independent turning-theorem assumption or a triangulation.
Proof sketch or proof idea
Proof
(Lifting the tangent angle and comparing nearby curves)
The lifted angle is differentiable locally because a smooth local inverse of the circle parametrization exists around every tangent direction. Differentiating gives . Integrating gives the first formula. Reparametrization and reversal follow from the chain rule, and reflection reverses determinants and angles.
For homotopy invariance, normalized tangents depend continuously on the compact parameter square. At two sufficiently close homotopy times their tangent directions are uniformly nonopposite. There is then a unique continuous relative angle , obtained from the two directions by the local angle of their quotient. Since both tangent paths close, has equal endpoint values. Adding to an angle lift changes its total increment by zero. Thus is locally constant in the homotopy parameter, and connectedness makes it constant throughout. This argument needs only continuous nonvanishing derivatives, not a homotopy of simple curves.
Concept lensCombinatorial
Counting the times the tangent points one way
The turning number can be read off by counting. Fix a unit vector such that at every parameter where (almost every direction qualifies). Those parameters are isolated, hence finitely many on the closed curve, and
because each such point is a crossing of the angle lift through a level , upward when and downward when . For the figure-eight below no tangent points straight up, so the sum is empty and .
The requirement that the derivative never vanish is not a technicality.
Counterexample mode
A homotopy through a cusp changes the turning number
Claim: turning number is unchanged along any continuous family of closed curves. In complex notation let for . Then , whose angle is . For the factor stays in the right half-plane, so its argument has net change and . For it winds once around , so . Position and derivative vary continuously in , yet jumps. The failure is at , , where : the curve has a cusp and the tangent direction is undefined. Requiring a nowhere-zero derivative restores invariance.
The turning theorem itself needs a new idea: compare the tangent with every secant direction at once.
Proof
(The secant-direction triangle)
Reparametrize a simple curve by arclength on , choosing at a point with minimal vertical coordinate. There its unit tangent is horizontal. On the triangle
define the unit secant
except at the corner . Simplicity ensures the denominator is otherwise nonzero. Extend on the diagonal by , and at by .
These extensions are continuous. Near a diagonal point, the difference quotient is the average of over a shrinking interval and tends uniformly to its tangent. Near , write and , with . Matching endpoint derivatives gives
After normalization the direction tends to , independently of the relative rates at which tend to . Thus is a continuous map from the entire closed triangle to the circle.
The triangle is simply connected and locally path connected, so the proved circle lifting theorem supplies one real angle for . Along the left edge put , using the endpoint extensions. Every chord from the lowest point lies in the closed upper half-plane, so lies in the closed upper semicircle. It begins at and ends at . Its angle increment is therefore when points right and when it points left: the upper semicircle has the single continuous angle coordinate , so no additional full rotations are possible there.
Along the top edge, , including the endpoint extensions. Negation adds a fixed to an angle lift, so the increment along this edge from to is again . Since is one function on the triangle, its increment from to along left then top is . Along the diagonal that same increment is the total tangent turning. Therefore
For positive Jordan orientation, the interior at the lowest point is above the supporting line and lies to the left of the rightward tangent, so the sign is positive. Without invoking any inside/outside theorem, the calculation already proves the unsigned conclusion . This separates the geometric argument from its named orientation input.
Proof X-ray
Where simplicity and the lowest point enter
A tangent loop by itself can wind any integer number of times; the proof gains control by enlarging the domain. Simplicity is used exactly once, to make the secant defined and continuous off the diagonal, so that it lives on the whole triangle and has a single angle lift . The key move is the choice of a lowest base point: it confines the left edge to the closed upper semicircle, so its increment is exactly one half-turn, , with no extra rotations. The identity makes the top edge repeat that increment, and the argument closes because a single-valued has the same increment along the two-edge route and along the diagonal. For a circle traversed twice the secant map is undefined at the repeated points, and that is precisely where the count escapes.
The remaining two theorems are short consequences.
Proof
(Absolute curvature and the disk formula)
The absolute-curvature bound follows from the integral triangle inequality. For a positively oriented simple curve, the difference between absolute and signed integrals is twice the integral of the negative part of . Since is continuous, this vanishes exactly when is nonnegative everywhere. The reversed case is identical. This proves the equality condition without silently equating it to an unproved global convexity theorem.
For the surface consequence, write the boundary coordinate curve as and its metric as in the positive patch. The Gram–Schmidt frame has tangent components obtained from the coordinate velocity by the matrix
Both diagonal entries are positive. The matrices remain upper triangular with positive diagonal, so the normalized vectors give a homotopy of closed nonzero direction loops. Thus the tangent angle relative to this orthonormal frame has the same winding as the planar coordinate tangent, namely . This avoids assuming that a varying metric frame leaves winding unchanged without proof.
The already derived identity is along the boundary. Its angle term integrates to by the preceding argument. Stokes and then give
proving the claimed smooth-boundary disk formula. Curvilinear triangles have corners, so they are not silently included in this smooth statement.
Worked examples
Worked example 1.3.8
(Repeated circles and orientation changes)
For , and nonzero integer , the speed is and , so the signed curvature is . Thus the curvature integral is and . When the map is not simple, although its image is a circle. Reversal or reflection changes the sign; changing speed through an increasing regular parametrization does not change .
Worked example 1.3.9
(An ellipse attains the absolute-curvature bound)
For with , the determinant of is and the speed cubed is . Curvature is therefore positive. The ellipse is simple and positively oriented, so its signed and absolute curvature integrals are , even though curvature varies. The turning theorem evaluates this integral without requiring an antiderivative of the displayed expression.
Worked example 1.3.10
(A regular figure-eight has turning number zero)
Take . Its derivative never vanishes: when the second component is . The curve crosses at , so is not simple.
Its tangent satisfies , where reflects across the vertical axis. Reflection reverses angle increments, so the second half's tangent increment is the negative of the first half's. Matching the lifts at the joining point gives total increment . Thus and the signed curvature integral is . It cannot be regularly homotoped to a once-traversed circle, even if self-intersections are allowed during that homotopy.
Yes: by the Whitney–Graustein theorem, two regular closed plane curves are regularly homotopic exactly when their turning numbers agree.
A natural next question
Does the turning number decide regular homotopy?
Yes: by the Whitney–Graustein theorem, two regular closed plane curves are regularly homotopic exactly when their turning numbers agree.
The invariance proved above gives one direction. The converse is the Whitney–Graustein theorem: if , there is a regular homotopy from to . A proof must deform the derivatives through nonzero loops and also keep the closing condition , which an arbitrary deformation of the tangent loop does not preserve. So the figure-eight is regularly homotopic to every regular closed curve with , and to no circle.
Worked example 1.3.11
(A curvature cap forces a length bound)
If a simple closed curve has with , then , so . Equality forces a circle: orient positively, so . The integrand is continuous and nonnegative, hence everywhere. Its tangent angle is , and integrating the tangent gives
for a fixed center . This is a radius circle, traversed once because . Thus the bound includes a proved equality case, not just a heuristic statement that a curve must bend enough to close. Equivalently, every such curve of length has a point with absolute curvature at least ; the maximum exists by continuity and compactness. Equality for that maximum again forces a circle, so a noncircular ellipse cannot attain it. The case permits no such closed curve.
Common mistakes
Common mistake
(A simple image is not a simple parametrization)
Traversing a circle twice has a simple image but violates the theorem's injectivity hypothesis, and its turning number is . Returning to the same direction only gives an integer winding, not automatically one turn.
Common mistake
(Reading shape from a zero turning number)
Signed curvature may cancel; absolute curvature cannot. Zero turning number does not mean a straight curve, as the figure-eight shows. The absolute lower bound is empty when .
Common mistake
(Using the wrong limit at the exceptional corner)
At the secant triangle's exceptional corner the limiting direction is , not . Confusing it with the diagonal extension would remove the very half-turn that drives the proof. Piecewise corners also need separate treatment rather than pretending their tangent is continuous.
Summary
Angle lifting converts curvature into an integer tangent winding, invariant under regular homotopy. Simplicity creates a continuous secant map on a triangle; a supporting line limits its edge turning and forces winding . This proves the absolute-curvature bound and supplies the formerly imported smooth turning step in one-chart disk Gauss–Bonnet.
Exercises
Quick check
(Compute the turning number of the unit normal and explain why it agrees with that of .)
Solution · Answer
Multiplying directions by the fixed rotation adds to an angle lift at all times. Its endpoint increment is unchanged, so the normal loop has the same . No new geometric assumption is needed.
Quick check
(Can a regular closed curve with everywhere positive signed curvature have turning number ? Must it be simple?)
Solution · Answer
The continuous positive curvature has positive integral, so the integer is positive and cannot be . Simplicity does not follow: a circle traversed twice has positive curvature and . Even the theorem is a necessary simple-curve condition, not a supplied converse characterization.
Quick check
(Prove a simple regular closed curve cannot have everywhere.)
Solution · Answer
Its signed curvature integral would be , contradicting the turning theorem's . Equivalently makes its unit tangent constant, so the arclength displacement cannot close for positive .
Quick check
(Explain why passing through a curve with one zero derivative may change turning number, and why the regular-homotopy proof no longer applies.)
Solution · Answer
Normalization is undefined at the zero. There is then no continuous tangent map on the full parameter square and no uniformly nonopposite comparison across that time. The integer-invariance proof loses its hypothesis; it does not prohibit a change through such a singular curve, and the family above shows the change really happens.
Quick check
(For a smooth positively oriented boundary in a flat one-chart surface region, compute its total geodesic curvature. Does flatness force the boundary itself to be geodesic?)
Solution · Answer
The smooth disk formula with gives total geodesic curvature . A geodesic boundary would have everywhere and contradict that value. Flatness concerns the surface, not the turning of an arbitrary boundary drawn on it.