Motivation
The Gaussian curvature condenses two principal curvatures into a single number. To understand what that number omits, examine the bending of the surface in individual tangent directions. Euler's formula describes every normal curvature from two perpendicular principal directions and also shows why the sign of the product distinguishes local shapes.
Definitions
Normal curvature of a direction
Let be an oriented regular surface, , and the chosen normal. The self-adjoint shape operator has an orthonormal eigenbasis with real eigenvalues .
Definition 3.2.1
(Normal curvature, principal directions and umbilics)
The normal curvature of a unit tangent vector is . The eigendirections of are the principal directions at . The point is umbilic if .
Every unit tangent vector can be written , and the normal curvature then depends only on .
Theorem 3.2.2
(Euler's formula)
In the notation above, for ,
Proof
(Expanding the second form in a principal basis)
Since and ,
because removes the cross terms and .
It follows that the largest and smallest normal curvatures are the larger and smaller eigenvalues. If the two eigenvalues agree, every tangent direction is principal; such a point is called umbilic. If they differ, precisely the two eigendirections (as unoriented lines) achieve the extreme values.
A normal section supplies a geometric interpretation. Intersect the surface locally with the plane spanned by and . At this plane is transverse to the surface because their tangent spaces span ; the implicit function theorem supplies a local regular intersection curve tangent to . Parametrize it by arc length. Its acceleration lies in the section plane and is perpendicular to , so at it is a multiple of . The signed curvature of this plane curve relative to is therefore exactly . The sign reverses when reverses; the existence of an asymptotic direction, defined below, does not.
Theorem/proposition
Signs and special directions
At a point where , the two principal curvatures are nonzero and have the same sign: the point is elliptic. When , their signs differ and the point is hyperbolic. When but the shape operator is nonzero, it is parabolic under the convention used here. If both eigenvalues vanish, we call it a flat point; terminology varies in how flat points are grouped.
Definition 3.2.3
(Asymptotic direction)
An asymptotic direction at is a nonzero tangent direction with .
How many asymptotic directions a point has depends only on its type.
Theorem 3.2.4
(Asymptotic directions by point type)
At a hyperbolic point Euler's formula gives two distinct unoriented asymptotic lines: if , then . At an elliptic point there are none, because both terms have the same nonzero sign. At a parabolic point there is one such line, the kernel of ; at a flat point every direction is asymptotic. These statements concern a point and do not, by themselves, assert the existence of a global curve following those directions.
The proof is the factorization of the quadratic carried out in the section on asymptotic directions and equations below.
Worked examples
Example: a saddle
Worked example 3.2.5
(A saddle)
At the origin of with upward normal, and have principal curvatures and . Thus
The two asymptotic lines are and . In fact those lines lie on this particular graph, since there. Do not infer that asymptotic curves on every surface are straight lines: zero normal curvature is only an instantaneous condition on the normal component of acceleration.
Concept lensGeometric
The second form as the shape of nearby slices
Place at the origin with tangent plane , principal directions along the - and -axes, and pointing up. Taylor's theorem writes the surface near as
Slice by the planes and divide coordinates by . Near any bounded region the rescaled slices approach the conic , the Dupin indicatrix: an ellipse at an elliptic point, two conjugate hyperbolas with the asymptotic lines as common asymptotes at a hyperbolic point, and a pair of parallel lines, on one side only, at a parabolic point.
Finding principal directions from the two forms
In a coordinate basis, the first and second forms have symmetric matrices
A principal direction with coordinate vector satisfies , so the possible principal curvatures solve
It would be incorrect to find the ordinary eigenvalues of alone unless the coordinate basis is orthonormal. Since is positive definite, the operator is self-adjoint with respect to the -inner product, even though its matrix need not be symmetric in the ordinary Euclidean coordinate inner product. The spectral theorem applied on the tangent plane ensures real eigenvalues and an orthonormal eigenbasis. The sum and product of the two roots are and .
Counterexample mode
The eigenvalues of B alone are not principal curvatures
Claim: the principal curvatures are the ordinary eigenvalues of the second-form matrix . Take the cylinder with outward normal . Then and , so . Also gives , and , so . Its ordinary eigenvalues are and ; for that is , not the true value. The correct equation is , giving and . The claim holds only when the coordinate basis is orthonormal at the point, that is .
Example: tilted coordinates on a plane
Worked example 3.2.6
(Tilted coordinates on a plane)
Take . Then , which is not the identity, and . Both generalized eigenvalues are zero: every direction is asymptotic. The coordinate lines are not perpendicular, but the surface is a plane. Nonzero off-diagonal entries of the first form record skew coordinates rather than curvature.
Example: a quadratic graph with rotated principal axes
Worked example 3.2.7
(A quadratic graph with rotated principal axes)
For , the gradient vanishes at the origin. There and
The eigenvectors and have eigenvalues and . Normalizing them gives an orthonormal principal basis. Thus , , and the origin is elliptic; Euler's formula bounds every unit-direction normal curvature between one and three. The coordinate axes themselves are not principal directions here. Diagonalizing the second form relative to the first is a geometric operation, not an instruction to discard its mixed term.
The explorer below diagonalizes relative to for a graph or for forms you enter, rotates a direction to compare with Euler's formula, and marks the asymptotic directions.
What the point classification does and does not say
A hyperbolic point has tangent directions of opposite signed normal curvature. This suggests a saddle, but the terminology records a condition at one point; it should not be taken as a global description of an entire surface. At an umbilic point the principal directions are not uniquely determined, even though their common curvature is. A sphere has every point umbilic; a plane has every point flat. An umbilic point with zero curvature has every tangent direction asymptotic, illustrating that “principal” and “asymptotic” answer different questions.
Extremal curvature and the metric
Euler's formula can also be read as an optimization statement.
Concept lensAnalytic
Principal curvatures as extremes of a ratio
For a nonzero tangent vector put , which is unchanged when is multiplied by a nonzero scalar. In a principal basis it is a weighted average of and with nonnegative weights summing to one. Therefore lies between the two principal curvatures, and equality at an endpoint occurs exactly on the corresponding eigenspace, unless both eigenvalues agree, in which case is constant. The principal curvatures are thus the maximum and minimum of over directions, and the principal directions are where they are attained.
This description matters in a calculation with an oblique coordinate system, where a coordinate vector's length is measured by the first form, not by the ordinary sum of the squares of its coordinate entries.
Common mistake
(Maximizing the second form without normalizing)
Asking to maximize without fixing is meaningless: replacing by multiplies the value by four although the direction is unchanged. Maximize the ratio , or equivalently solve , which builds in the normalization and gives the same geometric extrema in every regular chart.
The characteristic polynomial of the shape operator is . Its discriminant is . Thus for either normal orientation, with equality precisely at an umbilic point. The inequality remains true at a hyperbolic point, where ; it does not say that is positive. Reversing the normal changes to and leaves the inequality unchanged.
A point on a graph
Consider at the origin with upward normal. The first derivatives vanish, so the coordinate tangent vectors are orthonormal there. The Hessian, and hence the second-form matrix at that point, is
Its characteristic equation is , giving and . Both are positive. Therefore , , and the point is elliptic. The largest possible unit-direction normal curvature is and the smallest is . A direction along the first coordinate axis has curvature two, which lies between those values but is not an extremum: the mixed term tilts the principal directions away from the coordinate axes.
Umbilics and a local warning
At an umbilic point , so every unit direction has normal curvature . The sphere has this property at every point.
Common mistake
(One umbilic point does not make a sphere)
A statement at one point does not classify a whole surface. The graph has zero gradient and Hessian at the origin, so it is umbilic there with for the upward normal. The cubic term changes the surface away from that point. Knowing a finite amount of data at a point is not, by itself, enough to identify a neighborhood with a piece of a sphere or plane.
Nor should a vanishing principal curvature be confused with vanishing curvature in every direction. A cylinder has a zero principal curvature along each generator and a nonzero one around a circular cross-section. A direction tilted away from the generator generally has nonzero normal curvature, as Euler's formula shows. The sign and magnitude are properties of a direction together with the chosen normal; the existence of the zero direction survives reversing the normal.
Relating asymptotic directions to equations
At a hyperbolic point, use the principal basis and write a nonzero direction as . The zero-normal-curvature equation becomes . If the first eigenvalue is positive and the second negative, it factors into two distinct real linear equations after taking a positive square root. Each equation defines one line through the origin of the tangent plane. This algebra explains why there are two unoriented lines rather than four distinct directions: replacing by describes the same line.
At an elliptic point the same quadratic expression is definite and vanishes only for the zero vector, which is not a direction. At a parabolic point it is a nonzero square times one coefficient, so its zero set is the one-dimensional kernel. These are pointwise conclusions. Constructing an actual asymptotic curve requires solving a differential equation for a direction field and checking regularity and domain; the algebraic classification alone does not produce a global curve.
A normal-curvature comparison
Suppose at a point the principal curvatures are and , and a unit tangent makes angle with the first principal direction. Euler's formula gives normal curvature . At angle zero the curvature is four; at a right angle it is one. The middle direction need not be a principal direction even though its curvature is the arithmetic mean. If instead the curvatures are and , a direction with zero normal curvature satisfies . The two solutions modulo determine two distinct tangent lines.
These examples also show why the orientation convention must be fixed before quoting signs. Reversing the normal turns the first pair into , negating every directional normal curvature. It does not change which directions attain extrema, although the roles of maximum and minimum reverse. For the second pair it also preserves the same zero lines. The geometric directions and the signed values are related but not interchangeable data.
The eigenvalue calculation is pointwise: moving to a neighboring point changes the tangent plane and may change both eigenvalues and eigenvectors. Even when the two principal curvatures are distinct at each point, a consistent choice of oriented eigenvectors over a large region requires additional care. The unoriented principal line is the geometric object used in the preceding classification.
Exercises
Quick check
(A point has principal curvatures and . Find its Gaussian and mean curvatures, its type, and its asymptotic directions relative to a principal basis.)
Solution · Answer
and , so the point is hyperbolic. Euler's formula gives , equivalently , that is ; angles differing by determine the same unoriented line.
Quick check
(A cylinder of radius has principal curvatures and for the outward normal. Find the normal curvature of a unit direction making angle with the circular principal direction.)
Solution · Answer
Euler's formula with along the circle and gives . It vanishes along a generator, where modulo ; it equals along the circular direction.
Further practice
Quick check
(At a point the form matrices in a coordinate basis are and . Find the principal curvatures, and .)
Solution · Answer
The equation gives and . Thus and . The ordinary eigenvalues of would ignore the metric.
Quick check
(A point has . How many asymptotic directions does it have? Is it umbilic?)
Solution · Answer
Euler's formula gives normal curvature in every unit direction, so none is asymptotic. The equal principal curvatures make the point umbilic.