Motivation
A surface can be studied locally by two coordinates, but distances in the parameter plane need not be distances on the surface. The first fundamental form corrects this distortion. A second quadratic form records the normal part of the surface's bending in space.
Definitions
Tangent vectors and the first form
Definition 2.1.1
(Regular parametrized surface and first fundamental form)
A regular parametrized surface is a map , where is open and are linearly independent at every point. Locally it represents a surface patch; for an embedded surface the map is also required to be a homeomorphism onto its image. Its tangent plane is spanned by , and a local unit normal is
For a tangent vector , the first fundamental form is , where , , and . The Gram determinant is positive. Thus a coordinate curve has speed , and area element . These expressions account for the parametrization; alone is generally not area on the surface.
The three coefficients can also be packaged into a single matrix, and that packaging explains several of their properties at once.
Concept lensAlgebraic
The first form as a Gram matrix
Let be the matrix with columns , and write for the coordinates of . Then
Positive definiteness is immediate: for , because is injective. The identity is Lagrange's identity for the two columns. This matrix is the metric matrix written in the coordinate-change theorem below, where the change of coordinates becomes a congruence .
Theorem/proposition
The second form
Definition 2.1.2
(Second fundamental form)
With the chosen normal, put , , and . Reversing reverses but leaves unchanged.
The definition is justified by the geometric quantity it computes along curves.
Theorem 2.1.3
(The second form gives normal acceleration)
If a unit-speed surface curve has tangent at a point, its normal acceleration there is . Consequently the normal curvature of a nonzero tangent vector depends only on the direction of , not on the speed used to represent it.
Proof
(Differentiate twice and dot with the normal)
By the chain rule , so has components . Differentiating again and using ,
The terms containing and are tangent, so they vanish after dotting with , leaving . Replacing by with multiplies both and by , so their quotient depends only on the direction.
Concept lensGeometric
The second form as height above the tangent plane
Taylor's formula at gives
because the first-order term is tangent. Thus is twice the quadratic part of the signed height of the surface over its tangent plane, measured along . For the cylinder this height is , matching . If is positive definite, the surface lies near on the side toward which points.
Worked examples
Example: a cylinder
Worked example 2.1.4
(A cylinder)
For let on a parameter rectangle of width less than . With outward normal ,
A circle around the cylinder has unit tangent and normal curvature under this outward-normal convention. A vertical generator has normal curvature zero. The cylinder and plane have the same local metric after the substitution , , even though their second forms differ. Consequently an isometry need not preserve extrinsic bending.
Changing coordinates without changing geometry
Theorem 2.1.5
(Changing coordinates without changing geometry)
Suppose and describe the same patch, where is a diffeomorphism between the parameter domains with Jacobian . Then the metric matrices satisfy
Proof
(Chain rule on the coordinate columns)
By the chain rule : the columns of are linear combinations of those of with coefficients from . Since the metric matrix is , we get , and taking determinants gives the second identity.
Consequently , exactly the Jacobian factor in the change-of-variables theorem for integration. This proves that the computed length and area do not depend on the coordinates when the usual integration hypotheses hold. It does not say that the numerical coefficients are unchanged: those coefficients describe the chosen basis of the tangent plane.
The normal selected by a cross product may change sign when . If an orientation is fixed independently, choose the sign consistently before comparing second forms. This is why an orientation-reversing coordinate change does not change intrinsic lengths but can reverse the raw cross-product normal.
Counterexample mode
Swapping coordinates reverses the raw second form
Claim to test: second-form coefficients computed with each parametrization's own cross-product normal are unchanged by every change of coordinates. Take and , so and . At the origin and , so . With this normal, and give , , whereas with its normal gives , . Both first forms are the identity there. The repaired statement: once the normal is fixed independently of the parametrization, the quadratic form on tangent vectors does not depend on coordinates; with raw cross-product normals this holds for , and changes sign when .
Example: a sphere and its coordinate singularities
Worked example 2.1.6
(A sphere and its coordinate singularities)
For , set
Restrict to an open interval of length at most to obtain a patch without the periodic identification. Calculation gives , , , so the area element is . The determinant tends to zero at the excluded poles and vanishes if the formula is extended there, but the sphere itself is regular there: these coordinates fail because . A different local graph chart covers each pole. Integrating the displayed density over , gives ; the omitted poles and seam do not affect the area. The periodic boundary values are not treated as one injective chart.
Example: a graph and its tangent plane
Worked example 2.1.7
(A graph and its tangent plane)
Let for on an open set, with upward normal
Then , , , while
Here denotes the second-form coefficient, to avoid confusing it with the graph function . The area density is . At a critical point of the graph function, the first-form matrix is the identity and the second-form matrix is its Hessian. Away from a critical point the Hessian alone is not the shape operator: the normalization and the inverse metric both matter.
Length, area and normal acceleration are different data
Common mistake
(Normal curvature is not the curvature of every curve in that direction)
A path on a surface has an acceleration in . Its normal component is determined by and its tangent; its tangential component describes intrinsic turning and can vary between paths with the same starting point and tangent. Thus is not generally the full curvature of every surface curve in that direction. For a unit-speed curve, the curvature vector decomposes orthogonally into tangential and normal parts. For a normal section, whose acceleration at the point is normal to the surface, its signed plane curvature equals the normal curvature directly.
Local graphs and the role of regularity
A familiar way to specify a surface is as a level set . Suppose is , vanishes at a point , and its gradient there is nonzero. The implicit function theorem says that, after possibly permuting coordinate axes, the level set near is the graph of a function of two variables. We use that theorem without proof. The gradient is perpendicular to the tangent plane: differentiating along any surface curve through gives .
The nonzero-gradient hypothesis is sufficient for this graph construction; if it fails, one cannot simply conclude either that the level set is singular or that it is regular. For instance, has zero gradient on the plane , although the level set is a perfectly regular plane. The same level set can be defined by , whose gradient is nonzero. The hypothesis concerns the chosen defining function as well as the geometric set.
A parametrization with independent derivatives is locally an immersion. If it is not one-to-one on a large domain, different parameter points can describe the same spatial point. For example has independent derivatives for all but repeats whenever changes by . Restricting to an open interval of length at most produces a regular embedded patch. Independent derivatives control local tangent geometry; they do not on their own make a global parametrization injective.
Computing a surface area from a graph
Worked example 2.1.8
(Area of a paraboloid cap)
Take the paraboloid over the disk , with . The graph formula gives the density . In polar coordinates in the parameter plane, , , the change-of-variables factor is . Hence the area is
The factor comes from changing coordinates on the parameter domain; the square root comes from the metric of the surface. They represent different geometric effects and both are needed. Although polar coordinates are singular at the center of the parameter disk, that single point does not affect the integral. The graph itself remains regular there. As becomes small the area approaches to leading order, consistent with the tangent plane approximating the surface near the origin.
Separating normal and tangential acceleration
Let be a unit-speed curve on a surface with chosen unit normal . At each point its acceleration decomposes uniquely as
The first term is normal and the bracketed term is tangent. The coefficient of the normal term is , determined by the surface, the normal, and the unit tangent . The tangent term depends on how the curve turns within the surface. Both are perpendicular to because differentiating gives . They are also perpendicular to each other, so the square of the curve's ordinary curvature is the sum of the squares of their lengths.
This decomposition explains why knowing one normal curvature does not determine an arbitrary curve's entire curvature. In the plane, the second fundamental form vanishes everywhere, but a circle drawn in the plane still has nonzero ordinary curvature: all its acceleration is tangential to the plane. On the cylinder, a vertical generator has both components zero when parametrized at constant speed, while a circular cross-section has normal acceleration and no tangential acceleration at each point. Both examples use the same definitions without requiring a picture to decide which quantity is being calculated.
The distinction also motivates geodesics. A constant-speed curve is a geodesic when its tangential acceleration vanishes; its normal acceleration may be nonzero because the surface bends in the ambient space. Thus a great circle is intrinsically straight on a sphere although its ordinary curvature in space is nonzero. This is a geometric condition along a curve, whereas the first and second forms are defined on the entire tangent plane at a point.
A regular change of orientation
Suppose a surface patch has normal and a second observer chooses . The first form does not use the normal, so lengths, angles and areas are unchanged. Every coefficient of the second form changes sign. Accordingly the normal acceleration coefficient of a fixed curve changes sign, but the normal acceleration vector does not: multiplying both the coefficient and the unit normal by minus one leaves the vector unchanged. This is a useful check on calculations, because a geometric vector cannot depend on an arbitrary sign convention for its scalar component.
For a nonzero tangent vector , the Cauchy–Schwarz inequality applied in the tangent plane gives . It defines the angle between two nonzero tangent directions by the usual normalized inner product. In coordinates, the expression includes the mixed coefficient . Coordinate lines meet at right angles exactly when at the point, provided both coordinate derivatives are nonzero. Orthogonality at one point does not mean the coordinate lines are perpendicular everywhere in a patch.
Example: a sloping plane
Worked example 2.1.9
(A sloping plane: nonzero without bending)
Let for fixed real . Then , and , while all second derivatives vanish. The plane has zero second form despite a metric matrix that may not be the identity. Its area density is . Over a rectangle , , the area is . Multiplying by alone would give the area of the parameter rectangle, not the sloping surface patch.
The example also warns against reading geometry off the coordinate coefficients.
Common mistake
(An oblique metric is not evidence of curvature)
A nonzero is not evidence of surface curvature. It records that the chosen coordinate curves meet obliquely in the tangent plane. An orthonormal change of tangent basis can diagonalize the metric at a single point, but that choice alone does not make a curved surface flat on a neighborhood. First and second forms answer different questions, and a coordinate simplification should not be mistaken for a geometric conclusion.
Exercises
Quick check
(Compute the area of the patch , on the cylinder of radius , for .)
Use the coefficients found in the cylinder example.
Solution · Answer
Here , , , so and the area is , half the lateral area of a cylinder of height .
Quick check
(For the graph , find at and there with the upward normal.)
Evaluate the graph formulas at a critical point of .
Solution · Answer
At the origin and , so . The upward normal is and , , so ; hence . Equality of the first form with the Euclidean form at one point does not imply the surface is flat nearby.
Further practice
Quick check
(For the graph , find the first-form coefficients at and the area element there.)
Compute and first.
Solution · Answer
Here and , so , , , and the area element is . The point is not a critical point of the graph function, so its coordinate metric is not the identity.
Quick check
(On the cylinder of radius , compute the length of the helical path , for , with constant.)
Insert , into the speed formula.
Solution · Answer
The squared speed is , so the length is . The curve may cross a chosen chart seam, but length can be computed on smaller patches and summed; the cylinder itself has no seam.