Work on an open rectangle containing , with smooth data. Read the first/second forms, intrinsic-curvature and curve-ODE reconstruction chapters. The parameter dependence of ODE solutions proved in the exponential-map chapter is used explicitly; no Frobenius or surface-realization theorem is imported.
Motivation
A positive metric tells us lengths and angles; a symmetric second form proposes how the normal should turn. These data cannot be chosen independently. Traversing a tiny coordinate rectangle in two orders must give the same frame, and this requirement produces Gauss and Codazzi.
The converse is more informative than a consistency check. We will solve for the tangent vectors and normal first, verify that their inner products are the prescribed metric, and then integrate the tangent vectors to obtain the position. Compatibility makes the two-dimensional integration possible. The proof is constructive on a rectangle and does not promise a single global realization on every manifold with locally compatible data.
Definitions
Definition 2.7.1
(Christoffel symbols, curvature and the Gauss–Codazzi conditions)
Let be a smooth positive-definite symmetric matrix and smooth symmetric. Indices range over , corresponding to , and repeated indices are summed. Define
Use the earlier curvature convention
For the second form write . The Gauss–Codazzi conditions are
For an immersion , the ordered-coordinate normal is and . Fixing this orientation is part of prescribing .
Definition 2.7.2
(Frame matrix and structure matrices)
Introduce a three-column frame and matrices
Unlike a Frenet frame, the first two columns need not be orthogonal or unit. Their desired Gram matrix is .
Theorem/proposition
Theorem 2.7.3
(Compatibility is a flat matrix equation)
The Gauss–Codazzi conditions are equivalent to
Every smooth immersed surface with ordered-coordinate normal satisfies them.
Flatness is exactly the integrability condition for a frame equation , , which leads to the following analytic statement.
Theorem 2.7.4
(Solving a flat system on a rectangle)
If smooth matrix fields satisfy , then for every initial matrix there is a unique smooth solution
It is proved below using ordinary differential equations only. Applying it with and produces the surface.
Theorem 2.7.5
(Fundamental theorem of local surface realization)
If the given positive metric and symmetric second form satisfy Gauss–Codazzi on , they are the first and second forms of a smooth immersion . An initial point and positively oriented initial frame with Gram matrix determine it uniquely. Any two such realizations differ by a unique proper rigid motion with the same coordinates. Injectivity of the immersion and realization on a nonrectangular global manifold are not asserted.
Proof sketch or proof idea
Proof
(Gauss–Codazzi is flatness of the frame system)
Direct substitution of the Christoffel formula gives metric compatibility in matrix form:
The tangent block is the usual identity . The tangent-normal entries cancel because . Commuting the mixed derivatives of in this identity gives
Let be the upper-left two-by-two block of . Thus is skew-symmetric and is determined by its single entry in position . Expanding the indicated entries of gives
The first uses . If Gauss holds then ; if Codazzi also holds then the lower-left block vanishes. The identity now forces the upper-right block and last diagonal entry to vanish too. Conversely gives the displayed conditions. This proves the scalar Gauss equation suffices in two dimensions, rather than imposing unspecified extra curvature equations.
For a surface, differentiating in its tangent-normal basis gives . Differentiating , gives . Hence . Since is invertible, equality of mixed derivatives forces , proving necessity.
Proof
(Integrate horizontally, then vertically)
To solve the general flat system, first solve , . For each , solve the vertical equation with . The continuous linear ODE existence/uniqueness theorem has been proved by factorial iteration in the preceding curve chapter. Its parameter differentiation, proved earlier from the integral equation, gives
One can apply that result to this nonautonomous equation by adjoining the time coordinate and treating as a constant state parameter. Put . Then
Linear uniqueness gives , so both required equations hold. They also bootstrap from to smoothness by repeated differentiation, since the coefficients are smooth. Any other solution must agree first on and then on every vertical line, proving uniqueness. The cancellation of the flatness term is the whole reason this horizontal-then-vertical construction works.
Proof
(Build the frame, check its Gram matrix, integrate)
Apply it with , . Choose with positive determinant and . Such a matrix exists explicitly: take , and at the initial point.
Let . Along each coordinate line it satisfies . The prescribed satisfies the same linear equations. Equality at the initial point propagates along the horizontal initial line by uniqueness, then along every vertical line, so on the rectangle. The same factorial argument applies to this linear matrix operator, bounded in norm by on compact intervals. Positivity of makes invertible; continuity preserves its positive determinant. Thus are independent, is their oriented unit normal, and their metric is exactly .
Symmetry of and now gives . Define
Then . Differentiating in and using the mixed identity gives . The integral construction is therefore path-independent in the precise way needed; no unproved Poincaré lemma is hidden here. The independent tangent vectors make an immersion, the Gram identity gives its first form, and gives its second form.
For two realizations, their initial frames have the same Gram matrix. Therefore is orthogonal: substitute into . Their positive determinants make it proper. The frames and solve the same system with the same initial value, so agree. Their integrated surfaces consequently differ by the corresponding rotation and a translation. Initial frame and position also force uniqueness of that motion.
Proof X-ray
Where each hypothesis enters the realization
The construction uses three different facts, each once. Gauss–Codazzi enters only through , in the identity : it makes the frame obtained by integrating horizontally and then vertically satisfy the horizontal equation on the whole rectangle. The key move is the Gram identity. Metric compatibility says that and solve the same linear system with the same initial value, so uniqueness forces , and the solved columns have the prescribed inner products rather than arbitrary ones. Symmetry of and in is used only for , which makes the two tangent columns partial derivatives of one position map. The rectangle is used because every vertical line meets the initial horizontal line, so the frame and the position integral are defined everywhere; on a cylinder with a period this last step fails, as the cylinder exercise below shows.
Worked examples
Worked example 2.7.6
(A plane metric can bend into a cylinder)
Take , with constant . Intrinsic curvature and are zero, and the constant data satisfy Codazzi. A realization is
Its unit tangents are and , with normal . Thus and the second form is exactly the proposed one. The cylinder radius is . The same metric with gives a plane: the metric alone does not determine the bending. On a sufficiently wide rectangle the cylinder parametrization repeats, illustrating why the theorem promises immersion rather than injectivity.
Worked example 2.7.7
(Codazzi alone does not make spherical bending compatible with a flat metric)
For , , all derivatives vanish, so Codazzi holds. But while . Gauss fails, so no such surface immersion exists on an open rectangle. Smoothness and symmetry of a proposed second form do not remove its curvature constraint.
Worked example 2.7.8
(Everywhere umbilic data force a plane or sphere)
Suppose . Codazzi and give . Multiplying respectively by and forces both derivatives of to vanish. On a connected rectangle it is a constant . If , the normal is constant and the surface lies in an affine plane. If , the normal equation is , so is constant. Hence the image lies on the sphere of radius about that constant center. Gauss simultaneously gives . This is a proved geometric consequence of compatibility, not an assumption that equal principal curvatures already identify a sphere.
Worked example 2.7.9
(Changing coordinates also changes the ordered normal)
Let be a coordinate change with invertible Jacobian . The new tangent columns are , so the first form is . A second differentiation gives
The additional term is tangent and contributes nothing when dotted with a normal. If the same geometric normal is retained, the second form therefore transforms as . Our convention, however, chooses the normal from the ordered coordinate cross product. That cross product is multiplied by , giving and
Thus reversing coordinate orientation introduces an extra sign when the ordered normal is reset. The determinant ratio remains unchanged because both form determinants acquire in dimension two. Gaussian curvature agrees in either orientation. This calculation explains precisely which convention underlies the proper-motion statement; it does not assert that every surface admits one global continuous normal. For example, simply interchanging swaps the metric rows and columns but also reverses the ordered normal. The second form must receive both the corresponding reordering and this sign. Keeping the original geometric normal instead removes the extra sign, so those two conventions must not be mixed.
Common mistakes
Common mistake
(Flatness of the frame system is not zero Gaussian curvature)
Flatness here concerns the combined tangent-normal system, not vanishing Gaussian curvature. Its extrinsic terms balance the intrinsic curvature through Gauss.
Common mistake
(Gauss and Codazzi cannot replace each other)
Gauss and Codazzi impose different restrictions; the two failed flat examples show neither can be discarded. A symmetric matrix field is not automatically a realizable second form. Likewise, a positive metric does not select its second form uniquely.
Counterexample mode
Gauss alone does not guarantee a surface
Claim to test: if , the data are realizable. For , take , . Both and vanish, so Gauss holds, but the Codazzi equation with reads , or . By the necessity part of the first theorem, these data cannot be realized. The repair is to impose Codazzi as well: changing the dependence to removes the obstruction, and an exercise below constructs the resulting generalized cylinder.
Common mistake
(The second form carries an orientation convention)
The normal is chosen by the ordered coordinates. Reflecting ambient space changes the sign of the corresponding second form, while Gaussian curvature remains unchanged. Proper rigid-motion uniqueness therefore includes an orientation convention.
The rectangle hypothesis is used in the two-stage integration and uniqueness argument. Local data do not automatically descend through global identifications, and the resulting immersion need not be embedded even on a large rectangle.
Summary
Gauss–Codazzi is exactly flatness of the combined tangent-normal matrix system. Ordinary differential equations solve the flat system on a rectangle, metric compatibility preserves its Gram matrix, and tangent symmetry integrates it into a surface. The result realizes the prescribed forms uniquely up to a proper rigid motion while keeping local and global claims distinct.
Exercises
Quick check
(Verify block by block, including the tangent-normal cancellation.)
Treat the tangent block, the tangent-normal entries and the last entry separately.
Solution
The tangent block is from the Christoffel formula. A tangent-normal entry is . The normal-tangent entries are its transposes and also zero; the last entry is zero, matching the derivative of the constant one in . Thus every block agrees, without treating the tangent columns as orthonormal.
Quick check
(Realize , for smooth , allowing zeros and changes of sign. Compute its normal and second form.)
Let the tangent turn by an angle whose derivative is .
Solution
Set and . The tangents are unit and orthogonal, the ordered normal is , and while the other second derivatives vanish. This proves the forms directly. Zeros of do not destroy surface regularity, because the tangent vectors remain independent; the positive-curvature condition of a space-curve Frenet theorem is not needed here.
Quick check
(For , let be the Hessian of a smooth scalar function . Show Codazzi holds automatically and identify the remaining Gauss condition. Does every such work?)
With all Christoffel symbols vanish.
Solution
With zero Christoffel symbols, Codazzi is , precisely equality of mixed third derivatives for a Hessian. Gauss requires . It is not automatic: has Hessian , failing it. Functions of one linear variable, , have rank-at-most-one Hessian and therefore supply compatible examples.
Quick check
(Under an ambient reflection compute the second form defined by the ordered-coordinate normal. Check what happens to Gauss and Codazzi.)
Track how the ordered cross product transforms under a reflection.
Solution
If with , then because cross products acquire the determinant sign. Thus , while is unchanged. In dimension two , so Gauss remains true. Codazzi is linear in , so it remains true as well. Fixing rather than its negative is what selects the proper-motion uniqueness statement.
Quick check
(On the flat cylinder with periodic coordinate of period , propose and . They satisfy all local conditions. Prove there is no global immersion of this cylinder with those forms, using its lift to .)
A flat metric with forces every second derivative of the lift to vanish.
Solution
Lift a hypothetical immersion to a map periodic in . The lifted metric has zero Christoffel symbols and , so the tangent equations give every second derivative of equal to zero. Hence for constant vectors with , . Periodicity would require , impossible. Local compatibility therefore does not eliminate global periods; the rectangle theorem made no such assertion.