This note supplies the existence proof left as an input in the first curve chapter. Let be a nonempty open interval and . Curvature is and strictly positive; torsion is continuous. No unproved linear ODE existence theorem is used below.
Motivation
Curvature and torsion are measured from a curve, but can they be prescribed first? The Frenet equations describe a moving orthonormal frame. Solve those equations from an initial frame, then integrate its tangent vector. The important check is that the solution remains an oriented orthonormal frame; a generic linear system would not preserve lengths or perpendicularity.
The resulting theorem determines a curve up to a proper rigid motion, with the arc-length parameter fixed. It says nothing by itself about closure. Periodic coefficients can produce a frame returning to its start while the curve drifts in space. We will give exact closing conditions and a helix that defeats the naive periodicity inference.
Definitions
Put the Frenet vectors into the columns of and define
The frame equations are , so , , . The coefficient matrix is skew-symmetric, .
Definition 1.2.1
(Oriented orthonormal frame and proper rigid motion)
A matrix whose columns form a frame is an oriented orthonormal frame if , that is and . A proper rigid motion is a map with and .
Reflections are excluded because signed torsion changes under ambient orientation reversal. Matrix norms below are operator norms for the Euclidean norm, so and orthogonal matrices have norm one.
The reconstruction is parametrized by arc length: and .
Definition 1.2.2
(Periodic closed curve)
A curve is a periodic closed curve of period if for all ; its derivatives then agree as well.
Merely joining endpoint positions is a weaker statement.
Theorem/proposition
Theorem 1.2.3
(Continuous linear frame equations have global interval solutions)
For continuous , the equation , , has a unique solution on all of . If is skew-symmetric and , then everywhere.
Theorem 1.2.4
(Fundamental theorem of space curves)
Given with , , an initial point and an oriented initial frame, there is exactly one unit-speed curve with those initial data and curvature/torsion . Any two curves with the same prescribed functions differ by a unique proper rigid motion, with the same parameter .
Theorem 1.2.5
(Two conditions for periodic closure)
Suppose the data are -periodic on . The reconstructed curve is -periodic exactly when
The first closes the frame; the second closes the displacement. Neither periodic coefficients nor the first condition alone implies the second.
Proof sketch or proof idea
Proof
(Picard iteration and the conserved Gram matrix)
We prove linear existence directly. On any compact interval containing , let bound , and let . Start with the constant matrix function and iterate
Induction by integration gives
The factorial denominator comes from integrating each power of . It also works to the left of by reversing the integral and bounding its absolute value. The exponential series makes the iterates uniformly Cauchy on , with a continuous limit . Boundedness of permits passing this uniform limit through the integral, giving . Hence is and solves the equation. No smallness condition on was needed.
If two solutions differ by , then and . Let bound on . Repeating this integral estimate times gives , which tends to zero. Thus uniqueness holds on . Solutions on different compact intervals through agree on overlaps, so their union defines the solution on all of . Coefficients can grow near an excluded endpoint; only boundedness on each compact interval was used.
For skew , differentiate the row Gram matrix:
It equals initially, hence always. A square matrix with is invertible with inverse , so its columns are orthonormal too. Its determinant is always and is continuous, hence retains the initial positive sign. Thus , not its negative. The skew structure supplied precisely the geometric constraint that a general linear equation lacks.
With the frame in hand, the curve is obtained by one more integration.
Proof
(Integrating the tangent and matching initial data)
Define
It is unit speed. Initially the frame is ; since and both factors are , is and is . Moreover with , so the actual Frenet normal is the constructed , and orientation makes the binormal the constructed . Its torsion is . Explicitly,
so the triple-product torsion formula also yields . This verifies the promised regularity and invariants rather than merely producing a formal frame.
For uniqueness with fixed initial data, the actual Frenet frame of any candidate solves the same linear system and hence equals . Integrating its tangent gives the same curve. For different initial data, let be the two frames and choose
Then and solves the same frame equation with the same initial matrix as . Thus , and integration gives . Any proper rigid motion carrying one parametrized curve to the other must carry its tangent, normal and cross-product binormal too, forcing this and then this .
Proof X-ray
Three separate checks between an ODE solution and a curve
The ODE theorem by itself produces only a matrix function. The first check is skew-symmetry: it makes constant, so the columns stay orthonormal, and continuity of keeps . The second is integration: has unit length, so is arc length. The third, and the one easily missed, is : only then is , so the constructed frame is the curve's actual Frenet frame and is its actual torsion. The same identification drives uniqueness, since any candidate's Frenet frame solves the same linear system. Drop any one check and the conclusion about prescribed torsion or rigid-motion uniqueness is unsupported.
The closing conditions use the same uniqueness once more, now comparing the solution with its own translate by one period.
Proof
(Comparing the frame with its shift by one period)
For periodic data, solves the same equation as . If the frames agree at , uniqueness makes them agree everywhere. Then the derivative of is zero, so it is the constant vector . Vanishing displacement gives periodic closure. Conversely a periodic curve has periodic tangent and acceleration; positive curvature then gives a periodic normal and binormal, and the displacement integral is zero. Both directions are proved.
Even when closure fails, periodic coefficients have a precise geometric consequence. Put and . The frame functions and solve the same equation and agree at , so are equal. Integrating their tangents with the chosen yields
Thus one data-period applies a fixed proper rigid motion, rather than necessarily returning the curve to its original position. Repeated periods apply that same affine map repeatedly. This explains the helix's drift without treating it as an isolated exception. To obtain an actual period, the whole frame must return and the remaining translation must vanish, precisely the two stated conditions. The rotation acts on the left of , because it acts in ambient space. Placing it on the right instead would change moving-frame components and would not generally solve the same equation unless the relevant matrices commute. No classification theorem for rigid motions is needed for this conclusion.
Worked examples
Worked example 1.2.6
(Constant invariants give a helix, not automatic closure)
For constants , , put . The curve
has speed one, curvature and torsion , as direct differentiation or the earlier helix formulas verify. Its frame repeats after , but its vertical displacement then is . If the curve never closes. For it is a circle of radius . Constant data are periodic for every period, so coefficient periodicity plainly cannot by itself determine a closed curve.
Worked example 1.2.7
(Planar reconstruction by a turning angle)
For positive curvature and zero torsion, choose , set , and integrate . The left normal satisfies , while is constant. Thus a one-dimensional angle integral already solves the planar case. Taking on gives and integrals of sine/cosine of a quadratic. Existence does not require an elementary antiderivative. Extending that signed-curvature recipe through zero is possible in the plane, but the positive-curvature spatial Frenet theorem does not apply at the inflection.
Worked example 1.2.8
(A finite endpoint with infinitely rotating tangent)
On prescribe , . In complex planar notation take
Then has unit length, its turning rate is , and the curve is planar. The curve approaches zero as , yet its tangent has no limit because makes arbitrarily many turns: along it is one, while along it is minus one. It therefore has no extension through that endpoint. The global-on- theorem is fully satisfied; it did not promise extension through points where the prescribed curvature becomes unbounded.
Worked example 1.2.9
(A prescribed speed changes the frame equation)
Suppose a different parameter has prescribed positive speed and remains in . Reconstruct the unit-speed curve first and set . It is regular , with speed , and the chain rule gives . Omitting this factor would prescribe turning per unit time instead of per unit length. For a radius- circle and , the curve is and has speed while retaining curvature . Thus a change in speed is accommodated by reparametrization, not by altering the geometric invariants.
Common mistakes
Common mistake
(Allowing reflections in the uniqueness statement)
A positive curvature function and continuous torsion determine a parametrized shape up to a proper rigid motion, not an absolute position or starting frame. An arbitrary orthogonal transformation may reflect space and change torsion's sign. Reversing the parameter has a different effect, computed in the exercises.
Counterexample mode
Prescribed torsion is lost where curvature vanishes
Claim: the frame ODE realizes any continuous , even with . Take , and initial frame . The system becomes , , , with solution , , . The frame ODE is solved, but is a straight line: , so the curve has no Frenet normal, and the same line arises for every choice of . The prescribed torsion merely spins the unused around . The hypothesis is what identifies the solved frame with the curve's own.
Common mistake
(Reading global properties into interval existence)
Solving on every compact subinterval is enough for existence throughout the open interval; it is not a guarantee of endpoint smoothness, periodic closure or embeddedness. A reconstructed unit-speed curve may intersect itself.
With two parameters the frame equations become partial differential equations, solvable only under an integrability condition.
A path to generalization
Why surfaces need compatibility equations and curves do not
With two parameters the frame equations become partial differential equations, solvable only under an integrability condition.
A curve's frame depends on one variable, so every continuous coefficient matrix yields a solution. Along a surface parametrized by , an adapted frame satisfies two equations and . Equality of mixed partials, , forces . Written through the first and second fundamental forms, this compatibility is the Gauss and Codazzi equations, and the fundamental theorem of surfaces says that on a simply connected domain compatible data determine a surface up to a proper rigid motion.
Summary
Factorial Picard estimates prove continuous linear frame existence and uniqueness on the whole interval. Skew-symmetry preserves the oriented frame, whose integrated tangent realizes the prescribed curvature and torsion with exact regularity. Initial frames account for proper rigid motions; global closure additionally requires return of both frame and position.
Exercises
Quick check
(For constant curvature and torsion , compute the helix radius, angular rate, frame period and vertical advance per turn.)
Solution · Answer
Here , radius , angular rate five, and frame period . The axial velocity is , so the advance per turn is . The squared speed is . Nonzero advance prevents closure despite constant, hence periodic, invariants.
Quick check
(Prove a positive-curvature unit-speed curve has identically zero torsion if and only if its image lies in an affine plane.)
Solution · Answer
If , , so is constant and , giving an affine plane. Conversely, for a curve lying in a fixed plane, its tangent and positive-curvature normal lie in that plane. The binormal is one of the two fixed unit normals; continuity on the interval prevents switching signs. Hence and . Positive curvature is needed to identify the normal used here.
Quick check
(Why does a general linear frame equation fail to preserve orthonormality? Solve with as a counterexample.)
Solution · Answer
The solution is , verified by differentiation and the initial value. Its columns have length , not one except at zero. The linear coefficient here is , not skew-symmetric. Linearity gives solvability, while skew-symmetry gives the conservation law needed for a geometric frame.
Quick check
(Compute how curvature and torsion change under an orthogonal ambient map of determinant , and under parameter reversal .)
Solution · Answer
Orthogonal maps preserve the cross-product norm in the curvature formula, so curvature stays the same. The torsion numerator's scalar triple product is multiplied by the determinant, while its denominator is unchanged; a reflection therefore reverses torsion. Under reversal, the three derivatives become . The two minus signs cancel in the triple product, so torsion is and curvature . Signed planar curvature instead reverses sign, which must not be confused with spatial torsion.
Quick check
(Two skew frame systems have the same initial frame and point. Suppose between and . Prove the frame error is at most and the curve position error at most .)
Solution · Answer
Let the orthogonal frames be and put . Then and , so . Integration gives . Since , this is the frame bound. Its first-column bound controls ; integrate once more from the common initial point to obtain the quadratic position bound. Orthogonality removes an exponential growth factor that a generic linear-system estimate would introduce.