Curvature of Space Curves

How sharply does a road bend? Can we measure that sharpness with a single number at every point? This is the motivating question behind the curvature of a space curve. It is one of the oldest and most productive ideas in differential geometry. A curve can twist through three-dimensional space in complicated ways. Yet at each point it has a well-defined "instantaneous circle" that best approximates its bending. The reciprocal of that circle's radius is the curvature. This single scalar quantity captures, locally, everything about how a curve deviates from being a straight line. The idea matters because curvature is the first invariant that survives the passage from a curve's parametrization to its actual geometric shape. Many different parametrizations can trace out the same trace in space. But curvature — properly defined using arc length — depends only on the geometric curve. It does not depend on how fast or in what direction we traverse it. This makes curvature the natural starting point for building further invariants. It is the seed from which the entire local theory of curves grows. The big picture is this: a regular curve parametrized by arc length has unit-speed tangent vectors. So all the information about turning is encoded in how the direction of the tangent changes. Curvature measures exactly the rate of this directional change. This one idea — differentiate the unit tangent, measure the size of the result — unifies three pictures. It links the geometric picture (osculating circles), the analytic picture (a nonnegative function along the curve), and the mechanical picture (centripetal acceleration for unit-speed motion). This article gives the precise definition of curvature via arc-length parametrization. It builds geometric intuition through circles and tangent-vector diagrams. It proves the central theorem relating curvature to the acceleration vector for a general (not necessarily unit-speed) parametrization. It also works through two concrete asymmetric examples: a non-circular helix-like curve and a plane curve with varying curvature. It closes by showing why the theorem needs each hypothesis. Finally, it points ahead — without developing it — to the richer The Frenet-Serret Frame that curvature helps generate.

Key takeaways

Curvature measures how fast the unit tangent vector turns as you move along a curve at unit speed.
It is defined intrinsically via arc-length parametrization, so it depends only on the curve's shape, not its speed.
Geometrically, curvature at a point equals 1/R, where R is the radius of the best-fitting (osculating) circle.
For a curve traversed at any speed, curvature is computable from a formula involving velocity and acceleration, no reparametrization needed.
A straight line has curvature identically zero; a circle of radius R has constant curvature 1/R.
Key idea: curvature is the magnitude of the derivative of the unit tangent with respect to arc length — it quantifies bending, not speed.

Where this lives today

Curvature of space curves is the entry point to the differential geometry used throughout mathematics and its applications. In computer-aided design and computer graphics, curvature governs the smoothness of splines and the visual quality of rendered curves and surfaces; designers explicitly bound curvature to avoid unwanted "kinks" in car bodies or font outlines. In robotics and autonomous vehicle path planning, curvature constraints on planned trajectories translate directly into physical limits on steering angle and speed, since sharper curvature demands tighter turning or lower velocity to avoid excessive lateral acceleration. In physics, for a particle moving along a curve at unit speed, the curvature vector $T'(s)$ is precisely the centripetal (normal) component of acceleration, linking this purely geometric quantity to Newtonian mechanics. In elasticity theory, the bending energy of thin rods (the Kirchhoff rod model) is expressed as an integral of the square of curvature along the rod's centerline, so curvature directly measures stored mechanical energy. Finally, curvature is the essential building block for the The Frenet-Serret Frame, which packages curvature together with torsion to give a complete local classification of space curves up to rigid motion — the natural next stage of the theory built directly on the ideas developed here.

Ask the tutor: How does adding torsion to curvature let us reconstruct an entire space curve, and why does curvature alone suffice only for plane curves?

\alpha : I \to \mathbb{R}^3

Intuition

Picture driving along a road embedded in space. At each instant your car has a heading, recorded as a unit vector $T$ pointing in your direction of travel. If the road is straight, your heading never changes: $T$ is constant, and its derivative is zero. If the road bends — even gently — $T$ rotates as you move, and it rotates faster per unit distance traveled where the bend is sharper. Curvature is precisely the speed of this rotation of $T$ , measured per unit of arc length traveled, so it isolates "how much turning is happening" from "how fast you happen to be going." Because $T(s)$ is always a unit vector, $T(s) \cdot T(s) = 1$ for all $s$ , and differentiating gives $T'(s)\cdot T(s) = 0$ : the derivative of the tangent is always perpendicular to the tangent itself. This is the geometric seed of the normal direction that will later organize into the full Frenet frame. The length $\|T'(s)\|$ , unencumbered by any tangential component, is exactly the curvature. The circle picture makes this concrete. A circle of radius $R$ , traversed at unit speed, has tangent vector rotating through angle $2\pi$ over an arc length of $2\pi R$ ; the rate of rotation (in radians per unit arc length) is $1/R$ . This matches $\kappa = 1/R$ for a circle of radius $R$ , and it explains why curvature is often introduced as "one over the radius of the best-fitting circle" — the osculating circle — at each point of a general curve.

Figure

Tangent vectors turning along a curve of varying curvature, compared to a circle of radius R

\gamma : J \to \mathbb{R}^3

\gamma(t) = \alpha(s(t))

\gamma_1 = \alpha \circ \phi_1

Worked examples

Example 1: A non-circular helix. Let $\gamma(t) = (3\cos t,\ 3\sin t,\ 2t)$ for $t \in \mathbb{R}$ . This is a helix with radius $3$ and "pitch parameter" $2$ (asymmetric numbers chosen deliberately, not the common radius-1 case). Compute derivatives:

\gamma'(t) = (-3\sin t,\ 3\cos t,\ 2), \qquad \gamma''(t) = (-3\cos t,\ -3\sin t,\ 0).

Speed:

\|\gamma'(t)\|^2 = 9\sin^2 t + 9\cos^2 t + 4 = 9 + 4 = 13, \qquad \|\gamma'(t)\| = \sqrt{13}.

Notice the speed is constant even though this is not the arc-length parametrization (arc length would require speed exactly $1$ ; here it's $\sqrt{13}$ , a nontrivial constant). Cross product:

\gamma'(t) \times \gamma''(t) = \begin{vmatrix} \mathbf{i} & \mathbf{j} & \mathbf{k} \\ -3\sin t & 3\cos t & 2 \\ -3\cos t & -3\sin t & 0 \end{vmatrix}.

Expanding:

\mathbf{i}: (3\cos t)(0) - (2)(-3\sin t) = 6\sin t,

\mathbf{j}: -\big[(-3\sin t)(0) - (2)(-3\cos t)\big] = -(6\cos t) = -6\cos t,

\mathbf{k}: (-3\sin t)(-3\sin t) - (3\cos t)(-3\cos t) = 9\sin^2 t + 9\cos^2 t = 9.

So $\gamma'(t) \times \gamma''(t) = (6\sin t,\ -6\cos t,\ 9)$ , and

\|\gamma'(t)\times\gamma''(t)\|^2 = 36\sin^2 t + 36\cos^2 t + 81 = 36 + 81 = 117.

\|\gamma'(t)\times\gamma''(t)\| = \sqrt{117} = 3\sqrt{13}.

Apply the theorem:

\kappa(t) = \frac{3\sqrt{13}}{(\sqrt{13})^3} = \frac{3\sqrt{13}}{13\sqrt{13}} = \frac{3}{13}.

So this helix has constant curvature $3/13$ , independent of $t$ . This matches the known fact that a helix $(a\cos t, a\sin t, bt)$ has curvature $a/(a^2+b^2)$ ; here $a=3, b=2$ gives $3/(9+4) = 3/13$ . ✓ (a good numerical self-check, not assuming the general helix formula in the derivation itself).

Example 2: A plane curve with visibly varying curvature. Let $\gamma(t) = (t, t^3, 0)$ for $t \in \mathbb{R}$ — a twisted cubic collapsed into the $xy$ -plane (chosen asymmetric: no circular symmetry, curvature should genuinely vary and vanish at one point).

\gamma'(t) = (1, 3t^2, 0), \qquad \gamma''(t) = (0, 6t, 0).

Speed:

\|\gamma'(t)\| = \sqrt{1 + 9t^4}.

Cross product (only the $\mathbf{k}$ -component survives, since both vectors lie in the $xy$ -plane):

\gamma'(t) \times \gamma''(t) = (0,0,\ (1)(6t) - (3t^2)(0)) = (0, 0, 6t).

\|\gamma'(t) \times \gamma''(t)\| = 6|t|.

By the theorem:

\kappa(t) = \frac{6|t|}{(1+9t^4)^{3/2}}.

Check specific values. At $t=0$ : $\kappa(0) = 0$ — the curve $y = x^3$ has an inflection point at the origin, where curvature genuinely vanishes, matching the geometric fact that inflection points have zero curvature. At $t = 1$ : $\kappa(1) = \dfrac{6}{(1+9)^{3/2}} = \dfrac{6}{10^{3/2}} = \dfrac{6}{10\sqrt{10}} = \dfrac{3}{5\sqrt{10}} = \dfrac{3\sqrt{10}}{50} \approx 0.1897$ . At $t = 1/3$ : $9t^4 = 9/81 = 1/9$ , so $1 + 9t^4 = 10/9$ , and $(10/9)^{3/2} = \dfrac{10\sqrt{10}}{27}$ . Then

\kappa(1/3) = \frac{6 \cdot (1/3)}{10\sqrt{10}/27} = \frac{2 \cdot 27}{10\sqrt{10}} = \frac{54}{10\sqrt{10}} = \frac{27}{5\sqrt{10}} = \frac{27\sqrt{10}}{50} \approx 1.708.

So the curvature rises from $0$ at the inflection point to a local maximum somewhere between $t=1/3$ and $t=1$ and then decays back toward $0$ as $|t|\to\infty$ (since the denominator grows like $t^6$ while the numerator grows like $t$ ). This is a genuinely non-constant curvature function, illustrating the theorem's full generality beyond circles and helices.

\gamma(t) = (t^3, t^2, 0)

Scope, edge cases & common traps

Situation	What happens	Common error
Straight line $\gamma(t) = p + tv$	$\kappa \equiv 0$ everywhere	Forgetting to check $\gamma'' = 0$ gives numerator $0$ directly
Circle of radius $R$	$\kappa \equiv 1/R$ constant	Confusing radius and diameter, off by factor $2$
Curve with a cusp	Curvature undefined at the cusp	Plugging into the formula and reporting a spurious finite value
Non-arc-length parameter	Must use the $\gamma'\times\gamma''$ formula, not $\\|\alpha''\\|$ directly	Applying $\kappa = \\|\gamma''(t)\\|$ without accounting for speed
Inflection point of a plane curve	$\kappa = 0$ at that point, sign change on either side (in the signed curvature sense)	Assuming curvature is never zero for a "curvy-looking" curve
Reparametrizing with reversed orientation	$\kappa$ (unsigned, as defined here) is unaffected; only the direction convention for later frame vectors changes	Believing curvature can become negative under this definition

The mathematician behind it

Origins · 1 minJean Frédéric Frenet and how this was discovered

Jean Frédéric Frenet

1816–1900 · French mathematician, mid-19th century differential geometry of curves in space

Frenet was a French mathematician and later an academic administrator (rector of the Academy of Lyon), whose 1847 doctoral thesis formalized the systematic study of how a curve's tangent, normal, and binormal vectors evolve along the curve. Working independently and around the same time, Joseph Serret extended and published these ideas in a more complete form in 1851. Together their names attach to the moving frame that organizes curvature and the closely related invariant torsion into a coherent local theory of space curves — the natural next chapter after curvature itself is understood as the rate of turning of the tangent.

History

- 1847 — Jean Frédéric Frenet's doctoral thesis introduces formulas describing the derivatives of the tangent, normal, and binormal vectors of a space curve.

- 1851 — Joseph Alfred Serret independently publishes a more complete and systematic version of these formulas.

- 1874 — The now-standard vector notation and terminology for the moving trihedron (frame) becomes widespread in French differential geometry texts.

- 1909 — Gaston Darboux's treatises on the general theory of surfaces situate curvature of curves within the broader intrinsic and extrinsic geometry of curves lying on surfaces.

- 1940s–1950s — Modern textbook treatments (Struik, later do Carmo) fix the now-standard arc-length-based definitions of curvature used universally today.

Why must we use arc length in the definition of curvature?

Arc-length parametrization guarantees unit speed, $\|\alpha'(s)\| = 1$ , so any change in the tangent vector $\alpha'(s)$ reflects pure turning rather than a mix of turning and speeding up or slowing down. If we differentiated a curve traversed at varying speed, $\gamma''(t)$ would combine a tangential acceleration term (change in speed) with the true bending information, contaminating the measurement. Using arc length isolates the geometric bending completely, making curvature a property of the trace alone — independent of how it is traced. The general formula with $\gamma'\times\gamma''$ divided by $\|\gamma'\|^3$ is precisely the bookkeeping device that removes the speed-dependence for us, without literally requiring a reparametrization.

Can curvature ever be negative under this definition?

No. As defined here, curvature is $\kappa(s) = \|T'(s)\|$ , a Euclidean norm, and norms are always nonnegative by definition. This is sometimes called unsigned curvature. For plane curves specifically, one can define a signed curvature that can be positive or negative depending on whether the curve bends left or right relative to a chosen orientation, but that is a distinct, plane-curve-specific refinement built on top of this unsigned quantity, not a contradiction of it.

What does it mean geometrically for curvature to equal zero at a point?

Zero curvature at a point means the unit tangent's derivative vanishes there, so to second order the curve is momentarily behaving like a straight line — it has no instantaneous bending. This happens at every point of an actual straight line, but it can also happen at isolated points of a curved path, called inflection points, such as $t=0$ for $\gamma(t) = (t,t^3,0)$ worked out above. At such points the "osculating circle" degenerates to a circle of infinite radius, i.e., a straight line, consistent with $\kappa = 1/R \to 0$ as $R \to \infty$ .

How is curvature related to the osculating circle precisely?

At a point where $\kappa(s_0) \neq 0$ , the osculating circle is the unique circle that matches the curve's position, tangent direction, and rate of turning at that point — it lies in the plane spanned by $T(s_0)$ and $T'(s_0)$ , has radius $R = 1/\kappa(s_0)$ , and its center lies at $\alpha(s_0) + R\, N(s_0)$ , where $N(s_0) = T'(s_0)/\|T'(s_0)\|$ is the unit normal direction. This circle is the best possible circular approximation to the curve to second order near $s_0$ , which is exactly why $1/\kappa$ is interpreted as a "radius of curvature."

Does curvature change if I traverse the curve backward?

No, not under this unsigned definition. Reversing orientation replaces the parameter $s$ by $-s$ (or more generally composes with an orientation-reversing reparametrization), which flips the sign of the tangent vector $T$ but not the magnitude of its derivative's size relative to arc length; one can check directly that $\|T'(s)\|$ is unchanged because reversing $s\to -s$ introduces two sign flips (one from differentiating $T = -T_{\text{original}}$ and one from $d/d(-s) = -d/ds$ ) that cancel. Curvature genuinely only depends on the trace and its bending, not on direction of travel.

Is curvature enough to determine a space curve completely?

Not by itself. Curvature alone determines the shape of a plane curve up to rigid motion (via the fundamental theorem of plane curves), but a curve in three-dimensional space can bend within a plane exactly as prescribed by $\kappa$ while also twisting out of that plane in infinitely many ways. To pin down a space curve completely (up to rigid motion), one needs a second invariant called torsion, which measures the failure of the curve to stay in a single plane; together curvature and torsion form the data behind the Frenet–Serret theory.

Why does the formula divide by ‖γ′(t)‖ cubed, and not some other power?

Tracking units and scaling makes this transparent. If you rescale the parameter so speed is multiplied by a constant $c$ (replace $t$ by $t/c$ ), then $\gamma'$ scales by $c$ and $\gamma''$ scales by $c^2$ , so $\gamma' \times \gamma''$ scales by $c^3$ . To get a quantity that is invariant under this rescaling (since curvature should not depend on how fast you move), you must divide by something scaling as $c^3$ as well, and $\|\gamma'\|^3$ scales exactly as $c^3$ . The proof makes this precise: the factor $(s'(t))^3$ appears explicitly in Step 2 and is exactly canceled in Step 5.