Vector-Valued Functions

Vector-valued functions map a scalar parameter (usually time

t

) to a vector in

R^{n}

, written

r (t) = ⟨ x (t), y (t), z (t)⟩

. They trace curves in space; their calculus (limits, derivatives, integrals) is performed component-wise, yielding velocity, acceleration, arc length, and curvature. They are the foundational language of kinematics, differential geometry of curves, and parametric modeling. Developed through the 19th-century work of Hamilton, Grassmann, and Gibbs.

A vector-valued function (or vector function) is a function whose domain is a set of real numbers and whose range is a set of vectors. In the most common case, $r : I \subseteq R \to R^{n}$ assigns to each parameter value $t$ a position vector, so as $t$ varies the tip of the vector sweeps out a space curve. They matter because nearly every moving object, signal, or geometric path is most naturally described as one vector function of a single variable, unifying the separate scalar equations of motion into one compact object. The calculus of these functions is delightfully simple: differentiation and integration occur independently in each coordinate. A surprising fact is that the derivative of a constant-length vector function is always perpendicular to the function itself — which is exactly why the unit tangent and unit normal vectors are orthogonal, and why circular motion has centripetal acceleration.

The Core Insight

A vector-valued function is one curve carried by several scalar functions. Every analytic operation — limit, derivative, integral, Taylor expansion — passes through the vector and acts on each component separately, because the standard basis vectors $i, j, k$ are constant. The geometry (tangent, normal, curvature) then emerges by reparametrizing and renormalizing these derivatives.

Formal Statement

Vector-Valued Function

A vector-valued function of a real variable is a map

r (t) = ⟨ f (t), g (t), h (t)⟩ = f (t) i + g (t) j + h (t) k,

⟳a vector in

R^{3}

is a linear combination of the constant orthonormal basis

i, j, k

; allowing the coefficients to be functions of

t

produces the vector function.

where $t$ ranges over an interval $I \subseteq R$ (the parameter domain), and $f, g, h : I \to R$ are the component (coordinate) functions. The domain of $r$ is the intersection of the domains of $f$ , $g$ , $h$ . The image ${r (t) : t \in I}$ is a space curve $C$ , and the equations $x = f (t), y = g (t), z = h (t)$ are its parametric equations.

Limit of a Vector Function

The limit is taken component-wise:

t \to a lim r (t) = ⟨ t \to a lim f (t), t \to a lim g (t), t \to a lim h (t) ⟩,

⟳convergence in

R^{n}

under the Euclidean norm is equivalent to convergence in each coordinate, since

∣ x_{i} - L_{i} ∣ \leq ∥ x - L ∥ \leq \sum_{j} ∣ x_{j} - L_{j} ∣

provided each component limit exists. $r$ is continuous at $a$ iff $lim_{t \to a} r (t) = r (a)$ , equivalently iff each component function is continuous at $a$ .

Derivative and Integral

The derivative is the component-wise (and limit) definition

r^{'} (t) = h \to 0 lim \frac{r ( t + h ) - r ( t )}{h} = ⟨ f^{'} (t), g^{'} (t), h^{'} (t)⟩,

⟳substitute the components into the difference quotient; the limit distributes across the three coordinates by equation [2].

and the definite integral is likewise

\int_{a}^{b} r (t) d t = ⟨ \int_{a}^{b} f (t) d t, \int_{a}^{b} g (t) d t, \int_{a}^{b} h (t) d t ⟩ .

⟳a Riemann sum of vectors is the vector of component Riemann sums because vector addition and scalar multiplication act coordinate-wise.

The vector $r^{'} (t)$ , when nonzero, is tangent to the curve and points in the direction of increasing $t$ .

Differentiation Rules

Let $u, v$ be differentiable vector functions, $c$ a scalar, $ϕ (t)$ a differentiable scalar function. Then:

\frac{d}{d t} [u \cdot v] = u^{'} \cdot v + u \cdot v^{'}, \frac{d}{d t} [u \times v] = u^{'} \times v + u \times v^{'} .

⟳write the dot/cross product in components, apply the ordinary scalar product rule term by term, and regroup; for the cross product the factor order must be preserved since it is non-commutative.

\frac{d}{d t} u (ϕ (t)) = ϕ^{'} (t) u^{'} (ϕ (t)) (Chain Rule) .

⟳apply the scalar chain rule to each component of

u \circ ϕ

Constant-Length Orthogonality

If $∥ r (t) ∥$ is constant on an interval, then $r^{'} (t) ⊥ r (t)$ for all $t$ :

∥ r (t) ∥ = c ⟹ r (t) \cdot r^{'} (t) = 0.

⟳differentiate

r \cdot r = c^{2}

using the dot-product rule [5] to get

2 r \cdot r^{'} = 0

Why It Holds / Derivation

Component-Wise Derivative

Goal: show equation [3], that the limit-definition derivative equals the vector of component derivatives.

Write $r (t) = ⟨ f (t), g (t), h (t)⟩$ and form the difference quotient

\frac{r ( t + h ) - r ( t )}{h} = ⟨ \frac{f ( t + h ) - f ( t )}{h}, \frac{g ( t + h ) - g ( t )}{h}, \frac{h ( t + h ) - h ( t )}{h} ⟩ .

This is valid because vector subtraction and scalar division (by $h$ ) are performed coordinate-by-coordinate.
Take the limit $h \to 0$ . By the limit theorem [2], the limit of the vector equals the vector of the component limits.
Each component limit is, by definition, the ordinary derivative of that scalar function: $f^{'} (t)$ , $g^{'} (t)$ , $h^{'} (t)$ .
Therefore $r^{'} (t) = ⟨ f^{'} (t), g^{'} (t), h^{'} (t)⟩$ , proving the components differentiate independently. $■$

Arc Length Formula

Goal: establish that the length of a smooth curve

r (t)

t \in [a, b]

, is

L = \int_{a}^{b} ∥ r^{'} (t) ∥ d t = \int_{a}^{b} f^{'} (t)^{2} + g^{'} (t)^{2} + h^{'} (t)^{2} d t .

⟳approximate the curve by inscribed polygonal chords and pass to the limit using the Mean Value Theorem on each component.

Partition $[a, b]$ into $a = t_{0} < t_{1} < \dots < t_{n} = b$ . The polygonal approximation has length $\sum_{i} ∥ r (t_{i}) - r (t_{i - 1}) ∥$ .
By the Mean Value Theorem applied to each component on $[t_{i - 1}, t_{i}]$ , there exist $t_{i}^{*}$ with $f (t_{i}) - f (t_{i - 1}) = f^{'} (t_{i}^{*}) Δ t_{i}$ , and similarly for $g, h$ .
Hence $∥ r (t_{i}) - r (t_{i - 1}) ∥ \approx f^{'2} + g^{'2} + h^{'2} Δ t_{i}$ (the MVT sample points may differ per component, but they coincide in the limit by continuity).
The sum is a Riemann sum for $\int_{a}^{b} ∥ r^{'} (t) ∥ d t$ ; refining the partition ( $max Δ t_{i} \to 0$ ) gives the integral.
This limit is independent of the parametrization (an orientation-preserving change of parameter leaves $L$ invariant by the substitution rule). $■$

William Rowan Hamilton | 1805–1865

The Irish mathematician Hamilton invented quaternions in 1843 (famously carving the multiplication rule into Brougham Bridge), creating the first algebra in which vectors could be added, multiplied, and differentiated as objects in their own right. He coined the words "scalar" and "vector" and developed the differential operator $\nabla$ (nabla). Although modern vector calculus stripped quaternions down to their 3-dimensional vector part, Hamilton's insistence that a directed quantity could be a single mathematical entity — and that one could take $d r / d t$ — is the direct ancestor of the vector-valued function.

Building Intuition

Picture a particle flying through space. At each instant $t$ its location is a single point, captured by the arrow $r (t)$ from the origin. Run the clock forward and the arrowhead draws a curve. Now zoom in on a tiny time step: the displacement $r (t + h) - r (t)$ is a short chord. Divide by $h$ and shrink $h$ — the chord swings into the tangent line, and its length scales to the speed. That is why $r^{'} (t)$ is simultaneously the velocity (physics) and the tangent vector (geometry).

The orthogonality theorem [7] has a vivid picture: if a point is confined to a sphere of fixed radius (constant $∥ r ∥$ ), it can only move along the surface, never radially — so its velocity must be tangent to the sphere, i.e. perpendicular to the radius. This single idea generates the entire moving frame: the unit tangent $T$ has constant length 1, so $T^{'}$ is perpendicular to it, defining the normal direction $N$ .

A space curve $\mathbf{r}(t)$ with position, velocity (tangent), and the secant chord that becomes the tangent in the limit.

Worked Examples

Velocity, Speed, and Arc Length of a Helix

Given:

r (t) = ⟨ 3 cos t, 3 sin t, 4 t ⟩

t \in [0, 2 π]

(units: meters;

t

in seconds for the kinematic reading). Step 1 — Velocity. Differentiate component-wise (eq. [3]):

r^{'} (t) = ⟨ - 3 sin t, 3 cos t, 4 ⟩ .

Step 2 — Speed. Take the magnitude:

∥ r^{'} (t) ∥ = 9 sin^{2} t + 9 cos^{2} t + 16 = 9 + 16 = 25 = 5.

The speed is constant:

5

m/s. Step 3 — Arc length over one full turn (eq. [8]):

L = \int_{0}^{2 π} 5 d t = 5 (2 π) = 10 π .

Step 4 — Acceleration (second derivative):

r^{''} (t) = ⟨ - 3 cos t, - 3 sin t, 0 ⟩, ∥ r^{''} ∥ = 3.

∥ r^{'} ∥ = 5 m/s, L = 10 π \approx 31.42 m, ∥ r^{''} ∥ = 3 m/s^{2}

Notice that although the speed is constant, the acceleration is nonzero — it changes direction, not magnitude (cf. theorem [7]).

Projectile as a Vector Function

Given: A ball launched from the origin with initial velocity

v_{0} = ⟨ 20, 0, 15 ⟩

m/s under gravity

a = ⟨ 0, 0, - 9.8 ⟩

m/s². Find

r (t)

, the time of flight, and range. Step 1 — Integrate acceleration (eq. [4]) to get velocity, using

v (0) = v_{0}

v (t) = \int a d t = ⟨ 20, 0, 15 - 9.8 t ⟩ .

Step 2 — Integrate velocity to get position, with

r (0) = 0

r (t) = ⟨ 20 t, 0, 15 t - 4.9 t^{2} ⟩ .

Step 3 — Time of flight. Set

z = 0

(other than

t = 0

15 t - 4.9 t^{2} = 0 \Rightarrow t = \frac{15}{4.9} \approx 3.06 s .

Step 4 — Horizontal range. The

x

-coordinate at landing:

x = 20 \times 3.06 \approx 61.2 m .

r (t) = ⟨ 20 t, 0, 15 t - 4.9 t^{2} ⟩, t_{flight} \approx 3.06 s, range \approx 61.2 m

Common Misconceptions

The derivative of the magnitude is NOT the magnitude of the derivative. In general

\frac{d}{d t} ∥ r (t) ∥ \neq = ∥ r^{'} (t) ∥.

The correct rule (from differentiating

r \cdot r

) is

\frac{d}{d t} ∥ r ∥ = \frac{r \cdot r ^{'}}{∥ r ∥}

. The left side measures how the distance from the origin changes (radial speed); the right side measures the total speed. They agree only when motion is purely radial. Other frequent errors:

Forgetting that the cross-product rule [5] is order-sensitive: $u^{'} \times v + u \times v^{'}$ , never $v \times u^{'} + \dots$ .
Treating the constant of integration as a scalar; in eq. [4] the antiderivative constant is a vector $C$ .
Confusing the curve (the image set) with the parametrization (the function). Many functions trace the same curve.

Consider $r (t) = ⟨ cos t, sin t ⟩$ on the unit circle. Here $∥ r (t) ∥ = 1$ is constant, so

\frac{d}{d t} ∥ r (t) ∥ = 0,

yet

r^{'} (t) = ⟨ - sin t, cos t ⟩, ∥ r^{'} (t) ∥ = 1 \neq = 0.

The magnitude of the position never changes, but the velocity is decidedly nonzero — a concrete demonstration that

\frac{d}{d t} ∥ r ∥ \neq = ∥ r^{'} ∥

, and a direct illustration of theorem [7]:

r \cdot r^{'} = - cos t sin t + sin t cos t = 0

Connections

Vector-valued functions sit at a crossroads of several theories.

They are the single-variable special case of Parametric Surfaces and of general maps $R^{m} \to R^{n}$ studied in Multivariable Calculus.
Their derivatives generate the Frenet–Serret frame $(T, N, B)$ studied in Curvature and Torsion and the broader Differential Geometry of Curves.
As models of motion they feed directly into Newtonian Mechanics and Kinematics, where $r, v, a$ are position, velocity, acceleration.
Integrating a vector field along such a curve is the subject of Line Integrals and the Fundamental Theorem of Line Integrals.
The component-wise limit structure is justified rigorously by Convergence in Normed Spaces.

Real-World Applications

Spacecraft & orbital mechanics: A satellite's trajectory is $r (t)$ ; NASA's trajectory software integrates $\ddot{r} = - μ r /∥ r ∥^{3}$ to plan flybys (e.g., Voyager's grand tour).
Computer animation & CGI: Keyframe interpolation uses vector-valued Bézier and Catmull–Rom splines $r (t)$ to move characters and cameras smoothly through 3D scenes (Pixar, game engines).
CNC machining & 3D printing: Tool paths are programmed as parametric curves; the controller samples $r (t)$ and commands stepper motors along $x, y, z$ .
Robotics: A robot arm's end-effector follows a planned trajectory $r (t)$ ; velocities $r^{'} (t)$ feed the inverse-kinematics Jacobian.
GPS & inertial navigation: Position is reconstructed by integrating measured acceleration twice (eq. [4]) — the projectile example generalized to vehicles and aircraft.
Roller-coaster & highway design: Engineers use clothoid (Euler spiral) vector functions so curvature changes linearly, avoiding sudden lateral jerk.
Electromagnetism: The path of a charged particle in a magnetic field, governed by $\ddot{r} = \frac{q}{m} \dot{r} \times B$ , is solved as a vector function (helical motion).

Historical Development

1799–1828 — Caspar Wessel, Jean-Robert Argand, and Carl Friedrich Gauss represent complex numbers as directed plane segments, seeding the idea of an arrow as a computable object.
1843 — William Rowan Hamilton invents quaternions, defining "vector," "scalar," and the operator $\nabla$ ; vectors gain an algebra.
1844 — Hermann Grassmann publishes Die lineale Ausdehnungslehre, an abstract theory of $n$ -dimensional vector spaces and the exterior product — far ahead of its time.
1860s — James Clerk Maxwell uses quaternion notation in his electromagnetic work, exposing physicists to vector methods.
1881–1884 — J. Willard Gibbs distributes Elements of Vector Analysis to students at Yale, extracting the modern dot and cross products from quaternions.
1893 — Oliver Heaviside independently develops streamlined vector calculus for circuit and field theory; he and Gibbs win the "quaternion vs. vector" debate against Tait.
1901 — Edwin Bidwell Wilson, based on Gibbs's lectures, publishes Vector Analysis, the textbook that standardized $i, j, k$ notation and the calculus of $r (t)$ for generations.
1900s–1920s — The Frenet–Serret apparatus (Frenet 1847, Serret 1851) is fully absorbed into differential geometry, formalizing the moving frame of a space curve.
1960s–present — Pierre Bézier and Paul de Casteljau (at Renault and Citroën) develop parametric vector curves for automobile body design, launching computer-aided geometric design.

Subtleties & Limits

Several subtleties separate "a vector function" from "a nice curve," and some questions reach into active research:

Smoothness vs. cusps. A function can be infinitely differentiable yet its image have corners where $r^{'} (t) = 0$ (e.g. the cusp of $r (t) = ⟨ t^{3}, t^{2} ⟩$ ). A curve is called smooth only where $r^{'} \neq = 0$ ; characterizing geometric smoothness from analytic data is genuinely delicate.
Reparametrization invariance. Curvature and arc length are geometric (parametrization-independent), but proving this rigorously for all admissible reparametrizations requires care; the arc-length parametrization exists but is rarely expressible in closed form.
Rectifiability. Not every continuous vector function has finite length; fractal curves (e.g. the Koch curve, the path of Brownian motion) are continuous but nowhere differentiable and have infinite length on every interval — the boundary between rectifiable and non-rectifiable curves is studied in geometric measure theory.
Frame existence. The Frenet frame fails where curvature vanishes; the Bishop (rotation-minimizing) frame repairs this, but optimal frame transport along arbitrary curves remains a topic in computational geometry and graphics research.