Directional Derivatives via Gradient — Directional Derivatives via the Gradient - Calculus 3

SECTION 1

Historical Context & Motivation

The study of how functions change has driven mathematics since the invention of calculus in the seventeenth century. For single-variable functions, the derivative captures the rate of change along the one available direction—the real line. However, once mathematicians and physicists began investigating surfaces, temperature fields, and potential functions defined over two or three variables, a richer question emerged: how does a function change when we move in an arbitrary direction through its domain? Answering that question required new tools—partial derivatives, the gradient vector, and ultimately the directional derivative.

1744 • Euler's Partial Derivatives

Leonhard Euler formalized partial differentiation, enabling the study of functions of several variables and laying the algebraic groundwork for measuring change along coordinate axes independently.

1813 • Cauchy Introduces the Gradient Concept

Augustin-Louis Cauchy explored rates of change in multiple directions while studying potential theory, contributing early ideas about combining partial derivatives into a single vector quantity.

1846 • Hamilton's Nabla Operator

William Rowan Hamilton introduced the ∇ (nabla) operator and systematized the gradient as a vector field, unifying the partial derivatives of a scalar function into one compact symbol.

1871 • Maxwell & Physical Gradients

James Clerk Maxwell's electromagnetic theory made heavy use of the gradient, showing that the electric field is the negative gradient of the electric potential—a directional derivative in every direction simultaneously.

1900s • Modern Multivariable Calculus

Directional derivatives became a standard topic in analysis and applied mathematics, essential in optimization, fluid dynamics, machine learning, and any field requiring sensitivity analysis along non-axis directions.

The central question this lesson addresses is elegant in its simplicity: if you stand on a surface described by z = f(x, y) and begin walking in some direction u, how steep is the terrain beneath your feet? Partial derivatives answer this only along the coordinate axes. The gradient, combined with a unit direction vector, answers it for every possible direction.

SECTION 2

Core Principles & Definitions

Before we can compute directional derivatives via the gradient, we need to establish the foundational concepts that connect them. The directional derivative generalizes the notion of a partial derivative to arbitrary directions, while the gradient packages all partial derivative information into a single vector. The bridge between them is the dot product, which projects the gradient onto the chosen direction.

Partial Derivatives

The partial derivative f_x measures the rate of change of f in the x-direction while holding y constant. Similarly, f_y captures change in the y-direction. These are directional derivatives along the standard basis vectors.

The Gradient Vector

The gradient of a scalar function f is the vector ∇f = ⟨f_x, f_y⟩. It points in the direction of steepest ascent and its magnitude equals the maximum rate of change at that point.

Unit Direction Vector

A direction in ℝ² is specified by a unit vector u = ⟨cos θ, sin θ⟩ where θ is the angle measured from the positive x-axis. The vector must satisfy ‖u‖ = 1 so the derivative measures pure rate of change, not also speed.

Dot Product as Projection

The dot product ∇f · u computes the scalar projection of the gradient onto the direction u. Geometrically, this extracts how much of the gradient 'acts' along u, yielding the directional derivative D_uf.

Differentiability Requirement

The gradient formula for directional derivatives requires that f be differentiable at the point of interest—not merely that partial derivatives exist. Differentiability guarantees that the tangent plane exists and approximates f well locally.

✦

KEY TAKEAWAY

Think of the gradient as a compass needle that always points uphill on a terrain map. The directional derivative in any direction u is simply how much of that uphill pull aligns with u. If you walk directly uphill (along ∇f), you experience the steepest climb. If you walk perpendicular to ∇f (along a level curve), the terrain is flat—a zero directional derivative. Walking at an angle gives you something in between, proportional to cos θ, where θ is the angle between ∇f and u.

SECTION 3

Visual Explanation — The Gradient and Directional Derivatives

The diagram shows level curves of a function f with the point P at the center of interest. The gradient vector ∇f (violet) points perpendicular to the level curves toward steepest ascent. The unit direction vector u (cyan) indicates the desired direction. The directional derivative

D_u

f (green dashed) is the scalar projection of ∇f onto u.

In the diagram above, observe that the gradient ∇f is always perpendicular to the level curves, pointing toward increasing values of f. The directional derivative in the direction of u is determined entirely by the angle θ between ∇f and u. When θ = 0, the directional derivative reaches its maximum value of ‖∇f‖ because you are walking directly uphill. When θ = π/2, the directional derivative is zero—you are walking along a level curve where f does not change. When θ = π, the directional derivative equals −‖∇f‖, the steepest descent. This cosine dependence, D_uf = ‖∇f‖ cos θ, is a direct consequence of the dot product formula ∇f · u.

SECTION 4

Mathematical Framework

We now formalize the relationship between the gradient and the directional derivative. Let f : ℝ² → ℝ be differentiable at a point (a, b), and let u = ⟨u₁, u₂⟩ be a unit vector. The directional derivative of f at (a, b) in the direction of u is defined by a limit, but when f is differentiable the computation reduces to a simple dot product.

LIMIT DEFINITION

D_u f(a, b) = lim_{h→0} [f(a + h u₁, b + h u₂) − f(a, b)] / h

Here h is a scalar step size along the direction u = ⟨u₁, u₂⟩. This limit measures the instantaneous rate of change of f as we move from (a, b) along u.

GRADIENT FORMULA (THE KEY RESULT)

D_u f(a, b) = ∇f(a, b) · u = f_x(a, b) u₁ + f_y(a, b) u₂

∇f(a, b) = ⟨f_x(a, b), f_y(a, b)⟩ is the gradient vector; u = ⟨u₁, u₂⟩ is a unit vector with ‖u‖ = 1. This formula holds whenever f is differentiable at (a, b).

GEOMETRIC FORM

D_u f = ‖∇f‖ cos θ

θ is the angle between ∇f and u. Maximum rate of change = ‖∇f‖ (when θ = 0). Zero change when θ = π/2 (along level curves). Maximum decrease = −‖∇f‖ (when θ = π).

THREE-VARIABLE EXTENSION

D_u f(a, b, c) = ∇f · u = f_x u₁ + f_y u₂ + f_z u₃

For f : ℝ³ → ℝ, the gradient becomes ∇f = ⟨f_x, f_y, f_z⟩ and u = ⟨u₁, u₂, u₃⟩ must satisfy u₁² + u₂² + u₃² = 1.

⚠

Common Pitfall

The direction vector must be a unit vector for the gradient formula to yield the correct directional derivative. If you are given a non-unit vector v, first normalize it by computing u = v / ‖v‖ before taking the dot product with ∇f. Forgetting this normalization step is one of the most frequent errors in multivariable calculus.

SECTION 5

Geometric Interpretation — The Gradient and Level Curves

The geometric relationship between the gradient and level curves is one of the most illuminating results in multivariable calculus. At every point where ∇f ≠ 0, the gradient is orthogonal to the level curve passing through that point. This orthogonality is not a coincidence—it follows directly from the chain rule applied to a parameterization of the level curve. Along a level curve, f is constant, so its total differential is zero, which forces ∇f to be perpendicular to the tangent vector of the curve.

The cosine curve shows how the directional derivative varies with the angle θ between the gradient and the unit direction vector. The maximum occurs at θ = 0 (along ∇f), the zero crossings at θ = π/2 and 3π/2 (along level curves), and the minimum at θ = π (steepest descent).

Angle θ	Direction relative to ∇f	$D_u$ f	Interpretation
0	Same as ∇f	‖∇f‖	Maximum rate of increase (steepest ascent)
π/4	45° from ∇f	‖∇f‖ × (√2/2)	About 70.7% of maximum rate
π/2	Perpendicular to ∇f	0	Along level curve; no change in f
π	Opposite to ∇f	−‖∇f‖	Maximum rate of decrease (steepest descent)

Directional derivative values at key angles

SECTION 6

Worked Example

Finding the Directional Derivative of f(x, y) = x²y − 3xy + y² at (2, 1) in the Direction of v = ⟨3, 4⟩

1
Step 1 — Compute the Partial Derivatives

We need f_x and f_y. Differentiating f(x, y) = x²y − 3xy + y² with respect to x: f_x = 2xy − 3y. Differentiating with respect to y: f_y = x² − 3x + 2y.

f_x = 2xy − 3y, f_y = x² − 3x + 2y
2
Step 2 — Evaluate the Gradient at (2, 1)

Substitute x = 2 and y = 1 into the partial derivatives. f_x(2, 1) = 2(2)(1) − 3(1) = 4 − 3 = 1. f_y(2, 1) = (2)² − 3(2) + 2(1) = 4 − 6 + 2 = 0.

∇f(2, 1) = ⟨1, 0⟩
3
Step 3 — Normalize the Direction Vector

The given direction is v = ⟨3, 4⟩, which is not a unit vector. Compute its magnitude: ‖v‖ = √(3² + 4²) = √(9 + 16) = √25 = 5. The unit vector is u = v / ‖v‖ = ⟨3/5, 4/5⟩.

u = ⟨3/5, 4/5⟩ = ⟨0.6, 0.8⟩
4
Step 4 — Compute the Directional Derivative via the Dot Product

Apply the gradient formula: D_uf(2, 1) = ∇f(2, 1) · u = ⟨1, 0⟩ · ⟨3/5, 4/5⟩ = (1)(3/5) + (0)(4/5) = 3/5.

$D_u$ f(2, 1) = 3/5 = 0.6
5
Step 5 — Interpret the Result

The directional derivative is 3/5, meaning that at the point (2, 1), the function f increases at a rate of 0.6 units per unit step in the direction of ⟨3, 4⟩. Since the gradient at this point is ⟨1, 0⟩ (pointing purely in the x-direction), and our direction has a positive x-component, the positive result makes geometric sense. The maximum possible rate of change at (2, 1) is ‖∇f‖ = 1, achieved by moving in the direction ⟨1, 0⟩.

SECTION 7

Comparisons — Partial vs. Directional Derivatives

It is instructive to compare the directional derivative with its simpler cousins—partial derivatives—and to note both the strengths and limitations of the gradient-based formula. Understanding when the formula applies (and when it does not) is essential for correct usage in applications.

Feature	Partial Derivative	Directional Derivative via Gradient
Direction	Only along coordinate axes (x or y)	Any unit vector u in the domain
Formula	$f_x$ or $f_y$ (single derivative)	∇f · u (dot product)
Output	Scalar (rate of change along axis)	Scalar (rate of change along u)
Requirement	Partial derivative exists at the point	f must be differentiable at the point
Special case	$f_x$ = D_{⟨1,0⟩} f, $f_y$ = D_{⟨0,1⟩} f	Reduces to partial derivatives when u is a standard basis vector
Limitation	Cannot capture off-axis behavior	Requires differentiability, not just existence of partials

Partial derivatives vs. directional derivatives

✦

KEY TAKEAWAY

Think of partial derivatives as measuring the slope of a terrain when you walk due north or due east. The directional derivative via the gradient lets you measure the slope in any compass direction—and the gradient is your master slope gauge, encoding the steepness and heading of the steepest path in a single vector. In engineering optimization, this is precisely why gradient descent works: at each step you move opposite to the gradient to decrease a cost function as rapidly as possible.

SECTION 8

Connection to Advanced Theory

The directional derivative and the gradient are not endpoints of multivariable calculus—they are gateways to deeper structures. In higher dimensions, the gradient generalizes to the total derivative (or Jacobian), and directional derivatives become essential in functional analysis and differential geometry. Here we briefly survey the most important connections that a student of Calculus 3 will encounter next.

Concept in This Lesson	Advanced Extension	Where You'll See It
∇f (gradient of scalar)	Jacobian matrix J for vector-valued functions F : ℝⁿ → ℝᵐ	Linear algebra, differential equations, robotics
$D_u$ f = ∇f · u	Fréchet derivative Df(a)(h) for functions on Banach spaces	Functional analysis, PDE theory
‖∇f‖ = max rate of change	Gradient descent / ascent algorithms	Machine learning, optimization, neural networks
∇f ⊥ level curves	Normal vectors to surfaces, tangent planes, implicit function theorem	Differential geometry, constrained optimization (Lagrange multipliers)
Directional derivative along curves	Lie derivatives, covariant derivatives on manifolds	General relativity, advanced differential geometry

From directional derivatives to advanced mathematics

Perhaps the most immediate application students encounter after this lesson is Lagrange multipliers. When optimizing f subject to a constraint g(x, y) = c, the key insight is that at an optimum, the gradient ∇f must be parallel to ∇g—any directional derivative of f along the constraint curve must vanish, which happens precisely when ∇f is perpendicular to the constraint curve. This perpendicularity condition is nothing more than requiring D_uf = 0 for every tangent direction u of the constraint.

SECTION 9

Practice Problems

Question 1: CONCEPTUAL

Explain why the directional derivative of a differentiable function f at a point P in the direction perpendicular to ∇f(P) is always zero. What does this tell us about the geometry of level curves?

Show answer

By the gradient formula, D_uf = ∇f · u = ‖∇f‖ cos θ. When u is perpendicular to ∇f, θ = π/2 and cos(π/2) = 0, so D_uf = 0. Geometrically, this means that moving perpendicular to the gradient keeps f constant—you are traveling along a level curve. This confirms the fundamental result that the gradient is always orthogonal to level curves of f.

Question 2: BASIC CALCULATION

Let f(x, y) = 3x² + y². Find the directional derivative of f at the point (1, 2) in the direction of the vector v = ⟨1, 1⟩.

Show answer

First compute the gradient: f_x = 6x, f_y = 2y, so ∇f(1, 2) = ⟨6, 4⟩. Next, normalize v: ‖v‖ = √(1² + 1²) = √2, so u = ⟨1/√2, 1/√2⟩. Then D_uf(1, 2) = ⟨6, 4⟩ · ⟨1/√2, 1/√2⟩ = 6/√2 + 4/√2 = 10/√2 = 5√2 ≈ 7.07. The function increases at a rate of 5√2 per unit distance in the direction ⟨1, 1⟩.

Question 3: INTERMEDIATE

Let f(x, y) = eˣ sin y. At the point (0, π/3), find the direction in which f increases most rapidly and the maximum rate of increase.

Show answer

Compute the partials: f_x = eˣ sin y and f_y = eˣ cos y. At (0, π/3): f_x = e⁰ sin(π/3) = √3/2 and f_y = e⁰ cos(π/3) = 1/2. So ∇f(0, π/3) = ⟨√3/2, 1/2⟩. The direction of maximum increase is along the gradient itself, i.e., the unit vector u = ∇f / ‖∇f‖. The maximum rate is ‖∇f‖ = √((√3/2)² + (1/2)²) = √(3/4 + 1/4) = √1 = 1. The function increases most rapidly in the direction ⟨√3/2, 1/2⟩ at a rate of 1 per unit distance.

Question 4: APPLIED

The temperature in a room is modeled by T(x, y, z) = 100 − x² − 2y² − z² (in °C), where distances are in meters. A sensor is located at the point (1, 1, 2). In which direction should you move the sensor to experience the most rapid temperature increase, and what is that rate?

Show answer

Compute the gradient: T_x = −2x, T_y = −4y, T_z = −2z. At (1, 1, 2): ∇T = ⟨−2, −4, −4⟩. The direction of fastest temperature increase is the unit vector in the direction of ∇T: ‖∇T‖ = √(4 + 16 + 16) = √36 = 6, so u = ⟨−2/6, −4/6, −4/6⟩ = ⟨−1/3, −2/3, −2/3⟩. Move toward the origin (negative x, y, and z) to increase temperature most rapidly. The maximum rate of temperature increase is 6 °C per meter.

Question 5: CRITICAL THINKING

Suppose f : ℝ² → ℝ has the property that D_uf(a, b) = 3 in the direction u = ⟨1, 0⟩ and D_vf(a, b) = −1 in the direction v = ⟨0, 1⟩. Assuming f is differentiable at (a, b), determine ∇f(a, b) and then find the directional derivative of f at (a, b) in the direction making a 60° angle with the positive x-axis.

Show answer

The directional derivative along ⟨1, 0⟩ is just f_x(a, b) = 3, and along ⟨0, 1⟩ is f_y(a, b) = −1. Therefore ∇f(a, b) = ⟨3, −1⟩. For the direction at 60° from the x-axis, u = ⟨cos 60°, sin 60°⟩ = ⟨1/2, √3/2⟩. Then D_uf = ⟨3, −1⟩ · ⟨1/2, √3/2⟩ = 3/2 − √3/2 = (3 − √3)/2 ≈ 0.634. This problem illustrates a powerful idea: knowing directional derivatives along two independent directions (especially the coordinate directions) is sufficient to reconstruct the entire gradient, and hence the directional derivative in any direction.

SUMMARY

Lesson Summary

Summary

The directional derivative D_uf measures the instantaneous rate of change of a scalar function f in the direction of a unit vector u. When f is differentiable, this derivative can be computed elegantly via the gradient formula: D_uf = ∇f · u = f_x u₁ + f_y u₂. The gradient vector ∇f encodes all directional information: it points in the direction of steepest ascent, its magnitude ‖∇f‖ gives the maximum rate of change, and it is always perpendicular to the level curves of f.

The geometric form D_uf = ‖∇f‖ cos θ reveals that the directional derivative is maximized when u aligns with ∇f (θ = 0), vanishes when u is tangent to a level curve (θ = π/2), and is most negative when u opposes ∇f (θ = π). Always remember to normalize your direction vector before computing the dot product. These ideas extend naturally to three and higher dimensions and form the foundation for gradient-based optimization, Lagrange multipliers, and the analysis of vector fields throughout applied mathematics and the sciences.