Tangent Planes & Linear Approximation — Tangent Planes and Linear Approximations - Calculus 3

SECTION 1

Historical Context & Motivation

The idea of approximating a complicated curve by a straight line tangent to it dates back to antiquity, but the formal extension of this principle to surfaces in three-dimensional space required centuries of mathematical development. In single-variable calculus, the tangent line at a point provides the best local linear approximation to a curve. The natural question that drove mathematicians toward multivariable calculus was whether an analogous construct—a flat plane rather than a line—could serve the same purpose for surfaces defined by functions of two or more variables. This question became increasingly urgent as physics, engineering, and geodesy demanded tools for analyzing phenomena that depend on multiple independent quantities simultaneously.

1684 • Leibniz's Differential Calculus

Gottfried Wilhelm Leibniz publishes his foundational work on differentials, introducing the notation dy = f′(x)dx. His framework of infinitesimally small changes laid the conceptual groundwork for extending differentiation beyond single-variable functions.

1736 • Euler's Multivariable Analysis

Leonhard Euler systematically investigates functions of several variables, developing early notions of partial differentiation. His work on surfaces and their geometric properties foreshadows the concept of tangent planes.

1797 • Lagrange & the Total Differential

Joseph-Louis Lagrange formalizes the total differential for functions of multiple variables, providing the algebraic machinery needed to express tangent planes as linear approximations of surfaces.

1829 • Gauss's Differential Geometry

Carl Friedrich Gauss publishes the Disquisitiones generales circa superficies curvas, rigorously defining tangent planes to arbitrary surfaces and connecting local flatness to curvature via the first and second fundamental forms.

1900s • Modern Multivariable Calculus

The tangent plane becomes a standard tool in advanced calculus courses and differential geometry. Linear approximation in higher dimensions is formalized through the total derivative and the Jacobian matrix framework.

The fundamental question these developments address is deceptively simple: given a surface z = f(x, y), how can we find the unique plane that best approximates the surface near a given point? Answering this question not only generalizes the single-variable tangent line but also opens the door to linearization—a technique that replaces a potentially intractable nonlinear function with a manageable linear one, valid in a neighborhood of the point of tangency. This idea pervades modern science and engineering, from error propagation in experimental measurements to optimization algorithms in machine learning.

SECTION 2

Core Principles & Definitions

Before constructing the tangent plane formula, it is essential to establish the foundational ideas that connect single-variable differentiation to the multivariable setting. The transition from tangent lines to tangent planes rests on the notion that a surface z = f(x, y) can be locally described by its rates of change in each independent direction. Just as the slope of a tangent line captures the instantaneous rate of change of a curve, the partial derivatives f_x and f_y capture the rates of change along the x- and y-directions respectively, and together they determine the orientation of the tangent plane.

Partial Derivatives as Slopes

The partial derivative f_x(a, b) gives the slope of the tangent line to the curve formed by intersecting the surface z = f(x, y) with the plane y = b. Similarly, f_y(a, b) gives the slope along x = a.

Differentiability Implies Tangent Plane

A function f(x, y) is differentiable at (a, b) if it can be well-approximated by a linear function near that point. Differentiability is stronger than merely having partial derivatives; it guarantees the tangent plane exists and is unique.

The Normal Vector

The tangent plane at (a, b, f(a,b)) has a normal vector n = ⟨f_x, f_y, −1⟩. This vector is perpendicular to every direction lying within the tangent plane.

Linear Approximation (Linearization)

The tangent plane equation itself serves as the linearization L(x, y) of f near (a, b). This linear function replaces the surface with a flat plane for computations involving small deviations from the point of tangency.

The Total Differential

The total differential dz = f_x dx + f_y dy estimates the change in z along the tangent plane. It is the multivariable analogue of dy = f′(x) dx.

✦

KEY TAKEAWAY

Think of a tangent plane as a tiny sheet of glass pressed against a curved surface at a single point. Just as a magnifying glass makes a small patch of the Earth look flat, zooming in on a differentiable surface makes it indistinguishable from its tangent plane. The partial derivatives act like the tilt angles of the glass: f_x controls the east–west tilt, and f_y controls the north–south tilt. Together, they uniquely determine the plane's orientation.

SECTION 3

Visual Explanation — The Tangent Plane in 3D

The violet curves represent cross-sections of the surface z = f(x, y). The golden quadrilateral is the tangent plane at the point P = (a, b, f(a, b)). The cyan dashed line shows the tangent line in the x-direction (slope = f_x), the pink dashed line shows the tangent line in the y-direction (slope = f_y), and the green arrow is the normal vector n perpendicular to the tangent plane.

The diagram above illustrates the central geometric idea. At any point P on a smooth surface, two tangent lines can be drawn: one obtained by slicing the surface with a plane parallel to the xz-plane (holding y constant), and the other by slicing with a plane parallel to the yz-plane (holding x constant). These two tangent lines, taken together, span a unique plane—the tangent plane. The normal vector n is obtained by taking the cross product of the direction vectors of these two tangent lines, yielding ⟨f_x, f_y, −1⟩. Notice that near P, the surface and the plane are nearly indistinguishable—this is precisely the regime where the linear approximation is valid.

SECTION 4

Mathematical Framework

We now derive the equation of the tangent plane and the associated linear approximation rigorously. Consider a function z = f(x, y) that is differentiable at the point (a, b). Differentiability at (a, b) means that there exist constants A and B (which turn out to be f_x(a, b) and f_y(a, b)) such that the error in the linear approximation vanishes faster than the distance from (a, b).

TANGENT PLANE EQUATION

z = f(a, b) + fₓ(a, b)(x − a) + f_y(a, b)(y − b)

Here (a, b) is the point of tangency, f(a, b) is the height of the surface at that point, fₓ(a, b) = ∂f/∂x evaluated at (a, b), and f_y(a, b) = ∂f/∂y evaluated at (a, b). This is the equation of a plane passing through the point (a, b, f(a, b)).

LINEAR APPROXIMATION (LINEARIZATION)

L(x, y) = f(a, b) + fₓ(a, b)(x − a) + f_y(a, b)(y − b)

The linearization L(x, y) is numerically identical to the tangent plane equation. When we write f(x, y) ≈ L(x, y) for (x, y) near (a, b), we are using the tangent plane as a surrogate for the surface.

TOTAL DIFFERENTIAL

dz = fₓ(a, b) dx + f_y(a, b) dy

Setting dx = x − a and dy = y − b, the total differential dz gives the change in z along the tangent plane. The actual change is Δz = f(x, y) − f(a, b), and the approximation quality is measured by how closely dz matches Δz.

DIFFERENTIABILITY CONDITION

lim_{(x,y)→(a,b)} [f(x,y) − f(a,b) − fₓ(a,b)(x−a) − f_y(a,b)(y−b)] / √[(x−a)² + (y−b)²] = 0

This is the formal definition of differentiability. It states that the error in the linear approximation goes to zero faster than the distance from (a, b). A sufficient condition for differentiability is that fₓ and f_y exist and are continuous in a neighborhood of (a, b).

💡

Derivation via the Normal Vector

An alternative derivation uses the fact that the surface z = f(x, y) can be rewritten as the level surface F(x, y, z) = f(x, y) − z = 0. The gradient ∇F = ⟨fₓ, f_y, −1⟩ is normal to this level surface. The tangent plane at (a, b, f(a, b)) is then fₓ(a, b)(x − a) + f_y(a, b)(y − b) − (z − f(a, b)) = 0, which rearranges to the tangent plane equation above.

SECTION 5

The Total Differential & Error Estimation

One of the most practical applications of the tangent plane is error estimation via the total differential. When physical measurements have small uncertainties Δx and Δy, the total differential provides a first-order estimate of the resulting uncertainty in f(x, y). In practice, engineers and scientists routinely use dz ≈ Δz to propagate measurement errors through formulas without computing the exact (and often complicated) value of f at the perturbed inputs.

Left: A side view showing the surface (violet curve) and its tangent line (gold). The actual change Δz (red segment) is the vertical distance along the surface, while the differential dz (green segment) is the vertical distance along the tangent line. The two are nearly equal for small Δx.

The diagram illustrates a crucial distinction. The total differential dz is computed using only the partial derivatives at the point of tangency—no re-evaluation of f is needed. The actual change Δz, by contrast, requires knowing f at two points. When the increments Δx and Δy are small, the difference between Δz and dz is negligible, making the differential an efficient and reliable estimation tool. This is especially valuable when f involves complex expressions (square roots, exponentials, trigonometric compositions) that are cumbersome to evaluate by hand at perturbed inputs.

ℹ

Extending to n Variables

For a function w = f(x₁, x₂, …, xₙ), the total differential generalizes to dw = (∂f/∂x₁) dx₁ + (∂f/∂x₂) dx₂ + ⋯ + (∂f/∂xₙ) dxₙ. The tangent 'plane' becomes a tangent hyperplane in ℝⁿ⁺¹. The conceptual framework is identical: partial derivatives determine the tilt in each coordinate direction.

SECTION 6

Worked Example

Let us find the tangent plane and linear approximation for the function f(x, y) = x² + 2xy − y² at the point (1, 2), and then use the linearization to approximate f(1.05, 1.95).

Tangent Plane and Linear Approximation of f(x, y) = x² + 2xy − y²

1
Step 1 — Evaluate f at the Point

Compute f(1, 2) = (1)² + 2(1)(2) − (2)² = 1 + 4 − 4 = 1. So the point of tangency is (1, 2, 1).

f(1, 2) = 1
2
Step 2 — Compute Partial Derivatives

fₓ(x, y) = ∂/∂x [x² + 2xy − y²] = 2x + 2y. Evaluating: fₓ(1, 2) = 2(1) + 2(2) = 6. Similarly, f_y(x, y) = ∂/∂y [x² + 2xy − y²] = 2x − 2y. Evaluating: f_y(1, 2) = 2(1) − 2(2) = −2.

fₓ(1, 2) = 6, $f_y$ (1, 2) = −2
3
Step 3 — Write the Tangent Plane Equation

Substituting into z = f(a, b) + fₓ(a, b)(x − a) + f_y(a, b)(y − b): z = 1 + 6(x − 1) + (−2)(y − 2). Expanding: z = 1 + 6x − 6 − 2y + 4 = 6x − 2y − 1.

Tangent plane: z = 6x − 2y − 1
4
Step 4 — Write the Linearization

The linearization is the same expression viewed as a function: L(x, y) = 1 + 6(x − 1) − 2(y − 2), or equivalently L(x, y) = 6x − 2y − 1. This is our linear approximation of f near (1, 2).

L(x, y) = 6x − 2y − 1
5
Step 5 — Approximate f(1.05, 1.95)

Using the linearization: L(1.05, 1.95) = 6(1.05) − 2(1.95) − 1 = 6.30 − 3.90 − 1 = 1.40. For comparison, the exact value is f(1.05, 1.95) = (1.05)² + 2(1.05)(1.95) − (1.95)² = 1.1025 + 4.095 − 3.8025 = 1.395. The error is |1.40 − 1.395| = 0.005, which is quite small.

L(1.05, 1.95) = 1.40 ≈ f(1.05, 1.95) = 1.395 (error = 0.005)

💡

Checking with the Total Differential

We can verify using dz: with dx = 0.05, dy = −0.05, we get dz = 6(0.05) + (−2)(−0.05) = 0.30 + 0.10 = 0.40. Since f(1,2) = 1, the approximation gives f ≈ 1 + 0.40 = 1.40, consistent with Step 5.

SECTION 7

Strengths, Limitations & Common Pitfalls

Aspect	Strengths	Limitations
Computational Cost	Only requires f, fₓ, and $f_y$ at a single point—no iterative evaluation needed.	Must compute partial derivatives, which can be complex for implicit or composite functions.
Accuracy	Excellent for small displacements from the point of tangency; error is O(‖Δ‖²).	Accuracy degrades rapidly as (x, y) moves far from (a, b); no built-in error bound without second derivatives.
Existence	Guaranteed whenever f is differentiable; sufficient condition is continuity of partial derivatives.	Partial derivatives can exist without f being differentiable (e.g., f(x,y) = xy/√(x²+y²) at origin).
Geometric Intuition	Provides a concrete, visualizable flat surface that captures local behavior of the function.	Cannot capture curvature; misses saddle points, bowls, and ridges unless quadratic terms are included.

Strengths and limitations of tangent plane approximation

✦

KEY TAKEAWAY

Linear approximation is the multivariable analog of using the tangent line in single-variable calculus, and it shares the same fundamental trade-off: simplicity versus range of validity. Think of it like a GPS signal: the flat-Earth approximation built into local GPS calculations works perfectly for navigating a city, but it would produce absurd errors if you tried to use it to plan a transatlantic flight. Similarly, the tangent plane is powerful and elegant within its domain of validity, but it must be used judiciously, with awareness of how far (x, y) is from the point of tangency.

SECTION 8

Connection to Higher-Order Approximation & Advanced Theory

The tangent plane is the first-order term of the multivariable Taylor expansion. Just as single-variable Taylor polynomials can be extended to quadratic, cubic, and higher-degree approximations, the linear approximation L(x, y) can be refined by including second-order terms involving f_xx, f_xy, and f_yy. The second-order Taylor approximation introduces a quadratic surface (paraboloid, hyperboloid, etc.) that captures the curvature the tangent plane misses. This is precisely the structure analyzed in the second derivative test for classifying critical points of multivariable functions.

Feature	Linear Approximation (1st Order)	Quadratic Approximation (2nd Order)
Formula	L = f + fₓΔx + $f_y$ Δy	Q = L + ½[ $f_{xx}$ Δx² + $2f_{xy}$ ΔxΔy + $f_{yy}$ Δy²]
Geometry	Flat plane tangent to surface	Osculating quadric surface (captures curvature)
Error Order	O(‖Δ‖²) — error ∝ square of displacement	O(‖Δ‖³) — error ∝ cube of displacement
Derivatives Needed	fₓ, $f_y$ (2 partial derivatives)	fₓ, $f_y$ , $f_{xx}$ , $f_{xy}$ , $f_{yy}$ (5 values)
Use Cases	Quick estimates, error propagation, Newton's method initialization	Classifying critical points, curvature analysis, tighter approximation

Linear vs. quadratic approximation comparison

Looking forward, the tangent plane concept generalizes further in differential geometry, where the tangent space at a point on a manifold is the vector space of all tangent vectors at that point. In optimization, the gradient ∇f = ⟨fₓ, f_y⟩ extracted from the tangent plane drives gradient descent algorithms, while the Hessian matrix (second derivatives) refines these steps via Newton's method. Understanding the tangent plane is therefore not merely a calculus exercise—it is the entry point to a rich landscape of ideas in analysis, geometry, and computation.

SECTION 9

Practice Problems

Question 1: CONCEPTUAL

Explain why the existence of both partial derivatives fₓ(a, b) and f_y(a, b) at a point is not sufficient to guarantee that the tangent plane exists at that point. What additional condition is needed, and why?

Show answer

The partial derivatives fₓ and f_y only measure the rates of change along the x- and y-coordinate directions. They can exist even when the function behaves badly along other directions (e.g., diagonal approaches). The tangent plane exists when f is differentiable at (a, b), meaning the linear approximation error vanishes faster than the distance from (a, b). A sufficient condition is that the partial derivatives be continuous in a neighborhood of (a, b). Without differentiability, the plane defined by the partial derivatives may not actually be tangent to the surface in every direction.

Question 2: BASIC CALCULATION

Find the equation of the tangent plane to the surface z = e^xy at the point (1, 0).

Show answer

First, f(1, 0) = e⁰ = 1. Next, fₓ = ye^xy, so fₓ(1, 0) = 0 × e⁰ = 0. Also,

f_y

= xe^xy, so

f_y

(1, 0) = 1 × e⁰ = 1. The tangent plane equation is z = 1 + 0(x − 1) + 1(y − 0) = 1 + y. This makes geometric sense: at (1, 0), the surface is flat in the x-direction but rising at rate 1 in the y-direction.

Question 3: INTERMEDIATE

Let f(x, y) = √(x² + y²). Find the linearization L(x, y) at the point (3, 4), and use it to approximate f(3.02, 3.97).

Show answer

f(3, 4) = √(9 + 16) = 5. Computing partials: fₓ = x/√(x² + y²), so fₓ(3, 4) = 3/5 = 0.6.

f_y

= y/√(x² + y²), so

f_y

(3, 4) = 4/5 = 0.8. The linearization is L(x, y) = 5 + 0.6(x − 3) + 0.8(y − 4). Substituting: L(3.02, 3.97) = 5 + 0.6(0.02) + 0.8(−0.03) = 5 + 0.012 − 0.024 = 4.988. The exact value is √(9.1204 + 15.7609) = √24.8813 ≈ 4.9881, so the error is approximately 0.0001, demonstrating excellent accuracy for small displacements.

Question 4: APPLIED

The resistance R in a parallel circuit with two resistors is given by R = R₁R₂/(R₁ + R₂). If R₁ = 100 Ω and R₂ = 200 Ω, each measured with a possible error of ±1 Ω, use the total differential to estimate the maximum error in the computed parallel resistance.

Show answer

First, R(100, 200) = (100)(200)/300 = 200/3 ≈ 66.67 Ω. The partial derivatives are: ∂R/∂R₁ = R₂²/(R₁ + R₂)² = 40000/90000 = 4/9 ≈ 0.444, and ∂R/∂R₂ = R₁²/(R₁ + R₂)² = 10000/90000 = 1/9 ≈ 0.111. The total differential is dR = (4/9)(±1) + (1/9)(±1). For maximum error, both terms add: |dR|_max = 4/9 + 1/9 = 5/9 ≈ 0.556 Ω. This tells us the parallel resistance is approximately 66.67 ± 0.56 Ω, a relative error of about 0.83%.

Question 5: CRITICAL THINKING

Consider f(x, y) = (x² − y²)/(x² + y²) for (x, y) ≠ (0, 0), with f(0, 0) = 0. Show that both fₓ(0, 0) and

f_y

(0, 0) exist and find their values, but explain why f is not differentiable at the origin. What does this imply about the existence of a tangent plane at (0, 0)?

Show answer

By the limit definition: fₓ(0,0) =

\lim_{h→0}

[f(h,0) − f(0,0)]/h =

\lim_{h→0}

(h²/h²)/h =

\lim_{h→0}

1/h, which diverges—so actually fₓ(0,0) does not exist via this computation. Let us reconsider: f(h, 0) = h²/h² = 1 for all h ≠ 0, so fₓ(0,0) =

\lim_{h→0}

(1 − 0)/h, which diverges. This means even the partial derivatives fail to exist in the usual sense. However, if we instead examine g(x,y) = xy/(x² + y²) with g(0,0) = 0, then gₓ(0,0) =

\lim_{h→0}

0/h = 0 and

g_y

(0,0) = 0, but g is not continuous (hence not differentiable) at the origin since g(t,t) = 1/2 ≠ 0. This demonstrates that having partial derivatives equal to zero does not guarantee differentiability, and no tangent plane exists because the surface has a discontinuity that no plane can approximate linearly.

SUMMARY

Summary

The tangent plane to a surface z = f(x, y) at a point (a, b, f(a, b)) is the unique flat surface that best approximates f near (a, b). Its equation, z = f(a, b) + fₓ(a, b)(x − a) + $f_y$ (a, b)(y − b), is determined entirely by the partial derivatives at the point of tangency. This equation simultaneously defines the linearization L(x, y), a linear function that replaces the surface for computations near (a, b).

The total differential dz = fₓ dx + $f_y$ dy provides a practical tool for error estimation, approximating the change in f without re-evaluating the function. The tangent plane exists whenever f is differentiable (a condition stronger than merely having partial derivatives), and it generalizes to the multivariable Taylor expansion when higher-order accuracy is needed. Mastery of tangent planes equips you with both a geometric visualization tool and the algebraic foundation for optimization, sensitivity analysis, and advanced differential geometry.