Computing Partial Derivatives - Calculus 3

SECTION 1

Historical Context & Motivation

The study of functions that depend on more than one variable arose naturally from problems in physics and geometry during the eighteenth century. Quantities like temperature in a room, gravitational potential in space, and the shape of a vibrating membrane all depend on multiple independent coordinates, and scientists needed a calculus capable of handling such dependence. The concept of the partial derivative emerged as the key tool: differentiate with respect to one variable while holding all others fixed, thereby decomposing a complex multivariable relationship into manageable single-variable slices.

The notation and formal machinery of partial differentiation evolved over more than a century, shaped by contributions from several mathematical giants. What began as an ad hoc technique for solving specific physical problems eventually crystallized into a cornerstone of modern analysis, differential geometry, and applied mathematics. Understanding this historical trajectory illuminates why partial derivatives are defined the way they are and why the notation can sometimes seem redundant or overloaded.

1734 • Euler's Early Multivariable Work

Leonhard Euler systematically studied functions of two variables and introduced the idea of differentiating with respect to one variable while treating the other as a constant. His work on fluid mechanics and the calculus of variations made multivariable techniques indispensable.

1788 • Lagrange's Mécanique Analytique

Joseph-Louis Lagrange formalized the use of partial derivatives in classical mechanics, expressing equations of motion in terms of generalized coordinates. His formulation demonstrated that partial differentiation was not merely a trick but a foundational operation in theoretical physics.

1811 • Notation Standardization

Adrien-Marie Legendre introduced the curly-d symbol ∂ to distinguish partial derivatives from ordinary derivatives, though widespread adoption took decades. Carl Gustav Jacob Jacobi later popularized and refined this notation in his work on determinants.

1823 • Cauchy's Rigorous Foundations

Augustin-Louis Cauchy placed limits and continuity on rigorous footing, which in turn gave partial derivatives a precise ε–δ definition. This rigor clarified when partial derivatives exist and when they can be interchanged (a subtlety that Euler had occasionally glossed over).

1900s • Modern Multivariable Calculus

The twentieth century saw partial derivatives become central to general relativity, quantum mechanics, thermodynamics, and machine learning. Today, computing partial derivatives — including automatic differentiation in neural networks — is one of the most performed operations in computational science.

The central question that partial derivatives answer is deceptively simple: how does a multivariable function change when we nudge just one of its inputs? Answering this question precisely requires us to extend the limit-based definition of the derivative to higher dimensions, and doing so opens the door to gradients, directional derivatives, tangent planes, and the rich geometric landscape of multivariable analysis.

SECTION 2

Core Principles & Definitions

Before computing partial derivatives, it is essential to internalize several foundational ideas that distinguish multivariable differentiation from its single-variable counterpart. These principles govern how we set up the calculation, interpret the result, and recognize when the operation is valid.

Hold Other Variables Constant

When computing ∂f/∂x, treat every variable except x as a fixed constant. This reduces the problem to an ordinary single-variable derivative in x, which you already know how to compute from Calculus 1.

Limit Definition Extends Naturally

The partial derivative ∂f/∂x at a point (a, b) is defined as lim(h→0) [f(a+h, b) − f(a, b)] / h. The only difference from Calc 1 is that the other coordinate stays fixed at b.

One Function, Multiple Partial Derivatives

A function f(x, y) has two first-order partial derivatives: ∂f/∂x and ∂f/∂y. A function of n variables has n first-order partials, each obtained by differentiating with respect to one variable while holding the remaining n−1 constant.

All Standard Differentiation Rules Apply

The power rule, product rule, quotient rule, and chain rule all carry over unchanged — you simply apply them to the 'active' variable while treating constants (the held-fixed variables) as you would treat the number 3 or π.

Existence ≠ Continuity

A function can have both partial derivatives at a point yet fail to be continuous there. Continuity of partials (not just existence) is what guarantees differentiability in the multivariable sense, a subtlety without analogue in single-variable calculus.

✦

KEY TAKEAWAY

Think of a partial derivative like adjusting a single knob on a mixing console while leaving all other knobs untouched. The partial derivative ∂f/∂x tells you how the output signal changes in response to that one knob (x), regardless of where the other knobs (y, z, …) are set. In an engineering context, this is precisely how sensitivity analysis works: isolate one input parameter to quantify its individual effect on a system's output.

SECTION 3

Visual Explanation — Slicing a Surface

The geometric interpretation of a partial derivative is both elegant and intuitive. Consider the surface z = f(x, y) in three-dimensional space. To compute ∂f/∂x at a point (a, b), we slice the surface with the vertical plane y = b. This slice produces a curve in the xz-plane, and the partial derivative is simply the slope of the tangent line to that curve at x = a. Likewise, ∂f/∂y is obtained by slicing with the plane x = a and measuring the slope of the resulting curve in the yz-plane. The diagram below illustrates these two orthogonal slices on a paraboloid.

The cyan curve represents the cross-section obtained by fixing y = b, whose tangent slope is ∂f/∂x. The pink curve represents the cross-section obtained by fixing x = a, whose tangent slope is ∂f/∂y. Both tangent lines pass through the yellow point (a, b, f(a,b)) on the surface.

Notice that each partial derivative captures rate of change in only one direction. The x-partial measures how steeply the surface rises or falls as you walk in the x-direction (east–west), while the y-partial measures steepness in the y-direction (north–south). Together, the two partial derivatives at a point determine the tangent plane to the surface at that point, which is the best linear approximation to f near (a, b). This connection between partial derivatives and linear approximation is one of the most powerful ideas in multivariable calculus.

SECTION 4

Mathematical Framework

We now establish the formal definitions and the standard differentiation rules as they apply to partial derivatives. If you are comfortable with single-variable differentiation, you will find that the computational mechanics are almost identical — the only novelty is the bookkeeping of which variables are active and which are held constant.

LIMIT DEFINITION — PARTIAL WITH RESPECT TO x

∂f/∂x (a, b) = lim[h→0] ( f(a + h, b) − f(a, b) ) / h

Here f is a function of two variables, (a, b) is the point of evaluation, and h is the increment in the x-direction only. The variable y remains fixed at b throughout the limit process.

LIMIT DEFINITION — PARTIAL WITH RESPECT TO y

∂f/∂y (a, b) = lim[h→0] ( f(a, b + h) − f(a, b) ) / h

Symmetric to the previous definition: now x stays fixed at a while y is incremented by h.

Notation Conventions

Several notations for partial derivatives coexist in the literature and across textbooks. The Leibniz-style notation ∂f/∂x is the most common in multivariable calculus courses. The subscript notation f_x is compact and widely used, especially when writing second-order partials like f_xy. One also sees D₁f or ∂₁f for the derivative with respect to the first argument. All of these are interchangeable; adopt whichever your course uses and be consistent.

COMMON NOTATION EQUIVALENCES

∂f/∂x = fₓ = D₁f = ∂₁f

All four symbols denote the same object: the partial derivative of f with respect to its first argument.

Key Differentiation Rules Recap

When differentiating with respect to x, treat y (and any other variable) as a constant. Then the standard rules apply without modification. The power rule gives ∂/∂x [xⁿ] = nxⁿ⁻¹. The product rule gives ∂/∂x [u·v] = uₓ·v + u·vₓ where subscript x denotes partial differentiation. The chain rule gives ∂/∂x [g(u(x,y))] = g′(u)·uₓ. Each of these is identical to the Calc 1 version; the only discipline required is remembering that terms involving only y differentiate to zero, just as the derivative of 5 with respect to x is zero.

PARTIAL DERIVATIVE OF A PRODUCT

∂/∂x [u(x,y) · v(x,y)] = (∂u/∂x) · v + u · (∂v/∂x)

This is the standard product rule applied in the x-direction. An analogous formula holds for ∂/∂y.

SECTION 5

Techniques & Common Function Types

While the conceptual framework is uniform — hold all variables constant except the one you are differentiating with respect to — the practical execution varies with the algebraic form of the function. In this section we catalog the most common function types encountered in Calculus 3 and illustrate how partial differentiation proceeds in each case. The diagram below provides a quick-reference flowchart.

Flowchart showing the partial derivative computation process. Start by treating all non-target variables as constants, then identify the function type — polynomial/rational, exponential/logarithmic, or trigonometric — and apply the corresponding Calc 1 differentiation rules.

Polynomials and Rational Functions

Polynomial functions in several variables are the most straightforward to differentiate. Each term is a product of a coefficient, powers of x, and powers of y (and possibly other variables). When computing ∂f/∂x, terms that do not involve x vanish entirely, and for the remaining terms you apply the power rule to the x-factor while carrying the y-factors along as constants. For example, if f(x, y) = 4x³y² − 7xy + y⁵, then ∂f/∂x = 12x²y² − 7y, because 4 × 3 = 12, x³ becomes x², y² and y are constants that stay, and y⁵ disappears.

Exponential and Logarithmic Functions

Functions like f(x, y) = e^x²y or f(x, y) = ln(x² + y²) require the chain rule. Identify the inner function u(x, y), compute ∂u/∂x as if u were a polynomial, and then multiply by the outer derivative. For the exponential, the outer derivative of e^u is itself, so ∂f/∂x = e^u · ∂u/∂x. For the logarithm, the outer derivative of ln(u) is 1/u, giving ∂f/∂x = (1/u) · ∂u/∂x.

Trigonometric Functions

For f(x, y) = sin(xy²), the chain rule again takes center stage. The outer function is sin, whose derivative is cos, and the inner function is u = xy². Since ∂u/∂x = y² (treating y as a constant), we obtain ∂f/∂x = y² cos(xy²). Notice how y² behaves exactly like a numerical coefficient during x-differentiation — this is the recurring theme in partial derivative computation.

SECTION 6

Worked Example — A Multivariable Function

Let us compute both first-order partial derivatives of the function f(x, y) = x²e^3y + sin(xy) step by step. This function combines a product of polynomial and exponential pieces with a trigonometric piece, exercising several rules simultaneously.

Computing ∂f/∂x for f(x, y) = x² $e^{3y}$ + sin(xy)

1
Step 1 — Identify Terms and Hold y Constant

We write f as the sum of two terms: Term A = x²e^3y and Term B = sin(xy). When computing ∂f/∂x, the variable y is treated as a constant. In Term A, e^3y is therefore a constant multiplier. In Term B, the inner function of sin is u = xy, where y acts as a constant coefficient of x.
2
Step 2 — Differentiate Term A

Since e^3y is a constant with respect to x, we apply the power rule to x²: ∂/∂x [x²e^3y] = 2x · e^3y.

∂(Term A)/∂x = 2xe^3y
3
Step 3 — Differentiate Term B via Chain Rule

Let u = xy. Then ∂u/∂x = y (since y is constant). The outer function is sin(u), whose derivative is cos(u). By the chain rule, ∂/∂x [sin(xy)] = cos(xy) · y = y cos(xy).

∂(Term B)/∂x = y cos(xy)
4
Step 4 — Combine

Adding the two results:

∂f/∂x = 2xe^3y + y cos(xy)

Computing ∂f/∂y for the Same Function

1
Step 1 — Hold x Constant

Now x is the constant. In Term A = x²e^3y, the factor x² is constant and e^3y depends on y. In Term B = sin(xy), the inner function is u = xy with ∂u/∂y = x.
2
Step 2 — Differentiate Term A

Apply the chain rule to e^3y: the derivative of e^3y with respect to y is 3e^3y. Multiplied by the constant x², we get 3x²e^3y.

∂(Term A)/∂y = 3x²e^3y
3
Step 3 — Differentiate Term B

With u = xy, we have ∂u/∂y = x. Therefore ∂/∂y [sin(xy)] = cos(xy) · x = x cos(xy).

∂(Term B)/∂y = x cos(xy)
4
Step 4 — Combine

Adding the two results:

∂f/∂y = 3x²e^3y + x cos(xy)

💡

Verification Tip

A useful sanity check is to evaluate your partial derivative at a specific point, say (1, 0). For ∂f/∂x at (1, 0): 2(1)e⁰ + 0·cos(0) = 2. Then compute f(1 + 0.001, 0) − f(1, 0) and divide by 0.001. If you get approximately 2, your algebra is likely correct. This numerical finite-difference check catches sign errors and missing terms.

SECTION 7

Common Pitfalls & Best Practices

Computing partial derivatives is mechanically straightforward once the principle of holding other variables constant is internalized, but several common mistakes trip up students. The table below catalogs the most frequent errors alongside the correct approach and a mnemonic for avoiding them.

Common Mistake	Correct Approach	Why It Matters
Differentiating y-terms when computing ∂f/∂x (e.g., writing ∂/∂x[y³] = 3y²)	∂/∂x[y³] = 0, since y is a constant with respect to x.	Introducing phantom terms corrupts the gradient and any downstream calculation (tangent planes, optimization).
Forgetting the chain rule on composite inner functions (e.g., ∂/∂x[ $e^{xy}$ ] = $e^{xy}$ instead of $ye^{xy}$ )	Identify the inner function u = xy. Then ∂/∂x[ $e^u$ ] = $e^u$ · ∂u/∂x = $ye^{xy}$ .	Missing the chain rule factor is equivalent to dropping a constant multiplier — the result is off by a factor.
Confusing ∂f/∂x with df/dx when f depends on x and y(x) (implicit dependence)	∂f/∂x holds y fixed. df/dx uses the total derivative and includes (∂f/∂y)(dy/dx).	In thermodynamics and constrained optimization, mixing these up leads to incorrect equations of state.
Treating constants like π or e as variables	π and e are numerical constants. Only named variables (x, y, z, t, …) defined in the problem are variables.	Rare but confusing — especially in physics where e might denote charge or Euler's number depending on context.
Misapplying the product rule when only one factor depends on x	If v does not depend on x, then ∂/∂x[u·v] = v · ∂u/∂x. No second term is needed.	Writing an unnecessary zero term is harmless algebraically but signals conceptual confusion that may compound in harder problems.

Common partial derivative errors and corrections

✦

KEY TAKEAWAY

The single most important habit in partial derivative computation is to explicitly identify, before you start differentiating, which symbols are variables and which are constants in the current context. Write a brief note like 'treating y as constant' at the top of your work. This simple annotation prevents the majority of errors and is analogous to declaring variable types in a programming language before performing operations on them.

SECTION 8

Connection to Higher-Order Derivatives & Advanced Theory

First-order partial derivatives are only the beginning of the story. Just as you can differentiate a single-variable function repeatedly to obtain f′, f″, f‴, etc., you can differentiate a multivariable function's partial derivatives to obtain second-order partial derivatives and beyond. For a function of two variables, there are four second-order partials: f_xx, f_yy, f_xy, and f_yx. The famous Clairaut's theorem guarantees that if these mixed partials are continuous, then f_xy = f_yx, a symmetry that simplifies computation considerably.

Concept	Relies on Partial Derivatives	Where You Will Encounter It
Gradient Vector ∇f	The gradient assembles all first-order partial derivatives into a vector: ∇f = ⟨∂f/∂x, ∂f/∂y, …⟩. It points in the direction of steepest ascent.	Directional derivatives, optimization, gradient descent in machine learning.
Tangent Plane & Linearization	The tangent plane to z = f(x,y) at (a,b) is z = f(a,b) + fₓ(a,b)(x−a) + $f_y$ (a,b)(y−b), built directly from partial derivatives.	Local linear approximation, differentials, error estimation.
Hessian Matrix	The matrix of all second-order partial derivatives. For f(x,y), it is a 2×2 matrix with entries f_xx, f_xy, f_yx, f_yy.	Second derivative test for multivariable optimization, convexity analysis.
Jacobian Matrix	For vector-valued functions F: ℝⁿ → ℝᵐ, the Jacobian is the m×n matrix of all partial derivatives ∂Fᵢ/∂xⱼ.	Change of variables in multiple integrals, nonlinear systems, inverse and implicit function theorems.
Partial Differential Equations (PDEs)	PDEs express relationships among a function and its partial derivatives: e.g., the heat equation $u_t$ = k·u_xx.	Physics, engineering, finance (Black–Scholes), fluid dynamics (Navier–Stokes).

How partial derivatives connect to advanced topics in Calculus 3 and beyond

Mastering the computation of first-order partial derivatives is therefore not an end in itself but a prerequisite for essentially every topic that follows in multivariable and vector calculus. The mechanical skill of treating other variables as constants and applying Calc 1 rules will be exercised hundreds of times as you progress through gradients, Lagrange multipliers, surface integrals, and Stokes' theorem. Building speed and accuracy now pays dividends throughout the remainder of the course.

SECTION 9

Practice Problems

Question 1: CONCEPTUAL

Explain in your own words why ∂/∂x [y³ + 5y] = 0. What principle justifies this result, and how does it differ from computing d/dx [x³ + 5x]?

Show answer

When computing the partial derivative with respect to x, every expression that does not contain x is treated as a constant, and the derivative of a constant is zero. The expression y³ + 5y involves only y, which is held fixed during x-differentiation, so each term contributes zero. By contrast, d/dx [x³ + 5x] = 3x² + 5 because x is the variable of differentiation and the power rule applies to both terms. The key principle is that partial differentiation isolates one variable's contribution to the rate of change, effectively projecting the full multivariable behavior onto a single coordinate axis.

Question 2: BASIC CALCULATION

Let f(x, y) = 5x³y² − 2xy⁴ + 7y. Compute ∂f/∂x and ∂f/∂y.

Show answer

For ∂f/∂x, treat y as constant: ∂/∂x[5x³y²] = 15x²y², ∂/∂x[−2xy⁴] = −2y⁴, ∂/∂x[7y] = 0. Therefore ∂f/∂x = 15x²y² − 2y⁴. For ∂f/∂y, treat x as constant: ∂/∂y[5x³y²] = 10x³y, ∂/∂y[−2xy⁴] = −8xy³, ∂/∂y[7y] = 7. Therefore ∂f/∂y = 10x³y − 8xy³ + 7. Each term is handled by the power rule applied to the active variable while the inactive variable's factors remain untouched.

Question 3: INTERMEDIATE

Let g(x, y) = ln(x² + y²). Compute both first-order partial derivatives and then verify that ∂²g/∂x∂y = ∂²g/∂y∂x (Clairaut's theorem).

Show answer

Apply the chain rule with u = x² + y². Then ∂g/∂x = (1/u)(2x) = 2x/(x² + y²) and ∂g/∂y = 2y/(x² + y²). For the mixed partials, compute ∂/∂y[2x/(x² + y²)] using the quotient rule: numerator derivative is 0·(x² + y²) − 2x·(2y) = −4xy, denominator squared is (x² + y²)², giving ∂²g/∂y∂x = −4xy/(x² + y²)². Similarly, ∂/∂x[2y/(x² + y²)] = [0·(x² + y²) − 2y·(2x)]/(x² + y²)² = −4xy/(x² + y²)². The mixed partials are equal, confirming Clairaut's theorem, which is expected since g has continuous second-order partials wherever x² + y² ≠ 0.

Question 4: APPLIED

The temperature at a point (x, y) on a metal plate is modeled by T(x, y) =

100e^{−x²−2y²}

. Find ∂T/∂x and ∂T/∂y. At the point (1, 0), in which coordinate direction is the temperature decreasing faster? By how much?

Show answer

Using the chain rule with u = −x² − 2y²: ∂T/∂x =

100e^{−x²−2y²}

·(−2x) = −

200xe^{−x²−2y²}

and ∂T/∂y =

100e^{−x²−2y²}

·(−4y) = −

400ye^{−x²−2y²}

. At (1, 0): ∂T/∂x|(1,0) = −200(1)

e^{−1}

= −200/e ≈ −73.6, and ∂T/∂y|(1,0) = −400(0)

e^{−1}

= 0. At this particular point, temperature decreases only in the x-direction (at a rate of about 73.6 degrees per unit length); the y-partial is zero, meaning the temperature is momentarily flat in the y-direction. Physically, (1, 0) lies on the y = 0 axis of symmetry of the Gaussian profile, where the y-gradient vanishes.

Question 5: CRITICAL THINKING

Define f(x, y) = xy(x² − y²)/(x² + y²) for (x, y) ≠ (0, 0) and f(0, 0) = 0. Using the limit definition, show that both

f_x

(0, 0) and

f_y

(0, 0) exist and find their values. Then compute f_xy(0, 0) and f_yx(0, 0) and comment on whether Clairaut's theorem applies.

Show answer

By the limit definition,

f_x

(0,0) = lim[h→0] f(h,0)/h = lim[h→0] [h·0·(h²−0)/(h²+0)]/h = 0. Similarly

f_y

(0,0) = lim[h→0] f(0,h)/h = 0. To find f_xy(0,0), first compute

f_y

(x,0) for x ≠ 0 using the formula: f(x,h)/h as h→0 gives

f_y

(x,0) = lim[h→0] xh(x²−h²)/[h(x²+h²)] = x. Then f_xy(0,0) = lim[h→0] [

f_y

(h,0)−

f_y

(0,0)]/h = lim[h→0] h/h = 1. By symmetry,

f_x

(0,y) = −y, so f_yx(0,0) = lim[h→0] [

f_x

(0,h)−

f_x

(0,0)]/h = lim[h→0] −h/h = −1. Here f_xy(0,0) = 1 ≠ −1 = f_yx(0,0), so the mixed partials are not equal. Clairaut's theorem does not apply because the mixed partial derivatives are not continuous at the origin. This classic counterexample underscores that the hypothesis of continuity in Clairaut's theorem is essential, not merely technical.

SUMMARY

Lesson Summary

Summary

Computing partial derivatives extends single-variable differentiation to functions of multiple variables by holding all variables constant except the one being differentiated. The formal definition is a straightforward limit — lim[h→0] [f(a+h, b) − f(a, b)] / h for ∂f/∂x — and in practice you apply the same power rule, product rule, quotient rule, and chain rule from Calculus 1, simply treating the inactive variables as numerical constants. Geometrically, each partial derivative represents the slope of a cross-sectional slice of the surface z = f(x, y) along one coordinate axis.

Key pitfalls include accidentally differentiating constant terms, forgetting the chain rule on composite inner functions, and confusing partial derivatives with total derivatives. First-order partial derivatives serve as the building blocks for the gradient vector, tangent planes, the Hessian matrix, and partial differential equations — making fluency in their computation an essential skill for the remainder of multivariable calculus and its applications across science and engineering.