Chapter 6 · Introduction

Chapter 6 — Functions of Several Variables

Calculus in $\mathbb{R}^n$: limits along every path, partial and directional derivatives, the total derivative as a linear map, the inverse and implicit function theorems, and the search for maxima and minima with and without constraints.

Most quantities in science depend on more than one variable: the temperature of a plate depends on two coordinates, the pressure of a gas on volume and temperature, the loss of a neural network on millions of weights. For two centuries mathematicians differentiated such functions "one variable at a time" and assumed that nothing new could happen. Then the examples arrived. In 1873 Hermann Schwarz worked out exactly when the order of two partial differentiations may be swapped, and a decade later Peano published a function whose two mixed partial derivatives at the origin are $-1$ and $+1$. Peano also showed that a function can have a minimum along every straight line through a point and still not have a minimum there. And it became clear that a function of two variables can possess both partial derivatives at a point at which it is not even continuous.

The cure, worked out by Weierstrass, Stolz, Fréchet and others, is to differentiate all variables at once: $f$ is differentiable at $a$ when $f(a+h)-f(a)$ is a linear function of $h$ up to an error much smaller than $\|h\|$. That linear map $Df(a)$, written as the Jacobian matrix, carries all of one-variable calculus over to $\mathbb{R}^n$: the chain rule becomes multiplication of matrices, the mean value theorem becomes an inequality, Taylor's theorem acquires a Hessian. Its deepest consequence is that an invertible derivative makes a map locally invertible — the inverse and implicit function theorems, which say when a system of nonlinear equations can be solved and when an equation describes a smooth curve or surface. The chapter ends with optimisation: critical points, the Hessian test and Lagrange multipliers.

Limits in ℝⁿ→Partial & directional derivatives→Total derivative Df(a)→Chain rule & MVT→Schwarz & Taylor→Inverse function theorem→Implicit function theorem→Hessian test→Lagrange multipliers
🎯 By the end of this chapter you can…
  • prove or disprove limits and continuity at a point of $\mathbb{R}^2$ using paths, polar bounds and iterated limits — and know which of these can only disprove;
  • decide, for a function given by a formula away from the origin, whether its partial derivatives, its directional derivatives and its total derivative exist at the origin, and whether it is $C^1$;
  • compute Jacobian matrices, gradients and directional derivatives, and use the chain rule, Euler's theorem for homogeneous functions and the mean value inequality;
  • decide whether mixed partial derivatives are equal (Schwarz's theorem) and write down second-order Taylor polynomials;
  • state the inverse and implicit function theorems with all their hypotheses, apply them to compute derivatives of inverse and implicit functions, and recognise their local character;
  • classify critical points with the Hessian, recognise when the test is inconclusive, and solve constrained extremum problems with Lagrange multipliers.
Teaser — Zoom In Until the Function Looks Linear

The left panel shows the blow-up $q_\rho(s)=\dfrac{f(a+\rho s)-f(a)}{\rho}$ on the square $s\in[-1,1]^2$, in colour bands (red positive, blue negative). The right panel shows the linear map $s\mapsto\nabla f(a)\cdot s$ in the same colours. Press Play to zoom in ($\rho\to0$). For a differentiable function the left picture turns into the right one; for the others it never changes, whatever the zoom.

ρ—
∇f(a)—
max |q_ρ(s) − ∇f(a)·s|—

How this is examined

CSIR NET returns to this chapter in almost every paper. Part B typically asks for a directional derivative or the maximal rate of change at a point, the value of an expression such as $xf_x+yf_y$ for a homogeneous function, the Jacobian of a composite or of a change of variables, an implicit derivative, the nature of the critical points of a polynomial in two variables, or the maximum of a function (often a quadratic form) on a circle or sphere. Part C, where several options may be correct, prefers functions defined by cases at the origin and asks which of continuity, existence of partial or directional derivatives, differentiability, continuity of the partials and equality of mixed partials hold; it also tests whether the inverse function theorem gives a global inverse, what can happen when the implicit function theorem's hypothesis fails, and whether a semidefinite Hessian or minima along all lines decide anything. Both sections collect the standard counterexamples in their ⚠️ boxes and drill these question types in their 🎯 Exam corners. GATE MA asks very similar questions.

Sections in this chapter

✅ Before you start

You will need $\varepsilon$–$\delta$ limits and continuity (§3.1), the one-variable mean value theorem and Taylor's theorem with remainder (§3.3), Cauchy sequences and completeness (§2.1), and from linear algebra: linear maps and their matrices (§8.2), determinants and invertibility (§8.3), and the definiteness of a symmetric matrix through its eigenvalues or leading minors (§10.3 — the facts used here are restated where they are needed). Open sets, compactness and connectedness in $\mathbb{R}^n$ are treated fully in Chapter 7 (§7.2, §7.3); here only the basic facts are used.