Chapter 33 — Key Takeaways

The one idea to carry forward: a smooth transformation looks locally like its Jacobian matrix, and the absolute value of that matrix's determinant is the local area- (2D) or volume- (3D) scaling factor. Change of variables is bookkeeping for that local stretching — the multivariable generalization of $du = g'(x)\,dx$.

1. The Jacobian Matrix and Determinant (Section 33.3)

For a transformation $T:(u,v)\mapsto(x,y)$, the Jacobian matrix is the matrix of first partial derivatives, and the Jacobian determinant (often just "the Jacobian") is its determinant:

$$J_T = \frac{\partial(x,y)}{\partial(u,v)} = \begin{pmatrix} x_u & x_v \\ y_u & y_v \end{pmatrix}, \qquad \det J_T = x_u y_v - x_v y_u.$$

The matrix is the best linear approximation to $T$ near a point — the multivariable derivative from Chapter 30.

2. Geometric Meaning: A Scaling Factor (Section 33.6)

$\lvert\det J_T\rvert$ is the local area/volume scaling factor. A tiny cell $du\,dv$ maps to a parallelogram of area $\lvert\det J_T\rvert\,du\,dv$. The derivation: the sides of the cell map to the partial-derivative vectors $\mathbf{T}_u$ and $\mathbf{T}_v$, and the area of the parallelogram they span is the absolute value of the determinant whose columns are those vectors. A smooth map is locally linear, and a linear map scales area by its determinant.

3. The Change of Variables Formula (Section 33.4, Section 33.7)

$$\iint_R f(x,y)\,dA = \iint_S f\big(x(u,v),y(u,v)\big)\,\lvert\det J_T\rvert\,du\,dv,$$

and in three dimensions, with $dV = \lvert\det J_T\rvert\,du\,dv\,dw$,

$$\iiint_R f\,dV = \iiint_S f\big(x(u,v,w),\dots\big)\,\lvert\det J_T\rvert\,du\,dv\,dw.$$

The factor $\lvert\det J_T\rvert$ is the successor of the single-variable stretch factor $g'(x)$; the absolute value is the only genuinely new wrinkle — the sign of the determinant records orientation, which area discards.

4. The Single-Variable Analogy (Section 33.2)

The structural parallel is exact. In 1D, "how much does a small piece scale?" has the one-number answer $\lvert g'(x)\rvert$; in higher dimensions it is still one number, the absolute value of a determinant.

5. The Standard Jacobians — Derived, Not Memorized (Section 33.3, Section 33.7, Section 33.13)

Why each factor is what it is: - Polar $r$: a wedge of fixed angular width spans more arc the farther out you go, so area grows with $r$. - Cylindrical $r$: polar in the $xy$-plane with $z$ untouched; the bottom row $(0,0,1)$ of the Jacobian reduces the $3\times3$ determinant to the polar $2\times2$. - Spherical $\rho^2\sin\phi$: the $\rho^2$ says volume grows like the radius squared (a shell's area $\propto \rho^2$); the $\sin\phi$ pinches wedges thin near the poles ($\phi = 0,\pi$) and fattest at the equator ($\phi = \pi/2$). It correctly vanishes at $\phi = 0$ and $\rho = 0$, where the coordinates degenerate.

6. Inverses Multiply to One (Section 33.8)

If $T$ is invertible, then $J_{T^{-1}}J_T = I$ (chain rule), and since the determinant is multiplicative,

$$\det J_{T^{-1}} = \frac{1}{\det J_T}.$$

This is the multivariable echo of $\frac{dx}{du}\cdot\frac{du}{dx}=1$. Practical use: when computing $x,y$ in terms of $u,v$ is messy but the reverse is easy, compute $\det\partial(u,v)/\partial(x,y)$ and take the reciprocal. By the Inverse Function Theorem, $\det J_T \neq 0$ at a point certifies a local smooth inverse there; where $\det J_T = 0$, the map collapses dimensions (e.g. the polar origin).

7. Choosing a Transformation (Section 33.4)

• Awkward region bounded by a family of lines/curves → introduce variables that label the family, turning boundaries into coordinate lines (a parallelogram becomes a rectangle).
• Awkward integrand depending on $x^2+y^2$ → polar; spherically symmetric → spherical; axially symmetric → cylindrical.
• Cross terms in a quadratic → a linear (rotation) change of variables that decorrelates the variables, with a constant Jacobian.
• Transformations can be chained: e.g., stretch an ellipse to a disk ($\lvert\det J\rvert = ab$), then go polar ($\lvert\det J\rvert = \rho$); the two factors multiply.

8. Probability Density Transformations (Section 33.9)

For $(U,V) = T(X,Y)$, demanding that total probability be preserved forces

$$p_{UV}(u,v) = p_{XY}\big(x(u,v),y(u,v)\big)\,\lvert\det J_{T^{-1}}\rvert,$$

where $T^{-1}$ expresses the old variables in terms of the new. A density is "probability per unit area," so when area rescales, the density rescales inversely — and the Jacobian is that rescaling. This single formula underlies the Box–Muller transform (uniform → Gaussian) and, in $n$ dimensions, normalizing flows in machine learning, whose engineering goal is a transformation expressive enough to be useful yet structured (often triangular $J$) so $\det J$ is cheap.

9. Common Errors

• Dropping the absolute value: the area element uses $\lvert\det J\rvert$, never the signed $\det J$.
• Wrong direction: the Jacobian factor multiplies the new differentials. If you integrate over $(u,v)$, use $\lvert\det\partial(x,y)/\partial(u,v)\rvert$. Confusing it with its reciprocal is the single most common slip.
• Forgetting the Jacobian entirely: the classic wrong answer.
• Ignoring degenerate points ($\det J = 0$): usually a set of zero area that does not affect the integral, but where the geometry genuinely breaks.

10. Connections

• Chapter 15 — single-variable $u$-substitution; the Jacobian generalizes $du = g'(x)\,dx$.
• Chapter 26 — polar coordinates as a curve-description tool; here we see why they carry a factor of $r$ in integration.
• Chapter 28 — the parallelogram-area-as-determinant (2D cross product) and the scalar triple product (3D) that justify the area/volume interpretation.
• Chapter 30 — the Jacobian matrix is the multivariable derivative; the inverse rule is the chain rule for Jacobians.
• Chapter 32 — this chapter derives the polar, cylindrical, and spherical volume elements that Chapter 32 used on faith.
• Part VII (Chapters 34–38) — the Jacobian is the change-of-coordinates engine for every line and surface integral, and through differential forms (Section 33.12) it is the seed of the generalized FTC in Chapter 38.

What's Next

Chapter 34: Vector Fields opens Part VII — functions $\mathbf{F}(\mathbf{x})$ assigning a vector to every point, the language of fluid flow, electric and gravitational fields, and force. From there, line integrals and Green's Theorem (Ch. 35), surface integrals (Ch. 36), and Stokes' and the Divergence Theorems (Ch. 37) each build a higher-dimensional Fundamental Theorem of Calculus — and each quietly leans on the Jacobian whenever coordinates change.

The single sentence to remember. When a problem is hard, change your point of view — and let the Jacobian pay the toll for the change.