ES.1802, October 8, 2026

Overview

The lesson connects Jacobians to area scaling in coordinate changes, then develops the multivariable critical-point test from the gradient, Taylor approximation, and Hessian. It applies these ideas to an ellipse-to-circle substitution and to classifying minima, maxima, saddle points, and inconclusive cases using the Hessian’s eigenvalue signs.

Key takeaways

For a coordinate map (x,y) → (u,v), the absolute determinant of ∂(u,v)/∂(x,y) gives the local area-scaling factor; the inverse map is used when rewriting dx dy in terms of du dv.
A multivariable critical point requires both f_x = 0 and f_y = 0, but those equations alone cannot distinguish a maximum, minimum, or saddle.
The Hessian is the matrix of second partial derivatives; when the mixed partials agree, it is symmetric and its eigenvalue signs classify the local curvature.
At a critical point, a positive-definite Hessian indicates a local minimum, a negative-definite Hessian indicates a local maximum, and an indefinite Hessian indicates a saddle.
For f(x,y) = x² + 4y² + 3 + 4x − 4y − 2xy, the critical point is (−2, 0), and its Hessian [[2, −2], [−2, 8]] is positive definite.
If the Hessian has a zero eigenvalue, the second-derivative test does not decide the critical point’s type.

Chapters

0:00 Why Jacobians Matter for Coordinate Changes

A coordinate substitution defines u and v as functions of x and y, with Jacobian matrix ∂(u,v)/∂(x,y).
The example 3x² + y² = 9 becomes the circle u² + v² = 9 under u = √3x and v = y.
Simplifying an ellipse into a circle makes the region easier to handle, but area calculations need a conversion factor.

4:00 Transforming Tiny dx and dy Area Elements

In the original xy plane, the small area element is dA = dx dy, spanned by the x- and y-direction differentials.
Applying the Jacobian to those two directions produces transformed vectors that generally span a parallelogram in uv coordinates.
The determinant measures how the transformation changes the area of that small parallelogram.

9:00 The Jacobian Determinant as an Area-Scaling Factor

The transformed area satisfies dA_uv = |det(∂(u,v)/∂(x,y))| dA_xy.
For integration written in uv coordinates, the inverse transformation supplies the corresponding factor for dx dy in terms of du dv.
Because a double integral adds many small area elements, the local Jacobian area factor also converts the area of the full region.

18:40 Critical Points and the Multivariable First-Derivative Test

A critical point occurs where both first partial derivatives, f_x and f_y, are zero—that is, where ∇f is the zero vector.
A flat tangent plane can occur at a local maximum, a local minimum, or a saddle point, so the gradient alone does not classify the point.
The one-variable second-derivative analogy is negative curvature at a maximum and positive curvature at a minimum.

25:00 Finding the Critical Point of a Quadratic Surface

The worked function is f(x,y) = x² + 4y² + 3 + 4x − 4y − 2xy.
Setting f_x = 2x + 4 − 2y and f_y = 8y − 4 − 2x to zero gives the critical point (−2, 0).
The level curves are ellipses shifted away from the origin; the plotted surface indicates a minimum at the critical point.

31:00 Taylor’s Quadratic Approximation Around a Critical Point

Near a point p, the quadratic approximation combines f(p), the linear gradient term, and a second-order term involving the Hessian.
The second-order contribution has the form ½ Δxᵀ H Δx, where Δx is the displacement from p.
At a critical point, the gradient term vanishes, making the Hessian’s quadratic form central to classification.

33:00 Building the Hessian from Second Partial Derivatives

The gradient is the first-derivative vector; differentiating that vector produces the Hessian matrix.
For f(x,y), H = [[f_xx, f_xy], [f_yx, f_yy]], containing the second partial derivatives.
When mixed partials are equal, f_xy = f_yx, the Hessian is symmetric, which gives it useful eigenvalue properties.

40:00 Positive and Negative Definiteness Identify Extrema

A symmetric matrix H is positive definite when zᵀHz > 0 for every nonzero vector z; equivalently, all its eigenvalues are positive.
At a critical point, a positive-definite Hessian makes the quadratic change positive in every direction, indicating a local minimum.
A negative-definite Hessian has all negative eigenvalues and indicates a local maximum.

45:00 Indefinite Hessians, Saddle Points, and Inconclusive Tests

An indefinite Hessian has both positive and negative eigenvalues, so the function curves up in some directions and down in others: the critical point is a saddle.
If any Hessian eigenvalue is zero, the second-derivative test is inconclusive and higher-order behavior may be needed.
For the quadratic example, H = [[2, −2], [−2, 8]] is positive definite, confirming a minimum at (−2, 0).

Summary, takeaways, and chapters were generated by AI from the video's transcript and may contain errors. The video belongs to its creator, ChristianAltoSax.