Understanding the behavior of a function at its critical points is a cornerstone of multivariable calculus. Still, the second derivative test for multivariable functions provides a systematic method for this classification using the eigenvalues of the Hessian matrix, or more practically, the determinant of that matrix. In real terms, while finding where the gradient vanishes identifies potential maxima, minima, or saddle points, classifying these points requires a deeper look at the function's curvature. This test extends the familiar concavity check from single-variable calculus into higher dimensions, offering a powerful tool for optimization problems in economics, engineering, machine learning, and physics.
The Foundation: Critical Points and the Hessian Matrix
Before applying the test, we must establish the necessary groundwork. For a function $f(x, y)$ of two variables with continuous second partial derivatives, a critical point $(a, b)$ occurs where the gradient vector is zero:
$ \nabla f(a, b) = \langle f_x(a, b), f_y(a, b) \rangle = \langle 0, 0 \rangle $
Once these points are located, the Hessian matrix (or simply the Hessian) captures the second-order behavior. For a function of two variables, it is a $2 \times 2$ symmetric matrix defined as:
$ H(x, y) = \begin{bmatrix} f_{xx}(x, y) & f_{xy}(x, y) \ f_{yx}(x, y) & f_{yy}(x, y) \end{bmatrix} $
By Clairaut’s theorem (equality of mixed partials), assuming continuity, $f_{xy} = f_{yx}$, making the matrix symmetric. This symmetry guarantees that the eigenvalues of the Hessian are real numbers, which correspond to the principal curvatures of the surface at that point. The determinant of the Hessian, often denoted as $D(x, y)$ or $\Delta(x, y)$, is the product of these eigenvalues:
$ D(x, y) = \det(H) = f_{xx}(x, y)f_{yy}(x, y) - [f_{xy}(x, y)]^2 $
This scalar quantity $D$ is the linchpin of the second derivative test.
The Second Derivative Test for Two Variables
The classification logic relies on the signs of the eigenvalues. Since calculating eigenvalues explicitly can be cumbersome, the test uses the determinant $D$ and the sign of $f_{xx}$ (which shares the sign of the first eigenvalue if $D > 0$) as proxies.
The Theorem: Let $(a, b)$ be a critical point of $f(x, y)$ with continuous second partial derivatives. Define $D = f_{xx}(a, b)f_{yy}(a, b) - [f_{xy}(a, b)]^2$ Which is the point..
- If $D > 0$ and $f_{xx}(a, b) > 0$: The eigenvalues are both positive. The function has a local minimum at $(a, b)$. The surface is concave up in all directions (bowl-shaped).
- If $D > 0$ and $f_{xx}(a, b) < 0$: The eigenvalues are both negative. The function has a local maximum at $(a, b)$. The surface is concave down in all directions (dome-shaped).
- If $D < 0$: The eigenvalues have opposite signs. The function has a saddle point at $(a, b)$. The surface curves up in one direction and down in another (shaped like a horse saddle or a Pringles chip).
- If $D = 0$: The test is inconclusive. The eigenvalues include at least one zero. The point could be a local max, min, saddle point, or none of these. Higher-order analysis (Taylor expansions) or graphical inspection is required.
Why $f_{xx}$? When $D > 0$, the product of eigenvalues is positive, meaning they share the same sign. Since the trace of the Hessian ($f_{xx} + f_{yy}$) equals the sum of the eigenvalues, checking the sign of $f_{xx}$ (or $f_{yy}$) reveals the sign of both eigenvalues. You could equivalently check $f_{yy}$; the result is identical Most people skip this — try not to..
Geometric Intuition: Quadratic Forms and Taylor Expansion
To truly grasp why this works, consider the second-order Taylor approximation of $f$ near a critical point $(a, b)$. Because the first derivatives are zero, the linear terms vanish:
$ f(a+h, b+k) \approx f(a, b) + \frac{1}{2} \left( f_{xx}h^2 + 2f_{xy}hk + f_{yy}k^2 \right) $
The expression in parentheses is a quadratic form $Q(h, k) = \mathbf{v}^T H \mathbf{v}$ where $\mathbf{v} = \langle h, k \rangle$. The classification of the critical point reduces entirely to the definiteness of this quadratic form:
- Positive Definite ($Q > 0$ for all $\mathbf{v} \neq 0$): Local Minimum. This occurs iff eigenvalues ${content}gt; 0$ iff $D > 0$ and $f_{xx} > 0$.
- Negative Definite ($Q < 0$ for all $\mathbf{v} \neq 0$): Local Maximum. This occurs iff eigenvalues ${content}lt; 0$ iff $D > 0$ and $f_{xx} < 0$.
- Indefinite ($Q$ takes both signs): Saddle Point. This occurs iff eigenvalues have opposite signs iff $D < 0$.
This perspective connects the algebraic test directly to the geometry of the surface.
Worked Example: Classifying Critical Points
Let's apply the test to the function $f(x, y) = x^3 - 3xy^2 + y^3$. This function is a classic "monkey saddle" variant.
Step 1: Find Critical Points Compute first partial derivatives: $ f_x = 3x^2 - 3y^2 $ $ f_y = -6xy + 3y^2 = 3y(y - 2x) $
Set $f_x = 0 \implies x^2 = y^2 \implies y = \pm x$. Set $f_y = 0 \implies y = 0 \text{ or } y = 2x$.
- Case $y=0$: $x=0$. Critical point: $(0,0)$.
- Case $y=x$: $x = 2x \implies x=0$. Same point $(0,0)$.
- Case $y=-x$: $-x = 2x \implies 3x=0 \implies x=0$. Same point $(0,0)$.
The only critical point is $(0,0)$.
Step 2: Compute Second Partial Derivatives $ f_{xx} = 6x $ $ f_{yy} = -6x + 6y $ $ f_{xy} = -6y $
Step 3: Evaluate at $(0,0)$ and Compute $D$ $ f_{xx}(0,0) = 0 $ $ f_{yy}(0,0) = 0 $ $ f_{xy}(0,0) = 0 $ $ D(0,0) = (0)(0) - (0)^2 = 0 $
Step 4: Interpret Since $D = 0$, the second derivative test is inconclusive. We cannot classify $(0,0)$ using this method alone. Analyzing cross-sections (e.g., $y=x$ gives $f=x^3$; $y=0$ gives $f=x^3$) reveals it is an inflection point/saddle-like behavior, but not a standard saddle point. This highlights
This highlights the limits of the second derivative test. When the Hessian determinant is zero, the quadratic approximation collapses to higher-order terms, so the usual sign analysis is no longer sufficient Which is the point..
For the function in question, the Taylor expansion about ((0,0)) reduces to [ f(x,y) \approx x^3
