When does the chain rule hold even if the inner derivative is zero?

The model tracks a small input change $dx$ through successive local linear approximations. First, the inner function $h(x)=x^2$ scales the input change by its derivative $h'(x)=2x$.

Conditions: Functions $h$ and $g$ are differentiable at the relevant points.; The input change $dx$ is infinitesimal.; The outer derivative is evaluated at the inner function's output $h(x)$, not at $x$.

By tracking a small input change $dx$ through nested mappings $x \to h(x) \to g(h(x))$. For $x=1.5$, $h=x^2$, and $g=\sin$, the inner rate is $h'(1.5)=3$, scaling $dx$ to $dh \approx 3dx$.

Conditions: Inner function $h(x)$ is differentiable at $x$; Outer function $g(u)$ is differentiable at $u=h(x)$; Example uses $h(x)=x^2$ and $g(u)=\sin u$ near $x=1.5$

Leibniz notation $\frac{dy}{dx} = \frac{dy}{du}\frac{du}{dx}$ resembles algebraic cancellation, but derivatives are limits, not ordinary fractions. The rigorous justification relies on expressing increments exactly as $\Delta y = [g'(u)+\epsilon]\Delta u$ with $\epsilon \to 0$.

Conditions: Functions are differentiable.; The inner derivative $du/dx$ can be zero.; The argument requires rigorous limit definitions, not informal algebra.

The outer derivative $g'$ must be evaluated at the output of the inner function, i.e., at $h(x)$, not at the original input $x$. This reflects the sequential nature of composition: the change in $x$ affects $h$, and the sensitivity of $g$ depends on the current value of $h$.

Conditions: Computing derivative of $g(h(x))$; Applying chain rule $g'(h(x))h'(x)$