5.3 Applications of the chain rule

Optimization and Sensitivity Analysis

The chain rule becomes especially useful when you need to optimize functions that depend on intermediate variables, or when you want to understand how sensitive a system's output is to changes in its inputs. Both of these tasks rely on tracking how changes propagate through layers of dependent variables.

Optimization Techniques

Optimization finds the maximum or minimum values of a function, often subject to constraints. The chain rule enters the picture whenever the function you're optimizing is built from composed or interconnected pieces.

• Directional derivatives measure the rate of change of a function in a specific direction. You compute them by taking the dot product of the gradient vector $\nabla f$ with a unit vector $\hat{u}$ pointing in your chosen direction: $D_{\hat{u}}f = \nabla f \cdot \hat{u}$. This tells you the direction of steepest ascent (the gradient direction) or steepest descent (the negative gradient direction).
• Rates of change in multivariate functions are captured by partial derivatives. If $f$ depends on variables that themselves depend on other variables, the chain rule lets you compute the total rate of change. The total differential expresses the overall change in $f$ due to small changes in all input variables:

$df = \frac{\partial f}{\partial x}\,dx + \frac{\partial f}{\partial y}\,dy + \cdots$

This is a direct consequence of applying the chain rule to each path of dependence.

Sensitivity Analysis Methods

Sensitivity analysis asks: if you nudge an input variable, how much does the output change? The chain rule is the mechanism that answers this question when variables are linked through intermediate relationships.

• Local sensitivity analysis examines the impact of small perturbations around a specific operating point. You calculate partial derivatives of the output with respect to each input and evaluate them at that point. Larger partial derivatives mean the output is more sensitive to that input.
• Global sensitivity analysis explores how inputs affect the output across their entire range, not just near one point. Common techniques include variance-based methods (Sobol indices) and screening methods (Morris method). These go beyond single-point derivatives but still rely on understanding how changes propagate through the model.
• Sensitivity analysis matters for decision-making, risk assessment, and model validation. It identifies which variables are critical to control and reveals how robust your model's predictions are to uncertainty in the inputs.

Applications in Various Fields

Thermodynamics Applications

Thermodynamic systems involve quantities (pressure, temperature, volume, entropy) that are deeply interconnected. The chain rule is essential for relating changes in one thermodynamic variable to changes in another through intermediate state functions.

• Optimization appears in designing heat exchangers and configuring power cycles for maximum efficiency or minimum energy consumption.
• Sensitivity analysis helps assess how uncertainties in thermodynamic properties (like specific heat or thermal conductivity) and operating conditions affect system performance.
• Directional derivatives apply when analyzing thermodynamic potentials such as internal energy $U$, enthalpy $H$, and entropy $S$. For example, the chain rule lets you express how $G$ (Gibbs free energy) changes with temperature at constant pressure when $G$ depends on $H$ and $S$, which themselves depend on $T$ and $P$. This determines the direction of spontaneous processes and phase transitions.

Economics Applications

In economics, most quantities of interest (profit, cost, revenue) depend on variables that are themselves functions of other market conditions. The chain rule tracks these dependencies.

• Optimization is used to maximize profits, minimize costs, and allocate resources. Examples: a firm choosing production levels where marginal cost equals marginal revenue, or an investor optimizing a portfolio's risk-return tradeoff.
• Sensitivity analysis assesses how robust economic models and policies are to changes in assumptions. For instance, how sensitive is projected GDP growth to a change in the assumed interest rate?
• Rates of change are fundamental for computing marginal costs, marginal revenues, and elasticities. If total cost $C$ depends on output $q$, which depends on input prices, the chain rule gives you $\frac{dC}{dp} = \frac{dC}{dq} \cdot \frac{dq}{dp}$, connecting cost sensitivity to price changes through the production response.

Vector-Valued Functions and Parametric Surfaces

Vector-Valued Functions

Vector-valued functions map a real parameter to vectors in space. The chain rule for these functions lets you differentiate compositions involving vector outputs and scalar or vector inputs.

• A vector-valued function is written as $\vec{r}(t) = (x(t), y(t))$ in 2D or $\vec{r}(t) = (x(t), y(t), z(t))$ in 3D.
• The derivative $\vec{r}\,'(t) = (x'(t), y'(t), z'(t))$ describes the rate of change of the vector with respect to the parameter. In physics, if $\vec{r}(t)$ is position, then $\vec{r}\,'(t)$ is velocity and $\vec{r}\,''(t)$ is acceleration.
• If the parameter $t$ itself depends on another variable $s$, the chain rule gives $\frac{d\vec{r}}{ds} = \frac{d\vec{r}}{dt} \cdot \frac{dt}{ds}$. This comes up when reparametrizing curves or changing coordinate systems.
• Integrals of vector-valued functions calculate quantities like work done by a force along a curved path or the flux of a vector field through a surface.

Parametric Surfaces

Parametric surfaces extend vector-valued functions to two parameters, typically $u$ and $v$, mapping a region in the $uv$-plane to a surface in 3D space.

• A parametric surface is written as $\vec{r}(u, v) = (x(u, v),\, y(u, v),\, z(u, v))$.
• These can model complex shapes: spheres, tori, Möbius strips, and more.
• The partial derivatives $\vec{r}_u$ and $\vec{r}_v$ give tangent vectors to the surface at each point. Their cross product $\vec{r}_u \times \vec{r}_v$ produces the normal vector, which is used to compute surface area and curvature.
• The chain rule is critical here when the parameters $u$ and $v$ themselves depend on other variables. For example, if $u = u(s, t)$ and $v = v(s, t)$, then:

$\frac{\partial \vec{r}}{\partial s} = \frac{\partial \vec{r}}{\partial u}\frac{\partial u}{\partial s} + \frac{\partial \vec{r}}{\partial v}\frac{\partial v}{\partial s}$

This is the multivariable chain rule applied to a vector-valued function of two intermediate variables.

• Surface integrals over parametric surfaces appear throughout physics and engineering, for instance when calculating the flux of a vector field through a curved boundary or the work done along a curve lying on a surface.