Calculus III: Chain Rule for Multivariable Functions

Moving from single-variable to multivariable calculus, you face a core challenge: how do small changes ripple through interconnected systems? The Chain Rule for Multivariable Functions is your essential tool for answering this. It provides a systematic method for differentiating compositions of functions with multiple inputs and outputs, which is foundational for modeling everything from heat flow in an engine block to stress propagation in a bridge truss.

From Single to Multiple Links

To build intuition, recall the classic single-variable chain rule. If $y=f(u)$ and $u=g(x)$, then the derivative of the composition is $dy/dx=(df/du)\cdot (du/dx)$. It describes how a change in $x$ propagates through $g$ to affect $f$, and ultimately $y$.

The multivariable version generalizes this idea to more complex networks. Consider a function $z=f(x,y)$, where both intermediate variables $x$ and $y$ are themselves functions of a single independent variable $t$ (e.g., time). Here, $z$ depends on $t$ through two distinct channels. The Chain Rule with Intermediate Variables states that the total derivative of $z$ with respect to $t$ is the sum of the contributions from each path:

$$\frac{dz}{dt}=\frac{\partial f}{\partial x}\frac{dx}{dt}+\frac{\partial f}{\partial y}\frac{dy}{dt}$$

The notation is critical: you use ordinary derivatives ($d/dt$) for functions of one variable and partial derivatives ($\partial /\partial x$) for functions of multiple variables. This formula calculates the instantaneous rate of change of $z$ as $t$ changes, accounting for how $t$ alters $x$ and $y$ simultaneously.

Example: Suppose the temperature on a metal plate is $T(x,y)=x^{2}y$ degrees, and a sensor moves along the path $x(t)=\cos (t)$, $y(t)=\sin (t)$. How fast is the temperature at the sensor changing at $t=\pi /4$?

1. Compute the partials: $\partial T/\partial x=2xy$, $\partial T/\partial y=x^2$.
2. Compute the ordinary derivatives: $dx/dt=-\sin (t)$, $dy/dt=\cos (t)$.
3. Apply the chain rule:

$$\frac{dT}{dt}=(2xy)(-\sin (t))+(x^2)(\cos (t))$$

1. Substitute $x=\cos (\pi /4)=\sqrt{2}/2$, $y=\sin (\pi /4)=\sqrt{2}/2$, $t=\pi /4$:

$$\frac{dT}{dt}=\left(2\cdot \frac{\sqrt{2}}{2}\cdot \frac{\sqrt{2}}{2}\right)\left(-\frac{\sqrt{2}}{2}\right)+\left({\left(\frac{\sqrt{2}}{2}\right)}^2\right)\left(\frac{\sqrt{2}}{2}\right)=(1)\left(-\frac{\sqrt{2}}{2}\right)+\left(\frac{1}{2}\right)\left(\frac{\sqrt{2}}{2}\right)=-\frac{\sqrt{2}}{4}$$

Visualizing Dependencies with Tree Diagrams

As dependency chains grow more intricate—with multiple independent variables or layers of composition—tracking all paths becomes paramount. A Tree Diagram for Dependency Tracking is an indispensable organizational tool.

To construct one, start with the dependent variable (e.g., $z$) at the top. Draw branches down to each intermediate variable it directly depends on (e.g., $x$ and $y$). From each intermediate variable, branch further to the ultimate independent variables (e.g., $s$ and $t$). Each edge on the tree represents a derivative.

Rule: For each independent variable, find every path from the dependent variable down to that independent variable. For each path, multiply the derivatives along the edges. Then, sum the products for all paths leading to that same independent variable.

For example, if $z=f(x,y)$, with $x=g(s,t)$ and $y=h(s,t)$, then $z$ depends on $s$ and $t$. The tree has $z$ at the top, branching to $x$ and $y$. Both $x$ and $y$ then branch to $s$ and $t$.

• To find $\partial z/\partial s$:
• Path 1: $z\to x\to s$: $(\partial z/\partial x)(\partial x/\partial s)$
• Path 2: $z\to y\to s$: $(\partial z/\partial y)(\partial y/\partial s)$
• Sum: $\partial z/\partial s=(\partial f/\partial x)(\partial g/\partial s)+(\partial f/\partial y)(\partial h/\partial s)$
• Similarly, $\partial z/\partial t=(\partial f/\partial x)(\partial g/\partial t)+(\partial f/\partial y)(\partial h/\partial t)$

This method scales to any number of variables and layers, preventing missed terms in complex engineering models where a single output may depend on dozens of inputs through a network of equations.

Implicit Differentiation Revisited and Generalized

In single-variable calculus, you learned implicit differentiation for equations like $F(x,y)=0$. The multivariable chain rule provides a more powerful and unified perspective. Consider an equation $F(x,y,z)=0$ that implicitly defines $z$ as a function of $x$ and $y$ (i.e., $z=f(x,y)$). You can find $\partial z/\partial x$ directly.

Treat $F$ as a function of three intermediate variables: $x$, $y$, and $z$. However, $z$ itself is a function of $x$ and $y$. Imagine $F(x,y,z)$ where $x$ and $y$ are independent, and $z=f(x,y)$. The total derivative of $F$ with respect to $x$ must be zero, because $F$ is constant (equal to 0). Using the chain rule along all paths from $F$ to $x$:

1. Direct path from $F$ to $x$: $(\partial F/\partial x)$.
2. Path from $F$ to $z$ to $x$: $(\partial F/\partial z)(\partial z/\partial x)$.
3. There is no path from $F$ to $y$ to $x$ with respect to this derivative, because $y$ is treated as independent of $x$ for this partial.

Setting the total derivative to zero gives: $$\frac{\partial F}{\partial x}+\frac{\partial F}{\partial z}\frac{\partial z}{\partial x}=0$$ Solving yields the elegant formula: $$\frac{\partial z}{\partial x}=-\frac{\partial F/\partial x}{\partial F/\partial z}$$ provided $\partial F/\partial z\ne 0$. This Implicit Differentiation via Multivariable Chain Rule is far more efficient and less error-prone than the single-variable technique, especially for functions of many variables.

Applications: Coordinate Systems and Sensitivity

The power of the chain rule is fully realized in its engineering applications. Two of the most critical are coordinate transformations and sensitivity analysis.

Coordinate Transformations are ubiquitous. When analyzing a physical system, you often shift from Cartesian $(x,y)$ to polar $(r,\theta )$ coordinates because the geometry simplifies the equations (e.g., modeling heat diffusion in a disk). Suppose a measurement $V$ is given as a function $f(x,y)$. To express how $V$ changes with radius $r$, you need $\partial V/\partial r$. The chain rule provides the direct link: since $x=r\cos \theta$ and $y=r\sin \theta$, $$\frac{\partial V}{\partial r}=\frac{\partial V}{\partial x}\frac{\partial x}{\partial r}+\frac{\partial V}{\partial y}\frac{\partial y}{\partial r}=\frac{\partial V}{\partial x}\cos \theta +\frac{\partial V}{\partial y}\sin \theta$$ This framework allows you to transform partial differential equations between coordinate systems, a routine task in fluid dynamics and electromagnetism.

Sensitivity Analysis in Engineering Models is the practice of quantifying how uncertainty in a model's inputs affects its outputs. If your model output $P$ (e.g., power output, stress, cost) is a function $P(x_1,x_2,...,x_n)$ of $n$ design parameters, the partial derivative $\partial P/\partial x_i$ is the sensitivity coefficient. It measures the rate of change of $P$ with respect to parameter $x_i$, holding others constant.

In a complex model, parameters might be linked. For instance, material strength ($S$) might depend on temperature ($T$) and pressure ($P$), which themselves depend on operational time ($t$). To find how strength degrades over time, $dS/dt$, you would use the full chain rule network. This allows engineers to identify which input parameters have the greatest influence on performance or failure, guiding robust design and cost-effective manufacturing tolerances.

Common Pitfalls

1. Mixing Derivative Notations Incorrectly: The most frequent error is using a partial derivative symbol ($\partial$) when an ordinary derivative ($d$) is required, or vice versa. Remember: use $d/dt$ when differentiating a function of a single variable with respect to that variable, even if the function's formula involves multiple intermediate variables. Use $\partial$ when dealing with a function of multiple independent variables.

1. Missing Paths in the Chain: When constructing the chain rule for $\partial z/\partial s$, it's easy to forget that $z$ may depend on $s$ through more than one intermediate variable. Always draw a tree diagram for problems with three or more variables to ensure you sum contributions from every distinct dependency path.

1. Misapplying the Single-Variable Rule: In an equation like $z=f(x(t),y(t))$, a common mistake is writing $dz/dt=f^{\prime }(x(t),y(t))$. This is meaningless because $f$ is a multivariable function and doesn't have a single "prime" derivative. You must take partial derivatives of $f$ and multiply by the derivatives of the inner functions.

1. Ignoring the Implicit Function Theorem Conditions: When using the formula $\partial z/\partial x=-F_x/F_z$, you must verify that $F_z\ne 0$ at the point of interest. If $F_z=0$, then $z$ may not be a differentiable function of $x$ and $y$ at that point, and the implicit differentiation result is invalid.

Summary

• The multivariable chain rule calculates the derivative of a composition by summing the contributions of all possible dependency paths, which can be visualized using a tree diagram.
• It generalizes and streamlines implicit differentiation, providing the formula $\partial z/\partial x=-F_x/F_z$ for an equation $F(x,y,z)=0$.
• Its engineering applications are profound, enabling seamless coordinate transformations (e.g., Cartesian to polar) and critical sensitivity analysis to determine how model outputs respond to changes in inputs.
• Success requires meticulous attention to notation (partial vs. ordinary derivatives) and a disciplined approach to enumerating all paths of influence from independent to dependent variables.