The Chain Rule

The Chain Rule tells us how to differentiate a composite function — a function built by plugging one function into another, such as $\sin (x^2)$ or $\sqrt{3x+1}$. Its central claim is that derivatives multiply along the chain of dependence: each layer contributes its own rate of change, and the total rate is the product. This matters because most realistic functions are compositions, so without the Chain Rule the bulk of calculus would be inaccessible. A surprising fact is that the rule extends infinitely — you can compose ten functions and simply multiply ten derivatives — and it is the engine behind backpropagation, the algorithm that trains modern neural networks. The Leibniz notation $\frac{dy}{du}\cdot \frac{du}{dx}$ makes the rule look like the cancellation of fractions, a memorable mnemonic that is almost but not quite the literal truth.

The Core Insight

A composite function is a relay race of dependence. If a small nudge in $x$ changes $u$ at rate $\frac{du}{dx}$, and a small nudge in $u$ changes $y$ at rate $\frac{dy}{du}$, then the nudge in $x$ ripples through $u$ to $y$ at the product of those rates. Sensitivities compose by multiplication.

Formal Statement

Composite Function

Given two functions $g$ and $f$, the composite function $f\circ g$ is defined by

Here $g$ is the inner function, $f$ is the outer function. If we write $u=g(x)$, then $y=f(u)$, and $y$ depends on $x$ through the intermediate variable $u$.

The Chain Rule

Let $g$ be differentiable at $x$ and let $f$ be differentiable at $u=g(x)$. Then the composite $f\circ g$ is differentiable at $x$, and

In Leibniz notation, with $y=f(u)$ and $u=g(x)$:

The phrase "$du$ cancels" is a mnemonic, not a proof. The differentials $du$ in $\frac{dy}{du}$ and $\frac{du}{dx}$ are not ordinary numbers being divided. A correct proof must handle the case where $g^{\prime }(x)\ne 0$ separately from where the difference quotient's denominator could vanish.

Why It Holds / Derivation

Sketch via Difference Quotients

We want $\frac{dy}{dx}$ where $y=f(g(x))$. Start from the limit definition.

Gottfried Wilhelm Leibniz | 1646–1716

Leibniz co-invented calculus and, crucially, designed the notation $\frac{dy}{dx}$ that makes the Chain Rule look almost self-evident. His differentials $dy$, $du$, $dx$ were conceived as "infinitely small" increments, and within that framework $\frac{dy}{du}\cdot \frac{du}{dx}$ reads as honest fraction cancellation. Though his infinitesimals were later replaced by limits, the notation survived precisely because it encodes the Chain Rule so transparently. The form in equation [3] is sometimes called the Leibniz form of the rule.

Building Intuition

Imagine a chain of gears. Turn the input gear (variable $x$); it spins a middle gear ($u$) which is, say, 3 times faster; that middle gear spins the output gear ($y$) which is 2 times faster than the middle one. How fast does $y$ spin compared to $x$? $2\times 3=6$ times. Each link multiplies its gear ratio.

Equivalently, think of unit conversions. If you travel 60 miles per hour and there are 1.6 kilometres per mile, you travel $60\times 1.6=96$ km/h. The $\frac{\text{miles}}{\text{hour}}$ and $\frac{\text{km}}{\text{mile}}$ multiply, with "miles" cancelling — exactly the Chain Rule pattern.

A composite function as a relay: a nudge in x ripples through u to y, with rates multiplying.

Worked Examples

Differentiating a Power of a Polynomial

A Physical Rate: Expanding Balloon

Common Misconceptions

The Rule Is Not Mere Substitution of Derivatives

One might guess that $(f\circ g{)}^{\prime }(x)=f^{\prime }(g(x))$ alone, as if differentiating the outer function were enough. Take $f(u)=u^2$ and $g(x)=5x$, so $f(g(x))=25x^2$.

The true derivative is $50x$. But $f^{\prime }(g(x))=2(5x)=10x$ — wrong by a factor of 5. The missing factor is exactly $g^{\prime }(x)=5$, and $10x\times 5=50x$ ✓. This shows the inner derivative is indispensable, not optional.

Connections

The Chain Rule sits at the heart of differentiation's machinery and links to many neighbours. It is essential for Implicit Differentiation, where we treat $y$ as a hidden function of $x$ and differentiate $y^n$ as $n{y}^{n-1}\frac{dy}{dx}$. It powers Related Rates problems, as in the balloon example, by chaining a geometric derivative to a time derivative. It underlies the Derivative of Inverse Functions via $\frac{dx}{dy}=1/\frac{dy}{dx}$. Combined with the Fundamental Theorem of Calculus, it produces the technique of Integration by Substitution — substitution is the Chain Rule run backwards. Its multivariable cousin generalizes equation [3] to sums of partial-derivative products.

Real-World Applications

The Chain Rule appears wherever quantities depend on each other through intermediate variables:

1. Neural networks (backpropagation). Training deep networks computes gradients of a loss across dozens of composed layers by repeatedly multiplying local derivatives — a direct, large-scale Chain Rule.
2. Population dynamics. If a population $P$ depends on food supply $F$, and $F$ depends on rainfall $R$ over time, $\frac{dP}{dt}$ chains through $\frac{dP}{dF}$ and $\frac{dF}{dR}$.
3. Thermodynamics. Pressure depends on temperature, which depends on time; rates of pressure change in an engine cylinder are computed by chaining.
4. Economics — marginal analysis. Cost depends on output, output depends on labour hours; the marginal cost per hour multiplies the two sensitivities.
5. Radiometry / inverse-square fields. Light intensity depends on distance, distance depends on time for a moving sensor; $\frac{dI}{dt}$ uses the Chain Rule.
6. Kinematics. A particle's energy depends on speed, speed depends on time — chaining gives instantaneous power.
7. Pharmacokinetics. Drug effect depends on blood concentration, which depends on time since dosing; clinicians chain these to model response rates.

Historical Development

• 1676 — Leibniz, in private manuscripts, manipulates differentials $dy$, $du$, $dx$ in ways that effectively use the Chain Rule, though without a formal statement.
• 1684 — Leibniz publishes Nova Methodus, the first calculus paper, establishing the $\frac{dy}{dx}$ notation that makes the rule transparent.
• 1690s — The Bernoulli brothers (Jacob and Johann) and L'Hôpital apply chained differentials freely to curves and mechanics problems.
• 1734 — Berkeley's The Analyst attacks the logical vagueness of infinitesimals, including the loose "cancellation" justifying the rule, spurring demand for rigor.
• 1755 — Euler systematizes differential calculus in Institutiones Calculi Differentialis, treating composite differentiation as routine.
• 1821 — Cauchy's Cours d'analyse recasts derivatives via limits, providing the conceptual basis for a rigorous Chain Rule proof.
• 1860s–1870s — Weierstrass's epsilon–delta formalism allows fully rigorous statements and proofs of the rule.
• 1880s — Carathéodory's "increment" formulation of the derivative yields the clean, gap-free proof now common in textbooks.
• 1986 — Rumelhart, Hinton, and Williams popularize backpropagation, recasting the Chain Rule as the computational core of modern machine learning.

Subtleties & Limits

The Chain Rule as stated requires both functions to be differentiable at the relevant points. Subtleties arise at the edges:

• Non-differentiable inner functions. If $g$ has a corner (e.g. $g(x)=|x|$ at $x=0$), the composite may still be differentiable, but the Chain Rule formula does not directly apply — case-by-case analysis is needed.
• The "fake fraction" problem. Equation [3] looks like cancellation but isn't; in multivariable calculus the naïve cancellation breaks entirely, replaced by $\frac{dz}{dt}=\frac{\partial z}{\partial x}\frac{dx}{dt}+\frac{\partial z}{\partial y}\frac{dy}{dt}$ — a sum of products.
• Higher derivatives. The second derivative of a composite is not simply $f^{\prime \prime }(g)g^{\prime 2}$; it equals $f^{\prime \prime }(g)(g^{\prime }{)}^2+f^{\prime }(g)g^{\prime \prime }$, and the general $n$-th derivative is given by the intricate Faà di Bruno's formula, a still-studied combinatorial object.
• Non-standard analysis. Whether Leibniz's infinitesimals can be rigorously revived (Robinson, 1960s) so that the "cancellation" is literally valid remains a live area of mathematical logic.

See Also