[Intermediate] — The shortcuts, each derived rather than asserted — and the chain rule, which is the one that actually matters.
Before you start
- You can differentiate from first principles — defining the derivative.
- You know the index laws — indices and logarithms.
What you will be able to do after this lesson:
- Apply the power, sum, product, quotient and chain rules correctly.
- Derive the power and product rules from the definition.
- Recognise which rule a composite expression needs, and in what order.
- Explain why the chain rule is the foundation of backpropagation.
1. The rules
| Rule | Statement |
|---|---|
| Constant | |
| Power | |
| Constant multiple | |
| Sum | |
| Product | |
| Quotient | |
| Chain |
And the standard functions:
is its own derivative, and that is the definition of — the unique base whose exponential grows at exactly its own value. It is why appears everywhere growth is proportional to size.
2. Two derivations
The rules are not arbitrary. Deriving two of them shows where they come from and makes the rest believable.
The power rule for
The general case is the same argument with the binomial theorem. Expanding gives . The cancels, dividing by leaves , and the rest vanishes.
The product rule
Why it is not : consider area. If a rectangle has width and height , growing both slightly increases the area by two strips — one from the width change, one from the height — plus a tiny corner.
g' ┌───┬─┐ <- corner: f' x g', negligible
│ │ │
g │ fg│ │ <- strip: f x g'
├───┼─┤
└───┴─┘
f f' <- strip: f' x g
Divide by and take the limit: the corner term still has a in it and disappears, leaving . The two strips are the two terms.
3. The chain rule — the important one
In Leibniz notation it looks like fractions cancelling, which is the best way to remember it:
Read it as rates multiplying. If changes 3× as fast as , and changes 5× as fast as , then changes 15× as fast as .
Worked: differentiate .
Outer function: , whose derivative is . Inner: , whose derivative is .
The commonest mistake is forgetting the inner derivative — writing and stopping. The chain rule is exactly that missing factor.
NOTE
This rule is why neural networks can be trained.
A network is a deep composition: . To adjust a weight in an early layer you need the derivative of the final loss with respect to that weight — and the chain rule says that is the product of the local derivatives along the path.
Backpropagation is the chain rule applied systematically, computing those products from the output backwards and reusing shared sub-products. It is not a separate algorithm — it is bookkeeping over this one rule. See ai-ml and build your own neural network.
4. Choosing the rule
Work from the outside in. Ask what operation is applied last:
| Expression | Outermost operation | Rule |
|---|---|---|
| addition | sum | |
| multiplication | product | |
| division | quotient | |
| composition | chain | |
| composition (cube of sine) | chain |
and are different functions and need the chain rule applied in opposite orders. Confusing them is one of the most common errors in the topic.
5. Practice — problems
- Differentiate: (a) (b) (c) . (Write (b) and (c) as powers first.)
- Product rule: (a) (b) .
- Quotient rule: (a) (b) .
- Chain rule: (a) (b) (c) (d) .
- Combined: differentiate and , naming each rule as you use it.
- Derive the quotient rule from the product rule. Hint: write as and use the chain rule on .
Answers — open only after an attempt
- (a) . (b) . (c) .
- (a) . (b) .
- (a) . (b) .
- (a) . (b) . (c) . (d) .
- : product rule, with the chain rule on → .
: chain rule → . - . ∎ The quotient rule is not a separate rule — it is the product and chain rules combined, which is worth knowing if you ever forget it.
Before moving on
- State all seven rules from memory.
- Derive the power rule for a specific and explain the product rule geometrically.
- Identify the outermost operation and choose the right rule.
- Apply the chain rule without dropping the inner derivative.
Recap: The derivative rules are shortcuts for limits you could always compute directly, and each is provable from the definition — the power rule from the binomial expansion, the product rule from the two strips of a growing rectangle, the quotient rule from the product and chain rules together. Choosing a rule means identifying the operation applied last. The chain rule multiplies rates along a composition, and applying it systematically across a deep composition is exactly what backpropagation does.
Next: Applications — using derivatives to solve problems rather than to differentiate expressions.
Related
- Defining the derivative
- Build your own neural network — the chain rule, applied
Software engineering
Infrastructure & security
Computer science
Hardware & engineering