Differentiation rules from structure

In brief

A rectangular garden is being enlarged on both sides at once: its length is growing over time and, independently, its width is also growing over time. The area, length times width, is changing too, but not simply as the sum of the two individual rates, because the two changes interact, a longer garden makes the same rate of widening add more area than it would on a narrower plot. Differentiation rules exist to handle exactly this: real quantities are built out of sums, products, ratios, and functions plugged into other functions, and each way of combining pieces has its own precise rule for how the combined rate of change relates to the rates of the pieces. These rules are not conventions to memorize blindly, each one is a provable consequence of the derivative's limit definition, and knowing where each rule comes from is what lets you trust the derivative of a complicated model as much as you trust the derivative of a simple one.

The full treatment

First look: the growing garden

Let the garden's length at time t be L of t, and its width be W of t, both changing as time passes. The area is A of t, equal to L of t times W of t. Suppose at a particular moment the length is 10 meters and growing at 2 meters per year, and the width is 6 meters and growing at 1 meter per year. Naively adding the two rates, 2 plus 1, equals 3, suggests area grows at 3 square meters per year, but this is wrong, because it ignores that a change in length affects a 6-meter-wide strip while a change in width affects a 10-meter-long strip. The area actually gains, in a short interval, contributions from both the length's growth stretched across the current width, and the width's growth stretched across the current length: 2 times 6, plus 1 times 10, equals 12 plus 10, equals 22 square meters per year. This exact pattern, current rate of one factor times the current value of the other, summed over both factors, is the product rule, and the garden makes visible exactly why it takes this particular form rather than simple addition.

Building the idea: deriving the product rule from the limit

Start from the definition. The derivative of A equals L times W, at a point a, is the limit as h approaches zero of: (L of a plus h, times W of a plus h, minus L of a times W of a) divided by h. To make progress, insert and subtract a helpful middle term, L of a plus h, times W of a, inside the numerator; this changes nothing since it is added and then subtracted, but it splits the expression into two more manageable pieces: L of a plus h, times the quantity (W of a plus h, minus W of a), plus W of a times the quantity (L of a plus h, minus L of a), all divided by h. As h shrinks toward zero, L of a plus h approaches L of a because L is continuous, the quantity (W of a plus h minus W of a) divided by h approaches W prime of a by definition, and similarly the other piece approaches L prime of a times W of a. Adding the two limits gives exactly: L of a times W prime of a, plus L prime of a times W of a, the product rule, derived, not asserted.

The quotient and chain rules follow the same logic

The quotient rule, for a ratio of two functions, follows a similar algebraic manipulation of the difference quotient, and produces the form: the derivative of the numerator times the denominator, minus the numerator times the derivative of the denominator, all divided by the denominator squared; the squared denominator arises directly from how the quotient's difference quotient factors once cleared of fractions. The chain rule handles a different kind of combination, a function plugged inside another function, such as the cost of heating a house as a function of outdoor temperature, where outdoor temperature is itself a function of the time of day. Here the combined rate of change is the product of two rates: how sensitive the outer function is to its immediate input, times how sensitive that immediate input is to the true underlying variable. Formally, if a function is written as f of g of x, its derivative with respect to x equals f prime evaluated at g of x, times g prime of x. The chain rule can be motivated exactly like the garden: a change in x causes a proportional change in g of x, governed by g prime, and that changed value of g of x then causes a further proportional change in f, governed by f prime evaluated at the new point, so the two proportionalities multiply rather than add.

What the rules assume and where they can mislead if used blindly

Every rule here assumes each piece function is itself differentiable at the point in question; if the length function in the garden example had a sharp corner in its growth pattern at the exact moment in question, the product rule's inputs would not exist and the rule could not be applied there at all. A second real hazard is applying a rule mechanically and losing track of what it says physically: a learner who correctly computes L prime times W plus L times W prime as a formula, but cannot explain why it is the length's growth spread across the current width plus the width's growth spread across the current length, has gained a symbol-manipulation trick without the meaning that makes the result trustworthy in a new situation. The rules are meant to preserve, not replace, the physical interpretation of the derivative as local sensitivity.

Formal model: naming each rule precisely

Collect the rules with every symbol defined. For functions u of x and v of x, both differentiable at x, the sum rule states the derivative of u plus v equals u prime plus v prime. The product rule states the derivative of u times v equals u prime times v, plus u times v prime. The quotient rule states the derivative of u divided by v, provided v is not zero, equals (u prime times v, minus u times v prime), all divided by v squared. The chain rule states that for a composite function f of g of x, the derivative with respect to x equals f prime of g of x, times g prime of x, often written in Leibniz notation as: dy over dx equals dy over du, times du over dx, where y depends on an intermediate quantity u which in turn depends on x, a notation that visually suggests the two ratios chaining together, canceling the intermediate du as if it were an ordinary fraction, though this cancellation is a mnemonic, not a literal algebraic step.

Lineage

Gottfried Wilhelm Leibniz stated versions of the product and quotient rules as early as 1684 in his Nova Methodus, working from the same infinitesimal reasoning that founded his calculus notation, and Isaac Newton, working independently in England, arrived at equivalent results for combining fluxions of related quantities. Both approaches, though algebraically productive and correct in their conclusions, rested on the same unrigorized infinitesimal foundation that limited the derivative itself until the nineteenth century. Once Cauchy and Weierstrass placed the derivative on the epsilon-delta limit footing, each differentiation rule could be, and was, re-derived as an honest theorem from that limit definition, exactly the derivation carried out above for the product rule, closing the gap between a useful calculating trick and a proven mathematical fact.

The strongest case for it

These rules earn their place by making differentiation of realistic, compound models practical without sacrificing rigor: almost no quantity worth modeling, whether a garden's area, a heated house's fuel cost, or a rocket's changing mass and velocity together, is a single simple function, and returning to the raw difference-quotient limit for every compound expression would make calculus useless for real work. Because each rule is derived, not assumed, from the same limit definition underlying the derivative itself, a derivative computed using several rules in combination is exactly as rigorous as one computed directly from the limit, no accuracy or meaning is lost by taking the shortcut. The rules also compose: the chain rule alone, applied repeatedly, handles arbitrarily deep nestings of functions inside functions, which is precisely the structure of most real scientific and engineering models.

The strongest case against it

The most common failure is treating the rules as a lookup table to pattern-match against, rather than a description of how rates of change genuinely combine; a learner who forgets which factor multiplies which in the product rule, or forgets the squared denominator in the quotient rule, has usually lost the underlying story, the length spread across the width and vice versa, rather than merely mis-stated a formula. A second genuine limitation is that the rules require each component to be differentiable; a compound model built from a piece with a sharp corner or a jump at the point of interest cannot be differentiated there by these rules or by any other method, because the underlying pieces themselves fail the more basic requirement established in the derivative's own definition. A third misconception is applying the chain rule's Leibniz notation as if the intermediate quantity truly cancels like an ordinary fraction; the cancellation is a useful mnemonic for remembering the rule's shape, not a literal algebraic operation, and treating it as literal can lead to errors when the intermediate quantity is itself a vector or a more exotic mathematical object in later, more advanced settings.

Where it stands now

The sum, product, quotient, and chain rules have been proven theorems, not conventions, since the nineteenth century's rigorization of the derivative, and there is no live mathematical dispute about their correctness or scope for ordinary differentiable functions. What remains an active and valuable skill, rather than an open mathematical question, is applying the rules correctly to increasingly deeply nested and physically meaningful compound models while retaining, at every step, a clear sense of what each piece of the resulting derivative represents in the original problem.

Test yourself

A company's total advertising reach is modeled as R of t equals N of t times C of t, where N of t is the number of viewers reached at time t and C of t is the average number of purchases each viewer makes, and both N and C are changing over time independently. At a certain moment, N equals 40,000 viewers and is growing at 500 viewers per week, while C equals 1.2 purchases per viewer and is growing at 0.05 purchases per viewer per week. Use the product rule, with each term's physical meaning stated explicitly, to compute how fast total reach in expected purchases is growing at that moment. Then suppose instead that C of t is itself a function of an advertising budget B of t, which is changing over time, so that C depends on t only through B; explain, using the chain rule and without recomputing from the raw definition, how you would find how sensitive total purchases are to a change in the advertising budget directly.