The derivative as local sensitivity

In brief

A speedometer needle does not average your speed over the whole trip, it reports what is happening to your speed right now, this instant, even though "now" has no width and no duration to average over. That instantaneous reading is what the derivative captures in general: not the rate of change across some stretch, but the exact sensitivity of one quantity to another at a single, dimensionless point. The derivative matters because most real questions, how a bridge cable stretches under one more kilogram, how a population responds to one more predator, how profit shifts with one more unit sold, are asking exactly this: what happens right here, for a tiny nudge, not what happened on average over some arbitrary stretch.

The full treatment

First look: shrinking the interval on a falling object

Drop a ball and record its height over time. Between one second and two seconds after release, the ball falls from a height of a certain value to a lower one, and dividing the change in height by the one-second interval gives an average falling speed. Now shrink the interval: instead of one second to two seconds, measure from 1.0 seconds to 1.1 seconds, then from 1.0 to 1.01 seconds, then 1.0 to 1.001 seconds. Each time, compute the average rate of change over the ever-narrower interval. The numbers do not wander randomly, they settle down, homing in on a specific value as the interval shrinks toward zero width. That settled-on value is the instantaneous falling speed at exactly the 1.0 second mark, the number the speedometer would show if the ball carried one, and it is a single, exact number even though it was produced by an interval that shrank to nothing.

Building the idea: the difference quotient and its limit

Formalize the shrinking process. For a function f and a point a, pick a small nonzero number, call it h, representing the width of the interval, and form the difference quotient: (f of a plus h, minus f of a) divided by h. This is exactly the average rate of change of f from a to a plus h, written using h instead of two separate endpoints purely for convenience. As h is taken smaller and smaller, approaching zero without ever equaling it, the difference quotient may settle down to a single limiting value. If it does, that limit is called the derivative of f at a, and the process of finding it is called differentiation. Crucially, h is never actually set to zero, which would make the quotient zero divided by zero and meaningless; instead, the limit idea developed earlier, examining what a quantity is forced toward as an input approaches a forbidden value, is exactly what rescues this construction from nonsense.

What the derivative buys you: local sensitivity, not the whole shape

The derivative at a single point tells you the instantaneous rate there and nothing further away; it is deliberately local. This is a genuine limitation worth stating plainly: knowing the derivative at one point does not tell you the function's value anywhere else, nor guarantee anything about its behavior a short distance away, since the rate of change can itself be changing. What the derivative buys, in exchange for this narrowness, is precision: an exact answer to "how sensitive is the output to a small nudge in the input, right here," which an average computed over any wider interval can only approximate. This is why the derivative, not the average rate of change, is the tool for prediction: doubling a factory's raw material input by one unit changes output by approximately the derivative of the output function at the current input level, and that approximation gets better the smaller the intended nudge actually is.

Formal model: the derivative as a limit, and as a tangent slope

State the definition precisely. The derivative of f at a, written f prime of a, or using Leibniz's notation dy over dx evaluated at a, is defined as the limit, as h approaches zero, of the quotient (f of a plus h, minus f of a) divided by h, provided that limit exists. If it exists, f is called differentiable at a. Geometrically, recall that the difference quotient is the slope of the secant line connecting the points (a, f of a) and (a plus h, f of a plus h). As h shrinks toward zero, the second point slides along the curve toward the first, and the secant line rotates to become the tangent line, the single straight line that just grazes the curve at the point a without crossing it there. The derivative, then, is exactly the slope of that tangent line: it is the single number that best approximates the curve's steepness at one exact location.

Why continuity is necessary but the derivative demands more

A function must be continuous at a to have any chance of being differentiable there, because a jump or a hole would make the two-sided limit defining the derivative fail outright, the left and right difference quotients would not even agree on a target. But continuity alone is not enough. Consider the absolute value function at zero: it is perfectly continuous there, no jump, no hole, yet the difference quotient computed from the right, using positive h, always equals 1, while the difference quotient computed from the left, using negative h, always equals negative 1. The two one-sided limits disagree, so no single derivative exists at that sharp corner, even though the function is entirely continuous. Differentiability is therefore a strictly stronger, more demanding condition than continuity: every differentiable function is continuous, but not every continuous function is differentiable, and a sharp corner is the standard example of the gap between the two.

Lineage

The instantaneous rate of a changing quantity was the central object of Isaac Newton's 1671 Method of Fluxions, developed to describe the motion of physical bodies, where a "fluxion" was the instantaneous speed of a "fluent," a flowing quantity such as position. Gottfried Wilhelm Leibniz arrived independently at essentially the same idea a few years later, publishing in 1684 and introducing the dy over dx notation still in use, along with the systematic differentiation rules built on top of the derivative. Both men relied on intuitive, unrigorized talk of infinitesimally small quantities to justify shrinking the interval to zero, a gap that stood for over a century until the epsilon-delta limit definition, developed principally by Cauchy and Weierstrass in the nineteenth century, put the shrinking process on the fully rigorous footing used today, exactly the limit machinery this entry relies on.

The strongest case for it

The derivative earns its central place in mathematics and science because it answers, with exactness, the single most common practical question: what happens to the output for a small change in the input, right here. It underlies every prediction that says "for each additional unit," from the marginal cost of one more item produced to the sensitivity of a bridge's deflection to one more kilogram of load, to the instantaneous velocity a speedometer displays. It is also computable in practice, not merely definable in principle: once a handful of differentiation rules are established from this very limit definition, derivatives of complicated combined functions can be found quickly and mechanically, without recomputing a difference quotient from scratch every time, which is exactly the labor the next entry in this sequence saves.

The strongest case against it

The derivative is strictly local and it is a common and costly error to extrapolate it too far: the instantaneous rate at one point says nothing certain about behavior even a short distance away, and using it to predict a large change, rather than a small nudge, can be badly wrong if the rate itself is changing quickly nearby. A second real limitation is that not every function has a derivative everywhere, sharp corners, jumps, and vertical tangents all defeat it at specific points, so "differentiable" must be checked, not assumed, exactly as continuity had to be checked before it. A third common misconception is treating the derivative as measuring total change; it measures a rate, an instantaneous ratio, and recovering a total change from it requires the separate operation of accumulation, which is a later idea in this sequence, not something the derivative provides on its own.

Where it stands now

The definition of the derivative as the limit of the difference quotient has been the settled, unchallenged foundation of differential calculus since the nineteenth century's rigorization, with no live dispute about its correctness. What remains genuinely active, in applied settings far beyond pure mathematics, is deciding how far a local derivative can be trusted to stand in for actual behavior over a finite interval, a judgment call about the smoothness and predictability of the underlying process rather than a question the mathematics itself can settle.

Test yourself

A company's revenue as a function of the number of units sold is given by the formula: revenue equals 50 times units, minus 0.01 times units squared. Using the difference quotient definition directly, that is, by computing (revenue at 1000 plus h, minus revenue at 1000) divided by h and examining what happens as h shrinks toward zero, find the instantaneous rate of revenue change at exactly 1000 units sold. Then use that single number to predict, without recomputing revenue from scratch, approximately how much extra revenue one more unit sold at that point would bring in, and state explicitly why this prediction should not be trusted to estimate the effect of selling 500 more units instead of one.