mathronic

Essentials · Derivatives

Derivatives

A fifteen-minute read: what it is for, where it is used

The weather app on your phone also shows the times of sunrise and sunset. In Istanbul, the day on 28 September was 11 hours 54 minutes long; a week later, on 5 October, it was 11 hours 35 minutes. In one week the day got 19 minutes shorter.

The shortest day of the year is 21 December. Until then, the days will keep getting shorter.

The text below answers this question. When you have read it, come back to your guess: what held up, and what changed?

The length of the day changes every day, but not at the same rate every day. To know how it changes, you do not have to measure every single day. Day length follows a rule: it falls from the longest day in June to the shortest day on 21 December, then rises again, and repeats itself every year. The graph of this rule is roughly a wave, much like a sine curve.

To find the rate of change at a single instant, you take a small interval around that instant and divide the change over the interval by the interval's length. Then you look at the number these ratios approach as the interval shrinks; that number is a limit. You do not have to do this separately for every day. If you know the rule, the rate of change at every instant can also be read off a single rule; how it is read off is the subject of the book. The whole topic fits into one sentence:

If you know the rule for something, you also know how fast it is changing at every instant: change has a rule of its own.

This rule of change is called the derivative. If we call the length of day number x of the year f(x), its derivative is written f′ and read "f prime"; the small mark at the top is not an exponent. f′(x) is how fast the day length is changing on day x. On the graph, it is how steep the curve is at that point.

This gives the answer to the opening question. The wave's graph is steepest halfway between its peak and its trough, and it levels off at the peak and at the trough. Day length changes fastest around the days when day and night are roughly equal in length (around 20 March and 23 September); the 19 minutes a week seen in Istanbul between late September and early October is close to that fastest rate. As the shortest day approaches, the wave levels off: between 14 and 21 December the day gets only about one minute shorter. On the shortest day the change stops for an instant, and then the days start to get longer.

Where the value is at its peak or its trough, the change stops; the change is fastest where the value is halfway between them. Knowing the value on a single day is not the same as knowing how it is changing on that day; but if you know the whole rule, you can work out the rule of change as well.

Where it is used

Distorted high notes · in audio amplifiers

An audio amplifier takes the weak signal from a microphone and makes it bigger. But no amplifier can change its output as fast as it likes. The OPA991 amplifier, described in a 2026 application note from Texas Instruments, can change its output by at most 0.5 volts per microsecond, that is, per millionth of a second, when the signal is a smooth wave. The number of times a wave repeats per second is called its frequency; its unit is the hertz. In the note's experiment, a wave of 10 V peak-to-peak (10 volts from its peak to its trough) comes out clean at a frequency of 10,000 hertz; at 20,000 hertz it distorts and turns into a triangle.

Can you know at which frequency the distortion will start, without trying it?

A real sound is a mix of many waves, so the engineer works out the worst case: the highest note, at the loudest level, treated as a single wave. Once you know the wave's rule, you also know the rule for how fast it changes. The wave changes fastest as it passes through zero, and its rate of change at that moment is proportional to the frequency and to the wave's peak-to-peak height. By the note's formula, for a 10 V peak-to-peak wave, the amplifier's maximum rate is exceeded at about 15,900 hertz: between 10,000 and 20,000, exactly where the experiment showed it. That is why the engineer chooses an amplifier fast enough to keep up with the moment when the signal changes fastest.

There is a price: according to the note, amplifiers generally draw more current to be faster. The OPA991 has an internal speed-up circuit that kicks in during sudden jumps and partly breaks this link between speed and current; with a smooth wave, however, the limit stays at 0.5 volts per microsecond, and distortion starts when the wave needs a faster rate than that.

Known: the rule of the signal and the amplifier's maximum rate of change · Found: the frequency at which distortion starts · Where the rule breaks: if you measure distortion with an instrument rather than by eye, it appears to start at a lower frequency, because an instrument also picks up distortion too small to see by eye; if the signal contains sudden jumps, the speed-up circuit makes the distortion start somewhere else.

The supply line that collapses for an instant · on circuit boards

On a circuit board, many parts are powered from the same 3.3-volt line. Next to the parts sit small energy stores called capacitors, which keep the supply steady; their capacitance, measured in microfarads, tells you how much energy they can hold at a given voltage. In a 2014 application report from Texas Instruments, a 100-microfarad capacitor connected directly to the line draws 6.88 amperes for an instant, and the line's voltage collapses from 3.3 volts to 0.96 volts.

Before you switch something on, how do you know how much current it will draw at the moment you do?

The current a capacitor draws is proportional to how fast its voltage is changing at that moment: current = capacitance × rate of change of voltage. If the voltage rises all at once, its rate of change is very large, and so is the current. The engineer runs the calculation backwards: first you decide what current you can accept, then you work out how slowly the voltage must rise. In the report's example, this time for a 22-microfarad capacitor and a limit of 600 milliamperes, the rise to 3.3 volts must take at least 22 microfarads × 3.3 volts ÷ 0.6 amperes = 121 microseconds. A part called a load switch does this; in the report's measurement, the peak current, seen at the steepest moment of the rise, stays at 392 milliamperes. The price: switching on takes longer, and the board needs one more part.

Known: the capacitance, the voltage and the acceptable current · Found: the shortest time the voltage rise may take · Where the rule breaks: if the voltage does not rise at a steady rate, the current follows the rate at the steepest moment, not the average rate of the rise.

The power cut after a lightning strike · in the power grid

Imagine you are designing a protection relay that connects a solar panel on a house roof, or a small power station, to the grid. A relay is a switch that breaks a connection by itself when a certain condition occurs. Its job is this: if a fault cuts the station off from the grid, the station must not carry on running on its own; a line that stays live is dangerous to anyone who comes near it, and machines can be damaged when the grid comes back. The grid voltage is a wave that swings back and forth fifty times a second; as in the amplifier box, this number of repetitions per second is called the frequency.

How does the relay tell that it has been cut off from the grid, and what could fool it?

If, at the moment of disconnection, the station's output does not match the consumption in its area, the frequency in that area starts to change quickly. That is why some relays watch the rate of change of the frequency and break the connection when that rate crosses a threshold. In Great Britain on 9 August 2019, for relays installed with older settings, this threshold was 0.125 hertz per second.

At 16:52 that day, lightning struck a transmission line; according to the official investigation, this was not an unusual fault. The fault shifted the voltage wave forward or backward in time for an instant (this is called a phase jump), and a different kind of relay, one that watches for such shifts, disconnected about 150 megawatts from small power stations (a megawatt is a million watts). But within one second, two large power stations, independently of each other, reacted to the fault in an unexpected way and lost output: an offshore wind farm lost about 740 megawatts when its voltage control system became unstable, and the steam turbine of a gas-fired power station lost about 240 megawatts. With generation now below consumption, the grid's frequency fell, and it fell faster than 0.125 hertz per second. Once the threshold was crossed, the relays at small power stations that watch the frequency's rate of change disconnected roughly another 350 to 430 megawatts of generation. The relays worked as they had been set to; but the threshold was so sensitive that it could not tell "I have been cut off from the grid" apart from "a power station has dropped out somewhere on the large grid".

The losses exceeded the reserve power that the grid operator keeps ready for a sudden loss. The frequency held at 49.1 hertz; about a minute after the fault, when another unit at the same gas-fired station dropped out, it fell to 48.8 hertz, and automatic protection cut the power to about 1.1 million customers for between 15 and 45 minutes.

A rule change approved four days before the event would raise the threshold at older small power stations to 1 hertz per second: the relay would break the connection only if the rate of change stayed above the threshold for half a second. The same rule change also required the removal of the relays that watch for an instantaneous shift in the wave; according to the rule change, these relays were easily triggered by mistake during temporary faults. The two decisions point the same way: trust a change that lasts for a while, not the jump of a single instant. The deadline for completing the change in the field was 31 August 2022. The price: a less sensitive relay may notice a real disconnection later, or not at all, and the station may carry on running on its own as a live "island" cut off from the grid; the rule change discussed this risk openly and judged it acceptable.

Known: the frequency and its rate of change · Found: whether the station has been cut off from the grid · Where the rule breaks: if the threshold is too sensitive, an ordinary loss on the large grid is mistaken for a disconnection; if generation happens to match consumption, a disconnection may go unnoticed altogether.

An engineer looks not only at the value but also at the change: the engineer works out how fast something is changing, sees rapid change coming and acts on it in advance, and does not trust this calculation at a corner where the change is undefined.

Its limits

Before you rely on a derivative, ask three questions: can it be calculated, can it actually happen, can it be trusted?

Can it be calculated? Where a rule jumps all at once, the rate of change at that point is not defined. If we picture the grid's wave as shifting all at once at the moment of the lightning strike, the voltage has no rate of change at that instant. An instrument that calculates a rate from a short interval containing that instant will still come up with a number, but that number is not a rate.

Can it actually happen? If the voltage jumps all at once, the capacitor formula predicts an unlimited current. No circuit draws an unlimited current; the 6.88 amperes in the report are limited by what the supply can deliver.

Can it be trusted? Apps round sunrise and sunset times to the minute. So the difference between two days can be off by about a minute: 20 December shows 9 hours 13 minutes and 21 December 9 hours 14 minutes, even though the shortest day is 21 December and the days keep getting shorter until then. The true lengths of these two days differ by much less than a minute; the difference you see comes from rounding. It is more reliable to read the rate of change from the difference over a week than from the difference between two consecutive days. But as the change slows down, as in December, even the weekly difference shrinks to the size of the rounding; then you need a longer interval or data given to the second.

In the language of formulas Open this if you like; let us work out the amplifier's maximum rate with numbers.

The wave used in the note is the one below. Here the letter f is not a rule but the frequency (in hertz; this is how the note writes it); Vpp is the wave's peak-to-peak height (in volts), and t is time (in seconds):

v(t) = (Vpp / 2) · sin(2π · f · t)

According to the note's calculation, the maximum rate of change of this wave, reached as it passes through zero, is:

maximum rate = π · f · Vpp

For now, take this as the note's result. For a 10 V peak-to-peak wave at 10,000 hertz, this is π · 10,000 · 10 ≈ 314,000 volts per second; since a second contains a million microseconds, that is 0.314 volts per microsecond, below the amplifier's maximum rate of 0.5. At 20,000 hertz it becomes 0.628, which is too fast for it. To find exactly where the crossover lies, set the wave's maximum rate equal to the amplifier's: π · f · 10 = 0.5 volts per microsecond = 500,000 volts per second. From this:

f = 0.5 · 1,000,000 / (π · 10) ≈ 15,900 hertz

Test the formula with a small interval. At 10,000 hertz the wave starts from 0. After 1 microsecond, 2π · f · t = 2π · 10,000 · 0.000001 = 2π · 0.01. With a calculator set to radians, find (10 / 2) · sin(2π · 0.01) and compare it with 0.314. To how many digits do they agree? Shrink the interval to 0.1 microseconds and do the same calculation (remember to divide the change by 0.1); what happens to the difference from 0.314?

In your own words

What is the derivative for? Write it in your own words, with an example from your own life.

Two topics together

This topic builds on limits. If you have the rule, a limit gives you the rate at an instant exactly; if all you have is measurements, you have to choose the interval. The idea the two share is this:

Nothing changes at a single instant; but by looking at the changes around that instant, you know its rate exactly.

An engineer calculates change from the difference between two nearby measurements, but knows that as the interval shrinks, measurement error makes up a growing share of the calculated rate, and chooses the interval accordingly.

If you want to go deeper

The long version of this topic (the book) covers the same idea in detail over six chapters: each chapter starts with an everyday situation, first asks for your guess, and then lets you try things yourself with small experiments. Each chapter opens once you have written your guess in the one before. The chapters are in Turkish for now.

Sources: Day length: U.S. Naval Observatory, Astronomical Applications (Istanbul, 41.01 N, 28.97 E). Amplifier: TI SDAA309, C. Featherstone, "Full Power Bandwidth and Op Amp Distortion" (2026). Inrush current: TI SLVA670A, A. Kaknevicius and A. Hoover, "Managing Inrush Current" (2014, revised 2015). Grid: Ofgem, "9 August 2019 power outage report" (3 January 2020) · National Grid ESO, "Technical Report on the events of 9 August 2019" (6 September 2019) · Ofgem, decision on DC0079, "Frequency Changes during Large Disturbances and their Impact on the Total System (Phase 4)" (5 August 2019) · Electricity North West, Accelerated Loss of Mains Change Programme (2019–2022).