Zero (Root)

A function is a calculation rule: you feed in a number and get a number out. For some inputs, the result is exactly zero. Such inputs are called zeros. If you plot the function on a coordinate system, the zeros are the points where the curve crosses or touches the horizontal axis. An example: the rule “take a number, multiply it by itself, and subtract 9” yields zero at 3 and at -3. This function therefore has two zeros. Other functions have only one, infinitely many, or none at all.

Why engineers look for zeros

The question “where does something become zero?” underlies many practical tasks. When does a vehicle come to a stop? That’s the point where the speed is zero. When is an account balanced? Then the difference between income and expenses is zero. In all these cases, one is looking for the zero of a function.

Even more important is a trick: any equation can be rewritten as a search for a zero. If you want to know when two quantities are equal, you simply form their difference. When this difference becomes zero, the quantities are equal. That’s why a method that finds zeros is, in reality, a method that solves equations. That’s exactly what computers constantly need.

One special case is particularly valuable: zeros of the slope. The slope describes how strongly a quantity is currently increasing or decreasing. Where the slope is zero, an increase ends and a decrease begins. That’s where maxima and minima lie. Anyone looking for the lowest price or the least material consumption is, in the end, looking for a zero.

From trial and error to Newton’s method

For simple functions, zeros are found by rearranging the equation. For complicated functions, this often no longer works. Then one calculates approximately. The simplest method is a kind of systematic guessing game: you know a point with a positive result and one with a negative result. Somewhere in between, the curve must cross the axis. You check the midpoint and keep the half that still contains the sign change. After twenty steps, the interval is about a millionth as large as at the start.

Newton’s method works faster. You choose a starting value and draw a line at that point that has exactly the curve’s slope. This line hits the axis at a certain point. From there, you start again. The estimate becomes dramatically more accurate from round to round, often doubling the number of correct digits per step.

A common misconception is that such methods always work. Newton’s method can drift away with a poor starting value or oscillate back and forth between two values. Many also confuse zero and minimum: a zero is the point where the function value is zero, not the point with the smallest value. The two only coincide by chance sometimes.

Zeros in machine learning and engineering

An AI model learns by shrinking its own errors. The error is measured as a number that depends on thousands of adjustable values. Training searches for settings where this error is as small as possible. Mathematically, this means: one searches for points where the slope of the error is zero. Training large language models is therefore, at its core, a gigantic search for zeros.

Outside of AI, too, the principle is built in everywhere. A calculator extracts roots by internally computing a zero. Simulation programs in the automotive industry solve their equations this way. In finance, the yield of an investment results as the zero of a series of payments. Even a navigation device uses this method to determine its own position from satellite signals.

In school, the term usually comes up in connection with parabolas and the formula for quadratic equations. That’s the same idea in simple form. Anyone who has understood the principle there will recognize it later in almost every technical computational method.