Relativity, and why GPS needs it

The GPS in your phone works by asking satellites what time it is. If we built those satellites the way common sense says clocks should work, your blue dot would drift off by about ten kilometers every single day. It does not, because the engineers who built GPS treated time the way Einstein said it really is, not the way it feels. This is relativity as working infrastructure.

The one fact everything hangs on

Here is the strange, experimentally ironclad fact that starts it all: the speed of light in empty space is the same for every observer, no matter how fast they are moving. Throw a ball forward from a moving train and its speed, to someone on the ground, is the train's speed plus the throw. Shine a flashlight forward from that train and its light does not come out at the train's speed plus the speed of light. It comes out at exactly $c$, about 300,000 kilometers per second, for the person on the train and the person on the ground alike. This is the constancy of the speed of light. It sounds like a small technicality. It is not. Something has to give, and what gives is time itself. Light is the electromagnetic wave from electromagnetic waves and light, and its fixed speed is the seed of everything below.

A clock made of light, and why moving slows it

The cleanest way to see time bend is to build a clock out of the one thing whose speed cannot change. Bounce a photon straight up between two mirrors; one round trip is one tick. Now set that whole clock moving sideways and watch it from the ground. Play with the widget: slide the clock's speed up and the photon, seen from the ground, has to travel a stretched diagonal instead of a straight up-and-down.

Here is the trap springing shut. The diagonal path is longer. But the photon can only travel at $c$ - it cannot speed up to cover the extra distance in the same time. So the tick must take longer. From the ground, the moving clock ticks slower. This is not an illusion or a measurement error; the moving clock genuinely experiences less time. We call it time dilation.

The triangle in the widget is the entire proof, and it is just the Pythagorean theorem you already know. In one tick of the moving clock lasting a ground-time $t$, the photon climbs the fixed mirror gap (call the still-clock tick $t_0$, so that leg is $c{t}_0$) while the clock slides sideways by $vt$. The diagonal the light actually travels is $ct$. A right triangle ties them together:

Solve for $t$ and the sideways motion inflates every tick by one clean factor, written $\gamma$ (gamma). This part is derived, start to finish, from the picture:

At walking speed $\gamma$ differs from 1 in the fifteenth decimal place, which is why time feels absolute. At a GPS satellite's speed it differs just enough to matter, and that difference, multiplied by a day, is measured in microseconds.

Gravity bends time too

There is a second effect, and for GPS it is the bigger one. A clock deeper in a gravitational field - lower down, where gravity is stronger - runs slower than one higher up. This is gravitational time dilation, and it is a result of Einstein's general relativity that we will state rather than derive. A GPS satellite orbits about 20,000 kilometers up, where Earth's gravity is much weaker than at the ground. So its clock, from our point of view, runs faster than ours. Two effects, pulling opposite ways: its speed slows its clock, its altitude speeds it up.

The budget: two effects, one number, eleven kilometers

Switch the widget to the GPS tab. Here is the actual engineering ledger for a GPS satellite clock, per day, compared to a clock on the ground:

Thirty-eight microseconds sounds like nothing. But GPS finds you by timing how long a signal takes to arrive and multiplying by the speed of light - distance equals $c\times \text{time}$, exactly the trilateration from orbits and GPS. Light travels about 30 centimeters in a nanosecond, so a clock error of 38 microseconds becomes a distance error of $c\times 38\mu s\approx 11$ kilometers. Every day. The widget accumulates it live: leave the correction off and within roughly half a minute your position is already wrong by more than a car lane.

The fix is beautiful in its bluntness. The atomic clocks on GPS satellites are deliberately built to run slow on the ground, at 10.22999999543 MHz instead of a clean 10.23 MHz, so that once they are in orbit and speeding and high up, relativity nudges them to tick at exactly the right rate as seen from the ground. Relativity is not a correction bolted on afterward; it is machined into the hardware before launch.

The CS bridge: relativity is a distributed-systems problem

You already know that keeping clocks in sync across a distributed system is hard. GPS is the hardest version: the clocks are on spacecraft moving at different speeds and sitting at different gravitational potentials, so they drift apart by a systematic, predictable amount rather than random jitter. That is the good news for an engineer - a predictable skew can be modeled and removed exactly, which is precisely what the ground segment does.

C = 299_792_458          # speed of light, meters per second

net_drift_per_day = 38e-6  # seconds the satellite clock gains each day (45 - 7)

# position error is just how far light travels during the clock error
position_error_m = C * net_drift_per_day
print(round(position_error_m / 1000, 1))   # -> 11.4  kilometers per day

# time to blow past the ~5 m accuracy a car needs
rate_per_sec = net_drift_per_day / 86_400
print(round(5 / (C * rate_per_sec)))       # -> 38  seconds

Check yourself