The note that changes on approach, and the two ways of getting it

An ambulance passing at speed drops its pitch as it goes by, and the drop is not gradual — it happens over a second or two, centred on the moment of passing, and the note is steady on either side of it.

That last detail is the one worth noticing, because it rules out the explanation most people reach for first. The pitch is not falling because the vehicle is receding and the sound is getting fainter. Loudness and pitch are independent, the note approaching is constant and higher, the note leaving is constant and lower, and the whole change occurs while the vehicle is closest — which is precisely when its speed away from the listener passes through zero and then reverses.

Drawing where the crests are #

The whole effect is in that figure and needs no algebra to see.

A source emits a crest, which spreads outward at the speed the medium fixes as a circle centred on where the source was at the moment of emission. That last clause carries everything. The source then moves, emits the next crest from a new place, and that crest spreads as a circle centred there.

The result is a set of circles whose centres march forward. Ahead of the source, successive circles are closer together than they would be from a stationary source, because each was launched from further along. Behind, they are further apart. The spacing between crests is the wavelength, so the wavelength ahead is shortened and the wavelength behind is stretched.

The arithmetic is one line. In one period $T$ the wave advances $cT = \lambda$ and the source advances $vT$, so the gap between crests ahead is $\lambda - vT$ and behind is $\lambda + vT$. Since the medium carries all crests at the same speed $c$ regardless of who made them, the frequency heard is $c$ divided by the local wavelength:

$$f' = f\,\frac{c}{c - v} \ \text{(approaching)}, \qquad f' = f\,\frac{c}{c + v} \ \text{(receding)}.$$

An ambulance at 30 m/s with a 700 Hz siren is heard at 767 Hz approaching and 644 Hz receding — a drop of 123 Hz, close to three semitones, and one that happens entirely within the moment of passing.

The two cases that are not the same #

Here is the part that distinguishes this rung from the mnemonic, and it is a genuine physical asymmetry rather than a bookkeeping detail.

Move the observer instead of the source, and the derivation is completely different. The crests are now evenly spaced, since nothing was moving when they were emitted. What changes is the rate at which the observer runs into them, which adds their own speed to the closing rate:

$$f' = f\,\frac{c + v_o}{c}.$$

Both formulas increase the frequency when the two approach, and they are not the same formula. Approaching at 30 m/s in air, a moving source gives a factor of 1.0958 and a moving observer gives 1.0875 — a difference of three-quarters of a per cent, which for a 700 Hz siren is 6 Hz and comfortably audible as a mistuning — a difference far easier to hear as a beat against a reference than to judge as a pitch.

The asymmetry exists because there is a third participant that the description keeps quiet about: the air. Sound travels at a fixed speed relative to the medium, so “moving” has an absolute meaning here — moving with respect to the air — and the two cases genuinely differ in whether the source or the observer is the one doing it. Anyone who can measure the shift to better than a per cent can determine which of the two is moving through the air, without looking out of the window.

That is worth holding onto, because it is exactly the property that light does not have. The speed of light is the same for every observer, there is no medium to be at rest with respect to, and so the relativistic Doppler formula cannot possibly distinguish a moving source from a moving observer. It depends only on the relative speed, and it is

$$f' = f\sqrt{\frac{1 + \beta}{1 - \beta}},$$

which is the geometric mean of the two classical formulas. That the exact answer sits precisely between the two classical cases is not a coincidence: it is what happens when the asymmetry is removed by requiring the result to be symmetric under exchanging who is moving.

The frequency belongs to the source, and the wavelength does not #

The construction also makes clear which quantity is being changed, and the answer is not the obvious one.

Frequency belongs to the source and wavelength to the pairing of source and medium. The moving source has not changed its rate of vibration at all — it is emitting exactly as many crests per second as before. What it has changed is where each crest starts, and therefore how far apart they end up.

So the shift ahead of the source is fundamentally a change in wavelength, which is then read as a change in frequency by an observer standing in the medium. The shift for a moving observer is fundamentally a change in the rate of arrival, with the wavelength untouched. The two mechanisms are different, they produce nearly the same number, and the small discrepancy above is the fingerprint of that difference.

When the source keeps up with its own sound #

Pushing the speed toward that of the wave does something the formula announces plainly and the picture makes vivid.

As $v \to c$, the denominator $c - v$ goes to zero and the frequency ahead diverges. Every crest ever emitted arrives at nearly the same moment, because the source is nearly keeping pace with them. The energy that was spread over many seconds of emission arrives compressed into a fraction of one.

Past that speed the geometry changes character entirely.

Beyond Mach one the source outruns every crest it has emitted. The circles fall behind it, and there is a cone — with half-angle $\arcsin(c/v)$, computed and printed in the figure — that is tangent to all of them at once. Inside the cone the sound has arrived; outside it, nothing has been heard at all.

Two corrections to the popular account follow immediately from that picture. First, the boom is not an event that happens when the aircraft “breaks the sound barrier”: the cone exists for as long as the aircraft is supersonic, and it drags along behind it, so the boom is heard by everyone the cone sweeps over, continuously, along the whole flight path. Second, there is no barrier being broken. The divergence at Mach one is a divergence in the pressure amplitude of the crowding, which is a real and severe engineering problem, and not a discontinuity in the physics. The crowding is superposition doing exactly what it always does, with the phases arranged so that everything adds.

The same cone appears wherever something outruns the waves it makes. A boat faster than its own water waves leaves a wake with a definite half-angle. A charged particle moving through water faster than light moves in water — which is legal, since the speed limit is the vacuum speed — emits a cone of blue Cherenkov light, and the angle of that cone is a direct measurement of the particle’s speed — the one everyday case in which the vacuum speed limit is not the relevant one. Neutrino detectors are built around reading it.

What the effect is used for #

The shift is a measurement of one component of velocity, delivered instantly, without contact, and that has made it into an instrument in a remarkable number of places.

Speed cameras and radar bounce a wave off a moving object, which is the two-case problem run in succession — the object receives a shifted frequency as a moving observer, then re-emits it as a moving source — so the shift is doubled and the formula is the product of the two.

Doppler ultrasound measures blood flow by the shift off moving red cells, and the sign of the shift distinguishes flow toward the probe from flow away, which is why the images are red and blue.

Weather radar reads the radial velocity of raindrops, and a small region where the shift changes sign across a short distance is a rotating updraught. That signature is the basis of tornado warnings, and it is a direct reading of this construction.

Stellar spectra shift bodily, and the size of the shift gives the radial velocity. A star with a planet orbiting it wobbles around the shared centre of mass by a few tens of metres per second — the first exoplanet found around an ordinary star, in 1995, was detected by a 59 m/s wobble, a shift of two parts in ten million in the wavelength of every line at once. The technique works because “every line at once” is a far stronger signal than any single line.

And the expansion of the universe was found this way, or nearly. Distant galaxies show their spectral lines shifted toward longer wavelengths, by more the further away they are, which is Hubble’s law.

What the measurement costs #

Every one of those instruments is buying the same thing, and the price is identical in all of them: the shift reports one component of the velocity and is completely blind to the other two.

Only motion along the line joining source and observer changes the crest spacing. A source crossing the field of view at any speed whatever produces no classical shift at all. So a Doppler measurement is a projection, and recovering the actual velocity requires the angle to be supplied from somewhere else entirely.

That cost is paid, visibly, in each application. A radar gun aimed at an angle $\theta$ to a vehicle’s path reads $v\cos\theta$, which is always an underestimate — the cosine error, which is why enforcement practice requires near head-on geometry and why the error is in the driver’s favour rather than the operator’s. Doppler ultrasound has the same problem in a worse form: the angle between the beam and the vessel is not known independently, so the sonographer estimates it from the image, and because the correction divides by $\cos\theta$ the error grows steeply. At 60° a five-degree misjudgement shifts the reported velocity by about 15 per cent, which is the difference between a normal and a stenotic reading.

The astronomical case is the one where the missing component is famous. A star wobbling because of an orbiting planet is measured only along the line of sight, so what is determined is not the planet’s mass but $m \sin i$, with $i$ the unknown inclination of the orbit. Every mass from that method is a lower bound. A system seen nearly face-on produces almost no radial signal and is missed entirely, which biases the whole catalogue toward orbits seen edge-on — and the bias was not a detail, because it interacted with the early conclusion that hot Jupiters were common. Transit surveys, which have the opposite geometry requirement, were needed before the population could be described honestly.

The second cost is that the shift measures velocity, not distance or position, so it must be combined with something else to locate anything. Weather radar pairs it with the ranging that gives the echo’s distance; a Doppler navigation system integrates it to get displacement and accumulates error while doing so.

The third is subtler and worth naming because it limits precision rather than accuracy. Measuring a frequency to a given fractional precision requires observing for at least the reciprocal of that precision in cycles — the same time–bandwidth constraint that stops a short note from having a pitch. A shift of two parts in $10^7$, which is what an exoplanet wobble amounts to, therefore cannot be extracted from a brief look at a single line. It is recovered by cross-correlating thousands of lines across the whole spectrum at once, which buys the necessary precision from the number of lines rather than from the length of the observation.

Where the model stops #

Four limits, in increasing order of how badly the naive picture fails.

The medium must be still. Every formula above measures speeds relative to the medium, so a wind changes both the effective wave speed and which of the two cases applies. Sound carried downwind arrives with no shift attributable to the wind itself, since source and observer are both stationary in the air’s frame — but a wind shear, with different speeds at different heights, bends the sound and is why distant noise is louder downwind.

The speeds must be along the line joining the two. Only the component of velocity toward or away from the observer produces a classical shift. A source passing at closest approach is moving purely sideways at that instant and is heard, momentarily, at exactly its emitted frequency — which is why the pitch slides through the true note rather than jumping past it.

Relativity adds a shift with no classical counterpart. For light, a source moving purely sideways still shows a shift, because its clock is running slow and its emission rate is therefore reduced. This transverse Doppler effect is pure time dilation with no geometry in it at all, it was measured by Ives and Stilwell in 1938, and it is the cleanest confirmation of time dilation available from a spectrum.

And cosmological redshift is not this effect. This is the most consequential misreading in the list. Distant galaxies are not, in general, moving through space away from an observer; the space between is expanding, and the light is stretched in transit. The two descriptions agree numerically for nearby galaxies and diverge badly for distant ones — the naive Doppler reading of a high redshift gives a recession speed exceeding that of light, which is not a violation of anything but a sign that the formula has been applied outside its domain. The distinction is not pedantry: the two accounts predict different relationships between redshift and brightness, and the measurements agree with the expansion account.

Doppler himself, publishing in 1842, was arguing about something else entirely and was wrong about it — he proposed the effect to explain the colours of double stars, which it cannot do, since the required speeds are impossible and the eye’s colour response would not read the shift that way. The effect was confirmed in 1845 by Buys Ballot, who hired a locomotive, put trumpeters on it, and stationed musicians beside the track to write down what they heard. The measurement was correct and the theory it was proposed for was not, which is a more common pairing than the tidy accounts admit.

The ladder from here #

Later rungs: the relativistic formula derived rather than quoted, and the transverse case treated separately. Radar and the double shift. The Doppler width of a spectral line, which measures the temperature of the emitting gas because the emitters are moving in all directions at once. Beat frequencies as the practical way to measure a small shift, which is comparing two large numbers by their difference. The Mach cone and shock formation. Cherenkov radiation and its use as a velocity gauge. Doppler cooling, in which the shift is used to make an atom absorb preferentially when it moves toward the laser, and so to slow it. And the cosmological redshift treated on its own terms, where the stretching happens to the light rather than to the source.