Two axes, one speed, and a diagram that does the arguing

Special relativity has a reputation for being counterintuitive, and it is — but only in its conclusions. Its content is a single postulate and a diagram, and once the diagram is drawn correctly the conclusions stop requiring any intuition at all. They can be read off with a ruler.

The postulate is that light travels at the same speed for every observer, however they are moving. That sentence is the whole of the strangeness. Everything else follows from taking it seriously enough to draw.

Choosing units that make light diagonal #

The first decision is to plot $ct$ rather than $t$ up the vertical axis — time multiplied by the speed of light, so that both axes carry the same units.

The consequence is that light travels at exactly forty-five degrees. In one second it covers one light-second, so its path on the diagram has slope one, and this is true no matter who draws it. The postulate is now a geometric constraint on the picture rather than a fact to be remembered: any observer’s diagram must have light at forty-five degrees.

Everything else on the diagram is slower than light, so every material object’s path — its worldline — is steeper than forty-five degrees. An object at rest has a vertical worldline, since it stays at one place while time passes. An object moving fast has a worldline tilted toward the diagonal, and it can approach the diagonal without reaching it.

A worldline is not a trajectory. It is the whole history of an object, laid out at once, and the object does not move along it. This takes some getting used to and it is the source of most of the diagram’s power: questions about what happens over time become questions about the shapes of curves in a fixed picture.

The axes that scissor #

Now the part that could not be guessed. Another observer, moving past at speed $\beta$ (as a fraction of $c$), draws their own axes on the same diagram — and their axes are not perpendicular.

Their time axis is easy to place. It is the set of events happening where they are, at successive times, which is just their own worldline: a line of slope $1/\beta$.

Their space axis is the surprise. It is the set of events they consider simultaneous with the origin, and it comes out tilted by the same angle the other way, with slope $\beta$. Both axes rotate toward the light line, from opposite sides, by equal amounts.

The scissoring is forced, and the reason is worth stating because it is the whole mechanism. The light line must bisect the two axes, because light must have speed one in every frame. If one axis tilts toward the light line, the other must tilt by the same amount, or the bisection fails. The constancy of light speed is the symmetry of the scissor.

This is why the transformation between frames is a Lorentz transformation and not a Galilean one. In Galilean relativity the time axis tilts and the space axis does not — everyone agrees on simultaneity — and the light line does not stay put. Keeping the light line fixed costs exactly one thing: the agreement about what is simultaneous.

The cone, and what it forbids #

Strip away the frames and the diagram still carries its most important content: the shaded wedges.

The upper wedge is everything that can be reached from here and now by anything travelling at or below light speed. The lower is everything that could have reached here. Together they are the light cone, and they divide spacetime into three regions with entirely different characters.

Events inside the future cone can be caused by the event at the centre. Events inside the past cone can have caused it. And everything outside the cone — the left and right regions, which are much larger than they look — can neither cause nor be caused by it. No signal connects them. They are elsewhere, and the word has to be that vague because “at the same time” is not available.

The causal structure is the part of the diagram every observer agrees on. Tilt the axes as far as they will go and events inside the cone stay inside; events outside stay outside. That invariance is not a lucky property — it is what makes relativity consistent, because the order of two causally connected events is the same for everybody, while the order of two causally disconnected events is not.

So relativity does permit disagreement about which of two events happened first. It permits it exactly and only when nothing could have travelled between them, which is precisely the case in which nobody can be wrong about anything that matters.

The clock that makes it concrete #

The scissoring axes are geometry, and geometry can feel like a convention until something mechanical produces the same number.

A clock made of two mirrors and a light pulse makes the tilt concrete. Set it moving at the same half of light speed the tilted axes belong to and the light covers the hypotenuse instead of the height — and since its speed is fixed for both observers, the moving clock must take longer to tick. Nothing about mirrors enters the conclusion: any clock carried alongside must agree with this one, or the two could be compared to detect uniform motion.

This is the same $\gamma$ that sets the tilt of the axes, arrived at from Pythagoras rather than from a coordinate transformation. That two such different routes give the identical factor is the strongest indication that the diagram is describing something rather than encoding a choice.

Two events simultaneous for one observer are not simultaneous for another, and the tilted slice is the moving observer’s own space axis. The disagreement is not an artefact of the drawing — it is what the drawing is for, and it is the one consequence of the two postulates that a reader will not believe without a picture: the space axis has to tilt, because the light line has to bisect the two axes at every speed.

The second figure isolates the price paid for keeping light at forty-five degrees. Tilting the space axis means redefining which distant events count as “now”, and there is no way to keep the light line fixed without paying it.

The quantity everyone agrees on #

Different observers assign different times and different distances to the same pair of events. There is nonetheless a number they all compute identically:

$$s^2 = (c\Delta t)^2 - (\Delta x)^2.$$

The invariant interval. It plays the role that distance plays in ordinary geometry — the thing that survives a change of coordinates — and the minus sign is what makes spacetime geometry different from Euclidean geometry in every consequence.

The minus sign is not a technicality. It means the interval can be positive, negative or zero, and the three cases are the three regions of the cone. Positive: the events are timelike separated, one can cause the other, and the interval is the time a clock would measure travelling between them. Negative: spacelike, no causal connection, and the interval is a length measured by someone for whom they are simultaneous. Zero: lightlike, connected only by light, and — remarkably — the interval along a light ray’s path is always zero, however far it travels.

Rotations in ordinary geometry preserve $x^2 + y^2$ and turn circles into circles. Lorentz transformations preserve $t^2 - x^2$ and turn hyperbolas into hyperbolas, which is why the axes scissor rather than rotate: the “circles” of spacetime geometry are hyperbolas asymptotic to the light lines, and nothing crosses an asymptote.

Reading the standard results off the picture #

Once the axes are drawn, the famous effects are exercises with a straightedge.

The factor is within half a per cent of one until a tenth of light speed and then climbs without limit as the speed approaches the light line. That shape is why the whole subject went unnoticed for two centuries — every measurement anybody could make sat on the flat part — and why the diagram is the more useful object than the formula: the flat part is where the tilt is invisible and the steep part is where it is the only thing happening.

Time dilation. A moving clock’s ticks are marked along its tilted time axis. Projecting them onto the stationary observer’s time axis gives intervals longer by $\gamma$. The light clock shows the same result with a mechanism attached.

Length contraction. A moving rod’s length is measured between its two ends at one moment — and which moment depends on whose space axis is used. Slicing along the tilted axis cuts the rod’s worldsheet at an angle and gives a shorter length, by the same factor $\gamma$.

The velocity limit. Adding velocities corresponds to adding the tilts of axes, and tilts add in a way that never reaches forty-five degrees. Two speeds each nine-tenths of light combine to about 0.994, not 1.8. The light line is an asymptote in every frame, which is a much stronger statement than a speed limit.

The curve above says something else that is easy to miss: at everyday speeds, $\gamma$ is indistinguishable from one. At orbital speed it differs in the ninth decimal place — enough to matter for satellite navigation, which corrects for it, and far too little to matter for anything else. Relativity does not contradict everyday mechanics; it contains it, in a limit that covers all ordinary experience.

That containment is the standard relationship between a new theory and the one it replaces, and it is worth recognising as such. Newtonian mechanics is not wrong; it is the small-$\beta$ expansion of this, exact nowhere and excellent everywhere anybody was looking, in precisely the way the small-angle pendulum is the small-amplitude expansion of a harder problem. Both approximations fail quadratically at first, which is why both survived so long unchallenged.

What the minus sign costs #

Calling $s^2 = (c\Delta t)^2 - (\Delta x)^2$ a geometry invites every habit built up from ordinary geometry, and most of those habits are wrong here. The minus sign is one character and it invalidates them individually.

The interval is not a distance. A distance is zero only between a point and itself. This one is zero between every pair of events connected by a light ray, however far apart they are — the entire light cone is at zero interval from its apex. A geometry in which distinct points are at zero separation is not something Euclid would recognise, and no intuition about “nearby” survives it.

The triangle inequality reverses. In the plane, a detour is longer than the direct route. In spacetime, a detour through a third event is shorter in proper time than the straight worldline between the endpoints. That single reversal is the twin paradox, and it is the reason a free-falling path maximises rather than minimises. Anyone importing the ordinary inequality gets every one of these results backwards, and the mistake is invisible because the reasoning looks identical.

The circles are hyperbolas, and they are unbounded. The set of events at a fixed interval from the origin is a closed curve in ordinary geometry and an open one here, running off to infinity along the light lines. So the analogue of a rotation moves points arbitrarily far along a hyperbola, which is why an arbitrarily large boost is possible while an arbitrarily large speed is not.

The page lies about scale. This is the practical charge, and it catches everyone once. The tilted axes are drawn with the same unit spacing as the stationary ones, and their units are genuinely longer on the page — by a factor set by the invariant hyperbola. A ruler laid on the diagram measures a Euclidean length, and the diagram is about a quantity that is not one. The picture is exactly reliable for structure: which events are in which cone, which lines are parallel, which order things occur in for a given observer. It is unreliable for magnitude until the calibration hyperbolas are drawn in, and they almost never are.

So the diagram buys an enormous amount of intuition and charges for it in a currency that is easy to overlook. What it gives is a way to see that time dilation, length contraction and simultaneity disagreement are three readings of one tilt. What it takes is every quantitative instinct trained on maps.

Where the model stops #

The diagram is of flat spacetime, and flat spacetime is the special case that gives special relativity its name.

No gravity. Mass curves spacetime, and on a curved diagram the light cones tilt from place to place. Where they tilt far enough that all future directions point inward, the surface is a black hole’s horizon. Nothing in the flat picture hints at this, and general relativity is what replaces it.

No acceleration, in the axes. The tilted axes belong to an observer moving uniformly. An accelerating observer’s worldline is curved and their “space axis” rotates as they go, which is what makes the twin paradox a real asymmetry rather than a contradiction: one twin’s frame is not a single frame at all.

One spatial dimension. The diagram draws $x$ and suppresses $y$ and $z$, so the light cone is a wedge rather than a cone and phenomena involving direction — aberration, the Doppler effect off-axis, the fact that a fast-moving sphere looks rotated rather than flattened — cannot appear in it.

Classical, throughout. Worldlines are definite paths. Quantum mechanics does not permit that, and the combination of relativity with quantum theory is quantum field theory, in which the diagram survives as a backdrop and the sharp trajectories do not. The object that survives best is the field, which was already the right way to think about electromagnetism and turns out to be the right way to think about matter as well.

One more caution, about the drawing rather than the physics. The tilted axes are drawn at the same scale as the stationary ones, and they should not be: a unit along the tilted axis is longer on the page than a unit along the vertical one, by a factor that follows from the invariant hyperbola. Measuring lengths off the diagram with a ruler gives wrong answers for exactly this reason. The diagram is reliable for structure — what is inside which cone, which lines are parallel to which — and needs the calibration hyperbolas before it is reliable for quantities.

The diagram nobody wanted #

Minkowski had taught Einstein mathematics at the Zurich Polytechnic and remembered him as a student who skipped lectures. Three years after the 1905 paper, he gave an address in Cologne that opened by declaring that space by itself and time by itself were doomed to fade into mere shadows, and that only a union of the two would preserve an independent reality. The union was the geometry above.

Einstein’s first reaction was that it was superfluous learnedness — a mathematician dressing up a physical result in machinery it did not need. The complaint was not unreasonable. Every result in this essay can be obtained algebraically from the Lorentz transformation, and had been, and the diagram adds no prediction whatsoever. Minkowski died the following year, at forty-four, of a ruptured appendix, and never saw what happened next.

What happened is that Einstein spent the years from 1907 trying to include gravity, and found that he could not do it with algebra. Gravity turned out to be curvature, curvature is a statement about a geometry, and a geometry needs the object Minkowski had supplied: an invariant interval, defined pointwise, from which everything else is derived. By 1912 Einstein was working in exactly that language, with Marcel Grossmann teaching him the Riemannian machinery that generalises it, and general relativity as it exists is unthinkable without it.

The episode is a fair warning about reformulations. A restatement that adds no predictions is easy to dismiss on the grounds that it adds no predictions, and the grounds are correct. What a reformulation changes is which generalisations are reachable from it — and that cannot be assessed at the time, because the generalisation has not been made yet. Minkowski’s diagram was superfluous for special relativity and indispensable for what came after, and both statements were true when Einstein made the first one.

The ladder from here #

Later rungs: the Lorentz transformation written out, and the hyperbolic rotation it really is. Rapidity, which does add linearly, and why it is the better variable. The calibration hyperbolas that make the diagram metric. Velocity addition. The relativistic Doppler effect. Four-vectors, and the reformulation in which every equation is manifestly frame-independent. Energy and momentum as one four-vector, and $E = mc^2$ as its length. Collisions redone relativistically, where mass stops being conserved and energy does not stop. And the twin paradox, which is a statement about the lengths of two paths and contains no paradox once it is drawn.

Beyond the rungs listed above, this anchor eventually meets the field picture coming the other way: electric and magnetic fields turn out to be one object sliced by a frame, and the tilt of the axes in these figures is exactly what does the slicing.