THE PRINCIPLE OF LEAST ACTION

Throw a ball. It leaves your hand at point A, lands on the floor at point B, and the whole thing takes, say, one second. Between those two events, the ball traced out exactly one trajectory — the parabola we computed in FIG.04 from Newton's laws.

Stop. Think about how strange this is. You fixed the start, you fixed the end, you fixed the time. Of all the curves you could draw between A and B, the ball picked one. It could have dawdled up high and plunged at the last moment. It could have skimmed low and hopped up to catch B. It could have wriggled. It didn't. It drew a clean parabola.

There is. It was spotted in 1744 by a man who had recently measured the shape of the Earth by freezing to death in Lapland, and who believed the answer was theological. He was the first to state it in the language we still use. His name was Maupertuis, and the quantity he baptised is called the action.

But the idea is older. It is older than Newton. It is older than calculus. We need to start with light.

In 1662, Pierre de Fermat — the lawyer from Toulouse who in his spare time founded number theory and nearly invented calculus — published a letter claiming that he could derive the whole law of refraction from a single sentence.

The sentence was: light travels between two points along the path that takes the least time.

This sounded crazy. The accepted explanation, due to Descartes, was that light was a kind of projectile and the bending at the surface was a momentum argument. Fermat's statement was different in kind. It was not a mechanism. It said nothing about what light was. It said only that, whatever light is, it is lazy — and that laziness alone reproduces Snell's law.

The setup is clean. A point A sits in air (where light travels fast). A point B sits underwater (where light travels slower, by a factor of 1.33). Ask: where should the ray cross the surface? A straight line minimises distance, but distance is not time when the two media have different speeds. The ray should spend more of its journey in the fast medium and less in the slow one — and the right amount of bending is exactly the trade-off that minimises total time.

Drag the crossing point along the interface. At either extreme, the ray takes a long path in the slow medium or bends too sharply. Somewhere in the middle sits a minimum — and at that minimum, and nowhere else, the ratios sin θ₁ / v₁ and sin θ₂ / v₂ become equal. That equality is Snell's law. Fermat derived it from laziness, without needing to know light was a wave, or a particle, or anything at all.

Fermat did not know it, but he had found the first instance of a pattern that was about to appear everywhere.

Eighty years after Fermat, Maupertuis asked: could the same trick work for matter? Light picks the path of least time. Does a thrown ball pick the path of least something?

The quantity he proposed is called the action, and it is built out of two things we already know.

T is the kinetic energy, V is the potential energy, and their difference L = T − V is the Lagrangian of the system. The action S is what you get by integrating L along a candidate trajectory, from the start to the end. Its units are energy × time — joule-seconds — which is the same as Planck's constant. (Quantum mechanics will need us to remember this.)

Read the definition slowly. It is unlike everything else in classical physics. Force is a thing you feel at an instant. Acceleration is a thing you measure at an instant. Action is a thing you compute only after you have already picked a whole trajectory. It is a property of the path, not a property of a moment. A single candidate path gets a single number S, and a nearby path gets a different number. That one number is the score.

Here is the principle, in one sentence.

Two words need unpacking.

Stationary is a generalisation of minimum. If you perturb the path a tiny bit in any direction, S does not change — to first order in the perturbation. The physical path sits at the flat bottom of a valley in the landscape of all possible paths. It is called least action out of historical loyalty to Maupertuis, but the rigorous word is stationary: it can be a minimum, it can be a saddle, in rare cases a maximum. For almost every problem you will meet, it is a minimum, and the slogan holds.

Fixed events means: the start time, the start position, the end time, and the end position are all held constant as you vary the path. The principle does not determine these; they are the boundary conditions. It determines what happens between them.

That is the whole idea. The rest is consequences.

Point A is the floor at t = 0. Point B is also the floor, 2 seconds later, 20 metres further along. A ball is to travel between them under gravity. What trajectory does it follow?

Newton's answer: F = ma, acceleration is g downward, integrate twice with the boundary conditions, get a parabola that rises to 4.9 m and comes back down.

The action-principle answer: write down all candidate vertical trajectories y(t) with y(0) = 0 and y(2) = 0. For each, compute S = ∫₀² [½m ẏ² − mgy] dt. Find the one with stationary S. You get the same parabola.

The remarkable thing is that the parabola is a minimum. Flatten it — send the ball along a straight line at y = 0 — and S is less negative (larger) than on the parabola. Exaggerate it — send the ball up to 10 m — and S is again larger. Nudge it slightly either way and S creeps up. The parabola sits at the bottom of the valley in path-space.

Two routes. Same parabola. The fact that they agree is not a tautology — it is a theorem about L = T − V and Newton's laws. We will prove it next topic.

If the action principle just reproduced Newton's laws for point particles in gravity, it would be a curiosity. A cute reformulation. It is more than that.

It generalises. The principle has the same form in every coordinate system. Polar coordinates, spherical, the angle of a pendulum, the length of a spring — whatever variables describe your system, write L(q, q̇) in them and the stationary-S condition gives the correct equations of motion. Newton's F = ma is cluttered by constraint forces (the tension in a string, the normal force from a surface) that the action principle handles transparently by simply not including them — constrained coordinates do the job automatically.

It scales. A single particle gets one L. Ten thousand particles get one L. A vibrating string, a flexing drumhead, an elastic solid — all one L. A coupled system of fields filling all of space, like the electromagnetic field — one L. You add the Lagrangians together and the stationary-S condition handles everything at once. This is why Mécanique analytique, the 1788 book in which Lagrange worked out the machinery, could claim on its first page that it contained not a single diagram. It did not need diagrams. It had a function and a rule.

It crosses into new physics. Every major theory after Newton has been built by writing down a Lagrangian and extremising. Maxwell's equations come from a Lagrangian. General relativity comes from the Einstein–Hilbert action. The Standard Model of particle physics is, at its heart, the most celebrated L ever written — a few lines on a T-shirt. When physicists invent a new theory, they no longer start from forces. They start from an action, and the equations of motion fall out.

It plugs into Noether. We met this in FIG.08: every continuous symmetry of the Lagrangian is paired with a conserved quantity. That machinery requires a Lagrangian to work. Without L = T − V, Noether's theorem has nothing to chew on. The reason we can speak of conservation of energy, momentum, and angular momentum as shadows of symmetries of spacetime is precisely because the whole of mechanics can be re-expressed as the extremum of a scalar functional. The action principle is the skeleton on which every conservation law hangs.

We have the principle. We have the action S, the Lagrangian L = T − V, and the slogan: make S stationary. What we do not yet have is a mechanical procedure to find the stationary path.

In the worked example above, we hand-waved: Newton's second law ÿ = −g fell out of the extremal condition. That is not a coincidence. For any L(q, q̇), the condition δS = 0 is equivalent to a differential equation — the Euler–Lagrange equation — and once you have it, finding the stationary path is just ordinary calculus. No guessing, no variations, no candidate paths. Just a differential equation you can solve.

That is FIG.29. We will derive the Euler–Lagrange equation from δS = 0 in a page, and then apply it to every system we have met in the branch — pendulum, spring, orbit — and watch the same piece of machinery pump out their equations of motion, one after another.

After that, FIG.30 reformulates the whole thing in terms of momenta instead of velocities, and we meet the Hamiltonian — the same structure seen from a different angle, the one that becomes quantum mechanics when you squint at it hard enough.

The road from here to the frontier of modern physics runs through a single optimisation.