Hamilton's Principle of Stationary Action

Newton hands you a rule enforced instant by instant: at every tick, force equals rate of change of momentum. Hamilton's principle turns that local bookkeeping inside out. It looks at the whole trajectory at once, assigns to every conceivable path $q(t)$ a system could take between a fixed start and a fixed finish a single number — the action $S[q]=\int_{t_1}^{t_2} L\,\mathrm{d}t$ , the running total of the Lagrangian $L=T-V$ — and asserts that the path Nature takes is the one for which $S$ is stationary: a tiny wiggle of the path changes $S$ only at second order. The equations of motion are no longer a postulate; they are what "stationary" forces to be true point by point. The pay-off is threefold. First, $L$ is a scalar, so the action does not care what coordinates you use — Cartesian, polar, angle-of-a-pendulum — and the resulting Euler–Lagrange equations wear the same form in all of them, with constraint forces that do no work dropping out for free. Second, the principle outlives mechanics: replace $q(t)$ by a field $\phi(x)$ and $L$ by a Lagrangian density, and the same "make it stationary" move produces Maxwell's, Dirac's, and Einstein's equations. Third, every continuous symmetry of the action buys a conservation law (Noether), so energy, momentum, and angular momentum are read straight off the structure of $L$ . One word does all the work and is almost always misquoted: stationary, not least. For short times the physical path does minimise $S$ , but past a conjugate point it becomes a saddle — nearby paths have smaller action, yet the true path still obeys $\delta S=0$ . The deepest reason the principle holds at all is quantum: Feynman's amplitude sums $e^{iS/\hbar}$ over all paths, and as $\hbar\to0$ only the stationary path survives the interference. Classical mechanics is the shadow of that stationary phase.

Key takeaways

Hamilton's principle: a system moves between fixed configurations $q(t_1)$ , $q(t_2)$ along the path that makes the action $S=\int_{t_1}^{t_2}L\,\mathrm{d}t$ stationary, $\delta S=0$ , for every endpoint-fixed variation.
The Lagrangian $L=T-V$ is a scalar, so the principle is coordinate-independent; varying $S$ and applying the fundamental lemma yields the Euler–Lagrange equations of motion.
"Stationary" means $\delta S=0$ to first order — not necessarily least: past a conjugate point the true path is a saddle of $S$ , exactly computable for the harmonic oscillator (the turn is at half a period).
For a conservative system the principle is equivalent to Newton's second law, but generalises effortlessly to fields, giving Maxwell, Dirac, and Einstein as stationary-action equations.
Time-translation symmetry of the action makes the Hamiltonian $H=\dot q\,p-L$ conserved — the mechanical face of Noether's theorem.
The quantum justification is Feynman's path integral: the classical trajectory is the stationary-phase point of $\int\mathcal{D}q\,e^{iS/\hbar}$ as $\hbar\to0$ .
Key idea: promote the trajectory to the unknown — instead of solving differential equations directly, demand that the scalar action functional be stationary, $\delta S=0$ , and let the calculus of variations produce the equations of motion.

Where this lives today

Hamilton's "write an action and make it stationary" is not a historical curiosity but the working definition of a physical theory. Every fundamental interaction is now specified by a Lagrangian handed to the field version of equation (5): the Standard Model is a single Lagrangian density whose Euler–Lagrange equations are the Dirac, Yang–Mills, and Higgs equations, and general relativity is the stationary-action condition of the Einstein–Hilbert action $\int R\sqrt{-g}\,\mathrm{d}^4x$ . The same principle drives navigation and control — spacecraft trajectories are extremals, and minimum-fuel orbit transfers and robot motion planning are modern brachistochrones solved by Pontryagin's maximum principle. Lattice gauge theory evaluates Feynman's $\int\mathcal{D}q\,e^{iS/\hbar}$ numerically to compute hadron masses from first principles, while the phase-space form Hamilton distilled underlies the symplectic integrators that keep planetary simulations stable over billions of years. Two centuries after the Dublin memoirs, the most productive sentence in physics remains: form the action, and make it stationary.

The physical picture

Freeze the endpoints. A ball must leave your hand at $t_1$ and reach the fielder's glove at $t_2$ — same start, same finish, same elapsed time. Between them, imagine every curve the ball could trace. Each carries a running tally of kinetic minus potential energy, and the action is that tally over the whole trip. Hamilton's principle says the realised path is the balance point of this tally over the space of curves — the flat spot of a landscape whose "points" are entire trajectories. The first figure freezes the true path against two competitors sharing its endpoints.

Figure

Every path shares the fixed endpoints q(t₁) and q(t₂) over the same elapsed time; the physical trajectory in blue makes the action S = ∫(T − V) dt stationary, so δS = 0 to first order for every endpoint-fixed wiggle δq(t)

That "flat spot" is almost always given the wrong adjective. Early statements called it least action, but stationary is the honest word: for a short trip the true path is a genuine minimum, yet stretch the interval past a conjugate point and the same path becomes a saddle — some neighbours accumulate less action while the physical one still satisfies $\delta S=0$ . The cleanest place to watch this is the harmonic oscillator, where the sign of the second variation flips at exactly half a period. The second figure carries that computation.

Figure

Second variation of the harmonic-oscillator action along the wiggle η = sin(πt/T), for m = 0.5 kg and ω = 2 rad/s. Below the conjugate time T★ = π/ω = 1.571 s the true path is a minimum (δ²S > 0, blue); above it the same path is a saddle (δ²S < 0, red) — stationary, but no longer least

The core equations

The whole framework rests on one object: the action, a functional that eats an entire trajectory and returns one real number by integrating the Lagrangian along it. Everything downstream follows from demanding this number be stationary.

S[q]=\int_{t_1}^{t_2} L\big(q(t),\dot q(t),t\big)\,\mathrm{d}t \qquad (1)

where: $S[q]$ — the action, a real number assigned to each admissible trajectory $q(t)$ ; $L$ — the Lagrangian, for a conservative system $L=T-V$ ; $q,\dot q$ — a generalized coordinate and its time derivative; $t_1,t_2$ — the fixed initial and final times. Reading: the entire shape of $q$ between the endpoints fixes the value of $S$ . Hamilton's principle is the single physical postulate: of all curves joining the two fixed configurations, it selects the one whose action does not change to first order under any allowed deformation.

\delta S=0\qquad\text{for all }\delta q\text{ with }\delta q(t_1)=\delta q(t_2)=0 \qquad (2)

where: $\delta S$ — the first variation of the action; $\delta q$ — an arbitrary infinitesimal deformation that must vanish at the fixed endpoints. Reading: stationarity against every endpoint-preserving wiggle is the law of motion, before any differential equation appears. To make "for all $\delta q$ " computable we embed the candidate path in a one-parameter family, so stationarity collapses into the vanishing of an ordinary derivative.

q_\epsilon(t)=q(t)+\epsilon\,\eta(t),\qquad \eta(t_1)=\eta(t_2)=0 \qquad (3)

where: $\epsilon$ — a small real bookkeeping parameter; $\eta(t)$ — an arbitrary smooth test function vanishing at the endpoints, so $\delta q=\epsilon\eta$ . Reading: the family collapses to the physical path at $\epsilon=0$ , and $\delta S=\frac{\mathrm{d}}{\mathrm{d}\epsilon}S[q_\epsilon]\big|_{0}$ . Differentiating, then integrating by parts to strip the time-derivative off $\eta$ , gathers the first variation into one bracket multiplying the arbitrary $\eta$ — the form the fundamental lemma needs.

\delta S=\int_{t_1}^{t_2}\left(\frac{\partial L}{\partial q}-\frac{\mathrm{d}}{\mathrm{d}t}\frac{\partial L}{\partial\dot q}\right)\delta q\,\mathrm{d}t \qquad (4)

where: $\partial L/\partial q$ — the generalized force; $\partial L/\partial\dot q$ — the canonical momentum $p$ ; the boundary term $[\,p\,\eta\,]_{t_1}^{t_2}$ is already dropped using $\eta(t_1)=\eta(t_2)=0$ . Reading: since $\eta$ is arbitrary, the bracket must vanish everywhere. Applying the fundamental lemma gives the local law obeyed at every instant — the Euler–Lagrange equations, one per degree of freedom.

\frac{\mathrm{d}}{\mathrm{d}t}\frac{\partial L}{\partial\dot q_i}-\frac{\partial L}{\partial q_i}=0,\qquad i=1,\dots,n \qquad (5)

where: $q_i$ — the $i$ -th generalized coordinate of an $n$ -degree-of-freedom system. Reading: instant by instant, the rate of change of the momentum $p_i=\partial L/\partial\dot q_i$ balances the force $\partial L/\partial q_i$ — Newton's law, coordinate-free. Two derived quantities organise the rest of mechanics: the momentum conjugate to each coordinate, and the Hamiltonian assembled from it by a Legendre transform.

p_i\equiv\frac{\partial L}{\partial\dot q_i},\qquad H=\sum_i p_i\dot q_i-L \qquad (6)

where: $p_i$ — the momentum conjugate to $q_i$ ; $H$ — the Hamiltonian, which equals the total energy $T+V$ when $L=T-V$ , $T$ is quadratic in the velocities, and the constraints are time-independent. Reading: $H$ is the conserved quantity that appears whenever the action has no explicit time dependence. (If an endpoint is released rather than pinned, the boundary term of equation (4) survives and forces the natural condition $p=\partial L/\partial\dot q=0$ there.)

Derivation AFrom Hamilton's principle to the equations of motion

The goal is to convert the global statement "the action is stationary for every endpoint-fixed variation" into a single local differential equation valid at each instant. We embed the candidate trajectory in a one-parameter family, reduce stationarity to an ordinary derivative in the parameter, and integrate by parts so the whole first variation appears as one continuous bracket multiplying the arbitrary variation — at which point the fundamental lemma of the calculus of variations forces that bracket to vanish pointwise.

Step 1. Embed the path: write $q_\epsilon(t)=q(t)+\epsilon\eta(t)$ with $\eta(t_1)=\eta(t_2)=0$ , so the deformed action $S[q_\epsilon]$ is an ordinary function of one real variable $\epsilon$ , and Hamilton's principle becomes $\delta S=\frac{\mathrm{d}}{\mathrm{d}\epsilon}S[q_\epsilon]\big|_{\epsilon=0}=0$ for every admissible $\eta$ . Why: this trades "vary a whole function" for the familiar "set an ordinary derivative to zero", losing no content because $\eta$ is arbitrary.

Step 2. Differentiate under the integral by the chain rule, using $\partial q_\epsilon/\partial\epsilon=\eta$ and $\partial\dot q_\epsilon/\partial\epsilon=\dot\eta$ : $\displaystyle\delta S=\int_{t_1}^{t_2}\Big(\frac{\partial L}{\partial q}\,\eta+\frac{\partial L}{\partial\dot q}\,\dot\eta\Big)\mathrm{d}t$ . Why: the deformation reaches $L$ through both dynamical slots, position $q$ and velocity $\dot q$ , so both terms are tracked.

Step 3. Integrate the velocity term by parts: $\displaystyle\int_{t_1}^{t_2}\frac{\partial L}{\partial\dot q}\dot\eta\,\mathrm{d}t=\Big[\frac{\partial L}{\partial\dot q}\eta\Big]_{t_1}^{t_2}-\int_{t_1}^{t_2}\frac{\mathrm{d}}{\mathrm{d}t}\frac{\partial L}{\partial\dot q}\,\eta\,\mathrm{d}t$ , and drop the boundary bracket because $\eta(t_1)=\eta(t_2)=0$ . Why: parts is the only move that removes the derivative from the arbitrary $\eta$ , and the fixed endpoints kill the leftover surface term — release an endpoint and it becomes the natural condition $p=0$ .

Step 4. Collect the survivors into one integral, $\displaystyle\delta S=\int_{t_1}^{t_2}\Big(\frac{\partial L}{\partial q}-\frac{\mathrm{d}}{\mathrm{d}t}\frac{\partial L}{\partial\dot q}\Big)\eta\,\mathrm{d}t=0$ , and invoke the fundamental lemma: a continuous function that integrates to zero against every smooth $\eta$ vanishing at the ends is identically zero. Hence $\displaystyle\frac{\mathrm{d}}{\mathrm{d}t}\frac{\partial L}{\partial\dot q}-\frac{\partial L}{\partial q}=0$ . Why: were the bracket nonzero anywhere, bump $\eta$ up there and $\delta S\neq0$ , contradicting stationarity — so it vanishes at every interior instant. *Verified: reproduces equation (5); with $L=\tfrac12 m\dot q^2-V(q)$ it gives $m\ddot q=-V'(q)$ , Newton's second law (Goldstein Ch. 2, Landau–Lifshitz §2).*

Go deeper · Derivation B · 1 minEnergy conservation from a time-symmetric action

Derivation BEnergy conservation from a time-symmetric action

The aim is to show that when the Lagrangian carries no explicit time dependence — the action is unchanged by sliding the whole motion forward or backward in $t$ — the Hamiltonian $H=\dot q\,\partial_{\dot q}L-L$ is conserved along every solution. This is the mechanical instance of Noether's theorem, the Beltrami first integral of the variational problem. We differentiate $L$ totally along a trajectory, eliminate the position-derivative term using the equation of motion, and watch the remaining pieces assemble into a total time derivative that must vanish.

Step 1. Differentiate $L$ totally along a solution: $\displaystyle\frac{\mathrm{d}L}{\mathrm{d}t}=\frac{\partial L}{\partial q}\dot q+\frac{\partial L}{\partial\dot q}\ddot q+\frac{\partial L}{\partial t}$ , and impose $\partial L/\partial t=0$ so the last term drops. Why: no explicit time dependence is exactly invariance of the action under time translation — it removes the only term that could spoil the cancellation.

Step 2. Replace the position-derivative using Euler–Lagrange (5), $\displaystyle\frac{\partial L}{\partial q}=\frac{\mathrm{d}}{\mathrm{d}t}\frac{\partial L}{\partial\dot q}$ , giving $\displaystyle\frac{\mathrm{d}L}{\mathrm{d}t}=\dot q\frac{\mathrm{d}}{\mathrm{d}t}\frac{\partial L}{\partial\dot q}+\frac{\partial L}{\partial\dot q}\ddot q$ . Why: injecting the equation of motion is what makes the two surviving terms conjugate halves of a single product-rule derivative.

Step 3. Recognise the product rule: the right-hand side is exactly $\displaystyle\frac{\mathrm{d}}{\mathrm{d}t}\Big(\dot q\,\frac{\partial L}{\partial\dot q}\Big)$ , so $\displaystyle\frac{\mathrm{d}}{\mathrm{d}t}\Big(\dot q\,\frac{\partial L}{\partial\dot q}-L\Big)=0$ . Why: assembling the pieces into one total derivative is the whole theorem — a rate of change that is identically zero is a conservation law.

Step 4. Read off the conserved quantity $\displaystyle H=\dot q\,\frac{\partial L}{\partial\dot q}-L=\text{const}$ . For $L=\tfrac12 m\dot q^2-V(q)$ this is $H=\tfrac12 m\dot q^2+V=T+V$ , the total mechanical energy. Why: a time-independent Lagrangian conserves energy — the same structure that, under spatial and rotational symmetry, conserves momentum and angular momentum. *Verified: reproduces equation (6) as a first integral; $H=T+V$ for the standard mechanical Lagrangian (Landau–Lifshitz §6, Goldstein §2.6); it is the Beltrami identity with the independent variable taken to be time.*

Go deeper · Derivation C · 1 minThe classical path as the stationary phase of the quantum amplitude

Derivation CThe classical path as the stationary phase of the quantum amplitude

The purpose is to answer the question the principle itself cannot: why should Nature extremise an action at all? Feynman's path integral supplies the reason. The quantum amplitude to travel between two configurations sums over every path, each weighted by the pure phase $e^{iS/\hbar}$ ; expanding about a stationary path shows that as $\hbar\to0$ only the neighbourhood of $\delta S=0$ survives the interference — so the stationary-action path is the one not washed out.

Step 1. Write the propagator as a sum over histories, $\displaystyle K(q_2,t_2;q_1,t_1)=\int\mathcal{D}q\;e^{\,iS[q]/\hbar}$ , each path contributing unit modulus and a phase set by its action in units of $\hbar$ . Why: no single path is privileged a priori; the classical trajectory must emerge from interference, not from a postulate.

Step 2. Expand the action about a stationary path $q_{\text{cl}}$ where $\delta S=0$ : $\displaystyle S[q_{\text{cl}}+\eta]=S[q_{\text{cl}}]+\underbrace{\delta S}_{=0}+\tfrac12\delta^2S[\eta]+\cdots$ , so near $q_{\text{cl}}$ the phase is stationary to first order in $\eta$ . Why: a stationary point of $S$ is where neighbouring paths share nearly the same phase — the only place constructive interference can occur.

Step 3. Away from $q_{\text{cl}}$ , $\delta S\neq0$ , so the phase $S/\hbar$ sweeps through many multiples of $2\pi$ across an infinitesimal bundle of neighbours and the contributions cancel; near $q_{\text{cl}}$ the phase is flat and a whole tube of paths adds coherently. Why: rapidly oscillating phases integrate to nearly zero — only a region of stationary phase yields a surviving amplitude.

Step 4. Take the classical limit $\hbar\to0$ : the coherent tube's width shrinks like $\sqrt{\hbar}$ and the sum localises entirely on $\delta S=0$ . The particle "chooses" the stationary-action path because it is the only one quantum amplitude does not destroy. Why: Hamilton's principle is thereby derived, not assumed — classical mechanics is the $\hbar\to0$ stationary-phase shadow of quantum mechanics. *Verified: reproduces $\delta S=0$ as the saddle condition of $\int\mathcal{D}q\,e^{iS/\hbar}$ ; the stationary-phase width $\sim\sqrt{\hbar}$ and the classical limit match Feynman & Hibbs Ch. 2 and the WKB reduction.*

Limiting cases

Special structure of $L$ collapses the principle into familiar statements, and the quantum limit recovers it from a deeper theory.

Case	Result	Meaning
Free particle	$L=\tfrac12 m\dot x^2\Rightarrow m\ddot x=0$	No potential: stationary action gives uniform straight-line motion, the time-domain analogue of a geodesic in flat space.
Conservative system	$L=T-V\Rightarrow\frac{\mathrm{d}}{\mathrm{d}t}\frac{\partial L}{\partial\dot q}-\frac{\partial L}{\partial q}=0$	Recovers Newton's second law $m\ddot q=-\partial V/\partial q$ in any coordinates; constraint forces drop out.
Autonomous $L$ ( $\partial_t L=0$ )	$\dot q\,\partial_{\dot q}L-L=\text{const}$	Time-translation symmetry gives the conserved Hamiltonian — the Beltrami/energy first integral $E=T+V$ .
Short time interval	$\delta^2 S>0$ : path is a true minimum	Before the first conjugate point the physical path genuinely minimises $S$ — "least action" is locally correct.
Past a conjugate point	$\delta^2 S$ indefinite: path is a saddle	Beyond a kinetic focus (for the oscillator, half a period) neighbouring paths have less action; only $\delta S=0$ survives.
Classical limit $\hbar\to0$	$\int\mathcal{D}q\,e^{iS/\hbar}\to\delta S=0$	The stationary-phase point of Feynman's sum over paths selects the classical trajectory.

Common trapThe principle is one of least action

It is not — it is stationary action, and the distinction is real, not pedantic. Hamilton's principle demands only $\delta S=0$ . For short times the true path does minimise the action, but past a conjugate point (a kinetic focus) it becomes a saddle: neighbouring paths have strictly smaller action while the physical path still obeys $\delta S=0$ . The harmonic oscillator makes this quantitative — along $\eta=\sin(\pi t/T)$ the second variation is $\delta^2 S=(m/4)(\pi^2/T-\omega^2 T)$ , positive for $T<\pi/\omega$ and negative beyond, so the crossover sits at exactly half a period. A second, quieter trap: the endpoint variations must vanish, $\eta(t_1)=\eta(t_2)=0$ ; relax that and you get natural boundary conditions $p=0$ at the free end, not the Euler–Lagrange equations. "Least" is a hangover from Maupertuis; the universal statement is stationarity.

Worked examples

1 · The harmonic oscillator from stationary action (confidence builder)

A mass $m$ on a spring has Lagrangian $L=\tfrac12 m\dot x^2-\tfrac12 kx^2$ . Use Hamilton's principle to find the equation of motion and the conserved energy. Then put in $m=0.5\ \mathrm{kg}$ , $k=2\ \mathrm{N/m}$ , amplitude $A=0.1\ \mathrm{m}$ .

Step 1. Read off the momentum from the velocity slot: $p=\partial L/\partial\dot x=m\dot x$ , the object whose time derivative the Euler–Lagrange equation balances against the force.

Step 2. Read off the force from the position slot: $\partial L/\partial x=-kx$ , the familiar spring force.

Step 3. Assemble Euler–Lagrange (5): $\frac{\mathrm{d}}{\mathrm{d}t}(m\dot x)+kx=0$ , i.e. $\ddot x=-\omega^2 x$ with $\boldsymbol{\omega=\sqrt{k/m}}$ . With the numbers, $\omega=\sqrt{2/0.5}=\mathbf{2\ rad/s}$ and the period is $2\pi/\omega=\boldsymbol{\pi\approx3.14\ s}$ .

Step 4. Because $L$ has no explicit time dependence, Derivation 2 hands you a conserved Hamiltonian $H=\dot x\,p-L=\tfrac12 m\dot x^2+\tfrac12 kx^2=T+V$ . At the amplitude $x=A$ the mass is momentarily at rest, so $E=\tfrac12 kA^2=\tfrac12(2)(0.1)^2=\mathbf{0.010\ J}$ , constant throughout the motion. The scalar action reproduces SHM with $\omega=2$ rad/s, period $\pi$ s, and conserved energy $E=0.010$ J — with no free-body diagram drawn.

2 · When the action is not least: the conjugate point (fully worked)

For the same oscillator ( $m=0.5\ \mathrm{kg}$ , $\omega=2\ \mathrm{rad/s}$ ), the true motion between two fixed configurations obeys $\delta S=0$ . Is that stationary path a minimum of the action? Test it with the deformation $\eta(t)=\sin(\pi t/T)$ , which vanishes at both endpoints, and find the time $T$ at which the answer changes.

Step 1. Confirm stationarity first. For any admissible $\eta$ , $\delta S=\int_0^T(m\dot x\,\dot\eta-kx\,\eta)\,\mathrm{d}t$ ; integrating by parts and using $\ddot x=-\omega^2 x$ gives $\delta S=-\int_0^T m(\ddot x+\omega^2 x)\eta\,\mathrm{d}t=0$ . The physical path is stationary for every $\eta$ , at all $T$ — that is Hamilton's principle.

Step 2. Go to second order, where minimum-versus-saddle is decided. The quadratic piece of $S[x+\epsilon\eta]$ is $\delta^2 S=\tfrac12\int_0^T m\big(\dot\eta^2-\omega^2\eta^2\big)\,\mathrm{d}t$ . A minimum requires $\delta^2 S>0$ for every admissible $\eta$ ; a single $\eta$ with $\delta^2 S<0$ proves it is a saddle.

Step 3. Evaluate the two integrals for $\eta=\sin(\pi t/T)$ , using $\dot\eta=(\pi/T)\cos(\pi t/T)$ : $\int_0^T\dot\eta^2\,\mathrm{d}t=\pi^2/(2T)$ and $\int_0^T\eta^2\,\mathrm{d}t=T/2$ . Hence $\delta^2 S=\dfrac{m}{4}\Big(\dfrac{\pi^2}{T}-\omega^2 T\Big)$ .

Step 4. Find the sign change. $\delta^2 S>0$ iff $\pi^2/T>\omega^2 T$ iff $T<\pi/\omega$ . So the conjugate time is $\boldsymbol{T^\star=\pi/\omega=\pi/2\approx1.571\ s}$ — exactly half the oscillator period $\pi$ s.

Step 5. Read the numbers. At $T=1.0\ \mathrm{s}<T^\star$ : $\delta^2 S=0.125(\pi^2/1-4)=\mathbf{+0.734>0}$ , a minimum. At $T=2.0\ \mathrm{s}>T^\star$ : $\delta^2 S=0.125(\pi^2/2-8)=\mathbf{-0.383<0}$ , a saddle — the wiggle $\eta$ lowers the action below the physical path's value.

Step 6. Interpret it. For $T<1.571\ \mathrm{s}$ the true trajectory is the least-action path; for $T>1.571\ \mathrm{s}$ it is merely stationary, a saddle of $S$ . The physical motion never changes — it always satisfies $\delta S=0$ — but "least action" fails past half a period. This conjugate point is the mechanical cousin of an optical focal point, and the reason the correct principle is stationarity, not minimisation.

Physics of the portrait

Hamilton's principle bears the name of the man who, in two dense papers of the 1830s, saw that the whole of dynamics could hang from a single variational function — and who spent the rest of his life on the algebra that quantum mechanics would one day need.

Origins · 1 minWilliam Rowan Hamilton and how this was discovered

William Rowan Hamilton

1805-1865 · the 1834-1835 papers that recast all of mechanics as one variational principle

William Rowan Hamilton was a prodigy of Trinity College Dublin — appointed Royal Astronomer of Ireland and Andrews Professor of Astronomy in 1827, at twenty-two, before he had even graduated. His first triumph was in optics: treating a light ray as the extremal of a "characteristic function", he predicted conical refraction in biaxial crystals in 1832, confirmed by Humphrey Lloyd within months — a rare pure-mathematical prediction of a new optical phenomenon. Hamilton then saw that mechanics and optics were the same variational subject in different clothes. His two memoirs "On a General Method in Dynamics" (1834 and 1835) reformulated Lagrange's mechanics around a single principal function $S$ and stated the principle now bearing his name: the motion makes $\delta\int(T-V)\,\mathrm{d}t=0$ between fixed configurations at fixed times. From it he extracted the canonical equations $\dot q=\partial H/\partial p$ , $\dot p=-\partial H/\partial q$ , splitting Newton's second-order law into a symmetric first-order pair and inventing the phase-space picture that dominates modern physics. He was careful to distinguish his principle from the older "least action" of Maupertuis and Euler, which fixes energy rather than time. His later years went to the quaternions, discovered on Broome Bridge in 1843 — the first non-commutative algebra, ancestor of the vector and spinor formalism physics still runs on. When Schrödinger built wave mechanics in 1926 and Feynman the path integral in 1948, both reached first for Hamilton's optical–mechanical analogy: his characteristic function is the classical limit of the quantum phase.

History

1744

— Maupertuis proclaims a principle of least action for the abbreviated action ∫ p dq at fixed energy

1755

— Euler and Lagrange put least action on a rigorous footing; Lagrange introduces the variation symbol δ

1834

— Hamilton's first memoir On a General Method in Dynamics recasts mechanics around a principal function

1835

— Hamilton's second memoir states δ∫(T − V) dt = 0 and the canonical equations

1918

— Noether proves that every continuous symmetry of the action yields a conservation law

1948

— Feynman's path integral explains stationary action as the ℏ → 0 limit of ∫ Dq exp(iS/ℏ)

Going further

Texts. Goldstein, Poole & Safko, Classical Mechanics, 3rd ed., Ch. 2 (variational principles and Lagrange's equations) · Landau & Lifshitz, Mechanics, §§2–6 · Stone & Goldbart, Mathematics for Physics, Ch. 1 (Calculus of Variations) · Arfken, Weber & Harris, Mathematical Methods for Physicists, 7th ed., Ch. 22 · Feynman & Hibbs, Quantum Mechanics and Path Integrals, Ch. 2 (why the classical path is stationary).

Provenance. The first-variation-to-Euler–Lagrange derivation and the energy first integral are standard (Goldstein Ch. 2; Stone & Goldbart Ch. 1; Landau–Lifshitz §§2, 6). A verifier should: (1) reproduce the integration-by-parts + fundamental-lemma argument for equation (5); (2) confirm the energy first integral (6) from the total time-derivative of $L$ ; (3) recompute the SHO worked example — $\omega=\sqrt{k/m}=2$ rad/s, period $\pi$ s, $E=\tfrac12 kA^2=0.010$ J; (4) recompute the conjugate-point result — $\delta^2 S=(m/4)(\pi^2/T-\omega^2 T)$ for $\eta=\sin(\pi t/T)$ , giving $+0.734$ at $T=1$ s and $-0.383$ at $T=2$ s, with sign change at $T^\star=\pi/\omega=1.571$ s (half the period); (5) check the stationary-phase reduction $\int\mathcal{D}q\,e^{iS/\hbar}\to\delta S=0$ against Feynman & Hibbs Ch. 2. All numbers are elementary and machine-checkable.

Prerequisites. Functionals and Function Spaces · First Variation and Stationarity · Fundamental Lemma of the Calculus of Variations. Leads to. Euler–Lagrange Equation · Lagrangian Mechanics · Beltrami Identity and First Integrals · Geodesics as Variational Problems.

Ask the tutor: Want to see the stationary-action machinery turned loose on your own system — a double pendulum, a bead on a rotating wire, or a charged particle in a field — with the Lagrangian built and the equations of motion cranked out line by line?

Is Hamilton's principle the same as the "principle of least action"?

Not quite, and conflating them causes real confusion. Maupertuis' original least-action principle (1744) extremises the abbreviated action $\int p\,\mathrm{d}q$ over paths of fixed energy, with travel time left free. Hamilton's principle (1834–35) extremises a different object — $S=\int(T-V)\,\mathrm{d}t$ — over paths with fixed endpoints and fixed elapsed time, energy free to vary along competitors. They are related by a Legendre transform but are distinct statements. Both are, moreover, principles of stationary action, not least. Modern usage says "the principle of stationary action" precisely to retire the misleading word "least".

Why is the true path a saddle rather than a minimum for long times?

Because the second variation can go negative. Stationarity ( $\delta S=0$ ) fixes only the first-order change; whether the path minimises $S$ is decided by the sign of $\delta^2 S=\tfrac12\int m(\dot\eta^2-\omega^2\eta^2)\,\mathrm{d}t$ for the oscillator. The two terms compete: $\dot\eta^2$ favours a minimum, $-\omega^2\eta^2$ fights it. Over a short interval the derivative term dominates and $\delta^2 S>0$ for every wiggle. Once the interval exceeds the conjugate time $T^\star=\pi/\omega$ , a low-frequency wiggle makes $\delta^2 S<0$ and nearby paths have smaller action — the physical path is then a saddle, the mechanical analogue of an optical caustic.

Why does treating q̇ as independent of q not lead to a contradiction?

It is a bookkeeping convention, not a physical claim. When you form the partials $\partial L/\partial q$ and $\partial L/\partial\dot q$ , you regard $L(q,\dot q,t)$ as an ordinary function of three separate slots and differentiate slot by slot, exactly as with any $f(a,b,c)$ . The physical link " $\dot q$ is the time derivative of $q$ " is reimposed only afterwards, when you substitute an actual trajectory $q(t)$ and apply $\mathrm{d}/\mathrm{d}t$ in the Euler–Lagrange equation. Keeping the two steps apart is what makes the chain rule and the parts integration come out right; merging them early is a classic source of dropped terms and wrong signs.

How does Hamilton's principle recover Newton's second law?

Feed it the mechanical Lagrangian $L=\tfrac12 m\dot q^2-V(q)$ . The velocity slot gives the momentum $p=\partial L/\partial\dot q=m\dot q$ ; the position slot gives the force $\partial L/\partial q=-V'(q)$ . The Euler–Lagrange equation $\frac{\mathrm{d}}{\mathrm{d}t}(m\dot q)-(-V')=0$ is exactly $m\ddot q=-V'(q)=F$ — Newton's second law for a conservative force. The variational route is not a new postulate but a re-derivation of the old law, with two bonuses: it works verbatim in any coordinates because $L$ is a scalar, and constraint forces that do no virtual work never appear, so a pendulum's tension or a bead's normal force drops out automatically.

What is the Hamiltonian, and is it always the energy?

The Hamiltonian is the Legendre transform of the Lagrangian, $H=\sum_i p_i\dot q_i-L$ , re-expressed in the momenta $p_i$ instead of the velocities. It is conserved whenever $L$ has no explicit time dependence ( $\partial L/\partial t=0$ ), by the time-translation argument of Derivation 2. It equals the total energy $T+V$ only under an additional condition: the kinetic energy must be a homogeneous quadratic in the velocities and the constraints time-independent (scleronomic). In a rotating frame or with a moving constraint, $H$ can be conserved yet differ from $T+V$ , or equal $T+V$ yet not be conserved — the two properties are logically separate, a distinction worth internalising early.

Does Nature really "try all paths" and pick the stationary one?

No — that is a vivid metaphor, not a mechanism. Classically the particle follows one trajectory, the one obeying $\delta S=0$ ; nothing physically samples the alternatives. The "trying all paths" language belongs to the quantum picture, where Feynman's amplitude genuinely sums $e^{iS/\hbar}$ over every path — and even there the paths are not tried sequentially, they interfere. Away from the stationary path the phase $S/\hbar$ varies so fast that neighbours cancel in pairs; near it the phase is flat and a whole tube of paths adds coherently, a tube that shrinks onto the classical trajectory as $\hbar\to0$ . So stationary action is what survives quantum interference, not a teleological choice by the particle.