Hamiltonian Formulation for Fields

Classical mechanics has two complete languages, and this chapter has so far spoken only one. The Lagrangian story — density, action, variation — is where field theories are designed; the Hamiltonian story is where they live as dynamical systems: states, phase space, brackets, flows. This article completes the translation for fields, and the translation is the last classical stop before quantum field theory, because canonical quantization is nothing but this article's formalism with one substitution performed at the end. The construction is the chapter's dictionary applied to the machinery of the Hamiltonian chapters: every point of space carries a canonical pair — the field value $\phi(\mathbf{x})$ and its conjugate momentum density $\pi(\mathbf{x}) = \partial\mathcal{L}/\partial\dot\phi$ — so the phase space is infinite-dimensional, with two functions' worth of coordinates. The Legendre transform proceeds pointwise; the Hamiltonian becomes a functional $H[\phi, \pi]$ ; and Hamilton's equations pair the functional gradients of the previous article, $\dot\phi = \delta H/\delta\pi$ , $\dot\pi = -\delta H/\delta\phi$ . The equal-time bracket $\{\phi(\mathbf{x}), \pi(\mathbf{y})\} = \delta^3(\mathbf{x} - \mathbf{y})$ is the continuum Kronecker delta of canonical pairs — one independent pair per point, as the bead picture always promised. The payoff move — the single most consequential calculation in this article — is diagonalization. Expand the free field in Fourier modes and the coupled infinity of point-oscillators becomes an uncoupled infinity of normal-mode oscillators, one per wavevector, with frequencies $\omega_k = \sqrt{k^2 + m^2}$ : the Hamiltonian is revealed as an infinite bank of independent harmonic oscillators. Every fact of the chapter converges here — the dispersion relation is the oscillator spectrum, the mass gap is the lightest oscillator, the mode counting of the second article is the size of the bank. And the door this opens is unmistakable: quantize each oscillator the way quantum mechanics quantizes one, and energy arrives in packets $\hbar\omega_k$ — particles. The capstone article walks through that door; this one builds it, entirely within classical mechanics.

Key takeaways

Each spatial point carries a canonical pair: the field $\phi(\mathbf{x})$ and the momentum density $\pi(\mathbf{x}) = \partial\mathcal{L}/\partial\dot\phi$ (for Klein-Gordon, $\pi = \dot\phi$ ); phase space is the space of pairs of functions.
Pointwise Legendre transform: $\mathcal{H} = \pi\dot\phi - \mathcal{L}$ ; the total Hamiltonian is the functional $H[\phi, \pi] = \int\mathcal{H}\,\mathrm{d}^3x$ — for the scalar field, kinetic + gradient + potential energy densities.
Hamilton's field equations are paired functional gradients: $\dot\phi = \delta H/\delta\pi$ , $\dot\pi = -\delta H/\delta\phi$ — first-order in time, equivalent to the Euler-Lagrange second-order equation.
The canonical structure is the equal-time bracket $\{\phi(\mathbf{x}), \pi(\mathbf{y})\} = \delta^3(\mathbf{x}-\mathbf{y})$ ; any observable evolves by $\dot F = \{F, H\}$ (full bracket formalism: next article).
Fourier diagonalization: the free field is an infinite bank of independent harmonic oscillators, $H = \sum_{\mathbf{k}} \big(\tfrac12|\pi_{\mathbf{k}}|^2 + \tfrac12\omega_k^2|\phi_{\mathbf{k}}|^2\big)$ with $\omega_k = \sqrt{k^2 + m^2}$ .
Normal variables $a_{\mathbf{k}} \propto \omega_k\phi_{\mathbf{k}} + i\pi_{\mathbf{k}}$ evolve as pure phasors $a_{\mathbf{k}}(t) = a_{\mathbf{k}}(0)e^{-i\omega_k t}$ — the classical objects that quantization promotes to annihilation operators.
Classical field theory is valid when mode occupations are large ( $k_BT \gg \hbar\omega_k$ or macroscopic amplitudes); the oscillator bank is where "particle" will mean "one quantum of one mode."
Key idea: Hamiltonian field theory equips every point of space with a canonical pair and every wavevector with an independent harmonic oscillator — the finished classical structure whose mode-by-mode quantization is quantum field theory.

The core insight

The field's canonical formalism — pairs $(\phi(\mathbf{x}), \pi(\mathbf{x}))$ , functional-gradient Hamilton equations, equal-time bracket $\delta^3(\mathbf{x}-\mathbf{y})$ — diagonalizes in Fourier modes into an infinite bank of independent harmonic oscillators with $\omega_k = \sqrt{k^2+m^2}$ : classical mechanics' final form for fields, and the doorstep of quantization.

Where this lives today

The oscillator-bank picture is the working ontology of modern physics. Particle physics reads equation (5) with $\hbar$ on: every species is a field, every particle a mode quantum, and collider phenomenology is bookkeeping of energy exchanged between the banks of coupled fields. Condensed matter runs the same formalism at $\bar n$ both large and small: phonons, magnons, photons in cavities — and quantum optics manipulates individual oscillators of the electromagnetic bank so cleanly that squeezed states of specific modes (deformed ellipses in exactly this article's mode phase planes) are engineering products, most famously injected into LIGO to beat the vacuum fluctuations of the modes that carry gravitational-wave signals. Numerical relativity and lattice field theory compute in the canonical formalism: the ADM equations evolving black-hole spacetimes are Hamilton's equations for the gravitational field (with Dirac's constraints enforced at every step), and Hamiltonian lattice gauge theory is the formulation of choice for the new quantum simulators — cold atoms and trapped ions realizing small oscillator banks that emulate field dynamics beyond classical computation. And cosmology's boldest measured prediction lives here: inflationary perturbation theory quantizes the mode bank of a scalar field on an expanding background, and the CMB's temperature speckles are, in the standard account, the stretched-out descendants of individual $a_{\mathbf{k}}$ fluctuations — this article's phasors, photographed on the sky. Within the chapter, two doors remain: the Poisson-bracket article develops the algebra sketched at equation (4) into the full canonical toolkit, and the capstone — from the action principle to quantum field theory — performs the one substitution this article prepared and walks through.

The physical picture

The bead chain, one last time — because the Hamiltonian picture is where the chapter's founding metaphor becomes exact bookkeeping. Each bead of the discrete chain carried a coordinate and a momentum: a phase plane per bead, $2N$ dimensions in all. The dictionary sends the bead index to the spatial label, so the field's phase space has a plane per point: the state of the field at one instant is the pair of functions $(\phi(\cdot), \pi(\cdot))$ — the configuration and the velocity profile, which is exactly the initial data the wave equation always demanded. Dynamics is a flow on this space: Hamilton's equations read the two functional gradients of $H$ and rotate each point's pair accordingly, coupled to its neighbours through the gradient energy.

Figure

The field as an oscillator bank — position basis, mode basis, and the gapped spectrum that quantization will read as particles

Coupled — until you change basis. The physical picture that organizes everything: the position basis (one oscillator per point, coupled by springs) and the mode basis (one oscillator per wavevector, uncoupled) describe the same phase space, and the Fourier transform is the normal-mode diagonalization of the oscillations chapter performed on an infinite matrix. In the mode basis the field is an orchestra of strictly independent harmonic oscillators: mode $\mathbf{k}$ has displacement $\phi_{\mathbf{k}}$ , momentum $\pi_{\mathbf{k}}$ , and spring constant $\omega_k^2 = k^2 + m^2$ — stiffer for shorter wavelengths, with the mass supplying the floor stiffness that even the $k = 0$ mode feels. Each oscillator's phase point circulates on its own ellipse at its own frequency, forever, ignorant of all the others. Every classical solution of the free field — every pulse, packet, standing wave — is a chord: a fixed distribution of amplitude among the oscillators, each spinning at its own $\omega_k$ . This picture is worth possessing viscerally because of what happens when quantum mechanics arrives. A quantum harmonic oscillator of frequency $\omega$ has energy levels spaced $\hbar\omega$ ; quantizing the bank endows each mode with integer occupation — and a "particle of momentum $\hbar\mathbf{k}$ " is precisely one quantum of oscillator $\mathbf{k}$ , with energy $\hbar\omega_k = \sqrt{(\hbar k)^2 + m^2}$ (the dispersion relation is the relativistic energy-momentum relation, as the density article foretold). The field-versus-particle duality that vexed a century of physics dissolves into basis bookkeeping this article makes classical: fields are the position-basis description, particles count quanta in the mode basis, and the Hamiltonian formalism is the exact hinge between them.

The canonical machinery, pointwise

The construction shadows the Hamiltonian chapter of particle mechanics step for step, with the functional calculus supplying the derivatives. First, the momentum conjugate to the field at each point — a definition inherited directly from the previous article's $\pi(x) = \delta L/\delta\dot\phi(x)$ :

\pi(\mathbf{x}, t) = \frac{\partial \mathcal{L}}{\partial \dot\phi}(\mathbf{x}, t) \qquad (1)

where: for the scalar density $\mathcal{L} = \tfrac12\dot\phi^2 - \tfrac12|\nabla\phi|^2 - V(\phi)$ this gives $\pi = \dot\phi$ — each point-oscillator's momentum, in the bead reading (per unit volume: $\pi$ is a momentum density, dimensions bookkept by the lattice weight of the functional-derivative article). The Legendre transform then proceeds pointwise — trade $\dot\phi$ for $\pi$ at each $\mathbf{x}$ independently:

\mathcal{H}(\phi, \pi, \nabla\phi) = \pi\,\dot\phi - \mathcal{L} = \tfrac{1}{2}\pi^2 + \tfrac{1}{2}|\nabla\phi|^2 + V(\phi), \qquad H[\phi, \pi] = \int \mathcal{H}\,\mathrm{d}^3x \qquad (2)

where: the three terms of $\mathcal{H}$ are the chapter's energy anatomy — point-kinetic, neighbour-spring, local-well — and $H$ is a functional of the instantaneous state, constant along the motion for time-independent densities. (That $\mathcal{H}$ is the $T^{00}$ component of the stress-energy tensor, and that its conservation follows from time-translation symmetry, is established in the continuum-mechanics chapter's Hamiltonian-density article and the stress-energy article of this chapter; here $H$ 's job is to generate the dynamics.)

Derivation AHamilton's field equations as paired functional gradients

We derive the canonical equations of motion for the field from the phase-space form of the action, obtaining $\dot\phi = \delta H/\delta\pi$ and $\dot\pi = -\delta H/\delta\phi$ — mechanics' most symmetric equations, with functional gradients in place of partials. The derivation treats $\phi$ and $\pi$ as independent variational variables (the whole point of the phase-space formulation), and the two equations emerge from the two independent variations. We then verify the pair reproduces the Euler-Lagrange dynamics, closing the loop with the Lagrangian articles.

Step 1. Write the action in phase-space (first-order) form, using (2) to eliminate $\mathcal{L}$ :

S[\phi, \pi] = \int \mathrm{d}t\,\mathrm{d}^3x\;\big[\pi\,\dot\phi - \mathcal{H}(\phi, \pi, \nabla\phi)\big]

Why $\phi$ and $\pi$ are now independent: the Legendre transform's exchange is only complete if the momentum is freed from its definition and allowed to vary on its own; the definition $\pi = \dot\phi$ must be RECOVERED as an equation of motion, not assumed — the standard subtlety of phase-space variational principles, inherited intact from particle mechanics.

Step 2. Vary $\pi$ (freely, no boundary conditions needed — no $\dot\pi$ appears):

\delta_\pi S = \int \mathrm{d}t\,\mathrm{d}^3x\;\Big(\dot\phi - \frac{\delta H}{\delta\pi(\mathbf{x})}\Big)\,\delta\pi = 0 \quad\Longrightarrow\quad \dot\phi(\mathbf{x}) = \frac{\delta H}{\delta\pi(\mathbf{x})} \qquad (3a)

For the scalar field: $\delta H/\delta\pi = \pi$ , recovering $\dot\phi = \pi$ — the Legendre definition returned as dynamics, on schedule.

Step 3. Vary $\phi$ (vanishing at temporal endpoints; integrate $\pi\,\delta\dot\phi$ by parts in time, and the gradient terms of $\mathcal{H}$ by parts in space per the functional-derivative article):

\delta_\phi S = \int \mathrm{d}t\,\mathrm{d}^3x\;\Big(-\dot\pi - \frac{\delta H}{\delta\phi(\mathbf{x})}\Big)\,\delta\phi = 0 \quad\Longrightarrow\quad \dot\pi(\mathbf{x}) = -\frac{\delta H}{\delta\phi(\mathbf{x})} \qquad (3b)

with the working component formula $\delta H/\delta\phi = \partial\mathcal{H}/\partial\phi - \nabla\!\cdot\!\big(\partial\mathcal{H}/\partial(\nabla\phi)\big)$ — for the scalar field, $V'(\phi) - \nabla^2\phi$ .

Step 4. Close the loop. Combine: $\ddot\phi = \dot\pi = \nabla^2\phi - V'(\phi)$ — the Euler-Lagrange equation of the covariant-equations article, split into two first-order halves. The split is not cosmetic: initial data $(\phi, \pi)$ on one slice, evolution first-order in time, conserved $H$ generating the flow — the entire conceptual kit of the Hamiltonian chapters (flows, brackets, Liouville, canonical transformations) now applies to fields verbatim, with $\sum_i \to \int\mathrm{d}^3x$ and partials $\to$ functional derivatives.

Step 5. State the canonical structure that summarizes it all. Define the equal-time Poisson bracket of phase-space functionals (full development: the Poisson-brackets-for-fields article); its fundamental values on the coordinates themselves are

\{\phi(\mathbf{x}), \pi(\mathbf{y})\} = \delta^3(\mathbf{x} - \mathbf{y}), \qquad \{\phi, \phi\} = \{\pi, \pi\} = 0 \qquad (4)

where: the delta is the continuum Kronecker of independent canonical pairs — the bead statement $\{q_i, p_j\} = \delta_{ij}$ through the dictionary, and the object that quantization will replace by $i\hbar$ times the identity structure. Evolution of any functional: $\dot F = \{F, H\}$ , with (3a)–(3b) the special cases $F = \phi, \pi$ .

The formalism is complete; now the calculation that reveals what it built:

Go deeper · Derivation B · 2 minDiagonalizing the free field — an infinite bank of harmonic oscillators

Derivation BDiagonalizing the free field — an infinite bank of harmonic oscillators

We expand the free scalar field in Fourier modes in a box and show the Hamiltonian separates into independent harmonic oscillators, one per allowed wavevector, with frequencies given by the dispersion relation. This is the normal-mode diagonalization of coupled oscillators performed on the continuum — the chapter's single most consequential computation, converting "a field" into "a spectrum of oscillators" exactly, and setting the stage for quantization. Watch the gradient coupling become a per-mode spring constant: the only place $k$ enters.

Step 1. Periodic box of volume $L^3$ , allowed wavevectors $\mathbf{k} = 2\pi\mathbf{n}/L$ (the mode counting of the discrete-to-fields article). Expand both phase-space functions:

\phi(\mathbf{x}) = \frac{1}{\sqrt{L^3}}\sum_{\mathbf{k}} \phi_{\mathbf{k}}\,e^{i\mathbf{k}\cdot\mathbf{x}}, \qquad \pi(\mathbf{x}) = \frac{1}{\sqrt{L^3}}\sum_{\mathbf{k}} \pi_{\mathbf{k}}\,e^{i\mathbf{k}\cdot\mathbf{x}}

with $\phi_{-\mathbf{k}} = \phi_{\mathbf{k}}^*$ (reality of the field); the pairs $(\phi_{\mathbf{k}}, \pi_{\mathbf{k}})$ are the new canonical coordinates.

Step 2. Insert into $H = \int\big[\tfrac12\pi^2 + \tfrac12|\nabla\phi|^2 + \tfrac12 m^2\phi^2\big]\mathrm{d}^3x$ and use the orthogonality $\int e^{i(\mathbf{k}-\mathbf{k}')\cdot\mathbf{x}}\mathrm{d}^3x = L^3\delta_{\mathbf{k}\mathbf{k}'}$ . Each term diagonalizes — the gradient contributing $|\nabla\phi|^2 \to k^2|\phi_{\mathbf{k}}|^2$ :

H = \sum_{\mathbf{k}} \Big[ \tfrac{1}{2}|\pi_{\mathbf{k}}|^2 + \tfrac{1}{2}\,\omega_k^2\,|\phi_{\mathbf{k}}|^2 \Big], \qquad \omega_k^2 = k^2 + m^2 \qquad (5)

where: each summand is a harmonic oscillator Hamiltonian in the canonical pair $(\phi_{\mathbf{k}}, \pi_{\mathbf{k}})$ , with unit mass and spring constant $\omega_k^2$ . No cross terms survive: the modes are exactly independent — the translation invariance of the density guarantees the Fourier basis diagonalizes it, the continuum version of "symmetric matrices diagonalize in normal modes." Why the frequencies are the dispersion relation: the mode's spring constant collects the gradient penalty $k^2$ (neighbour-springs resisting short wavelengths) plus the well curvature $m^2$ (every point's local tether) — the third independent derivation of $\omega^2 = k^2 + m^2$ in this chapter, now as an oscillator spectrum.

Step 3. Solve the bank by the oscillator chapter's normal-variable trick. Define, per mode,

a_{\mathbf{k}} = \frac{\omega_k\,\phi_{\mathbf{k}} + i\,\pi_{\mathbf{k}}}{\sqrt{2\omega_k}} \quad\Longrightarrow\quad \dot a_{\mathbf{k}} = -i\,\omega_k\, a_{\mathbf{k}}, \qquad a_{\mathbf{k}}(t) = a_{\mathbf{k}}(0)\,e^{-i\omega_k t}

Each mode's phase point circulates uniformly; the general field solution is the superposition of phasors — plane waves $e^{i(\mathbf{k}\cdot\mathbf{x} - \omega_k t)}$ with frozen complex amplitudes, i.e. every solution of the free field equation, now organized canonically. The energy per mode is $H_{\mathbf{k}} = \omega_k\,|a_{\mathbf{k}}|^2$ : amplitude-squared times frequency, the classical precursor of " $n$ quanta of $\hbar\omega_k$ ."

Step 4. Open the door (and stop at the threshold). The bracket (4) transcribes to $\{a_{\mathbf{k}}, a^*_{\mathbf{k}'}\} = -i\,\delta_{\mathbf{k}\mathbf{k}'}$ . Canonical quantization — the capstone article's business — performs one substitution: brackets to commutators, $\{\,,\} \to [\,,]/i\hbar$ , making $a_{\mathbf{k}}, a^\dagger_{\mathbf{k}}$ ladder operators, the mode energies $\hbar\omega_k(n_{\mathbf{k}} + \tfrac12)$ , and the quanta — countable, indivisible excitations with energy-momentum $(\hbar\omega_k, \hbar\mathbf{k})$ — particles. Everything quantum about quantum field theory enters through that single substitution; everything else was already here, classical, in equation (5).

Figure

Mode phase planes, computed — equal-energy oscillators of the bank circulating at their own dispersion frequencies

Limiting cases

Case	Result	Meaning
single point (no gradients)	$\mathcal{H} = \tfrac12\pi^2 + V(\phi)$	one particle in a well: field theory contracts to mechanics
free field, mode basis	independent oscillators, $\omega_k = \sqrt{k^2+m^2}$	the exact diagonalization (5)
$m = 0$	$\omega_k = k$ : gapless bank	the rod/EM-like case; $k = 0$ mode free, not oscillating
$m^2 < 0$ at $\phi = 0$	$\omega_k^2 < 0$ for $k < \lvert m\rvert$ : growing modes	hilltop instability, mode by mode — re-expand at the true vacuum
interactions ( $\phi^4$ etc.)	mode oscillators couple; energy migrates between $\mathbf{k}$	no exact diagonalization: the hard, generic case
large occupations ( $k_BT \gg \hbar\omega$ )	classical field description accurate	why this chapter's classical treatment is legitimate
gauge fields	constraints among $(\phi_a, \pi_a)$ ; naive brackets fail	Dirac's constrained formalism (see FAQ)

Common trapManifest covariance is spent, not lost — and the Legendre step is not always innocent

Two structural cautions before trusting the canonical formalism blindly. First: the Hamiltonian formulation picks a time slicing — $\pi$ privileges $\dot\phi$ over spatial gradients, $H$ generates evolution along one observer's time, and the equal-time bracket lives on that observer's "now." Lorentz covariance is not broken (the dynamics generated is exactly the covariant field equations, and boosted observers build equivalent formalisms on their own slices) but it is no longer manifest: term-by-term covariance checks of Lagrangian practice are unavailable, which is why relativistic QFT keeps both formalisms — canonical for states and unitarity, Lagrangian/path-integral for symmetries — and translates constantly. Second: the Legendre transform assumed $\pi = \partial\mathcal{L}/\partial\dot\phi$ is solvable for $\dot\phi$ . When it is not — the momentum conjugate to the electromagnetic $A_0$ vanishes identically; gauge and reparametrization theories generally — the naive formalism produces constraints among the phase-space variables, the brackets must be repaired (Dirac brackets), and "time evolution" itself acquires gauge ambiguity. That machinery (constrained Hamiltonian dynamics) is precisely what Dirac built for quantizing electromagnetism and gravity; know that the scalar field's smooth story is the nice case, and that the warning light is a momentum you cannot invert.

Worked examples

1 · The canonical anatomy of one travelling mode (confidence builder)

The scalar field with $m = 2.0$ (natural units) carries $\phi = A\cos(kz - \omega t)$ with $k = 1.5$ , $A = 0.10$ . (i) Give $\pi(\mathbf{x}, t)$ and verify Hamilton's first equation. (ii) Compute the cycle-averaged energy density and its budget. (iii) Confirm consistency with the density article's energetics.

Step 1. $\pi = \dot\phi = A\omega\sin(kz - \omega t)$ with $\omega = \sqrt{k^2 + m^2} = \sqrt{6.25} = \mathbf{2.50}$ . Hamilton's first equation $\dot\phi = \delta H/\delta\pi = \pi$ : satisfied identically by construction — the content check is the second equation: $\dot\pi = A\omega^2\cos(\cdot)\times(-1)\cdots$ compute: $\dot\pi = -A\omega^2\cos(kz-\omega t)$ , while $\nabla^2\phi - m^2\phi = (-k^2 - m^2)A\cos(\cdot) = -\omega^2 A\cos(\cdot)$ ✓ — equation (3b) verified on the nose.

Step 2. Energy budget, averaging $\cos^2, \sin^2 \to \tfrac12$ : kinetic $\tfrac12\langle\pi^2\rangle = \tfrac14 A^2\omega^2 = \mathbf{0.0156}$ ; gradient $\tfrac14 A^2k^2 = \mathbf{0.0056}$ ; well $\tfrac14 A^2m^2 = \mathbf{0.0100}$ . Total $\langle\mathcal{H}\rangle = \mathbf{0.0312}$ per unit volume.

Step 3. Cross-check: identical numbers to the density article's example (computed there from the Lagrangian side) — as they must be, since $\mathcal{H} = T^{00}$ is the same energy read through the Legendre transform. The kinetic share equals the sum of the other two ( $0.0156 = 0.0056 + 0.0100$ ): the oscillator virial theorem, holding mode by mode in the bank.

2 · When is a field classical? Phonons versus Higgs (exam level)

(i) For the steel wire of the discrete-to-fields article ( $L = 1.00$ m, $v = 5048$ m/s), find the lowest mode's frequency and its thermal occupation at $T = 300$ K. (ii) Repeat the occupation estimate for one mode of a field with frequency floor $mc^2 = 125$ GeV. (iii) Conclude.

Step 1. Lowest mode: $\omega_1 = \pi v/L = \pi \times 5048 = \mathbf{1.59\times10^{4}}$ rad/s. Energy quantum: $\hbar\omega_1 = 1.055\times10^{-34} \times 1.59\times10^4 = \mathbf{1.67\times10^{-30}}$ J.

Step 2. Thermal budget per mode: $k_BT = 1.381\times10^{-23}\times300 = 4.14\times10^{-21}$ J. Occupation $\bar n \approx k_BT/\hbar\omega_1 = 4.14\times10^{-21}/1.67\times10^{-30} = \mathbf{2.5\times10^{9}}$ quanta. With billions of quanta per mode, the discreteness is utterly invisible: the classical field description of this chapter is not an approximation of convenience but accurate to parts per billion — which is why classical elasticity, acoustics, and this chapter's mechanics work.

Step 3. Higgs-like mode: $\hbar\omega \ge mc^2 = 125\ \mathrm{GeV} = 2.0\times10^{-8}$ J, versus $k_BT = 4.14\times10^{-21}$ J at room temperature: $\bar n \sim e^{-\hbar\omega/k_BT} \approx e^{-4.8\times10^{12}}$ — zero, beyond hyperbole. Such a field is never classically occupied in the current universe; it exists only through its quanta (produced two at a time at colliders) and its vacuum structure. Conclusion: one formalism, two regimes, decided by $\bar n = k_BT/\hbar\omega$ per mode: classical fields are the $\bar n \gg 1$ face of the oscillator bank, particles the $\bar n \sim$ few face — and equation (5) is the shared ledger.

Origins · 1 minPaul Dirac and how this was discovered

Paul Dirac

1902–1984 · Cambridge

Dirac built more of the canonical bridge between this article and quantum field theory than anyone. His 1925 doctoral insight — that Heisenberg's mysterious noncommuting products shadow the Poisson bracket, commutator $= i\hbar \times$ bracket — is the precise substitution this article's final derivation gestures at; "canonical quantization" is Dirac's recipe. Two years later he applied it to the electromagnetic field's oscillator bank, quantizing radiation mode by mode exactly in the manner of equation (5) and inventing, in that single 1927 paper, quantum electrodynamics, creation and annihilation operators, and the quantum theory of emission and absorption. When the naive Legendre transform failed for gauge fields — vanishing momenta, the constraint disease of this article's warning — Dirac again supplied the cure: his constrained-Hamiltonian formalism (1950s) and Dirac brackets remain the standard machinery for quantizing gauge theories and gravity, and his slim Lectures on Quantum Mechanics (1964) expounding it may be the most consequential 87-page book in theoretical physics. Famously economical of speech (colleagues defined a "dirac": one word per hour), he wrote physics the same way; asked once to summarize his philosophy, he went to a blackboard and wrote that physical laws should have mathematical beauty. The formalism of this article — pairs, brackets, oscillators, all waiting for one substitution — was, in his hands, exactly that.

History

1833

— Hamilton recasts mechanics as paired first-order equations generated by one function: the formalism awaiting fields.

1925

— Dirac identifies the commutator as

i\hbar

times the Poisson bracket — the quantization recipe in one line.

1926

— Born, Heisenberg and Jordan quantize the vibrating string's normal modes: the first quantum field, phonons before the name.

1927

— Dirac quantizes the electromagnetic oscillator bank: quantum electrodynamics and the photon as a mode quantum.

1950

— Dirac's constrained-Hamiltonian formalism repairs the canonical method for gauge theories — the toolkit for QED, Yang-Mills, and gravity.

Going further

References. Goldstein, Poole & Safko, Classical Mechanics, 3rd ed., ch. 13 (Hamiltonian formulation for continuous systems) · Landau & Lifshitz, The Classical Theory of Fields, §32 · José & Saletan, Classical Dynamics, ch. 9 · Dirac, Lectures on Quantum Mechanics (1964) — constrained Hamiltonian systems Key relations: $\pi=\partial\mathcal{L}/\partial\dot{\phi}$ , $\mathcal{H}=\pi\dot{\phi}-\mathcal{L}$ , and $T^{00}=\mathcal{H}$ . The Legendre transform cancels the $\mathrm{d}\dot{\phi}$ terms cleanly; combining Hamilton's equations recovers $\ddot{\phi}-\nabla^2\phi+V'=0$ ; and the bracket-evolution equivalence $\dot{\phi}=\{\phi,H\}$ follows directly from $\{\phi,\pi\}=\delta^3$ . The mode diagonalization (5) and normal variables $a_{\mathbf{k}}$ use the standard periodic-box setup with classical content only. Worked-example numbers: $\omega = 2.50$ ; energy budget $0.0156/0.0056/0.0100$ matching the Lagrangian-side computation; $\omega_1 = \pi\times5048 = 1.59\times10^4$ rad/s, $\hbar\omega_1 = 1.67\times10^{-30}$ J, $\bar n = 2.5\times10^9$ at 300 K; $125$ GeV $= 2.0\times10^{-8}$ J.

Builds on: The Lagrangian Density and Field Action · Covariant Euler-Lagrange Equations for Fields · Functional Derivatives and Field Variations · From Discrete Systems to Continuous Fields Leads to: Poisson Brackets for Fields · The Canonical Stress-Energy Tensor · Noether's Theorem for Continuous Symmetries · From the Action Principle to Quantum Field Theory

If every mode is an independent oscillator, why is field theory considered hard?

Because independence dies with the first interaction term, and everything hard in the subject lives in the corpse. Add $\tfrac{\lambda}{4!}\phi^4$ to the density and its Fourier transcription couples quartets of modes with $\mathbf{k}_1 + \mathbf{k}_2 + \mathbf{k}_3 + \mathbf{k}_4 = 0$ : the bank becomes infinitely many oscillators, all talking, and no basis diagonalizes them again. Classically, this is where the chapter's other thread re-enters: mode-coupled oscillator systems are exactly the arena of resonances, near-integrability and chaos — the Fermi-Pasta-Ulam-Tsingou experiment, the founding computation of nonlinear science, was precisely a coupled mode bank refusing to thermalize on schedule, and wave turbulence theory is statistical mechanics for such banks. Quantum mechanically, the coupled bank is perturbation theory's home: free-oscillator states dressed order by order in $\lambda$ , Feynman diagrams as the bookkeeping of mode-quartet interactions, renormalization as the discovery that the ultraviolet tail of the bank (the unbounded mode counting of the second article) back-reacts on every low- $k$ answer. So the honest summary: this article solves the free theory completely — and thereby defines the reference frame in which a century of work on interactions is organized. "Hard" starts one term later.

The Hamiltonian picks a time direction — does anything physical depend on the slicing choice?

No, and the statement can be made precise at three levels. Kinematically, phase spaces built on different simultaneity slices are related by canonical transformations: the boost generators (constructed from the stress-energy tensor's moments in the Poisson-brackets and stress-energy articles) implement the change of slicing within the formalism, and the Poincaré algebra's closure under brackets is the certificate that all slicings cohere. Dynamically, the evolution generated on any slice reproduces the same covariant field equations, so predictions about invariant questions — scattering outcomes, conserved charges, energies of bound configurations — are slicing-independent even though intermediate objects (equal-time brackets, the split of energy into kinetic/gradient/well) are not. Quantum mechanically the issue was historically sharper (equal-time commutators look frame-dependent), and its resolution — demonstrating that canonical QED's physical amplitudes are Lorentz invariant — was a genuine achievement (Schwinger and Tomonaga's covariant reformulations; the interaction picture on arbitrary spacelike surfaces). Two caveats define the boundary of the reassurance: on curved spacetimes without preferred slicings, mode banks defined by different time choices genuinely disagree about particle content (the Unruh and Hawking effects live exactly in that gap — "particle" is slicing-dependent even though the field is not); and numerically, bad slicings can make well-posed problems stiff — the art of gauge choice in numerical relativity is choosing time wisely inside a formalism that permits any choice.

What is the zero-point energy problem visible already at the classical-quantum threshold here?

Quantize the bank and each oscillator contributes ground-state energy $\tfrac12\hbar\omega_k$ ; summing over the modes — whose density, from the discrete-to-fields article, is $V\mathrm{d}^3k/(2\pi)^3$ , unbounded above — gives a vacuum energy density $\int \tfrac12\hbar\omega_k\,\mathrm{d}^3k/(2\pi)^3$ that diverges as the fourth power of the cutoff. This is the mode-counting infinity of the second article striking its first blow, and its status is instructively split. For differences and derivatives of the zero-point sum, the physics is real and measured: the Casimir effect (conducting plates modify the allowed mode bank between them; the vacuum-energy difference produces an attractive pressure, measured to percent accuracy) and Lamb-shift-type phenomena confirm that mode zero-point energies respond to boundaries exactly as the oscillator picture predicts. For the absolute value, the problem is the most famous open embarrassment in physics: gravity couples to total energy density, the observed cosmological constant corresponds to $\sim 10^{-9}$ J/m³, and naive cutoff estimates at the Planck scale overshoot by $\sim 120$ orders of magnitude. Within this course's classical scope, the honest takeaways: the classical field's energy (2) is well-defined and cutoff-free; the divergence is purchased with $\hbar$ times unbounded mode counting; and every modern approach (renormalization of the vacuum term, symmetry cancellations, anthropic selection) is a proposed rule for handling the bank's infinite basement.

How do the constraints of gauge theories concretely change this article's story?

Run the machinery on electromagnetism and watch where it stalls. The Lagrangian density $-\tfrac14 F^2$ yields momenta $\pi^\mu = \partial\mathcal{L}/\partial\dot A_\mu = -F^{0\mu}$ : the spatial components give $\boldsymbol{\pi} = \mathbf{E}$ (healthy — the electric field is the momentum conjugate to the vector potential), but $\pi^0 \equiv 0$ identically — the Legendre transform cannot be inverted for $\dot A_0$ , and the phase space is born with a constraint. Consistency of that constraint under evolution ( $\dot\pi^0 = 0$ ) generates a second one: $\nabla\!\cdot\!\mathbf{E} = \rho$ — Gauss's law appears not as an evolution equation but as a restriction on allowed initial data, one constraint per point of space. Dirac's classification then organizes the repair: both constraints are "first class," generating gauge transformations rather than dynamics; physical states are equivalence classes; and one either works with the constrained brackets (Dirac brackets after gauge fixing) or keeps the constraints as operators annihilating physical states in the quantum theory. The scalar field's clean narrative — every $(\phi(\mathbf{x}), \pi(\mathbf{x}))$ pair independent, brackets (4) unmodified — is thus the unconstrained special case; gauge theories, gravity (where the entire Hamiltonian is constraints — the ADM $H \approx 0$ that haunts quantum gravity), and even the relativistic particle live in the constrained general case. The conceptual headline survives everywhere: identify the true canonical pairs, and quantization is one substitution away.

Is the Hamiltonian always the energy — and is it always conserved?

For this chapter's fields, yes on both counts, but the identifications are theorems with hypotheses, not definitions, and knowing the hypotheses prevents real errors. $H =$ energy requires the standard structure: a Lagrangian density quadratic in $\dot\phi$ with no terms linear in it, and coordinates (here: the field variables themselves) chosen without explicit time dependence — then the Legendre transform reproduces kinetic-plus-potential, and $\mathcal{H} = T^{00}$ matches the stress-energy tensor's energy density. $H$ conserved requires no explicit time dependence in $\mathcal{L}$ : then $\mathrm{d}H/\mathrm{d}t = \partial H/\partial t = 0$ along the flow (the base articles' energy theorem — one line via Hamilton's equations, with the two gradient terms cancelling pairwise). Both hypotheses fail in physically important settings. A field in a time-dependent background — the paradigm is a scalar field in an expanding universe, where the metric stretches the mode wavelengths — has an explicitly time-dependent mode Hamiltonian: $H$ is neither conserved nor unambiguously "the" energy, each mode behaves as an oscillator with slowly varying frequency, and the right conserved-ish quantities are the adiabatic invariants of the Hamiltonian chapters ( $E_k/\omega_k$ , the classical seed of "comoving particle number"). Driven systems (a cavity field with moving mirrors) similarly trade $H$ -conservation for work done by the boundary — the dynamical Casimir setting. The reflex to keep: before quoting "energy conservation" for a field, check who owns the time dependence — the field's own dynamics (fine), or the stage it stands on (then reach for adiabatic invariants and pumping physics instead).

Classically the a-amplitudes are continuous — what physically forbids half a quantum after quantization, and is there classical residue of that discreteness?

The discreteness is spectral, not amplitudinal, and its origin is the same as in the one-oscillator quantum problem: the ladder structure of the operator algebra. Quantizing $\{a, a^*\} = -i$ to $[\hat a, \hat a^\dagger] = 1$ makes $\hat n = \hat a^\dagger\hat a$ have non-negative integer spectrum — a theorem of the algebra, with no classical counterpart because classical $|a|^2$ is a continuous coordinate, not an operator with a spectrum. Physical consequences follow the operator, not the expectation value: energy exchanged with matter comes in lumps $\hbar\omega_k$ (photoelectric threshold behaviour), correlation measurements register integer coincidences (a single photon never triggers two detectors), and "half a quantum" is not a state in the Hilbert space. The classical residue runs the other way: coherent states — eigenstates of $\hat a$ with eigenvalue the classical amplitude — reproduce this article's phasor dynamics exactly, with quantum fluctuations of fixed absolute size $\sim\sqrt{\hbar\omega}$ ; at large $|a|$ (the $\bar n \gg 1$ regime of the worked example) those fluctuations are relatively negligible, and the classical field of this chapter emerges as the coherent-state limit of the quantum bank. So the classical theory is not overthrown but located: it is the large-amplitude face of the oscillator algebra, which is why every formula of this article survives inside QFT as the expectation-value skeleton.