Optical-Mechanical Analogy

In 1834 William Rowan Hamilton published a fact so far ahead of its time that physics needed ninety years to catch up. He had spent a decade building a "theory of systems of rays" — geometrical optics organized around a single characteristic function whose level surfaces are wavefronts and whose gradients are rays. Then he noticed that mechanics fits the same mold exactly: surfaces of constant action play the wavefronts, particle trajectories thread them perpendicularly like rays, and the momentum $\sqrt{2m(E-V)}$ stands wherever optics writes the refractive index $n$ . Fermat's principle of least optical path and Maupertuis' principle of least abbreviated action become one variational statement wearing two costumes. Read literally, the analogy whispers something radical. Geometrical optics is not a fundamental theory — it is the short-wavelength limit of wave optics, blind to diffraction and interference. If classical mechanics is mathematically identical to geometrical optics, then classical mechanics might likewise be the short-wavelength limit of an undiscovered wave mechanics. Hamilton had no experimental reason to take that step; matter showed no hint of a wavelength. The hint arrived in 1924 when de Broglie proposed $\lambda=h/p$ , and in 1926 Schrödinger ran Hamilton's logic in reverse — hunting the wave equation whose eikonal limit is the Hamilton-Jacobi equation — and found wave mechanics. This page builds the dictionary quantitatively: wavefronts from Hamilton's Characteristic Function, the eikonal equation from the wave equation, a mechanical Snell's law, and the phase-versus-group-velocity subtlety that seeded matter waves. It is the conceptual heart of this chapter's bridge from classical to quantum.

Key takeaways

The time-independent Hamilton-Jacobi equation $|\nabla W|^2=2m(E-V)$ and the optical eikonal equation $|\nabla L|^2=n^2$ are the same equation under the dictionary $W\leftrightarrow L$ , $p=\sqrt{2m(E-V)}\leftrightarrow n$ .
Surfaces of constant $W$ are mechanical wavefronts; trajectories are the orthogonal rays, with $\mathbf{p}=\nabla W$ playing the role of $\nabla L=n\hat{\mathbf{s}}$ .
Fermat's principle $\delta\int n\,\mathrm{d}s=0$ and Maupertuis' principle $\delta\int p\,\mathrm{d}s=0$ are the two faces of one variational law; a particle refracts at a potential step by a Snell's law $p_1\sin\theta_1=p_2\sin\theta_2$ .
The wavefronts do not travel with the particle: they sweep along at phase speed $u=E/p$ while the particle moves at $v=p/m$ — the mismatch resolved only by wave packets and group velocity.
The analogy is exact ray-to-ray but silent about wavelength; supplying $\lambda_{\mathrm{dB}}=h/p$ (de Broglie, 1924) converts it into quantum mechanics (Schrödinger, 1926) — the machinery is in WKB Approximation and the Eikonal Connection.
Working fruits today: electron microscopes are literally refraction of matter waves by potential "lenses"; seismic tomography and gradient-index optics integrate the same Hamilton-Jacobi ray equations.
Key idea: surfaces of constant characteristic function are mechanical wavefronts, trajectories are their orthogonal rays, and the Hamilton-Jacobi and eikonal equations coincide under the dictionary refractive index equals momentum — classical mechanics is a geometrical optics awaiting its wave theory.

The core insight

Surfaces $W(\mathbf{q})=\text{const}$ of Hamilton's characteristic function are mechanical wavefronts; trajectories are the orthogonal rays with $\mathbf{p}=\nabla W$ playing the role of $\nabla L = n\,\hat{\mathbf{s}}$ . The Hamilton-Jacobi equation and the optical eikonal equation are the same equation under the dictionary $n \leftrightarrow \sqrt{2m(E-V)}$ .

Where this lives today

Every electron microscope is the analogy running as infrastructure: magnetic and electrostatic fields are gradient-index lenses for matter waves (Examples 1-2), and lens designers still compute with the electron-optical refractive index $n\propto\sqrt{E-V}$ . The same Hamilton-Jacobi ray machinery, with $n=c_0/c(\mathbf{r})$ , powers seismic travel-time tomography — imaging Earth's interior by integrating mechanical rays — and gradient-index (GRIN) optics, from smartphone camera lenses to the graded cores of optical fibers. Atom interferometry closes the historical loop: laser-cooled atoms with picometer de Broglie wavelengths diffract off light gratings, and the resulting interferometers measure gravity and rotation precisely enough to hunt dark energy — Maupertuis' $\int p\,\mathrm{d}s$ read as an interferometric phase. Conceptually, the analogy remains the cleanest statement of why semiclassical physics works: classical mechanics is the caustic-and-ray skeleton of quantum mechanics, sharp wherever actions dwarf $\hbar$ . Its quantitative completion is the WKB expansion of WKB Approximation and the Eikonal Connection; its historical completion — how the eikonal limit plus de Broglie's postulate pinned down the Schrödinger equation — is the story finished in Connection to the Schrodinger Equation.

The physical picture

Drop a stone in a pond and watch the rings spread: the crests are wavefronts, and the energy travels along paths perpendicular to every crest — the rays. Hamilton's observation is that mechanics already contains this picture. Take the family of surfaces on which the characteristic function $W$ is constant. The momentum of any trajectory is $\mathbf{p}=\nabla W$ : always perpendicular to the local surface, threading the family like rays through wavefronts. One function $W$ organizes an entire bundle of trajectories at once — every particle of energy $E$ launched perpendicular to one wavefront — exactly as one eikonal organizes a beam of light.

Figure

The same geometry twice — light rays crossing wavefronts in a graded-index medium, and particle trajectories crossing constant-W surfaces in a potential

The dictionary earns its keep at an interface. Light entering glass slows down, its wavefronts crowd together, and the ray bends toward the normal: Snell's law, $n_1\sin\theta_1=n_2\sin\theta_2$ . Now push a particle across a step where the potential drops. Its momentum $p=\sqrt{2m(E-V)}$ grows, its constant- $W$ surfaces crowd together, and the trajectory bends toward the normal: $p_1\sin\theta_1=p_2\sin\theta_2$ , derived below by the same wavefront-matching argument opticians use. High momentum is high refractive index. An electrostatic lens focuses electrons for exactly the reason a glass lens focuses light. But the analogy carries a built-in tension, and it is the productive kind. The wavefronts are not carried along by the particle: Derivation A shows they sweep through configuration space at phase speed $u=E/p$ , while the particle crawls at $v=p/m$ — for a free particle the front moves at half the particle's speed. Which one is "the motion"? In optics the same puzzle is resolved by wave packets: individual crests move at the phase velocity, but a localized pulse — the thing you can detect — moves at the group velocity. Demanding that a matter-wave packet's group velocity equal the particle velocity is a one-line calculation once $\lambda=h/p$ is granted, and it works. That consistency check, run by de Broglie and Schrödinger, is the hinge on which this chapter's whole classical-to-quantum bridge turns. Keep one caveat in view throughout: everything here is ray-to-ray. Hamilton proved classical trajectories are the rays of some wave geometry; he could not say what waves, or how long their wavelength. The analogy is silent exactly where diffraction, interference, and tunneling live. That silence is not a defect — it is a precise map of what quantum mechanics would have to add.

The dictionary, quantitatively

Start on the mechanics side. For a conservative system, Hamilton's principal function separates as $S(\mathbf{q},t)=W(\mathbf{q})-Et$ (the split constructed in Hamilton's Characteristic Function), and the time-independent Hamilton-Jacobi equation of Time-Independent Hamilton-Jacobi Equation fixes the spatial function $W$ :

|\nabla W|^2 = 2m\big(E-V(\mathbf{r})\big) \qquad (1)

where: $W$ — the characteristic function, whose level sets are the mechanical wavefronts; $\nabla W=\mathbf{p}$ — the momentum field, of magnitude $p=\sqrt{2m(E-V)}$ . Read geometrically: (1) prescribes how tightly the wavefronts pack at each point — spacing inversely proportional to the local momentum. Now the optics side. A scalar monochromatic wave in a medium of refractive index $n(\mathbf{r})$ obeys the Helmholtz equation, and in the short-wavelength regime its phase $L$ (the eikonal, from the Greek for image) satisfies

|\nabla L(\mathbf{r})|^2 = n^2(\mathbf{r}) \qquad (2)

where: $L$ — optical path length accumulated by the wave, so wavefronts are $L=\mathrm{const}$ and rays run along $\nabla L=n\hat{\mathbf{s}}$ . Derivation B extracts (2) honestly from the wave equation, and the extraction is worth watching: it is the same expansion that extracts classical mechanics from the Schrödinger equation in WKB Approximation and the Eikonal Connection. Equations (1) and (2) are one equation. The complete dictionary reads:

L \leftrightarrow W,\qquad n(\mathbf{r}) \leftrightarrow \sqrt{2m\big(E-V(\mathbf{r})\big)},\qquad \text{rays} \leftrightarrow \text{trajectories} \qquad (3)

where: the correspondence is exact at the level of ray geometry — every theorem of geometrical optics has a mechanical twin. A potential well is a high-index inclusion (a lens); a potential barrier is a low-index gap; the turning surface $V=E$ is total internal reflection, where $n\to0$ and rays cannot pass. The variational principles translate along with the field equations. Fermat's principle selects light rays as paths of stationary optical length; under the dictionary it becomes Maupertuis' principle of stationary abbreviated action:

\delta\!\int_A^B n\,\mathrm{d}s = 0 \quad\longleftrightarrow\quad \delta\!\int_A^B p\,\mathrm{d}s = 0 \qquad (4)

where: $\mathrm{d}s$ — arc length; both principles compare paths between fixed endpoints at fixed frequency/energy. Fermat (1662) and Maupertuis (1744) were arguing about theology and economy in nature a century apart; Hamilton showed they had proved the same theorem twice. Note what is extremized: path length weighted by $n$ or $p$ — not time. The distinction matters and is a favorite examiner's trap (see the FAQ). Finally, the kinematics of the wavefronts themselves. Fix a value of $S$ and follow that surface as time runs:

u = \frac{E}{|\nabla W|} = \frac{E}{p},\qquad v=\frac{p}{m},\qquad uv=\frac{E}{m} \qquad (5)

where: $u$ — the normal speed of the constant- $S$ wavefront (Derivation A); $v$ — the speed of the particle riding the ray. They are different functions of position: fronts race where the particle dawdles ( $u\propto1/p$ ) and dawdle where it races. Their product is fixed by the energy. This is the crack in the pure ray picture through which wave mechanics eventually entered.

Derivation AWavefronts of constant action propagate at speed E over p

We are deriving the propagation speed of the mechanical wavefronts — the level surfaces of Hamilton's principal function — for a conservative system. The strategy is direct kinematics: freeze a value of $S=W-Et$ , ask how far along its normal the surface must shift in time $\mathrm{d}t$ to keep $S$ constant, and read off the normal speed. The payoff is the mismatch between this phase speed and the particle speed, which the analogy alone cannot resolve.

Step 1. Fix $S(\mathbf{q},t)=S_0$ . Since $S=W(\mathbf{q})-Et$ , the frozen surface satisfies $W(\mathbf{q})=S_0+Et$ : as $t$ advances, the surface must migrate to configurations of ever larger $W$ — it sweeps outward through the nested family of $W$ -surfaces.

Step 2. Displace a point of the surface by $\mathrm{d}\mathbf{q}$ along the local normal $\hat{\mathbf{n}}=\nabla W/|\nabla W|$ . Keeping $S$ constant requires

\mathrm{d}W=\nabla W\cdot\mathrm{d}\mathbf{q}=|\nabla W|\,\mathrm{d}s=E\,\mathrm{d}t

Why the normal direction is the right one to track: motion of a surface tangent to itself is invisible; the only geometrically meaningful "speed of a wavefront" is along its normal, exactly as for wave crests in optics.

Step 3. Solve for the normal speed and insert $|\nabla W|=p$ from equation (1):

u=\frac{\mathrm{d}s}{\mathrm{d}t}=\frac{E}{|\nabla W|}=\frac{E}{\sqrt{2m(E-V)}}

Step 4. Compare with the particle: $v=p/m$ , so $uv=E/m$ everywhere, and for a free particle ( $E=p^2/2m$ ) the front moves at exactly half the particle speed, $u=v/2$ . Front and particle agree nowhere; slow particles sit inside fast wavefronts and vice versa. Why this mismatch is a feature: identical behavior occurs for any dispersive wave — individual crests (phase velocity) and localized packets (group velocity) travel differently. Requiring a matter-wave packet with $\omega=E/\hbar$ , $k=p/\hbar$ gives group velocity $v_g=\mathrm{d}\omega/\mathrm{d}k=\mathrm{d}E/\mathrm{d}p=p/m=v$ : the packet moves with the particle even though no crest does. Verified: $u=E/p$ , $v=p/m$ , $uv=E/m$ match Goldstein §10.8 and Landau-Lifshitz Mechanics §47; the free-particle check $u=v/2$ agrees with the non-relativistic de Broglie dispersion $\omega=\hbar k^2/2m$ .

Go deeper · Derivation B · 1 minThe eikonal equation from the wave equation

Derivation BThe eikonal equation from the wave equation

We are deriving the eikonal equation (2) — the optics half of the analogy — from the scalar Helmholtz equation, to exhibit exactly what "geometrical optics is the short-wavelength limit" means. The strategy is the eikonal ansatz $\psi=A\,e^{ik_0L}$ followed by ordering in the large parameter $k_0$ : the vacuum wavenumber plays the role $1/\hbar$ plays for matter waves, and watching the parallel is the point of the exercise.

Step 1. In a medium of index $n(\mathbf{r})$ , a monochromatic scalar wave obeys the Helmholtz equation

\nabla^2\psi+k_0^2\,n^2(\mathbf{r})\,\psi=0,\qquad k_0=\frac{2\pi}{\lambda_0}

Step 2. Insert $\psi=A(\mathbf{r})\,e^{ik_0L(\mathbf{r})}$ with real amplitude $A$ and real eikonal $L$ , and divide by $\psi$ :

\frac{\nabla^2A}{A}+2ik_0\,\frac{\nabla A\cdot\nabla L}{A}+ik_0\nabla^2L-k_0^2|\nabla L|^2+k_0^2n^2=0

Step 3. Order in $k_0$ . The $O(k_0^2)$ terms dominate when the wavelength is short compared to the scale of variation of $n$ and $A$ :

|\nabla L|^2=n^2

which is equation (2) — the eikonal equation, with rays along $\nabla L$ . Why the condition is geometric: the neglected terms are smaller by $\left(\lambda_0/\ell\right)$ , with $\ell$ the scale over which the medium or amplitude varies. Geometrical optics needs a slowly varying medium, not a weak one — the exact analogue of the WKB validity condition.

Step 4. The $O(k_0)$ terms give $2\nabla A\cdot\nabla L+A\nabla^2L=0$ , i.e. $\nabla\cdot(A^2\nabla L)=0$ : intensity $A^2$ is transported along the rays, conserved in every ray tube. This is optics' transport equation, twin of the probability-continuity equation at order $\hbar$ in WKB Approximation and the Eikonal Connection.

Step 5. Under the dictionary $k_0L\to S/\hbar$ , $n\to\sqrt{2m(E-V)}$ (up to constants), Steps 1-4 map term-for-term onto the $\hbar$ -expansion of the Schrödinger equation: the eikonal equation becomes Hamilton-Jacobi, the transport law becomes probability conservation. One computation, two theories. Verified: matches Born & Wolf, Principles of Optics §3.1 (eikonal and transport equations); the correspondence table to the WKB expansion follows Goldstein §10.8.

Figure

Refraction at a potential step — wavefronts crowd where momentum is high, and the trajectory bends toward the normal exactly like light entering glass

Go deeper · Derivation C · 1 minMechanical Snell law from wavefront matching

Derivation CMechanical Snell law from wavefront matching

We are deriving the refraction law for a particle crossing a potential step, using only the continuity of the wavefront family — no forces, no Newton. The strategy is the optician's: across the interface the function $W$ is continuous, so the component of $\nabla W$ tangent to the interface cannot jump; only the normal component can. This is worth doing precisely because the usual force-based derivation obscures that refraction is pure wavefront geometry.

Step 1. Let two half-spaces carry constant potentials $V_1$ (incidence side) and $V_2$ , meeting at a plane. In each region, equation (1) fixes the magnitude $|\nabla W|=p_i=\sqrt{2m(E-V_i)}$ : wavefront spacing is uniform in each region but different between them.

Step 2. $W$ is continuous across the interface (the wavefront family does not tear), so its derivative along the interface is the same on both sides:

(\nabla W)^{(1)}_{\parallel}=(\nabla W)^{(2)}_{\parallel}

Why only the tangential component is protected: continuity of a function across a surface constrains derivatives within the surface; the normal derivative is free to jump, and does — that jump is the impulsive normal force of the step.

Step 3. Write each side's tangential component via the angle $\theta_i$ between the trajectory (along $\nabla W$ ) and the interface normal: $|\nabla W|_\parallel=p_i\sin\theta_i$ . Equating:

p_1\sin\theta_1=p_2\sin\theta_2

— Snell's law with momentum as the refractive index, completing dictionary entry (3).

Step 4. Consistency check against Newton: the step exerts force only along the normal, so the tangential momentum $p\sin\theta$ is conserved — the same law from mechanics. The wavefront argument, however, generalizes instantly to continuously graded $V(\mathbf{r})$ , where it integrates to the full ray equation of gradient-index optics. Verified: form matches optical Snell's law under $n\to p$ and the conserved-tangential-momentum argument; consistent with Landau-Lifshitz Mechanics §47.

Limiting cases

case	result	meaning
free particle, $V=0$	$\|\nabla W\|=\sqrt{2mE}$ const: plane wavefronts, straight rays	the mechanical vacuum: $n=\mathrm{const}$ , no refraction
potential step	$p_1\sin\theta_1=p_2\sin\theta_2$	Snell's law with momentum as index; lower $V$ bends trajectories toward the normal
turning surface $V\to E$	$p\to0$ : the "index" vanishes, rays cannot penetrate	total internal reflection = classical turning point; caustics form here
$\lambda\ll\ell$ (optics)	eikonal equation exact	geometrical optics valid; diffraction is the $O(\lambda/\ell)$ correction
$\hbar\to0$ (mechanics)	Schrödinger $\to$ Hamilton-Jacobi	classical mechanics is the eikonal limit of wave mechanics — the analogy read backwards
relativistic de Broglie wave	$u=c^2/v>c$	with rest energy included, fronts outrun light; harmless, since signals travel at group velocity $v<c$

Common trapThe wavefronts are not the trajectories, and they do not move at the particle speed

Two conflations, both fatal. First: surfaces of constant $W$ are perpendicular to the motion, not along it — a wavefront is a set of simultaneous positions of many possible particles, not the path of one. Second: the fronts sweep at phase speed $u=E/p$ while the particle moves at $v=p/m$ ; for a free non-relativistic particle $u=v/2$ , and in de Broglie's relativistic treatment $u=c^2/v>c$ . Neither is a paradox — no energy or information rides a phase front — but confusing phase velocity with particle (group) velocity wrecks every calculation it touches. Resolving this mismatch honestly is precisely what forced wave packets, and with them wave mechanics, into existence.

Worked examples

Example 1 — electron refraction at an accelerating grid (confidence builder). An electron with kinetic energy $E=100$ eV crosses a grid into a region at potential $+50$ V (so $V_2=-50$ eV), striking the interface at $\theta_1=30.0°$ from the normal. Find the refraction angle and the effective index ratio.

Step 1. Momenta from equation (1): $p\propto\sqrt{E-V}$ , so $\dfrac{p_2}{p_1}=\sqrt{\dfrac{100-(-50)}{100}}=\sqrt{1.50}=1.225$ — the "glass" side is the high-momentum side.

Step 2. Snell's law: $\sin\theta_2=\dfrac{p_1}{p_2}\sin\theta_1=\dfrac{0.500}{1.225}=0.408$ , giving $\theta_2=24.1°$ — bent toward the normal, like light entering glass with $n_2/n_1=1.225$ .

Step 3. The matter wavelengths confirm the picture. $p_1=\sqrt{2m_eE}=\sqrt{2(9.109\times10^{-31})(100\times1.602\times10^{-19})}=5.40\times10^{-24}\,\mathrm{kg\,m/s}$ , so $\lambda_1=h/p_1=1.23\times10^{-10}\,\mathrm{m}=0.123$ nm; after the grid $\lambda_2=\lambda_1/1.225=0.100$ nm. Wavefronts crowd by exactly the index ratio, as the hero figure shows.

Example 2 — the wavelength of a microscope electron (exam level). A transmission electron microscope accelerates electrons through $U=300$ kV. Find the de Broglie wavelength (the accelerating energy is a substantial fraction of $m_ec^2=511$ keV, so include the relativistic correction) and compare with atomic spacings.

Step 1. The exact relation is $\lambda=\dfrac{h}{\sqrt{2m_eeU\left(1+\dfrac{eU}{2m_ec^2}\right)}}$ ; the bracket corrects the non-relativistic $p=\sqrt{2m_eeU}$ .

Step 2. Evaluate the pieces: $eU=300\,\mathrm{keV}=4.806\times10^{-14}$ J; $2m_e\,eU=2(9.109\times10^{-31})(4.806\times10^{-14})=8.76\times10^{-44}$ ; the relativistic bracket is $1+\dfrac{300}{2\times511}=1.294$ .

Step 3. So $p=\sqrt{8.76\times10^{-44}\times1.294}=\sqrt{1.133\times10^{-43}}=3.37\times10^{-22}\,\mathrm{kg\,m/s}$ and

\lambda=\frac{6.626\times10^{-34}}{3.37\times10^{-22}}=1.97\times10^{-12}\,\mathrm{m}

$\lambda\approx2.0$ pm — about a hundred times smaller than an atomic spacing ( $\sim0.2$ nm).

Step 4. The lesson of the analogy: resolution is not wavelength-limited ( $\lambda/d\sim10^{-2}$ ) but aberration-limited — electron lenses are refracting potentials (Example 1 iterated smoothly), and shaping $V(\mathbf{r})$ precisely is harder than grinding glass. Modern aberration-corrected instruments reach $\sim50$ pm, resolving single atoms: Hamilton's dictionary, operating at industrial scale.

Origins · 1 minWilliam Rowan Hamilton and how this was discovered

William Rowan Hamilton

1805–1865 · Irish mathematician who unified optics and mechanics under one characteristic function

Hamilton was appointed Astronomer Royal of Ireland while still an undergraduate at Trinity College Dublin. His Theory of Systems of Rays (1828) rebuilt optics around a single characteristic function; by 1834-35 he had recast dynamics in the same mold, giving mechanics its canonical equations and the principal function at the heart of this chapter. The optical-mechanical analogy was, to him, the deepest of his results — and the least used: contemporaries treated it as formal elegance. He predicted conical refraction (confirmed within months, a sensation at the time), invented quaternions in a flash of insight on Broome Bridge, and died thirty years too early to see Schrödinger vindicate his conviction that mechanics and wave optics are one subject.

History

1662

— Fermat derives refraction from a principle of least time, igniting a century of debate about economy in nature

1744

— Maupertuis states the principle of least action for particles; Euler and Lagrange make it precise

1828

— Hamilton's Theory of Systems of Rays organizes all of geometrical optics around one characteristic function

1834

— Hamilton transplants the construction to dynamics: trajectories are rays of the surfaces of constant action

1924

— de Broglie supplies the missing wavelength,

\lambda=h/p

, and the analogy becomes physics

1926

— Schrödinger, explicitly citing Hamilton, seeks the wave equation whose eikonal limit is Hamilton-Jacobi theory — wave mechanics

Going further

References. Goldstein, Poole & Safko, Classical Mechanics, 3rd ed., §10.8 (the analogy and wave mechanics) · Landau & Lifshitz, Mechanics, 3rd ed., §47 · Born & Wolf, Principles of Optics, ch. 3 (eikonal theory) · Hamilton, Theory of Systems of Rays (1828); On a General Method in Dynamics (1834) · Schrödinger, Ann. Phys. 79, 489 (1926), second communication — the explicit tribute to Hamilton · Taylor, Classical Mechanics, ch. 13

Provenance. A verifier should reproduce: (1) the eikonal equation $|\nabla L|^2=n^2$ from the Helmholtz equation (Born-Wolf §3.1); (2) $|\nabla W|^2=2m(E-V)$ from the time-independent Hamilton-Jacobi equation (Goldstein §10.3, Landau-Lifshitz §47); (3) the phase/group relations $u=E/p$ , $v=p/m$ , $uv=E/m$ and the free-particle check $u=v/2$ ; (4) mechanical Snell's law $p_1\sin\theta_1=p_2\sin\theta_2$ from tangential continuity of $\nabla W$ . Worked numbers computed for this article and hand-verifiable: electron at 100 eV, $30.0°\to24.1°$ with $p_2/p_1=1.225$ , $\lambda_1=0.123$ nm; 300 kV microscope electron $\lambda=1.97$ pm with relativistic bracket $1.294$ . All constants CODATA.

Builds on: Hamilton-Jacobi Equation · Hamilton's Principal Function · Hamilton's Characteristic Function · Time-Independent Hamilton-Jacobi Equation Leads to: Action-Angle Variables · Separation of Variables in Hamilton-Jacobi Theory · Liouville Integrable Systems

If Hamilton had the full analogy in 1834, why did nobody propose matter waves for ninety years?

Because the analogy, taken alone, gives no reason to. It says classical mechanics has the form of a ray optics; it does not say the rays belong to an actual wave, and in 1834 nothing in experiment hinted at matter diffracting or interfering. Physics had no anomaly the wave hypothesis would cure, and no way to guess the wavelength scale. What changed by 1924: quantized atomic spectra, the photoelectric effect, and Bohr-Sommerfeld's mysterious integer conditions — phenomena that look like interference conditions ( $\oint p\,\mathrm{d}q=nh$ reads as "an integer number of wavelengths $h/p$ around the orbit"). De Broglie's postulate turned Hamilton's formal wave into a physical one with a definite $\lambda$ , and within two years Schrödinger had the equation. The analogy was a loaded spring waiting for one experimental fact.

Is the analogy exact, or approximate with corrections?

State it carefully and it is exact — as a correspondence between ray theories. The time-independent Hamilton-Jacobi equation and the optical eikonal equation are the same PDE under the dictionary; every ray-optics theorem (Snell, Fermat, caustic geometry, lens equations) has an exact mechanical twin. What is approximate is each ray theory's relation to its own wave theory: geometrical optics errs by $O(\lambda/\ell)$ against wave optics, classical mechanics by $O(\hbar/S)$ against quantum mechanics. So the accurate slogan is: exact analogy between two limits. The parallel even extends one level up — wave optics and wave mechanics correspond too, but only approximately (light obeys a second-order-in-time equation for a vector field; the Schrödinger equation is first-order in time for a complex scalar), which is why the analogy is sharpest at the ray level.

De Broglie's phase velocity comes out faster than light. Is that a problem?

No — and working out why is instructive. Including rest energy, a de Broglie wave has $\omega=E_{\mathrm{tot}}/\hbar$ and $k=p/\hbar$ , so $u=\omega/k=E_{\mathrm{tot}}/p=c^2/v$ : superluminal whenever $v<c$ . But a single infinite plane wave carries no information — it is the same everywhere forever. Anything detectable (a particle, a signal, a modulation) is a packet, and packets move at the group velocity $v_g=\mathrm{d}\omega/\mathrm{d}k=\mathrm{d}E/\mathrm{d}p=v<c$ . Phase fronts are bookkeeping surfaces, free to move at any speed, exactly like the intersection point of scissors blades. The same resolution handles the milder non-relativistic surprise in Derivation A ( $u=v/2$ ): no observable rides a single crest.

Light slows down in glass, but the particle speeds up where the index is high. Doesn't that break the analogy?

It breaks only a false expectation, and the fix locates the analogy precisely. The dictionary maps index to momentum, and rays bend toward high $n$ in optics exactly as trajectories bend toward high $p$ (low $V$ ) in mechanics — the geometry matches perfectly. Speeds are a different question: in optics high $n$ means low phase speed $c/n$ , while in mechanics high $p$ means high particle speed $p/m$ but low wavefront speed $E/p$ . So the honest statement is: particle speed corresponds to nothing simple in optics, but wavefront (phase) speed corresponds exactly — both go as $1/n$ . The analogy is wavefront-to-wavefront and ray-to-ray; it never promised particle-speed-to-energy-speed. Keeping track of which velocity is which is the entire content of the pitfall box.

Fermat's principle is least time — Maupertuis' principle is not. How can they be the same?

Because Fermat's principle, stated correctly for a dispersive medium, is stationary optical path $\int n\,\mathrm{d}s$ at fixed frequency — which equals travel time only when the phase and signal speeds coincide. The mechanical twin under $n\to p$ is stationary $\int p\,\mathrm{d}s$ at fixed energy: Maupertuis. Neither principle extremizes the particle's travel time; indeed a mechanical "least time" principle is generally false (the fast path and the stationary-action path differ). The deeper common structure: both are reduced variational principles — time (optics: frequency) has been eliminated in favor of a conserved quantity, leaving a purely geometric comparison of paths. Hamilton's principle $\delta\int L\,\mathrm{d}t=0$ , which does involve time, is a different (unreduced) principle; the relation between the two is the Legendre-transform story of this chapter's action articles.

Does the analogy extend to magnetic forces, where the force depends on velocity?

Yes, and the extension is telling: it maps onto anisotropic optics. With a vector potential $\mathbf{A}$ , the abbreviated action becomes $\int(\mathbf{p}\cdot\mathrm{d}\mathbf{q})=\int(p\,\mathrm{d}s+q\mathbf{A}\cdot\mathrm{d}\mathbf{s})$ , so the effective index depends on direction of travel, not just position — like a crystal whose refractive index differs along different axes, or a moving medium that drags light. Electron optics handles magnetic lenses exactly this way. The quantum echo is profound: that direction-dependent phase $\frac{q}{\hbar}\int\mathbf{A}\cdot\mathrm{d}\mathbf{s}$ is the Aharonov-Bohm phase, observable in electron interferometers even where $\mathbf{B}=0$ . The analogy thus quietly anticipates gauge phases — geometry entering through the action — a thread this chapter picks up again with Hannay's angle and Berry's phase.

Where exactly does the ray picture fail, and what phenomena live in the gap?

The eikonal expansion needs the wavelength short compared to every length in the problem, and it fails at three characteristic places. Caustics and foci — where rays cross and the predicted intensity diverges — get finite, structured spots (Airy patterns in optics; turning-point wavefunctions in mechanics). Shadow boundaries — where geometrics says intensity jumps discontinuously — get diffraction fringes; mechanically, this is wavepacket spreading into classically forbidden angles. And barriers thin compared to the decay length get leakage: evanescent coupling in optics, tunneling for matter. Each failure is the wave theory refusing to honor an infinitely sharp ray-geometric feature. Semiclassical physics is the art of repairing these spots locally (Airy patches, Maslov phases) while keeping rays everywhere else — the program executed in WKB Approximation and the Eikonal Connection.