Local Field and the Lorentz Cavity

The previous article wrote a cheque it could not cash. The Clausius–Mossotti Relation took one physical input beyond $\mathbf{p}=\alpha\mathbf{E}_{\mathrm{loc}}$ — the claim that a molecule deep inside a polarized medium feels not the macroscopic field $\mathbf{E}$ but $\mathbf{E}_{\mathrm{loc}}=\mathbf{E}+\mathbf{P}/3\varepsilon_0$ — and deferred its derivation here. This article pays that debt, and audits the fine print. The problem is real and quantitative. The macroscopic $\mathbf{E}$ of Maxwell's equations is a spatial average over thousands of molecules; the field three ångströms from a water molecule swings by billions of volts per meter as you pass its neighbours. Which field does a molecule actually respond to? Neither the raw microscopic chaos (most of it averages away) nor the plain average (that smears out a systematic piece that does not cancel). Lorentz's construction threads the needle: draw — do not carve — a small sphere around the molecule, treat everything outside as smooth continuum, everything inside as discrete neighbours, and count. The continuum's ledger closes at $+\mathbf{P}/3\varepsilon_0$ , the exact mirror of the $-\mathbf{P}/3\varepsilon_0$ interior field we found in Uniformly Polarized Sphere; the discrete neighbours, in any cubic or statistically isotropic arrangement, sum to exactly zero. Mastering the construction buys you three things: the microscopic justification of Clausius–Mossotti, an honest map of where that relation must fail — water is the star witness, and we run its autopsy with real numbers — and the vocabulary (cavity field, reaction field, depolarization factor) that dielectric physics, polar-liquid theory, and ferroelectric design all speak.

Key takeaways

A molecule polarizes in response to the local field $\mathbf{E}_{\mathrm{loc}}$ — the field of everything except itself at its own site — not the macroscopic average $\mathbf{E}$ .
Lorentz's bookkeeping: draw a virtual sphere around the molecule; outside = continuum, inside = discrete neighbours. Three contributions: $\mathbf{E}$ , the sphere-removal (cavity) term, and the near sum.
The cavity term is $+\mathbf{P}/3\varepsilon_0$ — the exact complement of the $-\mathbf{P}/3\varepsilon_0$ inside a uniformly polarized sphere: hole plus sphere rebuilds intact medium.
The discrete near sum vanishes identically for cubic lattices and statistically isotropic fluids; a 5% tetragonal stretch already revives it at the 14% level — the $1/3$ is symmetry, not magic.
Result: $\mathbf{E}_{\mathrm{loc}}=\mathbf{E}+\mathbf{P}/3\varepsilon_0$ , the engine of the Clausius–Mossotti relation; for other bookkeeping shapes $\mathbf{E}_{\mathrm{loc}}=\mathbf{E}+L\mathbf{P}/\varepsilon_0$ with $L\in[0,1]$ .
Polar liquids break the construction — orientational correlations spoil the near-field cancellation; for water the naive theory predicts a catastrophe that does not happen, and Onsager's reaction field is the repair.
Key idea: draw a fictitious sphere around the molecule — the continuum outside contributes $\mathbf{E}+\mathbf{P}/3\varepsilon_0$ , the discrete dipoles inside sum to zero for cubic or isotropic media, so $\mathbf{E}_{\mathrm{loc}}=\mathbf{E}+\mathbf{P}/3\varepsilon_0$ .

The core insight

Surround the molecule with a fictitious Lorentz sphere; the continuum outside contributes the macroscopic field plus the cavity field $+\mathbf{P}/3\varepsilon_0$ , while the discrete dipoles inside sum to zero for cubic or isotropic media — giving $\mathbf{E}_{\mathrm{loc}}=\mathbf{E}+\mathbf{P}/3\varepsilon_0$ .

Where this lives today

The Lorentz factor never left the lab bench. Whenever an optical measurement on a dense medium must be translated into a molecular property, the conversion carries the local-field factor $(n^2+2)/3$ — hyperpolarizability measurements, Raman cross sections, and solvatochromic Stark-shift probes all quote it explicitly, and getting the factor wrong is a classic source of order-unity disagreement between groups. Modern polarizable force fields for molecular simulation (AMOEBA and its successors) are, in essence, machines that refuse the Lorentz shortcut: they compute $\mathbf{E}_{\mathrm{loc}}$ at every molecule explicitly, self-consistently, every femtosecond timestep — a brute-force honest version of this article, which is also how one computes Kirkwood's $g_K$ from first principles today. The failure modes are equally alive. Dielectric and terahertz spectroscopy of water, alcohols, and electrolytes is analyzed in Onsager–Kirkwood language; and the polarization catastrophe that water so pointedly refuses is exactly the instability that perovskite ferroelectrics embrace — the next article, Ferroelectrics and Spontaneous Polarization, turns this page's disease into a device.

Origins · 1 minLars Onsager and how this was discovered

Lars Onsager

1903–1976 · Yale University, Norwegian-American

Onsager met the local-field problem as a young chemist and, in his quietly devastating way, located the exact flaw in a construction that had stood for half a century: Lorentz's cavity lets a molecule's own reaction field — the polarization it induces in its surroundings — feed back as if it could help align the molecule, when that field is always parallel to the dipole and can exert no torque. His 1936 paper "Electric Moments of Molecules in Liquids" split the cavity field from the reaction field and tamed polar dielectrics at a stroke; famously, Debye — whose own theory it corrected — resisted the paper for years. Onsager had form as an uninvited corrector: as a 22-year-old he had opened a conversation with Debye by telling him his electrolyte theory was wrong. The reciprocal relations he derived in 1931 won him the 1968 Nobel Prize in Chemistry, and his exact solution of the two-dimensional Ising model (1944) is one of the great calculations of the century — three separate revolutions, each executed with the same taste for finding the one term everyone else had mis-counted.

History

1850

— Mossotti relates dielectric constant to the response of conducting spheres in an insulating matrix — the local-field idea in embryo

1878

— Lorentz derives the P over three-epsilon-zero correction and shows the near-field sum vanishes for cubic symmetry

1880

— Lorenz in Copenhagen publishes the refractive-index form independently; the optical relation becomes Lorentz–Lorenz

1912

— Debye grafts permanent dipoles onto the Lorentz field, predicting a polarization catastrophe for polar liquids

1936

— Onsager separates cavity field from reaction field; the spurious catastrophe for water disappears

1939

— Kirkwood introduces the correlation factor g, connecting hydrogen-bond structure to the dielectric constant

The physical picture

Start with how violent the microscopic landscape actually is. A water molecule carries a dipole moment $p=6.17\times10^{-30}\,\mathrm{C\,m}$ , and its nearest neighbour sits about $3.1\,\text{Å}$ away. On the dipole's axis at that distance the field is $2p/4\pi\varepsilon_0 a^3\approx3.7\times10^{9}\,\mathrm{V/m}$ — four orders of magnitude above a strong laboratory field. The macroscopic $\mathbf{E}$ we have used since Polarization Density P is deliberately blind to this: it is the spatial average over a volume holding many molecules, and averaging is precisely what makes the smooth field theory of the previous nine articles legal. But now we need the one question averaging cannot answer: how hard is one specific molecule being pushed? The molecule does not feel the blur. It feels the applied sources plus the sharp fields of every other dipole — and emphatically not its own field, for the same reason you cannot lift yourself by your own collar: a molecule's polarizability $\alpha$ is defined as its response to fields imposed on it from outside itself. Lorentz's move is to split the census by distance, using a sphere that is drawn, not carved. Far molecules (outside the sphere) are legion and individually weak: replace them by the smooth continuum they average into. Near molecules (inside) are few and violent: keep them discrete. The far-continuum piece must then be corrected, because the macroscopic average had already counted the inside region as continuum too — so we remove a sphere's worth of smooth $\mathbf{P}$ (that removal contributes the cavity term) and add the discrete neighbours back by hand (the near sum). Two of the three pieces turn out to be old friends, and the third, remarkably, is often zero.

Figure

The Lorentz construction — a drawn sphere splits the census into continuum outside, discrete neighbours inside

One caution before the mathematics, because it is the most conflated point in the chapter: this sphere is imaginary. No material is removed; the dielectric is intact everywhere. Dielectric Sphere in a Uniform Field solved the other problem — a real, carved cavity with matched boundary conditions — and got a different answer, $3\varepsilon_r\mathbf{E}_0/(2\varepsilon_r+1)$ . Keep the knife-versus-pencil distinction; we will enforce it again in the warning below.

Drawing the sphere: the decomposition

The local field is defined as the microscopic field at the molecule's site produced by everything except the molecule itself, averaged over configurations. The Lorentz partition organizes it as

\mathbf{E}_{\mathrm{loc}}=\mathbf{E}_{\mathrm{out}}+\mathbf{E}_{\mathrm{cav}}+\mathbf{E}_{\mathrm{near}} \qquad (1)

where: $\mathbf{E}_{\mathrm{out}}$ — the macroscopic field $\mathbf{E}$ itself, i.e. the average over the whole medium including the material inside the drawn sphere — which is precisely why the next term has to un-count that material; $\mathbf{E}_{\mathrm{cav}}$ — the correction for the continuum we must un-count inside the sphere, since the macroscopic average had included it; $\mathbf{E}_{\mathrm{near}}$ — the honest discrete sum over the actual neighbours inside. The molecule's own field appears in none of the three pieces. The next three derivations evaluate the second term, kill the third, and then ask what happens when the drawn surface is not a sphere.

Derivation AThe cavity term equals P over three epsilon-zero

We are deriving the field contributed by removing a sphere's worth of uniform polarization from the medium — the middle term of the Lorentz census. The strategy is direct: the removed region's absence is equivalent to a shell of bound charge on the drawn surface, $\sigma_b=\mathbf{P}\cdot\hat{\mathbf{n}}$ with the normal pointing into the hole, and we integrate that shell's field at the centre. Watch the two signs — the inward normal, and the patch field pointing from patch to centre — because they conspire to make the answer parallel to $\mathbf{P}$ , not opposed to it.

Step 1. Removing the smooth polarization inside the drawn sphere of radius $a$ is, by superposition, the same as adding a sphere of polarization $-\mathbf{P}$ . Its surface carries bound charge $\sigma_b=(-\mathbf{P})\cdot\hat{\mathbf{r}}=-P\cos\theta$ , with $\theta$ measured from the $\mathbf{P}$ direction: negative charge crowns the top of the hole, positive lines the bottom — exactly what the hero figure shows. Why superposition and not surgery: nothing is physically removed. We are subtracting, inside the sphere, the continuum contribution the macroscopic average had already counted, and a subtraction is an addition of the negative.

Step 2. A wall patch of area $\mathrm{d}A$ at angle $\theta$ produces at the centre a field of magnitude $\sigma_b\,\mathrm{d}A/4\pi\varepsilon_0a^2$ directed along $-\hat{\mathbf{r}}$ (from patch toward centre for positive $\sigma_b$ ). Only the component along $\mathbf{P}$ survives the azimuthal integral, picking up the projection $(-\hat{\mathbf{r}})\cdot\hat{\mathbf{z}}=-\cos\theta$ :

E_{\mathrm{cav}}=\frac{1}{4\pi\varepsilon_0}\int\frac{(-P\cos\theta)(-\cos\theta)}{a^2}\,\mathrm{d}A

Step 3. With $\mathrm{d}A=a^2\sin\theta\,\mathrm{d}\theta\,\mathrm{d}\phi$ the radius cancels — the first sign that the sphere's size is irrelevant — and the angular integral is elementary:

E_{\mathrm{cav}}=\frac{P}{4\pi\varepsilon_0}\,2\pi\int_0^\pi\cos^2\theta\,\sin\theta\,\mathrm{d}\theta=\frac{P}{4\pi\varepsilon_0}\cdot2\pi\cdot\frac{2}{3}=\frac{P}{3\varepsilon_0}

Step 4. Restoring the direction: the cavity term is $\mathbf{E}_{\mathrm{cav}}=+\mathbf{P}/3\varepsilon_0$ , parallel to the polarization. This is precisely the negative of the $-\mathbf{P}/3\varepsilon_0$ interior field of Uniformly Polarized Sphere — as it must be: a uniformly polarized sphere dropped into the hole restores intact uniform medium, so the two contributions cancel identically. Note: the hole-plus-sphere sum vanishes identically — $+\mathbf{P}/3\varepsilon_0$ from the cavity plus $-\mathbf{P}/3\varepsilon_0$ from the sphere equals zero, as it must for a drawn line to leave no seam — and the angular integral $\int_0^\pi\cos^2\theta\sin\theta\,\mathrm{d}\theta=2/3$ underpins the result throughout.

The cavity term did not depend on the sphere's radius. That scale-independence is what licenses the whole trick: any radius large enough for the outside to look like continuum and small enough for $\mathbf{P}$ to be locally uniform gives the same answer, so the arbitrary line we drew leaves no fingerprint on the physics.

The near sum: zero, by symmetry — with an honest counterexample

The remaining piece is the discrete sum over neighbours inside the drawn sphere,

\mathbf{E}_{\mathrm{near}}=\frac{1}{4\pi\varepsilon_0}\sum_i\frac{3(\mathbf{p}_i\cdot\hat{\mathbf{r}}_i)\hat{\mathbf{r}}_i-\mathbf{p}_i}{r_i^3} \qquad (2)

where: $\mathbf{p}_i$ — the dipole moment of neighbour $i$ (all equal to $p\hat{\mathbf{z}}$ in a uniformly polarized medium); $\mathbf{r}_i$ — its position relative to our molecule. Each term is the point-dipole field we derived back in Electric Dipole Moment, evaluated at the centre. This is the piece that cannot be handled by continuum arguments — and the piece on which the whole construction's honesty rests.

Go deeper · Derivation B · 2 minWhy the discrete neighbours contribute nothing in a cubic environment

Derivation BWhy the discrete neighbours contribute nothing in a cubic environment

We are proving Lorentz's key cancellation: $\mathbf{E}_{\mathrm{near}}=0$ when the neighbours occupy a cubic lattice — and, by configuration averaging, an isotropic fluid. The strategy is to reduce the vector sum to one scalar lattice sum and kill it with the symmetry that exchanges the $x$ , $y$ , and $z$ axes. Watch which symmetry actually does the work: it is cubic symmetry (axis interchange), a far stronger demand than mere periodicity — and we will exhibit a periodic lattice that fails it.

Step 1. With all $\mathbf{p}_i=p\hat{\mathbf{z}}$ , the $z$ -component of equation (2) reads

E_{\mathrm{near},z}=\frac{p}{4\pi\varepsilon_0}\sum_i\frac{3z_i^2-r_i^2}{r_i^5}

and the transverse components collect $\sum_i 3x_iz_i/r_i^5$ and $\sum_i 3y_iz_i/r_i^5$ .

Step 2. The transverse sums vanish by parity alone: a cubic lattice maps onto itself under $x_i\to-x_i$ , which flips each $x_iz_i$ term while preserving the lattice, so the sum equals its own negative. Why parity is not enough for the z-component: $3z_i^2-r_i^2$ is even in every coordinate, so no reflection kills it — we need the axes to be interchangeable.

Step 3. Cubic symmetry means the lattice looks identical down the $x$ , $y$ , and $z$ axes, so the three diagonal sums are equal shell by shell:

\sum_i\frac{x_i^2}{r_i^5}=\sum_i\frac{y_i^2}{r_i^5}=\sum_i\frac{z_i^2}{r_i^5}=\frac{1}{3}\sum_i\frac{r_i^2}{r_i^5}

Substituting, $\sum_i(3z_i^2-r_i^2)/r_i^5=(3\cdot\tfrac13-1)\sum_i r_i^{-3}=0$ . Every complete shell cancels exactly, not asymptotically.

Step 4. See it concretely on the six nearest neighbours (distance $a$ , take $a=1$ ): the two axial sites ( $z=\pm1$ ) each contribute $(3-1)/1=+2$ ; the four equatorial sites ( $z=0$ ) each contribute $(0-1)/1=-1$ . Ledger: $2(+2)+4(-1)=0$ . The next shell repeats the trick with twelve sites, and so on outward — the figure below draws this shell.

Step 5. For a liquid there is no lattice, but the ensemble average of the neighbour positions around any molecule is spherically symmetric, and the same $\tfrac13$ argument applied to the average kills $\langle\mathbf{E}_{\mathrm{near}}\rangle$ . The fine print — "provided the neighbours' orientations are uncorrelated with their positions" — is exactly the clause polar liquids violate, as the worked example will show. Summing $(3z_i^2-r_i^2)/r_i^5$ over all simple-cubic sites within balls of radius up to $20a$ (33,400 sites) gives zero to machine precision ( $<10^{-15}$ ) at every radius. The honesty counterexample: stretch the lattice tetragonally by just 5% ( $c/a=1.05$ ) and the sum jumps to $S=-0.55/a^3$ , an $\mathbf{E}_{\mathrm{near}}$ correction of about $-14\%$ of the Lorentz term itself — the cancellation is cubic symmetry at work, not a generic small effect.

Figure

Shell-by-shell cancellation — the six nearest neighbours pay a balanced ledger

With the middle term evaluated and the near sum dead, the census closes:

\mathbf{E}_{\mathrm{loc}}=\mathbf{E}+\frac{\mathbf{P}}{3\varepsilon_0} \qquad (3)

where: $\mathbf{E}$ — the macroscopic field at that point in the medium (applied sources plus all bound charge, computed by the continuum methods of this chapter); $\mathbf{P}$ — the local polarization density. Read it physically: in a dense polarized medium the molecule is pushed harder than the average field suggests, by exactly one third of $P/\varepsilon_0$ . Feeding equation (3) into $\mathbf{p}=\alpha\mathbf{E}_{\mathrm{loc}}$ and $\mathbf{P}=N\mathbf{p}$ is the two-line route to the Clausius–Mossotti Relation — done carefully in that article, spent from here on as settled.

Go deeper · Derivation C · 2 minThe shape of the drawn surface — needle, sphere, slab

Derivation CThe shape of the drawn surface — needle, sphere, slab

We are deriving what happens to the local-field correction when the bookkeeping surface is not a sphere — both to expose where the $1/3$ comes from and because real crystals sometimes force a different choice. The strategy: rerun the cavity-term logic for the two extreme shapes, a long needle aligned with $\mathbf{P}$ and a thin slab perpendicular to it, using the boundary-condition machinery we already own rather than fresh integrals. The point to watch is which face carries wall charge — that alone decides the answer.

Step 1. Needle along $\mathbf{P}$ : the removed region is a thin cylinder whose curved wall has $\hat{\mathbf{n}}\perp\mathbf{P}$ , so $\sigma_b=\mathbf{P}\cdot\hat{\mathbf{n}}=0$ there; wall charge sits only on the two far-away end caps, which subtend vanishing solid angle from the centre. Hence $\mathbf{E}_{\mathrm{cav}}\to0$ and

\mathbf{E}_{\mathrm{loc}}^{\text{needle}}=\mathbf{E}

Why this matches the boundary conditions: along the needle we are sampling the field through a path where only the continuous tangential component matters — the same $E_\parallel$ continuity rule of Boundary Conditions for D and E.

Step 2. Slab perpendicular to $\mathbf{P}$ : the removed region is a thin pancake whose two broad faces carry the full $\sigma_b=\mp P$ (negative sheet above, positive below, for the removed $-\mathbf{P}$ bookkeeping). Between two such sheets the field is uniform, $\sigma/\varepsilon_0$ , so

\mathbf{E}_{\mathrm{loc}}^{\text{slab}}=\mathbf{E}+\frac{\mathbf{P}}{\varepsilon_0}

Why this too is a boundary-condition statement: across the faces it is $D_\perp$ that is continuous, and adding $P/\varepsilon_0$ to $E$ is exactly what keeps $\varepsilon_0E+P$ unchanged.

Step 3. The sphere interpolates with the factor computed in the first derivation. All three are members of one family,

\mathbf{E}_{\mathrm{loc}}=\mathbf{E}+\frac{L}{\varepsilon_0}\mathbf{P},\qquad L_{\text{needle}}=0,\quad L_{\text{sphere}}=\tfrac13,\quad L_{\text{slab}}=1

with $L$ the same depolarization factor that governed the sphere family in Uniformly Polarized Sphere and Dielectric Sphere in a Uniform Field.

Step 4. Which $L$ is right? For a molecule in a cubic or isotropic environment, the sphere — because only the sphere respects the environment's symmetry, and the near sum then vanishes to complete the argument. In anisotropic crystals the honest procedure is the one our tetragonal counterexample forced: keep the spherical surface, and let the non-vanishing $\mathbf{E}_{\mathrm{near}}$ act as an anisotropic correction, direction by direction — which is how the local-field tensors of crystal optics arise (Anisotropic Dielectrics and the Permittivity Tensor picks this up). The needle and slab limits give depolarization factors $L=0$ and $L=1$ respectively, and the family satisfies the standard sum rule $\mathrm{tr}\,L=1$ across the three axes — $0+0+1$ for the slab's two transverse axes and its normal, and $\tfrac13\times3$ for the sphere.

Limiting cases

Case	Result	Meaning
Dilute gas, $N\alpha/\varepsilon_0\ll1$	$\mathbf{E}_{\mathrm{loc}}\approx\mathbf{E}$	The correction is second order in density; naive $\chi_e=N\alpha/\varepsilon_0$ is fine — which is why gas-phase measurements read $\alpha$ directly.
Dense nonpolar medium	$\mathbf{E}_{\mathrm{loc}}=\mathbf{E}+\mathbf{P}/3\varepsilon_0$	Full Lorentz correction; drives the Clausius–Mossotti relation and its density-invariant molar polarization.
Cubic lattice	$\mathbf{E}_{\mathrm{near}}=0$ exactly	Shell-by-shell symmetry cancellation — not an approximation.
5% tetragonal stretch	$\mathbf{E}_{\mathrm{near}}\approx-0.14\,(\mathbf{P}/3\varepsilon_0)$	The cancellation is fragile: mild anisotropy revives the near sum; local-field corrections become tensors.
Polar liquid (water)	Lorentz fails	Orientational correlations tie neighbour alignment to position; Onsager's reaction field replaces the naive $+\mathbf{P}/3\varepsilon_0$ for permanent dipoles.
Needle / slab bookkeeping	$L=0$ / $L=1$	The $1/3$ is the sphere's depolarization factor, not a law of nature.

Common trapA drawn sphere is not a carved cavity — and E-local never enters Maxwell's equations

Two fatal conflations live on this page. First: the Lorentz sphere is drawn with a pencil in intact material to organize a sum; its "cavity field" $+\mathbf{P}/3\varepsilon_0$ is a bookkeeping term. A carved cavity — material physically removed, boundary conditions matched — is a different problem with a different answer, $\mathbf{E}_{\mathrm{cav}}=3\varepsilon_r\mathbf{E}_0/(2\varepsilon_r+1)$ , solved in Dielectric Sphere in a Uniform Field. More than one textbook paragraph, and many an exam answer, silently swaps them. Second: $\mathbf{E}_{\mathrm{loc}}$ is the input to the molecular law $\mathbf{p}=\alpha\mathbf{E}_{\mathrm{loc}}$ and nothing else. Gauss's law $\nabla\cdot\mathbf{D}=\rho_f$ , the boundary conditions, the energy integrals — all take the macroscopic $\mathbf{E}$ and $\mathbf{D}$ . Put $\mathbf{E}_{\mathrm{loc}}$ into a field equation and you have double-counted the very dipoles the construction so carefully separated.

Worked examples

Example 1 — how big is the correction? (confidence builder). A nonpolar dielectric has $\varepsilon_r=2.0$ and sits in a macroscopic field $E=1.0\,\mathrm{MV/m}$ . Find the local field and the error of the naive (no-correction) susceptibility.

Step 1. The polarization is $P=\varepsilon_0\chi_eE$ with $\chi_e=\varepsilon_r-1=1.0$ , so $P=8.854\times10^{-12}\times1.0\times10^{6}=8.85\,\mathrm{\mu C/m^2}$ .

Step 2. The Lorentz correction is $P/3\varepsilon_0=\chi_eE/3=0.333\,\mathrm{MV/m}$ , so $E_{\mathrm{loc}}=1.333\,\mathrm{MV/m}$ — one third larger than the average field. Equivalently $E_{\mathrm{loc}}/E=1+\chi_e/3=(\varepsilon_r+2)/3=1.333$ , a form worth memorizing.

Step 3. Invert Clausius–Mossotti for the molecular side: $N\alpha/\varepsilon_0=3(\varepsilon_r-1)/(\varepsilon_r+2)=0.75$ . A naive theory that equates $\chi_e$ with $N\alpha/\varepsilon_0$ would predict $\chi_e=0.75$ against the true $1.0$ — a 25% underestimate already at $\varepsilon_r=2$ , from ignoring a field the molecule demonstrably feels.

Example 2 — the water autopsy (exam level). Water at $300\,\mathrm{K}$ : permanent moment $p=6.17\times10^{-30}\,\mathrm{C\,m}$ , density $997\,\mathrm{kg/m^3}$ , molar mass $18.0\,\mathrm{g/mol}$ , measured $\varepsilon_r=80$ . Run the Lorentz construction to its conclusion and diagnose the wreckage.

Step 1. Orientational polarizability, from the Langevin result of Permanent Dipoles and Orientational Polarization: $\alpha_{\mathrm{or}}=p^2/3k_BT=(6.17\times10^{-30})^2/(3\times1.381\times10^{-23}\times300)=3.06\times10^{-39}\,\mathrm{C\,m^2/V}$ .

Step 2. Number density: $N=\rho N_A/M=997\times6.022\times10^{23}/0.0180=3.33\times10^{28}\,\mathrm{m^{-3}}$ . The catastrophe parameter is $u=N\alpha_{\mathrm{or}}/3\varepsilon_0=3.33\times10^{28}\times3.06\times10^{-39}/(3\times8.854\times10^{-12})=3.84$ .

Step 3. Clausius–Mossotti demands $(\varepsilon_r-1)/(\varepsilon_r+2)=u$ , but the left side cannot exceed 1 for any $\varepsilon_r$ : at $u=3.84$ there is no physical solution — the formal inversion $\varepsilon_r=(1+2u)/(1-u)=-3.1$ is nonsense. Lorentz-fed water sits nearly four times past the ferroelectric threshold $u=1$ (it would stay past threshold up to $\sim1150\,\mathrm{K}$ at this density), yet real water is not a ferroelectric: it sits at $\varepsilon_r=80$ , which corresponds to $u_{\mathrm{needed}}=79/82=0.963$ — a factor 4.0 below what the construction supplies.

Step 4. The failed assumption is Step 5 of the cancellation derivation: water's neighbours are hydrogen-bonded, so their orientations correlate with their positions, and $\mathbf{E}_{\mathrm{near}}$ is anything but zero. Onsager's 1936 repair separates the field in a molecule-sized cavity into a cavity field (what the surroundings apply) and a reaction field (the surroundings' response to the molecule's own $\mathbf{p}$ , which is parallel to $\mathbf{p}$ and therefore cannot help align it against thermal noise — the term Lorentz's construction wrongly lets feed back). Running Onsager's equation with $n^2=1.77$ for water's electronic part gives $\varepsilon_r\approx28$ — divergence gone, magnitude still low. The remaining factor $\approx2.8$ is Kirkwood's correlation factor $g_K$ : hydrogen bonding makes neighbouring dipoles co-operate, and measured $g_K\approx2.5$ – $2.9$ closes the books on 80. Verdict: Lorentz impossible, Onsager sane, Onsager+Kirkwood correct.

Going further

References. Griffiths, Introduction to Electrodynamics, 4th ed., §4.4.3 (the local field) and Problem on Clausius–Mossotti · Jackson, Classical Electrodynamics, 3rd ed., §4.5 (Molecular Polarizability and Electric Susceptibility) · Purcell & Morin, Electricity and Magnetism, 3rd ed., Ch. 10 · Zangwill, Modern Electrodynamics, §6.5 (The Local Field and Clausius–Mossotti) · Ashcroft & Mermin, Solid State Physics, Ch. 27 (dielectric properties) Key results to keep in mind: the cavity-field angular integral $\int_0^\pi\cos^2\theta\sin\theta\,d\theta=2/3$ yields $E=P/3\varepsilon_0$ ; the Clausius–Mossotti form follows from the algebraic reduction $(\varepsilon_r-1)/(\varepsilon_r+2)=N\alpha/3\varepsilon_0$ ; the cubic-lattice near-field cancellation of $\mathbf{E}_{\mathrm{near}}$ holds shell by shell; and the numeric check gives $\varepsilon_r=2\Rightarrow N\alpha/\varepsilon_0=0.75$ . Throughout, $\varepsilon_0=8.854\times10^{-12}\,\mathrm{F/m}$ (CODATA).

Builds on: Polarization Density P · Field of a Polarized Object · Uniformly Polarized Sphere · Induced Dipoles and Atomic Polarizability · Bound Charges: Surface and Volume Leads to: Linear Dielectrics, Susceptibility and Permittivity · Permanent Dipoles and Orientational Polarization · The Electric Displacement D

If the near-sum cancellation needs a cubic lattice, why does Clausius–Mossotti work so well for liquids and gases, which have no lattice at all?

Because the argument survives averaging. In a fluid, any given snapshot has a nonzero near sum, but a molecule's average environment is spherically symmetric, and the same one-third symmetry argument applied to the ensemble average kills $\langle\mathbf{E}_{\mathrm{near}}\rangle$ . What the fluid must supply is statistical isotropy plus one quieter condition: the neighbours' orientations must not correlate with their positions. Nonpolar molecules (argon, CS $_2$ , benzene) satisfy this beautifully, which is why the density test in the Clausius–Mossotti article works from vapor to liquid. Hydrogen-bonded and polar liquids violate it — a water molecule's neighbours point where the hydrogen bonds tell them to — and that is precisely where the relation dies.

Why doesn't the molecule's own field count? It is by far the biggest field at its own location.

Polarizability is defined as the response to fields imposed from outside the molecule: $\alpha$ already encodes how the electron cloud arranges itself under its own internal forces. Counting the self-field would double-book physics that is inside $\alpha$ by construction. There is also a mechanical way to see it: a charge distribution exerts zero net force on itself (internal forces cancel by Newton's third law), so the self-field cannot push the molecule's charge centroid anywhere. The subtlety worth remembering is the reaction field — the molecule's field polarizes the surroundings, and their response acts back on the molecule. That is not self-field, it is a real external field; Lorentz's construction mishandles it, and Onsager's whole contribution was to book it separately.

Does the radius of the Lorentz sphere matter? I could have drawn it twice as big.

No — and that arbitrariness is the best evidence the construction is legal. The cavity term came out radius-independent (the $a^2$ cancelled in the wall integral), and the near sum vanishes for every complete shell, so enlarging the sphere just adds more zeros. The only requirements are mesoscopic: big enough that the outside genuinely averages into a continuum (many molecules), small enough that $\mathbf{P}$ is uniform across it. Any radius in that window gives the same $\mathbf{E}_{\mathrm{loc}}$ . If your answer depends on the drawn radius, you have made an error — usually by treating the drawn sphere as a real cavity with boundary conditions.

Why is the cavity term plus P over 3 epsilon-zero when the field inside a polarized sphere is minus P over 3 epsilon-zero?

They are the two halves of one identity. The macroscopic average inside the drawn sphere had counted a sphere's worth of continuum polarization; to replace it by discrete neighbours we must first subtract it, i.e. superpose a sphere of $-\mathbf{P}$ . The interior field of a uniformly polarized sphere is $-\mathbf{P}/3\varepsilon_0$ (as derived in Uniformly Polarized Sphere), so the subtracted sphere contributes $+\mathbf{P}/3\varepsilon_0$ . Check: hole plus sphere is intact medium, and $+\mathbf{P}/3\varepsilon_0-\mathbf{P}/3\varepsilon_0=0$ , as it must — the drawn line leaves no seam.

Is the local field measurable, or is it pure bookkeeping?

It is measurable, though always through a molecular messenger. Vibrational Stark-effect probes (a C=O or C≡N group whose frequency shifts linearly with the field it sits in) read the local field at their own site directly, and solvatochromic dyes do the coarser version; both consistently report fields larger than the macroscopic $E$ , in line with the $(\varepsilon_r+2)/3$ enhancement. Nonlinear-optical determinations of molecular hyperpolarizabilities must divide out local-field factors $(n^2+2)/3$ per photon before different solvents agree on one molecular number — the agreement after correction is itself a measurement of the Lorentz factor. What no experiment can do is put $\mathbf{E}_{\mathrm{loc}}$ into Maxwell's equations; it remains the molecule's private field.

What exactly is Onsager's reaction field, in one breath?

Place a molecule with dipole $\mathbf{p}$ in a real molecule-sized cavity. Two fields exist there: the cavity field — what the polarized surroundings transmit in from the applied field — and the reaction field — the surroundings' response to $\mathbf{p}$ itself, which points along $\mathbf{p}$ . Lorentz's single $+\mathbf{P}/3\varepsilon_0$ silently merges them, letting the reaction field masquerade as an aligning field. But a field parallel to the dipole exerts no torque: it cannot fight thermal disorder, only stretch the moment slightly. Onsager's 1936 bookkeeping feeds only the cavity field to the Langevin alignment average and books the reaction field separately — the spurious catastrophe vanishes, and water comes out at $\varepsilon_r\approx28$ instead of impossible (Kirkwood's correlation factor supplies the rest of 80).

Is the polarization catastrophe at N alpha equals 3 epsilon-zero ever real, or always an artifact?

Both, instructively. For polar liquids it is an artifact — water "predicts" $u=3.8$ and simply is not ferroelectric; the construction, not nature, failed. But for displacive solids, where the relevant $\alpha$ is ionic (ions shifting off-centre in a soft lattice mode, no orientational correlations to spoil the counting), the feedback is real physics: as the lattice softens, $N\alpha/3\varepsilon_0\to1$ , the susceptibility diverges, and the crystal locks into a spontaneously polarized state. That is barium titanate's story, and the mean-field Curie–Weiss law falls straight out of the diverging denominator. Ferroelectrics and Spontaneous Polarization runs that argument to completion — this chapter's one disease becoming its flagship device.

When would I ever use the needle or slab factors instead of one third?

Whenever the environment around the polarizable unit is genuinely anisotropic. Thin films and layered dielectrics see slab-like screening along the normal; chain polymers and needle-like inclusions see the $L\approx0$ limit along the chain; ellipsoidal grains in composites take intermediate $(L_1,L_2,L_3)$ with $L_1+L_2+L_3=1$ . In anisotropic crystals the honest route keeps the spherical surface and restores the non-vanishing near sum as a directional correction — our 5% tetragonal stretch shifting the correction by 14% is the toy version — which is one microscopic origin of the permittivity tensor closing this chapter (Anisotropic Dielectrics and the Permittivity Tensor).