Ferroelectrics and Spontaneous Polarization

Every dielectric we have met so far had one thing in common: switch off the field and the polarization dies. The Clausius–Mossotti Relation ended with a road sign: when $N\alpha/3\varepsilon_0\to1$ , the feedback loop between polarization and local field no longer needs the applied field at all, and the susceptibility diverges. Local Field and the Lorentz Cavity then showed water sailing past that threshold with no drama — a warning that the catastrophe is a mean-field prediction, not a guarantee. This article is where the prediction finally comes true. In a family of soft ionic crystals — barium titanate is the flagship — the runaway really happens, and the crystal settles into a state with spontaneous polarization: a $\mathbf{P}$ that persists at zero field, points along a crystal axis, and can be flipped by an external field. The chapter's one disease becomes its flagship device. Mastering ferroelectrics buys you three things. First, the payoff of the micro–macro story: the catastrophe, tamed by anharmonicity, is the ferroelectric transition, and the Curie–Weiss law falls out of the Clausius–Mossotti denominator in two lines. Second, your first working encounter with Landau theory — order parameters, symmetry breaking, double-well free energies — the language modern physics speaks about every phase transition from magnets to the Higgs field. Third, the devices: the multilayer capacitors in every phone, FeRAM memories, and the piezo actuators that position atomic-force microscopes all run on this page's physics.

Key takeaways

A ferroelectric carries a spontaneous polarization $\mathbf{P}_s$ at zero applied field, switchable between symmetry-equivalent states — hysteretic response, no single $\chi_e$ or $\varepsilon_r$ .
It is the polarization catastrophe made real: a soft ionic lattice pushes $u=N\alpha/3\varepsilon_0$ through 1, local-field feedback sustains $\mathbf{P}$ unaided, and anharmonicity ( $bP^4$ ) picks the amplitude.
Landau theory packages this as $f=\tfrac12 a(T-T_0)P^2+\tfrac14 bP^4$ : one minimum at $P=0$ above $T_c$ ; a double well at $\pm P_s$ below, with $P_s\propto(T_c-T)^{1/2}$ .
Curie–Weiss law above the transition: $\chi=C/(T-T_0)$ — for BaTiO $_3$ $C\approx1.5\times10^5\,$ K, so $\chi\approx15{,}000$ at $T-T_0=10\,$ K — and $1/\chi$ exactly twice as steep below.
Uniform polarization would store $P_s^2/2\varepsilon_0\approx3.8\,\mathrm{GJ\,m^{-3}}$ of depolarizing-field energy; the crystal shatters into head-to-tail domains, and the loop mostly measures domain-wall motion.
The loop area $\oint E\,\mathrm{d}P$ is the heat dissipated per unit volume per cycle — the ledger entry Energy Stored in Dielectric Systems deferred here.
Key idea: a double-well free energy below $T_c$ gives two stable states $\pm P_s$ at zero field; driving the system between them with an applied field produces a hysteresis loop with remanent polarization $P_r$ and coercive field $E_c$ .

The core insight

A ferroelectric carries a switchable spontaneous polarization: a double-well free energy below $T_c$ gives two stable states $\pm P_s$ at zero field, and traversing them with an applied field produces a hysteresis loop characterized by remanent polarization $P_r$ and coercive field $E_c$ .

Where this lives today

Open any phone and you are holding this article by the hundreds. Multilayer ceramic capacitors are doped BaTiO $_3$ with the permittivity peak near $T_c$ deliberately smeared across the operating range: three hundred one-micron layers of $\varepsilon_r\approx3000$ ceramic at a square millimetre give about $8\,\mu\mathrm{F}$ in a package smaller than a grain of rice. The price is the physics above — capacitance that sags under DC bias (the loop's slope flattens away from $E=0$ ) and drifts along the Curie–Weiss curve; X7R/Y5V datasheet codes are maps of this compromise. The switchable well is a memory cell: FeRAM stores bits as $\pm P_r$ and reads them by the 520-femtocoulomb switching-charge trick of Example 3 — non-volatile, nanosecond, radiation-hard, and long a niche because perovskites integrate awkwardly with silicon. The 2011 discovery of ferroelectricity in hafnia (HfO $_2$ , already the gate oxide of every transistor) reopened the game: ferroelectric FETs and negative-capacitance devices now chase logic that remembers its state with zero standby power. Wherever motion meets electronics — AFM positioners, inkjet nozzles, ultrasound transducers — poled PZT actuators cash in the piezoelectricity that broken inversion symmetry guarantees, and passive infrared motion sensors are pyroelectric detectors reading $\mathrm{d}P_s/\mathrm{d}T$ of the order parameter you now know how to derive.

The physical picture

Picture one unit cell of barium titanate as a box with a marble in it — the Ti $^{4+}$ ion, caged by an octahedron of oxygens. Above the Curie temperature the marble's potential well has a single minimum at the centre of the cage; the cell has inversion symmetry and cannot carry a dipole moment. Cool through $T_c\approx393\,$ K and the well buckles: the bottom bulges upward, leaving two off-centre minima, and the marble must roll into one of them. Every cell in a region rolls the same way — the displacement is cooperative, not random — and each cell now carries a permanent dipole moment. That is spontaneous polarization: order chosen by the crystal, not imposed by a field.

Figure

The Landau free energy buckles at the Curie temperature — one well becomes two

Contrast this with everything the chapter has shown. The polar molecules of Permanent Dipoles and Orientational Polarization carry moments too, but they merely lean statistically toward an applied field and forget it instantly when it is removed. A ferroelectric is different because its dipoles talk to each other: each cell's dipole raises the local field on its neighbours, biasing them the same way, and beyond a critical coupling the conversation sustains itself with no external field — spontaneous symmetry breaking, precisely the feedback loop Clausius–Mossotti warned about. Two consequences dominate. First, memory: two equivalent states $\pm P_s$ separated by a barrier means the polarization depends on history. Sweep the field up and back and you trace a hysteresis loop; at $E=0$ the crystal remembers which well it sat in. Second, domains: a uniformly polarized crystal would coat itself in bound charge $\sigma_b=P_s$ and pay a ruinous depolarizing-field bill, so it shatters into head-to-tail domains whose surface charges nearly cancel. What a hysteresis loop mostly measures is not marbles hopping wells cell by cell, but domain walls sweeping through the crystal.

The equations that define the state

Before anything nonlinear happens, one anchor survives untouched. The displacement field of The Electric Displacement D was a definition, not a constitutive assumption, so it holds verbatim for a ferroelectric:

\mathbf{D}=\varepsilon_0\mathbf{E}+\mathbf{P} \qquad (1)

where: $\mathbf{D}$ — electric displacement; $\varepsilon_0=8.854\times10^{-12}\,\mathrm{F\,m^{-1}}$ ; $\mathbf{E}$ — macroscopic field; $\mathbf{P}$ — polarization density. Everything built on the definition alone — $\nabla\cdot\mathbf{D}=\rho_f$ , the bound charges $\sigma_b=\mathbf{P}\cdot\hat{\mathbf{n}}$ and $\rho_b=-\nabla\cdot\mathbf{P}$ , the boundary conditions for $D$ and $E$ — survives. What dies is the linear closure $\mathbf{P}=\varepsilon_0\chi_e\mathbf{E}$ of Linear Dielectrics, Susceptibility and Permittivity. Its replacement is state-dependent — a small wiggle riding on the spontaneous part:

\mathbf{P}=\mathbf{P}_s+\varepsilon_0\chi'\,\mathbf{E} \qquad (2)

where: $\mathbf{P}_s$ — the spontaneous polarization, nonzero at $E=0$ and set by history; $\chi'$ — the differential susceptibility, the loop's local slope at that bias. Quoted "dielectric constants" of ferroelectrics are always this slope at some bias point — the polarization is multivalued in $E$ , so no single number can describe it. The thermodynamics governing which state the crystal occupies is the Landau–Devonshire free energy, an expansion in the order parameter $P$ — even in $P$ , because odd terms are forbidden by the inversion symmetry of the high-temperature phase (the very symmetry that gets broken):

f(P,T)=f_0+\tfrac12\,a\,(T-T_0)\,P^2+\tfrac14\,b\,P^4+\tfrac16\,c\,P^6-EP \qquad (3)

where: $a,c>0$ and $T_0$ is the Curie–Weiss temperature at which the quadratic curvature changes sign; $b>0$ gives a continuous (second-order) transition, $b<0$ with $c>0$ a discontinuous (first-order) one; the $-EP$ term couples the applied field to the order parameter. Read it physically: the $P^2$ coefficient is a temperature-controlled spring constant, and the transition is nothing more than that spring constant passing through zero. The quartic term is the anharmonic stiffening that catches the runaway — without it, the catastrophe would have no finish line. Equilibrium states minimize $f$ . Setting $\partial f/\partial P=0$ gives the equation of state of a ferroelectric,

E=a(T-T_0)P+bP^3+cP^5 \qquad (4)

where: each branch with $\partial^2f/\partial P^2>0$ is a locally stable state. This one curve contains the entire phenomenology — spontaneous polarization, Curie–Weiss divergence, hysteresis, coercive field — extracted in the four derivations below.

From catastrophe to order

The chapter's micro–macro machinery predicts this transition before any thermodynamics is written down. That is where we start, because it answers the question Landau theory alone cannot: why does the spring constant change sign?

Derivation AFrom the polarization catastrophe to spontaneous order and the Curie Weiss law

We take the Clausius–Mossotti feedback seriously at the point where it breaks. The relation predicts $\chi_e$ diverging as $u=N\alpha/3\varepsilon_0\to1$ ; our goals are to show that at $u=1$ polarization can sustain itself with no applied field, to identify what selects its amplitude (the linear theory cannot), and to extract $\chi(T)$ when a soft ionic lattice carries $u$ through 1. The strategy is pure self-consistency — the same loop closed in the Clausius–Mossotti Relation, now run at $E=0$ — then a linearization of $u(T)$ near threshold. Watch the marginal case $u=1$ : it is a knife edge, and honesty about it is what forces the quartic term on us.

Step 1. Recall the closed loop of the Clausius–Mossotti Relation: each cell polarizes according to the local field, $\mathbf{P}=N\alpha\mathbf{E}_{\mathrm{loc}}$ , with $\mathbf{E}_{\mathrm{loc}}=\mathbf{E}+\mathbf{P}/3\varepsilon_0$ in a cubic environment (derived in Local Field and the Lorentz Cavity). Eliminating $\mathbf{E}_{\mathrm{loc}}$ gave $\chi_e=(N\alpha/\varepsilon_0)/(1-u)$ — the naive answer amplified by the feedback factor $1/(1-u)$ .

Step 2. Ask the forbidden question: can $\mathbf{P}\neq0$ survive at $\mathbf{E}=0$ ? Setting $\mathbf{E}=0$ in the loop gives $\mathbf{P}=u\,\mathbf{P}$ , i.e. $(1-u)\mathbf{P}=0$ . For $u<1$ the only solution is $\mathbf{P}=0$ : the medium always relaxes. At $u=1$ , any $\mathbf{P}$ solves it — the feedback exactly replaces the applied field, and polarization sustains itself. Why diverging $\chi$ and the self-sustaining state are the same fact: a diverging response means finite output for vanishing input — the $u\to1$ limit of "at $u=1$ the output needs no input at all."

Step 3. The linear theory is marginal at threshold — it permits any amplitude and stabilizes none — and for $u>1$ it predicts naked runaway: each pass around the loop multiplies $P$ by $u$ . The resolution is that $\alpha$ is only the initial slope of the ionic restoring force, which stiffens at large displacement: per unit volume the drive–restoring balance reads $a'P+bP^3=E_{\mathrm{drive}}$ with $b>0$ the anharmonic stiffness. Feedback softens only the linear coefficient, $a'\to a'(1-u)$ , leaving the cubic term untouched, so at $E=0$ and $u>1$ the medium settles at $P_s=\sqrt{a'(u-1)/b}$ — finite, nonzero, direction chosen by chance. This is exactly the Landau equation of state (4) with $a(T-T_0)\propto(1-u)$ : the microscopic loop derives the sign change of Landau's spring constant.

Step 4. Temperature enters through the polarizability. In a displacive ferroelectric the relevant $\alpha$ is ionic — a soft lattice vibration whose restoring force weakens on cooling — so $u$ rises toward 1 as $T$ falls. Linearizing near threshold, $u(T)\approx1-(T-T_0)/C_1$ , the susceptibility of Step 1 becomes

\chi_e=\frac{3u}{1-u}\approx\frac{3C_1}{T-T_0}\equiv\frac{C}{T-T_0},

the Curie–Weiss law with $C=3C_1$ (the additive $-3$ is irrelevant against $\chi\sim10^4$ ). Why this is a big deal: the empirical Curie–Weiss law is the Clausius–Mossotti denominator read as a function of temperature — the chapter's mean-field bridge promoted from describing dielectrics to predicting a phase transition. *For BaTiO $_3$ the measured Curie constant $C\approx1.5\times10^5\,$ K gives $\chi\approx15{,}000$ at $T-T_0=10\,$ K — the giant permittivities observed near the transition — and the $u>1$ settling value reproduces the Landau $P_s$ of Derivation B with $a(T-T_0)\propto(1-u)$ .*

One honesty clause, owed to Local Field and the Lorentz Cavity: $u>1$ is necessary mean-field logic, not sufficient reality. Water's orientational polarizability puts its nominal $u$ near 3.8 — four times past threshold — yet water is no ferroelectric: orientational correlations demolish the Lorentz construction, and Onsager's reaction field eats the runaway. The argument is trustworthy where the polarizability is displacive — ions shifting in a soft lattice, nothing orientational to spoil the bookkeeping — exactly the perovskite family where ferroelectricity lives.

Go deeper · Derivation B · 1 minDouble well and bistability from the Landau free energy

Derivation BDouble well and bistability from the Landau free energy

Thermodynamics now takes over from microscopics: we extract the equilibrium states from the Landau free energy. The goals are to show that the sign change of the quadratic coefficient at $T_0$ converts one minimum into two, to find the spontaneous polarization and its temperature exponent, and to check stability honestly via the curvature. Take the second-order case ( $b>0$ , $c=0$ , $E=0$ ) and minimize $f(P)=\tfrac12 a(T-T_0)P^2+\tfrac14 bP^4$ . The payoff to watch: the well depth — the energy barrier protecting a stored bit.

Step 1. Stationary points. Differentiate and factor:

\frac{\partial f}{\partial P}=a(T-T_0)P+bP^3=P\big[a(T-T_0)+bP^2\big]=0.

Roots: $P=0$ always; and $P^2=a(T_0-T)/b$ , which is real only for $T<T_0$ .

Step 2. Above $T_0$ the bracket is positive for all real $P$ , so $P=0$ is the only stationary point, and its curvature $f''(0)=a(T-T_0)>0$ confirms a minimum: the paraelectric phase, marble at the centre of the cage.

Step 3. Below $T_0$ the curvature at the origin flips: $f''(0)=a(T-T_0)<0$ , so $P=0$ becomes a local maximum — the barrier — and the two off-centre roots take over as minima:

P_s(T)=\pm\sqrt{\frac{a(T_0-T)}{b}}\propto(T_0-T)^{1/2},

the mean-field order parameter with critical exponent $\beta=\tfrac12$ . The signs are exactly degenerate; the crystal chooses, and inversion symmetry is spontaneously broken. Why the exponent matters: square-root growth is fast near the transition and flat far below — $P_s$ doubles from $T_c-5\,$ K to $T_c-20\,$ K — which is why device-grade $P_s$ values are quoted at room temperature.

Step 4. Depth of the wells. Substitute $P_s^2=a(T_0-T)/b$ back into $f$ :

f(P_s)=-\frac{a^2(T_0-T)^2}{4b}.

This negative condensation energy is the barrier (measured from the well bottom to the $P=0$ hump) that a switching pulse must overcome and that thermal fluctuations must not — the physical basis of non-volatile retention in FeRAM. The curvature at the minima, $f''(P_s)=2a(T_0-T)>0$ , confirms their stability, and this factor of 2 is exactly the factor-of-two susceptibility signature derived next.

Go deeper · Derivation C · 1 minCurie Weiss on both sides and the factor of two

Derivation CCurie Weiss on both sides and the factor of two

The equation of state contains a sharp, checkable prediction: the susceptibility diverges on both sides of a second-order transition, but $1/\chi$ versus $T$ is exactly twice as steep below. We derive both branches from one line of algebra — differentiating the equation of state at the equilibrium polarization of each phase — and flag the V-shaped $1/\chi$ plot as the standard experimental test of the Landau picture. Keep the second-order case ( $b>0$ , $c=0$ ); the subtlety is where you evaluate the derivative: $P=0$ above, $P=P_s$ below.

Step 1. The inverse susceptibility is the slope of the equation of state $E=a(T-T_0)P+bP^3$ :

\frac{1}{\varepsilon_0\chi}=\frac{\partial E}{\partial P}=a(T-T_0)+3bP^2,

evaluated at the equilibrium $P$ of the phase in question.

Step 2. Paraelectric side: $P=0$ , so $1/(\varepsilon_0\chi)=a(T-T_0)$ , i.e. $\chi=C/(T-T_0)$ with $C=1/(\varepsilon_0 a)$ — the same Curie–Weiss law Derivation A produced microscopically, now with the Curie constant tied to the Landau coefficient.

Step 3. Ferroelectric side: substitute $P^2=P_s^2=a(T_0-T)/b$ ,

\frac{1}{\varepsilon_0\chi}=a(T-T_0)+3b\cdot\frac{a(T_0-T)}{b}=2a(T_0-T).

The $3bP_s^2$ term does not merely correct the curvature — it flips its sign and doubles its magnitude.

Step 4. The signature: $1/\chi$ against $T$ is a V hitting zero at $T_0$ with slope ratio exactly $-2$ . For BaTiO $_3$ 's $C=1.5\times10^5\,$ K: $\chi\approx15{,}000$ at $T_0+10\,$ K, $\approx7{,}500$ at $T_0-10\,$ K. Why experimentalists love this: absolute χ depends on sample and electrodes, but a slope ratio is dimensionless and calibration-free — one plot either shows the mean-field V or it does not. *The below- $T_c$ curvature $2a(T_0-T)$ equals $f''(P_s)$ from Derivation B, as it must — susceptibility is inverse curvature.*

Figure

The V-plot: inverse susceptibility hits zero at T0 with a slope ratio of exactly two

One refinement the V-plot already hints at: BaTiO $_3$ 's transition is weakly first order. With $b<0$ and $c>0$ in equation (3), the transition happens at $T_c=T_0+3b^2/16ac$ , about 12 K above the $T_0$ where the paraelectric well would lose stability; $P_s$ jumps in discontinuously (to $\sqrt{-3b/4c}$ ), with latent heat and a few kelvin of thermal hysteresis. The extrapolated $1/\chi$ line still aims at $T_0$ — the divergence is simply preempted at $T_c$ . And cubic-to-tetragonal at 393 K is only the first step of a cascade: further cooling tilts the polar axis twice more (tetragonal→orthorhombic near 278 K, orthorhombic→rhombohedral near 183 K).

Switching, hysteresis, and the loop

Below $T_c$ the equation of state $E=a(T-T_0)P+bP^3$ is an S-shaped curve: for $|E|$ small it has three solutions — two stable branches (the wells) flanking an unstable crest (the barrier). Sweeping the field, the crystal rides its current branch until that branch ends, then jumps. The jump field is the intrinsic coercive field $E_c$ ; the closed trajectory is the hysteresis loop.

Figure

The hysteresis loop traced on the Landau equation of state with switching jumps at the coercive field

Read the loop like a story. Start saturated at large positive $E$ : a single domain on the upper branch. Reduce $E$ to zero — the polarization barely sags, arriving at the remanent value $P_r$ (equal to $P_s$ in the single-domain model; somewhat below it in real, imperfectly aligned crystals). Push $E$ negative: the state persists, metastable, until the branch ends at $-E_c$ and the polarization crashes to the lower branch — the crystal has switched. The return sweep mirrors it. The trajectory is never retraced: the loop encloses area, and by the bookkeeping of Energy Stored in Dielectric Systems — which deferred exactly this ledger entry to us — the source's work $\oint E\,\mathrm{d}D\approx\oint E\,\mathrm{d}P$ per cycle per unit volume is that area, delivered irreversibly as heat. One honesty flag: the Landau jump describes coherent switching, every cell hopping the barrier at once, and it grossly overestimates real coercive fields. Measured $E_c$ in good BaTiO $_3$ is of order $10^5\,\mathrm{V\,m^{-1}}$ — two orders of magnitude or more below the intrinsic branch-end field — because real switching nucleates small reversed domains at electrodes and defects whose walls then sweep cheaply through the crystal. Hysteresis in practice is domain-wall traffic, which is why $E_c$ depends on thickness, defects, frequency, and fatigue history: a kinetic quantity wearing thermodynamic clothing. Why domains at all? Because uniform polarization is ruinously expensive, as promised in the physical picture — and the estimate deserves to be done properly.

Go deeper · Derivation D · 2 minWhy a ferroelectric shatters into domains

Derivation DWhy a ferroelectric shatters into domains

We estimate the electrostatic cost of staying single-domain and show that splitting into oppositely polarized stripes lowers it, at the price of domain-wall energy — the competition that sets the equilibrium domain size. Consider a plate of thickness $L$ , polarized normal to its faces at $P_s$ , in open circuit so nothing screens $\sigma_b=\pm P_s$ . The tools are the slab bookkeeping of Bound Charges: Surface and Volume and the field-energy accounting of Energy Stored in Dielectric Systems; watch the sheer size of the single-domain energy density.

Step 1. A uniformly polarized ideal slab carries $\sigma_b=\pm P_s$ on its faces — a parallel-plate configuration with interior depolarizing field $E_{\mathrm{dep}}=-P_s/\varepsilon_0$ opposing the polarization (the slab case $L=1$ of the depolarization factors catalogued in Local Field and the Lorentz Cavity).

Step 2. The stored energy density is enormous. With $P_s=0.26\,\mathrm{C\,m^{-2}}$ (BaTiO $_3$ , room temperature),

u_{\mathrm{dep}}=\frac{\varepsilon_0E_{\mathrm{dep}}^2}{2}=\frac{P_s^2}{2\varepsilon_0}\approx3.8\times10^{9}\,\mathrm{J\,m^{-3}},

with $E_{\mathrm{dep}}\approx2.9\times10^{10}\,\mathrm{V\,m^{-1}}$ — a field a thousand times any breakdown strength, an energy density comparable to a pressure of about 38,000 atmospheres. No crystal will pay this bill. Why the estimate is an upper bound: the huge depolarizing field would itself suppress P long before this energy is reached — which is precisely the point: uniform P at full strength cannot coexist with open-circuit boundaries.

Step 3. Split into stripes of alternating $\pm P_s$ with width $w\ll L$ . Viewed from farther than $\sim w$ the surface looks charge-neutral: the field no longer fills the slab, only skin layers of thickness $\sim w$ at the faces. The electrostatic energy per unit plate area falls from $\sim u_{\mathrm{dep}}L$ to $\sim u_{\mathrm{dep}}\,\alpha w$ , with $\alpha$ a geometry factor of order one.

Step 4. Walls are not free: stripes of width $w$ need $L/w$ walls per unit length at surface energy $\sigma_w$ each, costing $\sim\sigma_w L/w$ per unit plate area. Minimizing $\;\alpha\,u_{\mathrm{dep}}\,w+\sigma_w L/w\;$ over $w$ gives

w^\ast=\sqrt{\frac{\sigma_w L}{\alpha\,u_{\mathrm{dep}}}}\;\propto\;\sqrt{L},

the Kittel square-root law: thicker crystals carry proportionally wider (but relatively finer-grained) domains. Why electrodes change the picture: shorted metal faces screen σ_b with free charge, removing the depolarizing penalty — poled, electroded devices can stay single-domain while free-standing crystals cannot. *The values $u_{\mathrm{dep}}$ and $E_{\mathrm{dep}}$ follow directly from $P_s=0.26\,\mathrm{C\,m^{-2}}$ and $\varepsilon_0$ , and the $w\propto\sqrt{L}$ scaling is the standard Kittel domain-width result.*

Limiting cases

Case	Result	Meaning
$T\gg T_c$	$\chi=\dfrac{C}{T-T_0}$ , small	Deep paraelectric: an ordinary (if generous) linear dielectric; equation (2) reduces to Linear Dielectrics with $P_s=0$ .
$T\to T_0^+$ (second order)	$\chi\to\infty$ , $1/\chi$ slope $a\varepsilon_0$	Curie–Weiss divergence: the soft lattice pushes $u=N\alpha/3\varepsilon_0\to1$ and the Clausius–Mossotti denominator dies.
$T<T_c$ , $b>0$	$P_s=\sqrt{a(T_c-T)/b}$ , $1/\chi$ slope doubled	Continuous onset, $\beta=\tfrac12$ ; the V-plot with slope ratio $-2$ is the calibration-free mean-field test.
First order ( $b<0$ , $c>0$ )	$T_c=T_0+\dfrac{3b^2}{16ac}$ , $P_s(T_c)=\sqrt{-3b/4c}\neq0$	BaTiO $_3$ 's case: discontinuous jump, latent heat, thermal hysteresis; the $1/\chi$ extrapolation aims at $T_0$ , about 12 K below $T_c$ .
$u>1$ but orientational (water)	no transition	Necessary-not-sufficient honesty: the Lorentz construction fails for correlated polar liquids (Onsager's repair) — displacive lattices, not polar fluids, realize the catastrophe.
$E\gg E_c$	single domain, $\partial P/\partial E\to$ small ionic slope	Saturation: the loop closes onto the stable branch; the differential $\chi'$ collapses to the ordinary lattice response.

Common trapThere is no such thing as THE dielectric constant of a ferroelectric

Below $T_c$ the response is nonlinear, multivalued, and history-dependent: quoting a single $\varepsilon_r$ for BaTiO $_3$ is like quoting a single slope for the letter S. Any tabulated value is a small-signal slope $\chi'$ at one bias, temperature, and frequency — above $10^4$ near $T_c$ , a few hundred along the polar axis at room temperature. Two neighbouring traps: a merely polar material is not ferroelectric (water has dipoles and $u>1$ , yet no spontaneous order — cooperativity and switchability are the tests), and the intrinsic Landau coercive field is not a design number — real switching nucleates domain walls, and measured $E_c$ runs an order of magnitude or more below the coherent estimate.

Worked examples

Example 1 — estimate the spontaneous polarization of BaTiO $_3$ from its ionic displacements. In the tetragonal phase (cell $a=3.99\,$ Å, $c=4.04\,$ Å, volume $V=6.43\times10^{-29}\,\mathrm{m^3}$ ), neutron diffraction gives the classic displacements relative to the Ba frame: Ti $^{4+}$ shifts $+0.06\,$ Å along the polar axis, the apical O $^{2-}$ shifts $-0.09\,$ Å, the two equatorial O $^{2-}$ shift $-0.06\,$ Å each.

Step 1. Sum the point-charge dipole per cell, $p=\sum_iq_i\delta_i$ : Ti gives $(+4e)(+0.06\,\text{Å})=+0.24\,e$ Å; the apical oxygen $(-2e)(-0.09\,\text{Å})=+0.18\,e$ Å; the equatorial pair $2\times(-2e)(-0.06\,\text{Å})=+0.24\,e$ Å. Total $p=0.66\,e$ Å $=1.06\times10^{-29}\,\mathrm{C\,m}$ — every term adds: cations shift one way, anions the other.

Step 2. Divide by the cell volume: $P=p/V=1.06\times10^{-29}/6.43\times10^{-29}\approx0.16\,\mathrm{C\,m^{-2}}$ .

Step 3. Measured: $P_s\approx0.26\,\mathrm{C\,m^{-2}}$ . The point-charge estimate lands the right order but 40% low — expected, because ions are not rigid point charges: covalent Ti–O charge transfer makes the dynamical (Born) effective charge of Ti closer to $+7$ than the formal $+4$ . The punchline stands either way: sub-ångström displacements produce a quarter coulomb per square metre — a tiny structural distortion with an enormous electrical voice.

Example 2 — Curie–Weiss numbers and the factor of two. BaTiO $_3$ has Curie constant $C\approx1.5\times10^5\,$ K and $T_0\approx381\,$ K. Step 1. Ten kelvin above $T_0$ : $\chi=C/(T-T_0)=1.5\times10^5/10=15{,}000$ — four orders of magnitude beyond vacuum, which is why industry parks doped BaTiO $_3$ ceramics near (but safely above) their transitions. Step 2. A hundred kelvin above: $\chi=1{,}500$ , falling as $1/(T-T_0)$ along the blue arm of the V-plot. Step 3. Ten kelvin below $T_0$ (second-order idealization), the doubled slope gives $\chi=C/[2(T_0-T)]=7{,}500$ : same distance, half the response. Step 4. The Landau coefficient underneath: $a=1/(\varepsilon_0C)=7.5\times10^{5}\,\mathrm{V\,m\,C^{-1}\,K^{-1}}$ — the number you would feed equation (3).

Example 3 — the loop as a ledger: dissipation and the FeRAM bit. Take a BaTiO $_3$ -class crystal, $P_r\approx0.26\,\mathrm{C\,m^{-2}}$ , measured $E_c\approx1\times10^5\,\mathrm{V\,m^{-1}}$ .

Step 1. A full saturating cycle encloses, on the rectangular estimate, $W\approx4E_cP_r=4\times10^5\times0.26\approx1.0\times10^5\,\mathrm{J\,m^{-3}}$ — the heat bill Energy Stored in Dielectric Systems deferred to this article.

Step 2. Cycle it at 1 kHz and the volumetric heating is $10^8\,\mathrm{W\,m^{-3}}$ : a millimetre cube would draw a tenth of a watt just to keep flipping. Moral: nobody runs a ferroelectric around its full loop continuously — memories switch in isolated nanosecond pulses; capacitors sit in the small-signal regime far from $E_c$ .

Step 3. The same loop read as a memory: a $1\,\mu\mathrm{m}^2$ FeRAM cell switching from $-P_r$ to $+P_r$ releases $Q=2P_rA=0.52\times10^{-12}\,\mathrm{C}\approx520\,\mathrm{fC}$ onto its bit line — comfortably above sense-amplifier thresholds. A cell that does not switch releases only the small linear response; the difference between "flipped" and "did not flip" is the stored bit.

Origins · 1 minPierre Curie and how this was discovered

Pierre Curie

1859–1906 · Paris — symmetry as a physical law

Before the radium years made the family name synonymous with radioactivity, Pierre Curie was the physicist of crystal symmetry. In 1880, aged 21, he and his brother Jacques discovered piezoelectricity — squeeze quartz and charge appears — by reasoning from crystal classes to which effects symmetry permits, then building electrometers delicate enough to catch them. His doctoral work on magnetism gave the $\chi\propto1/T$ paramagnetic law its name, and "Curie temperature" now marks the death of spontaneous order in magnets and their electrical cousins alike. His deeper legacy is the principle bearing his name — the symmetries of causes must reappear in their effects. Spontaneous polarization is its textbook illustration: a cubic, inversion-symmetric crystal can show no polar effect until, below $T_c$ , the symmetry itself breaks and the forbidden becomes mandatory.

History

1880

— Jacques and Pierre Curie discover piezoelectricity in quartz, tourmaline, and Rochelle salt

1895

— Pierre Curie's thesis establishes the paramagnetic law χ ∝ 1/T, the template Curie–Weiss behaviour

1920

— Joseph Valasek demonstrates switchable polarization and hysteresis in Rochelle salt: ferroelectricity discovered

1945

— Wul, Goldman, and von Hippel's groups establish barium titanate as a robust perovskite ferroelectric

1949

— Devonshire's Landau-type free-energy theory organizes the BaTiO3 phase cascade

1960

— Cochran and Anderson recast the transition as a softening lattice mode, unifying microscopics and thermodynamics

2011

— Ferroelectricity found in doped hafnium oxide films, making the effect CMOS-compatible at the nanoscale

Going further

References. Griffiths, Introduction to Electrodynamics, 4th ed., §4.4 (linear dielectrics; ferroelectrics as nonlinear example) · Jackson, Classical Electrodynamics, 3rd ed., §4.6 (models of dielectrics and ferroelectricity) · Purcell & Morin, Electricity and Magnetism, 3rd ed., Ch. 10 (electric fields in matter) · Zangwill, Modern Electrodynamics, §6.5 (dielectric matter, ferroelectrics) · Kittel, Introduction to Solid State Physics, 8th ed., Ch. 16 (ferroelectric crystals, Landau theory) · Lines & Glass, Principles and Applications of Ferroelectrics and Related Materials (Oxford, 1977) The key results follow from the Landau–Devonshire free energy $f(P)=\tfrac12 a(T-T_0)P^2+\tfrac14 bP^4+\tfrac16 cP^6-EP$ : the double-well minima $P_s=\sqrt{a(T_0-T)/b}$ , the Curie–Weiss law $\chi=1/[\varepsilon_0 a(T-T_0)]$ , and the factor-of-2 slope ratio of $1/\chi$ across $T_c$ . The anchor numbers for BaTiO $_3$ are $T_c\approx393\,\mathrm{K}$ (120 °C), $P_s\approx0.26\,\mathrm{C\,m^{-2}}$ , and $\varepsilon_r$ up to $\sim10^4$ near $T_c$ . The relations $\mathbf{D}=\varepsilon_0\mathbf{E}+\mathbf{P}$ and $\sigma_b=\mathbf{P}\cdot\hat{\mathbf{n}}$ are exact definitions. The first-order shift $T_c=T_0+3b^2/(16ac)$ is the standard result for $b<0,\,c>0$ .

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

If the polarization catastrophe predicts a transition for water, why does barium titanate order while water does not?

The catastrophe argument is only as good as the Lorentz local field behind it. In water the polarizability is orientational — whole molecules rotating, strongly correlated with their neighbours — so the cubic-cancellation bookkeeping fails; Onsager's analysis (see Local Field and the Lorentz Cavity) shows most of the naive feedback is a dipole interacting with its own polarization cloud, which aligns nothing. In BaTiO $_3$ the polarizability is displacive — ions shifting in a softening lattice mode, no rotations to correlate — and there the mean-field loop is honest: $u\to1$ really marks the instability. Same formula, two verdicts; the difference is whether the construction's assumptions hold.

Is the spontaneous polarization of a real crystal at zero field actually equal to P_s from the double well?

Per domain, essentially yes; per crystal, usually no. Each domain sits in a well at $\pm P_s$ , but an unpoled crystal contains many domains whose contributions largely cancel — net $P$ near zero even though every unit cell is fully polarized. Poling (cooling through $T_c$ under field, or saturating the loop) aligns the domains; the remanent $P_r$ measured at $E=0$ approaches, but rarely equals, the single-domain $P_s$ , since some domains back-switch to relieve local stresses and depolarizing fields.

Why is the measured coercive field so much smaller than the field where the Landau branch ends?

The Landau estimate assumes every unit cell hops the barrier simultaneously — coherent switching — which maximizes the cost. Real crystals cheat: a tiny reversed nucleus forms at an electrode, defect, or existing wall, then grows, because moving a wall sideways costs far less field than flipping a region at once. Switching is nucleation-and-growth kinetics, so measured $E_c$ depends on thickness, temperature, pulse duration, defects, and fatigue history. Treat quoted coercive fields as sample properties, not material constants; only the thermodynamic quantities ( $P_s$ , $T_c$ , $C$ ) transfer cleanly between samples.

What exactly is soft about the soft mode?

A transverse optical lattice vibration in which the Ti ion and the oxygen cage oscillate against each other — precisely the displacement pattern that, frozen, becomes the ferroelectric distortion. Its restoring force weakens on cooling (that is what raises the ionic $\alpha$ and drives $u\to1$ ), so its frequency falls — softens — with $\omega_s^2\propto(T-T_0)$ mirroring the Landau coefficient, and at the transition the mode's displacement freezes in as the static order parameter. Cochran and Anderson's 1960 insight: Curie–Weiss dielectric data and neutron measurements of $\omega_s(T)$ are the same physics seen by two instruments.

Ferroelectric, piezoelectric, pyroelectric — how do the categories nest?

By crystal symmetry, strictly. Of the 32 crystal classes, 21 lack inversion symmetry; 20 of those are piezoelectric (stress induces polarization). Ten of the 20 are polar — a unique axis carries a built-in polarization, making them pyroelectric ( $P_s$ changes with temperature). Ferroelectrics are the subset of pyroelectrics whose $P_s$ can be reversed by an accessible field. Every ferroelectric is therefore pyroelectric and piezoelectric, but not conversely: quartz is piezoelectric yet not pyroelectric; tourmaline pyroelectric yet not switchable. Switchability is an operational test, which is why ferroelectricity is demonstrated with a hysteresis loop, not deduced from a structure table.

How is a ferroelectric different from an electret, which also holds polarization at zero field?

An electret (see the taxonomy of Polarization Density P) is frozen disequilibrium: charges or dipole alignments locked in by cooling a waxy or polymeric material under field — metastable, decaying over months to years, not reversibly switchable; reheat it and the order dies. A ferroelectric's $P_s$ is a thermodynamic equilibrium property of the crystal phase: it regenerates below $T_c$ , switches reproducibly between symmetry-equivalent states, and shows a genuine phase transition with Curie–Weiss behaviour. Same zero-field polarization on a datasheet; completely different physics underneath.

Does the ferroelectric analogy with ferromagnetism actually hold up?

As thermodynamics, remarkably well — which is why the borrowed name stuck despite there being no iron in barium titanate: order parameter, Curie temperature, Curie–Weiss susceptibility, domains, hysteresis, Landau description; the $P$ – $E$ and $M$ – $H$ loops are near-identical in shape. Microscopically it breaks: ferromagnetism is quantum exchange between electron spins, displacive ferroelectricity is classical ionic displacement in a soft lattice, and dipole–dipole coupling — negligible for room-temperature magnetism — is the essential feedback here. The magnetic side is told properly in the Magnetic Fields in Matter tier.

Ferroelectrics and Spontaneous Polarization

Where this lives today

The physical picture

The equations that define the state

From catastrophe to order

Switching, hysteresis, and the loop

Limiting cases

Worked examples

Going further

Before this

Where this leads