Bragg-Williams Approximation

Heat a bar of β-brass — the ordered alloy of copper and zinc in which Cu and Zn sit on interpenetrating sublattices like a three-dimensional checkerboard — and near 741 K something abrupt happens: the atoms lose their preference for unlike neighbours and scramble into a random solid solution. The specific heat spikes, the superlattice X-ray reflections fade, and the material passes continuously from order to disorder. Bragg and Williams, in 1934–35, gave the first quantitative theory of this order–disorder transition, and their method turned out to be the simplest and most transportable mean-field theory of cooperative behaviour on a lattice. The Bragg–Williams move is to count configurations as if the atoms (or spins) were placed independently, discarding all spatial correlations. The energy of a configuration is written using only the average occupation — the long-range order parameter — and the entropy is the entropy of a random arrangement with that average fixed. Two ingredients, an average-occupation energy and a random-mixing entropy, are combined into a free energy that depends on a single number, and equilibrium is found by minimizing it. The result is the same self-consistency equation $m=\tanh(\beta z J m)$ and the same $k_BT_c=zJ$ that the Weiss molecular-field theory and the general mean-field decoupling produce — because Bragg–Williams, Weiss, and van der Waals are one theory in three costumes. This article owns the configuration-counting route to mean field: the combinatorial energy and entropy, the free-energy functional and its pitchfork of minima, the order-disorder transition in alloys, and the demonstration that this classical free energy underlies the Curie–Weiss law and the van der Waals loop alike. Its very simplicity is the point of departure: because it neglects correlations, it is exact only in infinite dimension or infinite range, wrongly predicts a transition in one dimension, and gives critical exponents that fail below four dimensions — which is exactly the failure that Bethe–Peierls, Landau theory, and the renormalization group were built to repair.

Key takeaways

Bragg–Williams is mean-field theory built by configuration counting: the energy uses the average occupation $E=-\tfrac12 NzJm^2$ and the entropy is the random-mixing entropy $S(m)$ with the average fixed — all spatial correlations discarded.
Minimizing the free energy $F(m)=E-TS$ gives the self-consistency $m=\tanh(\beta zJm)$ , identical to the Weiss and general mean-field results, with a continuous transition at $k_BT_c=zJ$ .
Applied to binary alloys ( $\beta$ -brass, CuZn) it describes the order–disorder transition through a long-range order parameter that vanishes as $(T_c-T)^{1/2}$ .
Bragg–Williams, Weiss molecular field, and van der Waals all coincide at the mean-field level — the same free energy, the same pitchfork, the same classical exponents $\beta=\tfrac12,\ \gamma=1,\ \delta=3,\ \alpha=0$ .
Because correlations are neglected it is exact only in $d\to\infty$ or infinite range; it gives a spurious transition in $d=1$ , an inflated $T_c$ , and the wrong exponents below the upper critical dimension $d_c=4$ .
Key idea: replace every bond energy by its average $\langle s_is_j\rangle\to m^2$ and count arrangements as if spins were independent; the resulting free energy $F(m)=-\tfrac12 NzJm^2-TS(m)$ has one order parameter, and minimizing it gives $m=\tanh(\beta zJm)$ with $k_BT_c=zJ$ — mean field obtained purely by counting.

Where this lives today

Bragg–Williams is the first mean-field theory every physicist meets and the reference point against which fluctuation effects are measured. Its direct descendants run through condensed-matter and materials science: order–disorder transitions in intermetallics and high-entropy alloys, the CALPHAD thermodynamic databases used to design engineering alloys (which build on regular-solution and sublattice models that are Bragg–Williams at heart), and the mean-field starting points for spin glasses, binary fluids, and polymer blends (Flory–Huggins theory is the Bragg–Williams entropy applied to chains). Wherever a phase diagram is computed from a free energy of mixing, the Bragg–Williams skeleton is usually inside. Its deeper role is pedagogical and structural. Because it is exact in $d\to\infty$ and analytically complete, it defines the classical critical exponents whose failure below four dimensions launched modern critical phenomena. The Ginzburg criterion quantifies when the neglected fluctuations overwhelm the mean field; the Bethe–Peierls cluster method restores short-range order and fixes the 1D pathology; Landau theory abstracts the free-energy functional; and the renormalization group finally computes the true exponents that Bragg–Williams could only approximate. Every one of these is a response to a specific Bragg–Williams shortcoming — which is why the theory, wrong as it is near criticality, is where the whole subject begins.

The physical picture

Bragg–Williams treats the lattice like a crowd whose only shared quantity is its mood — the average alignment $m$ . It never asks who your specific neighbours are; it assumes each site independently draws its state from a distribution with mean $m$ , so the chance that a bond joins two aligned spins is just $(\tfrac{1+m}{2})^2+(\tfrac{1-m}{2})^2$ , the product of single-site probabilities. This is the whole approximation in one sentence: bond correlations are the products of site averages. The energy then depends only on $m$ , and the entropy is the entropy of randomly sprinkling $N_\uparrow=\tfrac{N(1+m)}{2}$ up-spins among $N$ sites. That random-mixing entropy is generous to order. It credits the system with the full disorder-counting even in the ordered phase — it never notices that below $T_c$ the spins are actually clumped — so it overcharges the entropy of the ordered state and thereby overestimates how easily order survives heating. This is why Bragg–Williams pushes $T_c$ upward and even conjures a transition in one dimension where none exists. The payoff is a free energy $F(m)$ that is a single well above $T_c$ and a double well below. The transition is the instant the lone minimum at $m=0$ splits into two symmetric minima at $\pm m_0$ — a pitchfork bifurcation. The order parameter grows as $(T_c-T)^{1/2}$ out of the fork, and the specific heat jumps (rather than diverges) at the split.

Figure

Free energy F(m) vs order parameter m as temperature crosses T_c: a single well at m=0 for T>T_c splits into a symmetric double well with minima at ±m₀ for T<T_c — the pitchfork of Bragg-Williams mean-field theory

The same double well, read in the language of a lattice gas, is the van der Waals free energy: the substitution $s_i=2n_i-1$ turns spins into site occupations, the ordering energy into an attraction, and the non-convex region of $F(\rho)$ into the van der Waals loop whose Maxwell construction carves out liquid–gas coexistence. And the second face of the theory is the alloy itself: the up/down spins become the two atomic species, and $m$ becomes the long-range order parameter $\xi$ measuring how strongly Cu and Zn keep to their own sublattices.

Figure

A binary AB alloy on two interpenetrating sublattices (bcc, β-brass CuZn): at low T the ordered CsCl structure keeps A and B on distinct sublattices (long-range order ξ ≈ 1); at high T the atoms randomize into a disordered solid solution (ξ = 0)

The core equations

We work with the Ising ferromagnet as the archetype; the alloy and lattice-gas forms follow by relabeling. The starting point is the microscopic energy: nearest-neighbour spins want to align, and an external field $h$ biases them.

H=-J\sum_{\langle ij\rangle}s_is_j-h\sum_i s_i,\qquad s_i=\pm1 \qquad (1)

where: $J>0$ — ferromagnetic nearest-neighbour exchange; $\langle ij\rangle$ — sum over the $Nz/2$ distinct nearest-neighbour bonds; $z$ — coordination number (neighbours per site); $h$ — field in energy units. This is exact and intractable — the sum over $2^N$ configurations does not factorize because the bonds couple sites. The Bragg–Williams approximation replaces each bond's correlation by the product of site averages, $\langle s_is_j\rangle\to m^2$ , so the energy collapses onto the single order parameter $m=\langle s_i\rangle$ .

E_{\mathrm{BW}}=-\tfrac12 NzJ\,m^2-Nhm \qquad (2)

where: $m$ — magnetization per spin; the factor $\tfrac12 Nz$ counts the bonds. Read physically: every one of the $Nz/2$ bonds contributes the average energy $-Jm^2$ , and the field couples to the total moment $Nm$ . All correlation information $\langle s_is_j\rangle_c=\langle s_is_j\rangle-m^2$ has been thrown away — that discard is the approximation. The entropy is the logarithm of the number of ways to place $N_\uparrow=\tfrac{N(1+m)}{2}$ up-spins among $N$ sites, treated as independent placements.

S_{\mathrm{BW}}=-Nk_B\Big[\tfrac{1+m}{2}\ln\tfrac{1+m}{2}+\tfrac{1-m}{2}\ln\tfrac{1-m}{2}\Big] \qquad (3)

where: $\tfrac{1\pm m}{2}$ — fraction of up/down spins; the bracket is the binary (Shannon) entropy of a coin with bias $m$ . This is the random-mixing entropy: it assumes maximal disorder consistent with the average, which is exactly why it over-credits the ordered phase and inflates $T_c$ . Combining energy and entropy at fixed $m$ gives the Bragg–Williams free energy per spin — a function of one variable whose minima are the equilibrium states.

\frac{f(m)}{k_BT}=-\frac{\beta zJ}{2}m^2-\beta hm+\frac{1+m}{2}\ln\frac{1+m}{2}+\frac{1-m}{2}\ln\frac{1-m}{2} \qquad (4)

where: $f=E_{\mathrm{BW}}/N-TS_{\mathrm{BW}}/N$ ; $\beta=1/k_BT$ . The competition is transparent: the $-\tfrac12\beta zJm^2$ term rewards order, the entropy term rewards $m=0$ , and temperature tunes the balance. Minimizing $f$ over $m$ closes the theory.

m=\tanh\!\big(\beta(zJm+h)\big),\qquad k_BT_c=zJ \qquad (5)

where: $zJm+h$ — the effective Weiss field felt by one spin. The self-consistency says the magnetization must reproduce the field it generates; a nonzero solution appears the moment the low- $m$ slope $\beta zJ$ reaches $1$ , i.e. at $k_BT_c=zJ$ . Near $T_c$ this gives the classical exponents.

m_0\simeq\sqrt{3}\,\Big(\tfrac{T_c-T}{T_c}\Big)^{1/2},\qquad \chi=\frac{1}{k_B(T-T_c)},\qquad m\big|_{T_c}\propto h^{1/3} \qquad (6)

where: $m_0$ — spontaneous magnetization just below $T_c$ ; $\chi$ — zero-field susceptibility. These encode $\beta=\tfrac12$ , $\gamma=1$ , $\delta=3$ ; the specific heat jumps, so $\alpha=0$ . The full exponent set and its dimensional failure are owned by aq-sm-mean-field-critical-exponents — here we only read the leading behaviour off the free energy.

Derivation AMinimizing the Bragg-Williams free energy: the self-consistency equation and T_c

The goal is to take the configuration-counting free energy of Eq. (4), minimize it over the single order parameter $m$ , and show that the equilibrium magnetization obeys $m=\tanh(\beta(zJm+h))$ with a continuous transition at $k_BT_c=zJ$ . The strategy is direct calculus of one variable: differentiate $f(m)$ , recognize the entropy derivative as an inverse hyperbolic tangent, invert to reach the fixed-point form, and then examine the stability of the disordered $m=0$ solution by expanding for small $m$ . The one thing to watch is that the transition is signalled not by the appearance of a new root but by the $m=0$ root turning from a minimum into a maximum — the sign of the quadratic coefficient of $f$ .

Step 1. Differentiate the free energy. With $h$ restored, $\dfrac{\partial(f/k_BT)}{\partial m}=-\beta(zJm+h)+\tfrac12\ln\dfrac{1+m}{1-m}$ . Setting it to zero gives $\tfrac12\ln\dfrac{1+m}{1-m}=\beta(zJm+h)$ . Why: the energy contributes $-\beta(zJm+h)$ and the binary entropy contributes the logarithm; equilibrium is the stationary point of the free energy, not of the energy alone.

Step 2. Recognize the arctanh. The identity $\tfrac12\ln\dfrac{1+m}{1-m}=\operatorname{artanh}m$ turns the condition into $\operatorname{artanh}m=\beta(zJm+h)$ . Why: the binary-entropy derivative is exactly the inverse of the tanh that will emerge — this is what makes the mean-field self-consistency a hyperbolic-tangent fixed point rather than something transcendental and opaque.

Step 3. Invert to the fixed-point form. Taking $\tanh$ of both sides, $m=\tanh(\beta(zJm+h))$ . This is Eq. (5): the equilibrium magnetization equals $\tanh$ of the effective field it itself produces. Why: the self-consistency is the whole content of mean field — the average $m$ must be the thermal response to the field $zJm+h$ generated by that same average.

Step 4. Locate T_c by stability of m = 0. At $h=0$ , expand $\tanh(\beta zJm)=\beta zJm-\tfrac13(\beta zJm)^3+\cdots$ . A nonzero root branches off when the linear coefficient reaches unity, $\beta zJ=1$ , i.e. $k_BT_c=zJ$ . Equivalently, the $m^2$ coefficient of $f$ is $\tfrac12(k_BT-zJ)$ , which changes sign at $k_BT_c=zJ$ : above it $m=0$ is the minimum, below it $m=0$ is a maximum flanked by the two ordered minima. Why: a continuous transition is exactly the loss of stability of the symmetric state; the quadratic coefficient of the free energy vanishing is the mean-field definition of criticality. *Verified: the average-occupation energy $E=-\tfrac12 NzJm^2$ , the entropy derivative $dS/dm=-Nk_B\operatorname{artanh}m$ , the free-energy derivative giving $m=\tanh(\beta zJm)$ , and $k_BT_c=zJ$ from the vanishing $m^2$ coefficient are all checked symbolically in verify_bragg-williams-approximation.py, alongside a direct numerical fixed-point solve of $m(T)$ below $T_c$ .*

Go deeper · Derivation B · 2 minOrder-disorder transition in a binary alloy and the universality of mean-field exponents

Derivation BOrder-disorder transition in a binary alloy and the universality of mean-field exponents

Here the goal is twofold: to translate the magnetic result into the alloy language of $\beta$ -brass, obtaining the long-range order parameter and its transition temperature, and to show that Bragg–Williams, Weiss, and van der Waals share one set of critical exponents — the classical values $\beta=\tfrac12,\gamma=1,\delta=3,\alpha=0$ . The route is to write the alloy energy in terms of an ordering energy $w$ and a long-range order parameter $\xi$ , count sublattice arrangements with the same binary entropy, and minimize; then to expand the free energy near $T_c$ into the Landau form and read the exponents, noting that the identical quartic well appears in every mean-field theory. Watch the bookkeeping factor: the alloy $T_c$ carries $w/2$ rather than $J$ , purely because of how the ordering energy is defined.

Step 1. Define the ordering energy and order parameter. For an AB alloy on two interpenetrating sublattices, let $\xi$ be the long-range order: the fraction of A atoms on their correct sublattice is $\tfrac{1+\xi}{2}$ . The energy gain for ordering is set by $w=\varepsilon_{AA}+\varepsilon_{BB}-2\varepsilon_{AB}$ , and $E(\xi)=E_0-\tfrac14 Nzw\,\xi^2$ . Why: $w>0$ means unlike neighbours are favoured, so the system lowers its energy quadratically in $\xi$ — the exact analogue of $-\tfrac12 NzJm^2$ with $J\leftrightarrow w/4$ .

Step 2. Add the sublattice mixing entropy and minimize. The entropy is again the binary form $S=-Nk_B[\tfrac{1+\xi}{2}\ln\tfrac{1+\xi}{2}+\tfrac{1-\xi}{2}\ln\tfrac{1-\xi}{2}]$ , and minimizing $F=E-TS$ gives $\xi=\tanh(zw\xi/2k_BT)$ with $k_BT_c=\tfrac12 zw$ . Why: correlations between sublattice occupancies are neglected exactly as spin correlations were, so the algebra is identical up to the relabeling $zJ\to \tfrac12 zw$ .

Step 3. Expand into the Landau free energy. Near $T_c$ , using $\tfrac{1+m}{2}\ln\tfrac{1+m}{2}+\tfrac{1-m}{2}\ln\tfrac{1-m}{2}=-\ln2+\tfrac{m^2}{2}+\tfrac{m^4}{12}+\cdots$ , the per-spin free energy becomes $f=f_0+\tfrac12 a\,m^2+\tfrac14 b\,m^4$ with $a=k_BT-zJ$ and $b=\tfrac13 k_BT$ . Why: the quartic Landau well is the universal shape of every mean-field free energy near criticality — the microscopic details survive only in the coefficients $a$ and $b$ , not in the exponents.

Step 4. Read the classical exponents. Minimizing gives $m_0^2=-a/b=3(T_c-T)/T$ so $\beta=\tfrac12$ ; $\chi^{-1}=a=k_B(T-T_c)$ so $\gamma=1$ ; at $T_c$ , $h=bm^3$ so $\delta=3$ ; and the energy kink gives a finite specific-heat jump, $\alpha=0$ . The very same $a m^2+bm^4$ arises from the Weiss theory and from expanding the van der Waals equation about its critical point, so all three theories give the identical exponents. Why: universality of the mean-field exponents is not a coincidence — it is the statement that every mean-field theory reduces near $T_c$ to the same quartic free energy, whatever its microscopic origin. *Verified: the sign-change of the $m^2$ coefficient at $k_BT_c=zJ$ , the $m^4>0$ stabilizing coefficient, $m_0^2=3(T_c-T)/T$ ( $\beta=\tfrac12$ ), the critical isotherm $\beta h=m^3/3$ ( $\delta=3$ ), $\gamma=1$ , and the van der Waals inflection ( $w^2$ coefficient vanishing, $w^3\neq0$ , giving $\beta_{\mathrm{vdW}}=\tfrac12$ ) are all checked in verify_bragg-williams-approximation.py.*

Limiting cases

The approximation is exact in some limits and systematically wrong in others; knowing which is the price of admission.

Case	Result	Meaning
Infinite dimension / infinite range ( $J\to J/N$ )	Bragg–Williams becomes EXACT	Each spin couples to the genuine mean of all others; correlations vanish identically, and mean field is the exact solution.
One dimension ( $z=2$ )	predicts $k_BT_c=2J>0$	Spurious: the exact Ising chain has $T_c=0$ because domain walls (finite energy, extensive entropy) destroy order at any $T>0$ .
Near criticality, $d<4$	$\beta=\tfrac12,\ \gamma=1,\ \delta=3$	Classical exponents independent of $d$ ; true 3D Ising has $\beta\approx0.326,\ \gamma\approx1.237$ . Wrong below the upper critical dimension $d_c=4$ .
High temperature ( $k_BT\gg zJ$ )	$\chi\simeq C/(T-T_c)$	Recovers the correct Curie–Weiss paramagnetic response; the mean-field error is small far from $T_c$ .
Alloy ( $\beta$ -brass, bcc $z=8$ )	$k_BT_c=\tfrac12 zw$	Order–disorder transition with $\xi\sim(T_c-T)^{1/2}$ ; captures the $\lambda$ -shaped heat-capacity peak qualitatively.

Common trapBragg-Williams retains short-range order — it does not

The tempting misreading is that Bragg–Williams keeps some local ordering because it has an interaction energy. It does not. The energy uses only the average occupation ( $\langle s_is_j\rangle\to m^2$ ), and the entropy is the entropy of a completely random arrangement with that average fixed — spins are placed independently even deep in the ordered phase. There is literally no short-range order in the theory: the probability of a like-neighbour bond is the product of single-site probabilities, full stop. This is precisely why it invents a transition in $d=1$ , inflates every $T_c$ , and gets the exponents wrong below four dimensions. The Bethe–Peierls approximation is the minimal repair — it treats a central spin and its $z$ neighbours as a correlated cluster, restoring the short-range order that Bragg–Williams throws away, and correctly gives $T_c=0$ for the 1D chain.

Worked examples

1 · Configurational entropy and free energy at fixed order (confidence builder)

We evaluate the Bragg–Williams entropy and the energy for a definite magnetization to see the order–disorder competition numerically.

Step 1. Take $m=0.6$ (a moderately ordered state). The up/down fractions are $\tfrac{1+m}{2}=0.8$ and $\tfrac{1-m}{2}=0.2$ . The binary entropy per spin is $s/k_B=-[0.8\ln0.8+0.2\ln0.2]=-[0.8(-0.2231)+0.2(-1.6094)]=-[-0.1785-0.3219]=\mathbf{0.5004}$ .

Step 2. Compare with full disorder $m=0$ : $s/k_B=-[0.5\ln0.5+0.5\ln0.5]=\ln2=\mathbf{0.6931}$ . Ordering to $m=0.6$ has cost $0.6931-0.5004=\mathbf{0.193\,k_B}$ of entropy per spin — the entropic penalty for tilting the coin.

Step 3. The energy gain per spin (relative to $m=0$ , at $h=0$ ) is $-\tfrac12 zJm^2=-\tfrac12 zJ(0.36)=-0.18\,zJ$ . Writing the free energy per spin $f/k_BT=-\tfrac12\beta zJm^2+(s/k_B-\ln2)\cdot(-1)$ ... more directly, $f/k_BT=-0.18\,\beta zJ+(0.5004-0.6931)$ relative pieces combine to $-0.18\beta zJ+0.193$ .

Step 4. Order is favoured ( $f<f_{m=0}$ ) when $0.18\,\beta zJ>0.193$ , i.e. $k_BT<0.933\,zJ$ . So this particular $m=0.6$ state beats the disordered state below about $0.93\,T_c$ — consistent with the true equilibrium $m(T)$ crossing $0.6$ somewhat below $T_c$ , confirming the free-energy balance drives the transition.

2 · Critical temperature and the molecular field of β-brass (exam level)

We extract the ordering energy of $\beta$ -brass from its measured transition temperature and check that it is a sane bond energy.

Step 1. $\beta$ -brass (CuZn) is bcc, so the coordination number is $z=8$ . Its order–disorder transition is at $T_c\approx741$ K. The Bragg–Williams relation for the alloy is $k_BT_c=\tfrac12 zw$ , so the ordering energy is $w=\dfrac{2k_BT_c}{z}=\dfrac{2(1.381\times10^{-23})(741)}{8}$ .

Step 2. Compute: numerator $2\times1.381\times10^{-23}\times741=2.047\times10^{-20}$ J; divide by $8$ to get $w=2.56\times10^{-21}$ J. In electron-volts, $w=2.56\times10^{-21}/1.602\times10^{-19}=\mathbf{0.0160\ eV=16.0\ meV\ per\ bond}$ .

Step 3. This is a physically reasonable ordering energy — a small fraction of an eV, the scale of chemical bonding differences between CuCu/ZnZn and CuZn pairs. In the equivalent magnetic language $k_BT_c=zJ$ gives $J=k_BT_c/z=\mathbf{7.98\ meV}$ , a sane exchange scale for such a high ordering temperature.

Step 4. The predicted long-range order parameter just below $T_c$ grows as $\xi_0=\sqrt3\,\big(\tfrac{T_c-T}{T_c}\big)^{1/2}$ . At $T=0.9\,T_c$ this gives $\xi_0=\sqrt3\times\sqrt{0.1}=1.732\times0.316=\mathbf{0.548}$ . Bragg–Williams thus predicts β-brass is already about 55% ordered only 74 K below its transition — an overestimate of the true, sharper 3D-Ising rise, exactly the mean-field overshoot near $T_c$ .

Origins · 1 minWilliam Lawrence Bragg and how this was discovered

William Lawrence Bragg

1890–1971, Australian-British physicist; Nobel Prize 1915 (shared with his father, and at 25 the youngest science laureate) · order-disorder theory of alloys with E. J. Williams, 1934-35

William Lawrence Bragg shared the 1915 Nobel Prize in Physics with his father William Henry Bragg for founding X-ray crystallography — the Bragg law $n\lambda=2d\sin\theta$ that lets diffraction read atomic positions. At 25 he remains the youngest science Nobel laureate. Two decades later, with Evan James Williams, Bragg turned that structural insight onto a thermodynamic question: why do ordered alloys such as $\beta$ -brass lose their superlattice order on heating? Their 1934–35 papers introduced the long-range order parameter and the mean-field free energy that now bear both names, treating the energy through the average occupation and the entropy through random mixing. The theory was the first quantitative account of a cooperative order–disorder transition and, recognized only later, the archetype of all mean-field theories of phase transitions — the same mathematics as Weiss's ferromagnet and van der Waals's fluid. Its neglect of correlations is its defining limitation, corrected by Bethe and Peierls and ultimately by the renormalization group, but the Bragg–Williams free energy remains the indispensable zeroth-order picture from which every richer theory departs.

History

1913

— W. L. Bragg and W. H. Bragg establish X-ray crystallography and the Bragg law.

1915

— the Braggs share the Nobel Prize in Physics; W. L. Bragg is 25.

1934

— Bragg and Williams publish the order-disorder theory of alloys in Proc. Roy. Soc. A.

1935

— a second Bragg-Williams paper extends the treatment; Bethe introduces the pair (cluster) improvement.

1944

— Onsager's exact 2D Ising solution shows mean-field exponents are quantitatively wrong at criticality.

Going further

Texts. Pathria & Beale, Statistical Mechanics, 3rd ed., Ch. 12 (Ising model, mean-field/Bragg–Williams). Kardar, Statistical Physics of Fields, Ch. 2–3 (mean-field theory, saddle point). Reif, Fundamentals of Statistical and Thermal Physics, Ch. 10. Sethna, Statistical Mechanics: Entropy, Order Parameters, and Complexity, Ch. 8. Landau & Lifshitz, Statistical Physics Part 1, §142–143 (ordering of alloys).

Provenance. Base article: AllQFT track `sm`, "Bragg-Williams Approximation" (STAT.INTER.BRAGW.1256); source Pathria & Beale, Statistical Mechanics, 3rd ed., Ch. 12. All symbolic and numeric claims re-verified here in verify_bragg-williams-approximation.py (the average-occupation energy, the binary-entropy derivative $dS/dm=-Nk_B\operatorname{artanh}m$ , the minimization giving $m=\tanh(\beta zJm)$ , $k_BT_c=zJ$ from the vanishing quadratic coefficient, the Landau exponents $\beta=\tfrac12,\gamma=1,\delta=3,\alpha=0$ , the van der Waals inflection giving $\beta_{\mathrm{vdW}}=\tfrac12$ , and the $\beta$ -brass ordering energy $w\approx16$ meV/bond); no base physics error found. The base uses coordination symbol $q$ ; this article uses $z$ (identical meaning), consistent with the sibling mean-field articles.

Prerequisites. Configuration Integral and the Interacting Partition Function · Van der Waals Equation of State · Maxwell Construction and Coexistence.

Leads to. Landau Theory of Phase Transitions · Curie-Weiss Law · Bethe-Peierls Approximation.

Why does Bragg-Williams overestimate the critical temperature?

Because its entropy is too generous to the ordered phase. It counts configurations as if spins were placed independently — the fully random multinomial $N!/(N_\uparrow!N_\downarrow!)$ — even deep in the ordered state where the spins are actually clumped. This credits the ordered phase with more entropy than it really has, so the free energy of the ordered state is artificially low and order survives to higher temperatures than it should. The neglected short-range correlations, which lower the true entropy of an ordered configuration, would pull $T_c$ back down. Bethe–Peierls, which restores those correlations, gives a lower and more accurate $T_c$ .

What exactly is the difference between Bragg-Williams and the Weiss molecular field?

Almost nothing — they are the same mean-field theory reached by two routes. Weiss postulates an effective internal field $B_{\text{eff}}=B+\lambda M$ proportional to the magnetization and solves the resulting single-spin problem. Bragg–Williams builds a free energy by counting configurations (average-occupation energy plus random-mixing entropy) and minimizes it. Both yield $m=\tanh(\beta(zJm+h))$ , the same $k_BT_c=zJ$ , and the same classical exponents. Historically Weiss (1907) came from magnetism and Bragg–Williams (1934) from alloys, but the identification $\lambda\leftrightarrow zJ$ shows they are one theory. Bragg–Williams makes the entropy and the variational structure explicit, which is why it generalizes cleanly to alloys and lattice gases.

How can the same free energy describe both a magnet and a liquid-gas transition?

Through the lattice-gas mapping $s_i=2n_i-1$ , which turns each spin into a site occupation number $n_i\in\{0,1\}$ . Up-spin becomes an occupied site (liquid-like), down-spin an empty site (gas-like), the magnetization becomes the density, and the ferromagnetic exchange becomes an attractive interaction between neighbouring particles. The double-well $F(m)$ becomes a non-convex $F(\rho)$ — precisely the van der Waals loop — and the Maxwell equal-area construction that carves out magnetic coexistence becomes the one that carves out liquid–gas coexistence. The upshot is that the van der Waals equation is itself a mean-field (Bragg–Williams-class) theory, which is why it shares the classical exponents.

Why is the transition continuous (second order) rather than first order here?

Because the free energy $F(m)$ is symmetric under $m\to-m$ at zero field, so its Landau expansion has no cubic term: $f=f_0+\tfrac12 a m^2+\tfrac14 b m^4$ with $b>0$ . As $a=k_B(T-T_c)$ passes through zero, the single minimum splits smoothly into two — the order parameter grows continuously from zero as $(T_c-T)^{1/2}$ , with no jump and no latent heat. A first-order transition would require either a cubic term (broken symmetry) or $b<0$ (needing an $m^6$ term to stabilize), neither of which occurs for the symmetric Ising ferromagnet. Switching on a field $h$ does make the magnetization jump as $h$ crosses zero below $T_c$ — that is the first-order line ending at the critical point.

What does "long-range order parameter" mean for the alloy, concretely?

It measures how strongly the two atomic species keep to their own sublattices. For $\beta$ -brass on the bcc lattice, define two interpenetrating simple-cubic sublattices $\alpha$ and $\beta$ . If every Cu sits on $\alpha$ and every Zn on $\beta$ (perfect CsCl order), the order parameter $\xi=1$ ; if Cu and Zn are randomly distributed with no sublattice preference, $\xi=0$ . Formally $\xi$ is (twice) the excess probability of finding an A atom on its correct sublattice above the random value $1/2$ . It plays exactly the role of the magnetization $m$ : the same free energy, the same $\tanh$ self-consistency, and the same $(T_c-T)^{1/2}$ onset. Experimentally it is read from the intensity of superlattice X-ray reflections, which vanish as $\xi\to0$ at $T_c$ .

If mean-field exponents are wrong, why is Bragg-Williams still taught and used?

Because it is right about almost everything except the immediate neighbourhood of the critical point, and it is analytically complete. It correctly predicts that a transition exists, its approximate location, the qualitative shape of the phase diagram, the existence of a specific-heat anomaly, and the high-temperature Curie–Weiss law. It becomes exact in infinite dimensions or with infinite-range interactions, and it is the foundation on which Landau theory, the Ginzburg criterion, and ultimately the renormalization group are built — each of which is a systematic correction to a specific Bragg–Williams failure. In materials science its regular-solution and sublattice descendants are the workhorses of alloy thermodynamics. You learn it first because everything else in critical phenomena is defined relative to it.

Where precisely does the assumption of independence enter the calculation?

In two places, and both are the same assumption wearing different clothes. First, in the energy: $\langle s_is_j\rangle$ is replaced by $\langle s_i\rangle\langle s_j\rangle=m^2$ , dropping the connected correlation $\langle s_is_j\rangle_c$ entirely. Second, in the entropy: the number of configurations at fixed magnetization is taken to be the random multinomial $N!/(N_\uparrow!N_\downarrow!)$ , which counts every arrangement of the up-spins as equally likely — again treating sites as statistically independent. These are consistent: an ensemble in which sites are independent has both $\langle s_is_j\rangle=m^2$ and multinomial counting. The theory is thus internally coherent; it is simply the coherent theory of a system without correlations, which real finite-dimensional systems are not.