Self-Consistency Equation and Spontaneous Symmetry Breaking

Here is a small paradox that sits at the heart of every phase transition. Write down the energy of a magnet with no external field and it is perfectly symmetric: flip every spin and nothing changes. Yet a real magnet below its Curie temperature picks a definite direction and points that way. A symmetric equation has produced an asymmetric answer. How? The resolution lives inside a single transcendental equation. Mean-field theory replaces the fluctuating neighbours of one spin by their thermal average, so each spin sees an effective field $Jzm$ set by the very magnetization $m$ it helps produce. Closing that loop — demanding that the spin's response reproduce the assumed average — gives the self-consistency equation $m=\tanh[\beta(Jzm+h)]$ . This article is about that equation as an object in its own right: a nonlinear fixed-point relation whose solution set changes character as temperature is lowered. The mean-field Hamiltonian that produces it is derived by the companion article aq-sm-mean-field-theory; here we take the equation as given and study what it does. What it does is undergo a bifurcation. Above $T_c$ the only solution is $m=0$ and it is stable. At $k_BT_c=Jz$ the trivial solution goes unstable and two symmetry-related solutions $\pm m$ branch off continuously — a supercritical pitchfork, the dynamical-systems fingerprint of spontaneous symmetry breaking. The Hamiltonian keeps its $\mathbb{Z}_2$ symmetry; the equilibrium state does not. Understanding how an infinitesimal field, a boundary, or the order of the thermodynamic limit selects one branch — and why fluctuations can no longer restore the symmetry once the system is large — is the conceptual core of ordering, from ferromagnets to superconductors to the Higgs field. The same fixed-point structure reappears in the BCS gap equation and the van der Waals loop; mastering it once unlocks all of them.

Key takeaways

Mean-field theory closes into one nonlinear fixed-point equation, $m=\tanh[\beta(Jzm+h)]$ , relating the order parameter to itself through the effective field it generates.
As a fixed point it has a stability criterion: $m^*$ is stable iff the slope $\beta Jz\,\mathrm{sech}^2(\beta Jzm^*)<1$ ; the trivial root $m=0$ is stable above $T_c$ and unstable below.
At $k_BT_c=Jz$ the equation undergoes a supercritical pitchfork bifurcation — $m=0$ loses stability and two symmetric ordered branches $\pm m$ are born.
Near $T_c$ the order parameter grows as $m\simeq\sqrt{3}\,(-t)^{1/2}$ , the classical exponent $\beta=1/2$ , independent of dimension.
The free energy $f(m)$ is even in $m$ at $h=0$ , with two degenerate minima below $T_c$ : the state must choose one, breaking the $\mathbb{Z}_2$ symmetry the energy retains.
This is spontaneous symmetry breaking — real only in the thermodynamic limit, where the barrier between the wells diverges and an infinitesimal $h\to0^+$ selects a branch.
Key idea: the symmetric self-consistency equation acquires a second pair of solutions at $T_c$ ; the physical state occupies one degenerate free-energy minimum, so an asymmetric magnet emerges from a symmetric law — spontaneous symmetry breaking as a pitchfork bifurcation.

The core insight

The symmetric self-consistency equation $m=\tanh(\beta Jzm)$ acquires a second pair of solutions at $k_BT_c=Jz$ through a supercritical pitchfork; below $T_c$ the free energy has two degenerate minima $\pm m$ , and the physical state occupies one of them — an asymmetric magnet emerging from a symmetric law, the essence of spontaneous symmetry breaking.

Where this lives today

The self-consistency-plus-symmetry-breaking template is one of the most reused ideas in physics. In superconductivity the BCS gap equation $\Delta=g\sum_k\frac{\Delta}{2E_k}\tanh(\beta E_k/2)$ has the identical bifurcation structure — $\Delta=0$ above $T_c$ , a continuously growing gap below, with $\Delta\sim(T_c-T)^{1/2}$ — and the broken symmetry is the $U(1)$ phase of the electron pair field. The Higgs mechanism of particle physics is the same double-well, promoted to a quantum field: the vacuum sits in one minimum of a symmetric potential, giving mass to the $W$ and $Z$ bosons and to the fermions. Cosmological phase transitions in the early universe, the nematic ordering of liquid crystals, and the buckling of a compressed rod are all pitchfork bifurcations of an order parameter. The theory's honest limitation is also its most productive legacy. Because it discards fluctuations, mean-field theory gives the same exponents $\beta=1/2,\gamma=1,\delta=3$ in every dimension — quantitatively wrong below the upper critical dimension $d_u=4$ , where real exponents differ (in 3D, $\beta\approx0.33$ ). Understanding precisely when fluctuations overwhelm the mean-field description — the Ginzburg criterion (aq-sm-ginzburg-criterion) — is the doorway to the renormalization group, which finally computes the true exponents. So the pitchfork studied here is both the first correct picture of ordering and the launch point for everything that corrects it.

The physical picture

Picture one spin in a lattice, but do not track its neighbours individually. In mean-field theory it feels only their average alignment, which acts like an extra field $Jzm$ pushing it to align. The spin responds — partially aligning — and its alignment contributes to the very average that produced the field. Equilibrium is the fixed point where the response exactly regenerates the assumed average. The whole drama of ordering is the drama of this loop closing on a nonzero value. The cleanest way to see the transition is graphical. Plot the line $y=m$ and the curve $y=\tanh(\beta Jz\,m)$ on the same axes. The curve always passes through the origin with initial slope $\beta Jz$ . When $T>T_c$ that slope is less than $1$ : the tanh starts below the line and the only crossing is at $m=0$ . As $T$ falls through $T_c=Jz/k_B$ the slope passes through $1$ ; for $T<T_c$ the tanh now leaves the origin faster than the line, rises above it, and — because it saturates at $1$ — must cross back down through the line at some $\pm m^*$ . Two ordered solutions appear, symmetrically, out of nothing.

Figure

Graphical solution of m = tanh(βzJm): for T > T_c the tanh (slope < 1 at origin) meets the line y = m only at 0; for T < T_c the steeper tanh crosses again at ±m*, the spontaneous magnetization

Now the symmetry-breaking itself. Flipping every spin ( $m\to-m$ at $h=0$ ) leaves the energy unchanged, so the two ordered solutions come as a mirror pair with exactly equal free energy. The system cannot lower its energy by preferring one; it must simply choose. The choice is made by an infinitesimal external field, by the boundary conditions, or by the order in which the thermodynamic limit is taken. Once made, it is locked: flipping the entire macroscopic magnetization would have to pass through the free-energy maximum at $m=0$ , and that barrier grows with system size, so a large magnet stays in its well essentially forever. The free-energy landscape below is the whole story in one image: a symmetric double well whose ball has rolled into one side.

Figure

The free energy f(m): a single well above T_c collapses to a symmetric double well below T_c; the state occupies one minimum (broken symmetry), and a small field h tilts the wells to select it

There is one subtlety worth flagging early, because it trips almost everyone: the self-consistency equation always has the root $m=0$ . Finding it does not mean there is no transition. Below $T_c$ that root is a free-energy maximum — a valid stationary point but an unstable one. You must always check stability and compare free energies of competing roots before declaring which is physical.

The core equations

The self-consistency equation is the closure of mean-field theory. We quote the mean-field Hamiltonian (derived in aq-sm-mean-field-theory by dropping the fluctuation product $\delta s_i\,\delta s_j$ ) and then impose that a single spin's thermal average reproduce the order parameter it assumes.

m=\tanh\!\big[\beta(Jzm+h)\big] \qquad (1)

where: $m=\langle s_i\rangle$ — the order parameter (magnetization per spin); $\beta=1/k_BT$ ; $J>0$ — ferromagnetic coupling; $z$ — coordination number (neighbours per site); $h$ — external field in energy units. The right side is the exact single-spin magnetization $\langle s\rangle=\tanh(\beta h_{\rm eff})$ in the effective field $h_{\rm eff}=Jzm+h$ ; setting it equal to $m$ closes the loop. Read physically, $(1)$ says "the average that a spin produces must equal the average it assumes." The transition temperature is fixed not by solving $(1)$ but by asking when its trivial solution first becomes unstable — when the slope of the right side at the origin reaches $1$ .

\left.\frac{\mathrm d}{\mathrm dm}\tanh(\beta Jzm)\right|_{m=0}=\beta Jz=1 \;\;\Longrightarrow\;\; k_BT_c=Jz \qquad (2)

where: $T_c$ — the critical temperature. For $\beta Jz<1$ (i.e. $T>T_c$ ) the tanh starts below the line $y=m$ and only $m=0$ solves $(1)$ ; for $\beta Jz>1$ ( $T<T_c$ ) the steeper start forces two extra crossings. The marginal slope- $=1$ case is exactly the bifurcation point. Note $T_c$ rises linearly with $z$ : more neighbours, more reinforcement, order surviving to higher temperature. Every stability question is answered by the free energy per spin, whose stationarity reproduces $(1)$ and whose curvature decides which roots are minima.

f(m)=\tfrac12 Jzm^2-k_BT\,\ln\!\big[2\cosh(\beta(Jzm+h))\big] \qquad (3)

where: $f$ — Helmholtz free energy per spin. The first term is the mean-field energy (with the double-counting correction $+\tfrac12 Jzm^2$ ); the second is the entropy of $N$ independent spins in the field $h_{\rm eff}$ . At $h=0$ this function is manifestly even in $m$ — the $\mathbb{Z}_2$ symmetry — and $\partial f/\partial m=0$ returns equation $(1)$ exactly. Expanding $(3)$ in small $m$ near $T_c$ exposes the universal Landau structure that governs the bifurcation.

f(m)=f_0+\tfrac12\,a(T)\,m^2+\tfrac14\,b\,m^4-hm+\cdots,\qquad a(T)=Jz\big(1-\beta Jz\big),\;\; b=\tfrac13\,\beta^3(Jz)^4>0 \qquad (4)

where: $a(T)$ — the quadratic coefficient, which changes sign at $\beta Jz=1$ (i.e. $T=T_c$ ), $a\propto t=(T-T_c)/T_c$ ; $b$ — the strictly positive quartic coefficient that stabilizes the ordered state. When $a>0$ the well is single; when $a<0$ it splits into a symmetric double well with minima at $m^{*2}=-a/b$ . This is the mean-field face of Landau theory (owned in full by aq-sm-landau-theory-phase-transitions).

The physical picture, quantified

Derivation AThe pitchfork bifurcation and the order-parameter exponent

The goal is to show that the self-consistency equation $(1)$ at $h=0$ undergoes a supercritical pitchfork bifurcation at $T_c$ : the trivial solution $m=0$ persists everywhere but loses stability at $T_c$ , and two ordered branches $\pm m^*$ emerge continuously with $m^*\propto(T_c-T)^{1/2}$ . The strategy is a small- $m$ Taylor expansion of the tanh, keeping the first nonlinear (cubic) term; factoring out the trivial root exposes the nontrivial branch algebraically, and the near- $T_c$ limit reads off the exponent. The one thing to watch is that the cubic term must be negative (a saturating nonlinearity) for the bifurcation to be supercritical — a soft, continuous onset of order rather than a jump.

Step 1. Expand the tanh to cubic order. With $K\equiv\beta Jz$ , the identity $\tanh x=x-\tfrac13x^3+O(x^5)$ turns $(1)$ into $m=Km-\tfrac13K^3m^3+\cdots$ . Why: near $T_c$ the ordered $m$ is small, so the leading correction to the linear response is what decides whether new solutions appear; the cubic term is the lowest that can bend the tanh back to re-cross the line.

Step 2. Factor out the trivial root. Rearranged, $0=m\big[(K-1)-\tfrac13K^3m^2+\cdots\big]$ . Why: $m=0$ is a solution at every temperature (it is the factor $m$ out front); the bracket carries all the physics of the ordered branch. The transition is exactly where the bracket's constant term $(K-1)$ changes sign.

Step 3. Solve the nontrivial branch. Setting the bracket to zero gives $m^{*2}=\dfrac{3(K-1)}{K^3}$ , which is real and positive only when $K>1$ , i.e. $T<T_c$ . Why: for $K<1$ there is no real ordered root and only the disordered state survives; at $K=1$ the two roots emerge from zero — the pitchfork. That the roots appear continuously (not by a jump) is the supercritical signature.

Step 4. Take the near- $T_c$ limit. Writing $K=T_c/T$ so that $K-1=(T_c-T)/T\approx-t$ with $t=(T-T_c)/T_c\to0^-$ , and $K\to1$ , gives $m^{*2}\approx3(-t)$ , hence $m^*\simeq\sqrt3\,(-t)^{1/2}$ . Why: the leading coefficient of $m^{*2}$ in $t$ is exactly $-3$ (verified symbolically), fixing both the prefactor $\sqrt3$ and the exponent. The order parameter turns on with a vertical tangent — infinitely fast in $t$ but continuously in value.

Step 5. Read off the exponent. Comparing with the definition $m\sim(T_c-T)^{\beta}$ gives the mean-field order-parameter exponent $\boxed{\beta=1/2}$ , independent of dimension $d$ and coordination $z$ . Why: nothing in this expansion knew about spatial dimension — that is exactly why the mean-field exponent is universal within the theory and, equally, why it is wrong below $d=4$ where fluctuations dominate (the full exponent set is owned by aq-sm-mean-field-critical-exponents). *Verified: the initial slope $=K$ , the branch $m^{*2}=3(K-1)/K^3$ , the leading behaviour $m^{*2}\to-3t$ (so $\beta=1/2$ , prefactor $\sqrt3$ ), and the reduced-temperature fixed-point table are all checked in verify_mean-field-self-consistency.py against Pathria & Beale and Kardar SPoF Ch. 3.*

Go deeper · Derivation B · 2 minStability, degenerate minima, and spontaneous symmetry breaking

Derivation BStability, degenerate minima, and spontaneous symmetry breaking

Here the goal is to prove that below $T_c$ the symmetric root $m=0$ is a free-energy maximum while the two ordered roots $\pm m^*$ are genuine, degenerate minima — so that the equilibrium state is doubly degenerate and the system must break the $\mathbb{Z}_2$ symmetry by occupying one well. The route is the Landau form $(4)$ : locate the stationary points, test the curvature $f''$ , and compute the free-energy gap between the ordered and disordered states. The subtle point is that degeneracy plus a diverging barrier — not any asymmetry in $f$ — is what makes the broken symmetry real and irreversible in the thermodynamic limit.

Step 1. Find the stationary points. From $(4)$ at $h=0$ , $f'(m)=a\,m+b\,m^3=0$ gives $m=0$ or $m^{*2}=-a/b$ . Why: the ordered roots exist only when $-a/b>0$ , i.e. $a<0$ ( $T<T_c$ ); above $T_c$ the only stationary point is the origin.

Step 2. Test the curvature. $f''(m)=a+3bm^2$ . At the origin $f''(0)=a$ : positive (minimum) for $T>T_c$ , negative (maximum) for $T<T_c$ . At the ordered roots $f''(\pm m^*)=a+3b(-a/b)=-2a>0$ for $a<0$ . Why: below $T_c$ the origin flips from valley to hilltop while the two flanking roots become the valleys — the double well opens. The maximum at $m=0$ is the free-energy barrier separating the broken-symmetry states.

Step 3. Compute the free-energy gap. $f(m^*)-f(0)=\tfrac12 a m^{*2}+\tfrac14 b m^{*4}=-\dfrac{a^2}{4b}<0$ . Why: the ordered state lies strictly below the disordered one, so it is the true equilibrium; the gap grows as $a^2\propto t^2$ near $T_c$ , vanishing smoothly at the transition (a continuous transition, no latent heat).

Step 4. Establish the degeneracy and the symmetry. Because $f(m,0)$ is even in $m$ , $f(+m^*)=f(-m^*)$ exactly: the two minima are energetically identical. Why: the $\mathbb{Z}_2$ symmetry of the Hamiltonian is inherited by the free energy, forcing the wells to be mirror images. The theory cannot prefer one; the system must.

Step 5. Break the symmetry. A field $h\to0^+$ tilts the wells by $-hm$ , lowering the $+m^*$ well and selecting it; in the thermodynamic limit the barrier $\sim N\,a^2/b$ between wells diverges, so once selected the state cannot tunnel back. Why: for any finite system the exact ensemble average is still $\langle m\rangle=0$ (both wells are sampled), so symmetry breaking is strictly a $N\to\infty$ phenomenon — the non-commuting of the limits $h\to0$ and $N\to\infty$ is precisely what "spontaneous" means. *Verified: $f(m,0)$ even in $m$ , $a=Jz(1-\beta Jz)$ sign-changing at $T_c$ , $b=\tfrac13\beta^3(Jz)^4>0$ , the ordered minima $m^{*2}=-a/b=3(K-1)/K^3$ , and $f''$ signs are all checked symbolically in verify_mean-field-self-consistency.py, matching Landau & Lifshitz §142–148 and Sethna Ch. 9.*

The same mathematics can be made exact, not approximate, for one special model — a fact that dignifies mean-field theory as the leading term of a controlled expansion rather than an uncontrolled guess. For the infinite-range (fully connected) Ising model $H=-\tfrac{J}{2N}\big(\sum_i s_i\big)^2-h\sum_i s_i$ , a Hubbard–Stratonovich (Gaussian) transformation writes $Z=\int\mathrm dm\,e^{-N\beta f(m)}$ with exactly the free energy $(3)$ . Because the system size $N$ multiplies the exponent, Laplace's method makes the integral dominated by the minimum of $f$ as $N\to\infty$ , with corrections of relative order $1/N$ . The saddle-point condition $\partial f/\partial m=0$ is then precisely the self-consistency equation $(1)$ — no approximation. Mean-field theory is therefore exact in the limit of infinite coordination ( $z\to\infty$ ), infinite range, or infinite dimension, and its fluctuation corrections are the systematic $1/z$ , $1/d$ expansion.

Limiting cases

The single equation $(1)$ contains every critical exponent of the theory in its various limits.

Case	Result	Meaning
High temperature, $T\gg T_c$	$m\simeq\dfrac{\beta h}{1-\beta Jz}=\dfrac{h}{k_B(T-T_c)}$	Linearizing the tanh gives Curie–Weiss; $\chi=\partial m/\partial h\sim(T-T_c)^{-1}$ diverges at $T_c^+$ , the exponent $\gamma=1$ .
Just below $T_c$ , $h=0$	$m\simeq\sqrt3\,\Big(\dfrac{T_c-T}{T_c}\Big)^{1/2}$	The cubic expansion gives the order-parameter exponent $\beta=1/2$ , independent of $d$ and lattice.
At $T_c$ , small field	$m\simeq(3\beta_c h)^{1/3}\propto h^{1/3}$	The critical isotherm: the cubic balance $\tfrac13 m^3\simeq\beta_c h$ gives $\delta=3$ .
Zero temperature, $T\to0$	$m\to1$	$\tanh(\beta Jz)\to1$ : full saturation. Mean-field captures the fully ordered ground state even where it mishandles low- $T$ excitations.
Infinite range / $z\to\infty$	self-consistency becomes exact	The Hubbard–Stratonovich saddle point is exact; fluctuations are $O(1/N)$ .

Common trapSpontaneous symmetry breaking is not the free energy losing its symmetry

The free energy $f(m)$ stays perfectly symmetric ( $f(m)=f(-m)$ at $h=0$ ) above and below $T_c$ — it never loses its $\mathbb{Z}_2$ symmetry. What changes is the number of minima: one symmetric minimum becomes two degenerate ones. Symmetry is broken by the state occupying a single well, not by any asymmetry appearing in $f$ . And for any finite system the exact statistical average is still $\langle m\rangle=0$ , because both wells are sampled. Symmetry breaking is real only in the thermodynamic limit, where the inter-well barrier diverges; equivalently, it requires taking $N\to\infty$ before $h\to0$ . Confusing "two minima exist" with "the system is magnetized" — or expecting a finite simulation to show a nonzero $\langle m\rangle$ without a symmetry-breaking field — is the classic error.

Worked examples

1 · Solving the fixed point at half the critical temperature (confidence builder)

We solve the $h=0$ self-consistency equation $m=\tanh(m/t)$ (reduced temperature $t=T/T_c$ ) at $t=0.5$ and check it against the near- $T_c$ approximation, to feel how the fixed point behaves deep in the ordered phase.

Step 1. In reduced units $\beta Jz=T_c/T=1/t$ , so the equation is $m=\tanh(m/t)$ . At $t=0.5$ this is $m=\tanh(2m)$ .

Step 2. Iterate from a seed $m_0=0.9$ : $\tanh(1.8)=0.9468$ , $\tanh(1.894)=0.9556$ , $\tanh(1.911)=0.9573$ , converging to $m^*=\mathbf{0.9575}$ .

Step 3. Compare the near- $T_c$ form $m\approx\sqrt{3(1-t)}=\sqrt{1.5}=1.225$ . This overshoots the true value — as it must, since $\sqrt{3(1-t)}$ is only the leading small- $(1-t)$ term and $t=0.5$ is far from $T_c$ . Deep in the ordered phase the tanh saturation caps $m$ near $1$ .

Step 4. The lesson: at $t=0.5$ the magnet is already $96\%$ saturated. The $(-t)^{1/2}$ law is a near- $T_c$ statement only; the full transcendental equation must be solved numerically away from the critical point.

2 · The critical isotherm exponent δ (exam level)

We extract the mean-field exponent $\delta$ from the response exactly at $T=T_c$ , where the linear term in the expansion vanishes and the field couples to the cubic nonlinearity.

Step 1. At $T=T_c$ , $\beta_c Jz=1$ , so the expanded self-consistency equation $m=\beta Jz\,m-\tfrac13(\beta Jz)^3m^3+\beta h+\cdots$ becomes, keeping leading terms, $m\approx m-\tfrac13 m^3+\beta_c h$ .

Step 2. The linear terms cancel identically (that is what $T=T_c$ means), leaving $\tfrac13 m^3\approx\beta_c h$ .

Step 3. Solving, $m\approx(3\beta_c h)^{1/3}$ . So the magnetization responds to a field as $m\propto h^{1/3}$ , i.e. $h\propto m^{\delta}$ with $\boxed{\delta=3}$ .

Step 4. Numerically, with $k_BT_c=Jz$ and a field $h=0.01\,Jz$ : $\beta_c h=0.01$ , so $m=(0.03)^{1/3}=\mathbf{0.311}$ . A one-percent field at $T_c$ produces a $31\%$ magnetization — the anomalously large response of the critical isotherm, a direct fingerprint of the vanishing linear restoring term.

Origins · 1 minspontaneous symmetry breaking and how this was discovered

spontaneous symmetry breaking

the ordered state breaks a symmetry the Hamiltonian retains — Weiss 1907, formalized by Landau 1937 and Nambu, Goldstone, Anderson, Higgs in the 1960s · degenerate vacua and order-parameter selection

Spontaneous symmetry breaking is the principle that a system's equilibrium state can carry less symmetry than the laws that govern it. Pierre Weiss first met it in 1907: his molecular-field magnet acquires a definite magnetization below $T_c$ even though the underlying interactions have no preferred direction. Lev Landau turned it into a general framework in 1937, introducing the order parameter — a quantity that is zero in the symmetric phase and nonzero in the broken one — and the free-energy expansion that bears his name. The concept then migrated to quantum field theory: Yoichiro Nambu (Nobel 2008) recognized it in superconductivity around 1960, Jeffrey Goldstone proved in 1961 that breaking a continuous symmetry produces massless modes (Goldstone bosons), and Philip Anderson, Peter Higgs, and others showed in 1964 how a gauge field evades that theorem to acquire mass — the mechanism behind the Standard Model's $W$ and $Z$ bosons and the Higgs particle found in 2012. The mean-field self-consistency equation is where a physics student first meets the whole circle of ideas in fully computable form: a symmetric equation, degenerate minima, and a state that must choose.

History

1895

— Pierre Curie measures the paramagnetic susceptibility, later summarized as the Curie law.

1907

— Weiss introduces the molecular field, predicting spontaneous magnetization below T_c and the Curie–Weiss law above it.

1937

— Landau formulates the order-parameter theory of continuous transitions, the free-energy expansion underlying the double well.

1960

— Nambu identifies spontaneous symmetry breaking as the mechanism of superconductivity.

1961

— Goldstone proves that breaking a continuous symmetry yields massless modes.

1964

— Higgs, Englert, and Brout show how a gauge symmetry breaking gives mass to force carriers.

2008

— Nambu shares the Nobel Prize for the mechanism of spontaneous symmetry breaking in subatomic physics.

Going further

Texts. Pathria & Beale, Statistical Mechanics, 3rd ed., Ch. 11–12 (Ising model, mean-field and Landau theory). Kardar, Statistical Physics of Fields, Ch. 2–3 (saddle point, mean field, Landau expansion). Sethna, Statistical Mechanics: Entropy, Order Parameters, and Complexity, Ch. 8–9 (order parameters, the double well). Landau & Lifshitz, Statistical Physics Part 1, §142–148. Goldenfeld, Lectures on Phase Transitions and the Renormalization Group, Ch. 4–5 (Hubbard–Stratonovich, saddle-point exactness).

Provenance. Base article: AllQFT track `sm`, "Self-Consistency Equation and Spontaneous Symmetry Breaking" (STAT.INTER.MFSC.1257); source Pathria & Beale, Statistical Mechanics, 3rd ed., Ch. 11–12. All symbolic and numeric claims re-verified in verify_mean-field-self-consistency.py: the fixed-point slope and $T_c=zJ/k_B$ , the pitchfork branch $m^{*2}=3(K-1)/K^3$ and exponent $\beta=1/2$ with prefactor $\sqrt3$ , the $\mathbb{Z}_2$ -even free energy with degenerate minima, the Landau coefficients $a=Jz(1-\beta Jz)$ , $b=\tfrac13\beta^3(Jz)^4>0$ , and the reduced-temperature fixed-point table. No base physics error found; the base's decoupling and Hubbard–Stratonovich saddle are standard and reproduced.

Prerequisites. Ising model and the partition function · Canonical ensemble and free energy · Van der Waals Equation of State · Maxwell Construction and Coexistence.

Leads to. Landau theory of phase transitions · Ginzburg criterion · Critical exponents and universality · Renormalization group.

The equation always has the solution m = 0. How can there be a phase transition if that root never disappears?

The root $m=0$ genuinely persists at every temperature — but its stability does not. Above $T_c$ it is the unique free-energy minimum (the disordered equilibrium). At $T_c$ it becomes marginally stable, and below $T_c$ it turns into a free-energy maximum: a valid stationary point of the equation but a locally unstable one. Simultaneously two new roots $\pm m^*$ appear as the true minima. So the transition is not the disappearance of the $m=0$ solution but the change in its stability and the birth of competing solutions. This is exactly why you must always test $f''$ and compare free energies rather than just solving the equation.

If I simulate a finite Ising magnet below T_c, will I measure a nonzero magnetization?

Not on average, if you wait long enough and apply no field. In a finite system both wells $\pm m^*$ are accessible: the system occasionally tunnels (flips its whole magnetization) over the barrier, so the long-time average of $m$ is zero, respecting the exact $\mathbb{Z}_2$ symmetry. What you do see is that $\langle|m|\rangle$ or $\langle m^2\rangle$ is large, and that the tunnelling time grows extremely fast with system size. In the thermodynamic limit the tunnelling time becomes effectively infinite, the system is trapped in one well, and $\langle m\rangle\neq0$ . Practically, simulators either measure $\langle|m|\rangle$ , or apply a tiny symmetry-breaking field and take $N\to\infty$ first.

Why is the exponent exactly 1/2, and why does that make it wrong?

The value $1/2$ comes purely from the cubic term in the Taylor expansion of the tanh (or equivalently the $m^4$ term in the Landau free energy): balancing the quadratic and quartic terms of a smooth, analytic free energy always gives $m\propto(-a)^{1/2}\propto(T_c-T)^{1/2}$ . Nothing in the derivation references spatial dimension, because mean-field theory has replaced the spatial neighbourhood by a single averaged field. That is exactly the flaw: near $T_c$ the correlation length diverges and long-wavelength fluctuations of the order parameter dominate — and those are what the mean field threw away. Below $d=4$ the true singular behaviour is controlled by fluctuations, giving non-classical exponents (e.g. $\beta\approx0.33$ in 3D). The mean-field $1/2$ is correct only at and above the upper critical dimension.

What exactly picks the +m well over the −m well?

Anything that breaks the $\mathbb{Z}_2$ symmetry, however slightly. In practice: an infinitesimal external field $h\to0^+$ (which tilts the wells and lowers one), the boundary conditions of the sample, a stray field, or the initial condition of a quench. The crucial mathematical statement is that the limits do not commute: $\lim_{h\to0^+}\lim_{N\to\infty} m(h,N)=+m^*$ , whereas $\lim_{N\to\infty}\lim_{h\to0} m=0$ . "Spontaneous" means the ordering survives as the field is removed after the system is made infinite. Which specific well is chosen is not predicted by the theory — only that one is chosen, and that the choice is stable.

How is this the same as the van der Waals liquid–gas loop?

Structurally identical. The van der Waals equation is the mean-field theory of a fluid: each molecule moves in the averaged attraction of all the others. Its isotherms below the critical temperature develop a region with three densities for one pressure — the exact analogue of the three roots $\{0,\pm m^*\}$ of the self-consistency equation. The unstable middle root (negative compressibility) mirrors the free-energy maximum at $m=0$ ; the two stable roots are the liquid and gas, mirroring $\pm m^*$ . Choosing between coexisting phases by comparing free energies is the Maxwell construction (aq-sm-maxwell-construction), the fluid counterpart of comparing $f(\pm m^*)$ . Same bifurcation, same equal-area logic, different order parameter.

Is the self-consistency equation ever exact, or always an approximation?

It is exact for the infinite-range (fully connected) model, where every spin couples equally to every other. There a Hubbard–Stratonovich transformation writes the partition function as $Z=\int\mathrm dm\,e^{-N\beta f(m)}$ with the free energy $f(m)$ we used; because the extensive factor $N$ multiplies the exponent, Laplace's method makes the integral exactly dominated by the minimum of $f$ as $N\to\infty$ , and the saddle condition is the self-consistency equation. For finite-range models in finite dimension it is an approximation whose error is a controlled expansion in $1/z$ (or $1/d$ ): the more neighbours a spin has, the more accurately their fluctuating sum is captured by its mean, and the better mean-field theory becomes. It grows exact as $z\to\infty$ .