Critical Exponents

Critical exponents are among the most precisely tested predictions in all of physics. The superfluid transition of helium-4 (the $\lambda$-transition, XY class, $n=2$) has a slightly negative specific-heat exponent $\alpha\approx-0.0127$ — a cusp, not a divergence — and it was measured in a 1990s microgravity experiment (to escape gravitational rounding of $T_c$) to confront the renormalization-group prediction at the fourth decimal place. The 3D-Ising exponents that govern the liquid–gas critical point and uniaxial magnets have, since 2012, been computed to extraordinary precision by the conformal bootstrap, which exploits the conformal symmetry that emerges exactly at the critical point. On the numerical side, finite-size scaling on Monte-Carlo lattices — replacing $\xi\to L$ so that $\chi_{\max}\sim L^{\gamma/\nu}$ — remains the workhorse for extracting exponents of models with no exact solution. The same universality that unifies magnets and fluids reaches into cosmology (inflationary phase transitions), high-energy physics (the QCD chiral and deconfinement transitions), and even the onset of turbulence and percolation — anywhere a system develops a diverging correlation length.

Cool a magnet toward $T_c$ and correlated domains — patches of aligned spins — grow. Their typical linear size is the correlation length $\xi$. Far from $T_c$, $\xi$ is a few lattice spacings; approaching $T_c$ it swells, and at $T_c$ it diverges. A system whose only length scale has run off to infinity has, effectively, no length scale, and a scale-free system can respond to perturbations in only one functional form — a power law, because $f(bx)=b^{p}f(x)$ is solved uniquely by $f(x)\propto x^{p}$. That single observation is why critical exponents exist at all: they are the powers of the only functions a scale-free system is allowed to use.

Why universal? Coarse-grain the system — average spins over a small block and rescale — and the microscopic clutter (lattice type, atomic species) fades, while the long-wavelength order parameter survives. Two systems that flow, under repeated coarse-graining, to the same fixed point of the renormalization group end up with identical exponents; the microscopics set only non-universal amplitudes and the value of $T_c$. Mean-field theory is what you get when you ignore the fluctuations entirely — replace each spin by its average neighbourhood — and it works when a correlation volume $\xi^{d}$ holds so many fluctuating degrees of freedom that the central limit theorem tames them. That is true for $d>4$; below four dimensions the fluctuations within one correlation volume are not self-averaging, and they shift the exponents away from the mean-field values. The punchline figure is the operational meaning of an exponent: it is a slope on a log–log plot. Plot the order parameter $m$ against $|t|$ on logarithmic axes and you get a straight line whose slope is $\beta$. Mean-field draws that line with slope $\tfrac12$; the true 3D-Ising magnet draws it with slope $0.326$ — a visibly steeper rise of the magnetization as you cool below $T_c$. The gap between those two slopes is precisely the failure of mean-field, made geometric.

The exponents are defined as one-sided logarithmic limits of a response as $t\to0$. Concretely, if a quantity $f$ behaves as $f\sim|t|^{p}$ near the transition, the exponent is $p=\lim_{t\to0}\ln f/\ln|t|$. The first three exponents describe the bulk thermodynamics: the specific heat $C$, the zero-field order parameter $m$, and the susceptibility $\chi=\partial m/\partial h$.

where: $C$ is the specific heat at constant field (or volume); $m$ is the order parameter in zero conjugate field, defined only on the ordered side $t<0$; $\chi$ is the order-parameter susceptibility. Positive $\alpha$ and $\gamma$ mean a divergence; positive $\beta$ means a continuous onset from zero. Reading it physically: as you cool through $T_c$, $m$ rises from zero as a power $\beta$, while both $C$ and $\chi$ blow up as the transition is approached from either side. The remaining three exponents describe the response to a field and the spatial structure of fluctuations — the critical isotherm and the correlation function.

where: $h$ is the field conjugate to the order parameter; $\delta$ governs the nonlinear response exactly at $T_c$ (where $\xi=\infty$, so linear response fails); $\xi$ is the correlation length with exponent $\nu$; $G(r)$ is the two-point correlation of order-parameter fluctuations, decaying with the "anomalous dimension" $\eta$ (at $t=0$ the exponential cap disappears and $G\sim r^{-(d-2+\eta)}$). The physical reading: $\delta\ne1$ signals that at criticality the system is infinitely susceptible, so even a whisper of a field produces a nonlinear magnetization. The engine that ties all six together is the scaling hypothesis: the singular part of the free energy is a generalized homogeneous function of the two scaling fields $t$ and $h$.

where: $f_s$ is the singular free-energy density (in units of $k_BT$); $\Phi$ is a universal scaling function of the single combination $h/|t|^{\Delta}$; and $\Delta$ is the gap exponent. The content of this one line is enormous: differentiating $f_s$ with respect to $t$ and $h$ generates every thermodynamic exponent, and demanding self-consistency yields the scaling relations. Two systems that share $\Phi$ share every exponent — this is universality made precise.

Example 1 — Verify all four scaling relations for the exact 2D-Ising exponents. The 2D-Ising exponents are known exactly (Onsager, Yang), so they are the sharpest possible test of the scaling relations. Take $\alpha=0,\ \beta=\tfrac18,\ \gamma=\tfrac74,\ \delta=15,\ \nu=1,\ \eta=\tfrac14$, in $d=2$.

Step 1. Rushbrooke $\alpha+2\beta+\gamma$: $0+2\cdot\tfrac18+\tfrac74=\tfrac14+\tfrac74=\tfrac{8}{4}=2$. Equals 2. ✓

Step 2. Widom $\beta(\delta-1)$: $\tfrac18(15-1)=\tfrac18\cdot14=\tfrac{14}{8}=\tfrac74=\gamma$. Equals $\gamma$. ✓

Step 3. Fisher $\nu(2-\eta)$: $1\cdot(2-\tfrac14)=\tfrac74=\gamma$. Equals $\gamma$. ✓

Step 4. Josephson (hyperscaling) $d\nu$: $2\cdot1=2=2-\alpha=2-0$. Equals $2-\alpha$. ✓ All four hold with exact equality, which is why these rational numbers, and not the mean-field ones, describe a two-dimensional Ising magnet.

Example 2 — Read $\beta$ off a log–log plot, and see the mean-field error. Suppose you measure the magnetization at three reduced temperatures below $T_c$: at $|t|=10^{-2}$, $m=0.1000$; at $|t|=10^{-4}$, $m=0.01000$; at $|t|=10^{-6}$, $m=0.001000$ (mean-field data). Extract $\beta$.

Step 1. On a log–log plot the slope is $\beta$: from $|t|=10^{-2}$ to $10^{-4}$, $\log m$ drops by $\log(0.1/0.01)=1$ decade while $\log|t|$ drops by $2$ decades, so $\beta=\tfrac{1}{2}$. $\beta=0.500$ (mean field). ✓

Step 2. Now the real 3D-Ising magnet, same reduced temperatures, has $m\propto|t|^{0.326}$. From $|t|=10^{-2}$ to $10^{-6}$ (4 decades), $\log m$ changes by $0.326\times4=1.304$ decades. So over the range where mean-field predicts $m$ to fall by a factor $10^{2}=100$, the real magnet's $m$ falls only by $10^{1.304}\approx20.1$. The true magnetization is larger — the order parameter is more robust than mean-field claims (steeper log–log slope $\Rightarrow$ faster onset below $T_c$, i.e. $m$ rises more sharply from zero as you cool).

Step 3. The gap is quantitative: at $|t|=10^{-6}$, $m_{\text{MF}}=10^{-3}$ while $m_{\text{3D}}=(10^{-6})^{0.326}=10^{-1.956}\approx0.0111$ — about 11 times larger. Mean-field systematically underestimates the ordered-phase magnetization near $T_c$ because it ignores the correlated fluctuations that stiffen the order.

For the scaling hypothesis and the derivation of the relations, see Stanley, Introduction to Phase Transitions and Critical Phenomena, and Kardar, Statistical Physics of Fields, ch. 2–5. The mean-field benchmark is derived in full in the sibling article aq-sm-mean-field-critical-exponents; the diverging length that underlies everything is aq-sm-correlation-length-divergence; and the renormalization-group explanation of universality — and the $\epsilon=4-d$ computation of the exponents — is developed in the RG chapter (aq-sm-epsilon-expansion, aq-sm-wilson-fisher-fixed-point). Goldenfeld, Lectures on Phase Transitions and the Renormalization Group, is the standard pedagogical bridge. Provenance (from the base article). Mean-field exponents $\alpha=0,\beta=1/2,\gamma=1,\delta=3,\nu=1/2,\eta=0$ are standard (Landau–Lifshitz Statistical Physics Part 1; Pathria & Beale ch. 12). The 2D-Ising exact exponents are the Onsager/Yang values (Stanley; Kardar Fields ch. 7). The 3D-Ising values $(\beta=0.3265,\gamma=1.2372,\nu=0.6300,\eta=0.0364,\delta=4.789,\alpha=0.110)$ are conformal-bootstrap/Monte-Carlo standards (El-Showk et al. 2014; Pelissetto & Vicari, Phys. Rep. 2002). The scaling relations Rushbrooke, Widom, Fisher, and hyperscaling are textbook (Stanley; Kardar). The one-loop result $\nu=\tfrac12+\tfrac{n+2}{4(n+8)}\epsilon$ is the standard Wilson–Fisher expansion (Kardar Fields ch. 5; Goldenfeld). Prerequisites: Continuous Phase Transitions, Order Parameter, Landau Theory of Phase Transitions, Correlation Function. Leads to: Scaling Hypothesis and Scaling Relations, Universality and Universality Classes, Correlation Length and Its Divergence.