Past exam of the phase transitions and collective phenomena course of the University of Cambridge

Historical Part III paper. Exact student-answer availability for 2012 remains unverified; papers with advertised student answers are skipped. The public original is linked from each paper.

Original paper: Public original PDF, physical pages 486–489; cover on page 486, Q1 on page 487, its experimental plot on page 488, Q2–Q3 on page 489.

Use $s=|t|$, $x=h/s^{\Delta }$ and separate scaling functions above and below the transition. Up to a nonsingular conversion between physical and dimensionless field, magnetization scaling-function matching givesOn the ordered side take $h\to 0^{+}$ after the thermodynamic limit, selecting one pure thermodynamic phase. A finite nonzero $g_{m,-}(0^{+})$ impliesThe critical isotherm must have a finite nonzero limit at fixed small $h$ as $s\to 0$. Thus $g_m(x)\sim C\mathrm{sgn}(x)|x{|}^{1/\delta }$ at large $|x|$. Cancellation of the residual power of $s$ requiresThe absolute value and branch choice matter: spontaneous zero-field magnetization is below $T_c$, whereas the symmetry-unbroken thermodynamic phase has $m(t,0)=0$. The field crossover exponent is not the scaling dimension of the magnetization field.

A further field derivative of the magnetization scaling form givesAt zero field, on a branch with a finite nonzero susceptibility amplitude, comparison with $\chi \sim s^{-\gamma }$ yieldsConsequently $\alpha +2\beta +\gamma =2$ and $\gamma =\beta (\delta -1)$, the Rushbrooke and Widom scaling relations. Regular analytic terms do not change the leading divergent susceptibility.

At fixed zero field, ${\partial }_T=T_c^{-1}{\partial }_t$. Therefore, away from $t=0$ on either branch,Thus the singular heat capacity has exponent $\alpha$. An arbitrary analytic background must be added to obtain the full heat capacity. Vanishing branch amplitudes or the marginal case $\alpha =0$ need care: a finite jump is compatible with this power counting, while a logarithmic singularity requires a logarithmic refinement of the assumed homogeneous form.

Hyperscaling adds the physical assumption that the singular free energy associated with one correlation volume is of order $k_B{T}_c$:At fixed scaled field $x=h/s^{\Delta }$, this predicts $f_s\sim s^{d\nu }$, whereas the given thermodynamic form has power $2-\alpha$. HenceThe product $g_f(x)[g_{\xi }(x){]}^d$ can be a finite dimensionless function of fixed $x$; it need not be constant. The homogeneity of $\xi$ alone is insufficient: the correlation-volume free-energy assumption supplies the connection. Above an upper critical dimension, a dangerously irrelevant coupling can invalidate this argument; at a marginal dimension logarithms may qualify pure powers.

The susceptibility is the integrated connected correlation function, up to a nonsingular field-normalization factor:Above the transition at zero field the subtraction vanishes. In an ordered thermodynamic phase it is essential; integrating the disconnected constant would incorrectly give a volume divergence unrelated to the desired response.

With a microscopic cutoff $a$ and $r=|\mathbf{x}|$, radial integration and $u=r/\xi$ giveIf $g(0)$ is finite, its large-distance tail decays sufficiently rapidly and $\eta <2$, the dimensionless integral approaches a finite nonzero constant. ThusShort-distance contributions give a regular background. This is the Fisher scaling relation; it concerns the fluctuating connected part of the source correlation ansatz.

The annotations on the original plot give $\beta =0.368$ and $\gamma =1.215$. Use those displayed values rather than fitting the linear-axis curve as though its slope were a logarithmic exponent. Exponent inference from a magnetic data collapse identifies $\Delta =\gamma +\beta$.ThereforeThese estimates inherit both the empirical collapse exponents and the assumed scaling identities, including three-dimensional hyperscaling. They are not independent high-precision measurements of five exponents. In particular, the small difference defining $\alpha$ and the subtraction defining $\eta$ amplify uncertainties. No experimental curve or source image is reproduced.

Spontaneous symmetry breaking means that a pure thermodynamic phase is less symmetric than the Hamiltonian even when the selecting external field is removed. For example,The direction of the infinitesimal field is held fixed in this limit. Reversing the limits in a finite symmetric sample normally gives zero. For the XY model, an ordered state chooses one direction from a circle of energetically equivalent directions.

A Goldstone mode is a long-wavelength fluctuation along this continuously degenerate family. A uniform angle rotation costs no energy; a slowly varying one costs only gradient energy. In the stated static phase-only model the quadratic kernel is $\bar{K}{q}^2$, tending to zero as $q\to 0$. There is one independent soft order-parameter phase coordinate for broken $O(2)$ rotation. The static Hamiltonian does not specify a dynamical dispersion $\omega (q)$; that requires a dynamical law. Magnitude fluctuations instead have a restoring local curvature and have been excluded from this low-energy model.

For an infrared-regulated phase covariance, fix the uniform order-parameter phase zero mode and introduce ultraviolet and infrared cutoffs before taking limits. With the Fourier conventionthe dimensionless quadratic Hamiltonian diagonalizes asCompleting the square with a source gives $Z[J]/Z[0]=\exp [\frac{1}{2}\int JGJ]$, with inverse kernelThere is no additional $k_{B}T$ factor: it is already included in $\bar{K}$ because the supplied Hamiltonian is $\beta H$.

It is simplest to evaluate the real-space dependence by its defining Green equation, $-\bar{K}{\nabla }^{2}G={\delta }^{(d)}$. Rotational invariance gives $r^{d-1}{G}^{\prime }(r)=$ constant for $r>0$. Integrating over a small sphere fixes $-\bar{K}{S}_d{r}^{d-1}{G}^{\prime }(r)=1$, where $S_d=2{\pi }^{d/2}/\Gamma (d/2)$. Integrating once more,In two dimensions, take the limit after absorbing the divergent $r$-independent piece into $C$:For $d\le 2$, $G$ itself has an infrared-divergent constant in the infinite system; it is not a finite thermodynamic covariance without a regulator. The physical phase-difference variance is finite at fixed separation and cancels that constant. These are smooth spin wave correlations; compact-field vortex sectors are considered separately below.

The magnetization correlator is an average of a phase difference, not merely its lowest Taylor expansion. Gaussian phase averaging giveswhereThe coincident value is ultraviolet-regularized at a microscopic scale $a$. At large $r$:

Thus two is the lower critical dimension for conventional positive-temperature continuous-symmetry order under the short-range assumptions. Two dimensions can still have quasi-long-range order and a thermodynamic transition; zero limiting magnetization does not forbid them. The $d>2$ spin wave result describes an admissible low-temperature ordered thermodynamic phase, not order at arbitrarily high temperature. The Mermin-Wagner theorem supplies the more general no-order statement below or at two dimensions.^[ref]

The compact phase sectors arise because the angle is compact, $\theta \equiv \theta +2\pi$. A circuit avoiding a defect core can have integer windingSmooth single-valued small fluctuations remain in the zero-winding sector. Vortices with $q\ne 0$ cannot be unwound without a core or a boundary crossing; the smooth Gaussian theory misses those sectors. For $\theta =q\arg (x+iy)$ outside a core of radius $a$,The positional entropy of a freely placed defect is approximately $S/k_B=2\log (L/a)$. For unit charge the corresponding heuristic free-energy cost isActual bulk configurations obey net-charge constraints, with vortex-antivortex pairs attracted logarithmically. At low temperature they remain bound; they renormalize the phase stiffness but do not screen away the algebraic correlations. Heating can unbind the pairs into a screening vortex plasma with short-range correlations: this is the Berezinskii–Kosterlitz–Thouless transition.

The sharp criterion uses the long-distance renormalized stiffness $K_R$, not a bare coefficient: $K_R(T_{\mathrm{BKT}}^{-})=2/\pi$, giving algebraic exponent ${\eta }_c=1/4$. Both phases have zero conventional spontaneous magnetization at positive temperature; they differ in vortex binding, correlations and stiffness.^[ref]

A coarse-grained scalar order parameter $\varphi (\mathbf{x})$ describes a short-range Ising-like transition with sign symmetry. Write the dimensionless Landau-Ginzburg functional asHere $r=a_{t}t$, with the relevant critical-mass shift understood when fluctuations are included. Higher local powers and gradients are negligible only within a justified long-distance expansion. The partition function is $Z=\int \mathcal{D}\varphi e^{-\mathcal{H}}$ with a microscopic cutoff. The local functional is not already the fluctuation-integrated thermodynamic free energy.

Mean-field theory replaces this integral by its stable uniform saddle, $r{\varphi }_0+u{\varphi }_0^3=h$. At $h=0$,The branch curvature is $r_L=r+3u{\varphi }_0^2$: it is $r$ above and $2|r|$ below. HenceTogether with the finite heat-capacity jump from $f_{\min }$, this yieldsThe equal-$|t|$ susceptibility amplitude ratio is two. Uniform saddles also miss domain walls and spatial texture. For an $n$-component order parameter, symmetry allows a vector generalization; ordered transverse directions additionally carry Goldstone fluctuations.

To include small fluctuations set $\varphi ={\varphi }_0+\delta \varphi$. The Gaussian action is $\frac{1}{2}\int [c(\nabla \delta \varphi {)}^2+r_L(\delta \varphi {)}^2]$, givingThe displayed $G(q)$ is a spectral covariance per Fourier delta, avoiding an extra volume factor. As $r_L\to 0$, a growing correlation length lets increasingly large correlated regions fluctuate. The singular Gaussian heat correction contains $\int d^{d}q/(c{q}^2+r_L{)}^2$, which is infrared singular for $d<4$ and logarithmic at four. This signals failure of a simple saddle expansion; the Gaussian heat exponent obtained by extrapolation is not the exact interacting critical exponent.

The Ginzburg criterion compares the fluctuation of a correlation-volume average with ${\varphi }_0^2$, rather than using all ultraviolet fluctuations at a point. Keeping $q\lesssim {\xi }^{-1}$ below the transition gives, up to finite convention-dependent constants,Dividing by ${\varphi }_0^2=|r|/u$ yieldsFor $d<4$, writing $r=a_{t}t$ gives a nonuniversal Ginzburg reduced-temperature windowInside $|t|\lesssim \mathrm{Gi}$, interacting fluctuations must be treated. For example, in three dimensions $\mathrm{Gi}\sim u^2/(c^3{a}_t)$ in this normalization. Above four dimensions the long-wavelength ratio vanishes at criticality; at four it is marginal and interaction-generated logarithms require further analysis.

The spatial gradient term supplies correlations absent from uniform Landau theory. The Ginzburg test bounds where its mean-field saddle remains self-consistent; it does not compute nonclassical exponents or a universal material-independent window.^[ref]

Renormalization compares the same long-distance physics at different observation scales. The starting object is a microscopic or cutoff statistical Hamiltonian ${\mathcal{H}}_g$ with every symmetry-allowed coupling $g$. Because a critical correlation length is much larger than a lattice spacing, repeatedly eliminating short-distance detail can reveal scale-invariant behavior.

In a real-space transformation choose a normalized blocking kernel $B(S|s)$ for retained variables. DefineSumming over $S$ proves exact preservation of $Z$ before truncation. The transformed Hamiltonian generally gains longer-range and many-body interactions and a field-independent constant. Observables must be blocked consistently. In momentum-shell language split $\varphi ={\varphi }_{<}+{\varphi }_{>}$, integrate $\Lambda /b<|q|<\Lambda$, then use $x^{\prime }=x/b$ and a field rescaling to restore cutoff and kinetic normalization. Coarse-graining, rescaling and tracking generated interactions are distinct steps.

Repeated transformations generate a flow $g^{\prime }=R_b(g)$ with composition $R_b{R}_c=R_{bc}$ when the same exact coarse-graining family is used. Losing degrees of freedom generally makes this a semigroup, despite its conventional group name. A renormalization-group fixed point is invariant after rescaling. At a continuous transition the large-distance theory approaches a critical renormalization-group fixed point; finite correlation length instead eventually flows into an ordinary thermodynamic phase regime.

Linearize in scaling coordinates $u_i$:For increasing observation length, $y_i>0$ is relevant and must be tuned, $y_i<0$ is irrelevant, and $y_i=0$ is marginal and needs nonlinear analysis. Discrete multipliers $b^{y_i}$ must be compared with one, not with zero. The stable manifold forms the critical surface after relevant directions have been set to zero.

After subtracting regular additive contributions, ordinary hyperscaling givesChoosing $b=|t{|}^{-1/y_t}$ or choosing the field as scale yieldsFor a critical two-point function $G(r)\sim r^{-(d-2+\eta )}$, the order parameter scaling dimension is $x_m=(d-2+\eta )/2$, and its conjugate field has $y_h=d-x_m=(d+2-\eta )/2$. This recovers the Q1 identities while distinguishing the gap exponent $\Delta$ from the field's scaling dimension $y_h$.

As a concrete dimensional check, start from a scalar gradient action. Restoring its coefficient uses ${\varphi }^{\prime }(x^{\prime })=b^{(d-2)/2}{\varphi }_{<}(b{x}^{\prime })$, so at the Gaussian fixed pointQuartic interactions are relevant below four dimensions. They can produce an interacting Wilson-Fisher fixed point, whose exponents differ from naive Gaussian counting. Above four, the quartic coupling is irrelevant at the Gaussian fixed point but dangerous for ordered observables: ${\varphi }_0^2=-r/u$ is singular when $u$ is dropped. Thus the simple two-variable hyperscaling formula is not universal above the upper critical dimension.

Different microscopic systems can flow to the same long-distance renormalization-group fixed point, explaining universality of exponents and normalized scaling forms. Critical temperatures, metric factors and regular free energies remain nonuniversal. Dimension, order parameter symmetry, interaction range and additional soft variables constrain the universality class. Truncations, perturbative expansions and numerics approximate the flow; they must not be confused with the exact elimination definition.^[ref]

For a ferromagnetic nearest-neighbour Ising model on a chain let $s_i=\pm 1$, $K=J/(k_{B}T)>0$ and $h=H/(k_{B}T)$. The dimensionless Hamiltonian is $-K{\sum }_i{s}_i{s}_{i+1}-h{\sum }_i{s}_i$. In one dimension, decimating every other spin closes exactly within the nearest-neighbour family, including its additive constant. This is the exact chain building block of the Migdal-Kadanoff approach; bond moving becomes an additional approximation in higher-dimensional lattices.

For exact zero-field Ising-chain decimation, sum over an intermediate spin $\sigma$ between two retained spins:Parallel and antiparallel end spins give $A{e}^{K^{\prime }}=2\cosh 2K$ and $A{e}^{-K^{\prime }}=2$. HenceFor an even periodic chain, $Z_N(K)=A^{N/2}{Z}_{N/2}(K^{\prime })$. Rescaling the lattice spacing $2a$ back to $a$ completes the renormalization step. Decimation of $b$ successive links gives $\tanh K^{\prime }=(\tanh K{)}^b$, as follows by multiplying the transfer matrix $b$ times.

Let $v=\tanh K$. Then $v^{\prime }=v^2$ on $0\le v\le 1$. The infinite-temperature renormalization-group fixed point $v=0$ attracts every $0\le v<1$; the zero-temperature renormalization-group fixed point $v=1$ is unstable to heating, with slope two. There is no intermediate ferromagnetic renormalization-group fixed point. Equivalently $K^{\prime }\simeq K^2$ for small $K$, while $K^{\prime }\simeq K-\frac{1}{2}\log 2$ for large finite $K$. Thus there is no finite-temperature ordered thermodynamic phase in the short-range chain.

The domain-wall scaling coordinate of an Ising chain near $T=0$ is the domain-wall Boltzmann factor $w=e^{-2K}$. Its recursion and the zero-field correlation length areSince $\xi (K^{\prime })/a=[\xi (K)/a]/2$, the recursion has the correct length scaling. Its exponent is one with respect to $w$; it is not a finite power of physical temperature about $T_c=0$. In the thermodynamic limit, a bond is a domain wall with probability $(1-\tanh K)/2\sim e^{-2K}$, giving the same exponentially large length scale. Their finite energy allows entropy to proliferate walls at any positive temperature.

The Ising-chain decimation constant matters for thermodynamics. With dimensionless free energy per spin $f=-\lim N^{-1}\log Z_N$,The second expression follows independently from transfer eigenvalues $2\cosh K$ and $2\sinh K$ and satisfies the recursion. Multiplying it by $k_{B}T$ restores the physical free energy. Dropping $A$ would preserve normalized retained-spin probabilities but lose the eliminated-spin entropy.

Finite-field Ising-chain decimation is also exact. Use the symmetric transfer matrixand read the new couplings fromHere $W_{++}=e^{2K+2h}+e^{-2K}$, $W_{--}=e^{2K-2h}+e^{-2K}$ and $W_{+-}=2\cosh h$. The latter off-diagonal entry, the field convention split between adjacent bonds and the constant must all be tracked consistently. Near the zero-temperature zero-field renormalization-group fixed point $h^{\prime }\simeq 2h$, while at $K=0$ the independent-spin field is unchanged.

In the higher-dimensional Migdal-Kadanoff approximation, moving bonds creates chains that can be decimated. With bond moving first on a $d$-dimensional hypercubic lattice and length factor $b$, the strengthened links have $b^{d-1}K$, giving the conventionFor $d=1$ no transverse bonds move, so this reduces to the exact chain result. In higher dimensions it discards geometric correlations and is generally approximate on the original lattice; moving and decimating in another order yields another approximation, not a contradictory exact chain formula.^[ref]