There is a moment in my workflow that I have repeated so many times it has become muscle memory. I sit down with a fresh set of \(M(H)\) isotherms from the VSM, construct the Arrott plot, and draw a straight line through the high-field region of the isotherm closest to \(T_C\). The y-intercept tells me about the spontaneous magnetization; the isotherm passing through the origin marks the Curie temperature. I have done this for every batch of my \(\text{Nd}_{0.7}\text{Sr}_{0.3}\text{MnO}_3\) nanoparticles. But for a long time, I did the procedure without asking the obvious question: why does plotting \(M^2\) against \(H/M\) produce straight lines at all? Who decided that these particular axes are the right ones, and where does the prescription come from?

The answer is Landau theory. And once you understand it, you realize something slightly unsettling: most of what an experimentalist does near a phase transition — Arrott plots, modified Arrott plots, the Banerjee criterion, even the classification into first- and second-order transitions — is just Landau theory in disguise. This post is my attempt to work through that theory from the ground up, the way I wish someone had explained it to me before I ever touched a magnetometer.

1. What Is a Phase Transition, Really?

Start with ice melting. The naive description is "water gets warmer and the solid becomes a liquid," but that misses the point entirely. What actually changes at 0 °C is the symmetry of how molecules arrange themselves. In ice, molecules sit on a crystal lattice — the system looks the same only if you shift it by specific lattice vectors. In liquid water, molecules are disordered, and the system looks the same under any translation. Melting is not primarily about energy; it is about the sudden loss of a preferred arrangement.

The same thing happens when iron loses its magnetism at the Curie temperature. Below \(T_C\), the atomic spins align along a common direction — there is a collective order. Above \(T_C\), thermal agitation wins and the spins point in random directions. But what does "random" mean precisely? It means that if you average the magnetic moment over the whole sample, you get zero: for every spin pointing up, there is, statistically, one pointing down.

This suggests the key observation that underlies everything else in this post. In every phase transition, there is a quantity that is exactly zero in one phase and nonzero in the other. We call it the order parameter. For a ferromagnet, the natural choice is the reduced magnetization \(m = \langle M \rangle / M_\text{sat}\): zero in the paramagnetic phase, finite below \(T_C\). For the liquid–gas transition, it is the density difference \(\rho_L - \rho_G\) between the coexisting liquid and gas. For a superconductor, it is the density of Cooper pairs (a more subtle, complex-valued object I will not develop here). The details differ wildly between systems, but the logical role is identical.

The order parameter is physics's way of asking: how much order is there, and in which direction?

The "which direction" part deserves attention, because it hides the deepest idea in this whole subject: spontaneous symmetry breaking. Above \(T_C\), a ferromagnet in zero field has no reason to prefer spin-up over spin-down. The underlying physics — the Hamiltonian, the interactions between spins — treats \(m\) and \(-m\) identically. Below \(T_C\), however, the material must magnetize in some direction, and it picks one. The equations remain perfectly symmetric; the state of the system does not. This is not a mathematical trick or a philosophical curiosity. It is what physically happens inside the material: a tiny fluctuation, an impurity, a stray field of a few millioersted decides which of the equivalent states the sample falls into, and once it has fallen, it stays.

2. Writing Down the Free Energy — Without Knowing the Microscopic Physics

Here is Landau's genius, and it takes a while to appreciate how audacious it is. To describe a phase transition, you do not need to know what the atoms are doing. You do not need the Hamiltonian, the exchange integrals, the band structure, or the crystal field. You only need to know one thing: the symmetry of the order parameter. Everything else follows from writing down the most general free energy that symmetry allows.

1

Expand the Free Energy in Powers of \(m\)

Near \(T_C\), the order parameter is small — the system is barely ordered, hovering close to the disordered phase. Whenever a quantity is small, a physicist's reflex is to expand in a power series. So we write the free energy density as

$$\mathcal{F}(m, T) = \mathcal{F}_0 + a_1\,m + a_2(T)\,m^2 + a_3\,m^3 + a_4(T)\,m^4 + \cdots - \mu_0 H\,m$$

where the last term is the coupling to an external magnetic field \(H\), which always favors alignment. At this point we know nothing about the coefficients. That is fine — symmetry will do the work.

2

Kill the Odd Powers with Symmetry

For a ferromagnet in zero field, flipping every spin in the sample (\(m \to -m\)) must leave the free energy unchanged, because there is no a priori reason to prefer up over down — nothing in the material distinguishes the two directions. But a term like \(a_1 m\) or \(a_3 m^3\) changes sign under this flip. The only way the free energy can respect the symmetry is if every odd coefficient vanishes identically: \(a_1 = a_3 = a_5 = \cdots = 0\). We are left with

$$\mathcal{F}(m, T) = \mathcal{F}_0 + a_2(T)\,m^2 + a_4(T)\,m^4 + a_6(T)\,m^6 + \cdots - \mu_0 H\,m$$

Notice what just happened: we discarded half the terms in the expansion using a physical argument, without a single calculation about atoms.

3

Give \(a_2\) Its Temperature Dependence

The coefficient \(a_2(T)\) controls whether the disordered state \(m = 0\) is stable. If \(a_2 > 0\), the free energy curves upward at the origin and \(m = 0\) is a minimum — the paramagnetic phase. If \(a_2 < 0\), the origin becomes a local maximum, and the system must roll away into an ordered state. The transition happens exactly where \(a_2\) changes sign, so the simplest assumption is a linear zero-crossing: \(a_2(T) = a_0\,(T - T_C)\) with \(a_0 > 0\). Keeping terms up to fourth order with a constant \(b > 0\) (needed so the free energy does not run off to \(-\infty\) at large \(m\)), we arrive at the canonical Landau free energy.

Landau Free Energy (Ferromagnet) $$\mathcal{F}(m, T) = \mathcal{F}_0 + a_0(T - T_C)\,m^2 + b\,m^4 - \mu_0 H\,m$$

It helps enormously to picture what this function looks like. Above \(T_C\), the quadratic term is positive and \(\mathcal{F}(m)\) is a single well centered at \(m = 0\): the system sits at the bottom, disordered. Exactly at \(T_C\), the quadratic term vanishes and the well becomes anomalously flat near the origin — the system barely knows where to sit, which is the microscopic origin of the giant response functions at criticality. Below \(T_C\), the origin puckers upward into a local maximum and two symmetric minima appear at \(\pm m_s\). The free energy landscape becomes the famous double well — the one-dimensional slice of the "Mexican hat" that appears everywhere from magnetism to the Higgs mechanism. The system must choose one of the two equivalent minima, and in choosing, it breaks the symmetry.

3. Deriving Critical Exponents from the Free Energy

So far this might look like formal bookkeeping. This section is where Landau theory earns its place: from that one polynomial, we can derive how the magnetization, the susceptibility, and the critical isotherm behave near \(T_C\). The recipe is always the same — the equilibrium state minimizes the free energy, so we set \(\partial \mathcal{F}/\partial m = 0\) and see what comes out.

Finding \(\beta\): how order grows below \(T_C\)

Set \(H = 0\) and minimize:

Minimization Condition (H = 0) $$\frac{\partial \mathcal{F}}{\partial m} = 2a_0(T - T_C)\,m + 4b\,m^3 = 0$$

One solution is always \(m = 0\). For \(T < T_C\), where that solution is unstable, dividing through by \(m\) gives the spontaneous magnetization:

Spontaneous Magnetization $$m_s = \sqrt{\frac{a_0(T_C - T)}{2b}} \;\propto\; (T_C - T)^{1/2}$$

The order parameter grows as a square root of the distance from \(T_C\). Comparing with the standard definition \(m_s \propto (T_C - T)^{\beta}\), we read off \(\beta = 1/2\). Note how the specific values of \(a_0\) and \(b\) only affect the prefactor — the exponent is universal within the theory.

Finding \(\gamma\): how the susceptibility diverges

The susceptibility measures how eagerly the system responds to a small field: \(\chi = \partial m/\partial H\big|_{H \to 0}\). The cleanest way to get it is to note that the inverse susceptibility is the curvature of the free energy at its minimum:

Inverse Susceptibility $$\chi^{-1} \propto \frac{\partial^2 \mathcal{F}}{\partial m^2} = 2a_0(T - T_C) + 12b\,m^2$$

Above \(T_C\), where \(m = 0\), this gives \(\chi \propto (T - T_C)^{-1}\) — the familiar Curie–Weiss law, with \(\gamma = 1\). Below \(T_C\), substituting \(m_s^2 = a_0(T_C - T)/2b\) gives \(\chi^{-1} = 4a_0(T_C - T)\): the same exponent \(\gamma = 1\), just a different prefactor. The physical picture is the flattening well I described above: as \(T \to T_C\), the curvature at the minimum vanishes, so an infinitesimal field produces an enormous response. Divergence of \(\chi\) is not a mystery — it is geometry.

Finding \(\delta\): the shape of the critical isotherm

Exactly at \(T = T_C\), the quadratic term is gone, and the equation of state \(\partial\mathcal{F}/\partial m = 0\) with the field term included reduces to

Critical Isotherm $$\mu_0 H = 4b\,m^3 \;\implies\; m \propto H^{1/\delta}, \quad \delta = 3$$

So at the critical temperature, magnetization grows as the cube root of the field. Three exponents, all from one polynomial and one minimization. Here is the scorecard against reality:

Exponent Physical meaning Landau prediction Real 3D ferromagnet
\(\beta\) How fast order grows below \(T_C\) 0.500 ≈ 0.326
\(\gamma\) How fast \(\chi\) diverges 1.000 ≈ 1.241
\(\delta\) Shape of the critical isotherm 3.000 ≈ 4.82
Landau theory predicts universal exponents. The universe, unfortunately, disagrees — and that disagreement turned out to be one of the most productive failures in the history of physics.

The mismatch is not experimental error. It is real, systematic, and it eventually forced physicists to invent the renormalization group. But hold that thought — because there is one special case where Landau's mean-field predictions in three dimensions are not approximately right but exactly right, and it happens to be the case my thesis lives in.

4. First-Order vs. Second-Order: The Role of \(b\)

Everything above assumed \(b > 0\). But nothing in the symmetry argument forces that — \(b\) is a material-dependent coefficient, and materials are allowed to make it small, zero, or negative. Whenever \(b\) can go negative, we must keep the next term in the expansion, \(c\,m^6\) with \(c > 0\), to keep the free energy bounded. Three qualitatively different worlds open up:

b > 0 • second-order m b = 0 • tricritical m b < 0 • first-order m
Free energy \(\mathcal{F}(m)\) below the transition for the three regimes of the quartic coefficient. The sign of one number decides the entire character of the transition.
When b = 0, something special happens. Mean-field theory changes its predictions — and in three dimensions, it turns out to be exactly right.

Why exactly right? For ordinary critical points, fluctuations invalidate mean-field theory below four dimensions. But at a tricritical point, the upper critical dimension — the dimension above which fluctuations become harmless — drops from four to three. Our world sits precisely at the boundary, so the tricritical mean-field exponents \(\beta = 1/4\), \(\gamma = 1\), \(\delta = 5\) hold in real 3D materials, up to logarithmic corrections. Landau theory, wrong almost everywhere, is vindicated at this one strange point.

This is not an abstract curiosity for me. My entire master's thesis lives at this point. In \(\text{Nd}_{0.7}\text{Sr}_{0.3}\text{MnO}_3\) nanoparticles, the transition sits at the crossover between first- and second-order behavior — the competition between double-exchange ferromagnetism and the phase-separation tendencies of manganites drives the effective \(b\) toward zero. When I first constructed the modified Arrott plot of my sample with \(\beta = 0.25\) instead of the mean-field 0.5, the isotherms snapped into straight parallel lines. That was the moment I knew the tricritical Landau free energy was the right description of my material — not the Heisenberg class, not the Ising class, but the singular point in between where \(b\) vanishes.

5. The Equation of State and the Origin of Arrott Plots

Now I can close the loop back to the bench, and answer the question from the opening paragraph. Take the standard Landau free energy (\(b > 0\)) and minimize it with the field term. The result is the magnetic equation of state:

Landau Equation of State $$\mu_0 H = 2a_0(T - T_C)\,m + 4b\,m^3$$

Divide both sides by \(m\):

Arrott Form $$\frac{\mu_0 H}{m} = 2a_0(T - T_C) + 4b\,m^2$$

Look at what this says. If you plot \(m^2\) on one axis and \(H/m\) on the other, each isotherm is a straight line. The slope is \(1/4b\) — the same for every isotherm, which is why the lines are parallel. The intercept is proportional to \(T - T_C\), which is why it changes sign exactly at the Curie temperature: the isotherm passing through the origin is \(T_C\). Every rule of thumb I learned as a recipe — parallel lines, origin-crossing isotherm, positive slopes for second order — is a line-by-line transcription of this one equation.

Arrott plots are not an empirical trick. They are a direct visualization of the Landau free energy landscape: each straight line is the equation of state at fixed temperature, and the geometry of the plot — slopes, intercepts, sign changes — maps one-to-one onto the coefficients of the free energy expansion.

The tricritical case follows the same logic with the \(m^6\) term. Setting \(b = 0\) and minimizing \(\mathcal{F} = a_0(T - T_C)m^2 + c\,m^6 - \mu_0 H m\) gives

Tricritical Equation of State $$\frac{\mu_0 H}{M} = 2a_0(T - T_C) + 6c\,M^4$$

Straight lines now require plotting \(M^4\) — not \(M^2\) — against \(H/M\). This is precisely the modified Arrott plot with \(\beta = 1/4\), \(\gamma = 1\), and it is why my NSMO isotherms refuse to straighten in the standard plot but become beautifully linear in the modified one. The plot is telling me, in geometric language, which term stabilizes the free energy of my material.

6. What Landau Theory Cannot Do (And Why That's Okay)

Honesty requires listing the failures, because they are severe. Landau theory ignores fluctuations entirely: it assumes the order parameter takes a single, uniform value, when in reality it fluctuates in space and time, and near \(T_C\) those fluctuations become correlated over enormous distances. Above four dimensions this is harmless; in our three-dimensional world it is catastrophic for ordinary critical points — which is exactly why the measured \(\beta \approx 0.326\) of a 3D Heisenberg-like ferromagnet disagrees with the predicted 0.5. The theory also cannot predict \(T_C\), \(a_0\), or \(b\) from microscopic physics; it parametrizes the transition rather than explaining it. And the closer you approach the critical point, where the correlation length diverges, the worse the theory gets — it fails precisely where a theory of critical phenomena should be at its best.

And yet. It provides a single universal framework that applies to every phase transition ever discovered — magnets, superconductors, liquid crystals, the electroweak vacuum. It correctly identifies the order of a transition from the sign of one coefficient. It becomes exact at the tricritical point in three dimensions. And every practical tool I use in the lab — every modified Arrott plot, every Kouvel–Fisher fit, every scaling collapse of \(\Delta S_M\) — is scaffolding built directly on Landau's expansion. The renormalization group did not discard Landau theory; it explained where the theory's boundaries lie and then built on top of it, using the same free-energy functional as its starting point.

Landau theory is not wrong. It is approximate — and knowing exactly where and why it fails is what makes it useful.
· · ·

7. What I Still Don't Fully Understand

In the spirit of learning in public, here are the gaps I am aware of in my own understanding — the places where I have used the machinery without being able to rebuild it from scratch.

Where do \(a_0\) and \(b\) actually come from? I know, in outline, that one can start from a Heisenberg Hamiltonian, apply a mean-field decoupling, and expand the resulting free energy to recover the Landau form with microscopic expressions for the coefficients. I have seen this done on a blackboard. I have never sat down and pushed the algebra through myself, term by term, and until I do, my understanding of what \(b\) "is" physically — and why manganites can drive it to zero — remains secondhand.

Anisotropy terms that break the \(m \to -m\) symmetry. My entire even-powers argument in Section 2 rests on the up–down symmetry of an idealized ferromagnet. Real manganites have crystal fields and magnetocrystalline anisotropy, and in some symmetry classes cubic invariants sneak into the expansion. How exactly those terms modify the transition — and whether they matter for my nanoparticles, where surface anisotropy is amplified — is something I can hand-wave about but not calculate.

Does the Landau expansion even converge? Goldenfeld remarks that the expansion should be understood as an asymptotic series rather than a convergent one. I have read the sentence several times. I have not tracked down the actual argument, and I suspect that understanding it properly would change how I think about the whole construction. It is on the list.

8. Further Reading

The references below are the ones I keep coming back to, in roughly the order I would recommend reading them.

When I started writing this post, I intended it to be a quick summary of a textbook chapter. It turned into something else: a realization that the plots I make every week are a seventy-year-old theory speaking to me in the language of straight lines. The details of my material — the double exchange, the phase separation, the nanoparticle surfaces — remain complicated and partially mysterious. But near the transition, none of that matters. Symmetry takes over, the free energy simplifies into a short polynomial, and knowing almost nothing turns out to be enough. That, I think, is the real lesson of Landau theory: it is the art of knowing exactly which details you are allowed to ignore.