I have come across the concept of phase transitions in various contexts. From simple phase transition between different states of matter like water to ice and so on, to phase transition in magnetic materials from the Ising model and metal-insulator transition like the one proposed Sir Nevill Mott.

What is the generalized abstract definition of a phase transition?

Can anyone explain, what exactly are phase transitions in these specific cases given above and more?

  • 4
    $\begingroup$ I think you should consult a book on thermodynamics or statistical thermal physics or both. Basic idea is that at a phase transition there is a re-ordering of the structure, leading to a discontinuity in the equilibrium entropy or one of its derivatives, as a function of some other macroscopic variable such as temperature or applied field. $\endgroup$ Feb 28, 2019 at 16:17
  • 2
    $\begingroup$ to enlarge slightly upon @AndrewSteane's comment, there's a lot of cool physics involved in phase transitions generally, and the study of phase transitions has yielded mathematical models of them which among other things furnish schemes for placing them in different classes and so forth. $\endgroup$ Feb 28, 2019 at 21:37
  • $\begingroup$ @nielsnielsen Could you tell me more about them? Are you referring to the Ising model and Lars Onsager's exact solution to the problem? $\endgroup$
    – Fracton
    Mar 1, 2019 at 3:06
  • $\begingroup$ @AndrewSteane How do you differentiate between these Phase transitions? $\endgroup$
    – Fracton
    Mar 1, 2019 at 3:07
  • $\begingroup$ @EverydayFoolish, I last studied them 45 years ago, so I am not the guy you should talk to- but there's lots written about them that's available out there. $\endgroup$ Mar 1, 2019 at 5:33

2 Answers 2


A very large class of phase transitions are characterized by the breaking of some symmetry. Usually one finds a quantity called the order parameter, and finds its scaling with respect to an energy scale for the system, like the temperature. Usually for a phase transition one finds that the order parameter is either discontinuous or one of its derivatives is discontinuous. Now finding an order parameter is more of an art than a procedure, so a more systematic way of studying phase transitions is within the frame work of Landau theory.

In this framework one studies the free energy of the system in terms of an averaged or course grained version of the system. The main assumption is that the free energy can be expanded in terms of a ‘mean field" $\phi$

\begin{equation} \mathcal{F}(\phi)= a_1 (\partial \phi)^2+\sum_n b_n \phi^n + \dots \end{equation} Where the coefficients are found empirically at a particular temperature. As one course grains the system more and more we get different effective descriptions at each step because the coefficients $a_i$, $b_i$ change (this is in effect changing the energy scale at which the system sits). This means that at different scales the graph of the free energy looks different and roughly has different symmetries. The most famous example of this is the "mexican hat" potential.

\begin{equation} V(\phi)= \mu \phi^2 + \lambda \phi^4 \end{equation} When $\mu>0$ this potential has one minimum at $\phi=0$, but as $\mu$ becomes negative the potential has two minima. In the case where there is one minimum to the potential, small fluctuations around this potential are stable and have a symmetry under change of sign $\delta \phi \rightarrow -\delta \phi$. When there are two minima this is no longer the case becase changing the sign takes you from one vacuum to the other, so in that sense the symmetry is "spontaneously broken".

This whole procedure is usually called the Renormalization group. Critical points of a system are described by fixed points of the RG flow, meaning that they are scale invariant and the correlation length of excitations becomes infinite. This whole formalism is basically the same as a (Wilsonian) quantum field theory. So actually studying a big class of phase transitions (the ones that come from breaking of a symmetry) is equivalent to studying quantum field theories, and the study of critical points is the study of what are called conformal field theories.

Now it actually turns out that there are some phase transitions that are not characterized by the breaking of any symmetry. A good example of a real world example of this are topological insulators. In this case the phase is characterized by some "topological charge". There is a mathematical framework to try to classify all such phases of matter in terms of some objects called (braided) tensor categories.

  • $\begingroup$ Hologuin, you make a very interesting point about quantum field theories and conformal field theories, in a general sense how are they different? $\endgroup$
    – Fracton
    Oct 6, 2020 at 9:59
  • 1
    $\begingroup$ Conformal field theories are quantum field theories with very special behaviors. They are scale invariant so they look the same at all scales. $\endgroup$ Oct 6, 2020 at 17:10

For me a phase transition is defined (somewhat strictly) as a non-analyticity in the thermodynamic potential. A function $f(x)$ is non-analytic at $x=x_0$ if it can not be expanded in powers of $x-x_0$. For example $f(x) = \sqrt{x+1}$ can be expanded around $x=0$

$$f(x) \cong 1 + \frac{x}{2} - \frac{x^2}{8} + \dots $$

but this does not work at $x=-1$ because the first derivative of $f(x)$ diverges at $x=-1$. A more realistic example would be a function like $f(x) = \sqrt{x^2+m^2}$ with $m$ to be interpreted as some parameter of the system. We see that $f(x)$ is analytic for all values of $x$, except when $m=0$ and then we get a cusp $f(x) = \left|x\right|$. This means that there is a special point in the phase diagram where the free energy is non-analytic. This is a phase transition.

For example, in the Ising model, the free energy is a continuous function of the magnetic field $h$, but is has a cusp at $h=0$. Then the magnetisation (which is the derivative of the free energy with respect to the magnetic field) is discontinuous. At a the critical point, the cusp goes away but the magnetisation behaves as a power law (with a non-integer exponent). A different non-analyticity emerges.

I like this definition because it is directly tied to the fact that something special happens at phase transitions. Indeed, in statistical physics the Free energy is computed by taking the logarithm of the partition function

$$ F = -k_{\text{B}} T \, \text{Ln}\left(\sum_{\text{states}} \text{e}^{-E_{\text{state}}/(k_{\text{B}T})}\right) \, .$$

This is a sum of finite elements. Then it is analytic almost by construction. The way out is that the system may contain infinitely many particles so that the sum has infinitely many elements. Think for example of the geometric series, $\sum_{n=0}^\infty x^n = 1/(1-x)$ which is not analytic at $x=1$ even though this does not appear in the individual elements of the sum. The finite sum $\sum_{n=0}^N x^n = (1-x^{N+1})/(1-x)$, is always perfectly analytic. Physically, this means that a phase transition can only occur in the thermodynamic limit. Systems without an infinity somwhere do not have phase transitions.

This is of course a bit strict and needs to be generalised some times. The idea however remains. For example

  • Dynamical phase transitions can emerge when we look at the state of a system after an infinite time evolution as a function of its microscopic parameters.
  • In Renormalisation Group (RG) methods, we purposefully avoid non-analiticities by introducing a cut-off, but we let the cut-off run to infinity.
  • In quantum phase transitions, we send the temperature to zero (see the exponentials in the above equation where $1/T$ goes to infinity).
  • In the Ginzburg-Laudau picture, the free energy is first given as an analytic function of the magnetic field and the magnetisation. The physical free energy depends however only on the magnetic field with the magnetisation being computed by looking for the minimum of the free energy. This minimisation produces the non-analyticity and is equivalent to finding the asymptotic equilibrium of a dynamical system.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.