Measure of Lee-Yang zeros

Question

Consider a statistical mechanical system (say the 1D Ising model) on a finite lattice of size $N$, and call the corresponding partition function (as a function of, say, real temperature and real magnetic field) $Z^{(N)}(t, h)$, where $t$ is temperature and $h$ - the magnetic field.

The partition function $Z$ is analytic (in finite volume $N$) and doesn't admit any zeros. However, as soon as one passes to complex field $h$ (or temperature, but let's consider complex field here), $Z$ admits zeros on the unit circle $S^1$ in $\mathbb{C}$.

Call the set of zeros $\mathcal{Z}_N$, where $N$ emphasizes finite lattice of size $N$. It is in general a nontrivial problem to decide whether the sequence of sets $\{\mathcal{Z}_N\}_{N\in\mathbb{N}}$ accumulates on some set in $S^1$, and if it does, to describe the topology of this limit set, which we'll call $\mathcal{Z}$.

Now, suppose that for a given system we proved that there does indeed exist a nonempty set $\mathcal{Z}$ such that $\mathcal{Z}_N\rightarrow\mathcal{Z}$ as $N\rightarrow\infty$ (in some sense - say in Hausdorff metric).

Is there a natural measure $\mu$ defined on $\mathcal{Z}$ that has physical meaning? If so, what sort of properties of this measure are physically relevant (say, relating to phase transitions)?

In my mind this is quite a natural question, because it translates into "Is there a natural way to measure the set where the system develops critical behavior?"

For example, one candidate would be the Hausdorff dimension. But I am interested more in something that would measure the density of zeros in a natural way (such as, for example, the density of states measure for quantum Hamiltonians).

EDIT: I know, of course, that the 1D Ising model is exactly solvable when interaction strength and magnetic field are constant. Here I implicitly assume that interaction (nearest neighbor, to keep it simple) and/or the magnetic field depend on lattice sites.

This post has been migrated from (A51.SE)

Comments

@Yvan: Also, in dimension one, as far as I know, depending on choice of interaction, distribution of zeros may also present a nontrivial problem, or am I wrong? In other words, I think even in dimension one, results and techniques are model dependent. Is this true?

This post has been migrated from (A51.SE)

---

@WNY: I guess that it depends what information you want to extract (obviously, if the coupling constants are uniformly bounded and decay fast enough with the distance between the corresponding spins, then there won't be a phase transition in the model). Now, independently of what you want to extract, I'd guess that the determination of the asymptotic locations of zeros is certainly difficult (and usually impossible), in general, when the interaction is not finite-range and periodic...

This post has been migrated from (A51.SE)

---

@YvanVelenik: I think I agree with what you wrote, but just to make sure: one can certainly be sure of the analyticity on the way to the thermodynamic limit --- and always the limit should be the last step, as is the whole point of the Y/L papers.

This post has been migrated from (A51.SE)

---

@WNY: approaching thermodynamics via the zeros seems very elegant, and I think people stochastically and independently re-discover this fact, work on it, and find that it's essentially intractable. Certainly, I know of no influential (physics) literature which builds in this direction --- despite having friends who work exactly on this. I don't know if that meta-observation is worth something ;-)

This post has been migrated from (A51.SE)

---

@genneth: I'm not sure I see which of my comments you are adressing... If you mean to say that we can be sure that the free energy is real analytic in finite systems, then yes, of course I agree. But if you don't make it quantitative, this does not tell you anything of interest about the limiting behaviour. Concerning your other comment (to WNY), the LY approach can be used to derive consequences of analyticity of the free energy to other quantities (say, exponential decay of correlations, for example); many interesting results were obtained this way a few decades ago.

This post has been migrated from (A51.SE)

---

I think the original paper by Yang and Lee explains this. I believe they actually show (or just assume) that in the thermodynamic limit there is a well-behaved density of zeros, and then use this density to calculate things like critical exponents.

This post has been migrated from (A51.SE)

---

I of course suspected that it should be (naturally) zero distribution. In fact, I think the distribution of zeros measure in the thermodynamic limit is just weak limit of distribution of zeros measures on finite lattices (as the lattice size grows to infinity). Thank you for the reference!

This post has been migrated from (A51.SE)