Skip to content
Tutorials Center
My learning

Topics

Lesson 38 of 39 · Projects and Practice

Project: Finite-Size Scaling and the Ising Critical Point

Locate the 2D Ising critical temperature from Monte Carlo data using the Binder cumulant and finite-size scaling of the susceptibility, and compare with Onsager's exact result.

A finite lattice has no true phase transition, yet simulations of finite lattices are how we study critical points. The bridge is finite-size scaling: how quantities change with the linear size L at the critical temperature carries the exponents, and certain combinations cross at T_c for all sizes. In this project you will estimate T_c of the 2D Ising model and the ratio γ/ν from your own simulations, and compare with the exact values T_c = 2.2692 J/k and γ/ν = 7/4.

The idea

Near T_c the only length scale is the correlation length ξ ~ |t|^(-ν), with t = (T - T_c)/T_c. When ξ exceeds L, the system cannot grow larger correlations, and every observable depends on the single ratio L/ξ. This gives scaling forms:

Text
m ~ L^(-β/ν) f_m(t L^(1/ν))
χ ~ L^(γ/ν)  f_χ(t L^(1/ν))

Two practical consequences:

  1. Susceptibility at T_c grows as L^(γ/ν); for the 2D Ising model this is L^(7/4).
  2. The Binder cumulant
Text
U_L = 1 - ⟨m⁴⟩ / ( 3 ⟨m²⟩² )

has no leading power of L in its scaling form, so curves of U_L(T) for different L cross at (approximately) T_c. Deep in the ordered phase U_L → 2/3 (since m = ±m₀), and in the disordered phase U_L → 0 (Gaussian m with ⟨m⁴⟩ = 3⟨m²⟩²).

Here m is the magnetization per spin of one configuration, m = (1/N) Σ s_i. The averages ⟨m²⟩ and ⟨m⁴⟩ are over configurations in the simulation, after equilibration.

Step 1: a fast simulator and the Binder cumulant

The simulator uses checkerboard Metropolis updates (see the Ising simulation lesson). Runs are short to fit in a browser; the statistical noise is therefore visible, and part of the exercise is to make it smaller.

Python Binder cumulant U_L(T) for L = 8, 12, 16

Press Run (or Ctrl + Enter) to see the result here.

Program input · what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

Expect curves that start near 0.65 at low T (the ordered limit is 2/3), fall toward 0 at high T, and fan out as L increases: larger lattices fall faster, so the curves cross near T_c ≈ 2.27. With this few sweeps and sizes the crossing point is imprecise (the noise is on the order of a few hundredths in U). For sharper results, raise n_meas, add more temperatures between 2.2 and 2.35 and use larger L; the Binder crossing for small sizes also moves slightly because of corrections to scaling.

Step 2: susceptibility at the critical temperature

At T = T_c the susceptibility per spin is χ = N(⟨m²⟩ - ⟨|m|⟩²)/T with N = L². The prediction is χ ∝ L^(7/4), so a log-log plot against L should have slope near 1.75. The code measures χ for L = 8, 12, 16, 24 and fits the slope.

Python Estimating γ/ν from the susceptibility at T_c

Press Run (or Ctrl + Enter) to see the result here.

Program input · what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

With the fixed seed above, the fit comes out near 1.9, somewhat above the exact 1.75. That is typical for runs this short: the data are noisy and corrections to scaling exist at small L. A result within about 0.15 of 1.75 is a good outcome, and more sweeps (or averaging several seeds) pulls it closer.

Step 3: extensions

  1. Average each run over several independent seeds, and use the scatter to put error bars on U_L and on the fitted γ/ν.
  2. Use L = 8, 16, 32 near the crossing and estimate T_c by the intersection of U_8 and U_16, U_16 and U_32. How does the estimate move as L grows?
  3. Estimate the exponent 1/ν from the slope of U_L at the crossing: dU/dT ∝ L^(1/ν) with exact ν = 1.
  4. Measure the autocorrelation time of m² at T_c for each L, and see critical slowing down.

Check your understanding

Quick check · 4 questions

of 4 answered You scored / 4 · perfect!

  1. 1. What property of the Binder cumulant makes it useful for locating T_c?

    U_L has no leading power of L in its scaling form, so different sizes cross at the critical point.
  2. 2. At T_c the susceptibility of an L × L Ising lattice scales as

    χ ∝ L^(γ/ν) with γ = 7/4 and ν = 1.
  3. 3. In the ordered phase the Binder cumulant of a large system tends to

    m = ±m₀ gives ⟨m⁴⟩ = m₀⁴ and ⟨m²⟩² = m₀⁴, so U = 1 - 1/3 = 2/3.
  4. 4. Why is a log-log plot natural for finite-size scaling?

    χ ∝ L^(γ/ν) becomes log χ = (γ/ν) log L + const.

Summary

  • Near T_c, observables depend on L/ξ: χ ~ L^(γ/ν) at T_c, and the Binder cumulant U_L = 1 - ⟨m⁴⟩/(3⟨m²⟩²) is size-independent at T_c.
  • Crossing of U_L curves estimates T_c; the slope of log χ against log L estimates γ/ν.
  • Short runs give rough numbers; the exact 2D values are T_c = 2.2692 J/k and γ/ν = 7/4.

Next: Project: A Hard-Disk Gas and the Maxwell-Boltzmann Distribution