Lattice Gas Model and Ising Mapping

The lattice gas is the simplest microscopic model of a fluid that still condenses. Space is chopped into cells, each either empty or holding a single particle — a crude hard core that forbids two particles in one cell — and neighbouring particles attract with energy $-\varepsilon$ . Despite the caricature, it captures the essential competition behind the liquid–gas transition: entropy favours dispersal, energy favours clustering, and below a critical temperature the system splits into coexisting dense and dilute phases. The central claim of this article is stronger than analogy: the lattice gas is the Ising model, wearing different clothes. Writing the occupation number $n_i\in\{0,1\}$ as $n_i=(1+\sigma_i)/2$ with $\sigma_i=\pm1$ turns the grand-canonical lattice-gas Hamiltonian, term by term, into the Ising Hamiltonian in a magnetic field — with an exchange coupling $J$ fixed by the attraction and a field $h$ fixed by the chemical potential plus a shift from the interaction. Every thermodynamic quantity transcribes: pressure to free energy, density to magnetization, compressibility to susceptibility. Because the correspondence is exact at the level of partition functions, every Ising result imports wholesale. This is why the liquid–gas critical point and the Curie point of a ferromagnet share the same universal behaviour. The fluid's coexistence curve is the image of the spontaneous-magnetization curve $\pm m(T)$ ; the spin-flip symmetry $\sigma_i\to-\sigma_i$ at $h=0$ becomes the particle–hole symmetry of the gas at half filling. Microscopic details — a magnet, a fluid, an alloy — wash out near criticality, leaving one set of exponents. The one-dimensional transfer matrix gives an exactly soluble lattice gas that never condenses; the Onsager solution gives the exact 2D coexistence curve; mean-field theory reproduces a van der Waals loop; and the Peierls argument proves condensation for $d\ge2$ . The lattice gas is the bridge carrying the whole machinery of Ising statistical mechanics into the theory of fluids and alloys.

Key takeaways

The lattice gas places at most one particle per site with nearest-neighbour attraction; its grand partition function is identical, term by term, to the Ising model in a field.
The dictionary is $n_i=(1+\sigma_i)/2$ : occupied $\leftrightarrow$ spin up, empty $\leftrightarrow$ spin down. Attraction $\varepsilon$ maps to ferromagnetic coupling $J=\varepsilon/4$ .
The chemical potential maps to the field, $h=\tfrac12(\mu+z\varepsilon/2)$ ; the field carries a $z\varepsilon$ shift, so half filling ( $\rho=1/2$ ) is $h=0$ , not $\mu=0$ .
Below $T_c$ the gas phase-separates into liquid and gas exactly as the magnet splits into up/down; the density gap is $\rho_l-\rho_g=m$ (the magnetization itself, not $2m$ — a factor the halved density range demands).
This puts the liquid–gas critical point in the Ising universality class, with the compressibility diverging as the magnetic susceptibility.
Key idea: The affine change of variables $n_i=(1+\sigma_i)/2$ maps the grand-canonical lattice gas exactly onto the Ising model with $J=\varepsilon/4$ and $h=\tfrac12(\mu+z\varepsilon/2)$ — liquid–gas physics is Ising physics.

The core insight

Occupation numbers and Ising spins are related by the affine change of variables $n_i=(1+\sigma_i)/2$ ; substituting it into the grand-canonical lattice-gas Hamiltonian reproduces the Ising Hamiltonian exactly, with $J=\varepsilon/4$ and $h=\tfrac12(\mu+z\varepsilon/2)$ where $z$ is the coordination number. Liquid–gas physics is therefore Ising physics.

Where this lives today

The lattice-gas mapping remains the standard bridge between fluid and magnetic criticality, and it powers a great deal of computation. Grand-canonical Monte Carlo simulations of condensation are, under the hood, Ising simulations: single-site (Glauber) updates model particle exchange with a reservoir, while particle-conserving (Kawasaki) spin-exchange dynamics model diffusion-limited phase separation and spinodal decomposition. The same Hamiltonian, reinterpreted with $n_i$ as "site occupied by species A," describes binary-alloy order–disorder transitions such as that of $\beta$ -brass. And the lattice-gas cellular automata of the late 1980s — discrete particles hopping on a lattice — reproduce the Navier–Stokes equations in the hydrodynamic limit, an idea that evolved into the lattice-Boltzmann methods now used across engineering fluid dynamics. From critical universality to industrial CFD, the through-line is that occupation numbers and spins are the same degrees of freedom.

The physical picture

Picture a chessboard of cells, each holding a ball or nothing. Two neighbouring balls lower the energy by $\varepsilon$ , mimicking the attractive tail of an intermolecular potential, while the one-ball-per-cell rule mimics hard-core repulsion. Now recolour: a filled cell is a black square, an empty cell white. The energetics of black–white patterns is exactly that of a ferromagnet that prefers like-coloured neighbours. At half filling the board has perfect black–white symmetry — swapping the colours changes nothing. This is particle–hole symmetry, the fluid's version of spin-flip symmetry $\sigma\to-\sigma$ at zero field. Below $T_c$ the symmetry spontaneously breaks: the board develops a black-rich (liquid) region and a white-rich (gas) region, just as a magnet picks up or down.

Figure

The lattice gas ↔ Ising dictionary: occupied cell = spin up, empty = spin down, attraction = ferromagnetic bond

The chemical potential is a knob biasing the board toward more or fewer balls — a magnetic field favouring one colour. Raising $\mu$ past the coexistence value at $T<T_c$ flips the whole board from gas-like to liquid-like, the discontinuous first-order condensation, exactly mirroring the magnetization jump as $h$ crosses zero. Because only the pattern statistics matter near criticality, a fluid, a magnet, and a binary alloy become indistinguishable: the lattice gas is the Rosetta stone translating between them.

Figure

Coexistence curve = spontaneous-magnetization curve: liquid and gas densities are the mirror images ½(1±m)

The core equations

The lattice-gas Hamiltonian is grand-canonical: particles are exchanged with a reservoir at chemical potential $\mu$ , and nearest neighbours attract.

\mathcal{H}_{\text{LG}}=-\varepsilon\sum_{\langle ij\rangle}n_in_j-\mu\sum_i n_i,\qquad n_i\in\{0,1\} \qquad (1)

where: $n_i$ is the occupation of site $i$ (0 empty, 1 occupied); $\varepsilon>0$ the nearest-neighbour attraction; $\mu$ the chemical potential; $\langle ij\rangle$ the sum over distinct bonds. Physically: $-\varepsilon n_in_j$ rewards adjacent particles, $-\mu n_i$ controls the mean density. The dictionary that unlocks everything is the unique order-preserving affine bijection between the occupation set $\{0,1\}$ and the spin set $\{-1,+1\}$ .

n_i=\frac{1+\sigma_i}{2},\qquad \sigma_i=2n_i-1\in\{-1,+1\} \qquad (2)

where: $\sigma_i=+1$ for an occupied site, $-1$ for empty. Physically: this is dimensionless and invertible, so it can be substituted directly into the Hamiltonian without any Jacobian — the configuration sum is a relabelling. Substituting the dictionary and collecting terms gives the Ising coupling and field. The coupling is a quarter of the attraction; the field mixes the chemical potential with a shift from the interaction.

J=\frac{\varepsilon}{4},\qquad h=\frac{\mu}{2}+\frac{z\varepsilon}{4}=\frac12\Big(\mu+\frac{z\varepsilon}{2}\Big) \qquad (3)

where: $J$ is the (ferromagnetic) Ising exchange, $h$ the Ising field (energy units), $z$ the coordination number. Physically: the $z\varepsilon$ piece is why $h=0$ is not $\mu=0$ — the mean interaction of a filled site with its $z$ neighbours acts like an internal field. The result is the Ising Hamiltonian plus a configuration-independent constant, which shifts the reference pressure but no phase boundary.

\mathcal{H}_{\text{LG}}=-J\sum_{\langle ij\rangle}\sigma_i\sigma_j-h\sum_i\sigma_i+E_0,\qquad E_0=-\frac{Nz\varepsilon}{8}-\frac{N\mu}{2} \qquad (4)

where: $N$ is the number of sites and $E_0$ the additive constant. Physically: $E_0$ is absorbed into the reference free energy; the physics lives entirely in the $\sigma\sigma$ and $\sigma$ terms. Densities and magnetizations are linearly related, which compresses the spin range $[-1,1]$ into the density range $[0,1]$ and fixes the critical density at one half.

\rho=\frac{1+m}{2},\qquad \rho_c=\tfrac12,\qquad m=\langle\sigma_i\rangle,\ \ \rho=\langle n_i\rangle \qquad (5)

where: $\rho$ is the density, $m$ the magnetization per spin. Physically: $m=0$ (disordered) maps to $\rho=1/2$ (the symmetric fluid), $m=\pm1$ to the empty/full limits. At coexistence ( $h=0$ , i.e. $\mu=-z\varepsilon/2$ ) the two spontaneous magnetizations $\pm m$ give the two coexisting densities; the density gap equals $m$ , not $2m$ — the halved range costs a factor of two relative to the magnetization gap.

\rho_{l,g}=\tfrac12\big(1\pm m(T)\big),\qquad \rho_l-\rho_g=m(T),\qquad m_+-m_-=2m(T) \qquad (6)

where: $\rho_l,\rho_g$ are the liquid and gas coexistence densities. Physically: the density gap $\rho_l-\rho_g=m$ inherits the exponent $\beta$ of the magnetization; the two are equal because $\rho_l-\rho_g=\tfrac12[(1+m)-(1-m)]=m$ . Finally, the exact square-lattice critical temperature transcribes Onsager's Ising result into fluid units through $J=\varepsilon/4$ .

k_BT_c=\frac{2J}{\ln(1+\sqrt2)}=\frac{\varepsilon}{2\ln(1+\sqrt2)}\approx0.567\,\varepsilon \qquad (7)

where: $T_c$ is the critical temperature of the square-lattice lattice gas. Physically: the same number that fixes a 2D magnet's Curie point fixes a 2D fluid's condensation point.

Derivation AExact reduction of the lattice gas to the Ising model

The goal is to substitute the affine dictionary $n_i=(1+\sigma_i)/2$ into the grand-canonical lattice-gas Hamiltonian and collect terms, showing that the bilinear piece becomes an Ising exchange with $J=\varepsilon/4$ , the linear piece an Ising field with $h=\mu/2+z\varepsilon/4$ , and the remainder a configuration-independent constant. The strategy is careful bond and site accounting: on a lattice of $N$ sites with coordination $z$ there are $Nz/2$ bonds, and in the sum $\sum_{\langle ij\rangle}(\sigma_i+\sigma_j)$ each site appears in $z$ bonds. This bookkeeping is exactly where the $z\varepsilon$ field shift comes from, so it is worth doing explicitly.

Step 1. Expand the interaction: $n_in_j=\tfrac14(1+\sigma_i)(1+\sigma_j)=\tfrac14(1+\sigma_i+\sigma_j+\sigma_i\sigma_j)$ . Each bond splits into a spin–spin term, two single-spin terms, and a constant. Why: the affine map turns a product of occupations into a product of $(1+\sigma)$ factors, whose expansion is the whole content of the mapping.

Step 2. Sum over the $Nz/2$ bonds. The constant gives $\tfrac14\cdot\tfrac{Nz}{2}$ . In $\sum_{\langle ij\rangle}(\sigma_i+\sigma_j)$ each site appears in $z$ bonds, so this equals $z\sum_i\sigma_i$ , contributing $\tfrac{z}{4}\sum_i\sigma_i$ . The bilinear part is $\tfrac14\sum_{\langle ij\rangle}\sigma_i\sigma_j$ . Why: counting how many bonds touch each site is what converts a bond sum of single spins into a site sum.

Step 3. Expand the chemical-potential term: $\sum_i n_i=\tfrac12\sum_i(1+\sigma_i)=\tfrac{N}{2}+\tfrac12\sum_i\sigma_i$ , a constant plus a uniform-field piece. Why: the same $\tfrac12(1+\sigma)$ substitution applied to the linear term.

Step 4. Collect. The bilinear term $-\varepsilon\cdot\tfrac14\sum\sigma_i\sigma_j=-J\sum\sigma_i\sigma_j$ gives $J=\varepsilon/4$ (attraction $\to$ ferromagnetic). The field term $-(\tfrac{z\varepsilon}{4}+\tfrac{\mu}{2})\sum_i\sigma_i=-h\sum_i\sigma_i$ gives $h=\tfrac{\mu}{2}+\tfrac{z\varepsilon}{4}$ . The leftover $E_0=-\tfrac{Nz\varepsilon}{8}-\tfrac{N\mu}{2}$ is configuration-independent. Why: both the chemical potential and the interaction's mean contribute to the field — this is the origin of the shift. *Verified: verify_aq-sm-lattice-gas-ising-mapping.py performs the substitution symbolically, matches coefficients to confirm $J=\varepsilon/4$ , $h=\mu/2+z\varepsilon/4$ , $E_0=-Nz\varepsilon/8-N\mu/2$ , and reproduces $k_BT_c=2J/\ln(1+\sqrt2)$ .*

Go deeper · Derivation B · 1 minCoexistence curve from spontaneous magnetization, and the density-gap factor

Derivation BCoexistence curve from spontaneous magnetization, and the density-gap factor

The goal is to translate the Ising coexistence physics at $h=0$ into fluid variables and to pin down the exact size of the density gap, correcting a factor-of-two that is easy to get wrong. The strategy is to note that phase coexistence in the magnet sits exactly on the line $h=0$ , where up and down are degenerate; mapping $h=0$ to $\mu=-z\varepsilon/2$ and the two magnetizations $\pm m$ to densities $\tfrac12(1\pm m)$ gives the liquid and gas branches, the rectilinear diameter, and the critical exponent. The subtlety is that the density range is half the spin range, so the density gap is $m$ , not $2m$ .

Step 1. Locate coexistence: $h=0\Rightarrow\mu_{\text{coex}}=-z\varepsilon/2$ . The fluid coexists at this one chemical potential for every $T<T_c$ — a vertical line in the $(\mu,T)$ plane, image of the $h=0$ axis. Why: the up/down degeneracy of the magnet is the liquid/gas degeneracy of the fluid.

Step 2. The two densities: $m=+m(T)\Rightarrow\rho_l=\tfrac12(1+m)$ and $m=-m(T)\Rightarrow\rho_g=\tfrac12(1-m)$ , mirror images about $\rho=1/2$ . Why: the two symmetry-broken magnetized states map to the dense and dilute phases.

Step 3. The rectilinear diameter is exactly horizontal: $(\rho_l+\rho_g)/2=\tfrac12=\rho_c$ , a consequence of exact particle–hole symmetry (real fluids show a slight tilt because that symmetry is only approximate). The density gap is $\rho_l-\rho_g=\tfrac12[(1+m)-(1-m)]=m$ — the magnetization itself, not $2m$ ; the magnetization gap $m_+-m_-=2m$ is halved by the $\tfrac12$ in $\rho=(1+m)/2$ . Why: densities live on $[0,1]$ , half the spin interval $[-1,1]$ , so every spin difference is halved when read as a density difference.

Step 4. The critical exponent is inherited: $\rho_l-\rho_g=m(T)\sim(T_c-T)^{\beta}$ , with $\beta=1/8$ exactly in 2D, $\beta\approx0.326$ in 3D, $\beta=1/2$ in mean field. Why: the density gap is proportional to the magnetization, so it vanishes with the same power. *Verified: verify_aq-sm-lattice-gas-ising-mapping.py confirms $\rho_l-\rho_g=m$ (and flags that the base's $2m$ was wrong), $\mu_{\text{coex}}=-z\varepsilon/2$ , $\rho_c=1/2$ , and $\beta$ values $1/8$ (2D), $0.326$ (3D).*

Limiting cases

The mapping is exact at the level of partition functions; the physical regimes correspond as follows.

Case	Result	Meaning
Half filling $\rho=1/2$	$h=0,\ \mu=-z\varepsilon/2$	The particle–hole symmetric line; below $T_c$ it is the coexistence diameter where liquid and gas have equal weight, image of the spontaneous-magnetization line.
Dilute gas $\rho\to0$	$h\to-\infty,\ m\to-1$	A strong negative field saturates spins down; large negative $\mu$ , the ideal-gas / virial limit.
Close-packed liquid $\rho\to1$	$h\to+\infty,\ m\to+1$	Spins saturate up; every cell occupied, the incompressible dense limit.
One dimension	$T_c=0$	The 1D Ising model has no finite- $T$ order, so the 1D lattice gas never condenses at $T>0$ .
High temperature $T\to\infty$	$\rho\to\dfrac{e^{\beta\mu}}{1+e^{\beta\mu}}$	Interactions negligible; each site is an independent two-state (hard-core) system, a Fermi–Dirac-like per-site occupancy.

Common trapThe field is not just μ — and the density gap is m, not 2m

Two traps live here. First, the magnetic field is NOT simply $h\propto\mu$ : it carries a $z\varepsilon$ shift, so $h=0$ (the symmetric coexistence line) is a specific subcritical chemical potential $\mu=-z\varepsilon/2$ , not $\mu=0$ . Forgetting the shift misplaces the coexistence curve. Second, because $\rho=(1+m)/2$ compresses the spin range $[-1,1]$ into the density range $[0,1]$ , the coexistence density gap is $\rho_l-\rho_g=m$ — the magnetization itself — while the magnetization gap is $m_+-m_-=2m$ . Writing the density gap as $2m$ double-counts the factor of two; both quantities carry the same exponent $\beta$ , but they differ by exactly a factor of two.

Worked examples

Example 1 — reading the coupling and field off the pair energies. For a square lattice ( $z=4$ ) with attraction $\varepsilon$ and chemical potential $\mu$ , find $J$ , $h$ , and the coexistence chemical potential.

Step 1. Coupling: $J=\varepsilon/4$ . If we choose $\varepsilon=4J$ then $J=J$ , the natural normalization used below. The attraction maps to a ferromagnetic bond one quarter its strength.

Step 2. Field: $h=\mu/2+z\varepsilon/4=\mu/2+4\varepsilon/4=\mu/2+\varepsilon$ . Setting $h=0$ gives the coexistence chemical potential $\mu_{\text{coex}}=-2\varepsilon=-z\varepsilon/2$ . The symmetric coexistence line sits at $\mu=-2\varepsilon$ , decidedly not at $\mu=0$ .

Step 3. Sanity check at half filling: at $\mu=-2\varepsilon$ and $T>T_c$ the density is $\rho=1/2$ (i.e. $m=0$ ), the critical density. Below $T_c$ the two branches open as $\rho_{l,g}=\tfrac12(1\pm m)$ . Everything is anchored to $\rho_c=1/2$ .

Example 2 — the 2D critical temperature and the density gap, with the base-error factor corrected. Compute the square-lattice condensation temperature in units of $\varepsilon$ , and the density gap just below it.

Step 1. Onsager's Ising value is $k_BT_c=2J/\ln(1+\sqrt2)$ . With $J=\varepsilon/4$ : $k_BT_c=\dfrac{\varepsilon}{2\ln(1+\sqrt2)}=\dfrac{\varepsilon}{2(0.8814)}=\mathbf{0.567\,\varepsilon}$ (equivalently $2.269\,J$ ).

Step 2. Just below $T_c$ the magnetization grows as $m\sim(T_c-T)^{1/8}$ (2D). The coexisting densities are $\rho_l=\tfrac12(1+m)$ and $\rho_g=\tfrac12(1-m)$ , so the density gap is $\rho_l-\rho_g=\tfrac12[(1+m)-(1-m)]=\mathbf{m}$ .

Step 3. This is the correction to a common slip: the magnetization gap is $m_+-m_-=2m$ , but the density gap is only $m$ , because the density range is half the spin range. For example if $m=0.6$ at some $T<T_c$ , then $\rho_l=0.8$ , $\rho_g=0.2$ , and $\rho_l-\rho_g=\mathbf{0.6}=m$ , not $1.2=2m$ . A draft that writes $\rho_l-\rho_g=2m$ would predict densities outside $[0,1]$ .

Origins · 1 minlattice gas and how this was discovered

lattice gas

introduced 1952 (Lee & Yang) · The lattice gas as a rigorous model of condensation was placed on its modern footing by T. D. Lee and C. N. Yang in 1952, in the second of their two celebrated papers on the statistical theory of phase transitions. Where the first paper proved the famous circle theorem about the zeros of the grand partition function, the second showed explicitly that the lattice gas is mathematically identical to the Ising model in a magnetic field, so that the entire machinery developed for magnetism — including Onsager's 1944 exact solution — could be imported directly into the theory of fluids. The idea itself has older roots in the cell models of Cernuschi and Eyring and in the order–disorder theories of alloys, but it was the Lee–Yang mapping that made the correspondence exact and general. The concept is now foundational: it underlies the universality connecting the liquid–gas critical point to the ferromagnetic Curie point, the lattice-gas cellular automata used to simulate hydrodynamics, and the grand-canonical Monte Carlo methods of modern computational fluid statistical mechanics.

History

1937

— Andrews's older liquid–gas critical-point data reinterpreted through lattice models

1944

— Onsager solves the 2D Ising model, later imported into fluid language

1952

— Lee and Yang prove the lattice gas is exactly the Ising model in a field

1952

— the Lee–Yang circle theorem constrains the grand-partition-function zeros

1958

— Yang and Lee's results extended to the general equivalence of fluids and magnets

1970

— universality of the liquid–gas and magnetic critical points confirmed experimentally

1986

— Frisch, Hasslacher and Pomeau launch lattice-gas automata for hydrodynamics

Going further

References. Pathria & Beale, Statistical Mechanics, 3rd ed., §12.1–12.4 (Lattice gas and Ising correspondence) · Kardar, Statistical Physics of Particles, Ch. 7; Statistical Physics of Fields, Ch. 1–2 · Reif, Fundamentals of Statistical and Thermal Physics, Ch. 10 · Sethna, Statistical Mechanics: Entropy, Order Parameters, and Complexity, Ch. 8 · Landau & Lifshitz, Statistical Physics Part 1, §150–151 · T. D. Lee & C. N. Yang, Phys. Rev. 87, 410 (1952) — statistical theory of the lattice gas and Ising model.

Provenance. The mapping algebra (eqCards 1–6 and Derivation 1) is the standard lattice-gas/Ising correspondence; verify_aq-sm-lattice-gas-ising-mapping.py recomputes the substitution $n_i=(1+\sigma_i)/2$ and confirms $J=\varepsilon/4$ , $h=\mu/2+z\varepsilon/4$ , $E_0=-Nz\varepsilon/8-N\mu/2$ against Pathria & Beale §12.1 and Kardar (Particles) Ch. 7. Base correction: the migrated draft wrote the coexistence density gap as $\rho_l-\rho_g=m_+-m_-=2m$ ; with $\rho=(1+m)/2$ the density gap is actually $\rho_l-\rho_g=m$ , while $m_+-m_-=2m$ is the magnetization gap. The verify script proves $\rho_l-\rho_g=m$ and $\rho_l-\rho_g=\tfrac12(m_+-m_-)$ ; the corrected relation is used throughout and flagged in the pitfall and cheat sheet. Exponents $\beta=1/8$ (2D) and $k_BT_c=2J/\ln(1+\sqrt2)\approx2.269\,J\approx0.567\,\varepsilon$ are Onsager's exact values; the 3D values $\beta\approx0.326,\gamma\approx1.24,\delta\approx4.8$ are standard high-precision Ising results. No fabricated numbers.

Builds on: The Ising Model: Definition and Hamiltonian · Order Parameter and Spontaneous Magnetization · Mean-Field (Weiss) Theory of the Ising Model. Leads to: Binary Alloy Order-Disorder Transition · Onsager Solution of the Two-Dimensional Ising Model · Peierls Argument for Long-Range Order.

Is the lattice gas the same model as the Ising magnet, or just similar?

It is exactly the same model, not merely similar. The substitution $n_i=(1+\sigma_i)/2$ is a change of variables, not an approximation, so the two partition functions are identical term by term once you match $J=\varepsilon/4$ , $h=\mu/2+z\varepsilon/4$ , and absorb a configuration-independent constant. Every Ising result — the 1D transfer-matrix solution, Onsager's 2D solution, mean-field theory, the Peierls argument — therefore applies to the lattice gas without any new work. This is the deepest content of the correspondence: a magnet and a fluid are the same statistical-mechanics problem in different variables.

Why does the field h contain a term with the interaction ε, not just μ?

Because both the chemical potential and the mean interaction act like a uniform bias on the spins. When you expand $n_in_j=\tfrac14(1+\sigma_i+\sigma_j+\sigma_i\sigma_j)$ , the single-spin terms $\sigma_i+\sigma_j$ summed over all bonds give $z\sum_i\sigma_i$ — the interaction contributes an effective field proportional to $z\varepsilon$ . Adding the chemical-potential piece $\mu/2$ gives the total field $h=\mu/2+z\varepsilon/4$ . Physically, a filled site is surrounded on average by other filled sites, and that internal attraction shifts the effective field away from the naive $\mu/2$ .

What exactly is the density gap at coexistence — m or 2m?

It is $m$ , the magnetization itself. Below $T_c$ the two phases correspond to $m=+m(T)$ and $m=-m(T)$ , so the magnetization gap is $m_+-m_-=2m$ . But densities are $\rho=(1+m)/2$ , which squeezes the spin interval $[-1,1]$ into $[0,1]$ , halving every difference: $\rho_l-\rho_g=\tfrac12[(1+m)-(1-m)]=m$ . So the density gap is exactly half the magnetization gap. Both carry the same critical exponent $\beta$ , but numerically the density gap is $m$ , and writing it as $2m$ would predict densities outside the physical range $[0,1]$ .

Why is the liquid–gas critical point in the same universality class as a magnet?

Because near $T_c$ only the long-wavelength statistics of the order parameter matter, and the lattice gas has exactly the same order parameter as the Ising magnet — the density deviation $\rho-\rho_c$ maps to the magnetization $m$ . Both have a single scalar order parameter with Ising ( $\mathbb{Z}_2$ ) symmetry, both live in the same spatial dimension, and both have short-range interactions. The renormalization group says these three features alone fix the universality class, so the compressibility of the fluid and the susceptibility of the magnet diverge with the same exponent $\gamma$ , and the coexistence curve and spontaneous magnetization vanish with the same $\beta$ .

What breaks the exact particle–hole symmetry in real fluids?

On the lattice the symmetry $n_i\to1-n_i$ (particle $\leftrightarrow$ hole) is exact, which forces a perfectly horizontal rectilinear diameter at $\rho_c=1/2$ . Real fluids are not lattice gases: molecules have finite size, the attraction has a range and angular dependence, and three-body forces exist. These break the exact symmetry, so the measured coexistence diameter is slightly tilted (the "law of the rectilinear diameter" is only approximate) and the critical density is not exactly at the midpoint. The universality class survives — the exponents are still Ising — but the exact half-filling symmetry is an artifact of the idealized model.

How is antiferromagnetic coupling relevant here?

If the effective coupling is antiferromagnetic ( $J<0$ ), the mapped system prefers alternating occupation rather than clustering — the lattice-gas analog is a sublattice-ordered structure, like the CsCl arrangement in an ordered binary alloy, rather than liquid–gas phase separation. So the sign of $J$ selects between two physically different transitions: ferromagnetic $J>0$ gives condensation / phase separation, antiferromagnetic $J<0$ gives an order–disorder (superlattice) transition. The binary-alloy article develops the antiferromagnetic case, where the Ising magnetization becomes the long-range chemical order parameter.