Antiferromagnetism and Frustration on Lattices

Cool a ferromagnet and every spin agrees with its neighbours; cool an antiferromagnet on a square lattice and every spin disagrees with its neighbours — a perfect checkerboard of up and down. Both order, and their thermodynamics is literally identical: flip every spin on one sublattice and the antiferromagnet becomes a ferromagnet, so it inherits the same free energy, the same critical temperature, the same specific heat. Only the name of the order parameter changes, from the uniform magnetization to the staggered magnetization that measures the checkerboard. Now put the same antiferromagnetic bonds on a triangle. Two spins can disagree, but the third is trapped: it cannot simultaneously oppose both of its neighbours. At least one bond must be violated no matter how you arrange the spins. This is geometric frustration — a conflict written into the lattice, not into any configuration. Its consequences are dramatic. The triangular Ising antiferromagnet has no ordering transition at all: it stays disordered down to absolute zero. And because so many arrangements tie for the lowest energy, it keeps a macroscopic ground-state entropy $s_0/k_B \approx 0.3231$ per spin even at $T=0$ — a residual entropy that a naive reading of the third law would forbid. Frustration is diagnosed loop by loop. A closed loop of bonds is frustrated when the product of the bond signs is negative, and this product is invariant under flipping any spin, so frustration is a fixed, gauge-invariant property of the lattice and its couplings. Whether antiferromagnetism orders is therefore decided by geometry: two-colourable (bipartite) lattices are secretly ferromagnets, while odd loops inject a frustration that can suppress the ordering temperature, degenerate the ground state, or destroy long-range order entirely — the seed of spin ice, spin liquids, and spin glasses.

Key takeaways

Antiferromagnetic (AFM) coupling $J<0$ in $H=-J\sum\sigma_i\sigma_j$ favours antiparallel neighbours; on a bipartite lattice the gauge flip $\sigma_i\to-\sigma_i$ on one sublattice turns the AFM into a ferromagnet, so thermodynamics is identical and only the order parameter changes.
The AFM order parameter on a bipartite lattice is the staggered magnetization $m_s=\frac1N\sum_i(-1)^i\langle\sigma_i\rangle$ , nonzero in the Néel state.
On a non-bipartite lattice (odd loops) the spins cannot satisfy every bond at once: this is geometric frustration, quantified per loop by $\Phi_p=\prod\mathrm{sgn}(J_{ij})$ , and it is gauge-invariant — it cannot be flipped away.
One AFM triangle has $6$ ground states (of $8$ ), $E_\text{min}=-|J|$ with exactly one frustrated bond, and $\langle\sigma_i\sigma_j\rangle_\text{nn}=-\tfrac13$ even at $T=0$ .
The full triangular Ising AFM never orders ( $T_c=0$ ) and keeps Wannier's residual entropy $s_0/k_B=0.3231$ per spin — a macroscopic $T=0$ degeneracy, echoed by the measured Pauling entropy of spin ice.
Key idea: whether antiferromagnetism orders is decided by lattice geometry — bipartite lattices unfrustrate (Néel order, same thermodynamics as the ferromagnet), while odd loops frustrate, killing finite- $T$ order and leaving macroscopic ground-state degeneracy and residual entropy.

The core insight

Whether antiferromagnetism orders is decided by lattice geometry: bipartite lattices unfrustrate (Néel order, identical thermodynamics to the ferromagnet), while odd loops frustrate, killing finite- $T$ order and producing macroscopic ground-state degeneracy and residual entropy.

Where this lives today

Geometric frustration is now a central organising theme of condensed-matter physics. Spin ices such as Dy $_2$ Ti $_2$ O $_7$ obey an "ice rule" mathematically identical to Pauling's water-ice problem, and their measured $T=0$ entropy matches Pauling's $\tfrac12\ln(3/2)$ — a laboratory confirmation that frustration produces macroscopic ground-state degeneracy. Frustrated magnets on the kagome and pyrochlore lattices are the leading candidates for quantum spin liquids, exotic phases with long-range entanglement and no symmetry breaking even at $T=0$ . Frustration also drives the hardest problems in computation. Adding bond disorder to a frustrated Ising model gives the Edwards–Anderson spin glass, whose ground state is NP-hard to find; the frustration function $\Phi_p$ marks the plaquettes carrying the irreducible energy cost and defines the rugged landscape that simulated annealing and quantum-annealing hardware are benchmarked against. The little triangle that cannot make up its mind is, quite literally, a template for a whole class of optimisation problems.

The physical picture

Picture three friends around a triangle, each determined to disagree with both neighbours. Two of them can disagree with each other, but the third is cornered — it must agree with someone. No rearrangement removes the conflict, because the conflict lives in the triangle itself, not in any particular seating. That is geometric frustration: a constraint the lattice makes impossible to satisfy globally, because the loop has odd length. Frustration trades a unique ground state for a vast plateau of equally good states. At $T\to0$ the frustrated magnet is not frozen into one configuration; it is free to wander a macroscopic set of tied arrangements, so it keeps entropy $S_0 \sim N k_B s_0$ even at absolute zero — like a crowd that never settles because every seating is equally imperfect. The sign-flip (gauge) view is clarifying. Flipping a spin reshuffles which bonds count as "satisfied," but it never changes whether a given loop is frustrated. Frustration is an intrinsic, gauge-invariant property of the lattice and its bond signs, not of any snapshot of the spins — which is exactly why you cannot cleverly relax it away. Bipartite versus non-bipartite is the whole story. If the lattice graph is two-colourable — no odd cycles, as for the square, honeycomb, or simple-cubic lattice — the antiferromagnet is secretly a ferromagnet, and it orders in the ordinary way. A single odd cycle anywhere injects frustration and can destabilise order entirely.

Figure

Frustration on a triangle: one bond must always lose

Figure

Residual entropy: the triangular AFM keeps disorder at T = 0

The second figure is the punchline: an ordinary (bipartite) antiferromagnet orders and reaches zero entropy, but the frustrated triangular antiferromagnet keeps a finite $s_0/k_B = 0.3231$ down to $T=0$ — a residual entropy sitting well below the fully disordered $\ln 2$ but sharply above zero. The measured Pauling entropy of spin ice is the same phenomenon in three dimensions.

The core equations

The Ising energy is a sum over bonds; the coupling sign on each bond decides whether that bond wants its spins parallel or antiparallel. We keep the general form so that antiferromagnets, ferromagnets, and mixed-sign (frustrated) systems are all special cases.

H = -\sum_{\langle ij\rangle} J_{ij}\,\sigma_i\sigma_j,\qquad \sigma_i=\pm1,\quad J_{ij}<0\ \text{(AFM)} \tag{1}

where: $J_{ij}$ — exchange coupling on bond $\langle ij\rangle$ ; $J>0$ ferromagnetic (favours parallel), $J<0$ antiferromagnetic (favours antiparallel); $\sigma_i$ — Ising spin at site $i$ ; $\langle ij\rangle$ — sum over nearest-neighbour bonds. The J-sign convention is inherited from The Ising Model: Definition and Hamiltonian by name: with $J<0$ , a satisfied (antiparallel) bond has energy $-J\sigma_i\sigma_j = -|J|$ . Whether a loop of bonds can be fully satisfied is captured by a single sign — the product of the coupling signs around it. This is Toulouse's frustration function, and it is the object that survives every spin flip.

\Phi_p \equiv \prod_{\langle ij\rangle\in p}\mathrm{sgn}(J_{ij}),\qquad \Phi_p=-1 \Rightarrow \text{plaquette frustrated} \tag{2}

where: $\Phi_p$ — product of bond signs around an elementary loop $p$ (the Toulouse frustration function); $p$ — an elementary plaquette of the lattice. A triangle of three AFM bonds has $\Phi_p=(-1)^3=-1$ : frustrated. A square of four AFM bonds has $\Phi_p=(+1)$ : not frustrated (it is bipartite). On a bipartite lattice the natural antiferromagnetic order parameter is not the uniform magnetization (which averages to zero) but the staggered magnetization, which weights the two sublattices with opposite sign. It is inherited by name from Order Parameter and Spontaneous Magnetization, adapted to the two-sublattice Néel pattern.

m_s = \frac{1}{N}\sum_i (-1)^{i}\,\langle \sigma_i\rangle,\qquad (-1)^i=\begin{cases}+1 & i\in A\\ -1 & i\in B\end{cases} \tag{3}

where: $m_s$ — staggered magnetization, nonzero in the Néel phase; $A,B$ — the two sublattices of a bipartite lattice; $N$ — number of sites. In the perfect Néel state $\langle\sigma_i\rangle=+1$ on $A$ and $-1$ on $B$ , so $m_s=1$ . The bridge between the two-sublattice antiferromagnet and the ordinary ferromagnet is an exact gauge transformation: flip every spin on sublattice $B$ . Every bond runs $A\!\to\!B$ , so this converts each antiferromagnetic bond into a ferromagnetic one, leaving the partition function untouched.

\sigma_i\to -\sigma_i\ (i\in B):\quad H_{\rm AFM}=-J\!\sum_{\langle ij\rangle}\sigma_i\sigma_j \;\longmapsto\; H_{\rm FM}=-|J|\!\sum_{\langle ij\rangle}\sigma_i'\sigma_j' \tag{4}

where: the AFM has $J=-|J|$ , and after the flip the coupling is $+|J|$ , i.e. ferromagnetic ( $H_{\rm FM}=-|J|\sum\sigma'\sigma'$ in the same $H=-J\sum\sigma\sigma$ convention). Because $Z$ is unchanged, the free energy, $T_c$ , and specific heat are identical to the ferromagnet; the ferromagnetic magnetization $\langle\sigma_i'\rangle$ becomes the AFM staggered magnetization $m_s$ . This map fails the instant an odd loop appears. The signature of frustration is a nonzero entropy at absolute zero. Wannier's exact solution of the triangular Ising antiferromagnet gives the residual entropy per spin as a lattice integral.

\frac{S_0}{N k_B} = s_0 = \frac{2}{\pi}\int_0^{\pi/3}\ln\!\big(2\cos x\big)\,\mathrm{d}x = 0.323066\ldots \tag{5}

where: $S_0$ — total ground-state entropy at $T=0$ ; $s_0$ — residual entropy per spin. The number sits between $0$ (a unique ordered ground state) and $\ln 2=0.6931$ (fully random spins), quantifying "strong but not total" disorder. The exact prefactor is $2/\pi$ and the upper limit is $\pi/3$ — a normalisation the AllQFT base got wrong twice (see the warning and the cheat sheet). Finally, naive two-sublattice mean-field theory linearises the self-consistent staggered field and predicts a Néel temperature. It is useful for bipartite antiferromagnets but silently wrong for frustrated ones.

k_B T_N^{\rm MF} = z\,|J|\,S(S+1)\big/3 \;\xrightarrow{\ S=1/2,\ \text{Ising}\ }\; z|J| \tag{6}

where: $z$ — coordination number (nearest neighbours); $T_N$ — Néel ordering temperature; $S$ — spin magnitude (the Ising limit gives $z|J|$ ). This assumes a globally consistent up/down sublattice pattern exists — true on the square lattice, false on the triangular one, where it spuriously predicts order.

Derivation ABipartite gauge map: an antiferromagnet is a hidden ferromagnet

We want to prove — not just assert — that on a bipartite lattice the antiferromagnet has exactly the same thermodynamics as the ferromagnet, and to see precisely why the argument collapses when an odd loop is present. The strategy is a change of variables (a $\mathbb{Z}_2$ gauge transformation) that flips every spin on one sublattice; because every nearest-neighbour bond crosses between sublattices, each bond's coupling changes sign, and the sum over spin configurations (hence $Z$ ) is untouched. Watch for the single load-bearing assumption — the absence of $A$ – $A$ or $B$ – $B$ bonds — which is exactly two-colourability.

Step 1. Split the sites of a bipartite lattice into two sublattices $A$ and $B$ such that every nearest-neighbour bond connects an $A$ site to a $B$ site. This is possible if and only if the lattice graph has no odd cycles (it is two-colourable): square, honeycomb, and simple-cubic all qualify.

Step 2. Introduce new spin variables $\sigma_i'=\sigma_i$ on $A$ and $\sigma_i'=-\sigma_i$ on $B$ . This is a relabelling of the summation dummy variables, so $\sum_{\{\sigma\}}=\sum_{\{\sigma'\}}$ and $Z$ is manifestly unchanged.

Step 3. Rewrite one bond. With $J=-|J|$ and $\sigma_j=-\sigma_j'$ (the $B$ end),

-J\,\sigma_i\sigma_j = -(-|J|)\,\sigma_i(-\sigma_j') = -|J|\,\sigma_i'\sigma_j'.

Every bond becomes $-|J|\sigma_i'\sigma_j'$ — a ferromagnetic bond in the $H=-J\sum\sigma\sigma$ convention. Why this is not just cosmetic: the two Hamiltonians share the same partition function term by term, so free energy $f$ , critical temperature, and specific heat are identical; only the observable that condenses changes — the ferromagnetic $\langle\sigma_i'\rangle$ is the antiferromagnetic staggered $m_s$ .

Step 4. See where it breaks. On the triangular lattice a triangle has three mutually adjacent sites; two-colouring is impossible (an odd cycle), so no consistent $A/B$ split exists and no such variable change converts all bonds to ferromagnetic. The antiferromagnet is then genuinely different. Verified: verify_antiferromagnetism-frustration.py symbolically checks $-J\sigma_i\sigma_j\to-|J|\sigma_i'\sigma_j'$ under the sublattice flip, and confirms the square-lattice consequence $k_BT_N=2|J|/\ln(1+\sqrt2)=2.2692\,|J|$ .

Go deeper · Derivation B · 1 minThe triangular plaquette: six ground states and one lost bond

Derivation BThe triangular plaquette: six ground states and one lost bond

The smallest frustrated unit already exposes the central phenomenon — macroscopic degeneracy — so we enumerate it exactly. Take three Ising spins on a triangle with antiferromagnetic bonds, $H=-J(\sigma_1\sigma_2+\sigma_2\sigma_3+\sigma_3\sigma_1)$ with $J=-|J|<0$ , and simply list all $2^3=8$ configurations, their energies, and their bond-satisfaction pattern. The goal is to pin down the ground-state energy, the degeneracy, and the microscopic correlation, all as exact integers or simple fractions — no approximation is needed at the plaquette level.

Step 1. Write the energy in terms of bond products. With $J=-|J|$ , $H=+|J|(\sigma_1\sigma_2+\sigma_2\sigma_3+\sigma_3\sigma_1)$ : a satisfied (antiparallel, $\sigma_i\sigma_j=-1$ ) bond contributes $-|J|$ , a violated (parallel) bond contributes $+|J|$ .

Step 2. Use the identity $(\sigma_1\sigma_2)(\sigma_2\sigma_3)(\sigma_3\sigma_1)=(\sigma_1\sigma_2\sigma_3)^2=+1$ . The product of the three bond values is always $+1$ , so the number of antiparallel ( $-1$ ) bonds is even: you can have $0$ or $2$ satisfied bonds, never $3$ .

Step 3. Minimise. The best you can do is two satisfied bonds ( $-|J|$ each) and one violated ( $+|J|$ ):

E_\text{min} = -|J|-|J|+|J| = -|J|.

The worst is all three parallel (both fully aligned states), $E_\text{max}=+3|J|$ .

Step 4. Count. Of the $8$ configurations only the two ferromagnetic ones (all-up, all-down) are at $E_\text{max}$ ; the remaining $\mathbf{6}$ are degenerate ground states — the violated bond can sit on any of the $3$ edges, times an overall spin flip. So $g_0=6$ . Why this matters: six tied ground states in a three-spin cluster is already an extensive count, $\tfrac13\ln 6 = 0.597$ per spin — an upper bound overshooting the correlated-lattice value because sharing bonds between triangles removes some freedom.

Step 5. Read off the microscopic correlation. In any ground state two bonds are $-1$ and one is $+1$ , so the average nearest-neighbour correlation is $\langle\sigma_i\sigma_j\rangle_\text{nn}=\tfrac13(-1-1+1)=-\tfrac13$ — far from the fully antiparallel $-1$ . Even at $T=0$ the triangular AFM is only one-third correlated. Verified: verify_antiferromagnetism-frustration.py enumerates all $8$ configurations, confirming $E_\text{min}=-|J|$ , $E_\text{max}=+3|J|$ , $g_0=6$ , product of bond values $=+1$ , and $\langle\sigma_i\sigma_j\rangle_\text{nn}=-1/3$ .

Go deeper · Derivation C · 1 minWannier residual entropy from the exact free energy

Derivation CWannier residual entropy from the exact free energy

The plaquette count suggests extensive $T=0$ entropy; Wannier's exact solution makes it a number. The route is thermodynamic: from the exact free energy $f(T)$ of the triangular Ising antiferromagnet, the entropy per spin is $s=-\partial f/\partial T$ , and its $T\to0$ limit is the residual entropy. We do not re-derive the full lattice-sum free energy here; the goal is to state the correct closed form of $s_0$ , verify its value, and expose the two normalisation slips that a careless transcription produces.

Step 1. Degeneracy sets the entropy. As $T\to0$ only the ground manifold survives, so $S(0)=k_B\ln\Omega_0$ with $\Omega_0=e^{Ns_0}$ , giving $s_0=\tfrac1N\ln\Omega_0$ . Each elementary triangle carries exactly one frustrated bond; the freedom to distribute those defects generates the count.

Step 2. The exact result. Wannier (1950) evaluated the lattice Green-function integral and obtained

s_0 = \frac{2}{\pi}\int_0^{\pi/3}\ln\!\big(2\cos x\big)\,\mathrm{d}x.

The prefactor is $2/\pi$ and the upper limit is $\pi/3$ .

Step 3. Evaluate numerically:

s_0 = 0.3230659\ldots \approx 0.3231.

Compare $\ln 2=0.6931$ (spins fully independent) and $0$ (ordered): the triangular AFM is partially but irreducibly disordered at absolute zero. Why the normalisation is a trap: writing the prefactor as $1/\pi^2$ gives $0.0514$ , and moving the upper limit to $\pi/6$ gives $0.2154$ — both wrong. Only $\tfrac2\pi\int_0^{\pi/3}$ returns $0.3231$ . (The AllQFT base carried both errors; corrected here.)

Step 4. Physical meaning. The defect bonds map one-to-one onto dimer coverings of the dual honeycomb lattice; counting dimer coverings is the exactly solvable combinatorial problem behind the number, and it is why the entropy is finite rather than $\ln 2$ . Verified: verify_antiferromagnetism-frustration.py computes $\tfrac2\pi\int_0^{\pi/3}\ln(2\cos x)\,dx=0.32307$ and explicitly checks that the $1/\pi^2$ and $\pi/6$ forms give the wrong $0.0514$ and $0.2154$ .

Limiting cases

How antiferromagnetic order behaves across lattices and limits — the deciding variable is always the presence or absence of odd loops:

Case	Result	Meaning
Square AFM (bipartite, $d=2$ )	$k_B T_N = \dfrac{2\vert J\vert}{\ln(1+\sqrt2)}\approx 2.269\,\vert J\vert$	Exact Onsager temperature; identical to the ferromagnet after sublattice rotation. True Néel order with $m_s\neq0$ below $T_N$ .
Triangular AFM (non-bipartite, $d=2$ )	$T_c=0,\quad s_0/k_B=0.3231$	No finite- $T$ transition; macroscopic ground-state degeneracy gives Wannier's residual entropy. A classical, algebraically correlated spin liquid at $T=0$ .
High temperature	$\langle\sigma_i\sigma_j\rangle\approx\beta J_{ij}\to0$	To leading order the sign of $J$ is irrelevant to thermodynamics; frustration first appears at order $\beta^3$ through odd-loop cumulants.
Two-sublattice mean field (bipartite)	$k_B T_N^{\rm MF}=z\vert J\vert$	Overestimates $T_N$ and ignores frustration entirely; on the triangular lattice it falsely predicts order where the exact answer is $T_c=0$ .

Common trapAn antiferromagnet is NOT just a ferromagnet with a flipped order parameter

The sublattice gauge trick $\sigma_i\to-\sigma_i$ that turns an AFM into an FM works only on bipartite lattices. On the triangular (or any non-bipartite) lattice it fails: the antiferromagnet is genuinely different, with $T_c=0$ and finite residual entropy $s_0$ . Two related traps: (1) the residual entropy $s_0=0.3231\,k_B$ does not violate the modern third law, because the degeneracy is macroscopic ( $\Omega_0=e^{Ns_0}$ ), not the "unique ground state" case the law addresses. (2) Wannier's number is $\tfrac2\pi\int_0^{\pi/3}\ln(2\cos x)\,dx$ — not $\tfrac1{\pi^2}\int_0^{\pi/3}$ ( $=0.0514$ ) and not $\tfrac2\pi\int_0^{\pi/6}$ ( $=0.2154$ ). Get the prefactor $2/\pi$ and the limit $\pi/3$ right or you land two-thirds too low.

Worked examples

Example 1 — Enumerate the frustrated triangle and read off its numbers.

Step 1. Take $|J|=1$ so $H=+(\sigma_1\sigma_2+\sigma_2\sigma_3+\sigma_3\sigma_1)$ for the AFM triangle. List energies: all-parallel $(\uparrow\uparrow\uparrow)$ gives $1+1+1=+3$ ; any single flip, e.g. $(\uparrow\uparrow\downarrow)$ , gives $\sigma_1\sigma_2=+1,\ \sigma_2\sigma_3=-1,\ \sigma_3\sigma_1=-1$ , sum $=-1$ .

Step 2. By symmetry the two fully-aligned states have $E=+3$ and the other six have $E=-1$ . So $E_\text{min}=-1$ , $E_\text{max}=+3$ , and the ground-state degeneracy is $\mathbf{g_0=6}$ .

Step 3. Nearest-neighbour correlation in a ground state: two bonds are $-1$ , one is $+1$ , so

\langle\sigma_i\sigma_j\rangle_\text{nn}=\tfrac13(-1-1+1)=\boldsymbol{-\tfrac13}.

Step 4. Independent-triangle entropy estimate: $\tfrac13\ln 6=\mathbf{0.597}$ per spin. This overshoots the true lattice value $0.3231$ because neighbouring triangles share spins — but it already proves extensive $T=0$ entropy. The exact answer is six ground states, $\langle\sigma_i\sigma_j\rangle=-1/3$ , residual entropy $\le 0.597$ per spin.

Example 2 — Compare the residual entropy against the disordered and ordered bounds, and against spin ice.

Step 1. Evaluate Wannier's integral $s_0=\tfrac2\pi\int_0^{\pi/3}\ln(2\cos x)\,dx$ . Numerically this is $s_0=\mathbf{0.32307}$ per spin.

Step 2. Bracket it: fully random spins give $\ln 2 = 0.69315$ , a unique ordered ground state gives $0$ . So the fraction of maximal entropy retained at $T=0$ is

\frac{s_0}{\ln 2}=\frac{0.32307}{0.69315}=\mathbf{0.466}.

The frustrated magnet keeps almost half of its high-temperature entropy at absolute zero.

Step 3. Compare spin ice, whose Pauling estimate is $s_0^{\rm ice}=\tfrac12\ln(3/2)=\mathbf{0.20273}$ per spin. Both are nonzero residual entropies of geometrically frustrated systems; the triangular Ising value is larger because its ground-state manifold is proportionally bigger.

Step 4. Number of ground states. For $N=100$ spins, $\Omega_0=e^{Ns_0}=e^{100\times0.32307}=e^{32.31}\approx\mathbf{1.1\times10^{14}}$ tied lowest-energy configurations — a concrete face of "macroscopic degeneracy."

Origins · 1 mingeometric frustration and how this was discovered

geometric frustration

concept: competing interactions on a lattice · The term describes a system whose bonds cannot all be simultaneously satisfied because of the lattice geometry — the defining feature of the triangular Ising antiferromagnet. The idea crystallised through Wannier's and Houtappel's 1950 exact solutions and Toulouse's 1977 gauge formulation, and now names an entire subfield spanning spin ice, spin liquids, and spin glasses.

History

1948

— Néel formalises antiferromagnetism and the two-sublattice picture (Nobel Prize 1970)

1950

— Wannier and Houtappel independently solve the triangular Ising antiferromagnet; Wannier finds the residual entropy s₀ ⁄ k_B = 0.3231

1964

— Stephenson computes the T = 0 algebraic (power-law) spin correlations of the triangular AFM

1973

— Wannier publishes an erratum refining the residual-entropy calculation

1977

— Toulouse introduces the gauge-invariant frustration function Φ_p for arbitrary sign patterns

1999

— Ramirez and collaborators measure the Pauling residual entropy of spin ice Dy₂Ti₂O₇

Going further

References. Pathria & Beale, Statistical Mechanics, 3rd ed., Ch. 11–13 (Ising model, lattice statistics) · Kardar, Statistical Physics of Fields, Ch. 7–8 (lattice models, frustration) · R. M. F. Houtappel, Physica 16, 425 (1950) — exact triangular Ising solution · G. H. Wannier, Phys. Rev. 79, 357 (1950); erratum Phys. Rev. B 7, 5017 (1973) — residual entropy · Sethna, Statistical Mechanics: Entropy, Order Parameters, and Complexity (frustration, residual entropy) · Landau & Lifshitz, Statistical Physics Part 1 (magnetism and the Ising model).

Provenance. Hamiltonian and bipartite gauge mapping: Pathria & Beale Ch. 11–12; Kardar SPF. Square-lattice $T_c=2J/\ln(1+\sqrt2)\approx2.269J$ from Onsager (verifier: $2/\ln(1+\sqrt2)=2.2692$ ). Frustration function and gauge invariance: Toulouse (1977), reproduced in Kardar/Sethna. Triangular Ising exact free energy: Houtappel Physica 16, 425 (1950); residual entropy $s_0/k_B=0.3231$ : Wannier Phys. Rev. 79, 357 (1950), erratum PRB 7, 5017 (1973). Nearest-neighbour correlation $-1/3$ from plaquette combinatorics; $T=0$ algebraic decay exponent from Stephenson (1964). Spin-ice Pauling entropy $\tfrac12\ln(3/2)=0.2027$ : Pauling (1935), Ramirez et al. (1999). Two base-error corrections are documented in the cheat sheet.

Builds on: The Ising Model: Definition and Hamiltonian · Order Parameter and Spontaneous Magnetization · Mean-Field (Weiss) Theory of the Ising Model · Peierls Argument for Long-Range Order. Leads to: Domain Walls and Interfacial Excitations · Bethe-Peierls Approximation · High- and Low-Temperature Series Expansions · Binary Alloy Order-Disorder Transition.

Is an antiferromagnet magnetic if its total magnetization is zero?

Yes — it is magnetically ordered, just with zero net moment. Below $T_N$ the spins lock into a regular staggered pattern (the Néel state), so the correct order parameter is the staggered magnetization $m_s$ , not the uniform $m$ . The uniform magnetization vanishes because the two sublattices cancel, but neutron diffraction sees the doubled magnetic unit cell directly, and the susceptibility has a characteristic cusp at $T_N$ . "No net moment" is not "no order."

Why does the triangular antiferromagnet not order even at T = 0?

Because there is no single spin arrangement it can commit to. On a bipartite lattice the ground state is unique (up to a global flip) and the system freezes into it. On the triangular lattice every triangle must sacrifice one bond, and the sacrificed bond can be chosen in many ways, so an exponentially large set of configurations ties for the lowest energy. With no unique state to select, no long-range order forms; the ground state is a fluctuating "classical spin liquid" with power-law correlations rather than a frozen crystal of spins.

How can entropy be nonzero at absolute zero without breaking the third law?

The modern (statistical) third law says the entropy per particle tends to zero if the ground state is unique or its degeneracy grows slower than exponentially in $N$ . Frustrated systems violate that premise: their ground-state count is $\Omega_0=e^{Ns_0}$ , exponential in $N$ , so $S_0/N=k_B s_0$ stays finite. This is not a loophole but a genuine physical situation, confirmed experimentally by the residual entropy of spin ice. The original Nernst statement was framed before such macroscopically degenerate systems were understood.

What exactly is the frustration function Φ_p, and how do I compute it?

For any closed loop $p$ of bonds, multiply the signs of the couplings around it: $\Phi_p=\prod_{\langle ij\rangle\in p}\mathrm{sgn}(J_{ij})$ . If $\Phi_p=-1$ the loop is frustrated (an odd number of bonds must be violated in any configuration); if $\Phi_p=+1$ it can be fully satisfied. For a pure antiferromagnet ( $J_{ij}<0$ everywhere), $\Phi_p=(-1)^{(\text{loop length})}$ : triangles ( $-1$ , frustrated), squares ( $+1$ , unfrustrated). It is gauge-invariant, so it is a property of the lattice and sign pattern, not of any spin configuration.

Does frustration always destroy magnetic order?

No — it weakens or reshapes it. On strongly frustrated lattices (triangular Ising, kagome) it can push $T_c$ to zero. On partially frustrated lattices it merely depresses the ordering temperature far below the mean-field $z|J|$ and can favour noncollinear or incommensurate spin structures (spirals) instead of simple Néel order. Many real materials — MnO, NiO on bipartite lattices — order normally, while face-centred-cubic and kagome antiferromagnets show suppressed, exotic, or absent order. The rule of thumb: the more (and the shorter) the odd loops, the stronger the frustration.

How does the triangular-lattice problem connect to dimers and to spin ice?

Each ground state of the triangular Ising antiferromagnet has exactly one frustrated bond per triangle; mark each frustrated bond as a "dimer" on the dual honeycomb lattice, and the ground states map one-to-one onto dimer coverings — an exactly solvable combinatorial problem that yields Wannier's entropy. Spin ice is the three-dimensional cousin: on the pyrochlore lattice a "two-in/two-out" ice rule plays the role of the "one frustrated bond per triangle" rule, and Pauling's water-ice counting gives its residual entropy $\tfrac12\ln(3/2)$ . Same idea, different geometry.

Why is the T = 0 nearest-neighbour correlation only −1/3 and not −1?

Because the triangle cannot make all bonds antiparallel. In a ferromagnet or a bipartite antiferromagnet, every satisfied bond has $\sigma_i\sigma_j=\pm1$ fully, and at $T=0$ correlations saturate. On the frustrated triangle, each ground state has two bonds at $-1$ and one forced to $+1$ , so averaging the three symmetry-equivalent bonds gives $\tfrac13(-1-1+1)=-\tfrac13$ . The value being pinned away from $-1$ at absolute zero is the microscopic fingerprint of frustration: the system is only partly correlated even in its lowest-energy states.