= Solution
Keep $N$ fixed and put $q=\tanh K$, so $q\to1^-$ as <temperature> tends to zero. For even $N$, the <finite-ring antiferromagnetic Ising correlation> becomes
$$
C_n=(-1)^n\frac{q^n+q^{N-n}}{1+q^N}
\longrightarrow\boxed{(-1)^n\quad(N\text{ even}).}
$$
Both alternating <ground states> have $S_1S_{n+1}=(-1)^n$, so their equal-weight <canonical ensemble> has the same <two-point correlation function> despite its zero <one-point correlation function>.
For odd $N$, the signs instead give
$$
C_n=(-1)^n\frac{q^n-q^{N-n}}{1-q^N}.
$$
Direct substitution gives an indeterminate ratio. Expanding at fixed $N,n$ with $q=1-\delta$,
$$
q^n-q^{N-n}=(N-2n)\delta+O(\delta^2),\qquad
1-q^N=N\delta+O(\delta^2).
$$
Consequently the <zero-temperature correlation of an odd antiferromagnetic Ising ring> is
$$
\boxed{\lim_{\beta J\to\infty}\langle S_1S_{n+1}\rangle
=\begin{cases}
(-1)^n,&N\text{ even},\\
(-1)^n\left(1-\dfrac{2n}{N}\right),&N\text{ odd}.
\end{cases}}
$$
The odd-ring factor has a direct <probability> interpretation. At zero <temperature>, the <canonical ensemble> gives equal weight to its $2N$ <ground states>. The single <frustrated bond in an antiferromagnetic Ising ring> is uniformly located, with <probability> $n/N$ of lying on the chosen $n$-bond path. If it is absent from that path, $S_1S_{n+1}=(-1)^n$; if it is present, this product is $-(-1)^n$. Averaging gives $(-1)^n[(1-n/N)-n/N]$.
Thus the decay with separation here comes from averaging a movable, compulsory <frustrated bond in an antiferromagnetic Ising ring>, even after additional thermally excited <frustrated bonds in an antiferromagnetic Ising ring> disappear. In particular $C_1=-1+2/N$ for odd $N$, and $JN C_1=-J(N-2)$ reproduces the <ground-state energy>. The formula also respects $C_n=C_{N-n}$, although its factor $1-2n/N$ changes sign halfway around the ring.
\Image[/past-exam-of-the-advanced-quantum-theory-path-integrals-and-many-particle-physics-course-of-the-university-of-oxford/2025-ring-parity.png]
{title=Ring parity, unavoidable frustration, and the zero-temperature correlation}
{description=Original <Ising spin> configurations show satisfied bonds in gray and the unavoidable odd-ring parallel bond in red. The plots show the exact <finite-ring antiferromagnetic Ising correlation>, with its alternating sign removed, and the zero-<temperature> limits for even and odd rings.}
The order and scale of limits matter. Taking the <thermodynamic limit> first at positive <temperature> gives the <exponential function> in part (c). Cooling an odd finite ring first leaves the linear factor $1-2n/N$. Both orders subsequently give $(-1)^n$ at fixed $n$, but this agreement is not uniform in separation: when $n/N$ stays nonzero, the odd-ring linear factor survives in the zero-<temperature> expression.
Back to article page
