The spin-one Haldane gap: rings, boundary fields and a topological index

← All digests

Family 268 · Mathematical physics
Importance 5/5 · Provisional editorial rating
Sources consulted 2026-10-25

1. The question

A quantum spin chain can have excitations whose energy becomes arbitrarily small as the chain grows. Alternatively, every nontrivial excitation can cost a fixed positive amount. That difference separates a gapless system from a gapped one.

Haldane predicted a striking distinction between antiferromagnetic Heisenberg chains with integer spin and those with half-odd-integer spin. For spin one, the expected excitation gap is positive. Physical arguments, numerical calculations and experiments supplied strong evidence, but they are different from a uniform mathematical lower bound for the exact Hamiltonian.

Family 268 contains two manuscripts claiming such bounds for the pure bilinear spin-one Heisenberg interaction. M1 treats even periodic rings. M2 treats odd open chains with a specified magnetic field at both endpoints, then uses Tasaki’s index theorem to identify a nontrivial topological index for the resulting boundary-selected infinite-chain states.

2. Background and essential definitions

A spin-one site has a three-dimensional state space $\mathbb C^3$. Its spin operators $S^x,S^y,S^z$ are Hermitian matrices satisfying the angular-momentum commutation rules and

$$(S^x)^2+(S^y)^2+(S^z)^2=2I.$$

The eigenvalues of $S^z$ are $-1,0,1$. On a chain, $S_j^\alpha$ acts on site $j$ and as the identity elsewhere.

For a ring of $n$ sites, with coupling constant one, the Hamiltonian is

$$H_n^{\mathrm{per}}=\sum_{j=0}^{n-1}\mathbf S_j\cdot\mathbf S_{j+1}, \qquad \mathbf S_n=\mathbf S_0.$$

The dot product means the sum of the three component products. The positive coupling is antiferromagnetic: neighboring spins favor antiparallel correlations. Quantum terms on overlapping bonds need not commute, so minimizing every bond separately is generally impossible.

Let $E_0(n)$ be the ground energy and $E_1(n)$ the next distinct energy. The periodic gap is $\gamma_n=E_1(n)-E_0(n)$. Each finite nonconstant matrix has a positive gap above its full ground eigenspace. The difficult assertion is a lower bound that survives arbitrarily large $n$.

For an open chain with sites $-L,\ldots,L$, M2 uses

$$H_L(h)=\sum_{j=-L}^{L-1}\mathbf S_j\cdot\mathbf S_{j+1} -h(S_{-L}^z+S_L^z).$$

There is no bond joining the ends, and both fields have the same sign. Here the two lowest eigenvalues are counted with multiplicity, so a positive gap also implies a unique ground state.

A thermodynamic state is described by expectations of local observables, matrices acting on finitely many sites. A subsequential local limit means these expectations converge along a selected sequence of longer chains, without assuming convergence of the entire sequence.

3. Development history and previous partial results

The Lieb–Schultz–Mattis theorem supplied a rigorous low-energy obstruction for the cyclic spin-one-half chain. Affleck and Lieb extended the half-odd-integer picture in their setting, retaining the possibility of ground-state degeneracy. Integer spin escapes that obstruction, but escaping a no-go theorem does not prove a gap.

Haldane’s 1983 field-theoretic work predicted the integer versus half-odd-integer distinction. The Affleck–Kennedy–Lieb–Tasaki model then provided a rigorous gapped spin-one example. Its bond interaction includes a biquadratic term: if $q=\mathbf S_j\cdot\mathbf S_{j+1}$, it is proportional, up to an additive constant, to $q+q^2/3$. The pure Heisenberg chain uses $q$ alone.

This distinction is essential. Stability of the AKLT gap under sufficiently small perturbations, as established by Yarotsky in the relevant class, controls a neighborhood of that interaction. It does not by itself give a gapped path all the way to the pure bilinear point.

White and Huse’s numerical study estimated the spin-one bulk gap near $0.41050$ in the unit-coupling normalization and identified effective spin-one-half behavior at free ends. Experiments in nearly one-dimensional spin-one materials supplied additional evidence. Boundary effects explain why a periodic gap does not automatically settle a particular open-chain gap.

Finite-size gap criteria also developed, notably for frustration-free chains. Those methods have powerful scope, but the pure Heisenberg model requires an argument suited to its competing interactions. M1 uses finite thermal traces and trial-state energy bounds followed by an analytic recurrence, rather than importing a frustration-free criterion unchanged.

For topology, Tasaki’s conditional theorem proves that the spin-one Heisenberg ground state has a nontrivial index under a precise fixed-boundary-field gap assumption. M2 claims to establish that assumption at one explicit field. The topological implication is credited to Tasaki, not presented as a new independent index theorem.

4. What the new manuscripts claim

Even periodic rings

M1 claims uniqueness of the periodic ground state for every even $n\geq60$, together with

$$\gamma_n>\frac4{105}\log\frac{80}{79} \qquad(n\geq60,\ n\text{ even}),$$

and the stronger eventual estimate

$$\gamma_n>\frac{\log20}{784} \qquad(n\geq2304,\ n\text{ even}).$$

Consequently the lower limit along even rings is at least $(\log20)/784>0$. This is the positive even-periodic formulation of the spin-one Haldane conjecture stated by the paper.

The finitely many smaller even rings have a positive gap above their full ground sector simply because they are finite-dimensional nonconstant Hamiltonians. Combining that fact with the uniform large-ring estimate gives a positive infimum over all even $n\geq4$. This observation does not provide a numerical bound or a uniqueness assertion for those smaller rings. M1 also states a corresponding bound for local excitations in subsequential periodic ground-state limits.

Odd open chains with one fixed field

M2 claims, at the fixed field $h=3/5$,

$$E_1(L,3/5)-E_0(L,3/5)>\frac{\log10}{392} \qquad(L\geq960).$$

The chain has $2L+1$ sites, so the threshold refers to half-length, not total length. The same-field endpoint Hamiltonian above is part of the theorem. It is not a free-end chain, nor a model with extra spin-one-half endpoint sites.

The boundary-selected index

For every subsequential local limit $\omega$ of those open-chain ground states, M2 derives a locally unique gapped ground-state inequality, invariance under rotations about the $z$-axis and site reflection, and Tasaki index $-1$.

The index is defined through a slowly varying local twist. In the source convention,

$$U_\varepsilon= \exp\left[-i\sum_{|j|\leq\pi/\varepsilon} (\pi+\varepsilon j)S_j^z\right].$$

After taking the infinite-chain state limit first, the index is $\lim_{\varepsilon\downarrow0}\omega(U_\varepsilon)$, which the cited theorem restricts to $\{-1,1\}$. The value $-1$ is the nontrivial alternative for this diagnostic.

This does not assume that the full finite-chain sequence converges, establish global uniqueness of every infinite-chain ground state, or identify all possible boundary constructions with these selected states.

5. Why the advance matters

Provisional importance: 5/5. The Haldane gap is a central example of a qualitative quantum effect that depends on spin parity. A uniform lower bound for the pure spin-one interaction would place this prediction on a mathematical footing in the stated finite-volume formulations.

The explicit lower bounds are much smaller than the familiar numerical gap estimate. Their purpose is different: they remain strictly positive at every sufficiently large allowed length. They are not claimed to locate the true gap accurately.

The second manuscript matters independently because boundaries can carry low-energy degrees of freedom even when the bulk is gapped. Its fixed-field result supplies the specific hypothesis needed for a nontrivial topological index. This connects spectral control to a robust property of selected infinite-chain states without conflating several different notions of string order, edge behavior and topological index.

6. The main idea

M1 reads one partition function in two directions. The physical expression is a trace of a thermal exponential, a sum of positive Boltzmann weights. A time-ordered bond-history expansion also expresses it as a spatial moment

$$Z_n(\beta)=\operatorname{Tr}(X_\beta^n)$$

of a compact self-adjoint transfer operator. A common energy shift is included in the source’s definition of $Z_n$; it cancels from the probability ratios used for gaps. For even $n$, the powers of the spatial eigenvalues contribute nonnegative weights, even though $X_\beta$ itself need not be positive.

Normalize each collection of weights to a probability distribution. Its purity is $\sum_i p_i^2$. Purity close to one means that one weight dominates. Squaring and renormalizing weights improves that concentration rapidly once a suitable initial threshold is reached.

On the physical side, squaring weights doubles inverse temperature $\beta$. On the spatial side, it doubles chain length. Exact identities couple these two operations. Starting from finite purity bounds, the manuscript propagates increasingly strong concentration as both scales double. Moment inequalities fill the intermediate even lengths.

Why does purity imply a gap? The ratio of an excited-state weight to a ground-state weight is $e^{-\beta(E-E_0)}$. A sufficiently strong exponential suppression as $\beta$ grows forces $E-E_0$ to stay positive. The argument tracks the full physical Gibbs distribution, so ground-state multiplicity is also relevant.

The finite starting estimates use explicitly specified thermal trace enclosures and matrix-product trial vectors. The source describes integer arithmetic, rational interval bounds and controlled polynomial truncation errors. These finite inputs initialize an analytic argument for all larger sizes; the digest reports that construction without rerunning its certificates.

For an open chain, the transfer expression becomes a boundary pairing rather than a trace. Spatial eigenspaces are weighted by overlap with a boundary vector, and that vector changes with temperature. Periodic concentration alone therefore leaves a missing step.

M2 controls boundary-weighted residual moments together with Gibbs purity for an auxiliary opposite-field chain. A rotation then relates the leading spatial contribution to the same-field Hamiltonian. The coupled recurrence yields the claimed uniform open-chain gap. Tasaki’s theorem subsequently fixes the twist-index sign: symmetry keeps the twist expectation real, and the gap prevents the interpolation from crossing zero in the relevant regime.

7. Scope, limitations, and open questions

These claims concern spin one, unit antiferromagnetic coupling and the specified pure bilinear interaction. They do not settle the entire integer-spin Haldane prediction, arbitrary anisotropies, all boundary conditions or odd periodic rings.

The open-chain field is fixed at $3/5$. A gap there should not be quoted as a free-boundary gap. The topological result concerns subsequential boundary-selected states and Tasaki’s particular twist index; stronger global uniqueness conclusions require additional hypotheses.

The papers provide explicit positive lower bounds, not the exact gap, a complete excitation spectrum or a uniform description of every perturbation of the model. Sharper quantitative bounds and extensions to other interactions remain separate questions from the claimed positive-gap resolution.

8. Sources and further reading

Haldane’s original field-theoretic paper and Tasaki’s conditional index theorem are linked in section 3. The manuscripts cite the Lieb–Schultz–Mattis and Affleck–Lieb obstructions, the AKLT construction, White–Huse numerical results and finite-size gap methods in their respective scopes. M1 supplies the periodic transfer and finite inputs used by M2; M2 proves the boundary-specific estimates before applying Tasaki’s theorem.

Source manuscripts

• M1. OpenAI, The periodic spin-one Haldane gap (September 24, 2026).
• M2. OpenAI, A boundary-field gap for the spin-one Heisenberg chain (September 24, 2026).

Sources consulted on 25 October 2026 at repository revision 1581d4c. Download the manuscript citations.