Many-Body Spectral Gaps

A many-body spectral-gap project asks whether an excitation gap remains positive under a specified limit, boundary condition, and Hamiltonian family. Finite-size diagonalization alone cannot settle that question. Reviewed 11 October 2026, this page includes a gap-certificate preprint revised on 2 October 2026 and distinguishes it from thermodynamic-limit criteria.

Helpful background. Lieb–Robinson bounds explain locality estimates; spectral problems introduces certification versus numerical approximation.

Specify the ground-state sector and the limit

For a finite Hamiltonian $H_L$ with ground energy $E_0(L)$, the gap is the distance from its ground-state sector to the next spectral value. A uniform-gap statement requires a positive bound independent of sufficiently large $L$ in a declared sequence.

Ground-state degeneracy, boundary modes, symmetry sectors, and boundary conditions matter. A low-lying edge excitation can close an open-chain gap while a bulk diagnostic remains finite. Comparing different sector restrictions can manufacture apparent agreement or disagreement.

For a nonnegative frustration-free Hamiltonian with ground energy zero,

$$H^2\ge\delta H$$

implies that nonzero spectral values are at least $\delta$. Turning this global inequality into local, verifiable certificates is a powerful route.

A finite calculation that really controls arbitrary length

Established criterion. Consider a translation-invariant chain with finite on-site dimension and nearest-neighbor orthogonal-projector interactions $h_i^2=h_i$. Assume frustration freeness for the periodic chains under consideration. Let $\gamma_n$ be the gap above the full ground space of the corresponding open chain of $n>2$ sites, containing $n-1$ bonds. The original Knabe criterion gives, for periodic length $L>n$,

$$\Delta_L\ge\frac{n-1}{n-2} \left(\gamma_n-\frac1{n-1}\right).$$

Thus a certified local gap exceeding $1/(n-1)$ proves a uniform positive bound. A negative right-hand side says nothing about gaplessness. Projector normalization, open versus periodic boundaries, and the number of sites versus bonds are part of the statement. Gosset and Mozgunov establish a stronger threshold for this setting; their paper is a useful source for comparing the original criterion with improvements. One must not transplant these constants into a different geometry or an arbitrary frustrated Hamiltonian.

For a complete check, take $h_i=(I-Z_iZ_{i+1})/2$. Each violated bond costs one energy unit. The open chain has two ground states and gap $\gamma_n=1$: a single domain wall suffices. On a periodic chain of at least three sites, domain walls occur in pairs and $\Delta_L=2$. Knabe’s bound returns $\Delta_L\ge1$, valid but not tight. This is already a successful certificate; the finite-volume gap and the rigorous lower bound are not expected to coincide. Rescaling every interaction by $J>0$ rescales all energies, but the numerical threshold must also be multiplied by $J$.

The example also separates ground-space degeneracy from gaplessness. Both ferromagnetic states belong to the ground sector. Calling their zero splitting the excitation gap would answer a different question.

Established barriers and active methods

Cubitt, Pérez-García, and Wolf prove undecidability for a sufficiently broad family of translationally invariant lattice Hamiltonians. This rules out a universal decision algorithm for that family; it does not say every physically interesting gap question is unanswerable.

Rai and colleagues’ published 2026 work develops semidefinite hierarchies for frustration-free thermodynamic gap certificates. Rao’s preprint, revised in October 2026, gives a convergent hierarchy for the first-two-eigenvalue gap $\lambda_1-\lambda_0$ of finite qubit Hamiltonians without assuming frustration freeness. Here eigenvalues are counted with multiplicity. This agrees with the gap above the ground-state sector when the ground state is unique; for a degenerate ground space it is zero even when the sector gap is positive. For example, $H=\operatorname{diag}(0,0,1,1)$ has $\lambda_1-\lambda_0=0$ and sector gap $1$. Match the certificate’s target to the question before applying it. A finite-system convergence theorem does not by itself furnish an efficient, size-uniform thermodynamic certificate.

Active directions include exploiting symmetry in certificates, reducing computational cost, proving finite-size criteria for additional interactions, and separating bulk gaps from boundary effects. A numerical extrapolation is a hypothesis generator until a rigorous criterion closes the inference. The evidence horizon includes the April 2026 Rai publication and the 2 October 2026 Rao preprint; it is a selected methodological assessment, not a complete catalog of models whose gap remains open.

An entry project: turn a numerical gap into a certificate

Baseline. First recover the commuting Ising-chain calculation above, including its degeneracy. Then use the spin-1 AKLT projector onto bond total spin two, with exactly the normalization in the selected certificate paper. Reproduce a published finite-size or SDP lower bound and compare small open and periodic chains separately. The AKLT gap is established; this stage tests the implementation.

Resources. A workstation or laptop for modest blocks, sparse linear algebra, semidefinite optimization, and basic operator inequalities. Keep the interaction normalization fixed.

Deliverable. Compare the rigorous lower bound, numerical finite-size gap, and computational cost as block size grows. Supply the certificate and independently check its positivity and residuals.

Failure criterion. A positive floating-point optimum is not a proof. Obtain an exact or interval-validated certificate with all linear constraints satisfied and positivity justified; small residuals alone can conceal negative directions near a kernel. Rai and colleagues’ certifiable formulation is an explicit starting point. If a certificate relies on frustration freeness and a perturbed Hamiltonian lacks it, the theorem no longer applies.

Research extension. Optimize the certificate using a symmetry decomposition, or find a parameter interval where a sharpened local inequality proves a uniform gap. A well-characterized region where the method fails is useful evidence about the method, not proof of gaplessness.

Set a first-stage resource budget in local block dimension, not total chain length: a dense spin-1 block has dimension $3^n$ before symmetry reduction. Stop increasing $n$ when memory use or enclosure errors erase the certification margin. At that point compare a smaller symmetry-reduced certificate or a different local decomposition. The original research target should be a proved bound or parameter interval absent from the selected literature, with independently checkable certificate data; a fresh numerical estimate of an already known gap is insufficient.

Choosing the question and update trigger

A precise proposal names the Hamiltonian, locality, geometry, ground-state structure, and limit. “Prove the spectral gap” without these is not a question. Favor a model where analytical structure and certification can reinforce numerical exploration.

Progress changes an assessment when a uniform bound is established, a gap-closing sequence is controlled, or a theorem extends the accessible class. Larger finite lattices alone increase evidence but do not logically replace the missing limit argument.

References

• T. S. Cubitt, D. Pérez-García, and M. M. Wolf, “Undecidability of the spectral gap,” Nature 528, 207–211 (2015). Full version.
• D. Gosset and E. Mozgunov, “Local gap threshold for frustration-free spin systems,” Journal of Mathematical Physics 57, 091901 (2016). Finite-size criteria.
• K. S. Rai, I. Kull, P. Emonts, J. Tura, N. Schuch, and F. Baccari, “A Hierarchy of Spectral Gap Certificates for Frustration-Free Spin Systems,” Quantum 10, 2065 (2026). Published article, full text and certifiable formulation.
• S. Rao, “A convergent hierarchy of spectral gap certificates for qubit Hamiltonians,” arXiv:2510.08427v2 (2 October 2026), preprint. Revised construction.

Entry points

Read one certificate theorem with every boundary and normalization assumption visible. Reproduce its known example before using it to label an unfamiliar model gapped.