The Uniformly Gapped Substack: Existence and Stability of the Thermodynamic Spectral Gap

1 Introduction

1.1 The gapped substack and why uniformity is the point

The program of which this is the third part organizes topological phases of matter around a single geometric object: the condensed moduli stack $\mathfrak{Ham}_{d,G}$ of $G$-symmetric local and quasi-local Hamiltonians on a lattice $L$ in spatial dimension $d$. Part I [1] makes the interaction space into a Banach space $\mathcal{B}_{F}$ through an $F$-function locality norm and promotes the Heisenberg dynamics to a morphism of condensed objects; Part II [2] attaches the quasi-local $C^*$-algebra, its compact state space, and the solid $K$-theory invariant of [3]. The present paper concerns the substack

$$\mathfrak{Gap}_{d,G}(S)\;=\;\bigcup_{\Delta>0}\; \{\,H\in\mathfrak{Ham}_{d,G}(S)\,:\,\mathop{\mathrm{inf}}_{s\in S}\operatorname{gap}(H_s)\ge\Delta\,\}, \qquad S\ \text{profinite},$$

the uniformly gapped systems.

Everything downstream (the stabilized phase $∞$-groupoid $\mathfrak{Phase}_{d,G}=\big(\mathfrak{Gap}_{d,G}[\mathcal{W}^{-1}]\big)^{\mathrm{st}}$, the phase set $\operatorname{Phases}_{d,G}=\pi_0\operatorname{Shape}(\mathfrak{Phase}_{d,G})$, and the invertible condensed phase spectrum $\mathbf{IP}^{\mathrm{cond}}_{d,G}$ built in Part IV [4]) is a construction on top of (1). The phase functor is only as good as the substack it is built from.

The word uniformly is not decorative. A profinite family $s↦ H_s$ in which each $H_s$ happens to be gapped, but with $\mathop{\mathrm{inf}}_{s}\operatorname{gap}(H_s)=0$, is exactly the kind of family that fails to define a robust invariant: as one moves through the base the gap can be driven arbitrarily small, correlations grow without bound, and the locally constant phase label $\nu_f$ of the program loses meaning at the degenerating points. The union over $\Delta>0$ in (1) records that a member of $\mathfrak{Gap}_{d,G}(S)$ is a family for which some strictly positive $\Delta$ bounds the gap below simultaneously over the whole probe. Keeping the $\mathop{\mathrm{inf}}_{s}$ explicit is the single most important discipline of this paper.

1.2 The undecidability wall

There is a hard theorem standing between us and any naive program to “compute the substack.” Cubitt, Pérez-García and Wolf [5] proved that the spectral gap is undecidable: there is a family of translation-invariant, nearest-neighbour Hamiltonians on a two-dimensional lattice, depending computably on a parameter, for which no algorithm decides whether the thermodynamic-limit system is gapped or gapless. The construction embeds the halting problem into the low-energy physics, so the question “is this system gapped?” is at least as hard as the halting problem, and in fact $\Pi_1$-complete in the arithmetic hierarchy. The undecidability survives many natural restrictions.

We state the consequence for this paper in the strongest terms, because it shapes every claim we make.

There is no general algorithm, decision procedure, or uniform criterion that determines membership in $\mathfrak{Gap}_{d,G}$ for arbitrary members of $\mathfrak{Ham}_{d,G}$. Existence of a thermodynamic gap is not, in general, a decidable property; it is an assumption that defines the substack $\mathfrak{Gap}_{d,G}$.

This is why the paper is titled around stability rather than existence. We do not offer a test for gappedness: no such test can exist in general. Instead we take gappedness as a standing hypothesis and ask a question undecidability does not forbid: given a uniformly gapped family, what can be said about its neighborhood, its correlations, and its behavior along the profinite base? The answers are stability theorems, and they are the mathematically load-bearing content of Part III. Any statement in this paper that could be read as deciding, bounding from below, or algorithmically certifying the gap of a general member of $\mathfrak{Ham}_{d,G}$ would contradict [5] and is, by construction, absent.

1.3 Contributions

Within the boundary set by Section 1.2, we prove three theorems and formulate three conjectures.

• Theorem III-A (Section 4). At a fixed finite volume the $n$-gap of a Hamiltonian is $2C_\Lambda$-Lipschitz in the interaction norm (equivalently $2$-Lipschitz in the operator norm); consequently the gapped locus is open at every finite resolution, and along any $S$-continuous family the map $s↦\operatorname{gap}(H_s^{\Lambda})$ is continuous. This is elementary (Weyl’s perturbation inequalities plus finite-volume norm equivalence), and it is the one place where openness of $\mathfrak{Gap}_{d,G}$ is unconditional. It is also the finite-volume statement that undecidability leaves untouched: a finite matrix has a computable gap; undecidability is a statement about the thermodynamic limit.

• Theorem III-B (Section 5). On the frustration-free / LTQO stratum, the stability theorems of Bravyi–Hastings–Michalakis [6,7], Michalakis–Zwolak [8], and Nachtergaele–Sims–Young [9] provide a perturbation threshold $\varepsilon_0>0$ independent of system size. We recast this as a statement about $\mathfrak{Gap}_{d,G}$: each point of the stratum has a condensed-open neighborhood, of interaction-norm radius $\varepsilon_0$ along quasi-local perturbation directions, lying inside $\mathfrak{Gap}_{d,G}$. Openness and stability of the substack hold on the stratum; we are careful to claim nothing beyond it.

• Theorem III-C (Section 6). A uniform gap over a family forces uniform exponential clustering. Combining Hastings–Koma [10] and Nachtergaele–Sims [11] with the uniform Lieb–Robinson velocity of Part I, a family with $\mathop{\mathrm{inf}}_s\operatorname{gap}(H_s)\ge\Delta$ has a correlation length $\xi$ bounded above uniformly over the profinite base. This is the structural feature separating uniformly gapped families from merely pointwise-gapped ones.

• Conjectures III-1, III-2, III-3 (Section 7). Openness of $\mathfrak{Gap}_{d,G}$ in the full condensed topology on the physically relevant stratum, with the uniform-gap sheaf condition detected by finite quotients (III-1); descent of uniform gappedness along profinite covers, i.e. disorder hulls $\Omega=Q^{\mathbb{Z}^d}$ (III-2); and the identification of quasi-adiabatic continuation with the internal path components realizing $\mathcal{W}$, so that $\pi_0\operatorname{Shape}\!\big(\mathfrak{Gap}_{d,G}[\mathcal{W}^{-1}]\big)$ equals the operator-algebraic set of gapped phases of Ogata [12] (III-3).

Section 8 illustrates the whole picture on the transverse-field Ising chain: exact diagonalization renders the gapless discriminant $\Sigma=\{g=1\}$, the gap persists away from it in the manner Theorem III-A predicts, and, crucially, no finite computation decides the thermodynamic gap, exactly as Section 1.2 requires.

1.4 Relation to companion papers

This is Part III of a six-part series developing topological phases of matter in the condensed-mathematics paradigm. The parts are modular: each takes the previous as input and produces new structure.

Part I, Condensed Locality: Lieb–Robinson Estimates and Quasi-Local Dynamics on the Moduli Stack of Hamiltonians [1], supplies the $F$-function Banach space $\mathcal{B}_{F}$ and the uniform Lieb–Robinson velocity that Theorem III-C consumes; the quasi-adiabatic and automorphic-equivalence tools it formalizes are the ones Conjecture III-3 invokes.

Part II, Positivity, $C^*$-Norms, and Condensed State Spaces of Quasi-Local Algebras [2], provides the infinite-volume GNS framework in which the thermodynamic gap of Section 2 is defined, and the crossed-product observable algebras through which the disorder hull of Conjecture III-2 acquires its $K$-theory.

Part IV, From Lattice Models to Effective Field Theories: Stabilization and the Invertible Condensed Phase Spectrum [4], is the immediate consumer of this paper: it group-completes the invertible sector of $\mathfrak{Phase}_{d,G}=(\mathfrak{Gap}_{d,G}[\mathcal{W}^{-1}])^{\mathrm{st}}$ into $\mathbf{IP}^{\mathrm{cond}}_{d,G}$. Its constructions presuppose that $\mathfrak{Gap}_{d,G}$ is a well-behaved, stable substack, which is what Theorems III-A–III-C, on their stratum, secure.

Part V, Physical Realizability of Bordism and Homotopy Classes by Gapped Lattice Systems [13], asks which abstract classes are realized by uniformly gapped lattice families; the “uniformly gapped” hypothesis it uses is the one defined here.

Part VI, Topological Phases of Matter in the Condensed-Mathematics Paradigm: A Modular Research Program [14], assembles the five modules and states the global Master Conjecture, of which Conjectures III-1–III-3 are the gap-theoretic components.

2 Mathematical framework

We recall only what is needed and fix notation consistent with the series. Details of the condensed formalism are in [15,16] and Part I; the operator-algebraic framework is in [12] and Part II.

2.1 Condensed probes and the Hamiltonian stack

A condensed set is a sheaf on the site of profinite sets with finite, jointly-surjective covers [15]. For a topological space $X$ its condensation $\underline{X}$ is defined on a profinite probe $S$ by $\underline{X}(S)=\operatorname{Cont}(S,X)$; on compactly generated spaces $X↦\underline{X}$ is fully faithful, so ordinary parameter spaces, tori, Banach spaces such as $\mathcal{B}_{F}$, and compact disorder hulls all embed into a category with good algebraic and homological behavior. We use light condensed sets throughout to avoid set-theoretic size issues in the profinite inverse limits [17].

For a profinite probe $S$, an element of $\mathfrak{Ham}_{d,G}(S)$ is an $S$-continuous family $s↦\Phi_s$ of $G$-symmetric quasi-local interactions with finite $F$-norm: equivalently, by Part I, a continuous map $S\to\mathcal{B}_{F}$ landing in the $G$-symmetric interactions. Each interaction $\Phi$ assigns to a finite $X\subset L$ a self-adjoint $\Phi(X)$ supported on $X$, and generates finite-volume Hamiltonians $H_\Lambda(\Phi)=\sum_{X\subseteq\Lambda}\Phi(X)$ for finite $\Lambda\subset L$ and a thermodynamic-limit dynamics and ground-state theory in the GNS sense of Part II. We write $H$ for the interaction, $H_s$ for the member at $s\in S$, and $H_\Lambda$ for a finite-volume restriction when the interaction is fixed.

2.2 Finite-volume gap and thermodynamic gap

Two notions of gap appear, and their relationship is where the subtlety lives.

At finite volume $\Lambda$, $H_\Lambda$ is a self-adjoint operator on the finite-dimensional Hilbert space $\mathcal{H}_\Lambda=\bigotimes_{x\in\Lambda}\mathcal{H}_x$. Let its eigenvalues, counted with multiplicity and in non-decreasing order, be $\lambda_0(H_\Lambda)\le\lambda_1(H_\Lambda)\le·s$. For an integer $n\ge 1$ we define the $n$-gap $$\operatorname{gap}_{\Lambda,n}(H)\;=\;\lambda_n(H_\Lambda)-\lambda_0(H_\Lambda),$$ the spectral gap above the lowest $n$ levels. When the ground state is non-degenerate one takes $n=1$; when there is a ground-state space of dimension $m$ (as in topological order or symmetry breaking) the physically relevant gap is the $m$-gap, the distance from the ground-state sector to the first genuinely excited level. The device of a fixed integer $n$ sidesteps the discontinuity that a changing ground-state multiplicity would otherwise introduce; we return to this in Section 4.

At the thermodynamic level, following Part II and [9,12], one works with an infinite-volume ground state $\omega$, its GNS representation $(\mathcal{H}_\omega,\pi_\omega,\Omega_\omega)$, and the GNS Hamiltonian $H_\omega\ge 0$ implementing the dynamics with $H_\omega\Omega_\omega=0$. The thermodynamic gap is $$\operatorname{gap}(H)\;=\;\sup\{\,\gamma\ge 0\,:\,\operatorname{spec}(H_\omega)\cap(0,\gamma)=\varnothing\,\},$$ the width of the spectral gap of $H_\omega$ above $0$ (with $\operatorname{gap}(H)=0$ if no such $\gamma>0$ exists). This is the quantity appearing in (1). When a system has a uniform finite-volume gap—$\operatorname{gap}_{\Lambda,m}(H)\ge\gamma$ for all $\Lambda$ with $\gamma$ independent of $\Lambda$—the thermodynamic gap satisfies $\operatorname{gap}(H)\ge\gamma$; but the converse can fail, and a thermodynamic gap need not be accompanied by a uniform finite-volume lower bound taken as an axiom. The Nachtergaele–Sims–Young framework [9] is designed precisely to work with the infinite-volume gap directly rather than to assume uniform finite-size bounds, and we adopt its stance.

2.3 The uniformly gapped substack

The filtration by $\Delta$ is genuine: the strata $\mathfrak{Gap}_{d,G}^{\ge\Delta}$ are nested, $\mathfrak{Gap}_{d,G}^{\ge\Delta'}\subseteq\mathfrak{Gap}_{d,G}^{\ge\Delta}$ for $\Delta'\ge\Delta$, and a family can belong to the union without belonging to any single stratum uniformly as the probe varies. The following elementary observation records the sheaf-theoretic shape of the object and will be sharpened, conjecturally, in Section 7.

2.4 What “open substack” should mean

Because $\mathcal{B}_{F}$ is a Banach space, its condensation $\underline{\mathcal{B}_{F}}$ is a condensed $\mathbb{R}$-vector space and $\mathfrak{Ham}_{d,G}$ sits over it. Openness of $\mathfrak{Gap}_{d,G}$ inside $\mathfrak{Ham}_{d,G}$ can be phrased in the condensed setting as follows. Call a subobject $U\subseteq\underline{\mathcal{B}_{F}}$ condensed-open along a direction class $\mathcal{D}$ if for every point $\Phi\in U$ there is $\varepsilon>0$ such that every $\Psi$ with $\Psi-\Phi\in\mathcal{D}$ and $\lVert \Psi-\Phi \rVert_{F}<\varepsilon$ again lies in $U$, and if this holds compatibly for probes: pulling back along any $g:S\to\underline{\mathcal{B}_{F}}$ whose image lies in the $\varepsilon$-ball around $\Phi$ lands in $U(S)$. For a full topology on $\mathfrak{Ham}_{d,G}$ one wants $\mathcal{D}$ to be all quasi-local directions and $\varepsilon$ uniform; the honest situation, developed in Section 5, is that we can prove condensed-openness of $\mathfrak{Gap}_{d,G}$ only on a stratum and only along the quasi-local perturbation class for which the stability theorems apply. Recording this scope precisely is not a weakness of the result but the exact shape undecidability forces the result to take.

3 The undecidability wall and the definitional stance

3.1 The Cubitt–Pérez-García–Wolf theorem

We restate the constraint precisely, in the form we will use.

Two features of Theorem 3.1 matter for us. First, the hardness is already present for translation-invariant, nearest-neighbour, two-dimensional systems—the most classical and most physical corner of $\mathfrak{Ham}_{d,G}$—so one cannot escape it by restricting to a tame-looking subclass defined by locality or symmetry alone. Second, the separation between the gapped and gapless cases is sharp, which forecloses the hope that some “soft” or approximate criterion could decide the gap up to an $\varepsilon$: the difficulty is not analytic delicacy near a threshold but genuine computational undecidability.

The undecidability persists in one spatial dimension [18] and under various symmetry restrictions, so it is not an artifact of high dimension. It is, however, a fragile phenomenon, in a way consonant with the viewpoint of this paper: the undecidable families are exactly the ones that are not stably gapped. Castilla-Castellano and Lucia [19] show that an arbitrarily small local perturbation of the one-dimensional construction restores decidability, by breaking the fine-tuned energy cancellation on which the halting encoding depends. Undecidability lives on the boundary between gapped and gapless— precisely the locus the stability theory of Section 5 stays away from, and precisely why that theory must assume, rather than certify, membership in the gapped substack.

3.2 Consequences for the substack

Theorem 3.1 has immediate and non-negotiable consequences for how $\mathfrak{Gap}_{d,G}$ may be studied.

Consequently, “$H\in\mathfrak{Gap}_{d,G}$” cannot be the output of a general computation; it is an input. We adopt the following stance, which we regard as forced rather than chosen.

Definitional stance. The uniformly gapped substack $\mathfrak{Gap}_{d,G}$ is defined by the gap hypothesis, not detected by a criterion. Membership is a hypothesis one places on a family. The theorems of this paper take the form “if a family (or a stratum point) is uniformly gapped and satisfies stated structural hypotheses, then its neighborhood, correlations, and descent behavior are controlled.” No theorem asserts, for a general family, that the hypothesis holds.

3.3 Why the program survives the wall

It might seem that undecidability is fatal to a program that puts $\mathfrak{Gap}_{d,G}$ at its center. It is not, for the same reason that undecidability of the halting problem is not fatal to the theory of computation: one develops the theory of the objects assuming the property, and proves structural and stability results about the class so defined. The moduli-theoretic viewpoint is well suited to this. A stack does not need a decision procedure for membership to be a useful object; it needs good functorial behavior, a sensible topology, and stability of its distinguished subobjects under the maps one cares about. Those are precisely what Theorems III-A–III-C provide, on the stratum where they can be proved. The undecidability wall bounds the global reach of the theory (one cannot hope for a single computable classification of all gapped systems), but it leaves the local and stratified theory intact, and it is the local and stratified theory that feeds Parts IV and V.

4 Finite-volume gap continuity

We begin with the one unconditional result: at finite volume, the gap is a Lipschitz function of the interaction, and the gapped locus is open. This is the finite-volume shadow of the openness we would like for $\mathfrak{Gap}_{d,G}$, and it is the part of the picture undecidability leaves entirely alone.

4.1 Weyl continuity of eigenvalues

To pass from operator norm to the interaction $F$-norm we use finite-volume norm equivalence, which is the elementary end of the Lieb–Robinson machinery of Part I.

4.2 Theorem III-A

4.3 Semicontinuity of the thermodynamic gap

At the thermodynamic level the gap is only lower semicontinuous in general, and even that requires care about the mode of convergence. We record the statement we can make and mark the pitfall.

5 Stability on the frustration-free / LTQO stratum

We now reach the heart of the paper: the recasting of the Bravyi–Hastings–Michalakis, Michalakis–Zwolak and Nachtergaele–Sims–Young stability theorems as a statement that $\mathfrak{Gap}_{d,G}$ contains condensed-open neighborhoods on a well-defined stratum. Throughout, we are scrupulous about hypotheses; the stratum is exactly where the hypotheses hold with constants uniform in the system size.

5.1 Frustration-freeness and local topological quantum order

Frustration-freeness is a genuine restriction. Generic gapped systems are not frustration-free; the class includes the paradigmatic exactly-solvable models (Kitaev’s toric code and honeycomb model [21], AKLT-type chains, group-cohomology fixed-point models [22]) and the parent Hamiltonians of matrix-product and PEPS states, but it excludes, for instance, generic Chern insulators. We never extend a frustration-free theorem to a non-frustration-free system.

The LTQO condition is the operator-algebraic expression of “topological” ground-state degeneracy: the degenerate ground states cannot be told apart by any local measurement. Michalakis and Zwolak [8] showed that, together with a uniform local gap, LTQO is essentially equivalent to stability of the gap and to an area law for the ground-state entanglement.

By construction $\mathfrak{S}\subseteq\mathfrak{Gap}_{d,G}$: the uniform finite-volume gap $\gamma$ forces $\operatorname{gap}(\Phi)\ge\gamma$ (Section 2.2). The stratum is the domain of the stability theorems.

5.2 The stability theorems, as cited

We quote the results we use in the form relevant here; the constants are those of the original sources.

The three results are complementary: [6,7] establish the finite-volume, size-independent threshold; [8] identifies LTQO as the precise condition and connects it to the area law; [9] lifts the conclusion to the thermodynamic gap directly. What they share, and what we need, is a single quantity: a perturbation radius $\varepsilon_0>0$ that does not shrink with system size.

5.3 Theorem III-B: stability as condensed-openness

5.4 Spectral flow and automorphic equivalence

The stability of Section 5.3 comes with a dynamical companion: within a uniformly gapped neighborhood the ground-state spaces are related by a quasi-local automorphism, the spectral flow. This is the rigorous meaning of “the same phase” along a gapped path, and it is the interface with Part I and with the equivalence class $\mathcal{W}$.

6 Uniform clustering from a uniform gap

The third theorem is structural: it isolates a property that uniformly gapped families have and pointwise-gapped families need not. A gap forces exponential decay of correlations; a uniform gap forces the decay rate to be uniform over the base. This is the sense in which $\mathfrak{Gap}_{d,G}$, and not the naive pointwise-gapped locus, is the correct object.

6.1 Gap implies clustering

The precise exponent and constants are as in [10]; the Lieb–Robinson input is the exponential-clustering theorem of [11]. We use them as cited.

6.2 Theorem III-C: uniformity over the base

7 The structure of the gapped substack: conjectures

Beyond the stratum, we do not have theorems; we have conjectures, and we state them as such, with the reasons they are plausible and the obstacles to proving them. Each is scoped to respect the undecidability wall.

7.1 Openness in the full condensed topology

Evidence and obstruction. Theorem III-B is exactly this statement restricted to $\mathfrak{S}$ and to quasi-local directions; the conjecture asserts that the phenomenon persists on the larger physical stratum and in all directions. The obstruction to a proof is precisely undecidability: openness in the full $\mathfrak{Ham}_{d,G}$, with a uniform radius, would let one certify gaps of nearby systems, and near a CPW-type embedding [5] the gap can turn on and off in a way no uniform radius can survive. Restricting to the physical stratum is what excludes those pathological directions; making “physical stratum” precise, so that the restriction is both honest and non-vacuous, is the mathematical work the conjecture packages. The finite-quotient detection clause is a descent statement and is closely tied to Conjecture III-2.

7.2 Uniform-gap descent along profinite covers

The disorder-hull thread of the program ([28,29] and Part II) makes the profinite structure of $S$ physical: for a finite local-configuration set $Q$ (the letter $Q$ denotes the disorder alphabet, kept deliberately distinct from the locality $F$-function of Part I) the hull $\Omega=Q^{\mathbb{Z}^d}$ is compact and totally disconnected, hence profinite, and a disordered family $\omega↦ H_\omega$ is literally a $\Omega$-point of $\mathfrak{Ham}_{d,G}$. Finite quotients of $\Omega$ are finite-resolution disorder data.

Evidence and obstruction. The “only if” direction is Proposition 2.3: a uniform gap over $\Omega$ restricts to a uniform gap over each $\Omega_i$. The content is the “if” direction—that compatible finite-resolution gaps at a common $\Delta$ assemble to a uniform gap over the limit. This is a genuine descent question. The optimistic reason to expect it: continuity of the gap along $\mathcal{B}_{F}$-convergent families (Theorem III-A at each finite volume) plus compactness of $\Omega$ suggests that a common finite-resolution bound should survive the limit. The obstruction: the gap is a thermodynamic, not finite-volume, quantity, and Proposition 4.4 shows only lower semicontinuity in general; the descent could fail if the low-lying spectrum reorganizes in the inverse limit. The conjecture asserts it does not, on the disorder hulls, at fixed $\Delta$. A proof would presumably route through the crossed-product algebra $C(\Omega)\rtimes\mathbb{Z}^d$ of Part II and a gap-labelling argument in the manner of Bellissard [28].

7.3 Quasi-adiabatic continuation realizes the equivalence class

Evidence and obstruction. Proposition 5.10 proves one direction locally: a uniformly gapped segment in a stratum gives a quasi-local $\mathcal{W}$-equivalence. The theory of automorphic equivalence in [26] and the classification program of [12] make the target set precise in $d=1$ and, for on-site finite symmetry, in $d=2$. The obstruction to the full statement is twofold: a general uniformly gapped path need not remain in a single stratum, so Proposition 5.10 does not immediately apply along it; and the identification of $\pi_0\operatorname{Shape}$ of the localized stack with the operator-algebraic set requires knowing that the condensed shape does not see more (or less) than the operator-algebraic equivalence, which is a comparison between two homotopy theories that has not been carried out. In low dimension, where [12] gives complete invariants, the conjecture is closest to reach; in higher dimension it is genuinely open, in step with the general state of the classification problem.

8 Numerical illustration: the
transverse-field Ising chain

We close the mathematical development with a concrete computation that makes the abstract picture visible: the spectral gap of the transverse-field Ising chain, computed by exact diagonalization. It renders the gapless discriminant $\Sigma$ as a one-dimensional picture and exhibits the stability that Theorem III-A quantifies. The accompanying Haskell program (Section 8.5) performs the computation and checks the properties as executable QuickCheck tests.

8.1 The model

On a chain of $N$ spins with open boundary conditions, consider $$H(g)\;=\;-\sum_{i=1}^{N-1} Z_i Z_{i+1}\;-\;g\sum_{i=1}^{N} X_i, \qquad g\in[0,2],$$ where $X_i,Z_i$ are Pauli operators at site $i$. The model has a global $\mathbb{Z}_2$ symmetry $P=\prod_i X_i$, $[H(g),P]=0$, and a quantum phase transition at $g=1$ separating a ferromagnetic phase ($g<1$, two-fold degenerate ground sector in the thermodynamic limit, the symmetry-broken states) from a paramagnetic phase ($g>1$, unique ground state). By the Jordan–Wigner transformation the model is a free-fermion system with single-particle dispersion $$\varepsilon_k(g)\;=\;2\sqrt{\,1+g^2-2g\cos k\,},$$ whose minimum over $k$ is $\min_k\varepsilon_k(g)=2\lvert 1-g\rvert$, the thermodynamic single-particle gap. We use open boundary conditions: they keep the sparse structure and the $\mathbb{Z}_2$ symmetry of the matrix simple and avoid the fermion-parity (Neveu–Schwarz / Ramond) sector bookkeeping that periodic boundaries introduce through the Jordan–Wigner transformation. Periodic boundaries would reduce finite-size edge effects, but the qualitative gap picture—the discriminant at $g=1$ and stability away from it—is independent of the boundary condition in the thermodynamic limit. Thus $$\Sigma=\{\,g\,:\,\operatorname{gap}(H(g))=0\,\}=\{\,g=1\,\},$$ a single point of the parameter interval: the gapless discriminant of this one-parameter family. Crossing $g=1$ is the topological/quantum phase transition; away from it the system is gapped, with gap growing linearly, $\approx 2\lvert 1-g\rvert$. The exact gap also exhibits the Kramers–Wannier self-duality $\min_k\varepsilon_k(g)=g\,\min_k\varepsilon_k(1/g)$, immediate from (5) since $2\lvert 1-g\rvert=g· 2\lvert 1-1/g\rvert$; the Haskell suite of Section 8.5 checks this identity.

8.2 Exact diagonalization and the discriminant

In the computational ($Z$-) basis, (4) is a real symmetric $2^N× 2^N$ matrix: the $ZZ$ term is diagonal, the transverse field $X_i$ is off-diagonal (a single spin flip). We diagonalize it exactly for small $N$ by a cyclic Jacobi eigenvalue routine, sweeping $g$ across $[0,2]$ and recording the low-lying spectrum. The correctness of the eigensolver is not taken on faith: the QuickCheck properties of Section 8.5 check the exact trace identities $$\sum_j\lambda_j(g)=\operatorname{Tr}H(g)=0, \qquad \sum_j\lambda_j(g)^2=\operatorname{Tr}H(g)^2=\big((N-1)+N g^2\big)\,2^N,$$ which follow from $\operatorname{Tr}(Z_iZ_{i+1})=\operatorname{Tr}(X_i)=0$ and the orthogonality of the Pauli strings, and which a wrong spectrum would violate. The cyclic Jacobi routine drives the off-diagonal Frobenius norm below $10^{-11}$, and for the dimensions used ($N\le 8$, matrices up to $256×256$) the eigenvalues are accurate to near machine precision—the residuals in the identities (7) stay below $10^{-9}$ across this range, reaching $\sim 10^{-13}$ at $N=4$. At substantially larger $N$ both the accumulated floating-point error and the $O(2^{3N})$ cost of dense diagonalization grow, which is why we cap the dense computation at $N=8$ and read the large-$N$ trend from the analytic reference (5) instead.

The finite-$N$ gap $\operatorname{gap}_{N}(g)=\lambda_1(g)-\lambda_0(g)$ and the $2$-gap $\lambda_2(g)-\lambda_0(g)$ together render (6). In the paramagnetic phase $g>1$ the ground state is unique and $\operatorname{gap}_{N}(g)$ is the physical gap, converging to $2(g-1)$ as $N$ grows. In the ferromagnetic phase $g<1$ the two lowest levels form the near-degenerate ground sector (their splitting decays exponentially in $N$), so the physical gap is the $2$-gap—precisely the $m$-gap of Remark 4.3 with $m=2$. In both phases the relevant gap develops a sharp minimum near $g=1$ that deepens with $N$: the finite-size fingerprint of the discriminant point (6). Figure 1 sketches the shape.

8.3 Perturbative stability away from criticality

To illustrate Theorem III-A and the stability picture, the program performs a perturbation experiment. Fix $g=1.5$, deep in the gapped paramagnetic phase, and add a random local field $V=\lambda\sum_i s_i Z_i$ with $s_i\in\{\pm 1\}$ and small $\lambda$. Exact diagonalization confirms $$\big\lvert\operatorname{gap}_{N}(H(1.5)+V)-\operatorname{gap}_{N}(H(1.5))\big\rvert\;\le\;2\,\lVert V \rVert \;\le\;2\lambda N,$$ the finite-volume Lipschitz bound of Theorem III-A (here $C_\Lambda$ absorbed into the per-site normalization). The gap stays bounded away from $0$ for all sampled perturbations: away from the discriminant the gapped locus is open and stability holds, exactly as Theorem III-A guarantees at finite volume. Near $g=1$, by contrast, the same size of perturbation moves the small gap by a comparable amount and can push the finite system across the rounded minimum—the numerical trace of why openness of the substack is a statement scoped away from the discriminant.

8.4 What the finite computation cannot do

It is essential to state what Figure 1 does not establish, on pain of contradicting Section 3. No finite-$N$ computation decides the thermodynamic gap. For the transverse-field Ising chain we happen to know the answer analytically—the model is exactly solvable, (5)—so the discriminant (6) is certain; but that certainty comes from the exact Jordan–Wigner solution, not from the diagonalization. For a general translation-invariant family the analogous finite-$N$ data would be genuinely uninformative about the thermodynamic limit, because by Theorem 3.1 no finite computation, however large $N$, can decide whether the gap survives. The Ising chain is an illustration of the structure of the gapped substack—its discriminant, its stability away from criticality, the Lipschitz bound— not a method for detecting gaps. The distinction is the entire moral of Part III.

8.5 The Haskell computation

The computation is implemented in Haskell in src/spectral-gap-stability/. The module TFIM builds the Hamiltonian (4) as a real symmetric matrix and diagonalizes it by cyclic Jacobi rotations; Main sweeps $g$ over $[0,2]$ for $N\in\{4,6,8\}$, prints the gap table, runs the perturbation experiment of Section 8.3, and exits with status $0$ on success. The module Properties encodes as QuickCheck properties: the exact trace identities (7) (which double as a validation of the eigensolver); symmetry of the Hamiltonian matrix; positivity of the gap in the paramagnetic phase at fixed $N$ (Section 8.2); the finite-volume Lipschitz bound (8) of Theorem III-A; and the monotone growth of the gap with $g$ in the paramagnetic phase. The properties are stated one per claim and cite the corresponding equation or theorem, so that the executable tests track the paper’s assertions.

9 Discussion

9.1 What the condensed paradigm contributes here

Part III does not prove a new stability theorem; the analytic content is [6–11]. Nor is the idea of studying the moduli space of gapped Hamiltonians and its topology new: it is developed, in ordinary topology and with effective-field-theory methods, by Hsin, Wang and collaborators [30]. What the condensed framing contributes is organizational, and it is not empty. First, it makes “uniformly gapped” a property of a family in a category where families, disorder hulls, and inverse limits are first-class objects: the $\mathop{\mathrm{inf}}_s$ of Definition 2.2 is the condition that a subobject of a condensed stack is cut out correctly, and Theorem III-C shows this is the condition with structural teeth. Second, it identifies exactly which classical theorems give openness of the distinguished subobject (Theorem III-B) and on which stratum, so that the downstream constructions of Parts IV and V know precisely what they may assume. Third, it places the undecidability wall where it belongs (as a statement about global decidability that leaves the local, stratified, functorial theory intact) rather than letting it either be ignored or overstated into paralysis.

9.2 The two walls

The program has two hard external constraints. The first is the undecidability of the gap [5], which is the governing constraint of this paper: it is why existence of a thermodynamic gap is a hypothesis and not a theorem, why openness is a stratum statement, and why Section 8.4 insists that no finite computation detects the gap. The second, the Kapustin–Fidkowski obstruction [23] to commuting-projector realizations of chiral phases, governs Part V [13] and does not bear directly here, except to note that the frustration-free / LTQO stratum of Definition 5.3 (on which we can prove stability) is close to the commuting-projector world where the second wall bites. Chiral phases (nonzero Hall conductance, [31]) are gapped but are not frustration-free commuting-projector systems, so they lie outside $\mathfrak{S}$; our stability theorem says nothing about them, and it should not, because their stability is a different and harder question.

9.3 The profinite-disorder thread

The disorder hull $\Omega=Q^{\mathbb{Z}^d}$ is where the condensed and profinite structure is not a repackaging of a smooth manifold but a match to the physical configuration space [24,28,29]. Conjecture III-2 is the gap-layer instance of the program’s recurring descent question: does a finite-resolution property, holding compatibly at every finite quotient of $\Omega$ with uniform constants, descend to the profinite limit? For gaps this is delicate precisely because the gap is thermodynamic. That the question is even well posed (that a disordered family is literally a $\Omega$-point of $\mathfrak{Ham}_{d,G}$, and uniform gappedness a condition on that point) is a contribution of the condensed viewpoint, and settling it would connect the gap-labelling of [28] to the substack $\mathfrak{Gap}_{d,G}$ directly.

9.4 Limitations

We list the limitations plainly. Theorem III-B is conditional on membership in a frustration-free / LTQO stratum, which is undecidable to verify in general and is a proper subclass of gapped systems. Theorem III-C assumes the uniform Lieb–Robinson data of Part I, which restricts to the $F$-function locality class and excludes genuinely long-range interactions. Proposition 4.4 gives only lower semicontinuity, and deliberately so. The three conjectures are unproven, and Conjecture III-1 in particular cannot be strengthened to openness in the full $\mathfrak{Ham}_{d,G}$ without contradicting [5]. None of these limitations is incidental; each marks a boundary that the undecidability of the gap, or the restriction to a provable stability class, places on what can be claimed.

10 Conclusion

The uniformly gapped substack $\mathfrak{Gap}_{d,G}$ is the object on which the condensed theory of topological phases is built, and it is governed by a hard theorem: the spectral gap is undecidable, so membership cannot be the output of a computation and must be taken as a defining hypothesis. Within that constraint we proved what can be proved. At finite volume the gap is Lipschitz and the gapped locus open (Theorem III-A). On the frustration-free / LTQO stratum the classical stability theorems give a size-independent perturbation threshold, which we recast as condensed-openness of $\mathfrak{Gap}_{d,G}$ along quasi-local directions (Theorem III-B), with an explicit spectral flow realizing the equivalence class $\mathcal{W}$ (Proposition 5.10). A uniform gap forces a uniform correlation length (Theorem III-C), the structural property that distinguishes uniformly gapped families from pointwise-gapped ones and that makes $\mathop{\mathrm{inf}}_s$ the correct condition in the definition of the substack. Beyond the stratum we recorded three conjectures: openness on the physical stratum with finite-quotient detection (III-1), uniform-gap descent along profinite disorder hulls (III-2), and the identification of quasi-adiabatic path components with the operator-algebraic gapped phases (III-3), each anchored to a theorem of this paper and each scoped to respect the undecidability wall. The transverse-field Ising computation made the discriminant $\Sigma=\{g=1\}$ and the stability away from it visible, while insisting that no finite computation detects the thermodynamic gap.

The substack is now a well-understood object on its provable stratum and a precisely delineated open problem beyond it. Part IV takes $\mathfrak{Gap}_{d,G}$ as given and group-completes its invertible sector into the condensed phase spectrum $\mathbf{IP}^{\mathrm{cond}}_{d,G}$; Part V asks which classes that spectrum records are realized by uniformly gapped lattice families. Both rest on the stability established here, and both inherit its scope.

Code availability

The Haskell source for the transverse-field Ising computation of Section 8 (exact diagonalization, the base-only cyclic Jacobi eigensolver, and the QuickCheck property suite) is available at github.com/YonedaAI/topological-phases-of-matter, in src/spectral-gap-stability/.

References