---
title: Kreuzer–Skarke Axiverse Overview
url: https://www.emergentmind.com/topics/kreuzer-skarke-axiverse
type: topic
---

# Kreuzer–Skarke Axiverse Overview

Searching arXiv for relevant papers on the Kreuzer–Skarke axiverse and foundational database.
The Kreuzer–Skarke axiverse is the ensemble of axion effective theories obtained from string compactifications on Calabi–Yau threefold hypersurfaces associated with the Kreuzer–Skarke classification of four-dimensional reflexive polytopes. In the underlying toric construction, a fine, regular, star triangulation of a reflexive polytope defines a toric ambient space, and a generic anticanonical hypersurface in that space is a Calabi–Yau threefold. The resulting geometric data—Hodge numbers, intersection tensors, Chern classes, and Mori and Kähler cone information—provide the inputs from which axion multiplicities, kinetic terms, instanton actions, and low-energy couplings are inferred [1411.1418]. Subsequent work turned this toric dataset into an axiverse in the technical sense: at large \(h^{1,1}\), the geometry generically produces many Ramond–Ramond four-form axions with highly suppressed nonperturbative masses, together with distinctive statistical regularities across enormous ensembles of compactifications [1808.01282].

## 1. Database origin and toric construction

The foundational object is the Kreuzer–Skarke list of all \(473{,}800{,}776\) reflexive polytopes in four dimensions, together with \(30{,}108\) distinct Hodge pairs \((h^{1,1},h^{2,1})\). In the toric hypersurface construction, a four-dimensional reflexive polytope \(\Delta\) and its dual \(\Delta^\ast\) encode singular ambient toric varieties in which Calabi–Yau threefold hypersurfaces can appear. A triangulation of \(\Delta^\ast\) yields a simplicial toric fourfold \(A\), and the Calabi–Yau threefold \(X\subset A\) is realized as an anticanonical divisor [1411.1418].

The companion database to the toric Calabi–Yau construction reports, for each threefold obtained from such triangulations, the Hodge numbers \(h^{1,1}(X)\), \(h^{2,1}(X)\), the Euler number \(\chi(X)=2(h^{1,1}-h^{2,1})\), the triple intersection numbers \(\kappa_{ijk}\), the second Chern class \(c_2(X)\), Mori cone generators, Kähler cone inequalities, toric divisor classes, the Stanley–Reisner ideal, ambient Chern classes, and triangulation-specific cone data. It also distinguishes between different triangulations that yield genuinely distinct Calabi–Yau geometries and triangulations that represent different Kähler phases of the same topology, gluing the latter when topological equivalence is established via matching of invariants such as \(c_2\) and \(\kappa_{ijk}\) [1411.1418].

The explicitly computed subset is complete for \(h^{1,1}\le 6\) and partial for \(h^{1,1}=7\), comprising \(23{,}568\) polytopes, \(101{,}673\) threefolds, and about \(85{,}000\) favorable geometries. The dependence on triangulation is not a minor technicality. For the \(h^{1,1}=3\) polytope \(\#95\), the database finds two distinct glued geometries with identical Hodge numbers \(h^{1,1}=3\), \(h^{2,1}=81\), but different \(\kappa_{ijk}\), \(c_2(X)\), and Mori/Kähler cones; one geometry is assembled from two triangulations, the other from one. This directly shows that Hodge data alone do not determine axion physics [1411.1418].

At the classification level, the Kreuzer–Skarke program relies on canonical normal forms for lattice polytopes. The PALP normal form was used by Kreuzer and Skarke to classify \(3\)D and \(4\)D reflexive polytopes, preventing double-counting and organizing the database by isomorphism classes. This infrastructure does not itself address axion physics, but it is essential for the uniqueness and scalability of the datasets on which KS axiverse studies depend [1301.6641].

## 2. Topological data, mirror symmetry, and cone geometry

For toric hypersurfaces, the Hodge numbers are determined combinatorially by Batyrev’s formulas. If \(l(\cdot)\) denotes the number of lattice points and \(l^\ast(\Theta)\) the number of interior lattice points of a face \(\Theta\), then for reflexive \(\Delta\) and \(\Delta^\ast\),
\[
h^{1,1}(X) = l(\Delta^\ast) - 5 - \sum_{\text{codim }\Theta^\ast = 1} l^\ast(\Theta^\ast) + \sum_{\text{codim }\Theta^\ast = 2} l^\ast(\Theta^\ast)\, l^\ast(\Theta),
\]
\[
h^{2,1}(X) = l(\Delta) - 5 - \sum_{\text{codim }\Theta = 1} l^\ast(\Theta) + \sum_{\text{codim }\Theta = 2} l^\ast(\Theta)\, l^\ast(\Theta^\ast).
\]
Mirror symmetry exchanges \(\Delta \leftrightarrow \Delta^\ast\), so \(h^{1,1}(X)=h^{2,1}(X^\vee)\) and \(h^{2,1}(X)=h^{1,1}(X^\vee)\) [1411.1418].

For favorable polytopes, inherited toric divisors generate \(H_4(X)\) and the effective cone studied in large-scale axiverse analyses. This favorable sector is especially important because it provides explicit divisor representatives and intersection data suitable for systematic scans [1808.01282].

A major structural result of the KS axiverse is that the Kähler cone becomes very narrow at large \(h^{1,1}\). In the large-\(h^{1,1}\) study of \(2{,}031{,}335\) Calabi–Yau hypersurfaces, including at least one triangulation for every polytope with \(240\le h^{1,1}\le 491\), the ambient-space Mori cone \(\mathcal M_V\) and its dual Kähler cone \(K_V\) were computed for all models, while triple intersection numbers were computed for \(h^{1,1}\le 100\). An outer approximation \(K_\cap\) to the Calabi–Yau Kähler cone \(K_X\) is defined by requiring positivity of the volumes of \(X\), all inherited divisors, and all toric intersection curves, yielding
\[
K_V \subset K_X \subset K_\cap .
\]
The corresponding stretched cones \(\widetilde K_V \subset \widetilde K_X \subset \widetilde K_\cap\) impose the condition that relevant cycle volumes are at least \(1\) in string units [1808.01282].

Quantitatively, the narrowness is expressed through the growth of the minimal Euclidean distance from the origin to the stretched cones:
\[
d_{\min}^V \propto (h^{1,1})^{3.1}, \qquad d_{\min}^\cap \propto (h^{1,1})^{2.5}.
\]
Under the condition that every effective curve has volume \(\ge 1\), the typical divisor and threefold volumes grow as powers of \(h^{1,1}\):
\[
\tau_{\mathrm{last}}^\cap \propto (h^{1,1})^{3.2}, \qquad \tau_{\mathrm{last}}^V \propto (h^{1,1})^{4.3},
\]
\[
\mathcal V^\cap \propto (h^{1,1})^{6.2}, \qquad \mathcal V^V \propto (h^{1,1})^{7.2}.
\]
Because the number of nonvanishing \(\kappa_{ABC}\) grows only linearly with \(h^{1,1}\), and the RMS size of nonzero intersections is roughly independent of \(h^{1,1}\), this scaling is attributed primarily to cone geometry rather than to explosive growth in intersection combinatorics [1808.01282].

## 3. From toric geometry to axion effective theories

The Kähler form is expanded in a divisor basis as \(J=t^i[D_i]\), and the toric data determine the standard geometric quantities
\[
\mathcal V=\frac{1}{6}\,\kappa_{ijk}\,t^i t^j t^k,\qquad
\tau_i=\frac{1}{2}\,\kappa_{ijk}\,t^j t^k,\qquad
\kappa_{ijk}=\int_X J_i\wedge J_j\wedge J_k.
\]
The Calabi–Yau condition gives \(c_1(X)=0\), while the second Chern class is obtained from the ambient toric data by adjunction, with \(c_2(X)=c_2(TA)|_X\) in the formulation used in the database [1411.1418].

In type IIB orientifolds, the principal axions of the KS axiverse arise from the Ramond–Ramond four-form \(C_4\) reduced on four-cycles. Writing the complexified Kähler moduli as \(T_i=\tau_i+i\phi_i\) or \(T_i=\tau_i+i\theta_i\), the axions are the imaginary parts of these moduli. The tree-level Kähler potential is
\[
K=-2\log \mathcal V,
\]
and the axion kinetic matrix is derived from the corresponding Kähler metric. The nonperturbative superpotential and potential are taken schematically as
\[
W = W_0 + \sum_p A_p e^{-2\pi Q_i^p T^i},
\]
\[
V(\theta)=\sum_I \Lambda_I^4\left[1-\cos(2\pi\,Q_i^I\theta^i+\delta_I)\right],
\]
with instanton actions scaling as \(S_I \simeq 2\pi\,Q_i^I\tau^i\). In the hierarchical approximation one uses \(m_i^2 \approx \Lambda_i^4/f_i^2\) [2507.12516].

The same toric data control the allowed large-volume regions. The Mori cone generators constrain the Kähler parameters through linear inequalities \(K^i{}_j t^j\ge 0\), so \(\kappa_{ijk}\) and the Kähler cone together determine the reachable \(\tau_i\), and hence the instanton actions and decay constants. The database quantity \(c_2(X)\cdot D_i\) can be combined with \(\kappa_{ijk}\) to compute
\[
\chi(\mathcal O_D)=\frac{1}{12}(D^3 + c_2(X)\cdot D),
\]
which is used as a diagnostic for divisor rigidity in instanton analyses [1411.1418].

Axion counting depends on compactification details. In a simplifying type IIB setup often used in KS ensemble studies, the orientifold is assumed not to project out Kähler moduli, so the number of axions is taken to be \(N=h^{1,1}\). More generally, the orientifold splits \(h^{1,1}\) into \(h^{1,1}_+\) and \(h^{1,1}_-\), affecting multiplicities, kinetic terms, and couplings, and the KS hypersurface database by itself does not supply the involution needed to determine that splitting. In type IIA orientifolds, axions predominantly arise from \(C_3\) reduced on three-cycles, with count \(b_3=2h^{2,1}+2\) up to orientifold parities [2507.12516].

## 4. Universal statistics in the KS axiverse

A later development was the identification of universal divisor-volume statistics across the KS landscape. In the large-volume regime, restricting to favorable toric hypersurfaces and prime toric divisors inherited from the ambient toric variety, one defines normalized divisor volumes by first rescaling along a Kähler-cone ray so that \(\min_I \tau^I=1\), and then setting
\[
\hat{\tau}^I := \frac{\log \tau^I}{\langle \log(\vec{\tau})\rangle_X}, \qquad
\langle \log(\vec{\tau})\rangle_X = \frac{1}{h^{1,1}+4}\sum_{I=1}^{h^{1,1}+4}\log \tau^I.
\]
Sampling favorable polytopes with \(h^{1,1}\in\{20,30,40,50,60,80,100,120,150,180,200\}\), with \(10^4\) polytopes per \(h^{1,1}\) for \(h^{1,1}\le 120\) and all favorable polytopes above that threshold, the resulting normalized distributions were found to be universal across geometries, triangulations, and random positions in the extended Kähler cone. The evidence was quantified using \(1\)-Wasserstein distances and adjusted \(r^2\), with pairwise percentage errors in \(W_1\) typically below \(\approx 17\%\) and decreasing with \(h^{1,1}\) [2507.12516].

The total normalized distribution develops a second peak at moderate \(h^{1,1}\), explained by superposition of sub-distributions associated with three toric divisor types: vertex, edge, and face. These sub-distributions are themselves universal, while the prevalence of face divisors grows with \(h^{1,1}\) [2507.12516].

To model this statistically, the paper proposes a geometry-independent single-function fit using the Gaussian Orthogonal Ensemble level-spacing distribution,
\[
p_{\mathrm{fit}}(\hat{\tau}) = \frac{\pi}{2}\,\hat{\tau}\,e^{-(\pi/4)\hat{\tau}^2},
\]
together with an \(h^{1,1}\)-dependent mean
\[
\langle \log(\vec{\tau})\rangle_{\mathrm{fit}}
= a - b\,(\log_{10} h^{1,1})^{-c},
\qquad
a \approx 3.433,\; b \approx 5.404,\; c \approx 4.050.
\]
Within the approximations adopted—large volume, single-instanton dominance, near-diagonal dominance in the Kähler metric, a simplified divisor basis, and volume domination by the largest \(2\)-cycle parameter—this model reproduces distributions of decay constants, masses, and the stringy PQ-quality estimator with good accuracy; for \(\Delta\theta_{\mathrm{stringy}}\), the fitted and exact ensemble distributions match with \(<5\%\) \(W_1\) percentage error [2507.12516].

A plausible implication is that much of the KS axiverse is governed statistically by divisor-volume hierarchies rather than by detailed, geometry-by-geometry inversion of the full Kähler metric. The universality claim, however, is explicitly tied to the large-volume regime and to generic saxion vevs in the extended Kähler cone, not to finely tuned stabilization scenarios such as KKLT- or LVS-like regions where many \(\tau_i\) are forced to be comparable [2507.12516].

## 5. Phenomenology and observational windows

The KS axiverse has been used to study axion couplings and masses in explicit large ensembles. In type IIB toy models constructed from the KS database, axion–photon couplings were found to be systematically suppressed relative to the inverse axion periodicity for two distinct reasons. First, couplings to the QED \(\theta\)-angle are suppressed for axion mass eigenstates that are light compared to the scale set by stringy instantons on the cycle supporting QED. Second, at large \(h^{1,1}\) the intersection matrix is sparse, so kinetic mixing is weak. In an ensemble of approximately \(200{,}000\) toy models constructed from \(131\) polytopes at \(h^{1,1}=50,100,200,491\), this yields many effectively invisible axions and only a handful with potentially observable photon couplings. The resulting phenomenology includes freeze-in production and decay of thermal axions, cosmic microwave background birefringence, X-ray spectrum oscillations, and supernova constraints on the QCD axion [2309.13145].

The same paper finds that the number of axions per model with \(g_{a\gamma\gamma}>g_{\mathrm{pl}}\), where \(g_{\mathrm{pl}}\equiv \alpha_{\mathrm{EM}}/(2\pi M_{\mathrm{pl}})\), remains small even at large \(h^{1,1}\). The maximum \(g_{a\gamma\gamma}\) increases with \(h^{1,1}\) because decay constants decrease, but the fraction of visible axions shrinks because kinetic isolation strengthens with \(h^{1,1}\). Under the assumptions used there, the QCD axion mass often reaches \(O(1\text{–}10)\,\mathrm{meV}\) at \(h^{1,1}=491\), near current SN1987A bounds [2309.13145].

A separate KS scan imposed grand-unification and proton-decay constraints on type IIB O3/O7 orientifolds. Sampling up to \(100\) favorable polytopes and \(500\) triangulations per polytope for each \(h^{1,1}\le 60\), producing approximately \(1.7\times 10^6\) compactifications, the study fixed a random prime toric divisor to host QCD, rescaled to \(\mathrm{Vol}(D_{\mathrm{QCD}})=25\), discarded geometries with any divisor volume \(<1\), and evaluated EFT quantities at the tip of the stretched Kähler cone. Imposing \(M_s\ge M_{\mathrm{GUT}}\), the fraction of compactifications satisfying this condition drops to zero for \(h^{1,1}\gtrsim 48\), leading to the conclusion that the KS axiverse compatible with these requirements contains at most \(\approx 47\) closed-string axions. In this restricted ensemble the QCD axion mass lies within
\[
3\times 10^{-11}\,\mathrm{eV} \lesssim m_a \lesssim 10^{-8}\,\mathrm{eV},
\]
and the majority of compactifications satisfying \(M_s\ge M_{\mathrm{GUT}}\) also satisfy the stated PQ-quality criterion, except at very small \(h^{1,1}\lesssim 5\) [2505.15884].

That study also reports that no KS candidates were found with \(m_a\le H_0\) and \(f_a\gtrsim 2.5\times 10^{17}\,\mathrm{GeV}\), so it finds no axions suitable for the DESI-evolving-dark-energy scenario within the scanned ensemble [2505.15884]. This suggests that apparently different phenomenological conclusions in the KS literature often reflect different ensemble cuts: generic large-\(h^{1,1}\) toy-model analyses, GUT-filtered scans, and fixed visible-sector assumptions probe different slices of the same geometric landscape.

## 6. Statistical inference on moduli space and current limitations

The KS axiverse has also been recast as a problem in Bayesian inference on Calabi–Yau moduli space. In that framework, one samples the invariant Weil–Petersson measure on Kähler moduli space using MCMC and normalizing flows, with explicit implementations up to \(h^{1,1}=30\). The real-parameter-space metric is
\[
M_{ij} = \frac{1}{4\mathcal V^2}\left(2\tau_i\tau_j - \mathcal V\,\kappa_{ijk} t^k\right),
\]
and the corresponding measure is
\[
d\mu_{\mathrm{WP}} \propto \sqrt{|\det M|}\prod_i dt^i,
\]
which behaves as \(d\mu_{\mathrm{WP}}\sim d\Omega_{n-1}\,d\mathcal V/\mathcal V\), i.e. log-flat in the Calabi–Yau volume. The sampled region is the stretched Kähler cone, with all effective curve volumes \(\ge 1\) and all prime toric divisor volumes \(\ge 1\), together with explicit volume cutoffs \(\mathcal V_{\min}=1\) and \(\mathcal V_{\max}\approx 6.2\times 10^{12}\) for \(g_s=0.5\) and \(M_{\mathrm{KK}}=10^9\,\mathrm{GeV}\) [2512.00144].

Using this prior, axion mass and decay-constant distributions were constructed for all inequivalent KS Calabi–Yau threefolds with \(h^{1,1}\le 5\). Under the restricted WP measure, Calabi–Yau volume histograms are log-flat between the cutoff scales, while divisor volume spectra depend strongly on divisor topology: K3 fibre divisors and point blowups tend to be smaller and less correlated with \(\mathcal V\), and a trimodality organized by vertex, edge, and face divisors persists under WP sampling [2512.00144].

The same framework was used to ask inverse questions. For mock haloscope detections at \(h^{1,1}=3\), a relatively heavy QCD axion at \(\bar m_{\mathrm{QCD}}=6\,\mu\mathrm{eV}\) with \(\tau_{\mathrm{QCD}}\approx 40\) statistically favors rigid divisors, blowup divisors, elliptic fibrations, and divisors shrinking on higher-dimensional boundaries of the cone. By contrast, DMRadio-like neV detections are described as typical at small \(h^{1,1}\) and therefore less constraining on topology [2512.00144].

A full forward model incorporating Planck 2018 temperature and polarization data together with eBOSS Ly\(\alpha\) compressed likelihoods was then constructed for a specific \(h^{1,1}=7\) geometry. In that example, the CY-informed posterior localizes the moduli-space region supporting an ultralight axion with \(\Omega_a/\Omega_{\mathrm{DM}}\approx 1\text{–}3\%\), \(m_a\in[10^{-25},10^{-23}]\,\mathrm{eV}\), and \(f_a\) peaked near \(\approx 10^{-2}M_{\mathrm{pl}}\), narrowing the EFT-like posterior from below in mass [2512.00144].

Across the literature, several limitations recur. The KS hypersurface database does not specify orientifold involutions, \(h^{1,1}_\pm\) splittings, O-plane data, or D-brane configurations; most ensemble studies work in the large-volume regime, often in the orientifold-even sector, and do not perform explicit moduli stabilization. Additional effects—ED3 zero-mode constraints, Freed–Witten consistency conditions, loop and higher-derivative corrections, backreaction, alignment, and swampland constraints—are repeatedly identified as relevant but not fully incorporated [1411.1418]. The present state of the subject is therefore best understood as a highly explicit geometric framework for statistical axion phenomenology, rather than a complete moduli-stabilized census of all string-realized axion sectors.

Source: https://www.emergentmind.com/topics/kreuzer-skarke-axiverse