AxiREPO: Integrated Frameworks Across Domains

• AxiREPO is a multifaceted framework label that integrates heterogeneous research components into cohesive, managed workflows across distinct domains.
• In health and software engineering, it supports reproducible reporting and automated model consolidation through explicit data, code, and dissemination pipelines.
• In astrophysics, AxiREPO employs advanced pseudo-spectral solvers to simulate fuzzy dark matter dynamics and cosmological phenomena with precision.

Searching arXiv for papers using the term “AxiREPO” across domains. “AxiREPO” is a name that appears in multiple, technically unrelated research contexts on arXiv. In the literature, it denotes at least three distinct kinds of frameworks: a reproducible-research reporting framework for health and social sciences centered on data, analytical codes, and dissemination (Vissoci et al., 2013); a quality-driven framework for automatically constructing reusable reference models from existing software analysis/design instances (Al-Khiaty et al., 2014); and an AREPO-embedded numerical framework for evolving ultralight-axion or fuzzy-dark-matter dynamics through pseudo-spectral Schrödinger–Poisson solvers, including cosmological and mixed-dark-matter simulations (May et al., 2022, Dome et al., 2024, Tocher et al., 26 Mar 2026). The shared label therefore does not identify a single canonical system. Instead, it designates domain-specific frameworks whose common feature is architectural integration: each links multiple technical components into a managed workflow.

1. Term usage across research domains

The term is used in at least three research areas with different meanings and technical objectives.

In health and social sciences, AxiREPO is introduced as a reproducible-research reporting framework applied to Big Clinical Data. Its aim is to make research outputs “fully traceable, rerunnable, and shareable,” so that others can inspect the data source, analytical code, and final reports in one integrated online environment (Vissoci et al., 2013).

In software engineering, AxiREPO is presented as a quality-driven framework for automatically building a reusable reference model from a collection of existing application instances. Its purpose is to consolidate multiple early-stage artifacts into a reference model that captures both common and variable analysis/design practices and can improve through reinforcement learning (Al-Khiaty et al., 2014).

In computational astrophysics and cosmology, AxiREPO is used as the fuzzy-dark-matter extension of the AREPO code. In that setting, it replaces standard collisionless dark-matter treatment with a pseudo-spectral Schrödinger–Poisson solver on a Cartesian mesh, enabling simulations of wave-like dark matter, including interference, solitonic cores, and suppression of small-scale structure (May et al., 2022, Tocher et al., 26 Mar 2026). A mixed-dark-matter gravity solver was later implemented in the same framework for cosmologies containing both ultralight axions and a dominant cold component (Dome et al., 2024).

A plausible implication is that “AxiREPO” should be interpreted contextually rather than as a single framework lineage. The identical label masks substantial differences in epistemic goal, system architecture, and validation standard.

2. AxiREPO as a reproducible-research reporting framework

In the health and social-science usage, the framework is explicitly organized around three axes: Data, Analytical codes, and Dissemination (Vissoci et al., 2013). The authors argue that reproducibility is not achieved by publishing data or code alone, but by integrating all three into a single workflow.

The data axis covers data formats and repositories. The paper discusses CSV, RDF / LOD / SPARQL, and JSON. CSV is described as simple and widely compatible but lacking built-in update mechanisms or security features. RDF / LOD / SPARQL are presented as supporting semantic-web and linked-data use, including automated updates and dynamic merging of datasets with shared entities. JSON is treated as a flexible data-interchange format. A complete data dictionary is identified as a key reproducibility requirement for public datasets, especially in clinical settings where variable meanings must remain unambiguous (Vissoci et al., 2013).

The analytical codes axis is the computational core. Reproducibility is said to require the exact analytical procedures to be available, including required packages, data import or connection steps, preprocessing steps, analytical commands, and code comments and descriptions. The framework uses R as the central orchestration layer connecting data files, repositories, statistical analysis, graphics, report generation, and interoperability with other software such as SAS, Stata, SPSS, Python, Java, databases, RDF, C/C++, and Weka. The authors state that they use RStudio to manage workspace, graphs, scripts, logs, and multiple project directories, and they treat GitHub as the main online repository for analytical code, collaboration, forking, templates, wiki-based method documentation, and web-based outputs from knitr (Vissoci et al., 2013).

The dissemination axis addresses public presentation and interaction. The framework supports automatically generated reports, tables and figures, HTML or PDF outputs, and interactive visualizations. The paper notes that knitr can translate code into reports mixing LaTeX and Markdown, with outputs generated directly from the same code that performs the analysis. It also mentions interactive graphics using rggobi, Shiny, and other CRAN-supported visualization tools. To integrate the full project, the authors create websites using Google Sites, each containing links to data repositories, code repositories, embedded reports, figures and graphs, and licensing information (Vissoci et al., 2013).

The overall integration strategy is centered on R, yielding a workflow summarized as data repository → R analysis scripts → output generation → dissemination website. The framework also requires that public documents and outputs carry a license, and the authors use Creative Commons Attribution-NonCommercial style licensing. This is presented as part of a broader management logic involving open-access repositories, alignment of data and code, publication of scripts with descriptions, and linkage of all outputs to a project website (Vissoci et al., 2013).

A major limitation is privacy. The paper states that health datasets often contain protected health information (PHI) under HIPAA, so public release presupposes appropriate de-identification and governance. Additional limitations include CSV’s lack of security and update mechanisms, dependence on users having the same R packages, and the fact that the framework is “still in progress” (Vissoci et al., 2013).

3. AxiREPO as a quality-driven reference-model construction framework

In software engineering, AxiREPO addresses a different problem: the automatic derivation of a reusable reference model from a set of similar application models. The framework is positioned at the intersection of Model-Driven Development (MDD) and Software Product Line Engineering (SPLE). From SPLE it borrows explicit treatment of commonality and variability; from MDD it borrows the idea that models are reusable assets at a high level of abstraction (Al-Khiaty et al., 2014).

The paper identifies six major challenges the framework is intended to address: efficient similarity assessment, efficient consolidation/merging, integration of quality as an orthogonal concern, a representation that preserves evolution/instantiation information, reinforcement learning for continuous improvement, and tool support for automation (Al-Khiaty et al., 2014). The problem statement is explicitly framed against ad hoc reuse patterns such as copy-and-modify, which produce duplicated artifacts, maintenance burden, and design smells.

AxiREPO is conceptually divided into four phases: Preprocessing / cleaning source models, Similarity assessment, Reference model construction and representation, and Reference model evolution (Al-Khiaty et al., 2014). The process is iterative rather than one-shot.

In preprocessing, source models are aligned against a catalog of well-known analysis/design patterns. Detected fragments are classified as correct instances or “spoiled” variants, and spoiled fragments are cleaned before merging. The paper states that this detection may use AI techniques to reduce search complexity, and that the pattern catalog itself can evolve through reinforcement learning (Al-Khiaty et al., 2014). This makes quality improvement a pre-consolidation concern.

In similarity assessment, the framework uses multi-view similarity rather than a single perspective. The paper explicitly names structural view, functional view, and behavioral view. The outcome is not merely a similarity score but a selection decision: some models are labeled candidate for merge, others not candidate for merge. This is treated as an optimization problem balancing reusable common elements, cohesion, redundancy, and domain coverage. The matching algorithm is described as likely heuristic-based and may use AI techniques such as Genetic Algorithms (Al-Khiaty et al., 2014).

In reference model construction, candidate models are merged into a super-set reference model that makes commonality mandatory/shared and variability explicit through variation points. The target model is intended to be complete, non-redundant, cohesive, instantiatable, and supportive of desired quality factors. It must also preserve enough information for instantiation, evolution, and synchronization with source models, reflecting the paper’s “bottom-up-top-down” view of merging (Al-Khiaty et al., 2014).

In reference model evolution, new instances are re-evaluated against the reference model. If similarity is above a threshold, the instance is treated as a positive example and merged; if it is below, it is treated as a negative example and may trigger evolution. The paper gives 50% as an illustrative threshold. Negative feedback can lead to revisiting previously excluded models, reassessing included parts, and removing elements that hurt cohesion or reuse. The framework therefore uses a reinforcement learning mechanism to improve quality, completeness, and representativeness over time (Al-Khiaty et al., 2014).

The paper is explicit about scope and evidence. It is a proposal and survey-driven framework paper, not a full implementation or empirical validation paper. It does not provide a complete implementation, formal proof of correctness, or quantitative evaluation of the complete framework (Al-Khiaty et al., 2014). A common misconception would be to treat it as an implemented automatic model-merging tool; the text instead presents it as a conceptual and architectural framework.

4. AxiREPO as an AREPO-based fuzzy-dark-matter solver

In astrophysical usage, AxiREPO is a numerical framework embedded in the AREPO code base for evolving fuzzy dark matter (FDM) or ultralight axion dark matter. Unlike standard AREPO runs that evolve collisionless particles with TreePM gravity, AxiREPO solves the full Schrödinger–Poisson (SP) system for a complex wave function on a uniform Cartesian mesh using a pseudo-spectral method with FFT-based spatial operations (May et al., 2022).

The governing equations are given in the literature. In one formulation, the dark-matter field obeys

$$i\hbar\,\partial_t \psi = - \frac{\hbar^2}{2m_a}\nabla^2\psi + m_a V\psi,$$

with gravitational potential

$\nabla^2 V = 4\pi G(\rho-\bar\rho),$

and density

$\rho_{\rm dm}=|\psi|^2$

together with the Madelung decomposition and energy splitting into quantum-pressure, bulk-flow kinetic, and gravitational terms (Tocher et al., 26 Mar 2026). In cosmological comoving coordinates, the same SP structure appears with scale-factor dependence:

$$i\hbar\,\partial_t \psi(t,\mathbf{x}) = -\frac{\hbar^2}{2m a(t)^2}\nabla^2\psi(t,\mathbf{x}) +\frac{m}{a(t)}\Phi\,\psi(t,\mathbf{x}),$$

$$\nabla^2\Phi(t,\mathbf{x}) = 4\pi G m\left(|\psi(t,\mathbf{x})|^2-\langle |\psi|^2\rangle(t)\right)$$

(May et al., 2022).

AxiREPO implements a pseudo-spectral kick-drift-kick scheme. In the isolated-halo study, the algorithm is described in four stages: compute the gravitational potential from density on the grid via FFTs; apply a half-step kick; drift in Fourier space; and finish with another half kick (Tocher et al., 26 Mar 2026). In mixed-dark-matter cosmologies, the paper describes a second-order symmetrised split-step pseudo-spectral “kick-drift-kick” method advancing the axion wavefunction through alternating gravitational phase kicks and FFT-based drifts from the Laplacian term (Dome et al., 2024).

The numerical constraints are stringent. The literature emphasizes that the timestep scales roughly as

$\Delta t \propto \Delta x^2,$

or, more precisely in cosmological form,

$$\Delta t < \min\!\left( \frac{4}{3\pi}\frac{m}{\hbar}a^2\Delta x^2, \; 2\pi\frac{\hbar}{m}a\frac{1}{|\Phi_{\max}|} \right),$$

and that the de Broglie wavelength must be spatially resolved everywhere relevant to the dynamics (May et al., 2022). This makes AxiREPO substantially more expensive than ordinary $N$-body CDM simulations and explains the need for a specialized module.

The framework’s scientific role is to capture wave effects that approximate methods miss: interference, quantum pressure, granularity, solitonic cores, and the associated suppression of small-scale structure (May et al., 2022). In hybrid gas-plus-FDM applications, gas remains on AREPO’s moving Voronoi mesh while the FDM sector is evolved on a fixed Cartesian grid, and the FDM potential is interpolated to gas cells (Tocher et al., 26 Mar 2026).

5. Cosmological and mixed-dark-matter applications

AxiREPO has been used for large cosmological SP simulations of FDM. One study performs a four-way comparison separating initial conditions from dynamics: FDM initial conditions with SP dynamics, FDM initial conditions with $N$-body dynamics, CDM initial conditions with SP dynamics, and CDM initial conditions with $N$-body dynamics (May et al., 2022). This design is used to disentangle the impact of the FDM transfer-function cutoff from the impact of full wave evolution.

That work reports the first direct measurement of the FDM halo mass function from full wave simulations, concluding that the small-scale FDM transfer-function cutoff strongly suppresses halo formation and that the halo abundance inferred from full SP simulations is broadly consistent with earlier $\nabla^2 V = 4\pi G(\rho-\bar\rho),$0-body-with-FDM-initial-conditions estimates (May et al., 2022). It also shows that FDM filaments are smooth, dense, and extended, unlike fragmented CDM filaments, and that this morphology makes halo finding difficult because smooth dense filaments connect haloes across the box while interference peaks can masquerade as bound structures (May et al., 2022).

A later mixed-dark-matter study uses a new MDM gravity solver implemented in AxiREPO to evolve cosmologies containing an ultralight axion component plus a dominant cold component (Dome et al., 2024). In that pipeline, AxiREPO generates the non-linear simulations from which halo statistics, density profiles, halo concentrations, and axion-in-halo mass relations are measured. Those measurements then calibrate an updated AxionHMcode semi-analytic halo model (Dome et al., 2024).

The mixed-dark-matter simulations are fully DM-only, use a fiducial axion mass

$\nabla^2 V = 4\pi G(\rho-\bar\rho),$1

vary the axion fraction over

$\nabla^2 V = 4\pi G(\rho-\bar\rho),$2

and evolve from $\nabla^2 V = 4\pi G(\rho-\bar\rho),$3 down to $\nabla^2 V = 4\pi G(\rho-\bar\rho),$4, with the analysis focused mainly on $\nabla^2 V = 4\pi G(\rho-\bar\rho),$5–4 (Dome et al., 2024). Haloes are identified with Rockstar using only the CDM particle distribution, a choice justified a posteriori by agreement of halo mass functions and concentration–mass relations with theoretical expectations (Dome et al., 2024).

The calibration results are quantitatively specific. The updated AxionHMcode remains within 10% for $\nabla^2 V = 4\pi G(\rho-\bar\rho),$6 on scales $\nabla^2 V = 4\pi G(\rho-\bar\rho),$7 at redshifts $\nabla^2 V = 4\pi G(\rho-\bar\rho),$8–3.5 around the fiducial mass, within 20% for $\nabla^2 V = 4\pi G(\rho-\bar\rho),$9 at $\rho_{\rm dm}=|\psi|^2$0 and $\rho_{\rm dm}=|\psi|^2$1, and can evaluate in under a minute on a single core (Dome et al., 2024). The paper explicitly frames this as enabling practical Bayesian parameter sampling and forecast analyses in mixed-dark-matter cosmologies.

6. Gas dynamics, fragmentation, and Cosmic Dawn

A distinct astrophysical use of AxiREPO couples dynamical FDM to primordial gas in isolated halo simulations. In that study, the authors use “the axirepo code within the framework of arepo” to evolve the ultra-light axion dark-matter field self-consistently with gas cooling, chemistry, and sink-particle formation (Tocher et al., 26 Mar 2026).

The simulation design distinguishes three cases: CDM, Frozen FDM, and Dynamic FDM. In Frozen FDM, AxiREPO evolves FDM to a virialized halo and the wave solver is then paused while the gas evolves in a static cored potential. In Dynamic FDM, the SP equations continue to be solved while the gas collapses (Tocher et al., 26 Mar 2026). This explicitly separates a geometry effect from a dynamics effect.

The paper reports that the delay in first sink formation, used as a proxy for the onset of runaway collapse or star formation, scales inversely with both halo mass and axion mass. For $\rho_{\rm dm}=|\psi|^2$2, the static cored geometry dominates the delay. For $\rho_{\rm dm}=|\psi|^2$3, the moving soliton and interference fluctuations create an additional dynamical barrier. In the $\rho_{\rm dm}=|\psi|^2$4, $\rho_{\rm dm}=|\psi|^2$5 case, the dynamic run is delayed by about $\rho_{\rm dm}=|\psi|^2$6 more than the frozen one (Tocher et al., 26 Mar 2026).

The mechanism is described in several parts. The soliton exhibits order-unity density fluctuations and a random walk with magnitude of order the de Broglie wavelength, with fluctuation scales written as

$\rho_{\rm dm}=|\psi|^2$7

(Tocher et al., 26 Mar 2026). These fluctuations inject kinetic energy into the gas and act as a stirring mechanism. The paper further defines the specific angular momentum as $\rho_{\rm dm}=|\psi|^2$8 and introduces a centrifugal support condition via

$\rho_{\rm dm}=|\psi|^2$9

with a critical angular momentum estimate

$$i\hbar\,\partial_t \psi(t,\mathbf{x}) = -\frac{\hbar^2}{2m a(t)^2}\nabla^2\psi(t,\mathbf{x}) +\frac{m}{a(t)}\Phi\,\psi(t,\mathbf{x}),$$0

In the dynamic FDM runs, gas within the soliton radius can reach or exceed $$i\hbar\,\partial_t \psi(t,\mathbf{x}) = -\frac{\hbar^2}{2m a(t)^2}\nabla^2\psi(t,\mathbf{x}) +\frac{m}{a(t)}\Phi\,\psi(t,\mathbf{x}),$$1, becoming rotationally stabilized against collapse (Tocher et al., 26 Mar 2026).

Chemically, the simulations use a 12-species primordial chemistry network with 45 reactions. Dynamic FDM redistributes $$i\hbar\,\partial_t \psi(t,\mathbf{x}) = -\frac{\hbar^2}{2m a(t)^2}\nabla^2\psi(t,\mathbf{x}) +\frac{m}{a(t)}\Phi\,\psi(t,\mathbf{x}),$$2-rich gas into extended $$i\hbar\,\partial_t \psi(t,\mathbf{x}) = -\frac{\hbar^2}{2m a(t)^2}\nabla^2\psi(t,\mathbf{x}) +\frac{m}{a(t)}\Phi\,\psi(t,\mathbf{x}),$$3 plumes, unlike the centrally confined $$i\hbar\,\partial_t \psi(t,\mathbf{x}) = -\frac{\hbar^2}{2m a(t)^2}\nabla^2\psi(t,\mathbf{x}) +\frac{m}{a(t)}\Phi\,\psi(t,\mathbf{x}),$$4-rich gas of the CDM case. The paper connects this to finite cooling times and outward transport by the evolving soliton potential (Tocher et al., 26 Mar 2026). It further argues that the resulting spread of cold gas over a larger volume lowers the Jeans mass in the outskirts,

$$i\hbar\,\partial_t \psi(t,\mathbf{x}) = -\frac{\hbar^2}{2m a(t)^2}\nabla^2\psi(t,\mathbf{x}) +\frac{m}{a(t)}\Phi\,\psi(t,\mathbf{x}),$$5

favoring fragmented star formation. Sink statistics support a transition from a small number of massive central sinks in CDM to many more low-mass sinks at larger radii in dynamic FDM (Tocher et al., 26 Mar 2026).

The broader conclusion is that AxiREPO reveals a suppression mechanism for early star formation beyond the usual FDM power-spectrum cutoff: even after haloes form, internal FDM wave dynamics further delay and fragment the gas collapse (Tocher et al., 26 Mar 2026). This is presented as relevant to Cosmic Dawn, reionization, the UV luminosity function, the 21-cm signal, and constraints on the axion mass.

7. Conceptual comparison and recurring architectural themes

Although the three AxiREPO usages are technically unrelated, they share a recurring systems pattern: each framework integrates heterogeneous artifacts into a single operational chain.

In the reproducible-research usage, the integrated chain is data repository → R analysis scripts → output generation → dissemination website (Vissoci et al., 2013). In the software-engineering usage, it is preprocessing → similarity assessment → reference-model construction → evolution (Al-Khiaty et al., 2014). In the astrophysical usage, it is the coupling of pseudo-spectral SP dark-matter evolution with either cosmological structure formation or AREPO-based gas dynamics (May et al., 2022, Tocher et al., 26 Mar 2026).

Another shared theme is explicit management of what might otherwise remain implicit. The health-science framework insists on public linkage of data, scripts, outputs, metadata, and licenses (Vissoci et al., 2013). The software-engineering framework insists on explicit representation of commonality, variability, and recoverability of instances from the merged model (Al-Khiaty et al., 2014). The astrophysical framework makes wave dynamics explicit rather than approximating them through modified initial conditions or static effective potentials (May et al., 2022, Tocher et al., 26 Mar 2026).

The principal differences lie in evidence and maturity. The health-science AxiREPO is presented as a practical, open-source-based reporting framework still requiring further improvement (Vissoci et al., 2013). The software-engineering AxiREPO is primarily conceptual and survey-driven, without a full implementation or empirical benchmark (Al-Khiaty et al., 2014). The astrophysical AxiREPO, by contrast, is deployed in large-scale numerical studies with quantitative performance, convergence, and scientific results, including halo mass functions, filament morphology, mixed-dark-matter halo-model calibrations, and fragmentation delays in primordial gas (May et al., 2022, Dome et al., 2024, Tocher et al., 26 Mar 2026).

A common misconception would be to assume that all occurrences of “AxiREPO” refer to the same framework family. The literature instead shows a reused label spanning reproducible reporting, software-model consolidation, and ultralight-dark-matter simulation. The precise meaning is therefore determined entirely by disciplinary context.