xAlpha: Automation in Horndeski Gravity

• The paper demonstrates that xAlpha automates the derivation of the complete second-order scalar perturbation system, reducing 300 possible coefficients to 150 nonzero ones.
• xAlpha, built on Mathematica with xAct and xPand, organizes perturbation equations into the α-basis to clearly expose screening behaviors such as Vainshtein, Chameleon, and Phaedrus screening.
• In finance, the term ‘xAlpha’ is also applied to dynamically aggregated alpha signals, highlighting its context-dependent use across disparate research fields.

Searching arXiv for the relevant papers on “xAlpha” and adjacent usages to ground the article. xAlpha is a polysemous designation used in distinct research contexts rather than a single universally standardized term. In contemporary arXiv literature, the name most prominently denotes a Mathematica package introduced in the study of luminal Horndeski gravity as a “package to compute and organise perturbation equations in scalar-tensor theories,” where it serves as the algebraic engine behind the derivation of the complete second-order perturbation system and the associated master screening equation (Sirera et al., 5 May 2026). In separate quantitative-finance works, “xAlpha” is used more generically for engineered, LLM-generated, or dynamically aggregated alpha signals within formulaic alpha mining and deployment frameworks (Chen et al., 1 Sep 2025, Pham et al., 10 Mar 2026). Because these usages belong to unrelated domains, the term is best understood through contextual disambiguation.

1. Principal usage in modified gravity

In the gravitational-physics literature, xAlpha is introduced in the paper “A Master Equation for Screening in Luminal Horndeski Gravity” as one of two publicly available software packages released alongside the theoretical results, the other being the Python module escut (Sirera et al., 5 May 2026). It is described as “the Mathematica backbone of this paper’s main technical result: the complete second-order perturbation system for luminal Horndeski gravity written in the α-basis and organised in a way that makes screening operators transparent” (Sirera et al., 5 May 2026).

The theoretical setting is the luminal Horndeski subclass, defined by $c_{\rm GW}=c$ and hence $\alpha_T=0$, with action

$$S=\int d^4x\sqrt{-g}\Big[\,G_4(\phi)R+K(\phi, X) -G_3(\phi, X)\Box\phi+{\cal L}_{\rm m}\Big],$$

where $X\equiv -\frac{1}{2}\phi_\mu\phi^\mu$, and matter is minimally coupled (Sirera et al., 5 May 2026). Cosmological perturbations are expressed in the Bellini–Sawicki $\alpha$-basis, with non-zero functions $\alpha_K,\alpha_B,\alpha_M$ and $M_*^2=2G_4$ (Sirera et al., 5 May 2026). Within this framework, xAlpha automates the derivation and organization of the full, unapproximated second-order scalar perturbation equations on a flat FLRW background.

The significance of this role is methodological. The paper states that “300 possible coefficients $A_i^{ab},B_i^{ab},C_i^{ab},D_i^{ab}$ exist in the general template,” while luminal Horndeski reduces this to 150 nonzero coefficients, all computed and stored by xAlpha (Sirera et al., 5 May 2026). This suggests that xAlpha functions not merely as a symbolic manipulator but as a structured reduction system for a perturbative operator basis that would otherwise be prohibitively cumbersome to handle analytically.

2. Computational structure and outputs

xAlpha is implemented in Mathematica and built on top of xAct and xPand (Sirera et al., 5 May 2026). In the configuration used in the paper, it takes as input the luminal Horndeski action, a flat FLRW background metric, a homogeneous scalar $\phi(t)$, and a choice of scalar perturbation variable, specifically

$Q\equiv H\,\delta\phi/\dot\phi,$

then expands the metric and scalar equations to second order in scalar perturbations in Newtonian gauge,

$\alpha_T=0$0

It organizes the resulting perturbation equations into a compact tensorial basis whose coefficients depend only on background quantities, and expresses those coefficients in the original $\alpha_T=0$1 language, in the $\alpha_T=0$2-basis, and in a further compactified set of $\alpha_T=0$3-combinations defined in Appendix C of the paper (Sirera et al., 5 May 2026).

The linear and second-order scalar perturbation system is written in terms of the fields

$\alpha_T=0$4

For example, the linear $\alpha_T=0$5 equation is given schematically by

$\alpha_T=0$6

while the second-order scalar equation includes terms such as

$\alpha_T=0$7

These expressions are central because the screening analysis depends on identifying which nonlinear operators survive after the quasi-static and weak-field limits are applied (Sirera et al., 5 May 2026).

The package also supports alternative scalar perturbation variables such as $\alpha_T=0$8 and $\alpha_T=0$9, together with mappings between the corresponding effective mass terms. One relation stated explicitly is

$$S=\int d^4x\sqrt{-g}\Big[\,G_4(\phi)R+K(\phi, X) -G_3(\phi, X)\Box\phi+{\cal L}_{\rm m}\Big],$$0

This establishes that xAlpha is not tied to a single perturbation-variable convention, but can translate between several conventions used in the literature (Sirera et al., 5 May 2026).

3. From perturbation algebra to screening diagnostics

A central function of xAlpha is to expose the coefficients entering the quasi-static scalar equation and, after elimination of the metric potentials, the master screening equation (Sirera et al., 5 May 2026). In the weak-field and quasi-static approximations, the metric equations remain linear while nonlinear corrections survive only in the scalar equation. After eliminating $$S=\int d^4x\sqrt{-g}\Big[\,G_4(\phi)R+K(\phi, X) -G_3(\phi, X)\Box\phi+{\cal L}_{\rm m}\Big],$$1 and $$S=\int d^4x\sqrt{-g}\Big[\,G_4(\phi)R+K(\phi, X) -G_3(\phi, X)\Box\phi+{\cal L}_{\rm m}\Big],$$2, the master screening equation becomes

$$S=\int d^4x\sqrt{-g}\Big[\,G_4(\phi)R+K(\phi, X) -G_3(\phi, X)\Box\phi+{\cal L}_{\rm m}\Big],$$3

with

$$S=\int d^4x\sqrt{-g}\Big[\,G_4(\phi)R+K(\phi, X) -G_3(\phi, X)\Box\phi+{\cal L}_{\rm m}\Big],$$4

All of these coefficients are direct outputs of xAlpha (Sirera et al., 5 May 2026).

This organization provides the direct mapping from a covariant luminal Horndeski Lagrangian to three screening behaviors recovered or identified in the paper: Vainshtein screening, Chameleon screening, and a new kinetic regime termed Phaedrus screening (Sirera et al., 5 May 2026). The relevant nonlinear operators are explicitly associated with distinct mechanisms: the cubic Galileon-type term with Vainshtein screening, the density-dependent nonlinear mass with Chameleon screening, and the $$S=\int d^4x\sqrt{-g}\Big[\,G_4(\phi)R+K(\phi, X) -G_3(\phi, X)\Box\phi+{\cal L}_{\rm m}\Big],$$5 and $$S=\int d^4x\sqrt{-g}\Big[\,G_4(\phi)R+K(\phi, X) -G_3(\phi, X)\Box\phi+{\cal L}_{\rm m}\Big],$$6 terms with Phaedrus screening.

For static, spherically symmetric configurations, the equation is written as

$$S=\int d^4x\sqrt{-g}\Big[\,G_4(\phi)R+K(\phi, X) -G_3(\phi, X)\Box\phi+{\cal L}_{\rm m}\Big],$$7

with

$$S=\int d^4x\sqrt{-g}\Big[\,G_4(\phi)R+K(\phi, X) -G_3(\phi, X)\Box\phi+{\cal L}_{\rm m}\Big],$$8

$$S=\int d^4x\sqrt{-g}\Big[\,G_4(\phi)R+K(\phi, X) -G_3(\phi, X)\Box\phi+{\cal L}_{\rm m}\Big],$$9

$X\equiv -\frac{1}{2}\phi_\mu\phi^\mu$0

The paper’s interpretation of these terms depends entirely on xAlpha’s coefficient extraction and operator organization (Sirera et al., 5 May 2026).

4. Workflow, implementation, and scope

The usage pattern reconstructed in the paper begins by specifying a theory either through $X\equiv -\frac{1}{2}\phi_\mu\phi^\mu$1, $X\equiv -\frac{1}{2}\phi_\mu\phi^\mu$2, and $X\equiv -\frac{1}{2}\phi_\mu\phi^\mu$3, or equivalently through the background functions $X\equiv -\frac{1}{2}\phi_\mu\phi^\mu$4, $X\equiv -\frac{1}{2}\phi_\mu\phi^\mu$5, and $X\equiv -\frac{1}{2}\phi_\mu\phi^\mu$6 and their derivatives (Sirera et al., 5 May 2026). xAlpha then derives the background equations $X\equiv -\frac{1}{2}\phi_\mu\phi^\mu$7, implements the perturbative expansion on flat FLRW in Newtonian gauge, and produces the linear and second-order coefficient sets $X\equiv -\frac{1}{2}\phi_\mu\phi^\mu$8 and $X\equiv -\frac{1}{2}\phi_\mu\phi^\mu$9 (Sirera et al., 5 May 2026).

The paper emphasizes that xAlpha is not itself the numerical solver for the nonlinear master equation. Instead, its outputs are exported to the Python package escut, which solves the radial equation numerically once the coefficients $\alpha$0 have been evaluated along a chosen background (Sirera et al., 5 May 2026). This division of labor makes xAlpha an algebraic front end rather than an end-to-end simulation environment.

Implementation details stated in the paper are concise. The language is Mathematica; dependencies are xAct and xPand; the code is public at https://github.com/sergisl/xAlpha (Sirera et al., 5 May 2026). The package is said to contain core modules for the covariant action and equations of motion, perturbation expansion routines, and post-processing routines that express coefficients in the $\alpha$1-basis and in the $\alpha$2-combinations (Sirera et al., 5 May 2026).

The stated limitations are equally specific. In the work under discussion, xAlpha is used fully only at second order; a complete third-order organization is left for future work. The current implementation is specialized to luminal Horndeski on flat FLRW and scalar perturbations, with tensors neglected for screening analysis. The paper also notes that the expressions are large, and that the authors mitigate this by factoring out powers of $\alpha$3 and $\alpha$4 and compressing combinations into $\alpha$5-variables (Sirera et al., 5 May 2026).

5. Relation to existing cosmology software

The paper situates xAlpha against existing modified-gravity tools by noting that hi_class and EFTCAMB implement modified gravity at linear level in the $\alpha$6- or EFT basis, but do not provide fully general second-order perturbation equations (Sirera et al., 5 May 2026). It also contrasts the xAlpha-based workflow with previous analytic studies of nonlinear perturbations in scalar–tensor theories, which often used other parameterizations, assumed non-luminal models, or imposed approximations such as quasi-static limits or Vainshtein dominance directly at the Lagrangian or equation level (Sirera et al., 5 May 2026).

What distinguishes xAlpha in this comparison is the combination of three elements: the full unapproximated second-order scalar perturbation system for luminal Horndeski in the $\alpha$7-basis, the systematic decomposition into operator coefficients depending on $\alpha$8, and the direct mapping from these coefficients to screening mechanisms (Sirera et al., 5 May 2026). A plausible implication is that xAlpha is intended to bridge the observational $\alpha$9-language used in survey-era cosmology with the nonlinear screening analysis needed for semi-analytical and simulation pipelines.

The paper states explicitly that the authors plan to use xAlpha-derived kernels as analytical input for fast modified-gravity simulations and semi-analytical tools, and mentions Hi-COLA as the kind of hybrid simulation code to which this bridge is relevant (Sirera et al., 5 May 2026). This places xAlpha within a broader program connecting theory-space parameterization, screening physics, and data-analysis workflows.

6. Alternative uses of “xAlpha” in quantitative finance

In quantitative-finance literature, “xAlpha” does not denote the Horndeski perturbation package. Instead, it appears as a generic label for an engineered or dynamically aggregated alpha signal. In “Adaptive Alpha Weighting with PPO: Enhancing Prompt-Based LLM-Generated Alphas in Quant Trading,” “xAlpha” is described as “essentially the combined, dynamically weighted signal built from many underlying LLM-generated alpha formulas” (Chen et al., 1 Sep 2025). There the underlying system uses the deepseek-r1-distill-llama-70b model to generate 50 formulaic alphas from OHLCV, technical indicators, sentiment, and index levels for Apple, HSBC, Pepsi, Toyota, and Tencent, and then applies PPO to adaptively weight them over time (Chen et al., 1 Sep 2025).

That framework defines a composite alpha

$\alpha_K,\alpha_B,\alpha_M$0

with the raw 50-dimensional PPO action clipped to $\alpha_K,\alpha_B,\alpha_M$1 and $\alpha_K,\alpha_B,\alpha_M$2-normalized before aggregation (Chen et al., 1 Sep 2025). The paper reports that the PPO-optimized strategy outperforms equal-weighted alpha portfolios and index benchmarks for Apple, HSBC, Pepsi, and Tencent, but underperforms for Toyota (Chen et al., 1 Sep 2025). In this usage, “xAlpha” is not a software package but a dynamically managed alpha ensemble.

A second finance usage appears in the “AlgoXpert Alpha Research Framework,” where the “xAlpha” framework is characterized as a deployment protocol built around an IS–WFA–OOS pipeline, majority-pass and catastrophic-veto rules, and defense-in-depth controls such as spread guards, leverage guards, circuit breakers, and a kill switch (Pham et al., 10 Mar 2026). Here the emphasis is not on alpha generation but on validation and deployment governance. This suggests that in finance, “xAlpha” can function as a broad brand-like label for an alpha research or execution framework rather than as a uniquely defined technical object.

These finance usages are conceptually unrelated to the cosmology package. The shared name reflects naming coincidence rather than methodological overlap.

7. Disambiguation and scholarly significance

Because “xAlpha” refers to unrelated entities across fields, interpretation depends entirely on disciplinary context. In gravitational theory, it denotes a Mathematica package for deriving and organizing second-order perturbation equations in luminal Horndeski gravity and for exposing screening operators in the $\alpha_K,\alpha_B,\alpha_M$3-basis (Sirera et al., 5 May 2026). In quantitative finance, it can refer to a composite alpha engine built from LLM-generated formulaic alphas and adaptively weighted with PPO (Chen et al., 1 Sep 2025), or to a staged protocol for validating and deploying alpha strategies under controls against overfitting and execution fragility (Pham et al., 10 Mar 2026).

The principal scholarly significance of xAlpha, in the sense anchored by the 2026 Horndeski paper, lies in the automation of otherwise intractable perturbative algebra. The package operationalizes a covariant-to-effective “dictionary” from the Lagrangian functions $\alpha_K,\alpha_B,\alpha_M$4 or the $\alpha_K,\alpha_B,\alpha_M$5-basis to the coefficient structure governing nonlinear screening (Sirera et al., 5 May 2026). The paper’s central claim is that, in many cases, these tools “enable the identification of the active screening type directly from a luminal Horndeski Lagrangian” (Sirera et al., 5 May 2026). This suggests that xAlpha is best regarded not simply as symbolic software, but as an infrastructure for theory diagnosis in modified gravity.