arXiv subjects

Daniel Tamayo

Publications and source records attributed to Daniel Tamayo.

At least 19 recordsLinked to original sources

arXiv2026

MrBERT: Modern Multilingual Encoders via Vocabulary, Domain, and Dimensional Adaptation

We introduce MrBERT, a family of 150M-300M parameter encoders built on the ModernBERT architecture and pre-trained on 35 languages and code. Through targeted adaptation, this model family achieves state-of-the-art results on Catalan- and Spanish-specific tasks, while establishing robust performance across specialized biomedical and legal domains. To bridge the gap between research and production, we incorporate Matryoshka Representation Learning (MRL), enabling flexible vector sizing that significantly reduces inference and storage costs. Ultimately, the MrBERT family demonstrates that modern encoder architectures can be optimized for both localized linguistic excellence and efficient, high-stakes domain specialization. We open source the complete model family on Huggingface.

cs.CL↗

arXiv2026

Canceling Effects of Conjunctions Render Higher Order Mean Motion Resonances Weak

Mean motion resonances (MMRs) are a key phenomenon in orbital dynamics. The traditional disturbing function expansion in celestial mechanics shows that, at low eccentricities, $p$:$p-q$ MMRs exhibit a clear hierarchy of strengths, scaling as $e^q$, where $q$ is the order of the resonance. This explains why first-order MMRs (e.g., 3:2 and 4:3) are important, while the infinite number of higher order integer ratios are not. However, this relationship derived from a technical perturbation series expansion provides little physical intuition. In this paper, we provide a simple physical explanation of this result for closely spaced orbits. In this limit, interplanetary interactions are negligible except during close encounters at conjunction, where the planets impart a gravitational "kick" to each other's mean motion. We show that while first-order MMRs involve a single conjunction before the configuration repeats, higher order MMRs involve multiple conjunctions per cycle, whose effects cancel out more precisely the higher the order of the resonance. Starting from the effects of a single conjunction, we provide an alternate, physically motivated derivation of MMRs' $e^q$ strength scaling.

astro-ph.EP↗

arXiv2026

Democratic heliocentric coordinates underestimate the rate of instabilities in long-term integrations of the Solar System

Wisdom-Holman (WH) integrators are symplectic operator-splitting methods widely used for long-term N-body simulations of planetary systems. Most implementations use either Jacobi coordinates or democratic heliocentric coordinates (DHC) for the Hamiltonian splitting, resulting in slightly different algorithms. In this paper we report results from numerical experiments, which show that integrations of the Solar System using DHC coordinates with typical timesteps of a few days suppress instabilities of the planet Mercury. We further show that this is due to an eccentricity dependent artificial numerical precession introduced by the DHC splitting. While the DHC splitting converges to the correct results at shorter timesteps of ~0.6 days, we argue that Jacobi coordinates remain reliable to significantly longer timesteps when orbits become moderately eccentric, and are thus a better choice when the innermost planet can reach high eccentricities.

astro-ph.EP↗

arXiv2026

Self-consistent Dynamical and Chaotic Tides in the REBOUNDx framework

At high eccentricities, tidal forcing excites vibrational modes within orbiting bodies known as dynamical tides. In this paper, we implement the coupled evolution of these modes with the body's orbit in the \texttt{REBOUNDx} framework, an extension to the popular $N$-body integrator \texttt{REBOUND}. We provide a variety of test cases relevant to exoplanet dynamics and demonstrate overall agreement with prior studies of dynamical tides in the secular regime. Our implementation is readily applied to various high-eccentricity scenarios and allows for fast and accurate $N$-body investigations of astrophysical systems for which dynamical tides are relevant.

astro-ph.EP↗

arXiv2025

A Unified, Physical Framework for Mean Motion Resonances

The traditional approach to analyzing mean motion resonances is through canonical perturbation theory. While this is a powerful method, its generality leads to complicated combinations of variables that are challenging to interpret and require looking up numerical coefficients particular to every different resonance. In this paper we develop simpler scaling relations in the limit where orbits are closely spaced (period ratios $\lesssim 2$) and interplanetary interactions can be approximated by only considering the close-approaches each time the inner planet overtakes the outer at conjunction. We develop geometric arguments for several powerful results: (i) that $p$:$p-q$ MMRs of the same order $q$ are all rescaled versions of one another (ii) that the general case of two massive planets on closely spaced, eccentric, co-planar orbits can be approximately mapped onto the much simpler case of an eccentric test particle perturbed by a massive planet on a co-planar circular orbit and (iii) that while the effects of consecutive conjunctions add up coherently for first-order ($p$:$p-1$) MMRs, they partially cancel for $p$:$p-q$ MMRs with order $q>1$, providing a physical explanation for why these higher order MMRs are weaker and can often be ignored. Finally, we provide simple expressions for the widths of MMRs and their associated oscillation frequencies that are universal to all closely spaced MMRs of a given order $q$, in the pendulum approximation.

astro-ph.EP↗

arXiv2025

Salamandra Technical Report

This work introduces Salamandra, a suite of open-source decoder-only large language models available in three different sizes: 2, 7, and 40 billion parameters. The models were trained from scratch on highly multilingual data that comprises text in 35 European languages and code. Our carefully curated corpus is made exclusively from open-access data compiled from a wide variety of sources. Along with the base models, supplementary checkpoints that were fine-tuned on public-domain instruction data are also released for chat applications. Additionally, we also share our preliminary experiments on multimodality, which serve as proof-of-concept to showcase potential applications for the Salamandra family. Our extensive evaluations on multilingual benchmarks reveal that Salamandra has strong capabilities, achieving competitive performance when compared to similarly sized open-source models. We provide comprehensive evaluation results both on standard downstream tasks as well as key aspects related to bias and safety.With this technical report, we intend to promote open science by sharing all the details behind our design choices, data curation strategy and evaluation methodology. In addition to that, we deviate from the usual practice by making our training and evaluation scripts publicly accessible. We release all models under a permissive Apache 2.0 license in order to foster future research and facilitate commercial use, thereby contributing to the open-source ecosystem of large language models.

cs.CL↗

arXiv2025

Mass-Editing Memory with Attention in Transformers: A cross-lingual exploration of knowledge

Recent research has explored methods for updating and modifying factual knowledge in large language models, often focusing on specific multi-layer perceptron blocks. This study expands on this work by examining the effectiveness of existing knowledge editing methods across languages and delving into the role of attention mechanisms in this process. Drawing from the insights gained, we propose Mass-Editing Memory with Attention in Transformers (MEMAT), a method that achieves significant improvements in all metrics while requiring minimal parameter modifications. MEMAT delivers a remarkable 10% increase in magnitude metrics, benefits languages not included in the training data and also demonstrates a high degree of portability. Our code and data are at https://github.com/dtamayo-nlp/MEMAT.

cs.CL↗

arXiv2025

Carving Out the Inner Edge of the Period Ratio Distribution through Giant Impacts

The distribution of orbital period ratios between adjacent observed exoplanets is approximately uniform, but exhibits a strong falloff toward close orbital separations. We show that this falloff can be explained through past dynamical instabilities carving out the period ratio distribution. Our suite of numerical experiments would have required $\sim 3$ million CPU-hours through direct N-body integrations, but was achieved with only $\approx 50$ CPU-hours by removing unstable configurations using the Stability of Planetary Orbital Configurations Klassifier (SPOCK) machine learning model. This highlights the role of dynamical instabilities in shaping the observed exoplanet population, and shows that the inner part of the period ratio distribution provides a valuable observational anchor on the giant impact phase of planet formation.

astro-ph.EP↗

arXiv2025

SPOCK 2.0: Update to the FeatureClassifier in the Stability of Planetary Orbital Configurations Klassifier

The Stability of Planetary Orbital Configurations Klassifier (SPOCK) package collects machine learning models for predicting the stability and collisional evolution of compact planetary systems. In this paper we explore improvements to SPOCK's binary stability classifier (FeatureClassifier), which predicts orbital stability by collecting data over a short N-body integration of a system. We find that by using a system-specific timescale (rather than a fixed $10^4$ orbits) for the integration, and by using this timescale as an additional feature, we modestly improve the model's AUC metric from 0.943 to 0.950 (AUC=1 for a perfect model). We additionally discovered that $\approx 10\%$ of N-body integrations in SPOCK's original training dataset were duplicated by accident, and that $<1\%$ were misclassified as stable when they in fact led to ejections. We provide a cleaned dataset of 100,000+ unique integrations, release a newly trained stability classification model, and make minor updates to the API.

astro-ph.EP↗

arXiv2024

Creating Pileups of Eccentric Planet Pairs Wide of MMRs Through Divergent Migration

Observed pileups of planets with period ratios $\approx 1\%$ wide of strong mean motion resonances (MMRs) pose an important puzzle. Early models showed that they can be created through sustained eccentricity damping driving a slow separation of the orbits, but this picture is inconsistent with elevated eccentricities measured through Transit Timing Variations. We argue that any source of divergent migration (tides, planet-disk interactions etc.) will cause planets that encounter an MMR to both jump over it (piling up wide of resonance) and get a kick to their free eccentricity. We find that the jumps in eccentricity expected from slow MMR crossings are sufficient (but mostly too large) to explain the free eccentricities measured through TTVs. We argue that this mechanism can be brought in line with observations if MMR crossings are not adiabatic and/or through residual eccentricity damping.

astro-ph.EP↗

arXiv2024

Orbital Migration through Atmospheric Mass Loss

Atmospheric mass loss is thought to have strongly shaped the sample of close-in exoplanets. These atmospheres should be lost isotropically, leading to no net migration on the planetary orbit. However, strong stellar winds can funnel the escaping atmosphere into a tail trailing the planet. We derive a simple kinematic model of the gravitational interaction between the planet and this anisotropic wind, and derive expressions for the expected migration of the planet. Over the expected range of parameters, we find typical migrations of a few tenths to a few percent inward. We argue that this modest migration may be observable for planet pairs near mean motion resonances, which would provide an independent observational constraint on atmospheric mass loss models.

astro-ph.EP↗

arXiv2024

Accelerating Giant Impact Simulations with Machine Learning

Constraining planet formation models based on the observed exoplanet population requires generating large samples of synthetic planetary systems, which can be computationally prohibitive. A significant bottleneck is simulating the giant impact phase, during which planetary embryos evolve gravitationally and combine to form planets, which may themselves experience later collisions. To accelerate giant impact simulations, we present a machine learning (ML) approach to predicting collisional outcomes in multiplanet systems. Trained on more than 500,000 $N$-body simulations of three-planet systems, we develop an ML model that can accurately predict which two planets will experience a collision, along with the state of the post-collision planets, from a short integration of the system's initial conditions. Our model greatly improves on non-ML baselines that rely on metrics from dynamics theory, which struggle to accurately predict which pair of planets will experience a collision. By combining with a model for predicting long-term stability, we create an ML-based giant impact emulator, which can predict the outcomes of giant impact simulations with reasonable accuracy and a speedup of up to four orders of magnitude. We expect our model to enable analyses that would not otherwise be computationally feasible. As such, we release our training code, along with an easy-to-use API for our collision outcome model and giant impact emulator.

astro-ph.EP↗

arXiv2023

Secular Dynamics of Compact Three-Planet Systems

The secular Laplace-Lagrange orbital solution, decomposing eccentricities into a set of uniformly precessing eigenmodes is a classical result that is typically solved numerically. However, in the limit where orbits are closely spaced, several simplifications make it possible to make analytical progress. We derive simple expressions for the eccentricity eigenmodes in a co-planar 3-planet system where the middle planet is massless, and show that these approximate the true eigenmodes of more general systems with 3 massive planets in various limits. These results provide intuition for the secular dynamics of real systems, and have applications for understanding the stability boundary for compact multi-planet systems.

astro-ph.EP↗

arXiv2023

On the Degree of Dynamical Packing in the Kepler Multi-planet Systems

Current planet formation theories rely on initially compact orbital configurations undergoing a (possibly extended) phase of giant impacts following the dispersal of the dissipative protoplanetary disk. The orbital architectures of observed mature exoplanet systems have likely been strongly sculpted by chaotic dynamics, instabilities, and giant impacts. One possible signature of systems continually reshaped by instabilities and mergers is their dynamical packing. Early Kepler data showed that many multi-planet systems are maximally packed - placing an additional planet between an observed pair would make the system unstable. However, this result relied on placing the inserted planet in the most optimistic configuration for stability (e.g., circular orbits). While this would be appropriate in an ordered and dissipative picture of planet formation (i.e. planets dampen into their most stable configurations), we argue that this best-case scenario for stability is rarely realized due to the strongly chaotic nature of planet formation. Consequently, the degree of dynamical packing in multi-planet systems under a realistic formation model is likely significantly higher than previously realized. We examine the full Kepler multi planet sample through this new lens, showing that ~60-95% of Kepler multi-planet systems are strongly packed and that dynamical packing increases with multiplicity. This may be a signature of dynamical sculpting or of undetected planets, showing that dynamical packing is an important metric that can be incorporated into planet formation modelling or when searching for unseen planets.

astro-ph.EP↗

arXiv2023

ASSIST: An Ephemeris-Quality Test Particle Integrator

We introduce ASSIST, a software package for ephemeris-quality integrations of test particles. ASSIST is an extension of the REBOUND framework and makes use of its IAS15 integrator to integrate test particle trajectories in the field of the Sun, Moon, planets, and 16 massive asteroids, with the positions of the masses coming from the JPL DE441 ephemeris and its associated asteroid perturber file. The package incorporates the most significant gravitational harmonics and general relativistic corrections. ASSIST also accounts for position- and velocity-dependent non-gravitational effects. The first order variational equations are included for all terms to support orbit fitting and covariance mapping. This new framework is meant to provide an open-source package written in a modern language to enable high-precision orbital analysis and science by the small body community. ASSIST is open source, freely distributed under the GNU General Public license, version 3.

astro-ph.EP↗

arXiv2023

Self-Consistent Spin, Tidal and Dynamical Equations of Motion in the REBOUNDx Framework

We have introduced self-consistent spin, tidal and dynamical equations of motion into REBOUNDx, a library of additional effects for the popular N-body integrator REBOUND. The equations of motion used are derived from the constant time lag approximation to the equilibrium tide model of tidal friction. These effects will allow the study of a variety of systems where the full dynamical picture cannot be encapsulated by point particle dynamics. We provide several test cases and benchmark the code's performance against analytic predictions. The open-source code is available in the most recent release of REBOUNDx.

astro-ph.EP↗

arXiv2022

The Yarkovsky effect in REBOUNDx

To more thoroughly study the effects of radiative forces on the orbits of small, astronomical bodies, we introduce the Yarkovsky effect into REBOUNDx, an extensional library for the N-body integrator REBOUND. Two different versions of the Yarkovsky effect (the "Full Version" and the "Simple Version") are available for use, depending on the needs of the user. We provide demonstrations for both versions of the effect and compare their computational efficiency with another previously implemented radiative force. In addition, we show how this effect can be used in tandem with other features in REBOUNDx by simulating the orbits of asteroids during the asymptotic giant branch phase of a 2 $M_{\odot}$ star. This effect is made freely available for use with the latest release of REBOUNDx.

astro-ph.EP↗

arXiv2022

celmech: A Python package for celestial mechanics

We present celmech, an open-source Python package designed to facilitate a wide variety of celestial mechanics calculations. The package allows users to formulate and integrate equations of motion incorporating user-specified terms from the classical disturbing function expansion of the interaction potential between pairs of planets. The code can be applied, for example, to isolate the contribution of particular resonances to a system's dynamical evolution and develop simple analytical models with the minimum number of terms required to capture a particular dynamical phenomenon. Equations and expressions can be easily manipulated by leveraging the extensive symbolic mathematics capabilities of the sympy Python package. The celmech package is designed to interface seamlessly with the popular $N$-body code REBOUND to facilitate comparisons between calculation results and direct $N$-body integrations. The code is extensively documented and numerous example Jupyter notebooks illustrating its use are available online.

astro-ph.EP↗