Mapping the Milky Way’s Dark Side

by

Author: Denis Avetisyan

New simulations and machine learning techniques are refining our understanding of dark matter’s distribution within our galaxy, paving the way for more sensitive direct detection experiments.

Despite uncertainties in modeling baryonic feedback, the distribution of local dark matter speeds-calculated for a dataset emulating galactic dynamics at 8.3 kpc and restricted to stellar masses between $10.60$ and $10.78$ and local standard of rest masses between $10.975$ and $11.085$-remains consistent whether those feedback parameters are varied or fixed at TNG fiducial values, and notably encompasses predictions from the Standard Halo Model, suggesting that inherent galactic variance and halo mass uncertainty dominate over the specific details of baryonic physics.
Despite uncertainties in modeling baryonic feedback, the distribution of local dark matter speeds-calculated for a dataset emulating galactic dynamics at 8.3 kpc and restricted to stellar masses between $10.60$ and $10.78$ and local standard of rest masses between $10.975$ and $11.085$-remains consistent whether those feedback parameters are varied or fixed at TNG fiducial values, and notably encompasses predictions from the Standard Halo Model, suggesting that inherent galactic variance and halo mass uncertainty dominate over the specific details of baryonic physics.

The DREAMS Project disentangles the impact of halo variance and baryonic feedback on the local dark matter velocity distribution.

Constraining the nature of dark matter requires precise modeling of its distribution within galaxies, yet significant uncertainties remain in predicting the local dark matter velocity distribution. This research, presented in ‘The DREAMS Project: Disentangling the Impact of Halo-to-Halo Variance and Baryonic Feedback on Milky Way Dark Matter Speed Distributions’, leverages a large suite of hydrodynamical simulations and machine learning to demonstrate that halo-to-halo variance is the dominant factor governing these uncertainties, dwarfing the effects of baryonic feedback. Our analysis provides a state-of-the-art prediction for the Milky Way’s dark matter speed distribution, revealing deviations from the Standard Halo Model and comparable uncertainties to those present in direct detection experiments. Will these improved constraints ultimately unlock the secrets of dark matter’s fundamental properties?

The Illusion of Smoothness: Mapping the Dark Matter Velocity Landscape

Direct detection experiments, designed to observe the faint interactions between dark matter particles and ordinary matter, fundamentally depend on predicting how quickly these particles are likely to be moving through our galaxy. This prediction often relies on the Standard Halo Model (SHM), a theoretical framework that treats the galactic dark matter distribution as a smooth, symmetrical halo with particles following a Maxwell-Boltzmann velocity distribution. Essentially, the SHM provides a baseline expectation for the speed of incoming dark matter – a crucial parameter in interpreting experimental data. Because the interaction rate between dark matter and detectors is highly sensitive to particle velocity, an inaccurate assessment of this speed distribution can lead to misinterpretations of results, potentially masking a true signal or creating false positives. Consequently, the SHM, while a convenient starting point, represents a significant assumption in the search for these elusive particles and motivates ongoing efforts to refine and validate its predictions.

The Standard Halo Model (SHM) posits a simple Maxwell-Boltzmann distribution to describe the velocities of dark matter particles surrounding our galaxy, but this assumption increasingly clashes with detailed galactic simulations. These simulations reveal a far more nuanced picture, demonstrating that dark matter isn’t smoothly distributed, but rather exists within complex substructures – streams and clumps created by gravitational interactions over cosmic time. Consequently, the local dark matter velocity distribution isn’t the neat bell curve predicted by the SHM; it’s likely characterized by multiple velocity peaks, anisotropic features, and even coherent streams passing through the solar system. This complexity means current dark matter detection experiments, designed under the SHM’s assumptions, may be significantly underestimating the expected signal, or even misinterpreting potential interactions, necessitating a move toward more sophisticated analysis techniques and simulations that accurately capture the true, messy reality of dark matter in our galactic neighborhood.

Determining the velocity distribution of local dark matter is paramount to the success of direct detection experiments. These searches rely on the rare interactions between dark matter particles and atomic nuclei, and the interaction rate is heavily influenced by how quickly these particles are moving through the galaxy. A miscalculation of the typical dark matter speed – whether it’s an overestimate or underestimate – can lead to false negatives or misinterpretations of potential signals. Currently, many experiments assume a simple Maxwell-Boltzmann distribution to model these velocities, but increasingly sophisticated simulations suggest this may be an oversimplification. Accurately characterizing the local dark matter velocity distribution, accounting for phenomena like tidal streams and dark matter substructures, will not only improve the precision of existing experiments but also guide the development of new, more sensitive detectors capable of finally unveiling the nature of this elusive substance.

The continued absence of conclusive dark matter detection, despite decades of increasingly sensitive experiments, demands a critical reevaluation of the theoretical frameworks guiding these searches. Current analyses often rely on simplified simulations of the Milky Way’s dark matter halo, potentially leading to misinterpretations of null results. The discrepancy between predicted interaction rates and observed data isn’t necessarily evidence against specific dark matter particle candidates; instead, it strongly suggests that the assumed astrophysical environments – particularly the local dark matter density and velocity distribution – are not accurately captured by existing models. Consequently, researchers are developing more sophisticated simulations incorporating factors like halo substructure, streams, and non-Maxwellian velocity distributions, alongside novel analysis techniques to account for these complexities and enhance the prospects for a definitive dark matter discovery.

An emulated dark matter density distribution closely matches the original simulation data, as demonstrated by a low Earth Mover’s Distance of 0.010 GeV cm⁻³.
An emulated dark matter density distribution closely matches the original simulation data, as demonstrated by a low Earth Mover’s Distance of 0.010 GeV cm⁻³.

Simulating the Darkness: DREAMS and the Quest for Realism

The DREAMS suite consists of 1024 cosmological simulations, each modelling a galaxy with a mass comparable to that of the Milky Way. These simulations were performed using the $N$-body code ACE, with a mass resolution of approximately $3 \times 10^6$ solar masses, allowing for detailed analysis of dark matter halo properties. The large number of simulated galaxies enables robust statistical studies of dark matter distributions and velocity dispersions, facilitating the characterization of dark matter substructure and its dependence on halo mass and formation history. Data products include fully resolved dark matter particle data, allowing for precise measurements of dark matter density profiles, velocity distributions, and the abundance of subhalos within the simulated Milky Way-mass halos.

Baryonic feedback, encompassing processes like stellar winds and active galactic nuclei (AGN) jets, significantly alters the distribution of dark matter within halos. These energetic outputs from stars and supermassive black holes deposit energy into the surrounding gas, driving outflows and suppressing star formation. Consequently, dark matter is displaced outwards, leading to less concentrated halos and modified velocity distributions compared to scenarios neglecting these effects. The DREAMS simulations incorporate models for both thermal and kinetic feedback, allowing for a quantitative assessment of how these baryonic processes shape the internal structure and dynamics of dark matter halos across a range of masses and redshifts. Accurate modelling of these feedback mechanisms is essential for bridging the gap between cosmological simulations and observations of galactic structure and kinematics.

Comparison with the TNG50 simulation suite was undertaken to validate the results produced by the DREAMS simulations and to quantify potential systematic uncertainties. TNG50, a well-established cosmological simulation, provides an independent dataset for assessing the fidelity of the DREAMS dark matter distributions and velocity dispersions. This involved a direct comparison of key statistical properties, such as halo mass functions, dark matter density profiles, and velocity anisotropy parameters, between the two simulation suites. Discrepancies observed between DREAMS and TNG50 are carefully analyzed to identify the source of systematic errors and to establish confidence intervals on the derived dark matter properties. The established nature of TNG50 allows for a robust assessment of any biases introduced by the specific methodologies employed within the DREAMS framework.

The DREAMS simulations utilize halo mass as a primary input parameter, directly influencing the resolution and characteristics of each simulated system. Specifically, the simulations are performed at varying resolutions dependent on the modelled halo mass, with more massive halos receiving higher resolutions to accurately capture internal dynamics and substructure. This dependency is crucial because the number of particles used to represent a halo scales with its mass; lower-mass halos are simulated with fewer particles, while higher-mass halos utilize significantly larger particle counts. Consequently, results from DREAMS are intrinsically linked to the chosen halo mass, necessitating careful consideration when interpreting findings and comparing simulations across different mass ranges.

Comparing dark matter speed distributions between the DREAMS and TNG50 simulations reveals that while DREAMS successfully emulates TNG50's distributions after rescaling to match the Milky Way's mass, it exhibits larger uncertainties for TNG50-2 galaxies and limitations in modeling the most massive halos.
Comparing dark matter speed distributions between the DREAMS and TNG50 simulations reveals that while DREAMS successfully emulates TNG50’s distributions after rescaling to match the Milky Way’s mass, it exhibits larger uncertainties for TNG50-2 galaxies and limitations in modeling the most massive halos.

From Particles to Probability: Reconstructing the Velocity Landscape

Kernel Density Estimation (KDE) was employed to determine the local dark matter velocity distribution from the extensive particle data generated by the DREAMS simulations. This non-parametric method estimates the probability density function by placing a kernel – a weighted function – on each particle and summing the contributions. The resulting distribution represents the likelihood of finding dark matter particles with specific velocities within a given region. For the DREAMS simulations, KDE allows for the reconstruction of a continuous velocity distribution from discrete particle data, providing a detailed description of the dark matter kinematics within simulated halos. The bandwidth of the kernel is a critical parameter, influencing the smoothness of the estimated distribution and requiring careful calibration to avoid over- or under-smoothing.

A Normalizing Flows Emulator was integrated into the workflow to accelerate and refine the reconstruction of dark matter velocity distributions. This emulator functions as a surrogate model, predicting the full dark matter speed distribution given key halo properties, specifically halo mass and the impact of baryonic feedback processes. By learning the mapping between these halo characteristics and the resulting velocity distribution from the DREAMS simulations, the emulator bypasses the computationally intensive Kernel Density Estimation (KDE) applied to individual particles. This approach significantly reduces processing time while maintaining a high degree of accuracy, facilitating broader exploration of parameter space and statistical analysis of halo-to-halo variations.

The Normalizing Flows Emulator functions by establishing a direct relationship between key halo properties – specifically, halo mass – and the resulting dark matter speed distribution. Critically, the emulator incorporates the impact of baryonic feedback processes, such as energy injection from supernovae and active galactic nuclei, which alter the gravitational potential and, consequently, the dark matter distribution. By modeling this dependence, the emulator facilitates a significant reduction in computational cost, enabling the rapid evaluation of dark matter distributions across a wide range of halo masses and baryonic feedback strengths, and thus allowing for efficient exploration of the parameter space governing halo formation and evolution.

Analysis of the DREAMS simulations, leveraging Kernel Density Estimation and a Normalizing Flows Emulator, enables quantitative assessment of dark matter distribution variance between dark matter halos. This is achieved by reconstructing the local dark matter speed distribution within numerous halos and comparing the resulting distributions. Statistical measures, such as the standard deviation and percentiles of velocity dispersion, are calculated for each halo, allowing for direct comparison of their dark matter content. The method accounts for halo mass and baryonic feedback effects, providing a standardized framework to isolate and quantify the intrinsic, natural variation in dark matter distributions, independent of these influencing factors.

The emulator accurately captures the relationship between gas speed distributions and baryonic parameters, with halo-to-halo variance dominating uncertainty, though halo mass contributes up to a 30% difference.
The emulator accurately captures the relationship between gas speed distributions and baryonic parameters, with halo-to-halo variance dominating uncertainty, though halo mass contributes up to a 30% difference.

Beyond Simplification: The Implications for Detection

Simulations of dark matter distribution in the Milky Way consistently demonstrate substantial departures from the predictions of the Standard Halo Model (SHM). The SHM, which posits a spherically symmetric and isotropic dark matter halo, often oversimplifies the complex gravitational interactions within the galaxy. These simulations reveal a more nuanced picture, characterized by anisotropies and asymmetries in the local dark matter velocity distribution. Specifically, the simulations suggest a non-Maxwellian velocity distribution with a steeper fall-off at high speeds and a possible substructure resulting from the accretion history of the galaxy. This means that the expected velocity dispersion of dark matter particles interacting with detectors on Earth is significantly different from what the SHM predicts, impacting the recoil energy observed in direct detection experiments. The deviation isn’t merely quantitative; it fundamentally alters the shape of the expected signal, potentially explaining the lack of conclusive detections despite years of searching.

Current direct detection experiments, designed to observe the faint interactions between dark matter particles and ordinary matter, rely on predictions derived from the Standard Halo Model (SHM). However, simulations reveal that the actual distribution of dark matter velocities in our galactic neighborhood deviates significantly from these SHM assumptions. These deviations directly impact the expected event rate – the number of detectable interactions – within these experiments. Specifically, a non-SHM distribution can suppress the predicted signal, potentially explaining the null results reported by numerous ongoing searches. A lower event rate means interactions are less frequent and harder to distinguish from background noise, effectively masking any potential dark matter signal. Therefore, accounting for these simulated deviations is crucial, as it may reconcile existing experimental data with theoretical expectations and illuminate the path towards a successful dark matter detection.

Refining models of the local dark matter velocity distribution is paramount to enhancing the capabilities of direct detection experiments. Current searches assume a smooth, Maxwellian distribution – the Standard Halo Model – but simulations suggest a more complex reality, featuring streams and substructures. By incorporating these nuances into experimental simulations, researchers can more accurately predict the expected signal, differentiating it from background noise and improving event selection. This precision is critical because the predicted event rate is highly sensitive to the assumed velocity distribution; even small inaccuracies can significantly impact the search sensitivity. Consequently, future experiments can be strategically designed, focusing on specific velocity ranges and energy thresholds to maximize their potential for discovering dark matter interactions and ultimately revealing the nature of this elusive substance.

Accurately characterizing the local dark matter velocity distribution is paramount to successfully interpreting the results of direct detection experiments and, ultimately, revealing the nature of dark matter. Current searches rely on assumptions about how fast dark matter particles are moving through our galaxy, often based on the Standard Halo Model. However, subtle variations in this distribution – influenced by gravitational forces from the galactic disk, the Sun, and even dark matter self-interactions – can significantly alter the expected signal. Without a precise understanding of these nuances, experiments may misinterpret null results or, conversely, falsely identify a signal. Refined models of the local dark matter distribution, therefore, serve as an essential tool for both validating existing constraints and optimizing the design of future detectors, bringing scientists closer to solving one of the most enduring mysteries in modern physics.

The emulator accurately reproduces DM speed distributions from the full simulation across various percentiles, demonstrating excellent agreement even when restricted to galaxies with stellar masses between 10.60 and 10.78 and LSR masses between 10.975 and 11.085.
The emulator accurately reproduces DM speed distributions from the full simulation across various percentiles, demonstrating excellent agreement even when restricted to galaxies with stellar masses between 10.60 and 10.78 and LSR masses between 10.975 and 11.085.

The DREAMS project, with its intricate hydrodynamical simulations, seeks to map the unseen architecture of dark matter within the Milky Way. It is a humbling endeavor, this attempt to quantify the unknowable. As Erwin Schrödinger observed, “We must be aware that the uncertainty principle is not a limitation of our knowledge, but a fundamental property of nature itself.” This sentiment echoes the challenges faced in modeling dark matter; the velocity distributions are not simply waiting to be discovered, but are inherently probabilistic. The simulations, while powerful, operate within the boundaries of computational precision and model assumptions – a finite grasp at an infinite reality. Any predictive power derived, any constraint placed on direct detection experiments, must acknowledge this inherent uncertainty, recognizing that even the most sophisticated theory can vanish beyond the limits of its own construction.

What Lies Beyond the Horizon?

The pursuit of dark matter’s local velocity distribution, as exemplified by the DREAMS project, feels less like solving a puzzle and more like charting an increasingly detailed map of ignorance. Each refinement of the simulated galactic halo, each application of machine learning, merely sharpens the edges of what remains unknown. It is a useful exercise, certainly, to constrain the parameter space available to direct detection experiments, but one should not mistake increased precision for genuine understanding. Every theory is just light that hasn’t yet vanished.

The reliance on hydrodynamical simulations, even with the sophistication demonstrated here, introduces inherent limitations. These models are, after all, constructions built upon assumptions – assumptions about star formation, feedback processes, and the very nature of dark matter itself. A discrepancy between simulation and observation will not necessarily reveal the dark matter; it might merely expose a flaw in the scaffolding upon which the simulation rests. The true signal may lie in the unmodeled physics, the effects currently relegated to the ‘unknown unknowns.’

Future work will undoubtedly push for higher resolution, larger simulation volumes, and more intricate baryonic physics. But perhaps the most fruitful path lies in embracing the unexpected. To actively seek out the anomalies, the deviations from predicted behavior. For it is in these cracks, these imperfections, that the universe often whispers its most profound secrets. Models exist until they collide with data, and the horizon is always closer than it appears.

Original article: https://arxiv.org/pdf/2512.04157.pdf

Contact the author: https://www.linkedin.com/in/avetisyan/

See also:

2025-12-06 14:07