Project Motivation
Understanding the nature of Dark Matter is one of the most compelling challenges in modern astrophysics and cosmology.
While the existence of Dark Matter is well-supported by an abundance of observational evidence, a clear description of its particle nature remains elusive. The standard cosmological paradigm identifies two key features in the description of Dark Matter. First, that it interacts gravitationally — but otherwise minimally — with other particles (i.e. it is collisionless). Second, that it had minimal thermal velocities in the early Universe (i.e. it is cold). Taken together, these assumptions characterize the very successful Cold Dark Matter (CDM) model.
Despite its success in explaining large-scale structures in the universe, CDM faces potential challenges on smaller scales. Some of the most significant discrepancies between CDM and galaxy formation models concern the properties of the inner-most regions of individual galaxies, as well as the overall population statistics of all of these galaxies taken together. Alternative Dark Matter models offer avenues for preserving the successes of CDM while potentially alleviating some of these tensions.
A critical challenge arises when investigating alternative Dark Matter models: their observable impact on galaxies is not only subtle, but can also be degenerate with astrophysical processes. This raises a central motivating question for the DREAMS project: Can we reliably identify signatures of the influence of alternative Dark Matter models in the face of complex galaxy formation?
There are a wealth of different particle physics models for Dark Matter and an expansive range of possibilities for its basic properties.
Are there one or many Dark Matter particles? What are their masses and spins? How do they interact with each other and the Standard Model? These are just a small number of the basic questions that remain unanswered. Galaxies offer a unique opportunity to use the known gravitational effects of Dark Matter to learn about its other particle-physics properties.
To break down this seemingly intractable problem, we divide the wealth of Dark Matter models into different categories, each of which captures a generic feature that leads to a relevant galaxy observable. For example, we can consider scenarios where the Dark Matter impacts how the galaxies form in the very early Universe and scenarios where it affects the internal structure of galaxies later in time. Thinking about Dark Matter in terms of these broad model categories allows us to draw generic conclusions when comparing simulation predictions to data and to learn about general Dark Matter properties that can ultimately be mapped back to specific theories of interest.
Simulating Galaxy Formation with Alternative Dark Matter
Cosmological simulations are powerful tools that can produce realistic galaxy populations resembling the observed Universe.
Modern galaxy formation simulations incorporate an array of important physical processes including gravity, gas dynamics, star formation, and feedback from stars and black holes. These simulations create a mapping from the early Universe initial conditions — based on the density fluctuations observed in the Cosmic Microwave Background — to the observable galaxy populations at later times. These simulations are now well-established elements within the field of galaxy formation owing to the leverage they provide when trying to probe the physics that drives the emergence of galaxy properties.
There are two key features that are critically important to the scientific utility of galaxy formation simulations. First, simulations enable full access to the time domain — meaning that we can explore galaxy populations at different times and even track galaxy populations as they evolve. Second, we have full control over the input physics and model assumptions. This second feature allows us to run numerical experiments: we can execute galaxy formation models while varying its parameters or those of the Dark Matter model — or both. This allows us to produce predictions for galaxy properties under the influence of varied Dark Matter models, while also contextualizing the relative impact of the Dark Matter model against variations in the galaxy formation physics.
At the core of the DREAMS project is a massive galaxy formation simulation effort. Our simulations include traditional cosmological boxes, as well as zoom-in simulations of both Milky Way and dwarf galaxies.
Applying Artificial Intelligence
One of the most powerful ways to learn about the nature and properties of dark matter is to study its signatures on astrophysical observables. Unfortunately, two major problems complicate that mission: 1) the observables with the highest signal-to-noise ratio are unknown, and 2) uncertain baryonic effects can mimic dark matter signatures.
Historically, the community has been searching, rather manually, for observables with high signal-to-noise and minimal baryonic contamination. However, recent advances in machine learning allow us to tackle this problem in a very different manner. If we can simulate the complex non-linear interaction between dark matter, gas, and stars for different universes created with different dark matter models and different strengths of astrophysics, we can train neural networks to learn to identify the observables with the highest S/N ratio while marginalizing over baryonic effects.
The simulations of the DREAMS project have been designed to exploit recent deep-learning techniques and to guide us in identifying the best observables to learn about the nature and properties of dark matter.