Abstract
We present a novel method for estimating the physical properties of a progenitor dwarf galaxy from the tidal stream of stars it leaves behind as it is disrupted by the Milky Way’s gravitational field. By leveraging the PetaFLOPS-scale distributed computing power of MilkyWay@home, we demonstrate that the mass and radial profiles of both stellar and dark matter components can be reconstructed with high precision. Our algorithm utilizes N-body simulations and a likelihood function based on the Earth Mover Distance (EMD) to optimize five key parameters: stellar mass, stellar radius, dark matter mass, dark matter radius, and orbital time. Notably, we show that information about extended dark matter halos can be extracted from stellar debris even when the dark matter extends significantly beyond the visible stellar component. This approach provides a vital alternative to mass estimation methods that rely on the assumption of virial equilibrium, which is inherently violated in tidal streams.
Introduction
When a dwarf galaxy falls into the Milky Way, the intense gravitational gradient—tidal forces—strips stars and dark matter from the smaller body, stretching them into elongated structures known as tidal streams. These streams, such as the Orphan Stream, serve as "galactic fossils" that encode the history of the progenitor galaxy and the gravitational potential of the Milky Way.
However, decoding this information is a significant challenge. Traditional methods for measuring the mass of dwarf galaxies often rely on the assumption of virial equilibrium—the idea that the galaxy is a stable, balanced system. Because tidal streams are by definition unstable and undergoing disruption, these traditional models are inapplicable. Furthermore, finding the specific initial conditions that result in a modern-day stream requires thousands of computationally expensive N-body simulations, a task that exceeds the capacity of standard local computing clusters.
Methodology
Distributed Computing with MilkyWay@home
To overcome the computational bottleneck, this research utilizes MilkyWay@home, a volunteer computing project. By harnessing the idle processing power of over 14,000 volunteers' home computers, the project accesses approximately 800 TeraFLOPS of power. This allows for the massive parallelization required to run tens of thousands of N-body simulations across a wide parameter space.
Simulation and Modeling
The progenitor galaxy is modeled using a Double Plummer Model, consisting of two distinct components: a visible stellar core and a much larger, invisible dark matter halo. Each simulation evolves 20,000 particles (10,000 stars and 10,000 dark matter particles) through a fixed Milky Way potential. To find the best-fit model, the researchers employed Differential Evolution, a genetic-style optimization algorithm that searches across five dimensions: stellar mass, stellar scale radius, dark matter mass, dark matter scale radius, and the time since the progenitor was at its furthest point in orbit (apocenter).
Earth Mover Distance (EMD)
A key innovation in this work is the use of the Earth Mover Distance (EMD) as a metric for comparing simulated streams to observed data. Traditional "bin-by-bin" comparisons are often "brittle"; if a simulation is slightly shifted, it may be penalized heavily even if the overall shape is correct. EMD treats the distribution of stars like piles of dirt and calculates the minimum "work" required to move the simulated distribution to match the target data. This creates a smoother likelihood surface, allowing the optimization algorithm to converge more reliably on the true parameters.
Results and Validation
The algorithm was tested using synthetic datasets designed to mimic the Orphan Stream. Starting from a broad search range (e.g., dark matter masses ranging from $10^4$ to $10^{10}$ solar masses), the system successfully recovered the "true" parameters of the simulated progenitor:
- Stellar Components: Stellar mass and scale radius were recovered with high accuracy, within 1% and 2% of the true values, respectively.
- Dark Matter Components: Despite being invisible, the dark matter mass was recovered within 20% and its scale radius within 10%.
- Efficiency: The distributed network reached convergence in approximately two to three weeks, evaluating roughly 50,000 different model configurations.
A significant finding is that the algorithm could accurately recover dark matter properties even when 97% of the dark matter was located outside the radius of the stars. This confirms that the gravitational influence of an extended dark matter halo leaves a distinct "fingerprint" on the distribution of visible stars in the resulting tidal stream.
Limitations and Future Work
While this study demonstrates a powerful proof-of-concept, it relies on several simplifying assumptions. The simulations assumed the Milky Way’s gravitational potential and the dwarf galaxy’s orbit were known exactly. In reality, these are variables that must be solved for. Additionally, the current model does not account for the gravitational influence of the Large Magellanic Cloud (LMC), which is known to perturb the Orphan Stream.
The next phase of this research involves expanding the algorithm to fit the Milky Way’s potential and the dwarf galaxy’s properties simultaneously. By applying this method to multiple tidal streams at once, the team aims to build a high-resolution map of the dark matter distribution throughout our Galaxy, providing new insights into the nature of dark matter and the evolution of the Milky Way.
Summary updated