Dark matter is the most abundant matter in the Universe, however what is its nature remains one of the greatest questions in modern science. At time of writing, nothing is known about its interactions other than it feels the pull of gravity. Specifically, it is assumed that dark matter does not self-interact. In-fact, if dark matter does self-interact it will alter the distribution of ALL matter in the Universe. However, measuring this effect is incredibly difficult as it is very subtle and can often be confused with other physical effects.
Galaxy clusters represents some of the highest concentrations of dark matter known and as such, subtle changes in its interacting properties can be detected. Not only this, their massive concentrations allow us to use gravitational lensing to map out its distribution in to two-dimensional maps. We can use these maps to infer the underlying interactions of dark matter.
To do this, I developed a suite of simulations that predict the distribution of dark matter for different self-interacting dark matter models. However, even with these simulations, inferring the dark matter model is difficult and requires innovative machine learning methods.
Convolutional Neural Networks (CNN) are an innovative away to analyse multi-dimensional images. Below is a introductory lecture I wrote on CNNs. We can use these innovative methods to constrain dark matter interactions. Using the simulations, the neural network can learn the difference between each model, such that when we feed in the true data it can tell us the true underlying model.
darkCNN is a simple CNN that trains on a sample of simulated clusters and can then estimate the underlying dark matter model. It is easily installed and understood. Simple clone the GitHub account and python setup.py install it!
Galaxy clusters represents some of the highest concentrations of dark matter known and as such, subtle changes in its interacting properties can be detected. Not only this, their massive concentrations allow us to use gravitational lensing to map out its distribution in to two-dimensional maps. We can use these maps to infer the underlying interactions of dark matter.
To do this, I developed a suite of simulations that predict the distribution of dark matter for different self-interacting dark matter models. However, even with these simulations, inferring the dark matter model is difficult and requires innovative machine learning methods.
Convolutional Neural Networks (CNN) are an innovative away to analyse multi-dimensional images. Below is a introductory lecture I wrote on CNNs. We can use these innovative methods to constrain dark matter interactions. Using the simulations, the neural network can learn the difference between each model, such that when we feed in the true data it can tell us the true underlying model.
darkCNN is a simple CNN that trains on a sample of simulated clusters and can then estimate the underlying dark matter model. It is easily installed and understood. Simple clone the GitHub account and python setup.py install it!