As cosmologists ponder the universe – and other possible universes – the data available to them is so complex and vast that it can be extremely challenging for humans to understand.
By applying the scientific principles used to create models for understanding cell biology and physics and the challenges of cosmology and the big data, Cornell researchers have developed a promising algorithm for mapping a multifaceted set of probabilities.
The new method, which researchers used to visualize models of the universe, could help solve some of the greatest mysteries of physics, such as the nature of dark energy or the probable characteristics of other universes.
"Science works because things behave much more simply than they are entitled to," said James Sethna, a professor of physics and senior author of Visualizing Probabilistic Models With Intensive Principal Component Analysis, published June 24 in Proceedings of National. Academy of Sciences. "Very complicated things end up doing a fairly simple collective behavior."
This, he says, is because not every factor in a system is significant. For example, millions of atoms may be involved in a physical collision, but their behavior is determined by a relatively small number of constants. Data on the universe collected by powerful telescopes, however, have so many parameters that it can be challenging for researchers to find out what measures are most important to reveal insights.
The algorithm – developed by the first author Katherine Quinn, MS # 16, Ph.D. 19 – allows researchers to visualize a large set of probabilities to look for patterns or other information that might be useful – and provides them with a better intuition to understand complex models and data.
"Because we have much larger, better datasets with terabytes and terabytes of information, it's getting harder to really understand them," Quinn said. "A person can not just sit and do it. We need better algorithms that can extract what interests us, without telling us what to look for. We can not just say, "Look for interesting universes." This algorithm is a way of unraveling information in a way that can reveal the interesting structure of the data. "
To further complicate the task of researchers, the fact that the data consist of ranges of probabilities, rather than raw images or numbers. "It's a more complicated problem to deal with," Quinn said.
Your solution takes advantage of different properties of probability distributions to visualize a collection of things that could happen. In addition to cosmology, his model has applications for machine learning and statistical physics, which also work in terms of predictions.
To test the algorithm, researchers used data from Planck's satellite of the European Space Agency and studied it with co-author Michael Niemack, associate professor of physics, whose laboratory develops instruments to study the formation and evolution of the universe by measuring microwave radiation . They applied the model to data on the cosmic microwave background – radiation reminiscent of the early days of the universe.
The model produced a map representing possible characteristics of different universes, of which our own universe is a point. This new method of visualizing the qualities of our universe highlights the hierarchical structure of the model dominated by dark matter and dark energy that fits so well into the background data of cosmic microwave. While the structure is not surprising, these views present a promising approach to optimizing cosmological measurements in the future, Niemack said.
Then the researchers will try to expand this approach to allow more parameters for each data point. The mapping of this data could reveal new information about our universe, other possible universes, or dark energy – which seems to be the dominant form of energy in our universe, but about which physicists still know little.
"We only use coarse models to explain what dark energy could be, or how it could evolve over time," Niemack said. "There are a lot of different parameters that could be added to the models, so we could visualize them and decide which are the important measures to prioritize to try to understand which dark energy model best describes our universe."
Other collaborators were PhD student Colin Clement and Cornell's former research associate, Francesco De Bernardis. The research was partially supported by the National Science Foundation and the Natural Sciences and Engineering Research Council of Canada.