Astrophysicists have deployed artificial intelligence techniques to produce a complex three-dimensional simulation of the universe. The model runs very fast (producing results within a few milliseconds) and it stands up as robust, based on the different challenges to which it is subjected. With accuracy, the model currently has a relative error rate of only 2.8 percent. The faster time and greater accuracy are when the model is compared with an earlier method called second-order perturbation theory, or 2LPT. The earlier model was not only slower, its error rate was 9.3 percent.
Developing D3M
Produced by the Simons Foundation, with research drawn from multiple universities, and called the Deep Density Displacement Model (abbreviated to D3M), the computer simulation can accurately model how the cosmos would look when parameters are altered. This includes altering the dark matter composition of the universe.
What’s of particular interest in terms of the artificial intelligence is that the simulation will adapt to the parameters entered even though the AI has never received training data in relation to the different parameters that are varied. Once the AI is provided with a set of displacement vectors for the particles, the simulation can model the direction and distance the particles should be heading as the universe expands.
According to lead researcher Professor Shirley Ho: “It’s like teaching image recognition software with lots of pictures of cats and dogs, but then it’s able to recognize elephants.”
She adds that what is exactly occurring remains a bit of a mystery, saying: “Nobody knows how it does this, and it’s a great mystery to be solved.”
D3M was developed by feeding the artificial intelligence over 8,000 different simulations drawn from what was then the highest-accuracy models available.
Importance of modelling the universe
The types of computer simulations, of which D3M is the most advanced, play an important part in theoretical astrophysics. While instruments like the Hubble Space Telescope peer to the farthest reaches of space and provide visual data and measurements that reveal how galaxies looked as they took shape in a young universe, computer modeling aids astrophysicists in interpreting what they are seeing to improve their understanding of how galaxies first formed.
The models help researchers to learn how the cosmos may evolve under various scenarios. An example is with assessing what happens if the dark energy pulling the universe apart varied over time; or how gravity shapes the universe, to determine just how gravity shifts billions of individual particles across the entire age of the universe.
Around 68 percent of the universe is dark energy. Dark matter makes up about 27 percent; and the remainder – include everything on Earth – is normal matter. Dark energy is an unknown form of energy which is is thought to permeate all of space, and which is thought to be the driver for accelerating the expansion of the universe.
Previously such computations took days to run, requiring several thousand of simulations. With the new model these assessments have become very fast.
Research paper
The research has been published in Proceedings of the National Academy of Sciences, with the research paper titled “Learning to predict the cosmological structure formation.”
Essential Science
This article is part of Digital Journal’s regular Essential Science columns. Each week Tim Sandle explores a topical and important scientific issue. Last week we learned about the development of so-called ‘Universal Memory’, which has the potential to replace Dynamic Random Access Memory (DRAM) and flash drives. The new computer memory system will lead to ultra-low energy consumption.
The week before we considered how the differences between morning or evening effects on the body in terms of exercise regimes. Here scientists from the University of Copenhagen have discovered that the impact of exercise appears to differ according to times of day.
