The structure and dynamics of the Universe are phenomena eminently complex to reproduce, so much and so well that the current simulations on the most powerful require several minutes to calculate a rendering that sacrifices must be a large share of precision. But recently, thanks to the Artificial Intelligence, astrophysicists Americans have developed a new model capable of simulating the universe and its evolution in a few milliseconds only, while maintaining a precision spectacular.
For the very first time, American scientists have made use of artificial intelligence to create a complex simulation in three dimensions of the Universe. Called Deep Density variation version, or D3M , it is so fast and accurate as of the astrophysicists who designed do not understand precisely the entire mechanism that underlies it.
The study has been published in the journal PNAS. The model accurately simulates how the seriousness shapes the universe on billions of years. Each simulation only takes 30 milliseconds – compared to the minutes that are in general required to other simulations. In addition, D3M has learned the lessons of the 8000 simulations of training that the team has provided him – in the extrapolating and surpassing, capable to adjust settings on which it had not even been formed.
“It is as lead software for recognition of images with many images of cats and dogs, and then it becomes able to recognize elephants,” says the astrophysicist Shirley Ho of the Institute Flatiron and Carnegie Mellon University. “Nobody knows how this happens, and it is a mystery that we will have to solve.”
Simulate the universe
The simulations are critical to complement the observations that we do of the Universe. By performing simulations that produce results that correspond to our observations, as well as the Simulations which do not, the cosmologists can determine scenarios are the most likely to have produced the universe in which we live.
But the complexity of the history of our Universe makes such simulations rather tedious, which means that they take time. A single study may require thousands of simulations to obtain useful statistical data. It is there that intervenes D3M, developed by an international team of astronomers. It calculates how, on $13.8 billion of years (the age of the Universe), the gravity moves billions of particles in the space.
With a software not managed by the IA, this could take up to 300 hours of calculation for a single simulation extremely precise; this is also possible in just a few minutes, but the accuracy would suffer greatly. To resolve this problem, the research team decided to develop a network of neurons to run simulations and has resulted in D3M by providing him with 8000 different simulations, from a model extremely precise.
D3M – a quick simulation, precise and capable to adapt
Once the drive D3M was completed and that the IA was working correctly, the model was ready for testing. The researchers have asked him to simulate a cube of Universe of approximately 600 million light-years from the side. To assess its results, the team has also executed the same simulation with a method extremely slow of several hundreds of hours and a method taking only a few minutes. As expected, the slow method has produced the most accurate result, while the rapid method has generated a relative error of 9.3%.
The new model, called D3M , is both faster and more accurate than an existing method called “theory of the disturbance of the second order”, or 2LPT. The colors represent the error of moving average in millions of light-years for each point of the grid, in relation to a model of high precision (although much slower). Appropriations: S. He et al. 2019 D3M has surpassed all rapid methods previously. Its simulation was performed in just 30 milliseconds and, compared to the model slow but super precise, she had only a relative error of 2.8%.
Even more impressive, even if it had been formed to a single set of parameters, the network of neurons could predict the formation of the structure of the Universe simulated on the basis of other parameters for which it had not even been led – for example, if the quantity of black material was variable. This means that the IA may have the flexibility to adapt to a range of tasks of simulation – well that before this, the team hopes to understand how she managed to do exactly what it is doing.