Recently, foreign media reported that a team led by Marco Bonici at the University of Waterloo has developed a software solution called Effort.jl. Instead of running for days on a supercomputer, the software can now map the universe in minutes on a standard laptop, without compromising accuracy.
Previously, researchers used "effective field theory of large-scale structure" to describe these structures. This theory integrates countless detailed processes into a statistical model to explain the large-scale order of the universe. However, its computational complexity is extremely large. As each new cosmic mapping project generates ever-larger amounts of data, traditional analysis methods are reaching their limits. Effort.jl is a simulator: it learns how the EFTofLSS model responds to input data and can then calculate these responses orders of magnitude faster.
The core advantage of Effort.jl lies not only in its speed but also in the fact that its predictions are "differentiable." This means that for each input value, it is possible to calculate how the result will change if a small adjustment is made to that value. This provides a "map" that guides how the model will behave when a specific parameter is adjusted.
This is where the modern statistical method Hamiltonian Monte Carlo sampler comes in. Unlike traditional methods that rely on blind trial and error, HMC leverages the gradient information provided by Effor.jl to indicate which directions small changes are most effective. This allows for a faster and more direct traversal of the space of possible solutions, leading to the optimal values of the cosmological parameters. Previously, HMC was difficult to implement in cosmological models because this information had to be obtained separately for each calculation. Effor.jl automatically provides this information, making HMC a practical option for the first time.
Tests on PT-Challenge and BOSS galaxy mapping data show that Effor.jl's results are nearly identical to those of established codes like pybird, but in a fraction of the time. In some cases, it even enables more detailed analysis, as modules previously omitted due to time constraints can now be included.
Effor.jl will be a critical tool in the face of the upcoming data deluge from DESI and Euclid. It combines the precision of classical theory with the efficiency of modern simulation techniques, laying a solid foundation for more detailed mapping of the cosmic web and deepening our understanding of dark energy and cosmic dynamics.
