Scientists have ruled out the idea that the universe is a simulation, but that didn’t stop another group of researchers from coming surprisingly close to creating one anyway, at least in spirit. A team led by physicist Marco Bonici at the University of Waterloo has developed a new emulator called Effort.jl, a machine learning tool capable of modeling the large-scale structure of the universe on nothing more than a personal laptop.
Published in the Journal of Cosmology and Astroparticle Physics, Effort.jl is described as a high-speed, differentiable emulator that mimics the behavior of complex cosmological simulations. It’s not a “Matrix-style” digital reality, but it does give scientists the power to reconstruct the cosmos thousands of times faster than traditional methods.
Effort.jl was built to tackle one of modern cosmology’s biggest bottlenecks: the sheer computational cost of analyzing data from galaxy surveys like DESI and the Euclid mission. These projects collect trillions of data points, charting the distribution of galaxies and dark matter across billions of light years. Normally, turning that data into insight takes days or even weeks of supercomputing time. Effort.jl can do it in minutes.
The system uses machine learning to replace the slowest calculations in a framework known as the Effective Field Theory of Large Scale Structure, which connects visible galaxies to the invisible web of dark matter that holds the cosmos together. By combining physics-based preprocessing with neural networks, the emulator predicts how galaxies cluster, a key metric known as the galaxy power spectrum, with near-perfect accuracy.
In tests, Effort.jl computed these results in just 15 microseconds on a single CPU core and reached Bayesian convergence in about ten minutes on a laptop, compared to several hours on high-end clusters. Its predictions matched those from existing models “within Monte Carlo noise,” meaning statistically indistinguishable results, only much faster.
What makes Effort.jl remarkable is that it isn’t a black box. While most AI systems work in ways even their creators can’t fully explain, this one is built around established physics, anchoring its learning to known laws of cosmic evolution. The result is a hybrid of artificial intelligence and hard science, fast, explainable, and precise.
The team even experimented with symbolic regression to find simplified analytical equations hidden within the data, cutting computation times from microseconds to nanoseconds. In other words, Effort.jl doesn’t just model the universe; it understands it.
If the recent anti-simulation study showed that reality can’t be computed, Effort.jl might be the next best thing, a tool that can recreate the universe’s behavior without needing infinite processing power. And unlike a hypothetical cosmic supercomputer, this one’s open source. Researchers can download it right now from GitHub and use it to explore everything from dark energy to galaxy formation.
It may not prove we’re living in a simulation, but it does let scientists build one of their own, and all it takes is a laptop.
