Limitless clean energy is one step closer to becoming a reality, following the latest nuclear fusion breakthrough.
Scientists from the UK and Austria have developed a new AI tool that can simulate the superheated plasma inside a fusion reactor.
The tool, dubbed GyroSwin, completes calculations within seconds that would usually take days on the world's most powerful supercomputers.
This could help scientists to understand how to harness the unpredictable power of fusion energy and build the world's first functioning reactors.
Fusion reactors replicate the processes found in the heart of the sun, where hydrogen atoms are smashed together and fused into helium.
However, to create a miniature star on Earth requires heating plasma to around 100,000,000°C and keeping it hot and dense enough for fusion to take place.
Since no material could withstand these temperatures, the plasma is trapped by powerful magnetic fields inside a doughnut–shaped device known as a tokamak.
With the help of GyroSwin's simulations, engineers should be able to fine–tune these magnetic fields to create a stable fusion reaction.
Nuclear fusion has the potential to create a nearly endless source of clean energy, and has previously been described as the 'holy grail' by scientists.
Two types of hydrogen – deuterium and tritium – are the only fuels needed, while the only byproduct is helium.
This means there are no mountains of long–lasting radioactive waste or greenhouse gas emissions to damage the planet.
The problem is that making fusion reactors a reality requires us to harness some of the most unpredictable forces in the universe.
Superheated plasma doesn't run around in a neat ring; it bounces and ripples in a process known as turbulence.
Co–creator of GyroSwin Dr Fabian Paischer, of the Johannes Kepler University in Linz, told Daily Mail: 'Plasma leaks out of its magnetic cage due to turbulence, meaning it loses its potency for the fusion reaction to occur.'
For this reason, fusion reactions tend to be extremely short–lived.
In fact, the current record for a sustained reaction sits at just 43 seconds!
To keep a fusion reaction going indefinitely, scientists will need extremely accurate simulations of how turbulence forms under different conditions.
Since the dynamics inside a plasma are so complex, you can't use the same kinds of simulations that we use to predict the weather or the flow of liquids.
The current best simulations track particles of plasma in five dimensions: three for their locations, one for their speed, and one for their direction relative to the magnetic field.
However, these simulations take days to complete, even when they are being run on the world's best supercomputers.
GyroSwin, which has been developed by the UK Atomic Energy Authority (UKAEA), Johannes Kepler University in Linz and an Austrian firm Emmi AI, offers a different solution.
Scientists first run extremely accurate – but expensive and slow – simulations on traditional supercomputers.
The results of these simulations are then used to train an AI so that it can learn to predict the subtle relationships between cause and effect.
Once the training is complete, GyroSwin can skip over the complex calculations and make predictions about the outcome of simulations in seconds rather than days.
These types of 'AI surrogate model' aren't new, but what makes GyroSwin so exciting is just how accurate it appears to be.
Dr Paischer says: 'GyroSwin is the first model that actually models the full plasma turbulence in all its beauty and across multiple scales.
'Previous approaches only tried to model turbulence in a reduced form, meaning they always neglect important information to make predictions more efficient at the cost of accuracy.'
Importantly, the model is already starting to show signs of capturing the underlying physics of plasma turbulence.
While the AI will still need some traditional simulations to keep improving its training, it could speed up the production of working nuclear reactors.
Rob Akers, Director of Computing Programmes at UKAEA, told Daily Mail: 'Fusion development is highly iterative, and credible designs can require very large numbers of simulations.
'Cutting turnaround from days to seconds can speed up design loops and "what if" exploration dramatically.
'It won’t solve fusion on its own, but it can materially speed up the engineering cycle –which is exactly what you need on the path to a working fusion machine.'
In its current form, GyroSwin is a proof of concept, but the researchers plan on scaling it up for more practical scenarios.
The goal is to use the AI to guide fusion reactors that are already running or are soon to be built.
That could include the MAST Upgrade experimental tokamak that is under construction near Oxford, or the UK's flagship STEP (Spherical Tokamak for Energy Production) project, which aims to build a functioning prototype reactor by the 2040s.
Although a real, fully functioning fusion reactor is still within the realm of science fiction, these fundamental breakthroughs bring it a little bit closer to reality.