SolutionsX: supergravity theories/solutions as code, with an agentic workflow for machine-verifiable physics
The paper introduces SolutionsX, a Mathematica package that transforms supergravity solutions and other xAct-based physics into machine-verifiable, executable code, enabling an agentic workflow where AI agents can automatically codify, verify, and extend complex theoretical results with high accuracy.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
The universe, at its most fundamental level, is often described not as a collection of solid objects, but as a complex interplay of fields and shapes. Physicists who study the deepest layers of reality, such as those working on supergravity, try to write down the rules that govern how gravity interacts with other forces. These rules are incredibly intricate, involving mathematics that stretches across many dimensions and requires tracking how quantities change as they twist and turn through space. For decades, scientists have relied on computer programs to help them crunch these numbers, but a persistent problem has remained: the results of these calculations are often trapped in the minds of the researchers or buried in dense academic papers. If a scientist wants to check a colleague's work, they often have to start from scratch, re-deriving every step by hand to ensure no small typo or forgotten convention has slipped in. This process is slow, prone to error, and makes it difficult to build a reliable library of knowledge that others can trust and use.
A new approach, introduced by a researcher at the Bulgarian Academy of Sciences, aims to change how these calculations are stored and shared. The work presents a software tool called SolutionsX, which transforms the way theoretical physics results are handled. Instead of leaving a solution as a static description in a paper, this tool turns the entire calculation into a living piece of code. When a physicist uses this system, they are not just writing down an answer; they are creating a digital record that can be loaded into a fresh computer session, run again, and verified instantly. The system acts as a strict, impartial referee, checking every step of the logic against the laws of physics before the result is ever accepted. This ensures that the work is not just a claim, but a reproducible fact. The tool is designed to handle the specific, messy details of supergravity, where the geometry of space and the behavior of particles are inextricably linked, but its design is flexible enough to apply to a wide range of complex physical theories.
The core innovation lies in how the software manages the "life" of a solution. In the past, a physicist might define a theory, propose a specific scenario like a black hole, and then perform a long series of calculations to see if the scenario fits the theory. With this new system, the theory and the solution are stored together as a single, organized package. When a user loads this package, the computer automatically rebuilds the entire mathematical environment, including all the definitions and rules needed for that specific problem. The system then runs through the calculations again, checking that the proposed solution actually satisfies the equations of motion. If it does, the solution is saved as a verified record. This record can then be shared with others, who can load it, inspect the steps, and even use it as a foundation to build new ideas. It turns a solitary calculation into a communal asset, where every entry is a trusted building block for future research.
To demonstrate the power of this method, the author followed a specific, complex solution known as a supersymmetric black hole through the entire workflow. The process began by loading a stored theory of five-dimensional gravity. The researcher then input the specific details of the black hole, defining its shape and the fields that surround it. The software took these definitions and computed the values of every relevant physical quantity, such as the curvature of space and the strength of the magnetic fields, breaking them down into their component parts. Once the numbers were calculated, the system immediately tested them against the fundamental equations of the theory. It checked if the black hole was stable and if it preserved the necessary symmetries required by the laws of physics. The result was a verified entry, saved in a database where it could be retrieved later. This entry was then used to show how this specific black hole is related to other known solutions, proving that one could be derived from the other by adjusting certain parameters. The entire process, from input to verification, was recorded in a way that anyone could replay and confirm.
The paper also explores how this system interacts with artificial intelligence, a field that has recently become interested in scientific discovery. The author developed a toolkit that allows an AI agent to use the software to solve physics problems on its own. In a series of tests, an AI was given the task of recreating solutions from published scientific papers. Without the new toolkit, the AI struggled, taking many hours and failing to reach a correct conclusion. However, when equipped with the toolkit, which provided clear instructions and a way to verify its work at every step, the AI successfully recreated the complex physics calculations in under three hours. The AI did not just guess; it wrote code, ran the calculations, and used the software to check its own answers. If a step failed, the system flagged the error, and the AI adjusted its approach. This experiment showed that when an AI is given a reliable, machine-verifiable environment, it can become a powerful collaborator, capable of handling the tedious and error-prone aspects of theoretical physics that have traditionally slowed down human progress.
The success of these tests suggests a shift in how scientific knowledge might be accumulated in the future. Instead of a library of papers that must be read and manually checked, the vision is a database of verified code. In this future, a researcher could ask a computer to find a solution to a new problem, and the system would search through a vast collection of existing, verified examples to find the right path. The AI would act as a tireless assistant, testing ideas against the strict rules of the software and only presenting results that have been proven to work. This does not replace the human scientist, who still provides the creativity and the initial questions, but it removes the bottleneck of verification. It allows researchers to build upon a foundation of absolute certainty, knowing that the results they are using have been checked by a machine that cannot be tired or distracted.
The paper concludes by outlining the current capabilities of the system and the steps needed to expand it further. While the software currently excels at handling supergravity and related theories, the author notes that there are many other areas of physics where this approach could be applied. The system is open for others to contribute, with a plan to grow the database of verified solutions. The ultimate goal is to create a resource where the boundary between human insight and machine verification is blurred, allowing for a faster, more reliable pace of discovery. By turning physics into code that can be run, checked, and shared, the work offers a new way to ensure that the laws of the universe are understood not just by a few, but by a community that can verify every step of the journey.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.