OUR GOALS
Why we share our work.
Our ideas on math, AI, open research, and working together.
01 / THE VISION
Building open mathematical theory
This website is devoted to open mathematical research, particularly major conjectures and long-term programmes whose central aim is to construct new theories. At the same time, it seeks to develop systematic foundations for existing mathematics: precise definitions, lemmas and theorems, complete proofs, logical dependencies, and a body of knowledge that can be searched, cited, revised, and extended over time.
The project is inspired by the Stacks Project and the structural vision of Nicolas Bourbaki. The Stacks Project demonstrates how open collaboration, permanent references, and sustained revision can support a vast, rigorous mathematical theory. Bourbaki exemplifies a different but related ambition: to reveal the underlying unity of mathematics through axiomatic methods and structural thinking.
Mathematical theories often arise from different motivations and develop different languages, methods, and foundational assumptions. We hope to clarify their connections, identify common structures, and bring them into coherent frameworks wherever that is mathematically justified.
Yet our ambition extends beyond the organization of established knowledge. We also want to create new mathematical theories.
Mathematical progress includes the invention of concepts, the discovery of unexpected connections, and the construction of theories that did not previously exist. The significance of a conjecture may lie not only in its eventual solution but also in the ideas and methods developed in the attempt to solve it.
We therefore want to organize significant conjectures, partial results, known obstacles, and long-term proof programmes into research projects that can develop continuously. An open conjecture need not remain an isolated question: it can become the starting point of a growing theory, with results and methods of independent value.
Unfinished mathematical research itself should become knowledge that can be made public, cited, reviewed, and maintained together.
Traditional mathematical literature primarily records completed results. Much of the work that makes those results possible occurs before publication: the formulation of a problem, the emergence of an idea, the exploration of competing approaches, partial proofs, and the gradual identification of decisive obstacles. We want those stages to be organized and preserved, so that others can understand how a theory develops and continue the work from what is already known.
Conjecture Atlas is therefore both an evolving reference library and an active, long-term mathematical research project.
02 / AI FOR MATHEMATICS
Why now?
Making large-scale theory building possible
In the past, a project of this kind was extraordinarily difficult to undertake. To organize an entire mathematical field, researchers often had to master extensive and scattered literature, reconcile distinct technical languages, and spend years reconstructing definitions, proofs, and the architecture of a theory.
Such work has not always received recognition commensurate with the effort it demands. Consolidating a field, unifying different approaches, or supplying complete proofs of foundational results may require years, while producing few of the quickly identifiable outcomes favored by conventional academic evaluation.
Original theory building faces a similar difficulty. Before a new framework has solved a widely recognized open problem, its value may be difficult to appreciate, even when it has substantial long-term potential. Foundational work that could shape the future of mathematics can therefore be among the hardest kinds of research to sustain.
Artificial intelligence is beginning to change this situation. AI systems can help researchers examine large bodies of literature, compare definitions and arguments across theories, trace intricate dependencies, explore conjectures, formulate proof strategies, and organize extensive mathematical projects. These capabilities substantially reduce the practical barriers to systematic and long-term theory building.
I believe their significance goes beyond efficiency. AI may change which kinds of mathematical research can be undertaken, and what a mathematical project can look like.
Instead of organizing every project around a result expected within a short period, we can begin with a major conjecture, develop its supporting theory over time, publish meaningful intermediate progress, and revise a research programme as our understanding evolves. Even before the original question is resolved, this work may produce mathematics of lasting value.
Solving major open problems remains an extraordinary achievement. But mathematics advances equally through new concepts, the unification of theories, and foundational structures from which many later results can grow. As AI-assisted reasoning improves, these larger projects may take a more prominent place in mathematical research.
AI-assisted research and rigorous review
This website will rely extensively on artificial intelligence, and we want to be explicit and positive about that choice. AI will participate in literature review, theoretical organization, exploration of proofs, consistency checks, the detection of errors, and the planning of long-term work, as well as in mathematical writing.
For large projects, repeated checking of hypotheses, theorem dependencies, and alternative arguments is especially valuable. AI makes sustained and systematic review more feasible. We hope to develop a process that combines AI-assisted analysis, independent mathematical scrutiny, and formal verification, with visible evidence for each published claim.
Our intended practice is to publish complete and self-contained proofs wherever possible; distinguish conjectures, conditional statements, working drafts, and independently reviewed results; record exact proof dependencies and source versions; and preserve corrections rather than silently replacing earlier text.
One of our long-term goals is to formalize and verify the mathematical content of the website using Lean and related proof assistants. Formalization is also a mathematical undertaking in its own right: it forces definitions to be clarified, implicit hypotheses to be exposed, and previously scattered results to become part of a coherent, reusable structure.
We want to build not merely a body of mathematics that can be read, but one that can be checked, corrected, and extended without losing its history.
03 / OPENNESS AND DISCOVERY
Mathematics should not be a closed competition
One of the most beautiful aspects of mathematics is the freedom to exchange ideas and explore difficult questions together. A thought may begin with one person's observation and become a deep theory through conversation, criticism, and collaboration.
Yet academic incentives have, to some extent, changed this relationship. Researchers must consider not only whether an idea is true or meaningful, but also who will publish it first and who will receive recognition for the result. Unpublished questions, promising approaches, and crucial technical insights can consequently become resources that people feel compelled to keep private.
These tensions are not new. In the nineteenth century, Felix Klein and Sophus Lie moved from close collaboration to a painful rupture; the mathematical competition between Klein and Henri Poincaré over automorphic functions also illustrates the longstanding importance of priority and reputation. Competition may stimulate discovery, but it can also damage the trust on which collaboration depends.
In some contemporary fields, important new ideas first circulate within particular seminars, collaborations, or small research networks. Researchers who cannot participate in these circles may struggle to learn of developments in time. People of comparable ability, creativity, and commitment can therefore begin from very different informational positions, and competitive evaluation may amplify those differences.
Access to an important mathematical question should not depend primarily on membership in a particular academic circle or privileged access to unpublished information. Good ideas deserve to be understood, discussed, and developed by more people.
Open mathematics in the age of AI
AI can help individuals enter unfamiliar fields, navigate difficult theories, follow large bodies of research, and attempt problems that would previously have required years of preparation. It is weakening some of the information barriers that have shaped mathematical research.
At the same time, AI creates new anxieties. When a system can rapidly elaborate an idea, explore possible proofs, or turn a line of thought into a manuscript, researchers may become even more reluctant to disclose unfinished work. The fear of being overtaken is understandable, especially under current systems of credit and publication.
But we should not allow that fear to determine the basic rules of mathematical exchange. We should not confuse the competitive goals encouraged by academic evaluation with the reasons we began studying mathematics.
Why did we begin? At least for me, the answer was not to publish before someone else or to claim exclusive possession of an important problem. We study mathematics to understand profound and beautiful structures, discover patterns that were previously unknown, and approach truth.
What matters is what we discover, what we understand, and whether those discoveries allow others to go further. If open discussion can improve an idea, if wider participation can advance a longstanding conjecture, or if a new theory can mature through shared effort, the result is valuable in itself.
Contributions also take many forms. A deep conjecture, a new strategy, the identification of a decisive obstruction, a bridge between theories, or a rigorous reconstruction of foundational proofs may all have enduring value. Such contributions deserve to be recorded, examined, and properly acknowledged.
Mathematical knowledge should be a common good that we can build, test, and inherit together.
04 / FROM PRINCIPLE TO PRACTICE
Making mathematical research public while it is happening
This project is an attempt to put those principles into practice. We want to make not only completed results available, but also worthwhile conjectures, unresolved difficulties, partial proofs, and the long-term programmes that may lead toward a solution. A research project should not have to wait for its final theorem before it can be useful to others.
Every carefully reviewed partial result can become a starting point for further work. A clearly identified obstruction may spare another researcher an unproductive attempt. A promising new method may be improved and generalized by people who did not take part in its original discovery.
We organize this work along two distinct tracks:
Integrating established theory
Stacks-style chapters of definitions, theorems, complete proofs, stable references, and carefully audited dependencies.
Explore foundations ↗CONJECTURES & NEW THEORYBuilding mathematics not yet complete
Motivations, precise conjectures, partial results, obstacles, competing approaches, and long-term research programmes.
Explore research ↗These tracks are related, but they have different purposes and standards of organization. Foundational integration follows the architecture of a theorem-and-proof library; original research follows the development of a question and the theory growing around it. Links between them should reveal real mathematical dependencies without erasing this distinction.
We welcome everyone interested in these questions to read, discuss, and contribute. We use GitHub Issues for public questions, corrections, references, counterexamples, candidate proofs, new approaches, and Lean formalizations. Contributions should remain attached to the precise statement and version to which they refer.
We also want a transparent record of authorship and contribution. Openness should not mean neglecting individual credit. On the contrary, a public, traceable history may allow more valuable mathematical work to receive the recognition it deserves.
We hope this website will grow into a genuine mathematical research community: a place where a conjecture may remain open for years, a developing theory may be refined repeatedly, and every meaningful advance can become part of a larger shared undertaking.
The purpose of mathematics is discovery and understanding, and both should be shared as widely as possible.
05 / READING THE ATLAS
What does each status mean?
A mathematical entry's kind and its current evidential status are separate. Public visibility never upgrades an open conjecture to a theorem.
- Conjecture / Question
- An open problem or proposed mathematical statement, with its exact conditions.
- Draft
- A working argument or record awaiting development or review.
- Conditional
- A conclusion that depends on additional identified hypotheses or inputs.
- Proof available
- A complete proof text is supplied for inspection; this does not imply independent certification.
- Reviewed
- A review is recorded against the exact source version and declared dependencies.
- Withdrawn
- A previously published entry remains in the history after withdrawal.
06 / VERSIONS & CITATIONS
A stable place to return to
Every research entry has a permanent identifier and address. The ordinary URL opens its current published form; a version-specific citation opens the historical text, including the statement and proof available at that stage.
When citing an exact argument, include its title, permanent identifier, version number and version URL. Publication records retain dates and change summaries.
Explore the revision history07 / FEEDBACK
Open discussion on GitHub
Questions, corrections, proof ideas, and LaTeX or Lean source ZIP files can be shared through public GitHub Issues. Anyone with a GitHub account may comment. Only the repository owner and approved collaborators can modify published site content.
GitHub Issues are public from the moment of submission, including all attached files. Do not upload confidential materials.
Read and share feedback ↗