Abstract. While formulating M-theory as such remains a wide open conceptual problem, crucial limiting aspects are well-defined and have evaded understanding for decades only due to their computational complexity. One of these is the identification of consistent higher curvature corrections to 11D supergravity, which are thought to reflect M-theoretic small-scale UV-corrections to its low-energy behaviour.
Pivotal among these corrections is the appearance of the -curvature polynomial as an electric source term for the higher Maxwell–Chern–Simons field (the C-field). That this term is singled out and part of a consistent higher curvature corrected supergravity theory has long been argued but has been established only partially, and only to first order in the deformation coefficient of .
This project aims to finally settle these computations, now that AI agents provide the required mathematical capacity. This program of computational M-theory is fairly open-ended and we invite and welcome further contributions by the community. So far the project has obtained two results, subject to further scrutiny; the first is as expected while improving considerably over existing results; the second is surprising and, if it holds water, may point to aspects of M-theory not anticipated before:
(1.) The curvature corrected theory exists to 2nd order in the deformation parameter and is unique up to field redefinitions. Consistency alone fixes the coefficient to the value previously predicted via string theory and its conjectured dualities.
(2.) But it does not exist beyond 2nd order, at least not via local classical corrections. We are now checking if local quantum corrections may lift the obstruction.
These results have been audited by independent machine agents but not reviewed yet by human experts nor formalized in a proof verifier. The project is ongoing.
The project files are currently kept in a Dropbox folder:
Overview: I8/1-papers/I8-project-overview.pdf
Order-2 existence: I8/1-papers/I8-order-two-paper.pdf
Order-3 obstruction: I8/1-papers/I8-order-three-paper.pdf
This is part of nLab materials.
How much of M-Theory can we work out now? Now that we have superhuman mathematical capabilities at our disposal, that is (cf. Avigad & Ballard 2026).
The remarkable answer is: A fair bit! And we should use the opportunity.
While the actual nature of M-theory (cf. Duff 1998, Duff 1999) remains notoriously mysterious (cf. Duff 1996 §6, Witten 2020)1: it is neither a quantum field nor a string theory nor entirely a “matrix model”), some of its aspects are sharply defined: In the regime where experimental resolution remains coarse compared to the Planck length (which will remain the only practically relevant case until accelerator technology has grown to galactic scales), M-theory is expected to be described locally by 11-dimensional supergravity (11D SuGra), corrected by a series in powers of of higher curvature corrections (cf. Cederwall et al. 2000 §1, Cederwall et al. 2001 §1, Cederwall et al. 2004 §1, Ozkan, Pang & Sezgin 2024 p. 13).
Here a higher curvature correction is a higher-order interaction in a theory of gravity. It determines how, at tiny scales comparable to , gravitons interact not just according to the rules of standard Einstein gravity but with further, more fine-grained behaviour. This is the hallmark of quantum gravity, which has been the holy grail of fundamental physics since the 1970s – both in its promise and fascination and in how it has remained tantalizingly elusive.
Indeed, while there is a well-defined mathematical prescription (a rare situation in high energy physics!) for how to determine the higher curvature corrections of 11D SuGra, and while supergravity researchers have worked on the problem for decades, even the first nontrivial order has been understood only partially. The reason is the dramatic complexity of the mathematical terms involved (Peeters, Vanhove & Westerberg 2000 p. 4: “horrendously complicated”).
(This is a general aspect of 11D supergravity: while its basic principles are extremely elegant and simple, the mathematical detail emerging from these principles is extremely rich and intricate.)
But this is exactly the situation where current frontier AI models excel: the principles are settled, but working out their consequences requires immense effort. Given enough compute, such problems now yield.
And yield it did. Within a few weeks, AI agents working with exact computer arithmetic, and auditing each other, settled the first two orders: the corrected theory exists and is unique through second order, and consistency alone fixes the coefficient to the value predicted via string theory. This already goes well beyond what decades of supergravity research had achieved. Moreover, the size and breadth of the resulting proofs indicate that they are beyond what could practically be done “by hand” (see below).
The third order brought a surprise: there is no consistent local classical continuation. Whatever M-theory is, beyond this order its low-energy behaviour cannot be captured by local classical curvature corrections to 11D supergravity alone. Something else must take over – quantum effects, global (flux-quantization) structure, or new degrees of freedom – exactly where M-theory is supposed to be more than supergravity.
So the question has changed: no longer whether the corrections can be computed, but what M-theory does where classical supergravity stops. With these computations now within reach, that question can be attacked concretely, and many others with it. We invite the community to join.
We give two overviews of the problem addressed here, the first expository, assuming just an intutive grasp of differential forms, the second with full technical detail.
This subsection gives a glimpse of what the project is about, for readers with a basic working knowledge of differential forms (cf. Frankel 2012 §2.1). It does not attempt technical precision; that is the job of the technical section below.
The Einstein equivalence principle says that the infinitesimal neighbourhood of any point in spacetime looks like Minkowski spacetime , with spatial coordinates and time coordinate . The Minkowski metric, which measures the proper relativistic length of tangent vectors, is
Since we are concerned here with the small-scale (UV) behaviour of gravity, it is sufficient to consider a contractible patch of spacetime. On it, the equivalence principle is made precise by an identification of every tangent space with Minkowski space, smoothly varying from point to point:
This field of one-forms is the coframe field, or vielbein. It is the gravitational field: the metric of spacetime is .
To compare nearby tangent spaces we need a connection, a matrix of one-forms with values in the Lorentz Lie algebra, . From the coframe and the connection one forms two field strengths:
In Einstein’s gravity the torsion is required to vanish, . This condition determines uniquely from (it is then the Levi-Civita connection). The curvature two-form is the field strength of gravity: it measures how much the parallel transport of tangent vectors around a small loop fails to return them to themselves, which is how tidal forces appear. Einstein's equation in vacuum says that the trace of the curvature components vanishes.
The electric and magnetic fields jointly form one flux density, a two-form:
In vacuum, Maxwell's equations say that both and its Hodge dual are closed:
It is useful to read this as two Bianchi identities, one for and one for its dual , together with a constitutive law that relates the two. This is the same structure as in the physics of materials, where Maxwell’s equations in a medium hold for the pair of fields and , while the constitutive law, for instance , encodes the properties of the medium. In the presence of electric charges the second equation acquires a source, ; magnetic charges would source the first.
The point to take away is this: the equations of motion of a field theory can be written as Bianchi identities for flux densities, together with a constitutive law. We will see that the same is true for 11D supergravity, and that its corrections can be understood as corrections to sources and to constitutive laws.
Supersymmetry extends spacetime by odd, or fermionic, directions. A superpoint
is a “space” with no ordinary points beyond one, but with coordinate functions that anticommute, . In particular this means that , hence that these odd functions witness an infinitesimal extension of the superpoint.
Now consider each point of Minkowski spacetime to be such a superpoint. The result is Minkowski super-spacetime:
where the odd coordinates transform as a realepresentation#RealIrreducibleSpinRepresentationInLorentzSignature) spinor representation of the Lorentz group. In eleven dimensions there is a minimal choice that leads to a consistent theory of gravity, the Majorana spinor with real components.
On a super-spacetime the equivalence principle says that the infinitesimal neighbourhood of each point looks like super-Minkowski spacetime. The coframe accordingly has an even part and an odd part :
The odd coframe is the gravitino field, the fermionic super-partner of the graviton. Supersymmetry enters through the torsion: on super-Minkowski spacetime one has , so the torsion constraint of plain 11D supergravity reads
where are the Dirac matrices. Since the are odd one-forms, products are symmetric in . In eleven dimensions the symmetric square of the branches as
and correspondingly the gravitino combines into exactly three kinds of bilinear forms:
These three bilinears govern all of what follows (cf. M2-brane, M5-brane).
Besides the coframe and the gravitino , 11D supergravity has one further field, the C-field. It is a higher analogue of the electromagnetic field: its flux density is a four-form instead of a two-form, and its Hodge dual is a seven-form (as ). On super-spacetime both flux densities have, besides their ordinary components, a fixed component built from the gravitino bilinears (9) above:
The remarkable fact of 11D superspace supergravity (suggested by Cremmer & Ferrara 80, Brink & Howe 1980, CDF 1991 §III.8.5; established in the following guise in GSS 2024 Thm. 3.1, cf. 2026 Thm. 3.16), is that the entire 11D theory is captured by Bianchi identities on these super-flux densities (10), just as for Maxwell’s equations in (5):
In particular, the constitutive law on ordinary spacetime is implied!
Compared to Maxwell’s equations (5) there is one key new feature. The C-field sources its own dual: the term is an electric charge density made from the field itself. This is the higher Chern–Simons self-interaction of the C-field.
Eleven-dimensional supergravity is expected to be only the leading approximation to the low-energy behaviour of M-theory, valid when all curvatures are small compared to the Planck scale. The first correction is expected to appear as a further electric source term:
Here is a specific invariant polynomial of degree four in the curvature two-form (3), namely
in terms of the Pontryagin forms and ; and is a constant proportional to the sixth power of the Planck length,
In words: the curvature of spacetime itself carries membrane charge.
This term is independently suggested by the cancellation of quantum anomalies on M5-branes and by string theory, and it is expected to come with a whole tower of further corrections in higher powers of , starting with the famous -correction to Einstein's equations.
Adding the source in (12) does not by itself give a consistent theory. The Bianchi identities on super-spacetime are very rigid, and to accommodate the new source the constitutive laws must be corrected as well: the duality receives corrections in powers of , and so do the torsion constraint and the gravitino components of the fluxes. Just as in electromagnetism, where a nonlinear medium has a nonlinear constitutive law, the corrected constitutive laws encode the corrected dynamics.
The question of the project is therefore: can the super-components of the torsion and flux fields be corrected, order by order in , so that the corrected Bianchi identities are still equivalent to a consistent set of corrected equations of motion? The answer found (below) is yes through second order in , where the corrections are essentially unique, and no at third order, as long as one asks for local, classical corrections. (We are working on whether allowing a 2-loop quantum correction may fix this.)
See section 2.2 of: I8/1-papers/I8-project-overview.pdf
See section 3 of: I8/1-papers/I8-project-overview.pdf
See section 4 of: I8/1-papers/I8-project-overview.pdf
Jeremy Avigad, Matthew Ballard: AI for research mathematics has arrived, Science 394 6820 (2026) 166-167 [doi:10.1126/science.ael6628]
Lars Brink, Paul Howe: Eleven-Dimensional Supergravity on the Mass-Shell in Superspace, Phys. Lett. B 91 (1980) 384 [doi:10.1016/0370-2693(80)91002-3]
Martin Cederwall, Ulf Gran, Mikkel Nielsen, Bengt E. W. Nilsson: Manifestly supersymmetric M-theory, J. High Energy Phys. 0010 (2000) 041 [arXiv:hep-th/0007035, doi:10.1088/1126-6708/2000/10/041]
Martin Cederwall, Ulf Gran, Mikkel Nielsen, Bengt E. W. Nilsson: Generalised 11-dimensional supergravity, in: Quantization, Gauge Theory and Strings: Proceedings of the International Conference Dedicated to the Memory of Professor E. S. Fradkin, Moscow (2001) [arXiv:hep-th/0010042]
Martin Cederwall, Ulf Gran, Bengt E. W. Nilsson, Dimitrios Tsimpis: Supersymmetric corrections to eleven-dimensional supergravity, J. High Energy Phys. 0505 (2005) 052 [arXiv:hep-th/0409107, doi:10.1088/1126-6708/2005/05/052]
Eugène Cremmer, Sergio Ferrara: Formulation of Eleven-Dimensional Supergravity in Superspace, Phys. Lett. B 91 (1980) 61 [doi:10.1016/0370-2693(80)90662-0]
Michael Duff: M-Theory (the Theory Formerly Known as Strings), Int. J. Mod. Phys. A 11 32 (1996) 5623-5642 [arXiv:hep-th/9608117, doi:10.1142/S0217751X96002583]
Michael Duff: A Layman’s Guide to M-Theory, in: V. Alexiades, G. Siopsis (eds.): Trends in Mathematical Physics, AMS/IP Studies in Advanced Mathematics 13, American Mathematical Society (1999) 184-213 [arXiv:hep-th/9805177, doi:10.1090/amsip/013]
Michael Duff (ed.): The World in Eleven Dimensions: Supergravity, Supermembranes and M-Theory, Institute of Physics Publishing, Bristol (1999) [ISBN:978-0-7503-0672-0, doi:10.1201/9781482268737]
Theodore Frankel: The Geometry of Physics – An Introduction, Cambridge University Press (3rd ed. 2012) [doi:10.1017/CBO9781139061377]
Grigorios Giotopoulos, Hisham Sati, Urs Schreiber: Flux Quantization on 11D Superspace, J. High Energy Phys. 2024 82 (2024) [arXiv:2403.16456, doi:10.1007/JHEP07(2024)082]
Grigorios Giotopoulos, Hisham Sati, Urs Schreiber: Higher Superspace Supergravity and its IR-Completions (2026) [arXiv:2607.20312]
Mehmet Ozkan, Yi Pang, Ergin Sezgin: Higher derivative supergravities in diverse dimensions, Phys. Rep. 1086 (2024) 1-95 [arXiv:2401.08945, doi:10.1016/j.physrep.2024.07.002]
Kasper Peeters, Pierre Vanhove, Anders Westerberg: Supersymmetric higher-derivative actions in ten and eleven dimensions, the associated superalgebras and their formulation in superspace, Class. Quant. Grav. 18 5 (2001) 843-890 [arXiv:hep-th/0010167, doi:10.1088/0264-9381/18/5/307]
Edward Witten: Adventures in Physics and Maths, Bhāvanā 4 4 (2020) [bhavana.org.in/adventures-in-physics-and-maths]
Currently, only the abstract, prologue and exposition are human-generated (by Urs Schreiber), the writeups on order-2 existence and order-3 obstruction as well as all the supporting evidence are curated contributions by GTP-6 Astra Codex agents (proofs & computation) and Opus 5.5 (strategy & organization). Together with exploration of some dead-ends, this was obtained in the course of the first two weeks of November 2026, with parallel agents running almost 24/7, working locally on a Dell Latitude 7440 notebook computer. Thanks to Nikita Golub for sharing further GPT usage.
The idea is to eventually absorb more of this raw/slop material, distill it, beautify it, and then build on it.
Witten 2020: “I have often been asked why I used the name M-theory to describe the richer theory that has the traditional string theories as limiting cases. M-theory was meant as a temporary name pending a better understanding. Some colleagues thought that the theory should be understood as a membrane theory. Though I was skeptical, I decided to keep the letter “m” from “membrane” and call the theory M-theory, with time to tell whether the M stands for magic, mystery, or membrane. Later, the membranes were interpreted in terms of matrices. Purely by chance, the word “matrix” also starts with “m”, so for a while I would say that the M stands for magic, mystery, or matrix.” ↩
Last revised on November 18, 2026 at 08:57:27. See the history of this page for a list of all contributions to it.