Minimal surfaces, Knots, and Neural Networks

A team of mathematicians has used a machine-learning framework to test a conjecture linking knot theory to the geometry of four-dimensional hyperbolic space, reporting that computational results align with the proposed relationship. The conjecture, formulated by Joel Fine, proposes a connection between the coefficients of the HOMFLY polynomial of a knot in the 3-sphere and a signed count of minimal surfaces in hyperbolic 4-space that meet the sphere at infinity at that knot [1][2]. The HOMFLY polynomial is a knot invariant that distinguishes between different types of knots. The conjecture specifies that these minimal surfaces should have a prescribed genus and self-intersection number [2]. Hyperbolic 4-space is a non-Euclidean geometry where space is curved, distinct from the flat, three-dimensional space of everyday experience [3]. To investigate Fine's conjecture, the researchers developed a novel computational approach using Physics-Informed Neural Networks, or PINNs, to solve the minimal surface equation in this curved hyperbolic space [2]. The framework constructs near-minimal surfaces that bound various families of knots in the 3-sphere [2]. The team also created an algorithmic method to detect self-intersections within these surfaces and compute their mathematical sign [2]. For every knot analyzed, the computationally discovered minimal surfaces and their self-intersection numbers perfectly aligned with the predictions of Fine's Conjecture, providing empirical evidence for it [2]. The work represents an intersection of pure mathematics and computational science, applying neural networks to a problem in differential geometry. The study was posted on the arXiv preprint server on May 25, 2026 [1]. The research was conducted under the auspices of arXivLabs, a framework for experimental projects on the arXiv platform [1].