Extending the Dirac physics of two-dimensional (2D) graphene into three dimensions (3D) carbon allotropes with higher-dimensional band degeneracies remains a central challenge in topological materials science. Here, we propose a general symmetry-engineering principle that systematically transforms graphene's Dirac cone into a 3D nodal surface via controlled layering and registry shift, and employ this principle to guide a machine-learning-accelerated inverse design. By integrating a crystal diffusion variational autoencoder(CDVAE) with a Crystal Transformer, we discover a novel, dynamically and mechanically stable carbon allotrope named \textbf{Netsene} (bct-C24), which crystallizes in the body-centered tetragonal \textit{I4/mcm} space group. First-principles calculations confirm that Netsene is a unique nested nodal-surface semimetal: it hosts a complex, double-bowl-shaped nodal-surface system around the Fermi level, protected by non-symmorphic symmetries, alongside Dirac-like linear crossings with Fermi velocities comparable to graphene (∼9×105~m/s). Its non-trivial bulk topology manifests in drumhead surface states, including a nearly flat band. Netsene provides a structurally robust, bulk platform that unifies ultrahigh carrier mobility, topological nodal surfaces, and potential correlation physics, demonstrating the power of theory-guided, machine-learning-accelerated discovery for engineering topological quantum phases.
@article{arxiv.2605.22141,
title = {Theory-Guided, Machine-Learning-Accelerated Discovery of a 3D Carbon Nested Nodal-Surface Semimetal},
author = {Shuaihua Zhang and Silei Guo and Jingxiang Liu and Baoxin Hu and Yanling Wu and Jun Li},
journal= {arXiv preprint arXiv:2605.22141},
year = {2026}
}