Korzhinskii-Net: Physics-Informed Neural Network for Sub-Surface Mineral Prospectivity Modelling

A physics-informed neural network that simulates subsurface fluid flow and heat transport has outperformed classical machine-learning models in mineral prospectivity mapping, according to a preprint posted to arXiv on 31 May 2026 [1]. The model, called Korzhinskii-Net, achieved a mean precision-recall area under the curve of 0.885 across five ore provinces, compared with 0.281 for the strongest classical baseline [1]. Korzhinskii-Net is a 2-D radial physics-informed neural network, or PINN, that couples Darcy flow, advective-diffusive heat transport, and a softplus-saturated reaction rate into a single differentiable forward model [1]. It is weakly supervised by surface and remote-sensing proxies [1]. The network is named after Dmitri S. Korzhinskii (1899–1985), whose theory of infiltration metasomatism provides the physical scaffold for the approach [1]. The authors evaluated the model on five ore provinces spanning four commodity classes: Norilsk (Ni-Cu-PGE), Pechenga (Ni-Cu sulphide), Udokan (sandstone-hosted Cu), Sukhoi Log (orogenic Au), and Mirny (kimberlitic diamond) [1]. The evaluation used a leakage-controlled 5-fold cross-validation protocol with hard ring-shaped negatives [1]. Korzhinskii-Net attained a mean fractional rank of 0.019, versus 0.413 for gradient boosting, the strongest classical baseline tested [1]. The improvement was consistent across all five provinces and all four commodity systems [1]. The work addresses a known limitation in operational mineral prospectivity pipelines, which the authors describe as “blind to the subsurface physics that actually localises ore” [1]. By embedding physical constraints directly into the learning architecture, Korzhinskii-Net recovers localisation patterns that pure feature-based learners systematically miss, even when constrained only by global open-data proxies [1]. The full pipeline and evaluation harness have been released as open source [1]. The preprint appeared on arXiv, an open-access repository of electronic preprints that, as of November 2024, receives about 24,000 submissions per month [6]. arXiv hosts papers across mathematics, physics, computer science, and related fields, and its contents are moderated but not peer-reviewed [6]. The repository passed the two-million-article milestone at the end of 2021 [6]. The Korzhinskii-Net paper was posted under the physics.geo-ph subject classification [1].