Evaporation–Crystallization Generative Computing: A Physics-Guided Framework for Three-Dimensional Structure Generation Through Solvent Removal, Nucleation, and Competitive Crystal Growth

Three-dimensional generative modeling has enabled the synthesis of complex geometric structures through neural implicit representations, point-cloud generation, volumetric diffusion, and explicit scene representations. However, many existing approaches generate geometry primarily through direct prediction or iterative transformation of a target representation. Natural mineral structures suggest an alternative computational perspective: three-dimensional geometry can emerge from the removal of a surrounding medium, the redistribution of dissolved material, the formation of crystal nuclei, and the competition of growing solids for space and available resources. This paper proposes Evaporation–Crystallization Generative Computing (ECGM), a physics-guided framework for generating three-dimensional structures through coupled solvent removal, solute transport, supersaturation-dependent nucleation, anisotropic crystal growth, and geometric competition. The proposed system maintains explicit fields for solvent, dissolved solute, supersaturation, crystalline material, and environmental conditions. A stochastic nucleation mechanism creates variable numbers of crystal seeds, while local growth rules determine their morphology and interaction. The final solid geometry is extracted from the evolved material state and can be represented as a mesh, voxel volume, point cloud, or collection of surface primitives. A differentiable or surrogate-assisted extension allows a learned controller to select environmental parameters and growth conditions according to a target geometry or structural objective. The paper formalizes the framework, presents governing equations and a computational algorithm, defines training and evaluation objectives, and proposes experiments against established generative and physics-based baselines. The framework is a conceptual proposal; its performance, generalization, and advantages over existing methods remain to be established experimentally.