DRR-3D: Dissolution–Redistribution–Reprecipitation for Process-Based Three-Dimensional Generation

Three-dimensional (3D) generative methods have achieved substantial progress in representing complex geometry through neural implicit fields, point-cloud generation, diffusion models, and explicit radiance or Gaussian representations. However, generating structures formed by the selective removal, transport, and subsequent redeposition of material presents a distinct modeling challenge. Such processes occur in karst landscapes, mineral deposits, dissolution cavities, porous rocks, and other natural and synthetic materials. This paper proposes Dissolution–Redistribution–Reprecipitation 3D (DRR-3D), a process-based generative framework in which 3D geometry emerges through coupled material removal, transport, and deposition rather than being generated exclusively as a final geometric configuration. The framework maintains a spatial representation of solid material, a mobile material field, local reaction conditions, and a transport structure. Dissolution transfers material from the solid phase into the mobile phase, transport redistributes mobile material, and reprecipitation converts transported material back into solid geometry. A conservation-aware update mechanism links these processes and supports changes in surface geometry and topology. The framework can be implemented using voxel grids, level sets, point-based representations, or hybrid neural representations. Optional learned components can estimate reaction rates, transport behavior, and deposition patterns from reference data. This paper formalizes the proposed framework, defines its governing variables and objectives, and develops an evaluation protocol for comparison with direct 3D generative models and established process-based simulators. DRR-3D is presented as a research hypothesis requiring implementation and empirical validation, rather than as an experimentally established improvement.