Geometry from Chertalo
Chertalo owns the model, revision and geometry authority. The physics domain consumes a GeometryPack with units, coordinate frames and provenance bound to the source revision.
Smetalo is the Chertalo computational-physics and engineering-compute domain. The REV13 architecture separates the physical model from the numerical method, solver and hardware: from geometry, materials and boundary conditions to HPC execution and traceable results.
REV13 architecture · updated October 2026
Chertalo owns the model, revision and geometry authority. The physics domain consumes a GeometryPack with units, coordinate frames and provenance bound to the source revision.
Materials, loads, boundary conditions and quantities of interest define the problem. Compute Model IR is the target solver-independent contract; method, discretization and execution have separate responsibilities.
A field or a completed process alone does not establish accuracy. Numerical verification and acceptance for a specific engineering use are separate decisions, bound to the inputs and revision.
REV13 defines a solver-independent, method-independent, HPC-native architecture. FEM is an important method; the scope expands through qualified adapters with explicit capabilities.
A study describes fields, equations, materials, loads, conditions and objectives. Form-CAS is the target engineering-formulation layer; SageMath provides a separate capability for symbolic and exact mathematical checks.
FEM, FVM, DG and methods such as CutFEM, BEM and IGA are architectural directions. Selection depends on physics, geometry, accuracy and the qualified capabilities of the backend. This list does not claim implemented support for every method.
A solver is an interchangeable backend reached through an adapter, not the place the physics lives. Each adapter declares which methods, element types, physics and limits it actually supports, and is qualified separately. We do not bind to a single solver, and the physical model is not reduced to its file format.
Smetalo declares resource requirements; Compute Fabric owns placement and execution. CPU, GPU, MPI and Slurm are target capabilities whose availability must be established for the selected solver and workflow.
Target architectural chain. Supported transformations require validation of mathematical meaning and data provenance.
An engineering study binds the CAD revision to the physical problem, execution plan and results. Smetalo owns the physics semantics; Chertalo retains geometry, topology and product-revision authority.
Execution has returned the required results and artifacts for precisely identified inputs.
Applicable checks assess balances, residuals, errors and agreement with a reference solution within the declared validity domain.
A separate engineering policy decides whether the evidence meets the requirements of the intended use. AI, the UI and a successful solver exit do not grant acceptance.
Existing geometry and solver adapters provide implementation foundations. The shared Compute Model IR, qualified method/backend selection and normalized ResultSet remain implementation and validation work. Architecture does not establish readiness for every workflow.
The first vertical connects geometry, material, loads and conditions to a qualified solver backend for structural or thermal studies. Which backend, the scope and the acceptance criteria are agreed for each pilot.
Study-to-Compute Model IR and backend lowering, execution, artifact provenance, convergence and result verification for the intended use.
Additional physics and numerical methods, further solver adapters, GPU and distributed execution, UQ, sensitivities and optimization. Each extension requires its own qualification.
Earlier NEMO articles describe a previous research line. Their counts and measurements do not represent the current Smetalo product. AI may assist setup and interpretation; numerical results come from the explicitly selected computational method and solver.
Discuss your geometry, physics, quantities of interest and pilot acceptance criteria.
Discuss a pilot workflow