NEWS.md

In tulpaMesh: Constrained Delaunay Triangulation Meshes for Spatial 'SPDE' Models

tulpaMesh 0.1.4

• tulpa_mesh() no longer returns vertices that belong to no triangle (#4). Lattice points from max_edge were laid over the bounding box of the domain, so points in its corners fell outside the constraint loop and received zero mass in fem_matrices(). The lattice is now generated only inside the domain, and the triangulation drops any vertex that no triangle references (for example input points outside a user boundary).
• extend now reaches the documented distance: the extended boundary is the outward buffer of the convex hull of coords by extend times the domain diameter, with rounded corners. Previously the hull was scaled about its centroid, which reached roughly half that distance.
• The outer value of max_edge = c(inner, outer) now sets the lattice spacing and boundary segment length in the extension region, as documented; it was previously ignored.
• tulpa_mesh(extend = 0) without a boundary now triangulates the convex hull of the points instead of failing with zero triangles.
• n_input_points now counts the supplied coordinates.

tulpaMesh 0.1.3

• Parallel FEM assembly (fem_matrices(parallel = TRUE)) no longer references Intel TBB symbols directly. Each worker now accumulates triplets into a chunk-local buffer that is merged under a mutex, so the package builds on platforms where RcppParallel falls back to its TinyThread backend (for example Alpine Linux with musl).
• Fixed tulpa_mesh(max_edge = ...) collapsing to zero triangles for some (max_edge, cutoff) settings and point counts. Vertex deduplication could merge extended-hull boundary vertices that fell within max_edge * 0.3 of each other, deleting a constraint edge and leaving the boundary loop open; the constrained triangulation then erased every triangle. Boundary and hole vertices are now protected from deduplication so the constraint loop stays closed.
• tulpa_mesh() now errors loudly when the triangulation yields zero triangles (for example a collinear point set) instead of returning an empty mesh that silently produces all-zero FEM matrices.

tulpaMesh 0.1.1

• Removed single quotes from person names and algorithm names in DESCRIPTION Title and Description fields per CRAN policy.
• Added Artem Amirkhanov (CDT library) and William C. Lenthe (predicates.h) to Authors@R with ctb and cph roles.
• Updated inst/COPYRIGHTS to separately document predicates.h (BSD-3-Clause).

tulpaMesh 0.1.0

Initial CRAN release.

Mesh Generation

• tulpa_mesh(): constrained Delaunay triangulation with formula interface, sf boundary support (POLYGON, MULTIPOLYGON, holes), hexagonal lattice refinement, Ruppert refinement (min_angle, max_area), and vertex deduplication.
• tulpa_mesh_sphere(): icosahedral geodesic meshes with 3D surface FEM.
• tulpa_mesh_1d(): temporal meshes with tridiagonal FEM matrices.
• tulpa_mesh_graph(): metric graph meshes for network SPDE models.

FEM Assembly

• fem_matrices(): P1 linear FEM producing mass (C), stiffness (G), and projection (A) sparse matrices. Supports lumped mass, barrier models, and parallel assembly via RcppParallel.
• fem_matrices_p2(): P2 quadratic FEM with 6-node triangular elements.
• fem_matrices_nonstationary(): spatially varying kappa/tau weighting.

Mesh Operations

• mesh_quality(), mesh_summary(), plot.tulpa_mesh(): diagnostics.
• subdivide_mesh(), subset_mesh(), mesh_components(): mesh manipulation.
• refine_mesh(): adaptive refinement from error indicators.
• as_tulpa_mesh(): convert fmesher/INLA meshes.
• barrier_triangles(): identify barrier regions from polygons.
• mesh_crs(), set_crs(): CRS support.

tulpaMesh documentation built on Nov. 2, 2026, 9:06 a.m.