1 AUTOMATIC MESH GENERATION 1

1 Automatic Mesh Generation

1.1 Mesh Definition

Mesh M is a discrete representation of geometric model in terms of its

geometry G, topology T , and associated attributes A.

M = {G, T, A}
• geometry - nodal coordinates
• topology - element types, adjacency relationships
• attributes - color, loading, boundary conditions, ...

Typeset in LATEX by Daniel Rypl

1 AUTOMATIC MESH GENERATION 2

1.2 Requirements on Mesh Generation

• generality (broad range of geometries and topologies)
• automation (minimum user intervention)
• validity (valid mesh)
• accuracy (accurate resolution)
• convergence (guaranteed convergence)
• quality (guaranteed quality)
• invariance (to model rigid body motions)
• flexible mesh density control (uniform, graded meshes)
• robustness (reliability)
• compactness (storage requirements)
• efficiency (computational speed)
• linear computational complexity
All these requirements can be hardly fulfilled at the same time and it is
necessary to compromise.
Typeset in LATEX by Daniel Rypl
1 AUTOMATIC MESH GENERATION 3

1.3 Mesh Validity

• topological compatibility
◦ mesh is topologically compatible with model entity E d
of dimension d
- if each mesh entity M d−1 classified to E d is shared exactly by
two mesh entities M d classified to E d
- if each mesh entity M d−1 classified to Em d−1
forming m times
boundary of model entity E d is shared exactly by m mesh entities
M d classified to E d
◦ mesh is topologically compatible if it is compatible with all model
entities
◦ topological incompatibilities
- topological redundancy
- topological holes

Typeset in LATEX by Daniel Rypl

1 AUTOMATIC MESH GENERATION 4

Topological compatibility (left), topological hole (middle), and topological re-

dundancy (right) on a curve mesh.

Topological compatibility (left), topological hole (middle), and topological re-

dundancy (right) on a surface mesh.

Typeset in LATEX by Daniel Rypl

1 AUTOMATIC MESH GENERATION 5

• geometrical similarity
◦ topologically compatible mesh is geometrically similar to model
entity E d (1 ≤ d ≤ 3) if for any two different mesh entities Mid
and Mjd classified to E d holds

Mid ∩ Mjd = ∅d if d = 3

Mid ∩∗ Mjd = ∅d if d = 1 or d = 2

Mesh violating (left) and satisfying (right) geometrical similarity.

Typeset in LATEX by Daniel Rypl

1 AUTOMATIC MESH GENERATION 6

1.4 Computational Complexity

• number of operations needed to generate N elements expressed in terms
of N and some appropriate constants
• corresponds to the time needed to generate N elements
• algorithm dependent
• average, worst

◦ O(1) - constant computational complexity (not achievable)

◦ O(N ) - linear computational complexity (ideal)
◦ O(N log(N )) - logarithmic computational complexity (acceptable)
◦ O(N 2 ) - quadratic computational complexity (unacceptable)
◦ ···

Typeset in LATEX by Daniel Rypl

1 AUTOMATIC MESH GENERATION 7

1.5 Mesh Quality

Elementary quality criteria

• element shape based criteria
◦ inscribed circle/sphere radius to circumscribed circle/sphere radius
ratio (simplices only)
◦ area/volume2 to perimeter2 /surface3 ratio
◦ min edge to max edge ratio (aspect ratio)
◦ dihedral angle criterion (min angle to max angle ratio)
◦ jacobian
• mesh topology based criteria
◦ nodal valence criterion
• mesh density based criteria
◦ deviation of real element spacing from desired one
Typeset in LATEX by Daniel Rypl
1 AUTOMATIC MESH GENERATION 8

Overall mesh quality

• arithmetic mean
• harmonic mean
• worst quality
• quality distribution function

1.6 Mesh Classification

• dimension - 1D mesh, 2D mesh, 3D mesh
• element type - triangular, tetrahedral, quadrilateral, hexahedral,
quad-dominant, mixed, ... mesh
• element aspect ratio - isotropic, anisotropic mesh
• mesh density - uniform, graded mesh
• topology - structured, unstructured mesh

Typeset in LATEX by Daniel Rypl

1 AUTOMATIC MESH GENERATION 9

1.7 Mesh Generation Method Classification

• manual and semi-automatic methods
◦ applicable to geometrically simple domains (usually 2D)
◦ enumerative methods (user supplied mesh entities)
◦ explicit methods (revolution, extrusion)
• mapping (parameterization) methods
◦ mapping from parameter space to the physical space
◦ explicit mapping - algebraic interpolation methods
◦ implicit mapping - PDE solution methods
• domain decomposition methods
◦ block decomposition methods (multiblock method)
◦ spatial decomposition methods (quadtree/octree method)

Typeset in LATEX by Daniel Rypl

2 STRUCTURED MESH GENERATION 10

• constructive methods
◦ applicable to arbitrary geometry and topology
◦ element creation - advancing front method
◦ point insertion - Delaunay method

2 Structured Mesh Generation

• implicit mesh topology
• applicable to topologically simple domains
• limited mesh density control
• typical for CFD

Methods
◦ algebraic methods
◦ PDE based methods
◦ multiblock methods
Typeset in LATEX by Daniel Rypl
3 UNSTRUCTURED MESH GENERATION 11

3 Unstructured Mesh Generation

• explicit mesh topology
• applicable to domains of arbitrary geometrical and topological complexity
• flexible mesh density control
• typical for structural analysis

3.1 Triangular and Tetrahedral Mesh Generation

3.1.1 Quadtree/Octree Based Methods

Algorithm
1. tree construction
2. mesh generation
3. mesh optimization

Typeset in LATEX by Daniel Rypl

3 UNSTRUCTURED MESH GENERATION 12

• quadtree/octree data structure

◦ hierarchic data structure
- vertical (top-down, bottom-up) traversal - O(N log(N ))
- horizontal (neighbour at the same level) traversal - O(1)
◦ root cell (quadrant/octant) - bounding box (square/cube)
◦ boundary refinement by recursive subdivision to equal cells
(4/8 quadrants/octants) up to sufficient resolution
- desired cell size, desired cell level
- geometry representation (curvature, boundary features)
◦ global refinement by recursive subdivision using mesh density
control (background mesh, grid, sources)
◦ one-level difference rule enforcement (tree balancing)
- each two cells sharing at least an edge are at the same or
subsequent levels of hierarchic tree data structure
Typeset in LATEX by Daniel Rypl
3 UNSTRUCTURED MESH GENERATION 13

parental octant
8 9 terminal octant
Level
2
15 16
12 root octant
13 14
7
0
10 11
1 1 2

5 6
2 3 4 5 6 7 8 9
1
3 10 11 12
3 4

4 13 14 15 16

Octree hierarchy.

Typeset in LATEX by Daniel Rypl

3 UNSTRUCTURED MESH GENERATION 14

One-level difference rule.

Typeset in LATEX by Daniel Rypl

3 UNSTRUCTURED MESH GENERATION 15

◦ terminal cell classification

- interior / boundary / exterior
- in/out test (boundary orientation, ray intersection, modeller queries)
- classification propagation
◦ interior refinement by recursive subdivision (to minimum level)
◦ evaluation of intersection of domain boundary with boundary cells

Filtering of intersection points.

Typeset in LATEX by Daniel Rypl

3 UNSTRUCTURED MESH GENERATION 16

◦ filtering of intersection points

- violation of topology not allowed
- modification of geometry permissible
(original geometry restored after mesh completion)
- reclassification of boundary cells
- association of cell corner with closest (boundary or intersection)
point (if close enough) - modification of cell shape
- merging two intersection points
- association of boundary point with closest intersection point
(if close enough) - modification of boundary geometry
◦ smoothing of corner nodes of interior cells
(repositioning of corners to barycenter of corners of all cells incident
to smoothed corner)

Typeset in LATEX by Daniel Rypl

3 UNSTRUCTURED MESH GENERATION 17

• mesh generation
◦ exterior cells - skipped
◦ interior cells - templates (predefined patterns of elements topologically
compatible with the cell)
- cell corner points - mesh nodes
- there is at maximum one midside point per octree edge
(consequence of one-level difference rule)
P4 4


- 2D: i=0 i = 24 = 16 templates (6 basic templates)
P12 12

- 3D: i=0 i = 212 = 4096 templates (78 basic templates)

Basic 2D templates.

Typeset in LATEX by Daniel Rypl

3 UNSTRUCTURED MESH GENERATION 18

◦ boundary cells - specific algorithm

- only interior parts of boundary cells subjected to triangulation
- discretization of relevant boundary of the cell
(must ensure compatibility with neighbouring cells)
- discretization of domain boundary within the cell
(must comply with model topology)
- discretization of interior of boundary cell (e.g. AFT)
• mesh optimization
◦ Laplacian smoothing of interior mesh nodes
- repositioning of nodes to barycenter of nodes of all elements
incident to smoothed node
- modifies mesh geometry
- preserves mesh topology
- iterative process (convergence after about 5 cycles)

Typeset in LATEX by Daniel Rypl

3 UNSTRUCTURED MESH GENERATION 19

Features (• - advantages, ◦ - disadvantages)

• very fast
• very robust
• reasonable quality meshes
(slivers avoided, guaranteed quality for interior cells)
• guaranteed convergence (for sufficiently simple boundary cells)
• validity guaranteed by proper use of templates
• favourable computational complexity O(N log(N )) −→ O(N )
• boundary discretization is part of output
• good cache usage (for appropriate cell ordering)

◦ less flexible mesh density control (less suitable for adaptive analysis)
- limited cell sizing (power of 2)
- limited mesh density gradation
(consequence of one-level difference rule)
Typeset in LATEX by Daniel Rypl
3 UNSTRUCTURED MESH GENERATION 20

◦ boundary layer of elements - worst quality

◦ not invariant with respect to rotations of the model
◦ cannot fully comply with given boundary triangulation
◦ hardly usable for anisotropic meshing

Alternatives and Extensions

• instead of templates using
◦ Delaunay based cell corners insertion
◦ element removal concept
• quadtree/octree data structure only used as control space
(mesh density control, spatial localization)
• generation of mixed meshes (there are no all-hexahedral templates)
• applicable to curved surface meshing (direct approach)
• extensible to generalized tree approach in parametric space

Typeset in LATEX by Daniel Rypl

3 UNSTRUCTURED MESH GENERATION 21

3.1.2 Advancing Front Based Methods

Algorithm
1. front setup
2. mesh generation
3. mesh optimization

• front data structure

◦ oriented interface between the meshed part and not yet meshed part
of the domain
◦ initially formed by all boundary facets (segments/faces)
◦ evolves during mesh generation until becomes empty
◦ may be multiple connected (is not allowed to overlap itself)
◦ front management (facet selection, insertion, removal)
Typeset in LATEX by Daniel Rypl
3 UNSTRUCTURED MESH GENERATION 22

• mesh generation (until the front is not empty)

1. facet selection: select facet f from the front
- geometrical criteria (length, area, angles)
- topological criteria (neighbour, minimize the front)
2. optimal point placement: find position of the optimal point
P opt to form together with the selected facet f a new
tentative element e
- 2D uniform mesh: equilateral triangle
- 3D uniform mesh: the most regular tetrahedron
- graded mesh: mesh density variation should be taken
into account
3. potential candidate selection: search for a point P in the mesh
to be used instead of P opt

Typeset in LATEX by Daniel Rypl

3 UNSTRUCTURED MESH GENERATION 23

- spatial search
- shape - circle (2D), sphere (3D) with center at P opt
- size - related to mesh density at P opt
- points in the neighbourhood ordered with respect to the
increasing distance from P opt

3
Popt Popt
2
1

f f

Optimal point placement. Potential candidate selection.

Typeset in LATEX by Daniel Rypl

3 UNSTRUCTURED MESH GENERATION 24

4. intersection check: check if the new element e is topologically valid;

if not, the point is rejected and the potential candidate selection is
repeated; if there is no more candidates, optimal point placement is
repeated with reduced element size
- spatial search
- shape - circle/rectangle (2D), sphere/parallelepiped (3D)
- size - large enough to be reliable, small enough to be efficient
- intersection check - tentative element against front
- enclosure check - tentative element against nodes
5. element forming: form the new element
6. front update: update the front
- remove existing facets used to form the new triangle from the front
- insert new facets used to form the new triangle to the front

Typeset in LATEX by Daniel Rypl

3 UNSTRUCTURED MESH GENERATION 25

Front propagation.

Typeset in LATEX by Daniel Rypl

3 UNSTRUCTURED MESH GENERATION 26

• mesh optimization
◦ Laplacian smoothing of interior mesh nodes
- repositioning of nodes to barycenter of nodes of all elements
incident to smoothed node
- modifies mesh geometry
- preserves mesh topology
- iterative process (convergence after about 5 cycles)
◦ topological transformations (to remove slivers)
- diagonal edge swapping, generalized face swapping, node merging
- modify mesh topology
- preserve mesh geometry

Typeset in LATEX by Daniel Rypl

3 UNSTRUCTURED MESH GENERATION 27

Computational Aspects
• front management
◦ hashing
• spatial search
◦ background grid
◦ background octree data structure
• intersection check
◦ bounding box intersection
◦ alternating digital tree

Typeset in LATEX by Daniel Rypl

3 UNSTRUCTURED MESH GENERATION 28

Features (• - advantages, ◦ - disadvantages)

• high quality graded meshes
• high quality boundary layer of elements
• flexible mesh density control
• validity guaranteed (if meshing completed)
• complies with given boundary discretization
• favourable computational complexity O(N log(N )) −→ O(N )
• theoretically invariant with respect to rigid body motions
• extensible to anisotropic meshing
• good cache usage (for appropriate front processing)
• appropriate for adaptive analysis (local remeshing)

◦ rather low speed

Typeset in LATEX by Daniel Rypl

3 UNSTRUCTURED MESH GENERATION 29

◦ convergence not guaranteed in 3D (requires node insertion to complete

mesh even in very simple 3D case - Schönhardt polyhedron)
◦ creation of slivers in mesh generation phase is not avoided
(slivers eliminated during mesh optimization)
◦ requires existence of boundary discretization

Alternatives and Extensions

• mesh density control by octree (lost of invariance) or background grid
• node acceptance driven by local Delaunay circle/sphere empty property
• extensible to curved surface meshing (direct approach)
• extensible to anisotropic meshing (using appropriate metric)
• extensible to boundary layer anisotropy (using offsetting)
• applicable to curve surface meshing (indirect approach)

Typeset in LATEX by Daniel Rypl

3 UNSTRUCTURED MESH GENERATION 30

3.1.3 Delaunay Based Methods

Delaunay Triangulation
Triangulation of convex hull S of points P i , i = 1, 2, 3, . . . , m in Rn , n ≥ 2
where for each simplex K holds Delaunay criterion

∀i : k SP i k ≥ ρ(K) S = {P ∈ Rn : k SP j k= ρ(K) for P j ∈ K}

is called Delaunay triangulation Tm .

• if the equality in Delaunay criterion is fulfilled also for P not incident
to K then the Delaunay triangulation is degenerated
• empty circle property – Delaunay criterion in 2D
• empty sphere property – Delaunay criterion in 3D

Typeset in LATEX by Daniel Rypl

3 UNSTRUCTURED MESH GENERATION 31

Voronoi Diagram
Set of cells Vi around m points P i , i = 1, 2, 3, . . . , m in Rn , n ≥ 2
defined as

Vi = {P ∈ Rn | ∀i 6= j : k P P i k ≤ k P P j k}

is called Voronoi Diagram.

Duality between Delaunay Triangulation and Voronoi Diagram

• corners of Voronoi cells are centers of discs (circles, spheres)
circumscribed to simplices in Delaunay triangulation
• faces of Voronoi cells correspond to faces of simplices
in Delaunay triangulation

Typeset in LATEX by Daniel Rypl

3 UNSTRUCTURED MESH GENERATION 32

Duality between Voronoi diagram (left) and Delaunay triangulation (right) in 2D.

Typeset in LATEX by Daniel Rypl

3 UNSTRUCTURED MESH GENERATION 33

Algorithm
1. initial Delaunay triangulation setup
2. mesh generation
• boundary node insertion
• recovery of domain boundary
• interior classification
• interior node insertion
3. mesh optimization

• initial Delaunay triangulation setup

– one or few simplices (completely surrounding the domain)
with a priori fulfilled Delaunay criterion
– can be degenerated
– 2D: typically 2 triangles of square bounding box
– 3D: typically 5 or 6 tetrahedrons of cubic bounding box
Typeset in LATEX by Daniel Rypl
3 UNSTRUCTURED MESH GENERATION 34

Initial Delaunay triangulation.

• mesh generation
◦ incremental point insertion algorithm - Delaunay kernel
- Bowyer (Watson) algorithm
Ti+1 = Ti − CP + BP
P - (i + 1)th point from a convex hull S
Tj - Delaunay triangulation of first j points from a convex hull S
CP - cavity, set of elements K from Ti whose circumball contains P
BP - ball, set of new elements formed by boundary facets of CP and P
Typeset in LATEX by Daniel Rypl
3 UNSTRUCTURED MESH GENERATION 35

P
P

Bowyer algorithm.

- cavity CP is star-shaped
- boundary facets of CP are visible from P

Typeset in LATEX by Daniel Rypl

3 UNSTRUCTURED MESH GENERATION 36

◦ insertion of boundary points

- all points of boundary discretization are incrementally insert
in the Delaunay triangulation using Bowyer algorithm
◦ recovery of domain boundary
- boundary Delaunay triangulation may be not boundary conforming
(domain boundary facets are not present in the triangulation)
- topological transformations (edge and face swapping)
not sufficient in 3D (Schönhardt polyhedron)
violate Delaunay property (unless applied to degenerated scheme)
−→ constrained Delaunay triangulation
- insertion of additional points on boundary (Steiner points)
◦ interior classification
- simplices of the constrained Delaunay triangulation are classified
either as interior or exterior with respect to the model topology

Typeset in LATEX by Daniel Rypl

3 UNSTRUCTURED MESH GENERATION 37

Recovery of boundary (left 3) and interior classification (right).

◦ insertion of interior points

- points along edges of interior simplices not complying with desired
mesh density or of poor shape
- barycenters of interior simplices not complying with desired mesh
density or of poor shape
- circumcenters of interior simplices not complying with desired mesh
density or of poor shape

Typeset in LATEX by Daniel Rypl

3 UNSTRUCTURED MESH GENERATION 38

- points created on basis of AFT

(front is the interface between interior simplices complying
and not complying with desired mesh density and quality)
- points are inserted using modified Bowyer algorithms
(cavity CP cannot propagate over boundary facets)

• mesh optimization
◦ Laplacian smoothing of interior mesh nodes
- repositioning of nodes to barycenter of nodes of all elements
incident to smoothed node
- modifies mesh geometry
- preserves mesh topology
- iterative process (convergence after about 5 cycles)

Typeset in LATEX by Daniel Rypl

3 UNSTRUCTURED MESH GENERATION 39

◦ topological transformations (to remove slivers)

- diagonal edge swapping, generalized face swapping, node merging
- modify mesh topology
- preserve mesh geometry

Computational Aspects
• Delaunay kernel - cavity construction
◦ spatial search
(adjacency search from simplex containing point being inserted)
◦ robustness of in-circle and in-sphere test
(rounding errors, ill-conditioned simplices, perturbation)
• boundary recovery
◦ topological issue (localization of missing facet)
◦ geometrical issue (localization of entities intersecting missing facet)

Typeset in LATEX by Daniel Rypl

3 UNSTRUCTURED MESH GENERATION 40

Features (• - advantages, ◦ - disadvantages)

• strong mathematical background
• high quality graded meshes
• rather high speed
• flexible mesh density control
• convergence guaranteed
• validity of raw mesh through the meshing process
• favourable computational complexity O(N )
• extensible to anisotropic meshing

◦ validity not guaranteed (boundary recovery necessary)

◦ creation of slivers in mesh generation phase is not avoided
(slivers eliminated during mesh optimization)

Typeset in LATEX by Daniel Rypl

3 UNSTRUCTURED MESH GENERATION 41

◦ not invariant with respect to rotations

◦ requires existence of boundary discretization
◦ does not comply with given boundary discretization

Alternatives and Extensions

• mesh density control by octree or background grid
• use of preplaced points (octree - high probability of degeneracy)
• extensible to anisotropic meshing (using appropriate metric)
• applicable to curved surface meshing (indirect approach)
• extensible to curved surface meshing (direct approach) with applying
Delaunay property in tangent plane at a given location

Typeset in LATEX by Daniel Rypl

3 UNSTRUCTURED MESH GENERATION 42

3.2 Quadrilateral and Hexahedral Mesh Generation

3.2.1 Grid Based Methods

• quadtree-like (2D)
◦ 9-tree over domain interior
◦ all-quad templates
• octree-like (3D)
◦ 27-tree over domain interior
◦ all-hexa templates with one exception
• isomorphism technique (used to mesh boundary region between the
domain boundary and interior tree)
• poor quality elements along boundary
• limited mesh density flexibility
• cannot handle domains with internal faces
(multiple region and multiple material domains)
Typeset in LATEX by Daniel Rypl
3 UNSTRUCTURED MESH GENERATION 43

3.2.2 Advancing Front Based Methods

Paving and Plastering

• whole layer of elements is constructed along the part of the front
at a time
• paving (2D)
◦ all-quadrilateral meshes of high quality
◦ even number of boundary segments must be maintained
in each closed part of the front
• plastering (3D)
◦ mixed meshes of good quality
◦ seams and wedges used to resolve a conforming mesh closure
in areas where layers would overlap or coincide with the boundary

Typeset in LATEX by Daniel Rypl

3 UNSTRUCTURED MESH GENERATION 44

Triangle Merging
• extension of 2D advancing front technique
◦ merging two consequently generated triangles to form a quad
◦ modification of the optimal point placement
◦ preserving even number of segments in each closed part of the front
• high quality mesh

3.2.3 Topology Based Methods

Whisker Weaving
• spatial twisted continuum
• constructs a dual (topology based) representation of mesh
(from a boundary discretization)
• identifies chords - chains of elements neighbouring by a facet
• mesh geometry is derived from topology
Typeset in LATEX by Daniel Rypl
3 UNSTRUCTURED MESH GENERATION 45

3.2.4 Postprocessing Based Methods

Simplex Splitting
• initial grid of half density
• triangles split into 3 quadrilaterals
• tetrahedrons split into 4 hexahedrons
• poor connectivity and quality (especially in 3D)

Triangle Merging
• initial triangular grid of half density
• merging neighbouring triangles forming well-shaped quadrilaterals
• one-level refinement
◦ triangles split into 3 quadrilaterals
◦ quadrilaterals split into 4 quadrilaterals
• high quality mesh
• unable to form single row mesh
Typeset in LATEX by Daniel Rypl
REFERENCES 46

References
[1] T.J. Baker, Automatic mesh generation for complex three-dimensional re-
gions using a constrained Delaunay triangulation, Engineering with Com-
puters 5 (1989), 161–175.
[2] T.D. Blacker, Paving: A new approach to automated quadrilateral mesh gen-
eration, International Journal for Numerical Methods in Engineering 32
(1991), 811–847.
[3] J. Bonet and J. Peraire, An alternating digital tree (ADT) algorithm for
3D geometric searching and intersection problem, International Journal for
Numerical Methods in Engineering 31 (1991), 1–17.
[4] A. Bowyer, Computing Dirichlet tessellations, Computer Journal 24 (1981),
162–167.
[5] P.J. Frey, H. Borouchaki, and P.L. George, Delaunay tetrahedrization us-
ing an advancing-front approach, Proceedings of 5th International Meshing
Roundtable, Sandia National Laboratories, 1996.
[6] H. Jin and R.I. Tanner, Generation of unstructured tetrahedral meshes by
REFERENCES 47

advancing front technique, International Journal for Numerical Methods in

Engineering 36 (1993), 1805–1823.
[7] B.P. Johnston, J.M. Sullivan, and A. Kwasnik, Automatic conversion of tri-
angular finite element meshes to quadrilateral elements, International Jour-
nal for Numerical Methods in Engineering 31 (1991), 67–84.
[8] C.K. Lee and S.H. Lo, A new scheme for the generation of graded quadrilat-
eral meshes, Computers and Structures 52 (1994), 847–857.
[9] S.H. Lo, A new mesh generation scheme for arbitrary planar domains, In-
ternational Journal for Numerical Methods in Engineering 21 (1985), 1403–
1426.
[10] R. Löhner and P. Parikh, Three dimensional grid generation by the advancing
front technique, International Journal for Numerical Methods in Fluids 8
(1988), 1135–1149.
[11] P. Möller and P. Hansbo, On advancing front mesh generation in three di-
mensions, International Journal for Numerical Methods in Engineering 38
(1995), 3551–3569.
[12] M.A. Price, C.G. Armstrong, and M.A. Sabin, Hexahedral mesh generation
REFERENCES 48

by medial axis subdivisions: I. solids with convex edges, 1994, Submitted to

International Journal for Numerical Methods in Engineering.
[13] R. Schneiders, A grid-based algorithm for the generation of hexahedral ele-
ment meshes, Engineering with Computers 12 (1996), 168–177.
[14] R. Schneiders, R. Schindler, and F. Weiler, Octree-based generation of
hexahedral element meshes, Proceedings of 5th International Meshing
Roundtable, Sandia National Laboratories, 1996.
[15] W.J. Schroeder and M.S. Shephard, A combined octree/Delaunay method for
fully automatic 3-D mesh generation, International Journal for Numerical
Methods in Engineering 29 (1990), 37–55.
[16] M.S. Shephard and M.K. Georges, Automatic three-dimensional mesh gen-
eration by the finite octree technique, International Journal for Numerical
Methods in Engineering 26 (1991), 709–749.
[17] , Reliability of automatic 3-D mesh generation, Computer Methods
in Applied Mechanics and Engineering 101 (1992), 443–462.
[18] M.B. Stephenson, S. Canann, and T.D. Blacker, Plastering: A new approach
to automated 3D hexahedral mesh generation, Tech. Report SAND 89-2192,
REFERENCES 49

Sandia National Laboratories, 1990.

[19] T.J. Tautges, T.D. Blacker, and S. Mitchell, The whisker weaving algorithm:
A connectivity-based method for constructing all-hexahedral finite element
meshes, International Journal for Numerical Methods in Engineering 39
(1996), 3327–3349.

[20] D.F. Watson, Computing the n-dimensional Delaunay tessellation with ap-
plication to Voronoi polytopes, The Computer Journal 24 (1981), 167–172.

[21] N.P. Weatherill and O. Hassan, Efficient three-dimensional Delaunay trian-

gulation with automatic point creation and imposed boundary constraints, In-
ternational Journal for Numerical Methods in Engineering 37 (1994), 2005–
2039.

[22] J.Z. Zhu, O.C. Zienkiewicz, E. Hinton, and J. Wu, A new approach to the de-
velopment of automatic quadrilateral mesh generation, International Journal
for Numerical Methods in Engineering 32 (1991), 849–866.