On the Link between Mesh Size Adaptation and Irregular Vertices
Daniel Zint
a
and Roberto Grosso
b
Visual Computing, Friedrich-Alexander Universit
¨
at Erlangen-N
¨
urnberg, Cauerstr. 11, 91058 Erlangen, Germany
Keywords:
Mesh Generation, Block-structured Grids, Irregular Vertices.
Abstract:
In numerical simulations and computer graphics meshes are often required to have varying element sizes.
High resolution, i.e. small elements, should be only used where necessary. The transition between element
sizes requires introducing irregular vertices. In this work, we examine the occurance of irregular vertices
in transition regions by setting up an advancing front triangulation that generates optimal transitions. We
establish a relation between the appearance of irregular vertices and the properties of the size function and
show that a linear transition between different element sizes can be achieved without any singularities on the
interior of the transition. Therefore, we can optimize triangulations by setting transition fronts accordingly.
These results are used to estimate properties of block-structured grids, e.g. how many blocks are required to
represent a given domain correctly.
1 INTRODUCTION
Unstructured and block-structured meshes are widely
used in computer graphics and in numerical sim-
ulations. They are favored over fully structured
meshes due to their adaptiveness to complex geom-
etry. Block-structured grids (BSGs) are on the rise
for the last decade. A BSG is a mesh where each el-
ement contains a fully structured mesh. In computer
graphics BSGs enable tensor product surface repre-
sentations, grid-based multi-resolution techniques, or
discrete pixel-based map representations (Campen,
2017). In numerical simulations, block structure
enables optimizations which reduce simulation time
drastically. For example, multigrid solvers can be
used which converge much faster than solvers that
work on unstructured meshes (Armstrong et al.,
2015).
Unstructured meshes are irreplaceable in many
applications, not just because of their adaptiveness to
complex geometry but also for their ability to adapt to
a size function. It is well known that there is a link
between changes in element size and the appearance
of irregular vertices. The more rapid the element size
changes the more irregularity is required. Neverthe-
less, at least as far as we know, this link was never in
the focus of research.
In this work, we study the appearance of irregu-
a
https://orcid.org/0000-0003-4491-1685
b
https://orcid.org/0000-0001-5965-5325
lar vertices in transition zones between different ele-
ment sizes. We describe how irregularities, element
quality, and the number of elements are connected.
We use this knowledge to achieve optimal transition
zones. These are used to remove irregular vertices
from meshes that were generated with Delaunay re-
finement or isotropic remeshing. Furthermore, we
gain a deeper understanding about BSG generation.
More precisely, we predict the properties of size-
adapted BSGs and show constraints that a size func-
tion might impose.
1.1 Singularity and Irregularity
The topology of a mesh causes singularities (Beaufort
et al., 2017). The Euler characteristic for an orientable
surface S embedded in R
3
is
χ = 2 2g b, (1)
where g is the genus of the surface and b the num-
ber of boundaries. Thus, for a sphere we have χ = 2,
for a disk χ = 1, and for a ring (disk with an interior
boundary) or a torus χ = 0. Theoretically, a singular-
ity corresponds to a vertex with no incident edges. In
practice a singularity is spread over several vertices.
It is more viable to talk about vertex and mesh irreg-
ularity.
Definition 1. The vertex irregularity ι
v
is the vertex
valence deducted with its optimal valence. For trian-
gle meshes the optimal valence for interior vertices
is 6 and for boundary vertices 4. For quad meshes
Zint, D. and Grosso, R.
On the Link between Mesh Size Adaptation and Irregular Vertices.
DOI: 10.5220/0010259200670074
In Proceedings of the 16th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2021) - Volume 1: GRAPP, pages
67-74
ISBN: 978-989-758-488-6
Copyright
c
2021 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
67
the optimal valence for interior vertices is 4 and for
boundary vertices 3.
A vertex in a triangle mesh with valence 5 has an
irregularity of ι
v
= 56 = 1 on the interior and ι
v
=
5 4 = 1 on the boundary. In a quad mesh a valence
5 vertex has an irregularity of ι
v
= 5 4 = 1 on the
interior and ι
v
= 5 3 = 2 on the boundary.
Definition 2. The mesh irregularity ι
m
is the sum of
all vertex irregularities,
ι
m
=
N
i=1
ι
vi
, (2)
where N is the number of vertices in the mesh and ι
vi
is the irregularity of the vertex with index i.
Assume the unit square is subdivided in two tri-
angles, Figure 1a. The Euler characteristic is χ = 1.
All vertices are on the boundary and thus their opti-
mal valence would be 4. Two vertices have valence 3
giving ι
v
= 1 and two have valence 2 which corre-
sponds to ι
v
= 2. Therefore, the mesh irregularity
is ι
m
= 2 ·(2) + 2 ·(1) = 6. This irregularity is
purely determined by the topology. We could refine
the mesh and flip edges which would create new ir-
regular vertices, but the mesh irregularity will always
remain the same as long as the genus or the number
of boundaries is not changed, see Figures 1b and 1c.
For investigating transition zones, we need more
detailed information about mesh irregularity. We dis-
tinguish between positive and negative irregularity.
Definition 3. The positive / negative mesh irregular-
ity ι
+
m
/ ι
m
is the sum of all positive / negative vertex
irregularities,
ι
+
m
=
N
i=1
(
ι
vi
if ι
vi
> 0
0 otherwise,
(3)
ι
m
=
N
i=1
(
ι
vi
if ι
vi
< 0
0 otherwise.
(4)
The unit square has ι
+
m
= 0 and a negative irregu-
larity of ι
m
= 6. This remains constant if we sub-
divide the mesh as all new vertices have optimal va-
lence. If edges are flipped, positive and negative ir-
regularity are increased. In Figure 1c we have ι
+
m
= 1
and ι
m
= 7. Note, that both change equally as they
sum up to ι
m
,
ι
m
= ι
+
m
+ ι
m
. (5)
1.2 Irregularity and Quality
Irregular vertices have a negative impact on mesh
quality. They impose an upper bound to the minimal
Table 1: Upper bound to minimal quality depending on ver-
tex valence. Quality is measured with mean ratio, minimal
angle, and ratio between longest and shortest edge.
valence q α
min
l
max
/l
min
3 0.60 30 1.73
4 0.87 45 1.41
5 0.97 54 1.18
6 1.00 60 1.00
7 0.98 51 1.15
8 0.94 45 1.31
9 0.90 40 1.46
10 0.85 36 1.62
11 0.81 33 1.77
12 0.76 30 1.93
15 0.65 24 2.40
20 0.51 18 3.20
25 0.42 14 3.99
quality of their incident elements. A common way to
measure quality is using the mean ratio metric,
q = 4
3
A
3
i=1
l
2
i
, (6)
where A is the signed area of the triangle and l
i
is the
length of its edges (Bank and Smith, 1997; Canann
et al., 1998; Amenta et al., 1999). We also consider
as quality measures the minimal angle, α
min
, and the
ratio between the longest and shortest edge of a tri-
angle, l
max
/l
min
. We only measure minimal quality as
for simulations, a badly shaped element might cause
numerical instabilities. In contrast, average quality
does not contain much information and is therefore
neglected.
We compare vertex valence to the optimal mini-
mal quality of its incident triangles, Table 1. This is
an upper bound, i.e. a vertex with valence 9 might
have a quality of 0.90 but in almost any configuration
it will be lower. In the optimal configuration all inci-
dent edges have the same length and the same angle
between them. Usually, geometric constraints pro-
hibit this setup.
Highly irregular vertices also impose anisotropy.
A triangle with a valence four vertex has at best a quo-
tient of longest over shortest edge of 1.41. Depending
on the numerical scheme used by a simulation this
might not be acceptable.
1.3 Related Work
Adapting to a size function was considered thor-
oughly in triangle mesh generation methods. The
most common procedure is to generate a Delaunay
triangulation of a domain and then refine where nec-
essary (B
¨
ansch, 1991; Shewchuk, 1997; Cheng et al.,
GRAPP 2021 - 16th International Conference on Computer Graphics Theory and Applications
68
(a) Unit Square
ι
m
= 2 ·(2) + 2 ·(1) = 6
(b) Refined
ι
m
= 2 ·(2) + 2 ·(1) + 5 ·0 = 6
(c) Flipped Edge
ι
m
= 1 ·(2) + 5 ·(1) + 1 ·1 = 6
Figure 1: The unit square contains four irregular vertices, two with ι
v
= 1 and two with ι
v
= 2, summing up to a mesh
irregularity of ι
m
= 6. Refining the unit square adds five vertices with ι
v
= 0. An edge flip affects the vertex irregularity
and therefore also the positive and negative mesh irregularity. Vertices with negative or positive irregularity are marked blue
or red correspondingly.
2012). Most of this methods add Steiner points to
the mesh and flip edges to re-establish the Delau-
nay triangulation. This iterative approach, called De-
launay refinement, is robust and generates unstruc-
tured meshes with high quality. Triangle is an early
software package that used the Delaunay refinement
method (Shewchuk, 1996). Subsequent work focused
on improving the placement of vertices (
¨
Ung
¨
or, 2004;
Persson, 2005; Erten and
¨
Ung
¨
or, 2009; Engwirda and
Ivers, 2016).
In computer graphics, Delaunay triangulations are
generated a bit differently with isotropic remeshing
(Botsch and Kobbelt, 2004; Alliez et al., 2008). The
steps split, collapse, and smooth are performed subse-
quently to achieve almost equilateral triangles. Other
remeshing methods are based on parametrization (Al-
liez et al., 2003). Just like Delaunay refinement,
remeshing methods can generate meshes with vary-
ing element sizes.
Advancing front techniques, sometimes called
paving, start at a mesh boundary and add a ring of el-
ements (Peraire et al., 1987; Blacker and Stephenson,
1991; L
¨
ohner and Parikh, 1988). This is repeated un-
til the whole domain is filled with elements. Advanc-
ing front techniques often deliver good results along
boundaries but elements in the interior might have low
quality. Advancing front techniques regained interest
when they were combined with cross fields and are
used especially for quad mesh generation (Remacle
et al., 2013; Georgiadis et al., 2017).
None of these methods keeps track of irregular
vertices. When the research focus of meshing moved
from triangles to quads, irregularities became a topic
of bigger concern. Whereas triangles are quite flex-
ible regarding irregular vertices, they are a serious
issue in quad meshes due to the lower optimal va-
lence. The introduction of cross fields was an im-
portant contribution (Klberer et al., 2007; Ray et al.,
2008; Bommes et al., 2009; Bommes et al., 2013;
Kowalski et al., 2013; Crane et al., 2010). Cross fields
are computed by solving a system of partial differen-
tial equations on the domain. Besides giving input
for advancing front techniques, cross fields also show
positions of singularities and therefore allow directly
the generation of block-structured meshes. A down-
side is that cross fields are not aware of size functions.
For mesh adaptation the grid structure needs to be re-
fined by adding further singularities, e.g. Armstrong
et al. presents patterns for multiblock mesh refine-
ment (Armstrong et al., 2018). Zint and Grosso try to
overcome the issue of cross fields by using a simplifi-
cation method on an underlying size function to cre-
ate blocks of the correct size (Zint et al., 2019). Their
claim to representing element size correctly leads to a
large number of blocks.
2 TRIANGULATING A
TRANSITION ZONE
In this section, we study transition zones between ar-
eas of different element sizes. First, we will consider
a direct transition which does not add any further ver-
tices. Usually, the length of a transition zone is given
and the mesh generator has to find the optimal trian-
gulation. First, we go the other way around and exam-
ine the behavior of triangulations when we adapt the
transition length. In a second step, we will add fronts
in the transition zone. Here, we go back to the orig-
inal problem and find the optimal triangulation for a
given transition length.
The examples are generated with an advancing
front method. Therefore, we will use the term front
for the vertical lines in a mesh. For example, the unit
square consists of two fronts, one on the left and one
on the right. Furthermore, transitions are described
as patterns. The transition pattern 4:1 has 4 elements
on the first front and 1 element on the last. If another
front with two elements is inserted in between, the
pattern is 4:2:1. The number of elements on the front
with index k is denoted by f
k
. Thus, the pattern with
On the Link between Mesh Size Adaptation and Irregular Vertices
69
(a) 2:1 (b) 3:1 (c) 3:2
(d) 4:1 (e) 4:2 (f) 4:3
Figure 2: Delaunay triangulation on transition with only
two fronts.
four fronts is f
1
: f
2
: f
3
: f
4
.
2.1 Direct Transition
We use relative irregularity for analyzing transition.
This is advantageous as it cancels out the influence of
geometry.
Definition 4. The relative vertex irregularity
˜
ι
v
is the
vertex irregularity in comparison to the unit square as
shown in Figure 1a. This only affects the four corner
vertices.
In Figure 2a the top left vertex has
˜
ι
v
= 1 be-
cause in the unit square this vertex has three outgoing
edges and here only two. Furthermore, the vertices in
the middle and bottom left and the bottom right have
˜
ι
v
= 0. The vertex in the top right has
˜
ι
v
= +1. If f
1
is increased, a vertex with
˜
ι
v
= 1 is added and the
irregularity of a vertex on the right side is increased
by one, Figure 2b. The opposite happens if f
2
is in-
creased, Figure 2c.
Definition 5. The relative mesh irregularity
˜
ι
m
is the
sum of the relative irregularity of all vertices,
˜
ι
m
=
N
i=1
˜
ι
vi
. (7)
Definition 6. The relative positive / negative mesh ir-
regularity
˜
ι
+
m
/
˜
ι
m
is the sum of all relative positive /
negative vertex irregularities,
˜
ι
+
m
=
N
i=1
(
˜
ι
vi
if
˜
ι
vi
> 0
0 otherwise,
(8)
˜
ι
m
=
N
i=1
(
˜
ι
vi
if
˜
ι
vi
< 0
0 otherwise.
(9)
The cases shown in Figure 2 which only consist of
two fronts already illustrate some information about
transitions:
Figure 3: Minimal mean ratio quality of different transitions
depending on the transition-length. The domain has a height
of 1.
1. The number of elements between two fronts f
k
and f
k+1
is
N
E
( f
k
, f
k+1
) = f
k
+ f
k+1
. (10)
2. The relative positive / negative mesh irregularity
is
˜
ι
±
m
= ±( f
1
f
2
). (11)
3. Negative irregularities are on the first front, posi-
tive on the second. Or more general, negative ir-
regularities are on the front with more elements.
This is independent of the triangulation, as long
as triangles are not degenerated.
Additionally, we study the element quality in tran-
sition regions. We observe that triangle quality de-
creases when f
1
or f
2
increases. Triangles need to
be compressed to make room for others. The quality
of the pattern 4:3 in Figure 2f would be higher if the
transition zone would be shorter. Each transition pat-
tern has its own optimal length. The minimal element
quality of the meshes in Figure 2 when stretched in
horizontal direction is plotted in Figure 3. Some tran-
sitions always have low quality like 3:1 or 4:1. Here
arises the question if we can design such transitions
with better shaped elements by adding vertices.
2.2 Transition with Multiple Fronts
In the following, we add additional fronts in between
the left and right boundary to allow a smoother transi-
tion. Therefore, we generalize the domain descrip-
tion. A domain consists of n vertical fronts with
f
1
> f
2
> ... > f
n
. The region between fronts is trian-
gulated according to the Delaunay criterion. Further-
more, we introduce the relative irregularity of fronts.
Definition 7. The relative irregularity
˜
ι
f k
of a front f
k
is the sum of the relative irregularity of all vertices in
this front,
˜
ι
f k
=
vf
k
˜
ι
v
. (12)
GRAPP 2021 - 16th International Conference on Computer Graphics Theory and Applications
70
(a) 4:3:2:1 (b) 4:3:1 (c) 4:2:1
Figure 4: Transition from 4 to 1 element with different pat-
terns.
Definition 8. The relative positive / negative irregu-
larity
˜
ι
+
f k
/
˜
ι
f k
of a front f
k
is the sum of all relative
positive / negative vertex irregularities in this front,
˜
ι
+
f k
=
vf
k
(
˜
ι
v
if
˜
ι
v
> 0
0 otherwise,
(13)
˜
ι
f k
=
vf
k
(
˜
ι
v
if
˜
ι
v
< 0
0 otherwise.
(14)
Figure 4 shows different transitions from 4 el-
ements to 1. In Figure 4a two fronts are added.
The second front has f
2
= 3 elements and the third
f
3
= 2. Thus, the number of edges is decreased
by one between all fronts. The first front (which
is the left boundary) has a relative irregularity of
˜
ι
f 1
= 1, the second and third front have no irregular-
ity
˜
ι
f 2
=
˜
ι
f 3
= 0, and the right boundary has one pos-
itive irregularity
˜
ι
f 4
= 1. Thus, the additional fronts
reduced the relative mesh irregularity from
˜
ι
±
m
= ±3
to
˜
ι
±
m
= ±1.
Inserting fronts can be seen as stacking transition
zones. In the 4:2:1 example in Figure 4c we have a
transition from 4:2 and another from 2:1. The number
of elements can be computed as
N
E
( f
1
, f
2
,..., f
n
) = N
E
( f
1
, f
2
) + N
E
( f
2
, f
3
) + ...
= ( f
1
+ f
2
) + ... + ( f
n1
+ f
n
)
= f
1
+ f
n
+ 2
n1
i=2
f
i
. (15)
The first transition generates 2 positive irregulari-
ties on the first front and two negative on the sec-
ond. The second transition generates 1 positive ir-
regularity on the second front and one negative on the
third. The positive and one negative irregularity on
the second front are canceling each other out giving
˜
ι
f 2
= 1 2 = 1. This generalizes to the relative ir-
regularity of a front f
k
,
˜
ι
f k
= f
k1
2 f
k
+ f
k+1
with 1 k n. (16)
This holds also for the boundary fronts when we as-
sume that element size is constant outside of the tran-
sition zone, i.e. f
0
= f
1
and f
n+1
= f
n
. From Equation
16 it follows that irregularities only appear when the
gradient of the size function is not constant.
Figure 5: Minimal quality of different transitions between
four and one element depending on the transition-length.
With Equation 16 and f
k1
> f
k
> f
k+1
for
1 < k < n we can compute the relative positive and
negative mesh irregularity:
˜
ι
f 1
= f
0
2 f
1
+ f
2
= f
1
+ f
2
< 0 (17)
˜
ι
f n
= f
n1
2 f
n
+ f
n+1
= f
n1
f
n
> 0 (18)
˜
ι
f k
0 if f
k1
+ f
k+1
2 f
k
(19)
˜
ι
+
m
=
n
k=1
(
˜
ι
f k
if
˜
ι
f k
> 0
0 otherwise.
(20)
˜
ι
m
=
n
k=1
(
˜
ι
f k
if
˜
ι
f k
< 0
0 otherwise.
(21)
˜
ι
+
m
=
n
k=2
˜
ι
f k
if f
k1
+ f
k+1
2 f
k
(22)
˜
ι
m
= f
2
f
1
if f
k1
+ f
k+1
2 f
k
(23)
Equations 19, 22, and 23 are of special interest as they
give hints about optimal transition zones. For the case
f
k1
+ f
k+1
= 2 f
k
we can avoid irregularities within
the transition zone completely. Thus, the descent in
edge size should be equal between all fronts which
corresponds to a linear size function. Non-linear size
functions will always impose irregularities. Figure 4a
shows an optimal transition. Figure 4b and 4c have
the same relative mesh irregularity.
In terms of quality, it might not always be the best
choice to opt for the lowest possible amount of irregu-
lar vertices. Figure 5 shows the minimal quality of the
meshes from Figure 4 when stretched along the hori-
zontal axis. The 4:3:2:1 pattern is preferable when the
transition is longer. If the transition should be more
rapid, the 4:2:1 mesh is the better choice. Thus, de-
pending on the gradient size of a size function one
might choose a different pattern. In any case, asym-
metric transitions like 3:1 should be avoided as they
impose low quality while having the same relative ir-
regularity as symmetric patterns.
On the Link between Mesh Size Adaptation and Irregular Vertices
71
(a) remeshing
(b) optimal
Figure 6: Transition from 40 to 10 elements once achieved
with remeshing, once by computing the optimal transition.
Negative irregularities are marked blue, positive irregulari-
ties are red.
3 COMPARISON TO ADAPTIVE
REMESHING
A simple way to apply a size function to a mesh is by
using adaptive remeshing like the one of Botsch and
Kobbelt (Botsch and Kobbelt, 2004). Even though
remeshing delivers overall good quality, it also gen-
erates many unnecessary irregular vertices. In Fig-
ure 6a we apply the remeshing of Botsch and Kobbelt
to a transition from 40 to 10 elements, assuming a
linear size function. The average triangle quality is
0.98 but the minimal quality only 0.39. By computing
the optimal transition we only get irregularities on the
left and right boundary and achieve a minimal qual-
ity of 0.79. The average quality is 0.86. Both meshes
have almost the same number of elements, remesh-
ing produces 1512 and the optimal contains 1 500 el-
ements. In Figure 7, we applied a size function which
decreases exponentially. Similar issues are observed
here. Remeshing only achieved a minimal quality of
0.33 and an average quality of 0.75 whereas the op-
timal transition has a minimal and average quality of
0.87. Remeshing generates 313 triangles, the optimal
transition has 381.
Also Delaunay refinement can be improved by
identifying transition zones and exchanging them
with the optimal transition. In Figure 8 we selected
(a) remeshing (b) optimal
Figure 7: Transition from 128 to 1 element with an expo-
nential size function once achieved with remeshing, once
by computing the optimal transition. Negative irregularities
are marked blue, positive irregularities are red.
(a) Delaunay triangulation (b) optimal
Figure 8: Remeshing a rectangular region in a mesh gener-
ated with Delaunay refinement reduces the irregularity sig-
nificantly.
a rectangular region of a mesh designed for fluid
simulations. Deploying optimal transition reduces
the amount of irregular vertices significantly from 63
down to 8. The minimal quality of this region de-
creases slightly from 0.76 to 0.75. This value could
be further improved by smoothing the boundaries of
the rectangular region. The number of elements goes
down from 205 to 189.
4 PROPERTY-ESTIMATION OF
BLOCK-STRUCTURED GRIDS
Optimal transitions can be used to estimate proper-
ties of BSGs. Assume, the marked region in Figure 9
should be subdivided into triangular blocks. The re-
gion is a transition from 28 to 4 with 10 or 11 fronts.
This can be approximated by a relation of 7:1 and one
or two interior fronts. Thus, the optimal pattern would
be either 7:5:3:1 or 7:4:1, Figure 10. Refining these
patterns two times with uniform subdivision results in
BSGs with 16 triangles per block that are comparable
to the unstructured mesh, Table 2.
The size function imposes that the region should
consist of at least 16 blocks. The minimal number
of blocks for a 7:1 transition would be 8, but besides
GRAPP 2021 - 16th International Conference on Computer Graphics Theory and Applications
72
Figure 9: A rectangular transition region within an unstruc-
tured triangle mesh. The transition gives information about
the number of blocks that is required to represent the do-
main correctly.
Table 2: BSG property estimation.
unstructured 7:5:3:1 7:4:1
# blocks 24 16
# triangles 297 384 256
q
min
0.75 < 0.98 < 0.90
˜
ι
+
m
40 2 3
generating large irregularities, this would also lead to
anisotropy and is therefore not advisable. Also a 7:2:1
pattern would be possible which would then result in
14 blocks but already 16 blocks have less triangles
than in the unstructured case. Reducing the number
of blocks only makes sense if we can refine the blocks
further. This is constrained by the coarsest bound-
ary which only contains four elements. Considering a
larger region does not solve this issue as the relation
between fine to coarse will remain 7:1.
If 16 triangles build one block, we can compute
the number of blocks that is required for representing
all 18578 triangles of the unstructured mesh,
(a) 7:5:3:1 (b) 7:4:1
(c) 7:5:3:1 refined (d) 7:4:1 refined
Figure 10: Transition from 7 to 1 element either with one or
two additional fronts.
n
block
=
18578
16
1161. (24)
This is just a rough estimation but it hints at the
minimal amount of blocks. It will not be possible
to represent this domain with 10, 100, or even 500
blocks if element size and quality should be pre-
served.
Depending on the simulation the size function
might only be a lower limit, i.e. elements might be
smaller but not larger than the value given by the size
function. In that case, much less blocks can be gener-
ated as the relation between fine and coarse can be set
to 4:1 or even 2:1 and blocks may contain more ele-
ments than just 16. To enable this transition the num-
ber of elements needs to be increased. If one fourth
of the triangles would be refined uniformly we would
get around 32510 triangles. Assuming that one block
would consist of 64 triangles, the mesh could be rep-
resented by approximately 510 blocks.
In general, the size function gradient determines
the required number of blocks. The larger the gradi-
ent the more blocks are required. This issue can not
be solved by smoothing as this only adapts the mesh
locally.
5 CONCLUSIONS
We studied the transition between different element
sizes in triangular meshes. We demonstrated that the
minimum number of irregularities in a transition zone
can be computed with a simple formula and studied
On the Link between Mesh Size Adaptation and Irregular Vertices
73
the influence of irregularities on mesh quality. We
also show that a transition requires no interior singu-
larities if the size function gradient is constant. With
this knowledge, we can remove irregularities from
meshes generated by remeshing or with Delaunay
refinement without decreasing quality significantly.
Furthermore, we present a technique to estimate the
required number of blocks for correctly representing
a domain while satisfying the constraints imposed by
a size function. In the future, we plan to extend this
concepts to more complex problems.
REFERENCES
Alliez, P., De Verdiere, E. C., Devillers, O., and Isenburg,
M. (2003). Isotropic surface remeshing. In 2003
Shape Modeling International., pages 49–58. IEEE.
Alliez, P., Ucelli, G., Gotsman, C., and Attene, M. (2008).
Recent advances in remeshing of surfaces. In Shape
analysis and structuring, pages 53–82. Springer.
Amenta, N., Bern, M., and Eppstein, D. (1999). Optimal
point placement for mesh smoothing. Journal of Al-
gorithms, 30(2):302–322.
Armstrong, C. G., Fogg, H. J., Tierney, C. M., and Robin-
son, T. T. (2015). Common themes in multi-block
structured quad/hex mesh generation. Procedia En-
gineering, 124:70 82. 24th International Meshing
Roundtable.
Armstrong, C. G., Li, T. S., Tierney, C., and Robinson, T. T.
(2018). Multiblock mesh refinement by adding mesh
singularities. In International Meshing Roundtable,
pages 189–207. Springer.
Bank, R. E. and Smith, R. K. (1997). Mesh smoothing using
a posteriori error estimates. SIAM Journal on Numer-
ical Analysis, 34(3):979–997.
B
¨
ansch, E. (1991). Local mesh refinement in 2 and 3 di-
mensions. IMPACT of Computing in Science and En-
gineering, 3(3):181–191.
Beaufort, P.-A., Lambrechts, J., Henrotte, F., Geuzaine, C.,
and Remacle, J.-F. (2017). Computing cross fields
a pde approach based on the ginzburg-landau theory.
Procedia engineering, 203:219–231.
Blacker, T. D. and Stephenson, M. B. (1991). Paving: A
new approach to automated quadrilateral mesh gener-
ation. International Journal for Numerical Methods
in Engineering, 32(4):811–847.
Bommes, D., Lvy, B., Pietroni, N., Puppo, E., Silva, C.,
Tarini, M., and Zorin, D. (2013). Quad-mesh gener-
ation and processing: A survey. Computer Graphics
Forum, 32(6):51–76.
Bommes, D., Zimmer, H., and Kobbelt, L. (2009). Mixed-
integer quadrangulation. ACM Trans. Graph., 28(3).
Botsch, M. and Kobbelt, L. (2004). A remeshing approach
to multiresolution modeling. In Proceedings of the
2004 Eurographics/ACM SIGGRAPH symposium on
Geometry processing, pages 185–192.
Campen, M. (2017). Partitioning surfaces into quadrilat-
eral patches: A survey. Computer Graphics Forum,
36(8):567–588.
Canann, S. A., Tristano, J. R., Staten, M. L., et al. (1998).
An approach to combined laplacian and optimization-
based smoothing for triangular, quadrilateral, and
quad-dominant meshes. In IMR, pages 479–494. Cite-
seer.
Cheng, S.-W., Dey, T. K., and Shewchuk, J. (2012). Delau-
nay mesh generation. CRC Press.
Crane, K., Desbrun, M., and Schr
¨
oder, P. (2010). Trivial
connections on discrete surfaces. In Computer Graph-
ics Forum, volume 29, pages 1525–1533. Wiley On-
line Library.
Engwirda, D. and Ivers, D. (2016). Off-centre steiner
points for delaunay-refinement on curved surfaces.
Computer-Aided Design, 72:157–171.
Erten, H. and
¨
Ung
¨
or, A. (2009). Quality triangulations with
locally optimal steiner points. SIAM Journal on Sci-
entific Computing, 31(3):2103–2130.
Georgiadis, C., Beaufort, P., Lambrechts, J., and Remacle,
J. (2017). High quality mesh generation using cross
and asterisk fields: Application on coastal domains.
CoRR, abs/1706.02236.
Klberer, F., Nieser, M., and Polthier, K. (2007). Quadcover
- surface parameterization using branched coverings.
Computer Graphics Forum, 26(3):375–384.
Kowalski, N., Ledoux, F., and Frey, P. (2013). A pde based
approach to multidomain partitioning and quadrilat-
eral meshing. In Proceedings of the 21st international
meshing roundtable, pages 137–154. Springer.
L
¨
ohner, R. and Parikh, P. (1988). Generation of three-
dimensional unstructured grids by the advancing-front
method. International Journal for Numerical Methods
in Fluids, 8(10):1135–1149.
Peraire, J., Vahdati, M., Morgan, K., and Zienkiewicz,
O. C. (1987). Adaptive remeshing for compressible
flow computations. Journal of computational physics,
72(2):449–466.
Persson, P.-O. (2005). Mesh generation for implicit geome-
tries. PhD thesis, Massachusetts Institute of Technol-
ogy.
Ray, N., Vallet, B., Li, W. C., and L
´
evy, B. (2008). N-
symmetry direction field design. ACM Transactions
on Graphics (TOG), 27(2):1–13.
Remacle, J. F., Henrotte, F., Baudouin, T. C., Geuzaine, C.,
B
´
echet, E., Mouton, T., and Marchandise, E. (2013).
A frontal delaunay quad mesh generator using the L
norm. International Journal for Numerical Methods
in Engineering, 94(5):494–512.
Shewchuk, J. R. (1996). Triangle: Engineering a 2d quality
mesh generator and delaunay triangulator. In Work-
shop on Applied Computational Geometry, pages
203–222. Springer.
Shewchuk, J. R. (1997). Delaunay refinement mesh gener-
ation. Technical report, Carnegie-Mellon Univ Pitts-
burgh Pa School of Computer Science.
¨
Ung
¨
or, A. (2004). Off-centers: A new type of steiner points
for computing size-optimal quality-guaranteed delau-
nay triangulations. In Latin American Symposium on
Theoretical Informatics, pages 152–161. Springer.
Zint, D., Grosso, R., Aizinger, V., and K
¨
ostler, H. (2019).
Generation of block structured grids on complex
domains for high performance simulation. Com-
putational Mathematics and Mathematical Physics,
59(12):2108–2123.
GRAPP 2021 - 16th International Conference on Computer Graphics Theory and Applications
74