parametrization of the domain where integer iso-values of the parameter fields
form the sides [6, 7, 8], using Riemann geometry [9, 10, 11], or like in our case,
constructing a cross-field structure that will guide the integral lines emanating
from singularities [12, 13, 14, 15, 16, 17].
Although leaning on heterogeneous approaches, all the above-mentioned
methods share the common challenge: dealing with the inevitable singularity
configuration. A singularity appears where a cross-field vanishes and it repre-
sents an irregular vertex of a quad layout/quad mesh [18], i.e., a vertex which
doesn’t have exactly four adjacent quadrilaterals. The singular configuration is
constrained by the Euler characteristic χ, which is a topological invariant of a
surface. Moreover, a suboptimal number or location of singularities can have
severe consequences: causing undesirable thin partitions, large distortion, not
an adequate number and/or tangential crossings of separatrices as well as limit
cycles [5, 7, 17].
Cross-field guided methods can be very useful and flexible but they typically
lack direct control over the positions of the singularities and the structures of the
quad layout [19]. Our cross-field formulation, with mathematical foundations
detailed in Section 2, offers a contribution to this issue through the concept of
user-imposed singularity configuration in order to gain direct control over their
number, location, and valence (number of adjacent quadrilaterals). The user
is entitled to use either naturally appearing singularities, obtained by solving
a non-linear problem [14, 17, 20, 21], using globally optimal direction fields
[22], or to impose its own singularity configuration, possibly with high valences,
as illustrated in Fig. 1. It is important to note that the choice of singularity
pattern is not arbitrary, though. Moreover, it is under the direct constraint
of Abel-Jacobi theory [9, 10, 11] for valid singularity configurations. Here, the
singularity configuration is taken as an input and an integrable isotropic cross-
field is computed by solving only two linear systems, Section 3. Computation
of the scalar field Hused for this cross-field generation bears some resemblance
to the one developed for unstructured mesh generation on planar and curved
surface domains in [23]. Finally, the preliminary results of the developed cross-
field formulation for an isotropic block-structured quad mesh generation are
outlined using the 3-step pipeline [15] in Section 4.
Computing only one scalar field H(a metric that is flat except at singular-
ities) imposes a strict constraint on singularities’ placement, i.e., fulfilling all
Abel-Jacobi conditions. In practice, imposing suboptimal distribution of sin-
gularities may lead to not obtaining boundary-aligned cross-field, disabling an
isotropic quad mesh generation, Section 4.1 and 4.2. To bypass this issue,
we develop a new cross-field formulation on the imposed singularity configura-
tion, which considers the integrability, while relaxing the condition on isotropic
scaling of crosses’ branches. Here, two independent metrics H1and H2are
computed instead of only one as in the Abel-Jacobi framework, enabling an
integrable 2D cross-field generation with anisotropic scaling, Section 5.
Lastly, final remarks and some of the potential applications are discussed in
Section 6.
2