takeintoaccounttheempty-ballcriterion.Now,nonDelaunay-conformingtriangulationsareun-suitedforsomeapplications,suchasmedialaxisormid-surfacecomputationorspecificfinitevol-umesmethods.
Apriori
redefinition,
i.e.
findinganewconstraint,topologicallyandgeometricallyequivalenttotheinitialone,suchthatitappearsintoanyDelaunaytriangulationoftherelatedsetofvertices,allowstoprovideDelaunay-conformingresultingtriangu-lations,aswellastoavoidthesepossiblydiverging
aposteriori
recoveryoperations.Therefore,itisdesirabletoinvestigatethisapproach,in2and3dimensionsandtrytoproposeefficientandreli-ablealgorithms.
Relatedwork
Aconvergent
apriori
two-dimensionalconstraintsredefinitionalgorithm,withcomplexitybound,hasbeenproposedby[Edelsbrunner,Tan-93].However,thisapproachshouldberegardedasatheoreticalproofratherthanapracticalredefinitionmethod,providinganewstronglyDelaunay-conformingconstraint.Afirstapproachtowardsanefficientalgorithmcanbefoundin[George,Borouchaki-1997],butitscon-vergenceisensuredonlyformanifoldconstraintswithoutacuteangles.[Pebay-98c]addressedtheproblemandprovidedaconvergentalgorithm,nowproposedasapublicdomainsoftware.'Thethree-dimensionalcase,however,ismuchmorecomplicatedsinceageometricresultin1R?cannotbeextendedto
1R
3.
Therefore,ithasremainedwidelyopenuntilrecently.[Pebay-98c]proposedan
apriori
classificationfunctionandtheoremal-lowingtodecidewhetherornotanyfacewillap-pearinanyDelaunay-triangulationoftherelatedsetofvertices.Basedonthisresult,aheuristicbutreliablealgorithminordertoredefineathree-dimensionalconstrainthasbeenintroduced.Sincethen,[Murphy
etal.-OO]
provedthatatleastonetheoreticalconvergentalgorithmexists,butdidnotimplementednorevaluatedit,sinceitiscom-putationallyimpractical.Therefore,thisimpor-tanttheoreticalresultdidnotsolvetheproblemoffindingareliableaswellasusablealgorithm.
Outline
Firstofall,thecontextofconstrainedDelaunay-triangulationsisshortlyrecalledSec-tion
1.
GeometricissuesofDelaunay-conformity
in3dimensionsarediscussedSection2;sincefacesclassificationhasbeenproposedin[Pebay-98c],itisverybrieflyrecalledinthiscommunication,whileitfocusesonthree-dimensionalsegments.
In
particular,thereasonswhytwo-dimensionalsegmentsclassificationdoesnotextendnaturallyto3dimensionsaredetailedandanew
apriori
Delaunay-conformitytheoremforthesesegmentsisgiven.Basedontheseresults,anewalgorithmicapproachtoredefinetheconstraintisproposedandexemplifiedinSection3.Finally,Section4summarizesongoingandfurtherdevelopments.
1.
CONTEXT
Theaimofthissectionistoshortlyrecallafewbasicdefinitionsandresultsrequiredforfurtherunderstanding,see[Frey,George-OO]forfurtherde-tails.Given
S
C
1R
3,
finitesetofdistinctpoints,itmaybeusefultorecallthata
conformingmesh
T
ofasubset
n
c
1R
3
isapolyhedralcovering-upof
n,
sothatanyfaceofanypolyhedraof
T
iseitherafaceofanotherelementorbelongsto
an.
In
particular,a
triangulation
isaconformingmeshwhichcontainsonlysimplicialelements.
Voronoldiagram
The
Voronoi"cell
ofany
Pi
E
S
isdefinedas:The1,2and3-dimensionalintersectionsofcellsarerespectivelycalled
Voronoi"points,edges
and
faces
of
S.
Naturally,the
Voronoi"diagram
of
S,
referredas
V(S),
isthesetof
Voronoi
points,edgesandpossiblyfacesof
S.
Forthesakeofconcision,a
k-
Voronoi-face
denotesanyk-dimensionalentityofthe
Voronoi
diagram.Noticethatif
S
doesnotcontainanyquintupleofcosphericalpoints,thenthepointsaresaidtobein
regularposition.
DelaunayTriangulation
TheDelaunaytrian-gulationcanbedefinedasthegeometricaldual,inasense,ofthe
Voronoi
diagram.Morepre-cisely,if
f
isak-faceofthe
Voronoi
diagramof
S,
suchthattheonly
Voronoi
cellscontaining
f
aretheonesassociatedtothepointsoftheset
FeS,
then
F
issaidtobea(3-k)-Delaunay-faceoftheconvexhullof
F.
NotethatifafaceisDelaunay-conforming,soareitsthreeedges,while