Professional Documents
Culture Documents
Annalisa Bua
IGA: local renement via T-splines Why it is non trivial On the denition of T-splines Reduced Regularity T-splines Analysis Suitable T-splines Construction of a projection and approximation properties
Journes Lions-Magenes e
2 / 31
Journes Lions-Magenes e
3 / 31
What is IGA?
Hughes, Cottrell, Bazilevs, CMAME, 2005
Journes Lions-Magenes e
4 / 31
What is IGA?
Hughes, Cottrell, Bazilevs, CMAME, 2005
IGA: compute the solution of PDEs with NURBS/Splines on NURBS geometries. Non Uniform Rational B-Splines F() =
i
Ci Ri ()
Journes Lions-Magenes e
4 / 31
What is IGA?
Hughes, Cottrell, Bazilevs, CMAME, 2005
Journes Lions-Magenes e
4 / 31
Not so easy...
Today for IGA to work, we need a geometry described in terms of patches, and each patch have to have a trivariate NURBS description.
Journes Lions-Magenes e
5 / 31
Not so easy...
Today for IGA to work, we need a geometry described in terms of patches, and each patch have to have a trivariate NURBS description. CADs descriptions are based on trimmed NURBS patches and are often leaking CAD systems do not have trivariate representations of geometries NURBS and B-splines have a tensor product structure...
Journes Lions-Magenes e
5 / 31
Not so easy...
Today for IGA to work, we need a geometry described in terms of patches, and each patch have to have a trivariate NURBS description. CADs descriptions are based on trimmed NURBS patches and are often leaking CAD systems do not have trivariate representations of geometries NURBS and B-splines have a tensor product structure... But,
Journes Lions-Magenes e
5 / 31
Not so easy...
Today for IGA to work, we need a geometry described in terms of patches, and each patch have to have a trivariate NURBS description. CADs descriptions are based on trimmed NURBS patches and are often leaking CAD systems do not have trivariate representations of geometries NURBS and B-splines have a tensor product structure... But, IGA is a big step towards interoperability:
Analysis-aware modeling (E. Cohen et al. 2010 ) Trivariate representation is possible for targeted applications (biomedical, rubber industry, car industry...) Use of local rened splines for both modeling and analysis: T-splines, LR splines, Hierarchical splines ...
Journes Lions-Magenes e
5 / 31
Not so easy...
Today for IGA to work, we need a geometry described in terms of patches, and each patch have to have a trivariate NURBS description. CADs descriptions are based on trimmed NURBS patches and are often leaking CAD systems do not have trivariate representations of geometries NURBS and B-splines have a tensor product structure... But, IGA is a big step towards interoperability:
Analysis-aware modeling (E. Cohen et al. 2010 ) Trivariate representation is possible for targeted applications (biomedical, rubber industry, car industry...) Use of local rened splines for both modeling and analysis: T-splines, LR splines, Hierarchical splines ...
B-splines k-renement
C p1
6 / 31
{Bi }i=1,...,N0
Journes Lions-Magenes e
7 / 31
{Bi }i=1,...,N0
Journes Lions-Magenes e
7 / 31
{Bi }i=1,...,N0
or F = generalized B-spline
Journes Lions-Magenes e
7 / 31
{Bi }i=1,...,N1
The geometry and its parametrization F is given by CAD The discrete space on is the push-forward of Spline/NURBS on the parametric domain.
Journes Lions-Magenes e
7 / 31
{Bi }i=1,...,N2
The geometry and its parametrization F is given by CAD The discrete space on is the push-forward of Spline/NURBS on the parametric domain.
Journes Lions-Magenes e
7 / 31
{Bi }i=1,...,N2
The geometry and its parametrization F is given by CAD The discrete space on is the push-forward of Spline/NURBS on the parametric domain. The geometry and its parametrization F does not change during mesh renement
A. Bua (IMATI-CNR Italy) Journes Lions-Magenes e 7 / 31
Multipatch domains
If geometries are described by mappings of several patches.
F1
1
F2
F3
Patch interfaces are normally treated just imposing C 0 . . . C p/2 regularity domain decomposition type structure
A. Bua (IMATI-CNR Italy) Journes Lions-Magenes e 8 / 31
single patch:
multi-patch:
Courtesy by T.J.R Hughes et al.
Journes Lions-Magenes e
9 / 31
Let Sh /Nh denote the space of mapped splines/NURBS of degree less or equal to p in each space direction and global regularity C ( p 1)
Journes Lions-Magenes e
10 / 31
Let Sh /Nh denote the space of mapped splines/NURBS of degree less or equal to p in each space direction and global regularity C ( p 1) Approximation properties in Sobolev spaces There exist two operators S : H () Sh and N : H () Nh such that for all u H (), t < p + 1 u S u where h is the mesh size.
Hs
+ u N u
Hs
< Ch
Journes Lions-Magenes e
10 / 31
Let Sh /Nh denote the space of mapped splines/NURBS of degree less or equal to p in each space direction and global regularity C ( p 1) Approximation properties in Sobolev spaces There exist two operators S : H () Sh and N : H () Nh such that for all u H (), t < p + 1 u S u where h is the mesh size. C (p )(
s) ... Hs
+ u N u
Hs
< Ch
p1 . 2
Journes Lions-Magenes e
10 / 31
Let Sh /Nh denote the space of mapped splines/NURBS of degree less or equal to p in each space direction and global regularity C ( p 1) Approximation properties in Sobolev spaces There exist two operators S : H () Sh and N : H () Nh such that for all u H (), t < p + 1 u S u where h is the mesh size. C (p )(
s) ... Hs
+ u N u
Hs
< Ch
p1 . 2
Indeed, this means a gain for higher regularity in terms of degrees of freedoms.
A. Bua (IMATI-CNR Italy) Journes Lions-Magenes e 10 / 31
Spline basis are not interpolatory, of course S is constructed by dual functionals. E.g., cubic splines: i = [si2 , si1 , si , si4 , si5 ] The 1D projector is S u =
i
Bi (i ) spline i (i ) dual
i (Bj ) = ij
i (u)Bi (s).
Journes Lions-Magenes e
11 / 31
Spline basis are not interpolatory, of course S is constructed by dual functionals. E.g., cubic splines: i = [si2 , si1 , si , si4 , si5 ] The 1D projector is S u =
i
Bi (i ) spline i (i ) dual
i (Bj ) = ij
i (u)Bi (s).
Journes Lions-Magenes e
11 / 31
Journes Lions-Magenes e
12 / 31
T-splines
Sederberg, et al, 2003, Sederberg, Cardon, Finnigan, North, Zheng, Lyche, 2004
Journes Lions-Magenes e
13 / 31
T-splines
Sederberg, et al, 2003, Sederberg, Cardon, Finnigan, North, Zheng, Lyche, 2004
Journes Lions-Magenes e
13 / 31
T-splines
Sederberg, et al, 2003, Sederberg, Cardon, Finnigan, North, Zheng, Lyche, 2004
Journes Lions-Magenes e
13 / 31
Figure 2: Car door typically, a large number of NURBS control points serve no purpose other than to satisfy topological constraints. They carry no Figure 2 shows anoth signicant geometric information. In Figure 1.a, all the red NURBS Sederberg, et al, 2003, Sederberg, Cardon, Finnigan, North, Zheng, Lyche, 2004 points in a NURBS are control points are, in this sense, superuous. model is geometrically only 1/3 as many contr
T-splines
Figure 1: Head modeled (a) as a NURBS with 4712 control points Sederberg et al 2004 and (b) as a T-spline with 1109 control points. The red NURBS control points are superuous.
Figure 3: NURB
Journes Lions-Magenes e
Superuous control not merely because they but also because they c as can be seen by comp Figure 3.a with that of can waste dozens of ho 13 points NURBS control / 31
T-splines
Sederberg, et al, 2003, Sederberg, Cardon, Finnigan, North, Zheng, Lyche, 2004 ref Renement strategy which ensures Sh Sh Possible diculties:
Linear independance Possible severe ll-in of the T-mesh Expensive (cycle on many elements) and not local There is no well dened de-renement strategy
a. B-spline surfaces
A. Bua (IMATI-CNR Italy)
b. T-splines
Journes Lions-Magenes e
T-splines by in idea, that we c splines are the plained in 5. T two or more B 6. NURCCs ar inition 13 /cubi of 31
T-splines
T-meshes can be very general partitions:
Journes Lions-Magenes e
14 / 31
We have: 1 = 2 + 1 3 !!! 3
Journes Lions-Magenes e
15 / 31
Journes Lions-Magenes e
16 / 31
Journes Lions-Magenes e
16 / 31
Journes Lions-Magenes e
16 / 31
by split them in 4
Journes Lions-Magenes e
17 / 31
C 2 basis functions
Journes Lions-Magenes e
18 / 31
C 2 basis functions
Journes Lions-Magenes e
19 / 31
Journes Lions-Magenes e
20 / 31
Journes Lions-Magenes e
20 / 31
ination ination DD DEGREE CASE: M M M E ODD DEGREE CASE: M T-splines Reduced regularity anchor at aat a2011of M M vertex one anchor B. Cho Kumar vertex of T-spline associated with the anchor one T-spline associated with the anchor
M, p =p3= 3 M,
M, M, p =p3= 3
ICES, ICES, Austin TX,2011 2011 / 187 / 18 Austin TX, Jan. Jan. 7
Journes Lions-Magenes e
21 / 31
ination E ODD DEGREE CASE: ination M M DD DEGREE CASE: M M Reduced regularity T-splines one anchoraat a2011of M M anchor at vertex B. Cho Kumar vertex of one T-spline associated with the anchor T-spline associated with the anchor
( + 1)2 T-splines attached to each vertex v on M, which are denoted v 2 by k 1 k ( construct C 1 cubic T-splines, by counting all knots 2 GivenBa ,T-mesh, we + 1)
times.
M, p =p3= 3 M,
M, M, p =p3= 3
ICES, ICES, Austin TX,2011 2011 / 187 / 18 Austin TX, Jan. Jan. 7
Journes Lions-Magenes e
21 / 31
ination E ODD DEGREE CASE: ination M M DD DEGREE CASE: M M Reduced regularity T-splines one anchoraat a2011of M M anchor at vertex B. Cho Kumar vertex of one T-spline associated with the anchor T-spline associated with the anchor
( + 1)2 T-splines attached to each vertex v on M, which are denoted v 2 by k 1 k ( construct C 1 cubic T-splines, by counting all knots 2 GivenBa ,T-mesh, we + 1)
times.
M, M, p =p3= 3
Journes Lions-Magenes e
21 / 31
Regular T-meshes
Idea: Restrict the class of reasonable T-meshes. We construct one-bay T-junction extensions:
one-bay
Journes Lions-Magenes e
22 / 31
Regular T-meshes
Idea: Restrict the class of reasonable T-meshes. We construct one-bay T-junction extensions:
one-bay
Denition We say that a T-mesh is regular if (closed) bay extensions do not intersect.
Journes Lions-Magenes e
22 / 31
Regular T-meshes
Idea: Restrict the class of reasonable T-meshes. We construct one-bay T-junction extensions:
Denition We say that a T-mesh is regular if (closed) bay extensions do not intersect.
Journes Lions-Magenes e
22 / 31
Regular T-meshes
A lot can be proved for reduced regularity splines... Characterization in terms of piecewise polynomials : if p is odd and bay extensions are away from the boundary S(p, M) = C (p, M) = {u C (p1)/2 : u|Q Qp Q M} Topological characterization of the embedding : M1 M2 and M1 M2 S(p, M1 ) S(p, M2 )
Journes Lions-Magenes e
23 / 31
Regular T-meshes
A lot can be proved for reduced regularity splines... Characterization in terms of piecewise polynomials : if p is odd and bay extensions are away from the boundary S(p, M) = C (p, M) = {u C (p1)/2 : u|Q Qp Q M} Topological characterization of the embedding : M1 M2 and M1 M2 S(p, M1 ) S(p, M2 ) So, looking at the mesh, we know if the spaces are embedded and if they contain the polynomials we want !
Journes Lions-Magenes e
23 / 31
Renement Algorithm
1. Start with a tensor product mesh 2. Mark elements for renement and rene 3. Detect intersections of bay extensions 4. Add to the mesh one of intersecting extensions and go back to 3. The algorithm stays local and seems to work well.
Journes Lions-Magenes e
24 / 31
S(p, M0 )
S(p, M1 )
S(p, M2 )
S(p, M3 )
Journes Lions-Magenes e
25 / 31
(d) The second T-mesh (a) The starting T-mesh (b) Initial solution
S(p, M0 )
(d) The second T-mesh The(e) Rened solut (g) third T-mesh (a) The starting T-mesh (b) Initial solution with 1806 DOF (c) E DOF
S(p, M1 )
S(p, M2 )
S(p, M3 )
Journes Lions-Magenes e
25 / 31
(a) A T-mesh M.
Journes Lions-Magenes e
26 / 31
(a) A T-mesh M.
(b) Extensions on M.
Journes Lions-Magenes e
26 / 31
(a) A T-mesh M.
(b) Extensions on M.
Denition An AS T-mesh is one on which no horizontal T-junction extension intersects a vertical T-junction extension AS T-splines are dened on AS T-meshes
A. Bua (IMATI-CNR Italy) Journes Lions-Magenes e 26 / 31
An AS T-mesh M Denition An AS T-mesh is one on which no horizontal T-junction extension intersects a vertical T-junction extension AS T-splines are dened on AS T-meshes
A. Bua (IMATI-CNR Italy) Journes Lions-Magenes e 26 / 31
Journes Lions-Magenes e
27 / 31
Journes Lions-Magenes e
27 / 31
Journes Lions-Magenes e
27 / 31
Journes Lions-Magenes e
27 / 31
Properties of AS T-splines: Positivity. Linear independence. Relatively local renement (numerics) Partition of unity. = ane invariance.
Journes Lions-Magenes e
27 / 31
Properties of AS T-splines: Positivity. Linear independence. Relatively local renement (numerics) Partition of unity. = ane invariance. Positivity and Partition of unity = convex hull.
Journes Lions-Magenes e
27 / 31
Properties of AS T-splines: Positivity. Linear independence. Relatively local renement (numerics) Partition of unity. = ane invariance. Positivity and Partition of unity = convex hull. All proofs are based on topological properties of the mesh... what about function spaces? Characterization in terms of piecewise polynomials Projections and approximation properties
Journes Lions-Magenes e
27 / 31
A property of AS T-splines
Beirao-da-Veiga, B., Cho, Sangalli 2011
Given a AS T-mesh M, we say that two T-splines partially overlap if their knot vectors come from a common knot vector at least in one direction.
Journes Lions-Magenes e
28 / 31
A property of AS T-splines
Beirao-da-Veiga, B., Cho, Sangalli 2011
Given a AS T-mesh M, we say that two T-splines partially overlap if their knot vectors come from a common knot vector at least in one direction. DO NOT overlap
Journes Lions-Magenes e
28 / 31
A property of AS T-splines
Beirao-da-Veiga, B., Cho, Sangalli 2011
Given a AS T-mesh M, we say that two T-splines partially overlap if their knot vectors come from a common knot vector at least in one direction. DO NOT overlap
DO overlap
Journes Lions-Magenes e
28 / 31
A property of AS T-splines
Beirao-da-Veiga, B., Cho, Sangalli 2011
Given a AS T-mesh M, we say that two T-splines partially overlap if their knot vectors come from a common knot vector at least in one direction. DO NOT overlap
DO overlap
Theorem A T-mesh M is AS if and only if is each couple of T-splines dened on M partially overlaps. The proof is not trivial . . .
A. Bua (IMATI-CNR Italy) Journes Lions-Magenes e 28 / 31
t and
And now it is trivial to see 1 ,1 (B2 ,2 ) = 1 ,2 1 ,2 if the two T-splines B1 ,1 and B2 ,2 (and thus their knot vectors) partially overlap.
Journes Lions-Magenes e
29 / 31
t and
And now it is trivial to see 1 ,1 (B2 ,2 ) = 1 ,2 1 ,2 if the two T-splines B1 ,1 and B2 ,2 (and thus their knot vectors) partially overlap. Main Theorem On a AS T-mesh, the functionals { , ( ) T-node} form a local dual , , basis to the set of T-splines.
A. Bua (IMATI-CNR Italy) Journes Lions-Magenes e 29 / 31
Let S(M) be the span of all T-splines on M, then: there exists a local stable projection : L2 ((0, 1)2 ) S(M); if 1 S(M), the set of T-splines forms a partition of unity; if the global cubic polynomial belongs to the space, then: ||f [f ]||L2 (Q) C (hQ )r |f |H r (Q) e e Q M, f H r ([0, 1]2 ).
Journes Lions-Magenes e
30 / 31
More on AS T-splines
AS T-splines can now be extended in several space dimensions Error estimates in the physical domain is feasible (in progress)
Journes Lions-Magenes e
31 / 31
More on AS T-splines
AS T-splines can now be extended in several space dimensions Error estimates in the physical domain is feasible (in progress) An ecient local renement has still to be understood..
Journes Lions-Magenes e
31 / 31
More on AS T-splines
AS T-splines can now be extended in several space dimensions Error estimates in the physical domain is feasible (in progress) An ecient local renement has still to be understood.. Error indicators should be adapted to regular functions.
Journes Lions-Magenes e
31 / 31
More on AS T-splines
AS T-splines can now be extended in several space dimensions Error estimates in the physical domain is feasible (in progress) An ecient local renement has still to be understood.. Error indicators should be adapted to regular functions.
Journes Lions-Magenes e
31 / 31