betaFeedback

WorksMathematics

Dimensionally absolutely stable but non-diagonalizable TT-meshes

Abstract

We construct an infinite family of rectangular TT-meshes whose conformality-vector-space dimension is constant over all structurally similar geometric realizations, although the meshes are not diagonalizable. For every integer d≥2d\ge 2, the construction has 2(d+2)2(d+2) \Tl-edges and intersection graph Kd+2,d+2K_{d+2,d+2}. Every \Tl-edge contains at least d+2d+2 vertices, so no edge is vanishable under the usual vertex-count criterion for bidegree (d,d)(d,d) and smoothness (d−1,d−1)(d-1,d-1). We prove that the conformality-vector space is one-dimensional throughout the structural similarity class. Its generator is the outer product of two vectors of barycentric interpolation weights, extended by four zero coordinates. Non-diagonalizability follows by examining the last edge in any proposed diagonalizing ordering. Thus, for the geometric structural similarity class specified here, diagonalizability is not a necessary condition for dimensional absolute stability.

Introduction

The dimension of a spline space over a TT-mesh is closely related to the rank of its conformality matrix. For splines of bidegree (d,d)(d,d) and smoothness (d−1,d−1)(d - 1,d - 1), the constraints associated with a TlTl-edge can be written as polynomial moment equations. Their coefficients depend on the coordinates of the vertices on that edge. This is part of a broader literature on dimension instability and stability of spline spaces over TT-meshes; see Li and Chen [1], Berdinsky et al. [2], and Guo et al. [3]. The smoothing cofactor-conformality method is developed in Li and Deng [4]. Stability properties for highest-order smoothness spaces over TT-meshes, including the notion of dimensional absolute stability, have been studied by Huang and Chen [5, 6].

Diagonalizability is an ordering property of the TlTl-edges. It provides a combinatorial mechanism for controlling these constraints. The converse question is whether coordinate-independent dimension, even throughout a structural similarity class, forces such an ordering to exist.

This paper gives a negative answer for the geometric structural similarity class defined below. For each d≥2d \ge2, we construct a rectangular TT-mesh with no vanishable TlTl-edges whose conformality-vector-space dimension is identically one throughout that class, but for which no diagonalizing ordering exists.

The construction uses a complete bipartite intersection pattern. This pattern forces a shared set of coordinates on all edges of each orientation. The resulting compatibility equations admit a rank-one factorization involving classical barycentric interpolation weights.

To make the scope of the result precise, we explicitly state the structural similarity and stability conventions used here. In particular, the realizations under consideration are geometric collections of horizontal and vertical segments in the plane, not abstract incidence systems with independently assigned local coordinates.

Definitions and conventions

Fix an integer d≥2d \ge2. The associated spline setting is bidegree (d,d)(d,d) with smoothness (d−1,d−1)(d-1,d-1).

A rectangular TT-mesh is a partition of a rectangle into axis-aligned rectangles, allowing TT-junctions. An interior ll-edge is a maximal connected straight union of interior mesh edges. A cross-cut has both endpoints on the boundary of the domain. A TlTl-edge has both endpoints at interior TT-junctions.

For an ll-edge ℓ\ell, let Vert⁡(ℓ)\operatorname{Vert}(\ell) be its set of vertices, including its endpoints, and write

n(ℓ)=#Vert⁡(ℓ).n(\ell) = \#\operatorname{Vert}(\ell).

A TlT l-edge is called vanishable if

n(ℓ)≤d+1.n(\ell) \le d + 1.

A TT-connected component is a connected collection of TlT l-edges, where connectivity is determined by shared vertices. Within such a component, a vertex belonging to two component edges is called a multi-vertex, and a vertex belonging to exactly one component edge is called a mono-vertex.

Geometric structural similarity

The following two definitions adapt the notion of a generalized TT-connected component and the structurally similar map of Huang and Chen [6] to the planar geometric setting used in this paper.

Definition 2.1 (Geometric realization). A geometric component consists of finitely many horizontal and vertical closed line segments in R2\mathbb{R}^{2}, together with a finite set of distinct marked vertices on each segment. Endpoints and intersections of component segments are marked.

Segments of the same orientation do not share a vertex. Segments of opposite orientations meet in at most one vertex. Every marked vertex belongs to one or two component segments. The segment intersection graph is connected.

The segments in Definition 2.1 represent the ll-edges of a generalized geometric component. Additional marked vertices represent mono-vertices. The common-coordinate requirement in Definition 2.1 (same orientation segments do not share a vertex) is essential for the analysis below; it is precisely the planar geometric counterpart of the common-coordinate property discussed by Huang and Chen [6].

Definition 2.2 (Structural similarity). Two geometric components CC and C~\widetilde{C} are structurally similar if there is a bijection

ϕ:E(C)⟶E(C~)\phi:\mathcal{E}(C)\longrightarrow\mathcal{E}(\widetilde{C})

between their edge sets such that

n(ϕ(ℓ))=n(ℓ)n(\phi(\ell))=n(\ell)

for every edge ℓ\ell, and

ℓ∩ℓ′≠∅⟺ϕ(ℓ)∩ϕ(ℓ′)≠∅\ell\cap\ell' \ne\varnothing\Longleftrightarrow\phi(\ell)\cap\phi(\ell') \ne\varnothing

for every pair of distinct edges.

This is the geometric counterpart of the structurally similar map of Huang and Chen [6]: the condition ℓ∩ℓ′≠∅⟺ϕ(ℓ)∩ϕ(ℓ′)≠∅\ell\cap\ell^{\prime} \ne\varnothing\Longleftrightarrow\phi(\ell) \cap\phi(\ell^{\prime}) \ne\varnothing plays the role of the incidence-preservation condition β(ℓi∩ℓj)=β(ℓi)∩β(ℓj)\beta(\ell_i \cap\ell_j) = \beta(\ell_i) \cap\beta(\ell_j) in the original definition.

The class of all such geometric realizations is denoted by [C]S[\mathcal{C}]_{\mathrm{S}}.

Structural similarity does not prescribe numerical coordinates, distances, or the order of mono-vertices. It also permits an exchange of the two coordinate directions.

Conformality-vector spaces

Assign one real scalar γv\gamma_v to each distinct vertex vv of a component C\mathcal{C}. On a horizontal edge, let zvz_v be the horizontal coordinate of vv; on a vertical edge, let zvz_v be its vertical coordinate.

The moment conditions on an edge ℓ\ell are

∑v∈Vert⁡(ℓ)zvqγv=0,q=0,…,d.(1)\sum_{v \in\operatorname{Vert}(\ell)} z_v^q \gamma_v = 0,\qquad q = 0,\ldots,d. \tag*{(1)}

At a multi-vertex, the same variable is used in the conditions for both incident component edges.

The conformality-vector space is

CVS⁡d(C)={(γv)v∈Vert⁡(C):(1) holds on every edge}.(2)\operatorname{CVS}_d(\mathcal{C}) = \left\{(\gamma_v)_{v \in\operatorname{Vert}(\mathcal{C})} : (1)\text{ holds on every edge}\right\}. \tag*{(2)}

Equivalently, CVS⁡d(C)=ker⁡Md(C)\operatorname{CVS}_d(\mathcal{C}) = \ker M_d(\mathcal{C}), where Md(C)M_d(\mathcal{C}) is the matrix of these moment conditions.

Definition 2.3 (Dimensional absolute stability; Huang and Chen [6]). A component C\mathcal{C} is dimensionally absolutely stable if dim⁡CVS⁡d(C~)\dim\operatorname{CVS}_d(\widetilde{\mathcal{C}}) is independent of C~∈[C]S\widetilde{\mathcal{C}} \in[\mathcal{C}]_{\mathrm{S}}.

A TT-mesh with a single TT-connected component is called dimensionally absolutely stable if that component has this property.

All meshes constructed below have exactly one TT-connected component.

Diagonalizability

Definition 2.4 (Diagonalizability; Li and Deng [4], see also Huang and Chen [6], Definition 2.3). A component with edges ℓ1,…,ℓt\ell_1,\ldots,\ell_t is diagonalizable if its edges admit an ordering such that

#(Vert⁡(ℓi)\⋃j<iVert⁡(ℓj))≥d+1,i=1,…,t.(3)\#\left(\operatorname{Vert}(\ell_i) \mathbin{\backslash} \bigcup_{j<i}\operatorname{Vert}(\ell_j)\right) \ge d+1,\qquad i=1,\ldots,t. \tag*{(3)}

A TT-mesh with one TT-connected component is diagonalizable if its component is diagonalizable.

Remark 2.5 (Scope of the definitions). The common-coordinate property of a planar geometric realization is essential below. The argument does not assert stability for a different model in which each edge is assigned its own independent set of local vertex coordinates. Accordingly, applying the result to an externally formulated conjecture requires matching that conjecture’s definitions with Definitions 2.12.1–2.42.4.

A classical Vandermonde identity

We use the following standard consequence of the divided difference formula and the Lagrange interpolation formula; see de Boor [7] and Berrut and Trefethen [8]. The explicit vector consists of the barycentric interpolation weights. We state the result without proof.

Lemma 3.1 (One-dimensional Vandermonde null space). Let r≥2r \ge2, and let z1,…,zrz_1,\ldots,z_r be pairwise distinct real numbers. Set

V(z1,…,zr)=(1⋯1z1⋯zr⋮⋱⋮z1r−2⋯zrr−2)∈R(r−1)×r.(4)V(z_1,\ldots,z_r)= \begin{pmatrix} 1 & \cdots& 1\\ z_1 & \cdots& z_r\\ \vdots& \ddots& \vdots\\ z_1^{r-2} & \cdots& z_r^{r-2} \end{pmatrix} \in\mathbb{R}^{(r-1)\times r}. \tag*{(4)}

Then rank⁡V=r−1\operatorname{rank} V = r-1 and ker⁡V\ker V is one-dimensional. More explicitly, if

P(z)=∏k=1r(z−zk),P(z)=\prod_{k=1}^{r}(z-z_k),

then

wj=1P′(zj),j=1,…,r,w_j=\frac{1}{P'(z_j)},\qquad j=1,\ldots,r,

defines a generator w=(w1,…,wr)Tw=(w_1,\ldots,w_r)^{T} of ker⁡V\ker V. Every component of ww is nonzero. Consequently, every nonzero vector in ker⁡V\ker V has no zero component.

Construction and non-diagonalizability

Let

r=d+2r = d + 2

and consider the rectangle

Ω=[0,r+3]×[0,r+1].\Omega= [0, r + 3] \times[0, r + 1].

Insert the two vertical cross-cuts

CL={1}×[0,r+1],CR={r+2}×[0,r+1].C_{\mathrm{L}} = \{1\} \times[0, r + 1], \qquad C_{\mathrm{R}} = \{r + 2\} \times[0, r + 1].

Insert the rr vertical segments

Vj={j+1}×[1,r],j=1,…,r,(5)V_j = \{j + 1\} \times[1, r], \qquad j = 1, \ldots, r, \tag*{(5)}

the two horizontal segments

H1=[1,r+2]×{1},Hr=[1,r+2]×{r},(6)H_1 = [1, r + 2] \times\{1\}, \qquad H_r = [1, r + 2] \times\{r\}, \tag*{(6)}

and the remaining horizontal segments

Hi=[2,r+1]×{i},i=2,…,r−1.(7)H_i = [2, r + 1] \times\{i\}, \qquad i = 2, \ldots, r - 1. \tag*{(7)}

Together with the boundary of Ω\Omega, these segments define a TT-mesh denoted by TdT_d.

Proposition 4.1. The construction defines a rectangular TT-mesh. Its Tl-edges are precisely

H1,…,Hr,V1,…,Vr.H_1, \ldots, H_r, V_1, \ldots, V_r.

They form one TT-connected component CdC_d, whose intersection graph is Kr,rK_{r,r}. Moreover,

n(H1)=n(Hr)=r+2,n(Hi)=r,2≤i≤r−1,n(Vj)=r,1≤j≤r.(8)\begin{aligned} n(H_1) &= n(H_r) = r + 2,\\ n(H_i) &= r, && 2 \leq i \leq r - 1,\\ n(V_j) &= r, && 1 \leq j \leq r. \tag*{(8)} \end{aligned}

In particular, no Tl-edge is vanishable.

Proof. The two cross-cuts partition Ω\Omega into three vertical strips. The two outer strips are rectangles. Within the middle strip, H1H_1 and HrH_r separate a lower rectangle, an upper rectangle, and a central rectangle.

The central rectangle is partitioned by the VjV_j into two side rectangles and r−1r-1 interior columns. The segments H2,…,Hr−1H_2,\ldots,H_{r-1} subdivide those interior columns into rectangles. Thus all mesh cells are rectangles.

Each H1H_1 and HrH_r ends at TT-junctions on the cross-cuts. Each VjV_j ends at TT-junctions on H1H_1 and HrH_r. Each remaining HiH_i ends at TT-junctions on V1V_1 and VrV_r. The listed segments are maximal in their respective directions.

Every HiH_i meets every VjV_j, and no two listed edges of the same orientation meet. Hence the intersection graph is Kr,rK_{r,r}.

Every edge contains rr multi-vertices. The two edges H1H_1 and HrH_r additionally contain two mono-vertices each, at the cross-cuts. There are no other mono-vertices. The asserted counts follow.

Finally, r=d+2r=d+2, so every TℓT\ell-edge contains more than d+1d+1 vertices.

Proposition 4.2. For every d≥2d \ge2, the TT-mesh TdT_d is not diagonalizable.

Proof. Suppose that a diagonalizing ordering exists and consider its last edge ℓ\ell. All other component edges precede ℓ\ell.

If ℓ\ell is a vertical edge or one of H2,…,Hr−1H_2,\ldots,H_{r-1}, all its vertices belong to preceding edges. Thus it contributes no new vertex.

If ℓ\ell is H1H_1 or HrH_r, all its multi-vertices belong to preceding vertical edges. Only its two mono-vertices remain.

In either case the number of remaining vertices is at most two. Since d≥2d \ge2, this is less than d+1d+1, contradicting (3).

Absolute stability

The essential point is to treat every member of the structural similarity class, not merely coordinate perturbations of the displayed mesh.

Lemma 5.1 (Geometry of the similarity class). Let C~∈[Cd]S\widetilde{C} \in[C_d]_{\mathrm{S}}. After relabeling edges and, if necessary, exchanging the coordinate directions, its edges can be denoted by

H1,…,Hr,V1,…,VrH_1,\ldots,H_r,V_1,\ldots,V_r

with the following properties:

  1. every HiH_i is horizontal and every VjV_j is vertical;

  1. every HiH_i meets every VjV_j;

  2. H1H_1 and HrH_r each contain two mono-vertices;

  3. all remaining vertices are the r2r^2 intersections Hi∩VjH_i \cap V_j.

Furthermore, the supporting lines of the VjV_j have pairwise distinct horizontal coordinates xjx_j, and the supporting lines of the HiH_i have pairwise distinct vertical coordinates yiy_i.

Proof. By Proposition 4.1, the edge-incidence graph of CdC_d is Kr,rK_{r,r}. Structural similarity preserves the edge-incidence relation, so C~\widetilde{C} has the same incidence graph Kr,rK_{r,r}. Since Kr,rK_{r,r} is connected, its bipartition is unique up to exchanging the two parts. In an axis-aligned geometric realization, the horizontal and vertical orientation classes give such a bipartition; hence, after possibly exchanging the coordinate directions, the two partite sets are precisely {H1,…,Hr}\{H_1,\ldots,H_r\} (horizontal) and {V1,…,Vr}\{V_1,\ldots,V_r\} (vertical). This proves properties 1 and 2.

Structural similarity also preserves the number of vertices on every corresponding edge. By (8), in CdC_d the two edges H1,HrH_1,H_r have r+2r+2 vertices while all other edges have rr vertices. Hence the same vertex counts hold in C~\widetilde{C}. Every edge meets exactly rr edges of the opposite orientation, and each such intersection is a distinct vertex, so every edge has exactly rr multi-vertices. An edge with rr vertices therefore has no mono-vertex, whereas an edge with r+2r+2 vertices has exactly two mono-vertices. This proves properties 3 and 4.

Finally, by Definition 2.1, two same-orientation segments do not share a vertex. If two vertical edges had the same horizontal coordinate, then their intersections with any horizontal edge would coincide, making them share a vertex, a contradiction. Hence the xjx_j are pairwise distinct. The same argument applies to the yiy_i. □

The labels H1H_1 and HrH_r in Lemma 5.1 identify the two special edges; they need not be the geometrically lowest and highest edges of an arbitrary realization.

Theorem 5.2 (Constant conformality-vector-space dimension). For every C~∈[Cd]S\widetilde{C} \in[C_d]_S,

dim⁡CVS⁡d(C~)=1.\dim\operatorname{CVS}_d(\widetilde{C}) = 1.

More explicitly, let

Px(z)=∏j=1r(z−xj),Py(z)=∏i=1r(z−yi),P_x(z)=\prod_{j=1}^{r}(z-x_j), \qquad P_y(z)=\prod_{i=1}^{r}(z-y_i),

and define

ai=1Py′(yi),bj=1Px′(xj).a_i = \frac{1}{P'_y(y_i)}, \qquad b_j = \frac{1}{P'_x(x_j)}.

Then every conformality vector has intersection coordinates

γij=μaibj,1≤i,j≤r,(9)\gamma_{ij} = \mu a_i b_j, \qquad1 \le i,j \le r, \tag*{(9)}

for some μ∈R\mu\in\mathbb{R}, and all four mono-vertex coordinates are zero.

Proof. Use the labeling from Lemma 5.1. Set

X=V(x1,…,xr),Y=V(y1,…,yr).X = V(x_1,\ldots,x_r), \qquad Y = V(y_1,\ldots,y_r).

Since d=r−2d = r - 2, these are exactly the moment matrices on edges having no mono-vertices.

By Lemma 3.1,

ker⁡X=span⁡{b},ker⁡Y=span⁡{a},\ker X = \operatorname{span}\{b\}, \qquad\ker Y = \operatorname{span}\{a\},

and every component of aa and bb is nonzero.

Let γij\gamma_{ij} denote the coordinate associated with Hi∩VjH_i \cap V_j. Each VjV_j has only its rr intersection vertices. Its moment conditions therefore give

(γ1j,…,γrj)T∈ker⁡Y.(\gamma_{1j},\ldots,\gamma_{rj})^T \in\ker Y.

Consequently, for some λj∈R\lambda_j \in\mathbb{R},

γij=aiλj.\gamma_{ij} = a_i\lambda_j.

Writing λ=(λ1,…,λr)T\lambda= (\lambda_1,\ldots,\lambda_r)^T, the intersection-coordinate matrix is aλTa\lambda^T.

There is an ordinary horizontal edge, for example H2H_2. Its moment conditions yield

0=X(a2λ)=a2Xλ.0 = X(a_2\lambda) = a_2X\lambda.

Since a2≠0a_2 \ne0, it follows that λ∈ker⁡X\lambda\in\ker X. Hence λ=μb\lambda= \mu b for some μ∈R\mu\in\mathbb{R}, which proves (9). All remaining ordinary horizontal equations are then automatically satisfied.

Consider either special horizontal edge HsH_s, where s∈{1,r}s \in\{1,r\}. Let its mono-vertices have horizontal coordinates psp_s and qsq_s, with corresponding conformality coordinates ηs\eta_s and θs\theta_s. These are distinct vertices on a horizontal line, so ps≠qsp_s \ne q_s. For every k=0,…,dk = 0, \ldots, d, the contribution of its intersection vertices is

∑j=1rxjkγsj=μas∑j=1rxjkbj=0.\sum_{j=1}^{r} x_j^k \gamma_{sj} = \mu a_s \sum_{j=1}^{r} x_j^k b_j = 0.

Thus the remaining conditions are

pskηs+qskθs=0,k=0,…,d.p_s^k\eta_s + q_s^k\theta_s = 0,\qquad k = 0,\ldots,d.

Taking k=0,1k = 0,1 gives

(11psqs)(ηsθs)=0.\begin{pmatrix} 1 & 1 \\ p_s & q_s \end{pmatrix} \begin{pmatrix} \eta_s \\ \theta_s \end{pmatrix} = 0.

The matrix is nonsingular because ps≠qsp_s \ne q_s, so ηs=θs=0\eta_s = \theta_s = 0.

We have shown that every conformality vector has the asserted form. Conversely, (9), extended by four zero mono-vertex coordinates, satisfies every moment condition because Xa=0Xa = 0 is not required: the relevant identities are Xb=0Xb = 0 and Ya=0Ya = 0. In particular, the horizontal equations use Xb=0Xb = 0 and the vertical equations use Ya=0Ya = 0.

Taking μ=1\mu= 1 gives a nonzero vector, and the solution space is therefore exactly one-dimensional.

Remark 5.3. With a fixed vectorization convention, the full space is

CVS⁡d(C~)=span⁡{(vec⁡(abT),0,0,0,0)}.\operatorname{CVS}_d(\widetilde{\mathcal C}) = \operatorname{span}\left\{ \left(\operatorname{vec}(ab^{T}),0,0,0,0\right) \right\}.

The four final entries correspond to the mono-vertices.

Theorem 5.4 (An infinite counterexample family). For every d≥2d \ge2, the rectangular TT-mesh Td\mathcal T_d has no vanishable TlTl-edges, is dimensionally absolutely stable in the sense of Definition 2.3, and is not diagonalizable.

Proof. Proposition 4.1 gives the vertex-count condition. Theorem 5.2 gives dimensional absolute stability, and Proposition 4.2 gives non-diagonalizability.

Corollary 5.5 (Rank of the conformality matrix). For every C~∈[Cd]S\widetilde{\mathcal C} \in[\mathcal C_d]_S, the matrix Md(C~)M_d(\widetilde{\mathcal C}) has 2r(r−1)2r(r-1) rows, r2+4r^2 + 4 columns, and rank

rank⁡Md(C~)=r2+3.\operatorname{rank} M_d(\widetilde{\mathcal C}) = r^2 + 3.

Its row-rank deficiency is

2r(r−1)−(r2+3)=(r−3)(r+1).2r(r-1) - (r^2 + 3) = (r-3)(r+1).

Proof. There are 2r2r edges, each contributing d+1=r−1d+1=r-1 moment equations. There are r2r^2 multi-vertices and four mono-vertices. The rank follows from Theorem 5.2 and the rank-nullity theorem. □

The lowest-degree example in the family

For d=2d=2, we have r=4r=4 and

Ω=[0,7]×[0,5].\Omega=[0,7]\times[0,5].

The cross-cuts are x=1x=1 and x=6x=6. The vertical TlTl-edges are

Vj={j+1}×[1,4],j=1,2,3,4,V_j=\{j+1\}\times[1,4], \qquad j=1,2,3,4,

and the horizontal TlTl-edges are

H1=[1,6]×{1},H4=[1,6]×{4},H_1=[1,6]\times\{1\}, \qquad H_4=[1,6]\times\{4\},
H2=[2,5]×{2},H3=[2,5]×{3}.H_2=[2,5]\times\{2\}, \qquad H_3=[2,5]\times\{3\}.

The relevant counts are

n(H1)=n(H4)=6,n(H2)=n(H3)=n(Vj)=4.n(H_1)=n(H_4)=6, \qquad n(H_2)=n(H_3)=n(V_j)=4.

Thus there are eight TlTl-edges and twenty component vertices. The conformality matrix has size 24×2024\times20 and rank 19.

Figure 1. The TT-mesh T2\mathcal{T}_2. Filled circles are multi-vertices of its TT-connected component. Open circles are its four mono-vertices. The two full-height interior lines are cross-cuts.

Since both coordinate sets are equally spaced with unit spacing, one may use

u=(−1,3,−3,1)Tu = (-1,3,-3,1)^{T}

as a generator of each relevant Vandermonde null space. Accordingly, an intersection-coordinate generator is

uuT=(1−33−1−39−933−99−3−13−31),uu^{T} = \begin{pmatrix} 1 & -3 & 3 & -1 \\ -3 & 9 & -9 & 3 \\ 3 & -9 & 9 & -3 \\ -1 & 3 & -3 & 1 \end{pmatrix},

extended by four zero mono-vertex coordinates.

This is the lowest-degree member of the present family. No claim is made that it minimizes the number of edges or vertices among all possible counterexamples.

Discussion

The example distinguishes constant rank from the existence of a diagonalizing ordering. Its conformality matrix has coordinate-independent rank throughout the specified structural similarity class, despite having unavoidable row dependencies.

The complete bipartite intersection pattern is central. It forces every vertical edge to meet the same set of horizontal supporting lines and every horizontal edge to meet the same set of vertical supporting lines. The vertical constraints first force the intersection matrix to have the form aλTa\lambda^{T}. One ordinary horizontal edge then forces λ∈span⁡{b}\lambda\in\operatorname{span}\{\mathbf{b}\}. The remaining four mono-vertex variables vanish.

The failure of diagonalizability has a separate combinatorial cause. Every component edge has fewer than d+1d+1 mono-vertices. Thus no edge can occupy the last position of a diagonalizing ordering.

The restriction d≥2d \ge2 is necessary for this last-edge argument. For d=1d=1, two remaining mono-vertices meet the threshold d+1=2d+1=2, and the argument no longer applies.

Under Definitions 2.1–2.4, the implication

dimensionally absolutely stable⟹diagonalizable\text{dimensionally absolutely stable} \Longrightarrow\text{diagonalizable}

is therefore false. Any application to another notion of generalized component must first verify the common-coordinate requirement described in Remark 2.5.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability

No external datasets were used. All constructions and calculations needed to establish the results are included in the article.

References

  1. [1]X. Li, F. Chen, On the instability in the dimension of spline spaces over T-meshes, Computer Aided Geometric Design 28 (7) (2011) 420–426.DOI
  2. [2]D. Berdinsky, M. Oh, T. Kim, B. Mourrain, On the problem of instability in the dimension of a spline space over a T-mesh, Computers & Graphics 36 (5) (2012) 507–513.DOI
  3. [3]Q. Guo, R. Wang, C. Li, On the problem of instability in the dimensions of spline spaces over T-meshes with T-cycles, Journal of Computational Mathematics 33 (3) (2015) 248–262.
  4. [4]X. Li, J. Deng, On the dimension of spline spaces over T-meshes with smoothing cofactor-conformality method, Computer Aided Geometric Design 41 (2016) 76–86. doi:10.1016/j.cagd.2015.12.002.DOI
  5. [5]B. Huang, F. Chen, On the stability of the dimensions of spline spaces with highest order of smoothness over T-meshes, Journal of Computational and Applied Mathematics 441 (2024) 115681. doi:10.1016/j.cam.2023.115681.DOI
  6. [6]B. Huang, F. Chen, A preliminary study on the dimensional stability classification of polynomial spline spaces over T-meshes, Computer Aided Geometric Design 130 (2026) 102609. arXiv:2508.06217v2, doi:10.1016/j.cagd.2026.102609.DOI
  7. [7]C. de Boor, Divided differences, Surveys in Approximation Theory 1 (2005) 46–69.arxiv.org/abs/math/0502036
  8. [8]J.-P. Berrut, L. N. Trefethen. Barycentric Lagrange interpolation. SIAM Review 46 (3) (2004) 501–517. doi:10.1137/S0036144502417715.DOI

Paper details

vlnc:W_34ckswum
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents