Sharp Dimension Bounds for Spline Spaces over T-meshes with Highest Order of Smoothness

Abstract: The dimension of a polynomial spline space of bi-degree $(d_1,d_2)$ over a T-mesh $\mathscr{T}$ with the highest order of smoothness $(d_1-1,d_2-1)$ depends on both mesh topology and geometric configurations. Under the assumption that the T-connected components of the T-mesh $\mathscr{T}$ contain no vanishable T $l$-edges, we develop explicit upper and lower bounds of the dimension of the polynomial spline space. By introducing a decoupling technique within the completely non-diagonalizable component (CNDC) of the T-mesh $\mathscr{T}$, we separate tightly coupled multi-vertex constraints and transform global conformality conditions into localized linear equations along each interior large edge. Based on the decoupling technique, a new dimension formula of the polynomial spline space is then presented, and from which sharp upper and lower bounds of the dimension are obtained. The bounds are sharp in the sense that different geometric realizations of T-meshes with the same topology can attain the lower and upper bounds for the dimension of the polynomial spline space. We further prove that the new formula is consistent with Mourrain's homological dimension formula, and a sharper lower bound is obtained by our method.

• The paper introduces a decoupling method that replaces the globally coupled conformality matrix with independent per-edge Vandermonde solves and a smaller condensed coupling matrix K.
• The resulting formula expresses spline-space dimension through rank(K), while combinatorial rank bounds and an O(s2^s) dynamic program provide sharp upper and lower estimates.
• The bounds are provably sharp across coordinate realizations of fixed mesh topology, and the condensed correction m−rank(K) is shown to equal Mourrain’s homological correction term.

Background and motivation

The paper studies the dimension of the bivariate polynomial spline space $S_d(\mathscr{T})$ of bi-degree $(d,d)$ with smoothness order $(d-1,d-1)$ — the highest possible order of smoothness — over a T-mesh $\mathscr{T}$. In this regime, the dimension is known to be unstable: it depends not only on mesh topology but also on the exact geometric coordinates of grid lines (2608.19839). This instability is an obstacle for applications such as isogeometric analysis and locally refinable spline constructions (T-splines, hierarchical B-splines, LR-splines, PHT-splines), all of which can be viewed as representations of polynomial splines over T-meshes.

The smoothing cofactor method reduces the dimension problem to computing the rank of a global conformality matrix $M(T(\mathscr{T}))$ attached to the T-connected component of the mesh. Prior work by Huang and Chen decomposed this component into a diagonalizable part (whose contribution to the dimension is stable and explicitly computable) and a completely non-diagonalizable component (CNDC), which carries all geometric instability. The present paper's contribution is a decoupling technique that makes the CNDC analysis tractable in general.

The decoupling framework

The central algebraic difficulty is that each multi-vertex (an intersection of two T $l$-edges) has a single smoothing cofactor constrained simultaneously by horizontal and vertical conformality conditions, destroying block structure. The authors split each multi-vertex cofactor $\delta_i$ into independent horizontal and vertical variables $(\delta_i^h,\delta_i^v)$. This yields:

• A decoupled conformality matrix $K_1$, block-diagonal with $s$ independent Vandermonde-like blocks of full row rank $(d,d)$0, whose null space can be written analytically via Lagrange basis polynomials on each edge.
• A coupling matrix $(d,d)$1 enforcing $(d,d)$2, and a condensed coupling matrix $(d,d)$3 of size $(d,d)$4, where $(d,d)$5 is the number of multi-vertices.

The main structural result is the rank identity

$(d,d)$6

which leads to the new dimension formula

$(d,d)$7

where $(d,d)$8 is the number of cross-cuts, $(d,d)$9 the total number of T $(d-1,d-1)$0-edges, and $(d-1,d-1)$1 the number of interior vertices. Notably, the CNDC edge count $(d-1,d-1)$2 cancels out entirely; the entire geometric instability is compressed into the small matrix $(d-1,d-1)$3. Computationally, the framework replaces expensive global row eliminations on the monolithic matrix by parallelizable per-edge analytic solves plus one small reduced system — a substantial algorithmic improvement over prior case-by-case methods.

Sharp upper and lower bounds

Two bounds follow from trivial rank estimates and a combinatorial construction on $(d-1,d-1)$4:

• Lower bound: bounding $(d-1,d-1)$5 above by its dimensions gives

$(d-1,d-1)$6

shown to be algebraically identical to the earlier bound obtained via the reduced multi-vertex matrix $(d-1,d-1)$7.

• Upper bound: the key combinatorial theorem establishes

$(d-1,d-1)$8

where $(d-1,d-1)$9 ranges over permutations of the CNDC edges and $\mathscr{T}$0 counts forward multi-vertex intersections. Via an equivalent backward-intersection formulation, this yields

$\mathscr{T}$1

with $\mathscr{T}$2. A dynamic program over subsets of edges computes $\mathscr{T}$3 in $\mathscr{T}$4 rather than $\mathscr{T}$5, and a simpler directional bound (ordering all horizontal before vertical edges or vice versa) is also given.

Sharpness is demonstrated constructively: for the running example with $\mathscr{T}$6, $\mathscr{T}$7, $\mathscr{T}$8, the gap between bounds is exactly 1, and two explicit coordinate realizations of the same topology attain $\mathscr{T}$9 and $M(T(\mathscr{T}))$0 respectively (e.g., uniform coordinates give $M(T(\mathscr{T}))$1; perturbing one coordinate gives $M(T(\mathscr{T}))$2). Thus neither bound can be improved without using additional geometric information.

Consistency with Mourrain's homological formula

A significant portion of the paper reconciles these results with Mourrain's homological dimension formula. The combinatorial parts are shown to coincide via the relative Euler relation and the identity $M(T(\mathscr{T}))$3. More substantively, the homological correction term satisfies

$M(T(\mathscr{T}))$4

so the homological correction and the residual cofactor correction are the same numerical invariant of a fixed realization, not merely analogous quantities. Moreover, Mourrain's optimized ordering correction over all $M(T(\mathscr{T}))$5 T $M(T(\mathscr{T}))$6-edges equals the minimum over permutations restricted to the $M(T(\mathscr{T}))$7 CNDC edges only, so the upper bound here coincides exactly with Mourrain's, while the lower bound $M(T(\mathscr{T}))$8 strictly refines his $M(T(\mathscr{T}))$9 by the term $l$0 whenever it is positive. The practical advantage is that the ordering optimization is confined to the unique component where rank variation can occur.

Limitations and open questions

The results carry a standing assumption: no vanishable T $l$1-edges exist, i.e., every T $l$2-edge contains at least $l$3 vertices. Edges violating this contribute no degrees of freedom and must be handled separately. The sharpness claim concerns attainability across different geometric realizations of a fixed topology, not equality for every realization; determining the actual dimension for a specific mesh still requires evaluating $l$4, and the exact value can vary with coordinates. The framework is developed only for bi-degree $l$5 with highest-order smoothness $l$6; extension to lower or mixed smoothness orders, where continuity constraints couple more complexly, remains open, as does generalization to trivariate splines over 3D T-meshes. Finally, the paper does not yet exploit the localized equations to construct stable locally supported bases over non-diagonalizable meshes, which would be needed for direct use in isogeometric analysis.

Conclusion

This paper provides a decoupling-based reformulation of the highest-order-smoothness spline dimension problem over T-meshes, reducing a globally coupled conformality system to independent per-edge Vandermonde solves plus a small condensed compatibility matrix $l$7. It delivers explicit lower and upper bounds that are provably sharp under coordinate perturbations, an efficient combinatorial procedure for the upper bound, and a precise identification of the homological correction term with the nullity of $l$8, together with a refined lower bound relative to Mourrain's formula. The remaining open problems — mixed smoothness, basis construction, and higher dimensions — define the natural continuation of this line of work.