Unisolvent Point Set¶

A finite sampling set for which evaluation on a declared finite-dimensional function space is injective, equivalently making interpolation uniquely solvable when the dimensions match.

Version: v3 · 2026-09-06 · History
Domain-specific #: 3035
Origin domain: mathematics
Subdomain: multivariate interpolation
Aliases: Unisolvent node set, Unisolvent interpolation set

Core Idea¶

A finite point set is unisolvent for a function space when the values of a function at those points uniquely determine the function within that space. If W is a finite-dimensional space of functions on a domain Ω and X = {x_1, …, x_N}, define the evaluation map.

E_X : W → F^N, E_X(w) = (w(x_1), …, w(x_N)).

The set X is W-unisolvent when E_X is injective: the only w ∈ W that vanishes at every point of X is the zero function. When N = dim W, injectivity is equivalent to bijectivity, so every vector of nodal data has exactly one interpolant in W. Equivalently, a basis-evaluation or generalized Vandermonde matrix is nonsingular.^[1]^[2]

The abstraction binds points to a declared approximation space. No configuration is unisolvent in isolation; changing polynomial degree, dimension, basis space, or admissible functions can change the verdict.

Structural Signature¶

Sig role-phrases:

the domain Ω — the set on which candidate functions are defined
the finite-dimensional function space W — the reconstruction class
the node set X — the locations at which values are sampled
the evaluation functionals — maps w ↦ w(x_i)
the evaluation map E_X — the joint observation operator
the zero-set test — no nonzero member of W vanishes on all nodes
the rank condition — full column rank of the basis-evaluation matrix
the square condition — |X| = dim W, when uniqueness also gives existence for arbitrary nodal data
the cardinal basis — Lagrange functions, when the square map is invertible
the geometric configuration constraint — node placement that prevents algebraic degeneracy

Recognition test. State W, construct or characterize E_X, and test injectivity. Distinctness of points alone suffices for ordinary univariate polynomial interpolation but not generally for multivariate spaces. A set is not unisolvent merely because one particular data vector happens to admit a unique-looking fitted function.

What It Is Not¶

Not a property of points alone. The same X may be unisolvent for one space and not another.
Not mere point distinctness. Multivariate configurations can be distinct yet lie on the zero set of a nonzero polynomial in W.
Not existence alone. Some data may be interpolable even when uniqueness fails.
Not least-squares identifiability by default. Oversampled full-column-rank systems are injective, but “poised” interpolation often reserves attention for the square case.
Not numerical stability. A nonsingular Vandermonde matrix can be extremely ill-conditioned.
Not a claim that the true phenomenon belongs to W. Unisolvence guarantees reconstruction within the chosen model class.
Not a basis-dependent property. Matrix entries change with basis; rank does not.

Scope of Application¶

Unisolvent sets appear in polynomial interpolation, finite elements, approximation theory, meshless methods, radial-basis augmentation, quadrature construction, and experimental design. They determine whether nodal degrees of freedom identify an element of a trial space.

For univariate polynomials of degree at most m, any m+1 distinct nodes are unisolvent. For polynomials in d variables of total degree at most m, the space dimension is binomial(d+m,m), but selecting that many distinct points does not guarantee unisolvence; their algebraic geometry matters.^[2]

With more nodes than basis functions, injectivity still guarantees that no two functions in W share all observations, although arbitrary data may not lie in the range of E_X. With fewer nodes than dim W, injectivity is impossible. These regimes should not be collapsed into the square interpolation case.

Clarity¶

Choose a basis φ_1, …, φ_M of W and form V_{ij}=φ_j(x_i). Then:

X is W-unisolvent exactly when rank(V)=M;
if N=M, the interpolation coefficients solve the square system Vc=y uniquely for every y exactly when det(V) ≠ 0;
if N>M, full column rank gives uniqueness of a function consistent with data, but exact consistency is not guaranteed for arbitrary y;
if N<M, a nontrivial nullspace prevents unisolvence.

This typing resolves a common ambiguity in informal definitions that say “unique interpolation” without specifying whether uniqueness is conditional on existence or required for all nodal data.

Manages Complexity¶

A function in W may be represented by many coefficient systems. Unisolvence replaces global equality checking by finitely many evaluations: if two candidates agree on a unisolvent set, they are the same member of W. In computation, the property turns nodal values into coordinates and enables cardinal bases.

The rank test also localizes design failures. Instead of blaming an interpolation algorithm, one can ask whether the node geometry annihilates a nonzero trial function. But full rank handles only uniqueness; approximation quality, noise amplification, and conditioning require additional criteria.

Abstract Reasoning¶

Name the space. Every conclusion must be indexed to W and its scalar field.

Use the kernel test. Search for a nonzero w ∈ W with w|_X=0; finding one disproves unisolvence.

Count before computing. If |X| < dim W, stop: injectivity is impossible.

Separate rank from conditioning. Small singular values can make a theoretically unique solution practically unstable.

Exploit basis invariance. Choose a convenient basis to test rank without treating that basis as part of the definition.

Inspect geometry. Collinearity, conic placement, symmetry, or other algebraic relations can create null functions in multivariate spaces.

Knowledge Transfer¶

The portable skeleton is finite identifiability: a family of probes jointly separates every pair of models in a class. This appears in sensor placement, experimental design, system identification, and tomography.

Literal unisolvent-point-set identity requires point evaluations of a declared function space. General linear functionals can be separating without being nodal, and statistical identifiability may be distributional rather than exact. Those are close structural relatives, not automatic instances.

The transfer lesson is to state both the hypothesis class and the observation operator before calling a reconstruction unique.

Examples¶

Canonical: three noncollinear nodes for affine functions¶

Let W = span{1,x,y} on R². Three noncollinear points produce a 3×3 evaluation matrix of full rank, so every three nodal values determine exactly one affine function. If the three points are collinear, a nonzero affine function defining their line vanishes at all of them, and unisolvence fails.

Mapped back: span{1,x,y} is the function space; the three locations are the nodes; the matrix is the evaluation map; noncollinearity ensures full rank; and the line equation supplies the null witness in the failure case.

Applied / In Practice: finite-element degrees of freedom¶

An element uses a polynomial trial space and nodal values as degrees of freedom. Before assembling the global system, the element designer proves that zero values at all nodes force the local polynomial to be zero. This proves the nodal functionals are unisolvent and permits construction of shape functions with Kronecker-delta values. A later conditioning analysis is still needed for numerical quality.

Mapped back: the trial polynomials form W; nodal degrees of freedom form E_X; the zero-data proof establishes injectivity; cardinal shape functions express the inverse map; and conditioning remains a separate requirement.

Structural Tensions¶

T1: Uniqueness vs existence. Injectivity guarantees at most one interpolant; arbitrary-data existence additionally needs surjectivity. Diagnostic: Is the evaluation map square or otherwise known onto?

T2: Algebraic validity vs numerical stability. Full rank can coexist with severe ill-conditioning. Diagnostic: What is the smallest singular value or relevant Lebesgue constant?

T3: Cardinality vs geometry. The correct number of nodes is necessary in the square case but not sufficient in several variables. Diagnostic: Does a nonzero trial function vanish on the configuration?

T4: Model-class certainty vs real-world truth. Data identify one member of W even if W is misspecified. Diagnostic: Is uniqueness being confused with accuracy?

T5: Basis convenience vs invariant property. A basis exposes a matrix but should not determine the conclusion. Diagnostic: Does the rank claim survive a change of basis?

T6: Exact data vs noise. Interpolation honors every value, while noise may favor oversampled approximation. Diagnostic: Is exact interpolation the right objective?

T7: Domain autonomy vs prime reduction. Injectivity and Identifiability explain the core. Diagnostic: Do point evaluations, polynomial/function spaces, node geometry, and cardinal bases still determine distinctive practice? If so, the node remains a useful domain abstraction.

Structural–Framed Character¶

The five-criterion aggregate is 0.05 (structural). Unisolvence is a formal property recognized through an invariant linear map. It is nonevaluative, cross-applicable within mathematics and computation, and not dependent on an institutional convention.

Structural Core vs. Domain Accent¶

Structural core: choose probes whose joint observation map is injective on a hypothesis class.

Domain accent: point evaluations, finite-dimensional function or polynomial spaces, Vandermonde matrices, node geometry, and Lagrange interpolation.

Generalization yields finite identifiability. The accent yields the recognized approximation-theory abstraction.

Injectivity is instantiated exactly by E_X. Identifiability expresses the epistemic reading: nodal data distinguish functions within W. Basis is related operationally because a basis turns evaluation into a matrix and its inverse into cardinal functions, but unisolvence itself is basis-invariant.

Relationships to Other Abstractions¶

Current abstraction Unisolvent Point Set Domain-specific

Parents (1) — more general patterns this builds on

Unisolvent Point Set is a kind of Injectivity Prime

Injectivity is instantiated exactly by E_X.

Hierarchy path (1) — routes to 1 parentless root

Unisolvent Point Set → Injectivity → Function (Mapping)

Neighborhood in Abstraction Space¶

Unisolvent Point Set sits in a sparse region of the domain-specific corpus (75^th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Functions, Maps & Integral Structure (10 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-09-08

Not to Be Confused With¶

a merely distinct point set
a well-spaced or quasi-uniform point set
a numerically well-conditioned interpolation set
uniqueness for one special data vector
least-squares solvability
polynomial poisedness stated without its exact convention
quadrature exactness

References¶

[1] Mariano Gasca and Thomas Sauer, “Polynomial Interpolation in Several Variables”, Advances in Computational Mathematics 12(4), 2000, 377–410. Surveys multivariate interpolation spaces, point configurations, and uniqueness of interpolation. registry ↩

[2] Max Gunzburger and Aretha L. Teckentrup, “Optimal Point Sets for Total Degree Polynomial Interpolation in Moderate Dimensions”, 2014. States the square unisolvence condition, its multivariate geometric dependence, and the equivalent nonsingular generalized-Vandermonde construction. registry ↩a ↩b