1 Definition

A weighted inner product is an inner product in which different components of vectors, or different regions of a function domain, are assigned distinct positive weights. These weights alter the way vector lengths, angles, orthogonality, and projections are measured, while preserving the algebraic structure needed for geometric interpretation. In finite dimensions, the weighting is usually encoded by a matrix; in function spaces, it is often represented by a nonnegative weight function inside an integral.

1.1 Inner product axioms

A weighted inner product must satisfy the usual axioms of an inner product. It is linear in one argument, conjugate symmetric in complex spaces, and positive definite. These conditions guarantee that the weighted form still defines a valid geometry on the vector space. The presence of weights does not change the axioms themselves; rather, it changes the specific numerical value assigned to each pair of vectors.

1.2 Weighted dot product in finite dimensions

In finite-dimensional vector spaces, the most familiar weighted inner product modifies the ordinary dot product by inserting a matrix or a vector of weights. If the weights are all equal, the standard dot product is recovered. When the weights vary, coordinates contribute unequally to the final value, which is useful when some variables are considered more important or more reliable than others.

1.2.1 Diagonal weight matrices

A common special case uses a diagonal matrix whose entries are positive weights. For vectors \(x\) and \(y\), the weighted inner product can be written as a sum of coordinatewise products, each multiplied by its corresponding weight. This is especially convenient when the coordinate system already separates the variables naturally, since the computation remains simple and transparent.

1.2.2 General positive definite matrices

More generally, a weighted inner product may be defined using any symmetric or Hermitian positive definite matrix. In this case, cross-terms between coordinates may appear, not just separate weights on each coordinate. This broader form allows the geometry to reflect correlations between components and is widely used in advanced linear algebra and optimization.

1.3 Weighted inner products on function spaces

On spaces of functions, weighting is typically introduced through an integral against a positive weight function. The integral measures the overlap between two functions while giving more or less influence to selected parts of the domain. Such inner products are common in approximation theory, orthogonal polynomials, and Hilbert space theory.

2 Properties

Weighted inner products retain the essential structural features of ordinary inner products. They support notions of norm, orthogonality, and projection, but the resulting measurements depend on the chosen weights. Many standard theorems remain valid once the appropriate weighted form is substituted for the usual one.

2.1 Linearity

Weighted inner products are linear in the designated argument, just as standard inner products are. This means that combining vectors before evaluating the inner product gives the same result as combining the corresponding inner products afterward. Linearity is fundamental for calculations involving bases, decompositions, and least-squares methods.

2.2 Symmetry or conjugate symmetry

In real vector spaces, a weighted inner product is symmetric: exchanging the two arguments does not change its value. In complex vector spaces, the corresponding property is conjugate symmetry. This ensures that the geometric interpretation remains consistent and that self-inner products are always real.

2.3 Positive definiteness

A weighted inner product assigns a strictly positive value to every nonzero vector paired with itself. This property is guaranteed when the weight structure is positive definite or, in the integral setting, when the weight function is positive almost everywhere. Positive definiteness prevents distinct nonzero vectors from having zero length.

2.4 Induced norm

Every weighted inner product determines a norm by taking the square root of the self-inner product. This weighted norm measures vector magnitude according to the chosen geometry. It behaves analogously to the Euclidean norm, but the lengths it produces depend on the distribution of weights.

2.5 Orthogonality

Two vectors are orthogonal in the weighted sense when their weighted inner product is zero. Weighted orthogonality often differs from ordinary orthogonality, since the weight may amplify or suppress certain components. This idea is central to weighted orthogonal bases and weighted projection methods.

3 Examples

Weighted inner products appear in many familiar settings. They may emphasize selected coordinates in finite-dimensional spaces, modify integrals in function spaces, or adapt to complex-valued vector spaces. These examples show how the same abstract idea takes different concrete forms.

3.1 Euclidean space with coordinate weights

In \(\mathbb{R}^n\), a weighted inner product can assign different importance to each coordinate. For instance, one coordinate may be weighted more heavily because it represents a more precise measurement or a more significant feature. The resulting geometry stretches the space in some directions and compresses it in others.

3.2 Polynomial spaces with integration weights

For polynomials on an interval, a weighted inner product often takes the form of an integral against a weight function. Certain classical polynomial families are orthogonal with respect to such weighted integrals. This makes the weighted inner product a natural tool for building approximation schemes and orthogonal expansions.

3.3 Complex vector spaces

In complex spaces, the weighted inner product includes complex conjugation in one argument and may involve a Hermitian positive definite matrix. This keeps the self-inner product real and positive. Complex weighted inner products are common in quantum mechanics, signal analysis, and other areas where phase information matters.

4 Geometric interpretation

Weights change the geometry of a space by altering how distances and angles are computed. Although the underlying vector space remains the same, the metric properties become anisotropic, meaning that some directions are measured more strongly than others. This leads to a geometry that can be visually understood as stretched or skewed relative to the standard one.

4.1 Distortion of lengths

A weighted inner product changes the length of a vector according to how much each component contributes under the chosen weights. Vectors with large components in heavily weighted directions acquire greater norm than they would under the usual dot product. Conversely, components in lightly weighted directions have less influence on the final length.

4.2 Angles under weighting

Angles are determined indirectly from the weighted inner product and the associated norms. Because the weighting changes both lengths and mutual overlap, the angle between two vectors may differ substantially from the standard Euclidean angle. This is especially noticeable when the weight matrix has strongly uneven entries.

4.3 Ellipsoidal unit balls

The set of vectors of weighted norm one is generally an ellipsoid rather than a sphere. This reflects the directional bias introduced by the weights. In matrix form, the shape of the unit ball provides a visual summary of how the weighted geometry stretches space.

5 Orthonormal bases

An orthonormal basis in a weighted inner product space consists of vectors that are mutually orthogonal and each have weighted norm one. Such bases simplify expansions, projections, and coordinate computations. The existence of weighted orthonormal bases follows from the same general principles that apply in ordinary inner product spaces.

5.1 Gram-Schmidt process with weights

The Gram-Schmidt process can be adapted by replacing the standard dot product with the weighted inner product. Each new basis vector is orthogonalized against the earlier ones using weighted projections, then normalized with the weighted norm. The resulting vectors form a weighted orthonormal system.

5.2 Weighted orthogonalization

Weighted orthogonalization is useful when basis vectors are already known but do not satisfy the weighted orthogonality condition. By systematically removing weighted components along previous vectors, one can construct a more convenient basis. This procedure is widely used in numerical algorithms and polynomial approximation.

6 Projections and least squares

Weighted inner products provide the natural framework for projections when certain errors should count more than others. They also underlie weighted least-squares methods, in which deviations are penalized unevenly across observations or coordinates. These techniques are central in data fitting and computational linear algebra.

6.1 Weighted projection formula

The projection of a vector onto a subspace is defined so that the residual is weighted-orthogonal to that subspace. In matrix settings, the projection operator depends on the weight matrix and the chosen basis for the subspace. This produces the best approximation relative to the weighted geometry.

6.2 Weighted least squares

Weighted least squares minimizes a weighted sum of squared residuals. Observations with larger weights influence the fitted solution more strongly than observations with smaller weights. The method is especially useful when measurements have unequal reliability or varying variances.

6.2.1 Normal equations

The weighted least-squares problem leads to a system of normal equations in which the weight matrix appears explicitly. Solving these equations yields the coefficient vector that minimizes the weighted residual norm. The formulation generalizes the standard least-squares equations by inserting the appropriate weights.

6.2.2 Error minimization

In weighted fitting, the error is measured by a norm that reflects the prescribed importance of each component. The optimizer chooses parameters that reduce the weighted discrepancy rather than the unweighted one. This can produce a better practical fit when some data points should influence the result more heavily.

7 Matrix representation

Weighted inner products are often represented by matrices, especially in finite-dimensional spaces. The matrix encodes how vectors interact under the weighted geometry and provides a compact way to compute inner products, norms, and projections. This representation also clarifies the role of basis changes.

7.1 Gram matrices

A Gram matrix records the pairwise weighted inner products of a collection of vectors. It summarizes the geometry of the set and indicates whether the vectors are linearly independent. In weighted settings, the entries reflect the chosen weighting scheme, not merely the coordinate arrangement.

7.2 Change of basis

Under a change of basis, the matrix representing the weighted inner product transforms accordingly. The underlying geometric structure remains the same, but its coordinate expression changes. This makes it possible to work in whichever basis is most convenient while preserving the weighted relations.

7.3 Relation to positive definite matrices

Every finite-dimensional weighted inner product can be represented by a positive definite matrix, and conversely every such matrix defines a weighted inner product. This correspondence is one of the most important facts about the subject. It connects geometric concepts directly to matrix theory and numerical computation.

8 Applications

Weighted inner products are used whenever different directions, variables, or data points should contribute unequally to a calculation. They are especially valuable in algorithms that require stable geometry, in statistical estimation, and in approximation methods. Their flexibility makes them a standard tool across several mathematical disciplines.

8.1 Numerical linear algebra

In numerical linear algebra, weighted inner products help define stable algorithms for decomposition, projection, and iterative solution methods. They are useful when the natural geometry of the problem is not Euclidean or when conditioning should be improved by rescaling. Weighted formulations can also reflect physical units or measurement scales.

8.2 Statistics and data fitting

Statistical estimation often uses weights to account for differing uncertainties among observations. A weighted inner product provides the geometric basis for weighted regression and related fitting procedures. This allows noisy observations to be downweighted and more reliable ones to have greater influence.

8.3 Signal processing

In signal processing, weighted inner products may model energy in selected frequency bands or time intervals more strongly than in others. They are useful in filtering, feature extraction, and matched analysis of signals with nonuniform importance. Weighted geometries can also help when sampling density varies across the domain.

8.4 Approximation theory

Approximation theory frequently relies on weighted inner products to construct orthogonal polynomials and best approximations under nonuniform criteria. The weights determine which parts of the domain matter most in the approximation error. This is important in both theoretical expansions and practical computation.

Weighted inner products are closely connected to several standard notions in linear algebra and analysis. These related ideas share the same geometric vocabulary but differ in the exact structure or assumptions involved. Understanding them clarifies the role of weighting in modern mathematics.

9.1 Standard inner product

The standard inner product is the unweighted case, usually given by the ordinary dot product or its complex analogue. It serves as the reference geometry from which weighted versions are derived. When all weights are equal, the weighted form reduces to this familiar setting.

9.2 Weighted norm

A weighted norm is the norm induced by a weighted inner product. It measures vector size according to the same weighting rule and is therefore not independent of the inner product itself. Weighted norms appear in analysis, optimization, and approximation.

9.3 Weighted Hilbert space

A weighted Hilbert space is a complete inner product space equipped with a weighted inner product. Completeness ensures that Cauchy sequences converge within the space, which is important for convergence arguments and spectral methods. These spaces are central in functional analysis and orthogonal expansions.

9.4 Bilinear and sesquilinear forms

A weighted inner product is a special case of a bilinear or sesquilinear form with additional positivity requirements. Such forms provide a broader language for measuring interactions between vectors. The inner product axioms select the forms that support geometry, lengths, and orthogonality.