Adjoint in weighted coordinates Created 2026-10-14 Updated 2026-10-31
Suppose a linear map between finite-dimensional Hilbert spaces has matrix , and their inner products in the chosen bases are and . The Gram matrices are positive-definite matrices. Writing for the ordinary conjugate transpose, the coordinate matrix of the adjoint operator is
Indeed, the defining inner product identity says for all . Thus simply transposing and conjugating is valid in orthonormal bases, where both Gram matrices are the identity matrix.
For the weighted inner product and the coordinate swap , the adjoint operator has matrix . With and , both inner products equal . The coordinate swap is its own ordinary transpose, but is not self-adjoint in this weighted inner product.
In the crystallographic convention, a Coxeter graph is admissible when it is realized by linearly independent unit vectors in a Euclidean space, with nonpositive mutual inner products and for distinct vertices. The number records a single, double or triple connection; equivalently, the displayed Gram matrix is positive-definite. In edge-label notation the nonzero connections have labels . Forgetting multiplicity, the underlying graph is a forest: for any cycle of distinct vertices, the sum of the corresponding unit vectors would have squared norm at most , contradicting their linear independence. Multiple lines on one connection are labels and are not length-two cycles. This is the convention of Wehler's Lie algebra notes, section 6.3.
Consider the squared error of approximating , , by a polynomial under a measure containing on . This error is finite exactly when . A nonzero constant term makes the integrand asymptotic to , which is not Lebesgue integrable; a vanishing constant term makes , giving an integrable integrand.
Consequently the full space of polynomials of degree at most is not contained in the weighted L2 space. Its admissible subspace is , or zero when . This finite-dimensional subspace has a unique orthogonal projection of , hence a unique best approximation. Indeed its Gram matrix is positive definite: a nonzero polynomial cannot vanish on an interval, so its squared norm is positive. Thus a singular weight can invalidate the proposed full approximation space while still allowing a unique finite-error minimizer in its admissible part. As usual, weighted L2 spaces identify functions equal almost everywhere.
Basis matrix 2026-10-25
A matrix whose columns are the coordinate vectors of an ordered basis in a fixed reference basis. It is an invertible matrix, and multiplication by it converts coordinates in the column basis to the reference coordinates: . For a real inner product space with orthonormal reference basis, its Gram matrix is , and its reciprocal basis has basis matrix .
Benjamin Recht 2026-11-05
Benjamin Recht coauthored the 2007 Random Fourier feature construction with Ali Rahimi. Randomly sampled features approximate a shift-invariant positive-definite kernel by ordinary inner products, permitting a linear learning method to approximate a kernel method without storing the whole pairwise kernel matrix.[ref]
For suitable normalized kernels, features use frequencies sampled from the kernel's spectral distribution and uniform phases. Their expected inner product recovers the kernel; finitely many features introduce an approximation error. The number of features controls a computational and statistical tradeoff rather than an exact identity for every finite sample.
B-spline Created 2026-09-24 Updated 2026-11-05
An order- B-spline is a local, nonnegative spline determined by consecutive entries of a nondecreasing spline knot sequence. Its polynomial pieces have degree at most , and its support is contained in . The Cox-de Boor recurrence gives a constructive definition from interval indicators.[ref]
For example, the order-two B-spline with spline knots is the triangular hat
Only nearby B-splines are nonzero at a given point. This local support makes spline interpolation matrices and spline Gram matrices sparse. On the basic knot interval, a completed B-spline basis has partition of unity; spline knot multiplicities determine the smoothness across joins.
Keep the same origin and the same vector , but use the nonperpendicular basis vectors and . They form a basis because the column matrix has nonzero determinant:
The reconstruction is . Its coordinates are read along lines parallel to the other axis, producing a parallelogram. These lines need not be perpendicular to the axis being measured. In contrast, the dot products are and ; treating those as coordinates would reconstruct , a different vector.
Figure 1.
The same vector reconstructed in an oblique basis: coordinates differ from dot products
. Details.
The dual basis supplies the right measuring rules for any basis.[ref] Here these coordinate functionals are and , the rows of . They give . Equivalently, their representing vectors , satisfy and extract the coordinates as . These are Riesz dual basis in an inner product space representatives, rather than the original basis vectors.
The Gram matrix records how the nonperpendicular directions interact:
For any real basis represented by a square invertible matrix , writing gives these identities immediately. One can recover the coordinates either as or by solving . In this example, if , the squared norm is , not . Oblique coordinates remain useful when natural directions are not perpendicular, but the Gram matrix must accompany them when computing lengths or dot products.
Use the standard polytabloids for tableaux with rows and , as an ordered basis . Adjacent transpositions act by
In characteristic three their common fixed line is . On its quotient every transposition acts by minus one, giving the sign representation. Thus the displayed sequence has the trivial representation and sign as its composition factors, each once. It is nonsplit: the minus-one eigenspace of is , which is not preserved by .
The natural tabloid form restricts to Gram matrix . Its radical in characteristic three is exactly that fixed line. Therefore is the sign quotient, while the trivial factor is . This gives an explicit small failure of semisimplicity in modular symmetric-group representation theory.
André-Louis Cholesky developed his calculation in the setting of geodetic adjustment: measured angles and lengths must satisfy geometric relations, while corrections should be small. The manuscript reproduced in Claude Brezinski's study is titled Sur la résolution numérique des systèmes d’équations linéaires and ends with the date 2 December 1910 at Vincennes. The manuscript reports a 56-equation calculation adjusting altitudes in an Algerian triangulation network. Its date establishes when this written account existed, rather than an exact invention date.[ref]
A related algebraic reason for the method's usefulness is that normal equations involve a Gram matrix. If a real matrix has linearly independent columns, then
so is a positive-definite matrix. That gives the structure needed by a Cholesky decomposition. The historical adjustment problem instead produces from independent constraint rows; the same argument applies to its transpose. In numerical least squares, explicitly forming normal equations squares the condition number of a full-column-rank design matrix; a QR decomposition can therefore be preferable for accuracy. Indeed, the extreme singular values give . The factorization remains useful when a positive-definite matrix is already the natural input.
For a symmetric zero-diagonal dissimilarity matrix , square each entry to obtain and put . Take the eigenvectors of corresponding to its largest positive eigenvalues. Multiplying them by square roots of these eigenvalues gives coordinates. If contains Euclidean distances, is a centered Gram matrix, so retaining all positive eigenvalues reconstructs those distances exactly. Retaining fewer approximates the Gram matrix, not generally the sum of squared errors in raw distances. Negative eigenvalues signal failure of an exact Euclidean representation. Coordinates are unchanged in meaning by orthogonal transformations.
For a matrix with at least two columns of unit Euclidean norm, its mutual coherence is the largest magnitude of an inner product between distinct columns. It measures the strongest pairwise ambiguity of linear measurements. The Cauchy-Schwarz inequality gives . The order-two restricted isometry constant equals the mutual coherence, because every two-column Gram matrix has eigenvalues .
Complement theta number Created 2026-10-05 Updated 2026-10-14
For a finite simple graph with at least one vertex, the complement theta number has the equivalent semidefinite program formulations
and
The Schur complement and prove the equivalence; any feasible is at least one, so division by is valid. For a graph with an edge, is the Gram matrix of a strict vector coloring. Consequently is the strict vector chromatic number, and an ordinary -graph colouring gives by assigning the colors the vertices of a regular simplex.
The consistency bound for separately measured heterodyne arm powers is the requirement that the corrected peak beat current not exceed the maximum permitted by the two separately measured arm currents:
Here are positive, dark-corrected currents and is the beat amplitude after correcting the electrical frequency response. The same collection geometry, detector responsivity and stable illumination must apply to all three measurements, as in separating power imbalance from mode mismatch in optical beating.
The Cauchy-Schwarz inequality gives a short physical proof. Write the collected optical fields as vectors in the detector-weighted inner product, with squared norms proportional to . Their cross term obeys and the beat amplitude is . Equivalently, their Gram matrix
is a positive semidefinite matrix, so its determinant cannot be negative. This is an overlap constraint rather than an empirical correction factor.
Measurement uncertainty matters when interpreting an apparent violation. In an original example, and give . Suppose the three readings have independent relative standard uncertainty. From , the relative uncertainty is approximately , giving . The excess above unity is only about standard uncertainties; it does not establish a physical overlap above one. Correlated calibration influences change the variance of a corrected heterodyne overlap estimate.[ref]
NIST's receiver experiment separately addressed optical mode/polarization matching, electrical losses and impedance mismatch.[ref] Interleaving the two arm-current measurements with the beat measurement helps detect source drift. Dark subtraction, detector nonlinearity or a changed collection aperture can also produce inconsistent inputs to the bound.
Constant-intersection family bound Created 2026-10-06 Updated 2026-10-07
If distinct members of a set family on have intersection size , then its size is at most for , and at most for , assuming . In the positive case, use the Gram matrix of the characteristic vectors of sets; a member of size is handled by disjoint remainders. In the zero case nonempty members are disjoint.
Coxeter Gram matrix Created 2026-09-24 Updated 2026-11-05
The Coxeter Gram matrix is the Gram matrix of the distinguished basis in the Geometric representation of a Coxeter group.
The crystallographic metric matrix is the Gram matrix of the three unit cell basis vectors. If has Cartesian columns , then
Here are the angles between ; ; and , respectively. The matrix converts fractional crystallographic coordinate differences into physical lengths: a fractional displacement has squared length .[ref]
For an original oblique cell, choose , , angstroms, and . Then
For ,
Ignoring the off-diagonal entries would instead give , overestimating this distance. The cross terms encode the cell angles.
The reciprocal crystallographic metric matrix is when the reciprocal basis has dot products , with no factor of . Thus the plane spacing for a nonzero Miller index column satisfies
A basis change gives and fractional coordinates , preserving . Distances do not change when the same crystal structure is described in another basis.[ref]
For normalized columns and , the restricted isometry constant is bounded by the cumulative coherence at . A restricted Gram matrix has diagonal one and every off-diagonal row sum at most . The Gershgorin circle theorem and the finite-dimensional spectral theorem bound its eigenvalues between and . At the distortion is zero, and at it is exactly the mutual coherence.
Decoupled coefficients of orthogonal polynomial projections Created 2026-10-18 Updated 2026-10-20
Let be a real function with , and let be the vector space of real polynomials of degree at most . Its orthogonal projection is the least-squares polynomial in an orthogonal-polynomial basis
Each coefficient depends on just one weighted inner product with and one squared norm. In particular, the coefficients stay the same when the allowed polynomial degree increases. This statement assumes a fixed weight, interval and normalization of the orthogonal polynomials.
To see why the least-squares normal equations decouple, first write a trial polynomial as in any basis of . The residual must be orthogonal to each permitted direction, giving
The Gram matrix usually couples the unknown coefficients. With , it becomes , leaving the separate equations . For the monomial basis on with , by contrast, is zero when is odd and when is even. Distinct even powers, or distinct odd powers, remain coupled.
There is also a proof of the minimum that uses no differentiation. The defining coefficients give for . Thus is orthogonal to all of . For every , the Pythagorean theorem in an inner-product space gives
The second term is nonnegative and vanishes exactly when , proving that the least-squares solution is unique. This is an integrated squared-error optimum; it does not assert a smallest error at every point or a smallest uniform norm.
For , the nested orthogonal projections satisfy the exact residual budget
Indeed, is a linear combination of mutually orthogonal polynomials, and is orthogonal to it. Every newly retained component removes exactly from the squared error. No infinite expansion or completeness assumption is needed for this finite identity. The Legendre least-squares projections of a cubic make every term explicit.
Directed subspace angle Created 2026-10-07 Updated 2026-10-25
For a nonzero closed subspace of a Hilbert space and a closed subspace of a Hilbert space , this angle measures the worst loss under projection from to . Its positive cosine is a uniform lower bound for . The order of the subspaces matters in general. This is distinct from the smallest angle between two subspaces, whose cosine is a supremum. For equal finite dimensions the cosine is the smallest singular value of the cross Gram matrix of orthonormal bases. When is zero, the infimum is over an empty unit sphere and does not define an angle in ; use a lower-bound formulation instead.
If the weight vector lies in the linear span of the training feature map outputs, write . Then
This is a dual representation: coefficients weight training examples rather than feature coordinates. They need not be positive, and need not be unique when the kernel matrix is singular. Some kernel methods instead write with under their own optimization constraints.[ref] The formula is an algebraic conversion for a weight vector in this linear span, not a claim about every possible weight vector. The representer theorem gives a broader reason this form appears in regularized kernel methods.
For the XOR separation by a product feature, let have rows . The Gram matrix of those rows, equivalently the kernel matrix, is
Its eigenvalues are , so it is a positive semidefinite matrix of rank two, not an invertible matrix. More generally, for any real coefficient vector and any finite set of inputs,
This proves the positive-semidefinite kernel condition for the displayed polynomial kernel on all of , rather than checking only this square.
Here , so the explicit choice gives and . Consequently
For the new input , the four polynomial kernel evaluations are . Their weighted sum is , and the sign function predicts . No explicit feature map is needed for this calculation. These coefficients were constructed to reproduce the chosen separator; no fitting algorithm or generalization guarantee is being asserted.
Figure 1.
Quadratic kernel Gram matrix and a dual prediction for a new input
. Both panels use the same ordered four square corners. The Gram matrix has eigenvalues eight, eight, zero, zero. The new-input column contains four kernel evaluations, whose signed weighted contributions sum to minus one half. Details.