Broyden's method 2026-10-25
A secant-updating method for a vector equation , named for Charles George Broyden. Solve , choose , and form , . For a nonzero step the displayed “good Broyden” update satisfies , learning the Jacobian matrix action along the observed step rather than evaluating a new full Jacobian matrix.
The update also has a least-change characterization in the Frobenius norm. Write . Any correction satisfying has squared row norms bounded below by , by the Cauchy-Schwarz inequality. Summing these bounds gives , with equality for . This proves the rank-one correction is the least-change update satisfying the secant constraint. Superlinear convergence needs local regularity and suitable initialization; it is not a global consequence of this algebraic identity.[ref]
Diamond norm of the transposition map Created 2026-09-24 Updated 2026-11-05
For the transposition map , , the induced trace norm is one, while the diamond norm is
The first equality follows because transposition preserves every singular value and hence the trace norm.
For the lower bound, apply to the normalized maximally entangled state projector . Its partial transpose is , with the swap operator. All singular values of are one, so while .
For the upper bound, partial transpose preserves the Frobenius norm . On matrices of size ,
A -dimensional ancilla suffices in the definition of the diamond norm, proving equality. [ref]
For , transposition is not completely positive: has a negative eigenvalue on the antisymmetric subspace. Thus this is a mathematical example of norm amplification by an entangled ancilla; transposition itself is not a physically implementable quantum channel.
Eckart-Young-Mirsky theorem Created 2026-10-14 Updated 2026-10-25
For a finite matrix with singular values in nonincreasing order, truncating its singular value decomposition after terms minimizes the Frobenius norm error, whose square is the sum of the discarded squared singular values. It also minimizes the operator norm error, which equals the next singular value, or zero if none remain. See Mazeika, Section 4.3: Low rank approximation.
The problem of replacing a matrix by one of lower matrix rank is explicit in Eckart and Young's 1936 paper, [ref]. The publication predates the modern image compression example, showing how a general matrix question can support later applications.
Energy release from an antiplane square-root coefficient Created 2026-10-31 Updated 2026-11-05
Use antiplane energy and normalize the tip singularity as plus a less singular term. The energy release rate is the small-circle flux . Here , and . Integrating over gives the displayed coefficient. The full three-dimensional antiplane stress has squared Frobenius norm , so its leading norm ahead of the tip is . These constants depend on the normalization of the stress intensity factor.
The dissipation potential , with the Frobenius norm, gives . An integrating factor proves the displayed memory formula. For , its nonlinear driving term is defined continuously as zero at . Mechanical dissipation is , which must be nonnegative on thermodynamically admissible histories.
Frobenius inner product Created 2026-10-06 Updated 2026-11-05
For complex matrices, the Frobenius inner product is
Here is the conjugate transpose, and this convention is conjugate-linear in the first argument. It is the ordinary Euclidean inner product on all matrix entries, so is positive for every nonzero matrix. Its induced norm is the Frobenius norm.
On real symmetric matrices it becomes . In particular, for a real symmetric matrix and a real vector ,
Thus a quadratic form evaluated at is a linear test of against a rank-one positive semidefinite matrix. This connects the geometry of matrices with quadratic measurements of vectors.
Hilbert-Schmidt norm 2026-10-06
The Hilbert-Schmidt norm of a Hilbert-Schmidt operator is the square root of the displayed sum. The value is independent of the orthonormal basis, by the Parseval identity. For a finite matrix, it is the Frobenius norm.
Hoffman–Wielandt inequality Created 2026-10-07 Updated 2026-10-14
For two normal matrices of the same size, a pairing of their eigenvalues has squared Euclidean discrepancy at most the squared Frobenius norm of their difference. For Hermitian matrices, the increasing real eigenvalue order gives such a pairing, because it minimizes squared discrepancy. For real symmetric matrices, the norm square is .
Low-rank approximation 2026-10-14
A low-rank approximation replaces an matrix by a matrix of matrix rank at most a chosen integer , with , while making an explicitly chosen error norm small. The point is to describe many entries using a few shared patterns: an image compression may retain broad brightness patterns while discarding finer ones. The approximation is useful when the saved storage or computation justifies the lost detail. A small matrix rank alone says nothing about the error.
The compact representation is a sum of outer products, equivalently a product of two thin matrices:
Here are the columns of , and the star denotes the conjugate transpose; over the real numbers use the transpose. The rank bound for a matrix product gives . Conversely, choose a basis of the column space of as the columns of , and put the coordinates of each column of into ; pad with zero columns if the matrix rank is smaller than . Each nonzero outer product is a rank-one matrix: it repeats a single column pattern at different strengths across the columns.
Keeping the two factors stores scalars instead of . This saves space only when . The usual singular value decomposition representation stores scalars because it keeps the singular values separately; absorbing them into one factor removes that extra array. These are representation sizes at equal scalar precision, before file overhead. See Uschmajew and Vandereycken, §9.2: thin factors and their storage, and Illinois CS 357, Low-rank approximation: storing singular vectors and values.
The factors also give without forming the full matrix . Direct evaluation uses scalar multiplications, compared with for a dense matrix applied to a vector. For , the two factors contain scalars rather than , and the same counts describe these multiplication costs. Constructing the factors is a separate cost. Moreover, a low-rank approximation need not be a sparse matrix: the constant matrix whose entries are all has matrix rank despite having no zero entries.
For the standard unweighted problem, write a singular value decomposition with and singular values . Retain the first paired left singular vectors and right singular vectors:
The Eckart-Young-Mirsky theorem says that this same minimizes both of the following error norms over all matrices with matrix rank at most :
where and an empty sum is zero. Here the Frobenius norm is , whereas the operator norm is specifically induced by the Euclidean norm, . The first aggregates errors in all entries; the second bounds the worst output error for a unit vector input. These formulas apply to real and complex matrices. [ref]
For , a relative Frobenius norm tolerance is met precisely when the discarded squared singular values sum to at most . Thus one can choose the smallest suitable before reconstructing the matrix. Many individually small discarded singular values can still have a large combined error. An operator norm tolerance instead tests only the largest discarded singular value, . The two questions can therefore select different ranks.
The following stipulated matrix makes the tradeoff visible. Define four real vectors of length :
The vectors are orthonormal, so this is already a singular value decomposition written as outer products, with positive singular values . All entries lie between and and can be displayed as brightnesses. The first component is uniform brightness, the second distinguishes broad blocks, and the last two supply finer alternating patterns. This is a designed arithmetic example, not measured image data.
Figure 1.
Shared patterns in a stipulated matrix
. The first three panels use the same brightness scale. Rank one retains uniform brightness; rank two adds the broad blocks. The signed residual has its own color scale to expose the discarded patterns. The factors apply a linear map through just intermediate coordinates. Details.
Here has every entry equal to , while has entries within the two diagonal blocks and in the other two blocks. For instance, becomes : its error is . The full residual is , whose Frobenius norm is and whose operator norm is . Since , the relative Frobenius norm error is . Two thin factors use scalars, or if the normalized left singular vectors, right singular vectors, and singular values are stored separately, instead of dense entries.
Figure 2.
Error falls as storage rises for the stipulated matrix
. The exact Frobenius norm and operator norm errors are plotted for . The storage panel counts the two thin factors, or the normalized left singular vectors and right singular vectors plus separate singular values, at equal scalar precision. Lines join discrete rank choices; they do not represent intermediate ranks or measured data. Details.
Retaining three components reduces the Frobenius norm error to while using scalars in two factors. Retaining all four reconstructs exactly but uses scalars, giving no saving in this generic representation; keeping the singular values separately uses . Our particular sign patterns could be encoded more economically, but that would exploit an additional structure beyond the general low-rank approximation. The striking feature is that numerical error and storage can both be read from the decomposition, rather than judged only by how smooth the picture looks.
The constraint is deliberately matrix rank at most . The truncation has . If , the best approximation is itself, with zero error. Requiring matrix rank exactly can destroy attainment: for and , the matrices have matrix rank for , and both error norms equal . Their error has infimum zero, but no rank-two matrix attains it. If , the displayed does have matrix rank exactly and attains the same minimum under either constraint.
Optimality also does not always mean uniqueness. For , the best Frobenius norm approximation is unique when . [ref] A positive tie across the cutoff permits different choices: for and , both and have Frobenius norm error and operator norm error . More generally, is optimal for every real unit vector in the first two coordinate directions. Rotating which tied direction is kept changes the approximation. Repeated singular values entirely inside the retained group do not change its sum of outer products. At the only feasible matrix is zero; at the unique zero-error approximation is , even if some singular values are tied at zero.
A strict gap does not guarantee uniqueness for the operator norm. For and , every with has matrix rank and error , which is optimal. Only minimizes the Frobenius norm, since . An unchanged worst direction can hide extra error in another direction under the operator norm.
Finally, these guarantees concern the stated norms and the sole matrix rank constraint. They do not establish that a discarded pattern is noise or irrelevant to a particular task: the finer patterns in our constructed example were put there intentionally. Adding entrywise weights or restrictions on the factors changes the problem, so the Eckart-Young-Mirsky theorem above does not automatically solve it. For historical context, the original 1936 matrix problem is documented in the existing Eckart-Young-Mirsky theorem section.
Matrix norm Created 2026-10-14 Updated 2026-11-05
A matrix norm is a norm on a vector space of matrices; common choices include induced operator norms and the Frobenius norm. A submultiplicative matrix norm additionally satisfies . The matrix 2-norm is the norm induced by Euclidean norms and equals the largest singular value.
For an orthogonal matrix , the identity gives . Thus multiplication by preserves the Euclidean norm. Applying this on both input and output sides proves for orthogonal matrices and the Euclidean operator norm. The analogous invariance of the Frobenius norm follows by summing column lengths or by a trace calculation.
The specific Helmholtz free energy satisfies . For the multiplicative decomposition of the deformation gradient , adopt , consistently with the material-first nominal stress tensor. Differentiating gives
Consequently
or
These relations follow from the chain rule; the transpose convention matters.
Differentiating the dissipation potential, with the Frobenius norm of , gives
An integrating factor yields the exponential memory from tensor relaxation:
For , the driving term is continuously zero at .
Taking the product explicitly defined in the question, gives , and hence its requested history functional is
There is a naming error in the PDF: with its convention, the actual second Piola-Kirchhoff stress tensor is , while is the transpose of the Mandel stress tensor, where . Thus if that stress measure is required. This agrees with the stress push-forward used in Question 5; the printed product is fully determined by the history above, whereas recovering also requires the deformation or an invertible elastic constitutive relation.
The multiplicative decomposition of the deformation gradient is used on histories where stays invertible. Thermodynamic admissibility additionally requires : positivity of the constants alone does not ensure this for arbitrary prescribed histories.
The Hoffman–Wielandt inequality says that for normal matrices with eigenvalues , there is a permutation such that
where is the Frobenius norm and denotes the conjugate transpose. For Hermitian matrices, the increasingly ordered real eigenvalues may be paired directly: this ordering minimizes the sum of squared distances over all permutations, by removing crossed pairs. In particular, for the real symmetric matrices in this question,
The square on the right is justified by symmetry. For arbitrary matrices, the appropriate expression is the conjugate-transpose product, not the ordinary square.
For partitions, the appropriate Hamming distance between unlabelled bipartitions is . It counts vertices assigned to the wrong group after the better global exchange of the two group names. The printed signed-indicator formula does not implement this exchange: negating a zero-one indicator is not taking its complement. It must be read as this partition distance, or written with the membership vectors.
Use the corrected leading eigenvector estimator from 4(b), and orient its sign to minimize . At every wrongly signed coordinate, this difference has magnitude at least . Therefore the sign rounding bound for a unit eigenvector gives
The last inequality uses the projector distance bound . Taking the expected value of this Frobenius norm estimate proves . The assertion relies on the corrected estimator and partition-distance definition.
For the matrix's Euclidean operator norm, orthogonal invariance of the Euclidean norm gives
Taking gives equality, so . For the Frobenius norm, the trace identity and the singular value decomposition give
Consequently
Here the Euclidean matrix norm means the operator norm induced by the vector Euclidean norm, rather than the entrywise Euclidean length, which is the Frobenius norm.
By the Eckart-Young-Mirsky theorem, truncate the singular value decomposition:
The same minimizes the spectral norm and Frobenius norm errors over matrices of matrix rank at most . This conventional meaning of rank- low-rank approximation includes the case , when .
Differentiability at means that a linear map satisfies
This map is unique and is the total derivative . In finite dimensions linear maps are continuous.
The Frobenius norm is submultiplicative. Expanding the noncommuting cube gives
The last four terms have norm at most . Thus
If , the Neumann series converges and is the inverse of , as follows by multiplying its partial sums and letting the final power tend to zero. Its terms from degree two onwards have norm at most . Hence .
For invertible , write . Whenever , this is invertible, proving the general linear group is open. The same expansion gives
Compose this inverse derivative with the cube derivative, setting . The result is
The three terms cannot generally be combined into a scalar-style , because matrix factors need not commute.
This is the cubic case of the Fréchet derivative of a negative matrix power.
Differentiability at means that there is a linear map with
This map is the total derivative . In the Frobenius norm, , by the Cauchy-Schwarz inequality applied to matrix entries. Expanding with the order of all factors preserved gives
The remainder after the terms linear in has norm at most . Hence
For , the Neumann series converges absolutely in the finite-dimensional space of matrices. Multiplying its partial sums by and passing to the limit shows that it is . Its remainder after has norm at most , proving the required expansion.
If is an invertible matrix and , then is an invertible matrix. Thus the general linear group is an open subset of the matrix space. The preceding expansion also gives
Apply the chain rule to , using the cubic total derivative at and this total derivative of inversion:
The three terms generally cannot be combined by commuting and .
Define the invariants by the characteristic polynomial
Because has positive eigenvalues , its invariants are the elementary symmetric functions in those eigenvalues:
The Frobenius norm gives . Write . Since , its singular values are , so
Finally . Therefore
These identities also give and .
The Green-Lagrange strain tensor has eigenvalues . Its Frobenius norm and trace are invariant under the orthogonal principal-axis change, so
For the homogeneous dilation , put . Then . Differentiating the energy through gives the second Piola-Kirchhoff stress tensor , and multiplication by gives the first Piola-Kirchhoff stress tensor. Thus
The normals of the reference and deformed spheres point in the same radial direction. As , the reference traction magnitude tends to zero, while the current traction magnitude tends to infinity, provided , as for physically stable Lamé parameters. The negative signs of the stress coefficients for indicate compression.
There is no contradiction: the current area is times the reference area, so the same force divided by the vanishing current area has the larger stress. Indeed . The limiting map at is not itself an admissible deformation in continuum mechanics; the statement concerns the approach through positive . A degenerate choice would not have the asserted growth without bound.