The required contractions are
Substituting into the axis decomposition of a symmetric matrix gives
Indeed , and . Adding reconstructs the given tensor.
Let , and . Taking the trace of the proposed decomposition gives
since and . Contracting twice with gives . Solving these two scalar equations yields
A single contraction gives , and . Hence
Finally set
This is manifestly symmetric. The formulas give ; substituting them into gives zero, and taking the trace gives . Since is symmetric, also holds.
A geometric form makes the axis decomposition of a symmetric matrix explicit. With the transverse orthogonal projection ,
Here is half the transverse trace and is its symmetric traceless rank-two tensor part. Every term was forced by contractions, so the decomposition is unique.
For a Cartesian change of components , a rank- tensor transforms according to
with repeated indices summed. An isotropic tensor has unchanged components under every proper rotation matrix.
For rank two write . Invariance under quarter-turns implies invariance under their squares, the coordinate half-turns
Each off-diagonal entry changes sign under at least one of these, so all off-diagonal entries vanish. A quarter-turn about the third axis then interchanges the first two diagonal entries, making them equal; a quarter-turn about the first axis equates the other two. Therefore
Conversely for every rotation. This proves the isotropic second-rank tensor classification without assuming that the original tensor was symmetric. Also is a scalar, since the trace is rotation-invariant.
At the origin, combine the rational coefficients before taking the limit:
Equivalently . The apparent poles at cancel, giving
The entire limiting profile is also finite:
which independently gives the same central value.
Put . The Fourier transform of a derivative turns the differential operator into , so
Use the partial fraction decomposition
The inverse-transform pair is the one-dimensional modified Helmholtz Green function. Hence
In terms of the original length scale,
This is the convolution of with the fourth-order Green kernel with two decay scales, here . It decays at both infinities. Although the individual exponential pieces have cusps, their derivative jumps through order three cancel, so the result has no unintended Dirac delta distribution at zero. The decaying solution is unique: the homogeneous solutions have no nonzero globally decaying combination.
Use the Fourier transform convention . Split at zero and integrate the decaying exponentials:
The transform is real and even because is real and even.
Specify the branch before defining a signed jump. Choose , and let be the value on the side with argument tending to from above, while is the value on the other side, with argument tending to from below. The entire function has the same limiting value on both sides, so
Thus the jump of a fractional power across a branch cut is
Equivalently the ratio of the limiting values is wherever is nonzero. Reversing the definition of the jump reverses its sign. Choosing a different branch multiplies both limits by the corresponding common phase.
Reflect the physical source across , writing . Its radius is still , and its spherical image is . The Dirichlet Green function inside a hemisphere is
Explicitly this four-source method of images is
Both terms vanish on the spherical cap. On the flat equatorial disc, reflection equates the two physical-source distances and also the two image-source distances, so both pairs cancel. The reflected physical source lies in the lower half-ball and the two inverted images lie outside the sphere. Consequently the only source in the upper half-ball is , with the correct coefficient:
The Dirichlet uniqueness for the Poisson equation then identifies this as the required Green function.
The three-dimensional fundamental solution of the Laplace equation, with , is
It is harmonic away from the source, and its gradient has total outward flux one through a small source-centred sphere.
For , place the image at , outside the unit ball. On , expansion of squared distances gives . The method of images therefore gives the Dirichlet Green function inside a sphere
The boundary value is zero by the distance identity, and the image term is harmonic throughout the physical ball. Thus there. At the central source, take the finite limit of the weighted image term:
Use the Dirichlet Green function with the sign convention
The normal derivative is taken with respect to the integration variable and the outward normal of . Green's second identity gives
The Dirac delta distribution evaluates the first volume term at . Substituting , on and there yields
Hence
One can justify applying the identity to the singular Green function by removing a small sphere around the source and then taking its radius to zero.
Project the initial condition onto the orthogonal spatial eigenfunctions. Their squared norms in the ordinary inner product are
The latter follows from the integrals of and being , with the sine-cosine cross term zero. Therefore
For square-integrable data the expansion converges to the data in the L2 norm at ; for the decaying high modes are smoothed. Initial data with zero projection have no growing term.
Use separation of variables in the form , with . Then and the spatial Sturm-Liouville problem is
These real separated Robin boundary conditions give a self-adjoint differential operator, so the eigenvalues are real. All three sign cases must be considered.
If , write . The first condition gives , while the second reduces to
Thus , and scaling the eigenfunction gives .
If , then . The first condition gives , and the second gives , so ; there is no zero eigenvalue.
If , write . The first condition gives , and the second becomes
Hence , giving the single negative eigenvalue and spatial eigenfunction . Completeness of this regular Sturm-Liouville problem now gives
This is Robin diffusion with a growing mode. The growth is compatible with the stated conditions: they prescribe the same coordinate derivative at both ends, not identical dissipative outward-normal conditions.
For , define the one-dimensional complex group representation
Explicitly , , and . Since , the map is well-defined on exponents modulo four and preserves products. Its four images are distinct, so it is also a faithful representation. The same construction applies to either of the other cyclic subgroups.
The three order-four subgroups are
In each case the generator has powers , so all three are cyclic and mutually isomorphic to , a cyclic group of order four. For the table is
The maps and are explicit group isomorphisms. These are all order-four subgroups: any such subgroup contains an element other than , and that element has order four and already generates the subgroup. They share the order-two subgroup and are each normal, being of index two.
The products in (a), together with and , give closure of the eight-element set. Associativity is inherited from matrix multiplication, is the identity element, and the inverses from (a), with , remain in the set. Thus the group axioms hold. This is the quaternion group .
In the element order specified in the question, the multiplication table is
Interpreting “character of the group” as the character of a representation specified by these matrices, take the trace of each matrix:
A character belongs to a representation; the underlying abstract group can have other characters.
The explicit matrices are
Direct matrix multiplication gives
The reverse products have the opposite signs: , , . Since is the identity matrix and ,
These are the usual quaternion multiplication relations in the complex matrix representation of quaternions.
The trigonometric addition formula gives
Dividing by and writing gives the recurrence relation
The first two members are and . If has degree and leading coefficient , then has degree and leading coefficient , whereas has lower degree and cannot cancel that term. Mathematical induction proves
For example the sequence begins , agreeing with the Chebyshev polynomial of the second kind.
Put , with . The chain rule gives
Consequently the original differential expression becomes . If , direct differentiation gives
Taking therefore yields
The apparent endpoint singularities are removable. Expansion at gives
Thus both endpoint regularity conditions hold. This is the Chebyshev polynomial of the second kind .

There are unlisted articles, also show them or only show them.