The upper-left block is a planar rotation matrix. Its real characteristic polynomial is , which is strictly positive for real because the angle excludes integer multiples of . Hence any real eigenvector has . The remaining equation is , giving
There are infinitely many nonzero solutions, spanning one independent eigenvector direction.
The zero values at select , and the zero value at selects . Expanding the constant upper-edge value as a Fourier sine series gives coefficients . The resulting harmonic function in a rectangle with one constant boundary value is
Each term satisfies Laplace's equation and the three zero boundary values. At , the series equals for . On compact subsets of the interior it and its derivatives converge exponentially, which justifies the termwise equation.
For , the top-corner boundary prescriptions conflict: a corner cannot be both zero on a vertical edge and on the upper edge. The bounded interior solution has the specified limits along the open sides; it has no single continuous extension at those two corners. Values assigned only at the isolated corners do not change the interior solution.
Figure 1.
Harmonic extension from one heated edge of a square
. Original normalized solution , with the upper open edge fixed at one and the other three open edges at zero. The top corner circles mark incompatible side limits, not prescribed continuous corner temperatures. Details.
For Laplace's equation, seek a nonzero separated product . Dividing by gives , so both sides are a separation constant. Choosing it as gives , , with trigonometric and hyperbolic functions, respectively. A zero constant gives linear factors; a positive constant interchanges the two types. Boundary conditions determine which modes are allowed, and linearity permits their superposition to match the remaining boundary data.
At , the Fourier series equals . Isolating the first term gives
Changing the sign of the remaining denominators proves
The absolute convergence justifies the evaluation and rearrangement.
Because the function is even, all sine Fourier coefficients vanish. Its average is zero, so . The Fourier cosine coefficients of a sine mode follow from the product-to-sum identity:
For , the integral is directly zero rather than an undefined substitution into a denominator. Thus
The Fourier series equals the continuous function at every point, including the corners. Its coefficients decay quadratically, so the series is also absolutely convergent.
The function is even and continuous, with period . On it is , so it rises from zero to one at , crosses zero at , reaches minus one at , and returns to zero at . Reflect this curve about the vertical axis and repeat periodically. The joins at integer multiples of have corners, but no jumps.
Figure 1.
Even periodic extension of a sine mode and its Fourier approximation
. Original plot over two periods. The dashed approximation uses the first twelve nonzero cosine coefficients; the exact function is continuous with corners at integer multiples of . Details.
The differential operator is , where . Set ; the product rule reduces the equation to . Twice integrating gives . The initial conditions yield , so
This substitution removes the repeated characteristic root before finding a particular solution.
Use . The constant forcing has particular solution , while the term satisfies the resonance condition and needs the trial . Since , the forced term is . Adding the homogeneous solution and imposing the initial conditions gives
Let . The finite matrix geometric series telescopes to . By the assumed convergence, its terms tend to zero, so the limit satisfies . Consequently . This is the finite-dimensional Neumann series argument; invertibility by itself would not imply convergence.
For the stated nilpotent matrix, and only the first three terms remain:
The dot product is . Each planar vector is perpendicular to because its third component vanishes. Each has norm one, so the three vectors form an orthonormal eigenbasis. This also follows from the spectral theorem for the real symmetric coefficient matrix.
The upper block has trace zero and determinant , so its characteristic polynomial is . The third coordinate has eigenvalue two. Angle-addition identities give the normalized eigenvectors
These formulas apply to every real , including values where an expression obtained by dividing by would fail. Geometrically the planar block is a reflection: it fixes the axis at angle and reverses its perpendicular direction.
A real normal matrix satisfies . For a symmetric matrix, and both products equal . For an antisymmetric matrix, and both equal . For an orthogonal matrix, both products are the identity. Thus each of the three classes consists of normal matrices.

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