Собствени стойност и собствени вектори на матрица

File:Mona Lisa eigenvector grid.png

In this shear mapping the red arrow changes direction but the blue arrow does not. The blue arrow is an eigenvector, and since its length is unchanged its eigenvalue is 1.

An eigenvector of a square matrix is a non-zero vector that, when multiplied by the matrix, yields a vector that differs from the original at most by a multiplicative scalar.

For example, if three-element vectors are seen as arrows in three-dimensional space, an eigenvector of a 3×3 matrix A is an arrow whose direction is either preserved or exactly reversed after multiplication by A. The corresponding eigenvalue determines how the length and sense of the arrow is changed by the operation.

Specifically, a non-zero column vector v is a right eigenvector of a matrix A if (and only if) there exists a number λ such that Av = λv. If the vector satisfies vA = λv instead, it is said to be a left eigenvector. The number λ is called the eigenvalue corresponding to that vector. The set of all eigenvectors of a matrix, each paired with its corresponding eigenvalue, is called the eigensystem of that matrix.<ref name=WolframEigenvector>Template:Cite web</ref><ref>Template:Citation</ref>

An eigenspace of A is the set of all eigenvectors with the same eigenvalue, together with the zero vector.<ref name=WolframEigenvector/>

The terms characteristic vector, characteristic value, and characteristic space are also used for these concepts. The prefix eigen- is adopted from the German word eigen for "self".<ref>See eigen or eigenvalue at Wiktionary.</ref> Having an eigenvalue is an accidental property of a real matrix (since it may fail to have an eigenvalue), but every complex matrix has an eigenvalue.

These concepts are naturally extended to more general situations, where the set of real scale factors is replaced by any field of scalars (such as algebraic or complex numbers); the set of Cartesian vectors $\mathbb {R} ^{n}$ is replaced by any vector space (such as the continuous functions, the polynomials or the trigonometric series), and matrix multiplication is replaced by any linear operator that maps vectors to vectors (such as the derivative from calculus). In such cases, the concept of "parallel to" is interpreted as "scalar multiple of", and the "vector" in "eigenvector" may be replaced by a more specific term, as in "eigenfunction", "eigenmode", "eigenface", "eigenstate", and "eigenfrequency". Thus, for example, the exponential function $f(x)=a^{x}$ is an eigenfunction of the derivative operator " ${}'$ ", with eigenvalue $\ln a$, since its derivative is $f'(x)=(\ln a)a^{x}=(\ln a)f(x)$.

Eigenvalues and eigenvectors have many applications in both pure and applied mathematics. They are used in matrix factorization, in quantum mechanics, and in many other areas.

Definition

Eigenvectors and eigenvalues of a real matrix

File:Eigenvalue equation.svg

Matrix A acts by stretching the vector x, not changing its direction, so x is an eigenvector of A.

Eigenvectors and eigenvalues depend on the concepts of vectors and linear transformations. In most cases, a vector can be assumed to be a list of real numbers. (For example, a vector can be the three coordinates of a point in three-dimensional space, relative to some Cartesian coordinate system. It helps to think of that point as the tip of an arrow whose tail is at the origin of the coordinate system.) Then a linear transformation for such vectors can be defined by a square matrix with one row and one column for each element of the vector. If we think of a vector x as a matrix with n rows and one column, then the linear operator defined by a matrix A with n rows and columns maps the vector x to the matrix product Ax. That is,

${\begin{bmatrix}x_{1}\\x_{2}\\\vdots \\x_{n}\end{bmatrix}}\mapsto {\begin{bmatrix}A_{1,1}&A_{1,2}&\ldots &A_{1,n}\\A_{2,1}&A_{2,2}&\ldots &A_{2,n}\\\vdots &\vdots &\ddots &\vdots \\A_{n,1}&A_{n,2}&\ldots &A_{n,n}\\\end{bmatrix}}{\begin{bmatrix}x_{1}\\x_{2}\\\vdots \\x_{n}\end{bmatrix}}\;=\;{\begin{bmatrix}y_{1}\\y_{2}\\\vdots \\y_{n}\end{bmatrix}}$

where, for each index i,

$y_{i}=A_{i1}x_{1}+A_{i2}x_{2}+\cdots +A_{in}x_{n}=\sum _{j=1}^{n}A_{ij}x_{j}$

Two vectors x and y are said to be parallel to each other if one is a multiple of the other; that is, if every element of one is the corresponding element of the other times the same scaling factor. For example, the vectors x with elements 1,3,4 and y with elements -20,-60,-80 are parallel, because each element of y is −20 times the corresponding element of x. If x and y are arrows in three-dimensional space, the condition "x is parallel to y" means that their arrows lie on the same straight line, and may differ only in length and direction along that line.

In general, multiplication of a non-zero vector x by a square matrix A yields a vector y that is not parallel to x. When these two vectors are parallel—that is, when Ax = λx for some real number λ—we say that x is an eigenvector of A. In that case, the scale factor λ is the corresponding eigenvalue.

In particular, multiplication by a 3×3 matrix A may change both the direction and the magnitude of an arrow in three-dimensional space. However, if the arrow x is an eigenvector of A with eigenvalue λ, the operation may only change its length, and either keep its direction or flip it (make the arrow point in the exact opposite direction). Specifically, the length of the arrow will increase if |λ| > 1, remain the same if |λ| = 1, and decrease it if |λ| < 1. Moreover, the direction will be precisely the same if λ > 0, and flipped if λ < 0. If λ = 0, then the length of the arrow becomes zero.

Examples

File:Eigenvectors.gif

The transformation matrix ${\bigl [}{\begin{smallmatrix}2&1\\1&2\end{smallmatrix}}{\bigr ]}$ preserves the direction of vectors parallel to ${\bigl (}{\begin{smallmatrix}1\\1\end{smallmatrix}}{\bigr )}$ (in blue) and ${\bigl (}{\begin{smallmatrix}1\\-1\end{smallmatrix}}{\bigr )}$ (in violet). The points that lie on the line through the origin, parallel to an eigenvector, remain on the line after the transformation. The vectors in red are not eigenvectors, therefore their direction is altered by the transformation. See also: An extended version, showing all four quadrants.

For the transformation matrix

$A={\begin{bmatrix}2&1\\1&2\end{bmatrix}},$

the vector

$\mathbf {x} ={\begin{bmatrix}3\\-3\end{bmatrix}}$

is an eigenvector with eigenvalue 1. Indeed,

$A\mathbf {x} ={\begin{bmatrix}2&1\\1&2\end{bmatrix}}{\begin{bmatrix}3\\-3\end{bmatrix}}={\begin{bmatrix}2\cdot 3+1\cdot (-3)\\1\cdot 3+2\cdot (-3)\end{bmatrix}}={\begin{bmatrix}3\\-3\end{bmatrix}}=1\cdot {\begin{bmatrix}3\\-3\end{bmatrix}}.$

On the other hand the vector

$\mathbf {x} ={\begin{bmatrix}0\\1\end{bmatrix}}$

is not an eigenvector, since

${\begin{bmatrix}2&1\\1&2\end{bmatrix}}{\begin{bmatrix}0\\1\end{bmatrix}}={\begin{bmatrix}2\cdot 0+1\cdot 1\\1\cdot 0+2\cdot 1\end{bmatrix}}={\begin{bmatrix}1\\2\end{bmatrix}},$

and this vector is not a multiple of the original vector x.

The identity matrix I (whose general element I_ij is 1 if i=j, and 0 otherwise) maps every vector to itself. Therefore, every vector is an eigenvector of I, with eigenvalue 1.

General definition

The concept of eigenvectors and eigenvalues extends naturally to abstract linear operators on abstract vector spaces. Namely, let V be any vector space with some scalar field K, and let T be a linear transformation mapping V into V. We say that a vector x of V is an eigenvector of T if (and only if) there is a scalar λ in K such that T(x) = λx.

This equation is called the eigenvalue equation for T. Note that T(x) means the result of applying the operator T to the vector x, while λx means the product of the scalar λ by x.<ref>See Template:Harvnb; Template:Harvnb</ref>

(Some authors say that x must be non-zero in the definition of eigenvector.<ref>Template:Citation</ref>)

Eigenspace and spectrum

If x is an eigenvector of T, then any scalar multiple αx of x with nonzero α is also an eigenvector, with the same eigenvalue as x, since T(αx) = αT(x) = α(λx) = λ(αx). Moreover, if x and y are eigenvectors with the same eigenvalue λ, then x+y is also an eigenvector with the same eigenvalue λ. Therefore, the set of all eigenvectors with the same eigenvalue λ, together with the zero vector, is a linear subspace of V, called the eigenspace of λ.<ref>Template:Harvnb</ref><ref>Lemma for the eigenspace</ref> If that subspace has dimension 1, it is sometimes called an eigenline.<ref>Schaum's Easy Outline of Linear Algebra, p111</ref>

The list of eigenvalues of T is sometimes called the spectrum of T. The order of this list is arbitrary, but the number of times that an eigenvalue λ appears is important: it is the dimension of the corresponding eigenspace, that is, the maximum number of linearly independent eigenvectors that have eigenvalue λ.

Any subspace spanned by eigenvectors of T is an invariant subspace of T.

Eigenvalues and eigenvectors of matrices

Characteristic polynomial

The eigenvalues of a matrix A are precisely the solutions λ to the equation

$\det(A-\lambda I)=0\,.$

Here det is the determinant of the matrix formed by A - λTemplate:Serif and Template:Serif is the n×n identity matrix. This equation is called the characteristic equation (or, less often, the secular equation) of A.

For example, consider the special case of a diagonal matrix A:

$A={\begin{bmatrix}a_{1,1}&0&\cdots &0\\0&a_{2,2}&\cdots &0\\\vdots &\vdots &\ddots &0\\0&0&0&a_{n,n}\end{bmatrix}}$

The characteristic equation of A would read:

$\det(A-\lambda I)=\det \left({\begin{bmatrix}a_{1,1}&0&\cdots &0\\0&a_{2,2}&\cdots &0\\\vdots &\vdots &\ddots &0\\0&0&0&a_{n,n}\end{bmatrix}}-\lambda {\begin{bmatrix}1&0&\cdots &0\\0&1&\cdots &0\\\vdots &\vdots &\ddots &0\\0&0&0&1\end{bmatrix}}\right)$

${\begin{aligned}&=\det {\begin{bmatrix}a_{1,1}-\lambda &0&\cdots &0\\0&a_{2,2}-\lambda &\cdots &0\\\vdots &\vdots &\ddots &0\\0&0&0&a_{n,n}-\lambda \end{bmatrix}}\\&=(a_{1,1}-\lambda )(a_{2,2}-\lambda )\cdots (a_{n,n}-\lambda )=0.\end{aligned}}$

The solutions to this equation are the eigenvalues λ_i = a_i,i (i = 1, ..., n).

Proving the aforementioned relation of eigenvalues and solutions of the characteristic equation requires some linear algebra, specifically the notion of linearly independent vectors: briefly, the eigenvalue equation for a matrix A can be expressed as

$A\mathbf {x} -\lambda I\mathbf {x} =\mathbf {0} \,,$

which can be rearranged to

$(A-\lambda I)\mathbf {x} =\mathbf {0} \,.$

If there exists an inverse

$\displaystyle (A-\lambda I)^{-1},$

then both sides can be left-multiplied by it, to obtain x = 0. Therefore, if λ is such that Template:Nowrap is invertible, λ cannot be an eigenvalue. It can be shown that the converse holds, too: if Template:Nowrap is not invertible, λ is an eigenvalue. A criterion from linear algebra states that a matrix (here: Template:Nowrap) is non-invertible if and only if its determinant is zero, thus leading to the characteristic equation.

The left-hand side of this equation can be seen (using Leibniz' rule for the determinant) to be a polynomial function in λ, whose coefficients depend on the entries of A. This polynomial is called the characteristic polynomial. Its degree is n, that is to say, the highest power of λ occurring in this polynomial is λⁿ. At least for small matrices, the solutions of the characteristic equation (hence, the eigenvalues of A) can be found directly. Moreover, it is important for theoretical purposes, such as the Cayley–Hamilton theorem. It also shows that any n×n matrix has at most n eigenvalues. However, the characteristic equation need not have n distinct solutions. In other words, there may be strictly less than n distinct eigenvalues. This happens for the matrix describing the shear mapping discussed below.

If the matrix has real entries, the coefficients of the characteristic polynomial are all real. However, the roots are not necessarily real; they may include complex numbers with a non-zero imaginary component. For example, a 2×2 matrix describing a 45° rotation will not leave any non-zero vector pointing in the same direction. However, there is at least one complex number λ solving the characteristic equation, even if the entries of the matrix A are complex numbers to begin with. (This existence of such a solution is known as the fundamental theorem of algebra.) For a complex eigenvalue, the corresponding eigenvectors also have complex components.

Algebraic and geometric multiplicities

Given an n×n matrix A and an eigenvalue λ_i of this matrix, there are two numbers measuring, roughly speaking, the number of eigenvectors belonging to λ_i. They are called multiplicities: the algebraic multiplicity of an eigenvalue is defined as the multiplicity of the corresponding root of the characteristic polynomial. The geometric multiplicity of an eigenvalue is defined as the dimension of the associated eigenspace, i.e. number of linearly independent eigenvectors with that eigenvalue. Both algebraic and geometric multiplicity are integers between (including) 1 and n. The algebraic multiplicity n_i and geometric multiplicity m_i may or may not be equal, but we always have m_i ≤ n_i. The simplest case is of course when m_i = n_i = 1. The total number of linearly independent eigenvectors, N_x, is given by summing the geometric multiplicities

$\sum \limits _{i=1}^{N_{\lambda }}{m_{i}}=N_{\mathbf {x} }.$

Over a complex vector space, the sum of the algebraic multiplicities will equal the dimension of the vector space, but the sum of the geometric multiplicities may be smaller. In this case, it is possible that there may not be sufficient eigenvectors to span the entire space – more formally, there is no basis of eigenvectors (an Template:Visible anchor). A matrix is diagonalizable by a suitable choice of coordinates if and only if there is an eigenbasis; if a matrix is not diagonalizable, it is said to be defective. For defective matrices, the notion of eigenvector can be generalized to generalized eigenvectors, and over an algebraically closed field a basis of generalized eigenvectors always exists, as follows from Jordan form.

The eigenvectors corresponding to different eigenvalues are linearly independent, meaning, in particular, that in an n-dimensional space the linear transformation A cannot have more than n eigenvalues (or eigenspaces).<ref name="Shilov_lemma">For a proof of this lemma, see Template:Harvnb; Template:Harvnb; Template:Harvnb; Template:Harvnb; and Lemma for linear independence of eigenvectors</ref> All defective matrices have fewer than n distinct eigenvalues, but not all matrices with fewer than n distinct eigenvalues are defective<ref>Template:Citation</ref> – for example, the identity matrix is diagonalizable (and indeed diagonal in any basis), but only has the eigenvalue 1.

Given an ordered choice of linearly independent eigenvectors, especially an eigenbasis, they can be indexed by eigenvalues, i.e. using a double index, with x_i,j being the j ^th eigenvector for the i ^th eigenvalue. The eigenvectors can also be indexed using the simpler notation of a single index x_k, with k = 1, 2, ..., N_x.

Worked example

These concepts are explained for the matrix

$A={\begin{bmatrix}2&1\\1&2\end{bmatrix}}.$

The characteristic equation of this matrix reads

$\det(A-\lambda I)=\det {\begin{bmatrix}2-\lambda &1\\1&2-\lambda \end{bmatrix}}=0\,.$

Calculating the determinant, this yields the quadratic equation

$\lambda ^{2}-4\lambda +3=0,\,$

whose solutions (also called roots) are $\lambda =1$ and $\lambda =3$. The eigenvectors for the eigenvalue $\lambda =3$ are determined by using the eigenvalue equation, which in this case reads

${\begin{bmatrix}2&1\\1&2\end{bmatrix}}{\begin{bmatrix}x\\y\end{bmatrix}}=3{\begin{bmatrix}x\\y\end{bmatrix}}.$

The juxtaposition at the left hand side denotes matrix multiplication. Spelling this out, this equation comparing two vectors is tantamount to a system of the following two linear equations:

$2x+y=3x\,$

$x+2y=3y.\,$

Both equations reduce to the single linear equation $x=y$. That is to say, any vector of the form (x, y) with y = x is an eigenvector to the eigenvalue λ = 3. However, the vector (0, 0) is excluded. A similar calculation shows that the eigenvectors corresponding to the eigenvalue $\lambda =1$ are given by non-zero vectors (x, y) such that Template:Nowrap beginy = −xTemplate:Nowrap end. For example, an eigenvector corresponding to $\lambda =1$ is ${\begin{bmatrix}-1\\1\end{bmatrix}}$ whereas an eigenvector corresponding to $\lambda =3$ is ${\begin{bmatrix}1\\1\end{bmatrix}}$. These vectors, placed as columns in a matrix, may be used to create a diagonalizable matrix.

Eigendecomposition

Let A be a square n × n matrix. Let q₁ ... q_k be an eigenvector basis, i.e. an indexed set of k linearly independent eigenvectors, where k is the dimension of the space spanned by the eigenvectors of A. If k = n, then A can be written

$\mathbf {A} =\mathbf {Q} \mathbf {\Lambda } \mathbf {Q} ^{-1}$

where Q is the square n × n matrix whose i-th column is the basis eigenvector q_i of A and Λ is the diagonal matrix whose diagonal elements are the corresponding eigenvalues, i.e. Λ_ii = λ_i. (See also change of basis.)

A matrix A is diagonalizable if and only if the space $\mathbb {R} ^{n}$ can be decomposed into a direct sum of eigenspaces of A, and the matrix has an eigenspace decomposition, or eigendecomposition, or spectral decomposition. If a matrix is not diagonalizable, then it is called defective, and, while it cannot be decomposed into eigenspaces, it can be decomposed into the more general concept of generalized eigenspaces, as discussed here.

Further properties

Let $A$ be an n×n matrix with eigenvalues $\lambda _{i}$, $i=1,2,\dots ,n$. Then

• Trace of A

$\operatorname {tr} (A)=\sum \lambda _{i}=\lambda _{1}+\lambda _{2}+\cdots +\lambda _{n}$.

• Determinant of A

$\operatorname {det} (A)=\prod \lambda _{i}=\lambda _{1}\lambda _{2}\cdots \lambda _{n}$.

• Eigenvalues of $A^{k}$ are $\lambda _{1}^{k},\dots ,\lambda _{n}^{k}$

These first three results follow by putting the matrix in upper-triangular form, in which case the eigenvalues are on the diagonal and the trace and determinant are respectively the sum and product of the diagonal.

• If $A=A^{H}$, i.e., $A$ is Hermitian, every eigenvalue is real. If it's also positive-definite, positive-semidefinite, negative-definite, or negative-semidefinite every eigenvalue is positive, non-negative, negative, or non-positive respectively.

• Every eigenvalue of a Unitary matrix has absolute value $|\lambda |=1$.

Examples in the plane

The following table presents some example transformations in the plane along with their 2×2 matrices, eigenvalues, and eigenvectors.

	horizontal shear	scaling	unequal scaling	hyperbolic rotation
illustration	Horizontal shear mapping	Equal scaling (homothety	Vertical shrink (k2 < 1) and horizontal stretch (k1 > 1) of a unit square.	'"`UNIQ--postMath-00000033-QINU`"'
matrix	${\begin{bmatrix}1&k\\0&1\end{bmatrix}}$	${\begin{bmatrix}k&0\\0&k\end{bmatrix}}$	${\begin{bmatrix}k_{1}&0\\0&k_{2}\end{bmatrix}}$	${\begin{bmatrix}\cosh \varphi &\sinh \varphi \\\sinh \varphi &\cosh \varphi \end{bmatrix}}$
characteristic equation	λ2 − 2λ+1 = (1 − λ)2 = 0	λ2 − 2λk + k2 = (λ − k)2 = 0	(λ − k1)(λ − k2) = 0	λ2 − 2λ cosh φ + 1 = 0
eigenvalues λi	λ1,2=1	λ1,2=k	λ1 = k1, λ2 = k2	λ1 = exp(φ), λ2 = exp(−φ),
algebraic and geometric multiplicities	n1 = 2, m1 = 1	n1 = 2, m1 = 2	n1 = m1 = 1, n2 = m2 = 1	n1 = m1 = 1, n2 = m2 = 1
eigenvectors	$\mathbf {u} _{1}=(1,0)$	All non-zero vectors	$\mathbf {u} _{1}=(1,0),\mathbf {u} _{2}=(0,1)$	$\mathbf {u} _{1}={\begin{bmatrix}1\\1\end{bmatrix}},\mathbf {u} _{2}={\begin{bmatrix}1\\-1\end{bmatrix}}.$

Shear

Shear in the plane is a transformation where all points along a given line remain fixed while other points are shifted parallel to that line by a distance proportional to their perpendicular distance from the line.<ref>Definition according to Weisstein, Eric W. Shear From MathWorld − A Wolfram Web Resource</ref> In the horizontal shear depicted above, a point P of the plane moves parallel to the x-axis to the place P' so that its coordinate y does not change while the x coordinate increments to become Template:Nowrap beginx' = x + k y,Template:Nowrap end where k is called the shear factor. The shear angle φ is determined by k = cot φ.

Repeatedly applying the shear transformation changes the direction of any vector in the plane closer and closer to the direction of the eigenvector.

Uniform scaling and reflection

Multiplying every vector with a constant real number k is represented by the diagonal matrix whose entries on the diagonal are all equal to k. Mechanically, this corresponds to stretching a rubber sheet equally in all directions such as a small area of the surface of an inflating balloon. All vectors originating at origin (i.e., the fixed point on the balloon surface) are stretched equally with the same scaling factor k while preserving its original direction. Thus, every non-zero vector is an eigenvector with eigenvalue k. Whether the transformation is stretching (elongation, extension, inflation), or shrinking (compression, deflation) depends on the scaling factor: if k > 1, it is stretching; if Template:Nowrap, it is shrinking. Negative values of k correspond to a reversal of direction, followed by a stretch or a shrink, depending on the absolute value of k.

Unequal scaling

For a slightly more complicated example, consider a sheet that is stretched unequally in two perpendicular directions along the coordinate axes, or, similarly, stretched in one direction, and shrunk in the other direction. In this case, there are two different scaling factors: k₁ for the scaling in direction x, and k₂ for the scaling in direction y. If a given eigenvalue is greater than 1, the vectors are stretched in the direction of the corresponding eigenvector; if less than 1, they are shrunken in that direction. Negative eigenvalues correspond to reflections followed by a stretch or shrink. In general, matrices that are diagonalizable over the real numbers represent scalings and reflections: the eigenvalues represent the scaling factors (and appear as the diagonal terms), and the eigenvectors are the directions of the scalings.

The figure shows the case where $k_{1}>1$ and $1>k_{2}>0$. The rubber sheet is stretched along the x axis and simultaneously shrunk along the y axis. After repeatedly applying this transformation of stretching/shrinking many times, almost any vector on the surface of the rubber sheet will be oriented closer and closer to the direction of the x axis (the direction of stretching). The exceptions are vectors along the y-axis, which will gradually shrink away to nothing.

Hyperbolic rotation

The eigenvalues are multiplicative inverses of each other.

Complex eigenvalues

A matrix with real entries may have no real eigenvalue: In the following example a complex number

$z=\cos \theta +i\sin \theta \ =\ \exp(i\theta )$ is employed. It lies on the unit circle in the complex plane. The rotation matrix

${\begin{pmatrix}\cos \theta &-\sin \theta \\\sin \theta &\cos \theta \end{pmatrix}}$ may be compared to the complex 1 × 1 matrix ${\begin{pmatrix}z\end{pmatrix}}.$

A 1 × 1 matrix is diagonal and thus expresses a relation of eigenvalue and eigenvector. It is common in algebraic geometry to extend the real field to the complex for enhanced expressions such as this one.

A rotation in a plane is a transformation that describes motion of a vector, plane, coordinates, etc., around a fixed point. Clearly, for rotations other than through 0° and 180°, every vector in the real plane will have its direction changed, and thus there cannot be any eigenvectors. But this is not necessarily true if we consider the same matrix over a complex vector space. The characteristic equation is a quadratic equation with discriminant D = 4 (cos² φ − 1) = − 4 sin² φ, which is a negative number whenever φ is not equal to a multiple of 180°. A rotation of 0°, 360°, … is just the identity transformation (a uniform scaling by +1), while a rotation of 180°, 540°, …, is a reflection (uniform scaling by -1). Otherwise, as expected, there are no real eigenvalues or eigenvectors for rotation in the plane. Instead, the eigenvalues are complex numbers in general. Although not diagonalizable over the reals, the rotation matrix is diagonalizable over the complex numbers, and again the eigenvalues appear on the diagonal. Thus rotation matrices acting on complex spaces can be thought of as scaling matrices, with complex scaling factors.

Calculation

The complexity of the problem for finding roots/eigenvalues of the characteristic polynomial increases rapidly with increasing the degree of the polynomial (the dimension of the vector space). There are exact solutions for dimensions below 5, but for dimensions greater than or equal to 5 there are generally no exact solutions and one has to resort to numerical methods to find them approximately. (In fact, since the roots of any polynomial can be expressed as eigenvalues of a companion matrix, the Abel–Ruffini theorem implies that there is no general algebraic solution for eigenvalues of 5×5 or larger matrices: any general eigenvalue algorithm is necessarily approximate, although in practice one can obtain any desired accuracy.<ref name=TrefethenBau/>) Worse, any computational procedure that starts by computing the coefficients of the characteristic polynomial can be very inaccurate in the presence of round-off error, because the roots of a polynomial are an extremely sensitive function of the coefficients (see Wilkinson's polynomial).<ref name=TrefethenBau>Template:Citation</ref> Efficient, accurate methods to compute eigenvalues and eigenvectors of arbitrary matrices were not known until the advent of the QR algorithm in 1961. <ref name=TrefethenBau/> Combining Householder transformation with LU decomposition offers better convergence than the QR algorithm.<ref>LU Householder Transformation</ref> For large Hermitian sparse matrices, the Lanczos algorithm is one example of an efficient iterative method to compute eigenvalues and eigenvectors, among several other possibilities.<ref name=TrefethenBau/>

History

Eigenvalues are often introduced in the context of linear algebra or matrix theory. Historically, however, they arose in the study of quadratic forms and differential equations.

Euler studied the rotational motion of a rigid body and discovered the importance of the principal axes. Lagrange realized that the principal axes are the eigenvectors of the inertia matrix.<ref>See Template:Harvnb</ref> In the early 19th century, Cauchy saw how their work could be used to classify the quadric surfaces, and generalized it to arbitrary dimensions.<ref name="hawkins3">See Template:Harvnb</ref> Cauchy also coined the term racine caractéristique (characteristic root) for what is now called eigenvalue; his term survives in characteristic equation.<ref name="kline807">See Template:Harvnb</ref>

Fourier used the work of Laplace and Lagrange to solve the heat equation by separation of variables in his famous 1822 book Théorie analytique de la chaleur.<ref>See Template:Harvnb</ref> Sturm developed Fourier's ideas further and brought them to the attention of Cauchy, who combined them with his own ideas and arrived at the fact that real symmetric matrices have real eigenvalues.<ref name="hawkins3" /> This was extended by Hermite in 1855 to what are now called Hermitian matrices.<ref name="kline807" /> Around the same time, Brioschi proved that the eigenvalues of orthogonal matrices lie on the unit circle,<ref name="hawkins3" /> and Clebsch found the corresponding result for skew-symmetric matrices.<ref name="kline807" /> Finally, Weierstrass clarified an important aspect in the stability theory started by Laplace by realizing that defective matrices can cause instability.<ref name="hawkins3" />

In the meantime, Liouville studied eigenvalue problems similar to those of Sturm; the discipline that grew out of their work is now called Sturm–Liouville theory.<ref>See Template:Harvnb</ref> Schwarz studied the first eigenvalue of Laplace's equation on general domains towards the end of the 19th century, while Poincaré studied Poisson's equation a few years later.<ref>See Template:Harvnb</ref>

At the start of the 20th century, Hilbert studied the eigenvalues of integral operators by viewing the operators as infinite matrices.<ref>See Template:Harvnb</ref> He was the first to use the German word eigen to denote eigenvalues and eigenvectors in 1904, though he may have been following a related usage by Helmholtz. For some time, the standard term in English was "proper value", but the more distinctive term "eigenvalue" is standard today.<ref>See Template:Harvnb</ref>

The first numerical algorithm for computing eigenvalues and eigenvectors appeared in 1929, when Von Mises published the power method. One of the most popular methods today, the QR algorithm, was proposed independently by John G.F. Francis<ref>Template:Citation and Template:Citation</ref> and Vera Kublanovskaya<ref>Template:Citation. Also published in: Template:Citation</ref> in 1961.<ref>See Template:Harvnb; Template:Harvnb</ref>

Generalizations

Left and right eigenvectors

Template:See also The word eigenvector formally refers to the right eigenvector  $x_{R}$. It is defined by the eigenvalue equation

$Ax_{R}=\lambda x_{R},\$

and is the most commonly used eigenvector. However, the left eigenvector $x_{L}$ exists as well, and is defined by

$x_{L}A=\lambda x_{L}.\$

In general, the left and right eigenvectors are usually different, matrix multiplication being noncommunative. But in the case of a Hermitian (or real symmetric) matrix $A$, the left and right eigenvectors are conjugate.

Infinite-dimensional spaces and spectral theory

Template:Details If the vector space is an infinite dimensional Banach space, the notion of eigenvalues can be generalized to the concept of spectrum. The spectrum is the set of scalars λ for which (T − λI)⁻¹ is not defined; that is, such that T − λI has no bounded inverse.

Clearly if λ is an eigenvalue of T, λ is in the spectrum of T. In general, the converse is not true. There are operators on Hilbert or Banach spaces that have no eigenvectors at all. This can be seen in the following example. The bilateral shift on the Hilbert space ℓ ²(Z) (that is, the space of all sequences of scalars … a₋₁, a₀, a₁, a₂, … such that

$\cdots +|a_{-1}|^{2}+|a_{0}|^{2}+|a_{1}|^{2}+|a_{2}|^{2}+\cdots \,$

converges) has no eigenvalue but does have spectral values.

In infinite-dimensional spaces, the spectrum of a bounded operator is always nonempty. This is also true for an unbounded self adjoint operator. Via its spectral measures, the spectrum of any self adjoint operator, bounded or otherwise, can be decomposed into absolutely continuous, pure point, and singular parts. (See Decomposition of spectrum.)

The hydrogen atom is an example where both types of spectra appear. The eigenfunctions of the hydrogen atom Hamiltonian are called eigenstates and are grouped into two categories. The bound states of the hydrogen atom correspond to the discrete part of the spectrum (they have a discrete set of eigenvalues that can be computed by Rydberg formula) while the ionization processes are described by the continuous part (the energy of the collision/ionization is not quantized).

Eigenfunctions

A common example of such maps on infinite dimensional spaces are the action of differential operators on function spaces. As an example, on the space of infinitely differentiable functions, the process of differentiation defines a linear operator since

$\displaystyle {\frac {d}{dt}}(af+bg)=a{\frac {df}{dt}}+b{\frac {dg}{dt}},$

where f(t) and g(t) are differentiable functions, and a and b are constants.

The eigenvalue equation for linear differential operators is then a set of one or more differential equations. The eigenvectors are commonly called eigenfunctions. The simplest case is the eigenvalue equation for differentiation of a real valued function by a single real variable. We seek a function (equivalent to an infinite-dimensional vector) that, when differentiated, yields a constant times the original function. In this case, the eigenvalue equation becomes the linear differential equation

$\displaystyle {\frac {d}{dx}}f(x)=\lambda f(x).$

Here λ is the eigenvalue associated with the function, f(x). This eigenvalue equation has a solution for any value of λ. If λ is zero, the solution is

$f(x)=A,\,$

where A is any constant; if λ is non-zero, the solution is the exponential function

$f(x)=Ae^{\lambda x}.\$

If we expand our horizons to complex valued functions, the value of λ can be any complex number. The spectrum of d/dt is therefore the whole complex plane. This is an example of a continuous spectrum.

Waves on a string

File:Standing wave.gif

The shape of a standing wave in a string fixed at its boundaries is an example of an eigenfunction of a differential operator. The admissible eigenvalues are governed by the length of the string and determine the frequency of oscillation.

The displacement, $h(x,t)$, of a stressed elastic chord fixed at both ends, like the vibrating strings of a string instrument, satisfies the wave equation

${\frac {\partial ^{2}h}{\partial t^{2}}}=c^{2}{\frac {\partial ^{2}h}{\partial x^{2}}},$

which is a linear partial differential equation, where c is the constant wave speed. The normal method of solving such an equation is separation of variables. If we assume that h can be written as the product of the form X(x)T(t), we can form a pair of ordinary differential equations:

$X''=-{\frac {\omega ^{2}}{c^{2}}}X$ and $T''=-\omega ^{2}T.\$

Each of these is an eigenvalue equation (the unfamiliar form of the eigenvalue is chosen merely for convenience). For any values of the eigenvalues, the eigenfunctions are given by

$X=\sin \left({\frac {\omega x}{c}}+\phi \right)\$ and $T=\sin(\omega t+\psi ).\$

If we impose boundary conditions (that the ends of the string are fixed with X(x) = 0 at x = 0 and x = L, for example) we can constrain the eigenvalues. For those boundary conditions, we find

$\sin(\phi )=0\$, and so the phase angle $\phi =0\$

and

$\sin \left({\frac {\omega L}{c}}\right)=0.\$

Thus, the constant $\omega$ is constrained to take one of the values $\omega _{n}={\frac {nc\pi }{L}}$, where n is any integer. Thus, the clamped string supports a family of standing waves of the form

$h(x,t)=\sin(n\pi x/L)\sin(\omega _{n}t).\$

From the point of view of our musical instrument, the frequency $\omega _{n}\$ is the frequency of the nth harmonic, which is called the (n-1)st overtone.

Associative algebras and representation theory

Template:Main More algebraically, rather than generalizing the vector space to an infinite dimensional space, one can generalize the algebraic object that is acting on the space, replacing a single operator acting on a vector space with an algebra representation – an associative algebra acting on a module. The study of such actions is the field of representation theory. To understand these representations, one breaks them into indecomposable representations, and, if possible, into irreducible representations; these correspond respectively to generalized eigenspaces and eigenspaces, or rather the indecomposable and irreducible components of these. While a single operator on a vector space can be understood in terms of eigenvectors – 1-dimensional invariant subspaces – in general in representation theory the building blocks (the irreducible representations) are higher-dimensional.

A closer analog of eigenvalues is given by the notion of a weight, with the analogs of eigenvectors and eigenspaces being weight vectors and weight spaces. For an associative algebra A over a field F, the analog of an eigenvalue is a one-dimensional representation $\lambda \colon A\to \mathbf {F}$ (a map of algebras; a linear functional that is also multiplicative), called the weight, rather than a single scalar. A map of algebras is used because if a vector is an eigenvector for two elements of an algebra, then it is also an eigenvector for any linear combination of these, and the eigenvalue is the corresponding linear combination of the eigenvalues, and likewise for multiplication. This is related to the classical eigenvalue as follows: a single operator T corresponds to the algebra F[T] (the polynomials in T), and a map of algebras $\mathbf {F} [T]\to \mathbf {F}$ is determined by its value on the generator T; this value is the eigenvalue. A vector v on which the algebra acts by this weight (i.e., by scalar multiplication, with the scalar determined by the weight) is called a weight vector, and other concepts generalize similarly. The generalization of a diagonalizable matrix (having an eigenbasis) is a weight module.

Because a weight is a map to a field, which is commutative, the map factors through the abelianization of the algebra A – equivalently, it vanishes on the derived algebra – in terms of matrices, if v is a common eigenvector of operators T and U, then $TUv=UTv$ (because in both cases it is just multiplication by scalars), so common eigenvectors of an algebra must be in the set on which the algebra acts commutatively (which is annihilated by the derived algebra). Thus of central interest are the free commutative algebras, namely the polynomial algebras. In this particularly simple and important case of the polynomial algebra $\mathbf {F} [T_{1},\dots ,T_{k}]$ in a set of commuting matrices, a weight vector of this algebra is a simultaneous eigenvector of the matrices, while a weight of this algebra is simply a k-tuple of scalars $\lambda =(\lambda _{1},\dots ,\lambda _{k})$ corresponding to the eigenvalue of each matrix, and hence geometrically to a point in k-space. These weights – in particularly their geometry – are of central importance in understanding the representation theory of Lie algebras, specifically the finite-dimensional representations of semisimple Lie algebras.

As an application of this geometry, given an algebra that is a quotient of a polynomial algebra on k generators, it corresponds geometrically to an algebraic variety in k-dimensional space, and the weight must fall on the variety – i.e., it satisfies defining equations for the variety. This generalizes the fact that eigenvalues satisfy the characteristic polynomial of a matrix in one variable.

Applications

Schrödinger equation

File:HAtomOrbitals.png

The wavefunctions associated with the bound states of an electron in a hydrogen atom can be seen as the eigenvectors of the hydrogen atom Hamiltonian as well as of the angular momentum operator. They are associated with eigenvalues interpreted as their energies (increasing downward: n=1,2,3,...) and angular momentum (increasing across: s, p, d,...). The illustration shows the square of the absolute value of the wavefunctions. Brighter areas correspond to higher probability density for a position measurement. The center of each figure is the atomic nucleus, a proton.

An example of an eigenvalue equation where the transformation T is represented in terms of a differential operator is the time-independent Schrödinger equation in quantum mechanics:

$H\psi _{E}=E\psi _{E}\,$

where H, the Hamiltonian, is a second-order differential operator and $\psi _{E}$, the wavefunction, is one of its eigenfunctions corresponding to the eigenvalue E, interpreted as its energy.

However, in the case where one is interested only in the bound state solutions of the Schrödinger equation, one looks for $\psi _{E}$ within the space of square integrable functions. Since this space is a Hilbert space with a well-defined scalar product, one can introduce a basis set in which $\psi _{E}$ and H can be represented as a one-dimensional array and a matrix respectively. This allows one to represent the Schrödinger equation in a matrix form.

Bra-ket notation is often used in this context. A vector, which represents a state of the system, in the Hilbert space of square integrable functions is represented by $|\Psi _{E}\rangle$. In this notation, the Schrödinger equation is:

$H|\Psi _{E}\rangle =E|\Psi _{E}\rangle$

where $|\Psi _{E}\rangle$ is an eigenstate of H. It is a self adjoint operator, the infinite dimensional analog of Hermitian matrices (see Observable). As in the matrix case, in the equation above $H|\Psi _{E}\rangle$ is understood to be the vector obtained by application of the transformation H to $|\Psi _{E}\rangle$.

Molecular orbitals

In quantum mechanics, and in particular in atomic and molecular physics, within the Hartree–Fock theory, the atomic and molecular orbitals can be defined by the eigenvectors of the Fock operator. The corresponding eigenvalues are interpreted as ionization potentials via Koopmans' theorem. In this case, the term eigenvector is used in a somewhat more general meaning, since the Fock operator is explicitly dependent on the orbitals and their eigenvalues. If one wants to underline this aspect one speaks of nonlinear eigenvalue problem. Such equations are usually solved by an iteration procedure, called in this case self-consistent field method. In quantum chemistry, one often represents the Hartree–Fock equation in a non-orthogonal basis set. This particular representation is a generalized eigenvalue problem called Roothaan equations.

Geology and glaciology

In geology, especially in the study of glacial till, eigenvectors and eigenvalues are used as a method by which a mass of information of a clast fabric's constituents' orientation and dip can be summarized in a 3-D space by six numbers. In the field, a geologist may collect such data for hundreds or thousands of clasts in a soil sample, which can only be compared graphically such as in a Tri-Plot (Sneed and Folk) diagram,<ref>Template:Citation</ref><ref>Template:Citation</ref> or as a Stereonet on a Wulff Net.<ref>Template:Citation</ref> The output for the orientation tensor is in the three orthogonal (perpendicular) axes of space. Eigenvectors output from programs such as Stereo32<ref>Stereo32</ref> are in the order E₁ ≥ E₂ ≥ E₃, with E₁ being the primary orientation of clast orientation/dip, E₂ being the secondary and E₃ being the tertiary, in terms of strength. The clast orientation is defined as the eigenvector, on a compass rose of 360°. Dip is measured as the eigenvalue, the modulus of the tensor: this is valued from 0° (no dip) to 90° (vertical). The relative values of E₁, E₂, and E₃ are dictated by the nature of the sediment's fabric. If E₁ = E₂ = E₃, the fabric is said to be isotropic. If E₁ = E₂ > E₃ the fabric is planar. If E₁ > E₂ > E₃ the fabric is linear. See 'A Practical Guide to the Study of Glacial Sediments' by Benn & Evans, 2004.<ref>Template:Citation</ref>

Principal components analysis

File:GaussianScatterPCA.png

PCA of the multivariate Gaussian distribution centered at (1,3) with a standard deviation of 3 in roughly the (0.878, 0.478) direction and of 1 in the orthogonal direction. The vectors shown are unit eigenvectors of the (symmetric, positive-semidefinite) covariance matrix scaled by the square root of the corresponding eigenvalue. (Just as in the one-dimensional case, the square root is taken because the standard deviation is more readily visualized than the variance.

The eigendecomposition of a symmetric positive semidefinite (PSD) matrix yields an orthogonal basis of eigenvectors, each of which has a nonnegative eigenvalue. The orthogonal decomposition of a PSD matrix is used in multivariate analysis, where the sample covariance matrices are PSD. This orthogonal decomposition is called principal components analysis (PCA) in statistics. PCA studies linear relations among variables. PCA is performed on the covariance matrix or the correlation matrix (in which each variable is scaled to have its sample variance equal to one). For the covariance or correlation matrix, the eigenvectors correspond to principal components and the eigenvalues to the variance explained by the principal components. Principal component analysis of the correlation matrix provides an orthonormal eigen-basis for the space of the observed data: In this basis, the largest eigenvalues correspond to the principal-components that are associated with most of the covariability among a number of observed data.

Principal component analysis is used to study large data sets, such as those encountered in data mining, chemical research, psychology, and in marketing. PCA is popular especially in psychology, in the field of psychometrics. In Q-methodology, the eigenvalues of the correlation matrix determine the Q-methodologist's judgment of practical significance (which differs from the statistical significance of hypothesis testing): The factors with eigenvalues greater than 1.00 are considered practically significant, that is, as explaining an important amount of the variability in the data, while eigenvalues less than 1.00 are considered practically insignificant, as explaining only a negligible portion of the data variability. More generally, principal component analysis can be used as a method of factor analysis in structural equation modeling.

Vibration analysis

File:Beam mode 1.gif

1st lateral bending (See vibration for more types of vibration)

Eigenvalue problems occur naturally in the vibration analysis of mechanical structures with many degrees of freedom. The eigenvalues are used to determine the natural frequencies (or eigenfrequencies) of vibration, and the eigenvectors determine the shapes of these vibrational modes. In particular, undamped vibration is governed by

$m{\ddot {x}}+kx=0$

or

$m{\ddot {x}}=-kx$

that is, acceleration is proportional to position (i.e., we expect x to be sinusoidal in time). In n dimensions, m becomes a mass matrix and k a stiffness matrix. Admissible solutions are then a linear combination of solutions to the generalized eigenvalue problem

$-kx=\omega ^{2}mx$

where $\omega ^{2}$ is the eigenvalue and $\omega$ is the angular frequency. Note that the principal vibration modes are different from the principal compliance modes, which are the eigenvectors of k alone. Furthermore, damped vibration, governed by

$m{\ddot {x}}+c{\dot {x}}+kx=0$

leads to what is called a so-called quadratic eigenvalue problem,

$(\omega ^{2}m+\omega c+k)x=0.$

This can be reduced to a generalized eigenvalue problem by clever algebra at the cost of solving a larger system.

The orthogonality properties of the eigenvectors allows decoupling of the differential equations so that the system can be represented as linear summation of the eigenvectors. The eigenvalue problem of complex structures is often solved using finite element analysis, but neatly generalize the solution to scalar-valued vibration problems.

Eigenfaces

File:Eigenfaces.png

Eigenfaces as examples of eigenvectors

Template:Main In image processing, processed images of faces can be seen as vectors whose components are the brightnesses of each pixel.<ref>Template:Citation</ref> The dimension of this vector space is the number of pixels. The eigenvectors of the covariance matrix associated with a large set of normalized pictures of faces are called eigenfaces; this is an example of principal components analysis. They are very useful for expressing any face image as a linear combination of some of them. In the facial recognition branch of biometrics, eigenfaces provide a means of applying data compression to faces for identification purposes. Research related to eigen vision systems determining hand gestures has also been made.

Similar to this concept, eigenvoices represent the general direction of variability in human pronunciations of a particular utterance, such as a word in a language. Based on a linear combination of such eigenvoices, a new voice pronunciation of the word can be constructed. These concepts have been found useful in automatic speech recognition systems, for speaker adaptation.

Tensor of moment of inertia

In mechanics, the eigenvectors of the moment of inertia tensor define the principal axes of a rigid body. The tensor of moment of inertia is a key quantity required to determine the rotation of a rigid body around its center of mass.

Stress tensor

In solid mechanics, the stress tensor is symmetric and so can be decomposed into a diagonal tensor with the eigenvalues on the diagonal and eigenvectors as a basis. Because it is diagonal, in this orientation, the stress tensor has no shear components; the components it does have are the principal components.

Eigenvalues of a graph

In spectral graph theory, an eigenvalue of a graph is defined as an eigenvalue of the graph's adjacency matrix A, or (increasingly) of the graph's Laplacian matrix (see also Discrete Laplace operator), which is either T−A (sometimes called the Combinatorial Laplacian) or I−T^−1/2AT^−1/2 (sometimes called the Normalized Laplacian), where T is a diagonal matrix with T_v,v equal to the degree of vertex v, and in T^−1/2, the v^th diagonal entry is deg(v)^−1/2. The k^th principal eigenvector of a graph is defined as either the eigenvector corresponding to the k^th largest or k^th smallest eigenvalue of the Laplacian. The first principal eigenvector of the graph is also referred to merely as the principal eigenvector.

The principal eigenvector is used to measure the centrality of its vertices. An example is Google's PageRank algorithm. The principal eigenvector of a modified adjacency matrix of the World Wide Web graph gives the page ranks as its components. This vector corresponds to the stationary distribution of the Markov chain represented by the row-normalized adjacency matrix; however, the adjacency matrix must first be modified to ensure a stationary distribution exists. The second smallest eigenvector can be used to partition the graph into clusters, via spectral clustering. Other methods are also available for clustering.

Basic reproduction number

See Basic reproduction number

The basic reproduction number ($R_{0}$) is a fundamental number in the study of how infectious diseases spread. If one infectious person is put into a population of completely susceptible people, then $R_{0}$ is the average number of people that one infectious person will infect. The generation time of an infection is the time, $t_{G}$, from one person becoming infected to the next person becoming infected. In a heterogenous population, the next generation matrix defines how many people in the population will become infected after time $t_{G}$ has passed. $R_{0}$ is then the largest eigenvalue of the next generation matrix.<ref>Template:Citation</ref><ref>Template:Citation</ref>

See also

• Nonlinear eigenproblem
• Quadratic eigenvalue problem
• Introduction to eigenstates
• Eigenplane
• Jordan normal form
• List of numerical analysis software
• Antieigenvalue theory

References

• Template:Citation.
• Template:Citation.
• Template:Citation.
• Template:Citation
• Template:Citation.
• Template:Citation.
• Template:Citation.
• Template:Citation.
• Template:Citation.
• Template:Citation.
• Template:Citation.
• Template:Citation.
• Template:Citation.
• Template:Citation.
• Template:Citation.
• Template:Citation.
• Template:Citation.
• Template:Citation.
• Template:Citation.
• Template:Citation.
• Template:Citation.
• Template:Citation.
• Template:Citation.
• Template:Citation.
• Template:Citation.
• Template:Citation.
• Template:Citation.
• Template:Citation.
• Pigolkina, T. S. and Shulman, V. S., Eigenvalue (in Russian), In:Vinogradov, I. M. (Ed.), Mathematical Encyclopedia, Vol. 5, Soviet Encyclopedia, Moscow, 1977.
• Template:Citation.
• Template:Citation.
• Curtis, Charles W., Linear Algebra: An Introductory Approach, 347 p., Springer; 4th ed. 1984. Corr. 7th printing edition (August 19, 1999), ISBN 0-387-90992-3.
• Template:Citation.
• Template:Citation.
• Template:Citation.

External links

• What are Eigen Values? — non-technical introduction from PhysLink.com's "Ask the Experts"
• Eigen Values and Eigen Vectors Numerical Examples – Tutorial and Interactive Program from Revoledu.
• Introduction to Eigen Vectors and Eigen Values – lecture from Khan Academy

Theory

• Template:Springer
• Template:Springer
• Template:Planetmath reference
• Eigenvector — Wolfram MathWorld
• Eigen Vector Examination working applet
• Same Eigen Vector Examination as above in a Flash demo with sound
• Computation of Eigenvalues
• Numerical solution of eigenvalue problems Edited by Zhaojun Bai, James Demmel, Jack Dongarra, Axel Ruhe, and Henk van der Vorst
• Eigenvalues and Eigenvectors on the Ask Dr. Math forums: [1], [2]

Online calculators

• arndt-bruenner.de
• bluebit.gr
• wims.unice.fr

Demonstration applets

• Java applet about eigenvectors in the real plane