Next: Matrix decompositions Up: Matrix decompositions and latent Previous: Matrix decompositions and latent   Contents   Index

# Linear algebra review

We briefly review some necessary background in linear algebra. Let be an matrix with real-valued entries; for a term-document matrix, all entries are in fact non-negative.

The rank of a matrix is the number of linearly independent rows (or columns) in it; thus, . A square matrix all of whose off-diagonal entries are zero is called a diagonal matrix; its rank is equal to the number of non-zero diagonal entries. If all diagonal entries of such a diagonal matrix are , it is called the identity matrix of dimension and represented by .

For a square matrix and a vector that is not all zeros, the values of satisfying

 (213)

are called the eigenvalues of . The -vector satisfying Equation 213 for an eigenvalue is the corresponding right eigenvector. The eigenvector corresponding to the eigenvalue of largest magnitude is called the principal eigenvector. In a similar fashion, the left eigenvectors of are the -vectors such that
 (214)

The number of non-zero eigenvalues of is at most .

The eigenvalues of a matrix are found by solving the characteristic equation, which is obtained by rewriting Equation 213 in the form . The eigenvalues of are then the solutions of , where denotes the determinant of a square matrix . The equation is an th order polynomial equation in and can have at most roots, which are the eigenvalues of . These eigenvalues can in general be complex, even if all entries of are real.

We now examine some further properties of eigenvalues and eigenvectors, to set up the central idea of singular value decompositions in Section 18.2 below. First, we look at the relationship between matrix-vector multiplication and eigenvalues.

Worked example. Consider the matrix

 (215)

Clearly the matrix has rank 3, and has 3 non-zero eigenvalues and , with the three corresponding eigenvectors
 (216)

For each of the eigenvectors, multiplication by acts as if we were multiplying the eigenvector by a multiple of the identity matrix; the multiple is different for each eigenvector. Now, consider an arbitrary vector, such as We can always express as a linear combination of the three eigenvectors of ; in the current example we have
 (217)

Suppose we multiply by :
 (218) (219) (220) (221)

End worked example.

Example 18.1 shows that even though is an arbitrary vector, the effect of multiplication by is determined by the eigenvalues and eigenvectors of . Furthermore, it is intuitively apparent from Equation 221 that the product is relatively unaffected by terms arising from the small eigenvalues of ; in our example, since , the contribution of the third term on the right hand side of Equation 221 is small. In fact, if we were to completely ignore the contribution in Equation 221 from the third eigenvector corresponding to , then the product would be computed to be rather than the correct product which is ; these two vectors are relatively close to each other by any of various metrics one could apply (such as the length of their vector difference).

This suggests that the effect of small eigenvalues (and their eigenvectors) on a matrix-vector product is small. We will carry forward this intuition when studying matrix decompositions and low-rank approximations in Section 18.2 . Before doing so, we examine the eigenvectors and eigenvalues of special forms of matrices that will be of particular interest to us.

For a symmetric matrix , the eigenvectors corresponding to distinct eigenvalues are orthogonal. Further, if is both real and symmetric, the eigenvalues are all real.

Worked example. Consider the real, symmetric matrix

 (222)

From the characteristic equation , we have the quadratic , whose solutions yield the eigenvalues and . The corresponding eigenvectors and are orthogonal. End worked example.

Subsections

Next: Matrix decompositions Up: Matrix decompositions and latent Previous: Matrix decompositions and latent   Contents   Index