In mathematics, a matrix norm is a vector norm in a vector space whose elements (vectors) are matrices (of given dimensions).
In what follows, will denote a field of either real or complex numbers.
Let denote the vector space of all matrices of size (with rows and columns) with entries in the field .
A matrix norm is a norm on the vector space . Thus, the matrix norm is a function that must satisfy the following properties:
For all scalars in and for all matrices and in ,
- (being absolutely homogeneous)
- (being sub-additive or satisfying the triangle inequality)
- (being positive-valued)
- iff (being definite)
Additionally, in the case of square matrices (thus, m = n), some (but not all) matrix norms satisfy the following condition, which is related to the fact that matrices are more than just vectors:
- for all matrices and in
A matrix norm that satisfies this additional property is called a sub-multiplicative norm (in some books, the terminology matrix norm is used only for those norms which are sub-multiplicative). The set of all matrices, together with such a sub-multiplicative norm, is an example of a Banach algebra.
The definition of sub-multiplicativity is sometimes extended to non-square matrices, for instance in the case of the induced p-norm, where for and holds that . Here and are the norms induced from and , respectively, and p,q ≥ 1.
There are three types of matrix norms which will be discussed below:
- Matrix norms induced by vector norms,
- Entrywise matrix norms, and
- Schatten norms.
Matrix norms induced by vector norms
Suppose a vector norm on is given (for K a field). Any matrix A induces a linear operator from to with respect to the standard basis, and one defines the corresponding induced norm or operator norm on the space of all matrices as follows:
In particular, if the p-norm for vectors (1 ≤ p ≤ ∞) is used for both spaces and ,
then the corresponding induced operator norm is:
These induced norms are different from the "entrywise" p-norms and the Schatten p-norms for matrices treated below, which are also usually denoted by
- Note: We have described above the induced operator norm when the same vector norm was used in the "departure space" and the "arrival space" of the operator . This is not a necessary restriction. More generally, given a norm on , and a norm on , one can define a matrix norm on induced by these norms:
- The matrix norm is sometimes called a subordinate norm. Subordinate norms are consistent with the norms that induce them, giving
Any induced operator norm is a sub-multiplicative matrix norm: ; this follows from
Moreover, any induced norm satisfies the inequality
‹The template Repeat is being considered for deletion.›
‹The template Repeat is being considered for deletion.›
where ρ(A) is the spectral radius of A. For symmetric or hermitian A, we have equality in (1) for the 2-norm, since in this case the 2-norm is precisely the spectral radius of A. For an arbitrary matrix, we may not have equality for any norm; a counterexample being given by , which has vanishing spectral radius. In any case, for square matrices
we have the spectral radius formula:
In the special cases of the induced matrix norms can be computed or estimated by
which is simply the maximum absolute column sum of the matrix;
which is simply the maximum absolute row sum of the matrix;
where in left hand side represents the largest singular value of matrix , and on the right hand side is the Frobenius norm. The first inequality can be derived from the fact that the trace of a matrix is equal to the sum of its eigenvalues. The equality holds if and only if the matrix is a rank-one matrix or a zero matrix.
For example, if the matrix is defined by
then we have
In the special case of (the Euclidean norm or -norm for vectors), the induced matrix norm is the spectral norm.
The spectral norm of a matrix is the largest singular value of i.e. the square root of the largest eigenvalue of the positive-semidefinite matrix :
where denotes the conjugate transpose of .
"Entrywise" matrix norms
These norms treat an matrix as a vector of size , and
use one of the familiar vector norms.
For example, using the p-norm for vectors, p ≥ 1, we get:
This is a different norm from the induced p-norm (see above) and the Schatten p-norm (see below), but the notation is the same.
The special case p = 2 is the Frobenius norm, and p = ∞ yields the maximum norm.
L2,1 and Lp,q norms
Let be the columns of matrix . The norm is the sum of the Euclidean norms of the columns of the matrix:
The norm as an error function is more robust
since the error for each data point (a column) is not squared. It is used in robust data analysis and sparse coding.
The norm can be generalized to the norm, p, q ≥ 1, defined by
When p = q = 2 for the norm, it is called the Frobenius norm or the Hilbert–Schmidt norm, though the latter term is used more frequently in the context of operators on (possibly infinite-dimensional) Hilbert space. This norm can be defined in various ways:
where are the singular values of . Recall that the trace function returns the sum of diagonal entries of a square matrix.
The Frobenius norm is the Euclidean norm on and comes from the Frobenius inner product on the space of all matrices.
The Frobenius norm is sub-multiplicative and is very useful for numerical linear algebra. This norm is often easier to compute than induced norms and has the useful property of being invariant under rotations, that is, for any rotation matrix . This property follows from the trace definition restricted to real matrices:
where we have used the orthogonal nature of (that is, ) and the cyclic nature of the trace (). More generally the norm is invariant under a unitary transformation for complex matrices.
It also satisfies
where is the Frobenius inner product.
The max norm is the elementwise norm with p = q = ∞:
This norm is not sub-multiplicative.
The Schatten p-norms arise when applying the p-norm to the vector of singular values of a matrix. If the singular values are denoted by σi, then the Schatten p-norm is defined by
These norms again share the notation with the induced and entrywise p-norms, but they are different.
All Schatten norms are sub-multiplicative. They are also unitarily invariant, which means that for all matrices and all unitary matrices and .
The most familiar cases are p = 1, 2, ∞. The case p = 2 yields the Frobenius norm, introduced before. The case p = ∞ yields the spectral norm, which is the operator norm induced by the vector 2-norm (see above). Finally, p = 1 yields the nuclear norm (also known as the trace norm, or the Ky Fan 'n'-norm), defined as
(Here denotes a positive semidefinite matrix such that . More precisely, since is a positive semidefinite matrix, its square root is well-defined.)
A matrix norm on is called consistent with a vector norm on and a vector norm on if:
for all . All induced norms are consistent by definition.
A matrix norm on is called compatible with a vector norm on if:
for all . Induced norms are compatible by definition.
Equivalence of norms
For any two matrix norms and , we have
for some positive numbers r and s, for all matrices A in . In other words, all norms on are equivalent; they induce the same topology on . This is true because the vector space has the finite dimension .
Moreover, for every vector norm on , there exists a unique positive real number such that is a sub-multiplicative matrix norm for every .
A sub-multiplicative matrix norm is said to be minimal if there exists no other sub-multiplicative matrix norm satisfying .
Examples of norm equivalence
Let once again refer to the norm induced by the vector p-norm (as above in the Induced Norm section).
For matrix of rank , the following inequalities hold:
Another useful inequality between matrix norms is
which is a special case of Hölder's inequality.
- ^ Carl D. Meyer,
Matrix Analysis and Applied Linear Algebra, §5.2, p.281,
Society for Industrial & Applied Mathematics, June 2000.
- ^ Ding, Chris; Zhou, Ding; He, Xiaofeng; Zha, Hongyuan (June 2006). "R1-PCA: Rotational Invariant L1-norm Principal Component Analysis for Robust Subspace Factorization". Proceedings of the 23rd International Conference on Machine Learning. ICML '06. Pittsburgh, Pennsylvania, USA: ACM. pp. 281–288. doi:10.1145/1143844.1143880. ISBN 1-59593-383-2.
- ^ Fan, Ky. (1951). "Maximum properties and inequalities for the eigenvalues of completely continuous operators". Proceedings of the National Academy of Sciences of the United States of America. 37 (11): 760–766. doi:10.1073/pnas.37.11.760. PMC 1063464 .
Golub, Gene; Charles F. Van Loan (1996). Matrix Computations – Third Edition. Baltimore: The Johns Hopkins University Press, 56–57. ISBN 0-8018-5413-X.
Roger Horn and Charles Johnson. Matrix Analysis, Chapter 5, Cambridge University Press, 1985. ISBN 0-521-38632-2.
- James W. Demmel, Applied Numerical Linear Algebra, section 1.7, published by SIAM, 1997.
- Carl D. Meyer, Matrix Analysis and Applied Linear Algebra, published by SIAM, 2000. 
- John Watrous, Theory of Quantum Information, 2.3 Norms of operators, lecture notes, University of Waterloo, 2011.
- Kendall Atkinson, An Introduction to Numerical Analysis, published by John Wiley & Sons, Inc 1989