Diagonal matrix: Difference between revisions

Revision as of 06:58, 26 February 2023

Latest revision as of 23:43, 27 June 2025

(25 intermediate revisions by 15 users not shown)

Line 1:

{{Use American English|date = March 2019}}

{{Short description|Matrix whose only nonzero elements are on its main diagonal}}

{{More footnotes needed|date=June 2025}}

In [[linear algebra]], a '''diagonal matrix''' is a [[matrix (mathematics)|matrix]] in which the entries outside the [[main diagonal]] are all zero; the term usually refers to [[square matrices]]. Elements of the main diagonal can either be zero or nonzero. An example of a 2×2 diagonal matrix is <math>\left[\begin{smallmatrix}

3 & 0 \\

Line 6 ⟶ 8:

\left[\begin{smallmatrix}

6 & 0 & 0 \\

0 & 05 & 0 \\

0 & 0 & 04

\end{smallmatrix}\right]</math>. An [[identity matrix]] of any size, or any multiple of it (is a diagonal matrix called a [[#Scalar matrix|''scalar matrix'']]), isfor aexample, diagonal matrix.<math>\left[\begin{smallmatrix}

0.5 & 0 \\

0 & 0.5 \end{smallmatrix}\right]</math>.

AIn [[geometry]], a diagonal matrix ismay sometimesbe calledused as a ''[[scaling matrix]]'', since matrix multiplication with it results in changing scale (size). Itsand determinantpossibly isalso the[[shape]]; productonly ofa itsscalar diagonalmatrix valuesresults in uniform change in scale.

==Definition==

As stated above, a diagonal matrix is a matrix in which all off-diagonal entries are zero. That is, the matrix {{math|1='''D''' = (''d''_''i'',''j'')}} with ''{{mvar|n''}} columns and ''{{mvar|n''}} rows is diagonal if

<math display="block">\forall i,j \in \{1, 2, \ldots, n\}, i \ne j \implies d_{i,j} = 0.</math>

However, the main diagonal entries are unrestricted.

The term ''diagonal matrix'' may sometimes refer to a '''{{visible anchor|rectangular diagonal matrix}}''', which is an ''{{mvar|m''}}-by-''{{mvar|n''}} matrix with all the entries not of the form {{math|''d''_''i'',''i''}} being zero. For example:

:<math display=block>\begin{bmatrix}

1 & 0 & 0\\

0 & 4 & 0\\

0 & 0 & -3\\

0 & 0 & 0\\

\end{bmatrix}</math> \quad \text{or} \quad <math>\begin{bmatrix}

1 & 0 & 0 & 0 & 0\\

0 & 4 & 0& 0 & 0\\

0 & 0 & -3& 0 & 0

\end{bmatrix}</math>

Line 46 ⟶ 49:

==Vector-to-matrix diag operator==

A diagonal matrix <{{math>\mathbf{|'''D'''}}</math> can be constructed from a vector <math>\mathbf{a} = \begin{bmatrix}a_1 & \dotsmdots & a_n\end{bmatrix}^\textsf{T}</math> using the <math>\operatorname{diag}</math> operator:

<math display="block">
\mathbf{D} = \operatorname{diag}(a_1, \dots, a_n).
</math>

This may be written more compactly as <math>\mathbf{D} = \operatorname{diag}(\mathbf{a})</math>.

The same operator is also used to represent [[Block matrix#Block diagonal matrices|block diagonal matrices]] as <math> \mathbf{A} = \operatorname{diag}(\mathbf A_1, \dots, \mathbf A_n)</math> where each argument <{{math>A_i</math>|'''A'''{{sub|''i''}}}} is a matrix.

The <math>\operatorname{{math|diag}</math>} operator may be written as:

<math display="block">
\operatorname{diag}(\mathbf{a}) = \left(\mathbf{a} \mathbf{1}^\textsf{T}\right) \circ \mathbf{I},
</math>

where <math>\circ</math> represents the [[Hadamard product (matrices)|Hadamard product]], and <math>\mathbf{{math|'''1'''}}</math> is a constant vector with elements 1.

==Matrix-to-vector diag operator==

The inverse matrix-to-vector <{{math>\operatorname{|diag}</math>} operator is sometimes denoted by the identically named <math>\operatorname{diag}(\mathbf{D}) = \begin{bmatrix}a_1 & \dotsmdots & a_n\end{bmatrix}^\textsf{T},</math> where the argument is now a matrix, and the result is a vector of its diagonal entries.

The following property holds:

<math display="block">
\operatorname{diag}(\mathbf{A}\mathbf{B}) = \sum_j \left(\mathbf{A} \circ \mathbf{B}^\textsf{T}\right)_{ij} = \left( \mathbf{A} \circ \mathbf{B}^\textsf{T} \right) \mathbf{1}.
</math>

== Scalar matrix ==

{{Confusing|section|reason=many sentences use incorrect, awkward grammar and should be reworded to make sense|date=February 2021}}

<!-- Linked from [[Scalar matrix]] and [[Scalar transformation]] -->

A diagonal matrix with equal diagonal entries is a '''scalar matrix'''; that is, a scalar multiple ''{{mvar|λ''}} of the [[identity matrix]] {{mvarmath|'''I'''}}. Its effect on a [[vector (mathematics and physics)|vector]] is [[scalar multiplication]] by ''{{mvar|λ''}}. For example, a 3×3 scalar matrix has the form:

<math display="block">

\begin{bmatrix}

Line 76 ⟶ 84:

</math>

The scalar matrices are the [[center of an algebra|center]] of the algebra of matrices: that is, they are precisely the matrices that [[commute (mathematics)|commute]] with all other square matrices of the same size.{{efn|Proof: given the [[elementary matrix]] <math>e_{ij}</math>, <math>Me_{ij}</math> is the matrix with only the ''i''-th row of ''M'' and <math>e_{ij}M</math> is the square matrix with only the ''M'' ''j''-th column, so the non-diagonal entries must be zero, and the ''i''th diagonal entry much equal the ''j''th diagonal entry.}} By contrast, over a [[field (mathematics)|field]] (like the real numbers), a diagonal matrix with all diagonal elements distinct only commutes with diagonal matrices (its [[centralizer]] is the set of diagonal matrices). That is because if a diagonal matrix <math>\mathbf{D} = \operatorname{diag}(a_1, \dots, a_n)</math> has <math>a_i \neq a_j,</math> then given a matrix <{{math>|'''M</math>'''}} with <math>m_{ij} \neq 0,</math> the <{{math>|(''i'', ''j'')</math>}} term of the products are: <math>(\mathbf{DM})_{ij} = a_im_{ij}</math> and <math>(\mathbf{MD})_{ij} = m_{ij}a_j,</math> and <math>a_jm_{ij} \neq m_{ij}a_i</math> (since one can divide by <math>m_{{mvar|m{{sub|ij}</math>}}}), so they do not commute unless the off-diagonal terms are zero.{{efn|Over more general rings, this does not hold, because one cannot always divide.}} Diagonal matrices where the diagonal entries are not all equal or all distinct have centralizers intermediate between the whole space and only diagonal matrices.<ref>{{cite web |url=https://math.stackexchange.com/q/1697991 |title=Do Diagonal Matrices Always Commute? |author=<!--Not stated--> |date=March 15, 2016 |publisher=Stack Exchange |access-date=August 4, 2018 }}</ref>

For an abstract vector space ''{{mvar|V''}} (rather than the concrete vector space <math>{{mvar|K^{{sup|n</math>}}}}), the analog of scalar matrices are '''scalar transformations'''. This is true more generally for a [[module (ring theory)|module]] ''{{mvar|M''}} over a [[ring (algebra)|ring]] ''{{mvar|R''}}, with the [[endomorphism algebra]] {{math|End(''M'')}} (algebra of linear operators on ''{{mvar|M''}}) replacing the algebra of matrices. Formally, scalar multiplication is a linear map, inducing a map <math>R \to \operatorname{End}(M),</math> (from a scalar ''{{mvar|&lambda;''}} to its corresponding scalar transformation, multiplication by ''{{mvar|&lambda;''}}) exhibiting {{math|End(''M'')}} as a ''{{mvar|R''}}-[[Algebra (ring theory)|algebra]]. For vector spaces, the scalar transforms are exactly the [[center of a ring|center]] of the endomorphism algebra, and, similarly, scalar invertible transforms are the center of the [[general linear group]] {{math|GL(''V'')}}. The former is more generally true [[free module]]s <math>M \cong R^n,</math>, for which the endomorphism algebra is isomorphic to a matrix algebra.

== Vector operations ==

Multiplying a vector by a diagonal matrix multiplies each of the terms by the corresponding diagonal entry. Given a diagonal matrix <math>\mathbf{D} = \operatorname{diag}(a_1, \dots, a_n)</math> and a vector <math>\mathbf{v} = \begin{bmatrix} x_1 & \dotsm & x_n \end{bmatrix}^\textsf{T}</math>, the product is:

<math display="block">\mathbf{D}\mathbf{v} = \operatorname{diag}(a_1, \dots, a_n)\begin{bmatrix}x_1 \\ \vdots \\ x_n\end{bmatrix} =

\begin{bmatrix}

a_1 \\

Line 94 ⟶ 102:

This can be expressed more compactly by using a vector instead of a diagonal matrix, <math>\mathbf{d} = \begin{bmatrix} a_1 & \dotsm & a_n \end{bmatrix}^\textsf{T}</math>, and taking the [[Hadamard product (matrices)|Hadamard product]] of the vectors (entrywise product), denoted <math>\mathbf{d} \circ \mathbf{v}</math>:

<math display="block">\mathbf{D}\mathbf{v} = \mathbf{d} \circ \mathbf{v} =

\begin{bmatrix} a_1 \\ \vdots \\ a_n \end{bmatrix} \circ \begin{bmatrix} x_1 \\ \vdots \\ x_n \end{bmatrix} =

\begin{bmatrix} a_1 x_1 \\ \vdots \\ a_n x_n \end{bmatrix}.

</math>

This is mathematically equivalent, but avoids storing all the zero terms of this [[sparse matrix]]. This product is thus used in [[machine learning]], such as computing products of derivatives in [[backpropagation]] or multiplying IDF weights in [[TF-IDF]],<ref>{{cite book |last=Sahami |first=Mehran |date=2009-06-15 |title=Text Mining: Classification, Clustering, and Applications |url=https://wwwbooks.google.com/books/edition/Text_Mining/?id=BnvYaYhMl-MC?gbpv=1&pg=PA14 |publisher=CRC Press |page=14 |isbn=9781420059458}}</ref> since some [[BLAS]] frameworks, which multiply matrices efficiently, do not include Hadamard product capability directly.<ref>{{cite web |url=https://stackoverflow.com/questions/7621520/element-wise-vector-vector-multiplication-in-blas |title=Element-wise vector-vector multiplication in BLAS? |author=<!--Not stated--> |date=2011-10-01 |website=stackoverflow.com |access-date=2020-08-30}}</ref>

== Matrix operations ==

The operations of matrix addition and [[matrix multiplication]] are especially simple for diagonal matrices. Write {{math|diag(''a''₁, ..., ''a''_''n'''')}} for a diagonal matrix whose diagonal entries starting in the upper left corner are {{math|''a''₁, ..., ''a''_''n''''}}. Then, for [[matrix addition|addition]], we have

<math display=block>

:{{math|diag(''a''₁, ..., ''a''_''n'')}} + {{math|diag(''b''₁, ..., ''b''_''n'')}} = {{math|diag(''a''₁ + ''b''₁, ..., ''a''_''n'' + ''b''_''n'')}}

\operatorname{diag}(a_1,\, \ldots,\, a_n) + \operatorname{diag}(b_1,\, \ldots,\, b_n) = \operatorname{diag}(a_1 + b_1,\, \ldots,\, a_n + b_n)</math>

and for [[matrix multiplication]],

<math display=block>\operatorname{diag}(a_1,\, \ldots,\, a_n) \operatorname{diag}(b_1,\, \ldots,\, b_n) = \operatorname{diag}(a_1 b_1,\, \ldots,\, a_n b_n).</math>

:{{math|diag(''a''₁, ..., ''a''_''n'')}} {{math|diag(''b''₁, ..., ''b''_''n'')}} = {{math|diag(''a''₁''b''₁, ..., ''a''_''n''''b''_''n'')}}.

The diagonal matrix {{math|diag(''a''₁, ..., ''a''_''n'''')}} is [[invertible matrix|invertible]] [[if and only if]] the entries {{math|''a''₁, ..., ''a''_''n''''}} are all nonzero. In this case, we have

<math display=block>\operatorname{diag}(a_1,\, \ldots,\, a_n)^{-1} = \operatorname{diag}(a_1^{-1},\, \ldots,\, a_n^{-1}).</math>

:{{math|diag(''a''₁, ..., ''a''_''n'')⁻¹}} = {{math|diag(''a''₁⁻¹, ..., ''a''_''n''⁻¹)}}.

In particular, the diagonal matrices form a [[subring]] of the ring of all ''{{mvar|n''}}-by-''{{mvar|n''}} matrices.

Multiplying an ''{{mvar|n''}}-by-''{{mvar|n''}} matrix {{mvarmath|'''A'''}} from the ''left'' with {{math|diag(''a''₁, ..., ''a''_''n'''')}} amounts to multiplying the {{mvar|i}}-th ''row'' of {{mvarmath|'''A'''}} by {{mathmvar|''a''_''i''}} for all {{mvar|i}}; multiplying the matrix {{mvarmath|'''A'''}} from the ''right'' with {{math|diag(''a''₁, ..., ''a''_''n'''')}} amounts to multiplying the {{mvar|i}}-th ''column'' of {{mvarmath|'''A'''}} by {{mathmvar|''a''_''i''}} for all {{mvar|i}}.

== Operator matrix in eigenbasis ==

{{Main|Transformation matrix#Finding the matrix of a transformation|Eigenvalues and eigenvectors}}

As explained in [[transformation matrix#Finding the matrix of a transformation|determining coefficients of operator matrix]], there is a special basis, {{math|'''e'''₁, ..., '''e'''_''n''}}, for which the matrix <{{math>|'''A</math>'''}} takes the diagonal form. Hence, in the defining equation <math display="inline">A \mathbf e_j{Ae}_j = \sum_i a_{i,j} \mathbf e_i</math>, all coefficients <math>a_{{mvar|a{{sub|i, j} </math>}}} with {{math|''i'' ≠ ''j''}} are zero, leaving only one term per sum. The surviving diagonal elements, <math>a_{{mvar|a{{sub|i,i j}}}}</math>, are known as '''eigenvalues''' and designated with <math>\lambda_i</math>{{mvar|&lambda;{{sub|i}}}} in the equation, which reduces to <math>A \mathbf e_i{Ae}_i = \lambda_i \mathbf e_i.</math>. The resulting equation is known as '''eigenvalue equation'''<ref>{{cite book |last=Nearing |first=James |year=2010 |title=Mathematical Tools for Physics |url=http://www.physics.miami.edu/nearing/mathmethods |chapter=Chapter 7.9: Eigenvalues and Eigenvectors |publisher=Dover Publications |chapter-url= http://www.physics.miami.edu/~nearing/mathmethods/operators.pdf |access-date=January 1, 2012|isbn=978-0486482125}}</ref> and used to derive the [[characteristic polynomial]] and, further, [[eigenvalues and eigenvectors]].

In other words, the [[eigenvalue]]s of {{math|diag(''λ''₁, ..., ''λ''_''n'')}} are {{math|''λ''₁, ..., ''λ''_''n''}} with associated [[eigenvectors]] of {{math|'''e'''₁, ..., '''e'''_''n''}}.

Line 133 ⟶ 142:

** A matrix is diagonal if and only if it is both [[triangular matrix|upper-]] and [[triangular matrix|lower-triangular]].

** A diagonal matrix is [[symmetric matrix|symmetric]].

* The [[identity matrix]] {{math|'''I'''_''n''}} and [[zero matrix]] are diagonal.

* A 1×1 matrix is always diagonal.

* The square of a 2×2 matrix with zero [[trace (linear algebra)|trace]] is always diagonal.

== Applications ==

Diagonal matrices occur in many areas of linear algebra. Because of the simple description of the matrix operation and eigenvalues/eigenvectors given above, it is typically desirable to represent a given matrix or [[linear operator|linear map]] by a diagonal matrix.

In fact, a given ''{{mvar|n''}}-by-''{{mvar|n''}} matrix {{mvarmath|'''A'''}} is [[similar matrix|similar]] to a diagonal matrix (meaning that there is a matrix {{mvarmath|'''X'''}} such that {{math|'''X'''⁻¹'''AX'''}} is diagonal) if and only if it has {{mvar|n}} [[linearly independent]] eigenvectors. Such matrices are said to be [[diagonalizable matrix|diagonalizable]].

Over the [[field (mathematics)|field]] of [[real number|real]] or [[complex number|complex]] numbers, more is true. The [[spectral theorem]] says that every [[normal matrix]] is [[matrix similarity|unitarily similar]] to a diagonal matrix (if {{math|1='''AA'''^∗ = '''A'''^∗'''A'''}} then there exists a [[unitary matrix]] {{mvarmath|'''U'''}} such that {{math|'''UAU'''^∗}} is diagonal). Furthermore, the [[singular value decomposition]] implies that for any matrix {{mvarmath|'''A'''}}, there exist unitary matrices {{mvarmath|'''U'''}} and {{mvarmath|'''V'''}} such that {{math|'''U'''^∗'''AV'''}} is diagonal with positive entries.

== Operator theory ==