Kronecker product

Template:Short description Template:CS1 config In mathematics, the Kronecker product, sometimes denoted by ⊗, is an operation on two matrices of arbitrary size resulting in a block matrix. It is a specialization of the tensor product (which is denoted by the same symbol) from vectors to matrices and gives the matrix of the tensor product linear map with respect to a standard choice of basis. The Kronecker product is to be distinguished from the usual matrix multiplication, which is an entirely different operation. The Kronecker product is also sometimes called matrix direct product.

The Kronecker product is named after the German mathematician Leopold Kronecker (1823–1891), even though there is little evidence that he was the first to define and use it. The Kronecker product has also been called the Zehfuss matrix, and the Zehfuss product, after Template:Ill, who in 1858 described this matrix operation, but Kronecker product is currently the most widely used term. The misattribution to Kronecker rather than Zehfuss was due to Kurt Hensel.

Definition

If A is an m × n matrix and B is a p × q matrix, then the Kronecker product A ⊗ B is the pm × qn block matrix:

$\mathbf{𝐀}\otimes \mathbf{𝐁}=\left[\begin{array}{ccc}{a}_{11}\mathbf{𝐁}& \cdots & a_{1n}\mathbf{𝐁}\\ ⋮& \ddots & ⋮\\ a_{m1}\mathbf{𝐁}& \cdots & a_{mn}\mathbf{𝐁}\end{array}\right],$

more explicitly:

$\mathbf{𝐀}\otimes \mathbf{𝐁}=\left[\begin{array}{cccccccccc}{a}_{11}{b}_{11}& a_{11}{b}_{12}& \cdots & a_{11}{b}_{1q}& \cdots & \cdots & a_{1n}{b}_{11}& a_{1n}{b}_{12}& \cdots & a_{1n}{b}_{1q}\\ a_{11}{b}_{21}& a_{11}{b}_{22}& \cdots & a_{11}{b}_{2q}& \cdots & \cdots & a_{1n}{b}_{21}& a_{1n}{b}_{22}& \cdots & a_{1n}{b}_{2q}\\ ⋮& ⋮& \ddots & ⋮& & & ⋮& ⋮& \ddots & ⋮\\ a_{11}{b}_{p1}& a_{11}{b}_{p2}& \cdots & a_{11}{b}_{pq}& \cdots & \cdots & a_{1n}{b}_{p1}& a_{1n}{b}_{p2}& \cdots & a_{1n}{b}_{pq}\\ ⋮& ⋮& & ⋮& \ddots & & ⋮& ⋮& & ⋮\\ ⋮& ⋮& & ⋮& & \ddots & ⋮& ⋮& & ⋮\\ a_{m1}{b}_{11}& a_{m1}{b}_{12}& \cdots & a_{m1}{b}_{1q}& \cdots & \cdots & a_{mn}{b}_{11}& a_{mn}{b}_{12}& \cdots & a_{mn}{b}_{1q}\\ a_{m1}{b}_{21}& a_{m1}{b}_{22}& \cdots & a_{m1}{b}_{2q}& \cdots & \cdots & a_{mn}{b}_{21}& a_{mn}{b}_{22}& \cdots & a_{mn}{b}_{2q}\\ ⋮& ⋮& \ddots & ⋮& & & ⋮& ⋮& \ddots & ⋮\\ a_{m1}{b}_{p1}& a_{m1}{b}_{p2}& \cdots & a_{m1}{b}_{pq}& \cdots & \cdots & a_{mn}{b}_{p1}& a_{mn}{b}_{p2}& \cdots & a_{mn}{b}_{pq}\end{array}\right].$

Using $//$ and $\mathrm{\%}$ to denote truncating integer division and remainder, respectively, and numbering the matrix elements starting from 0, one obtains

$(A\otimes B{)}_{pr+v,qs+w}=a_{rs}{b}_{vw}$

$(A\otimes B{)}_{i,j}=a_{i//p,j//q}{b}_{i\mathrm{\%}p,j\mathrm{\%}q}.$

For the usual numbering starting from 1, one obtains

$(A\otimes B{)}_{p(r-1)+v,q(s-1)+w}=a_{rs}{b}_{vw}$

$(A\otimes B{)}_{i,j}=a_{\lceil i/p\rceil ,\lceil j/q\rceil }{b}_{(i-1)\mathrm{\%}p+1,(j-1)\mathrm{\%}q+1}.$

If A and B represent linear transformations V₁ → W₁ and V₂ → W₂, respectively, then the tensor product of the two maps is a map V₁ ⊗ V₂ → W₁ ⊗ W₂ represented by A ⊗ B.

Examples

$\left[\begin{array}{cc}1& 2\\ 3& 4\\ \end{array}\right]\otimes \left[\begin{array}{cc}0& 5\\ 6& 7\\ \end{array}\right]=\left[\begin{array}{cc}1\left[\begin{array}{cc}0& 5\\ 6& 7\\ \end{array}\right]& 2\left[\begin{array}{cc}0& 5\\ 6& 7\\ \end{array}\right]\\ 3\left[\begin{array}{cc}0& 5\\ 6& 7\\ \end{array}\right]& 4\left[\begin{array}{cc}0& 5\\ 6& 7\\ \end{array}\right]\\ \end{array}\right]=\left[\begin{array}{cccc}1\times 0& 1\times 5& 2\times 0& 2\times 5\\ 1\times 6& 1\times 7& 2\times 6& 2\times 7\\ 3\times 0& 3\times 5& 4\times 0& 4\times 5\\ 3\times 6& 3\times 7& 4\times 6& 4\times 7\\ \end{array}\right]=\left[\begin{array}{cccc}0& 5& 0& 10\\ 6& 7& 12& 14\\ 0& 15& 0& 20\\ 18& 21& 24& 28\end{array}\right].$

Similarly:

$\left[\begin{array}{ccc}1& -4& 7\\ -2& 3& 3\end{array}\right]\otimes \left[\begin{array}{cccc}8& -9& -6& 5\\ 1& -3& -4& 7\\ 2& 8& -8& -3\\ 1& 2& -5& -1\end{array}\right]=\left[\begin{array}{cccccccccccc}8& -9& -6& 5& -32& 36& 24& -20& 56& -63& -42& 35\\ 1& -3& -4& 7& -4& 12& 16& -28& 7& -21& -28& 49\\ 2& 8& -8& -3& -8& -32& 32& 12& 14& 56& -56& -21\\ 1& 2& -5& -1& -4& -8& 20& 4& 7& 14& -35& -7\\ -16& 18& 12& -10& 24& -27& -18& 15& 24& -27& -18& 15\\ -2& 6& 8& -14& 3& -9& -12& 21& 3& -9& -12& 21\\ -4& -16& 16& 6& 6& 24& -24& -9& 6& 24& -24& -9\\ -2& -4& 10& 2& 3& 6& -15& -3& 3& 6& -15& -3\end{array}\right]$

Matrix equations

The Kronecker product can be used to get a convenient representation for some matrix equations. Consider for instance the equation AXB = C, where A, B and C are given matrices and the matrix X is the unknown. We can use the "vec trick" to rewrite this equation as

$({\mathbf{𝐁}}^{\text{T}}\otimes \mathbf{𝐀})\mathrm{vec}(\mathbf{𝐗})=\mathrm{vec}(\mathbf{𝐀}\mathbf{𝐗}\mathbf{𝐁})=\mathrm{vec}(\mathbf{𝐂}).$

Here, vec(X) denotes the vectorization of the matrix X, formed by stacking the columns of X into a single column vector.

It now follows from the properties of the Kronecker product that the equation AXB = C has a unique solution, if and only if A and B are invertible .

If X and C are row-ordered into the column vectors u and v, respectively, then

$\mathbf{𝐯}=(\mathbf{𝐀}\otimes {\mathbf{𝐁}}^{\text{T}})\mathbf{𝐮}.$

The reason is that

$\mathbf{𝐯}=\mathrm{vec}((\mathbf{𝐀}\mathbf{𝐗}\mathbf{𝐁}{)}^{\text{T}})=\mathrm{vec}\left({\mathbf{𝐁}}^{\text{T}}{\mathbf{𝐗}}^{\text{T}}{\mathbf{𝐀}}^{\text{T}}\right)=(\mathbf{𝐀}\otimes {\mathbf{𝐁}}^{\text{T}})\mathrm{vec}\left({\mathbf{𝐗}}^{\text{T}}\right)=(\mathbf{𝐀}\otimes {\mathbf{𝐁}}^{\text{T}})\mathbf{𝐮}.$

Applications

For an example of the application of this formula, see the article on the Lyapunov equation. This formula also comes in handy in showing that the matrix normal distribution is a special case of the multivariate normal distribution. This formula is also useful for representing 2D image processing operations in matrix-vector form.

Another example is when a matrix can be factored as a Kronecker product, then matrix multiplication can be performed faster by using the above formula. This can be applied recursively, as done in the radix-2 FFT and the Fast Walsh–Hadamard transform. Splitting a known matrix into the Kronecker product of two smaller matrices is known as the "nearest Kronecker product" problem, and can be solved exactly by using the SVD. To split a matrix into the Kronecker product of more than two matrices, in an optimal fashion, is a difficult problem and the subject of ongoing research; some authors cast it as a tensor decomposition problem.

In conjunction with the least squares method, the Kronecker product can be used as an accurate solution to the hand–eye calibration problem.

Related matrix operations

Two related matrix operations are the Tracy–Singh and Khatri–Rao products, which operate on partitioned matrices. Let the m × n matrix A be partitioned into the m_i × n_j blocks A_ij and p × q matrix B into the p_k × q_Template:Ell blocks B_kl, with of course Σ_i m_i = m, Σ_j n_j = n, Σ_k p_k = p and Σ_Template:Ell q_Template:Ell = q.

Tracy–Singh product

The Tracy–Singh product is defined as

$\mathbf{𝐀}\circ \mathbf{𝐁}={({\mathbf{𝐀}}_{ij}\circ \mathbf{𝐁})}_{ij}={\left({({\mathbf{𝐀}}_{ij}\otimes {\mathbf{𝐁}}_{kl})}_{kl}\right)}_{ij}$

which means that the (ij)-th subblock of the mp × nq product A $\circ$ B is the m_i p × n_j q matrix A_ij $\circ$ B, of which the (kTemplate:Ell)-th subblock equals the m_i p_k × n_j q_Template:Ell matrix A_ij ⊗ B_{kTemplate:Ell}. Essentially the Tracy–Singh product is the pairwise Kronecker product for each pair of partitions in the two matrices.

For example, if A and B both are 2 × 2 partitioned matrices e.g.:

$\mathbf{𝐀}=\left[\begin{array}{cc}{\mathbf{𝐀}}_{11}& {\mathbf{𝐀}}_{12}\\ {\mathbf{𝐀}}_{21}& {\mathbf{𝐀}}_{22}\end{array}\right]=\left[\begin{array}{ccc}1& 2& 3\\ 4& 5& 6\\ 7& 8& 9\end{array}\right],\mathbf{𝐁}=\left[\begin{array}{cc}{\mathbf{𝐁}}_{11}& {\mathbf{𝐁}}_{12}\\ {\mathbf{𝐁}}_{21}& {\mathbf{𝐁}}_{22}\end{array}\right]=\left[\begin{array}{ccc}1& 4& 7\\ 2& 5& 8\\ 3& 6& 9\end{array}\right],$

we get:

$\begin{array}{rl}\mathbf{𝐀}\circ \mathbf{𝐁}=& \left[\begin{array}{cc}{\mathbf{𝐀}}_{11}\circ \mathbf{𝐁}& {\mathbf{𝐀}}_{12}\circ \mathbf{𝐁}\\ {\mathbf{𝐀}}_{21}\circ \mathbf{𝐁}& {\mathbf{𝐀}}_{22}\circ \mathbf{𝐁}\end{array}\right]\\ =& \left[\begin{array}{cccc}{\mathbf{𝐀}}_{11}\otimes {\mathbf{𝐁}}_{11}& {\mathbf{𝐀}}_{11}\otimes {\mathbf{𝐁}}_{12}& {\mathbf{𝐀}}_{12}\otimes {\mathbf{𝐁}}_{11}& {\mathbf{𝐀}}_{12}\otimes {\mathbf{𝐁}}_{12}\\ {\mathbf{𝐀}}_{11}\otimes {\mathbf{𝐁}}_{21}& {\mathbf{𝐀}}_{11}\otimes {\mathbf{𝐁}}_{22}& {\mathbf{𝐀}}_{12}\otimes {\mathbf{𝐁}}_{21}& {\mathbf{𝐀}}_{12}\otimes {\mathbf{𝐁}}_{22}\\ {\mathbf{𝐀}}_{21}\otimes {\mathbf{𝐁}}_{11}& {\mathbf{𝐀}}_{21}\otimes {\mathbf{𝐁}}_{12}& {\mathbf{𝐀}}_{22}\otimes {\mathbf{𝐁}}_{11}& {\mathbf{𝐀}}_{22}\otimes {\mathbf{𝐁}}_{12}\\ {\mathbf{𝐀}}_{21}\otimes {\mathbf{𝐁}}_{21}& {\mathbf{𝐀}}_{21}\otimes {\mathbf{𝐁}}_{22}& {\mathbf{𝐀}}_{22}\otimes {\mathbf{𝐁}}_{21}& {\mathbf{𝐀}}_{22}\otimes {\mathbf{𝐁}}_{22}\end{array}\right]\\ =& \left[\begin{array}{ccccccccc}1& 2& 4& 7& 8& 14& 3& 12& 21\\ 4& 5& 16& 28& 20& 35& 6& 24& 42\\ 2& 4& 5& 8& 10& 16& 6& 15& 24\\ 3& 6& 6& 9& 12& 18& 9& 18& 27\\ 8& 10& 20& 32& 25& 40& 12& 30& 48\\ 12& 15& 24& 36& 30& 45& 18& 36& 54\\ 7& 8& 28& 49& 32& 56& 9& 36& 63\\ 14& 16& 35& 56& 40& 64& 18& 45& 72\\ 21& 24& 42& 63& 48& 72& 27& 54& 81\end{array}\right].\end{array}$

Khatri–Rao product

• Block Kronecker product
• Column-wise Khatri–Rao product

Face-splitting product

Mixed-products properties

$\mathbf{𝐀}\otimes (\mathbf{𝐁}\bullet \mathbf{𝐂})=(\mathbf{𝐀}\otimes \mathbf{𝐁})\bullet \mathbf{𝐂},$

where $\bullet$ denotes the Face-splitting product.

$(\mathbf{𝐀}\bullet \mathbf{𝐁})(\mathbf{𝐂}\otimes \mathbf{𝐃})=(\mathbf{𝐀}\mathbf{𝐂})\bullet (\mathbf{𝐁}\mathbf{𝐃}),$

Similarly:

$(\mathbf{𝐀}\bullet \mathbf{𝐋})(\mathbf{𝐁}\otimes \mathbf{𝐌})\cdots (\mathbf{𝐂}\otimes \mathbf{𝐒})=(\mathbf{𝐀}\mathbf{𝐁}\cdots \mathbf{𝐂})\bullet (\mathbf{𝐋}\mathbf{𝐌}\cdots \mathbf{𝐒}),$

${\mathbf{𝐜}}^{\text{T}}\bullet {\mathbf{𝐝}}^{\text{T}}={\mathbf{𝐜}}^{\text{T}}\otimes {\mathbf{𝐝}}^{\text{T}},$

where $\mathbf{𝐜}$ and $\mathbf{𝐝}$ are vectors,

$(\mathbf{𝐀}\bullet \mathbf{𝐁})(\mathbf{𝐜}\otimes \mathbf{𝐝})=(\mathbf{𝐀}\mathbf{𝐜})\circ (\mathbf{𝐁}\mathbf{𝐝}),$

where $\mathbf{𝐜}$ and $\mathbf{𝐝}$ are vectors, and $\circ$ denotes the Hadamard product.

Similarly:

$(\mathbf{𝐀}\bullet \mathbf{𝐁})(\mathbf{𝐌}\mathbf{𝐍}\mathbf{𝐜}\otimes \mathbf{𝐐}\mathbf{𝐏}\mathbf{𝐝})=(\mathbf{𝐀}\mathbf{𝐌}\mathbf{𝐍}\mathbf{𝐜})\circ (\mathbf{𝐁}\mathbf{𝐐}\mathbf{𝐏}\mathbf{𝐝}),$

${ℱ}(C^{(1)}x\star C^{(2)}y)=({ℱ}{C}^{(1)}\bullet {ℱ}{C}^{(2)})(x\otimes y)={ℱ}{C}^{(1)}x\circ {ℱ}{C}^{(2)}y,$

where $\star$ is vector convolution and ${ℱ}$ is the Fourier transform matrix (this result is an evolving of count sketch properties),

$(\mathbf{𝐀}\bullet \mathbf{𝐋})(\mathbf{𝐁}\otimes \mathbf{𝐌})\cdots (\mathbf{𝐂}\otimes \mathbf{𝐒})(\mathbf{𝐊}\ast \mathbf{𝐓})=(\mathbf{𝐀}\mathbf{𝐁}\cdot \mathbf{𝐂}\mathbf{𝐊})\circ (\mathbf{𝐋}\mathbf{𝐌}\cdots \mathbf{𝐒}\mathbf{𝐓}),$

where $\ast$ denotes the column-wise Khatri–Rao product.

Similarly:

$(\mathbf{𝐀}\bullet \mathbf{𝐋})(\mathbf{𝐁}\otimes \mathbf{𝐌})\cdots (\mathbf{𝐂}\otimes \mathbf{𝐒})(c\otimes d)=(\mathbf{𝐀}\mathbf{𝐁}\cdots \mathbf{𝐂}\mathbf{𝐜})\circ (\mathbf{𝐋}\mathbf{𝐌}\cdots \mathbf{𝐒}\mathbf{𝐝}),$

$(\mathbf{𝐀}\bullet \mathbf{𝐋})(\mathbf{𝐁}\otimes \mathbf{𝐌})\cdots (\mathbf{𝐂}\otimes \mathbf{𝐒})(\mathbf{𝐏}\mathbf{𝐜}\otimes \mathbf{𝐐}\mathbf{𝐝})=(\mathbf{𝐀}\mathbf{𝐁}\cdots \mathbf{𝐂}\mathbf{𝐏}\mathbf{𝐜})\circ (\mathbf{𝐋}\mathbf{𝐌}\cdots \mathbf{𝐒}\mathbf{𝐐}\mathbf{𝐝}),$

where $\mathbf{𝐜}$ and $\mathbf{𝐝}$ are vectors.

See also

• Generalized linear array model
• Hadamard product (matrices)
• Kronecker coefficient

External links

• Template:Springer
• Template:Planetmath reference
• The entry on the Kronecker, Zehfuss, or Direct Product of matrices has historical information.
• Software source in more than 40 languages.