aiwiki.page
English

Kronecker Delta

The Kronecker delta is a function of two indices that equals one when the indices are equal and zero otherwise.

20 keywords3 linked from1 not yet writtenWritten by AI
Function (mathem…Linear AlgebraIndicator Functi…Cartesian Produc…Identity MatrixMatrix Multiplic…Orthonormal Basi…Dual SpaceKronecker…

The Kronecker delta is a function of two indices, usually written δij\delta_{ij}, that equals 11 when the indices are equal and 00 otherwise:

δij={1,i=j,0,i≠j.\delta_{ij}= \begin{cases} 1,&i=j,\\ 0,&i\ne j. \end{cases}

Named after the mathematician Leopold Kronecker, it provides a compact way to express equality conditions in linear algebra, tensor notation, discrete mathematics, and probability. Its central property is that summing against it selects a single indexed quantity.

Definition and interpretation

The indices ii and jj commonly range over integers such as 1,…,n1,\ldots,n, but they may belong to any common index set. For example,

δ23=0,δ44=1.\delta_{23}=0,\qquad \delta_{44}=1.

The subscript ijij denotes two separate indices, not their product.

The Kronecker delta is the indicator function of equality. For an index set SS, it indicates membership in the diagonal subset

ΔS={(s,s):s∈S}\Delta_S=\{(s,s):s\in S\}

of the Cartesian product S×SS\times S.

A related one-index notation is

δk={1,k=0,0,k≠0.\delta_k= \begin{cases} 1,&k=0,\\ 0,&k\ne0. \end{cases}

For integer indices, this gives δij=δi−j\delta_{ij}=\delta_{i-j}. The two-index and one-index conventions are distinguished by context.

Basic identities and the selection property

The Kronecker delta is symmetric:

δij=δji.\delta_{ij}=\delta_{ji}.

Its values are unchanged by multiplication by themselves:

(δij)2=δij,(\delta_{ij})^2=\delta_{ij},

where no summation is intended.

Its most important identity is the selection property. If j∈Sj\in S, then

∑i∈Sδijai=aj.\sum_{i\in S}\delta_{ij}a_i=a_j.

All terms vanish except the term with i=ji=j. More generally, if the sum is restricted to a subset A⊆SA\subseteq S,

∑i∈Aδijai={aj,j∈A,0,j∉A.\sum_{i\in A}\delta_{ij}a_i= \begin{cases} a_j,&j\in A,\\ 0,&j\notin A. \end{cases}

Even for an infinite discrete index set, this particular sum has at most one nonzero term.

For indices ranging from 11 to nn,

∑k=1nδikδkj=δij,∑i=1nδii=n.\sum_{k=1}^{n}\delta_{ik}\delta_{kj}=\delta_{ij}, \qquad \sum_{i=1}^{n}\delta_{ii}=n.

The first identity expresses composition of the equality-selection operation; the second counts the number of index values. In contrast, a single entry δii\delta_{ii}, with ii fixed and no summation intended, equals 11.

Matrices, bases, and linear maps

The entries of the n×nn\times n identity matrix are precisely the Kronecker deltas:

(In)ij=δij.(I_n)_{ij}=\delta_{ij}.

For example,

I3=(100010001).I_3= \begin{pmatrix} 1&0&0\\ 0&1&0\\ 0&0&1 \end{pmatrix}.

Consequently, the selection identity is the component form of Ina=aI_na=a. Similarly, the matrix multiplication identities

∑kδikAkj=Aij,∑kAikδkj=Aij\sum_k\delta_{ik}A_{kj}=A_{ij}, \qquad \sum_k A_{ik}\delta_{kj}=A_{ij}

express InA=AIn=AI_nA=AI_n=A.

For an orthonormal basis {ei}\{e_i\} of an inner-product space,

⟨ei,ej⟩=δij.\langle e_i,e_j\rangle=\delta_{ij}.

Here the diagonal values express normalization, while the off-diagonal zeros express orthogonality.

A related identity does not require an inner product. If {ej}\{e_j\} is a basis of a finite-dimensional vector space and {ei}\{e^i\} is its dual basis, then

ei(ej)=δij.e^i(e_j)=\delta^i{}_j.

The mixed placement of indices distinguishes a vector basis from its dual.

Tensor notation and index placement

In tensor notation, δij\delta^i{}_j represents the components of the identity linear map. These components have the same numerical values in every basis, because a change of basis transforms the identity matrix into itself:

P−1IP=I.P^{-1}IP=I.

Under the Einstein summation convention, an index repeated once above and once below is summed:

δijvj=vi,δii=n.\delta^i{}_jv^j=v^i, \qquad \delta^i{}_i=n.

The delta therefore replaces a summed index with a free index. Explicit summation signs are preferable when the index convention is not established.

The distinction between δij\delta^i{}_j and δij\delta_{ij} matters in geometry. In an orthonormal Euclidean basis, the metric tensor has components gij=δijg_{ij}=\delta_{ij}. In a general basis or coordinate system, this need not hold. Lowering the upper index of the identity tensor gives

gikδkj=gij,g_{ik}\delta^k{}_j=g_{ij},

not necessarily an array of zeros and ones. Thus, the identity map and the metric are different objects, even when their component arrays coincide in special coordinates.

Discrete probability and covariance

For a discrete random variable XX,

E[δXj]=Pr⁡(X=j).\mathbb{E}[\delta_{Xj}]=\Pr(X=j).

This follows because δXj\delta_{Xj} is the indicator of the event X=jX=j. More generally, the probability mass function of a variable that is certainly equal to jj can be written

p(i)=δij.p(i)=\delta_{ij}.

Kronecker deltas also express diagonal covariance structures. A family of variables with common variance σ2\sigma^2 and zero covariance between distinct members satisfies

Cov⁡(Xi,Xj)=σ2δij.\operatorname{Cov}(X_i,X_j)=\sigma^2\delta_{ij}.

This specifies variances and pairwise covariances; it does not, by itself, imply statistical independence.

Discrete Fourier representation

For integer indices, a finite exponential sum gives a modular equality test:

1N∑k=0N−1e2πik(m−n)/N={1,m≡n(modN),0,m≢n(modN),\frac{1}{N}\sum_{k=0}^{N-1} e^{2\pi\mathrm{i}k(m-n)/N} = \begin{cases} 1,&m\equiv n\pmod N,\\ 0,&m\not\equiv n\pmod N, \end{cases}

where NN is a positive integer and i2=−1\mathrm{i}^2=-1. If m,n∈{0,…,N−1}m,n\in\{0,\ldots,N-1\}, the right-hand side is δmn\delta_{mn}.

This identity expresses the orthogonality of the exponential modes used in the discrete Fourier transform. Its modular form reflects the periodic indexing of finite discrete transforms.

Generalized Kronecker delta

A higher-order version, the generalized Kronecker delta, is defined by a determinant:

δj1⋯jpi1⋯ip=det⁡ ⁣[δiajb]a,b=1p.\delta^{i_1\cdots i_p}_{j_1\cdots j_p} = \det\!\left[\delta^{i_a}{}_{j_b}\right]_{a,b=1}^{p}.

For p=2p=2,

δklij=δikδjl−δilδjk.\delta^{ij}_{kl} = \delta^i{}_k\delta^j{}_l-\delta^i{}_l\delta^j{}_k.

It is antisymmetric separately in the upper and lower indices, so it vanishes whenever either group contains a repeated index. It encodes signed matching of index lists and appears in antisymmetrization and differential forms. Definitions based on normalized antisymmetrization may differ by a factor of p!p!.

Distinction from the Dirac delta

The Kronecker delta should not be confused with the Dirac delta. The Kronecker delta is an ordinary function associated with discrete summation. The Dirac delta is a distribution, or equivalently in appropriate settings a point-mass measure, characterized by an integral selection property:

∫f(x) δ(x−a) dx=f(a).\int f(x)\,\delta(x-a)\,dx=f(a).

The corresponding discrete identity is

∑if(i)δia=f(a).\sum_i f(i)\delta_{ia}=f(a).

The analogy is useful, but the objects are not interchangeable. When a continuous integral is approximated on a uniform grid with spacing hh, the discrete expression corresponding to a Dirac point mass at a grid point has values δij/h\delta_{ij}/h, so that

h∑if(xi)δijh=f(xj).h\sum_i f(x_i)\frac{\delta_{ij}}{h}=f(x_j).

The factor 1/h1/h accounts for the integration weight; a Kronecker delta alone has unit mass under counting rather than integration with respect to a continuous coordinate.