4

In a discrete $n$ dimensional vector space the Kronecker delta $\delta_{ij}$ is basically the $n \times n$ identity matrix. When generalizing from a discrete $n$ dimensional vector space to an infinite dimensional space of functions $f$ it seems natural to assume that the generalization of the Kronecker delta should be an identity operator $$ \operatorname{I} f = f $$ However it is said that the continuous generalization of the Kronecker delta is the Dirac delta function $$ \int_{-\infty}^\infty \delta(x - y) f(x)\, dx = f(y) $$ Why is that the case? What is "wrong" with the simple identity operator?

UPDATE: What I mean with "simple" identity operator is: Why is the identity operator not simply the scalar number "1", but the delta function instead?

UPDATE2: To make it more clear: Why is the continuous generalization of the Kronecker delta $$ \int_{-\infty}^\infty \delta(x - y) f(x)\, dx = f(y) $$ and not $$ 1 \cdot f(x) = f(x) $$ ?

asmaier
  • 2,763
  • 1
    $f \to \ \delta \ast f$ is the identity operator ($\ast$ being the convolution, continuous or discrete in 1D or dimension $n$ depending on the case, leading to many different $\delta$, one for each convolution operator) – reuns Feb 22 '16 at 21:52
  • Semantic comment to the post (v2): The identity operator is usually the map $x\mapsto x$ not the map $x\mapsto 1$. – Qmechanic Mar 27 '16 at 13:09
  • @Qmechanic: What I mean is the map/operator $x -> x$ can be represented as scalar number "1" in the sense that $x*1 = x$. – asmaier Mar 30 '16 at 10:01
  • Just to make sure: Is your $*$ symbol multiplication or convolution? – Qmechanic Mar 30 '16 at 10:26
  • @Qmechanic: I mean multiplication. – asmaier Mar 30 '16 at 12:34

2 Answers2

6

The identity operator is the same in both cases.

For the discrete case $$ \sum_{j=1}^n \delta_{ij} x_j = x_i $$ and in terms of operators, $I(\mathbf{x}) = \mathbf x$ .

For the continuous case $$ \int_{-\infty}^\infty \delta(x - y) f(x)\, dx = f(y) $$ and in terms of operators $I(f) = f$.

GEdgar
  • 117,296
0

I) Let there be given an linear operator $A:V\to V$, where $V$ is a vector space.

  1. If $V$ is finite-dimensional, given a choice of basis, the operator $A$ can be represented $$(Av)^i ~=~ \sum_j a^i{}_j v^j.\tag{1}$$ by a matrix $a^i{}_j$.

  2. If $V$ is infinite-dimensional, of the form of an appropriate function space, the operator $A$ can be represented $$ (Af)(x)~=~ \int \!dy ~a(x,y) f(y).\tag{2}$$ by an integral kernel $a(x,y)$

Notice that the discrete index $j$, the summation, and the vector components $v^j$ in eq. (1) have been replaced by a continuous index $y$, an integral, and a function value $f(y)$ in eq. (2), respectively.

II) Now consider the identity operator $A={\rm Id}_V:V\to V$. In both cases (1) and (2), we may view the identity operator ${\rm Id}_V$ as multiplication with the constant $1$.

  1. In case 1, the matrix $a^i{}_j=\delta^i_j$ is given by the Kronecker delta.

  2. In case 2, the integral kernel $a(x,y)=\delta(x,y)$ is given by the Dirac delta distribution.

Qmechanic
  • 13,259