The Dirac Delta
Source lecture(s): PHY621 Ch. 3
Intuition
The delta "function" is the continuous analogue of the Kronecker delta: the thing that picks out one value from a continuum, exactly as \(\delta_{ij}\) picks one term from a sum.
It is not a function. There is no function that is zero everywhere except one point and still integrates to 1. It is a distribution — an object defined entirely by what it does inside an integral, and any statement about \(\delta(x)\) that is not eventually integrated against a test function is meaningless.
Where it comes from in this course
Push Fourier series to an infinite interval and the sum becomes an integral; the completeness relation of the discrete basis becomes
That identity is Fourier inversion. Substituting it into \(f(x) = \int\delta(x-x')f(x')dx'\) reproduces the transform pair, and it is the continuous version of \(\sum_n|n\rangle\langle n| = \hat 1\) — see Dirac notation.
Properties worth knowing
The derivative rule comes from integrating by parts and throwing the derivative onto the test function — which is how all derivatives of distributions are defined. The last identity (the step function's derivative) is the one you use constantly in Green's functions.
As a limit
\(\delta\) is the limit of any normalised sequence that narrows:
The limit must be taken after integrating against the test function, never before. The Lorentzian form is the one that appears in scattering theory and in the Landau prescription; the \(\sin Lx/\pi x\) form is what makes Fourier inversion work.
Why physicists reach for it
- Point sources. A point charge is \(\rho = q\delta^3(\mathbf{r})\); a point mass, an impulse, a delta-function potential.
- Green's functions. The whole method is: solve \(\mathcal{L}G = \delta\), then build the response to any source by superposition, \(u = \int G f\). The delta is the "unit kick" whose response you record once and reuse forever.
- Convolution. \(\delta\) is the identity: \(f * \delta = f\).
- Orthonormality of continuous bases. \(\langle x|x'\rangle = \delta(x-x')\).
Common mistakes
- Evaluating \(\delta(0)\). It is not a number. If your algebra produces \(\delta(0)\) you have usually squared a delta, which is undefined.
- Forgetting the Jacobian. \(\delta(g(x))\) picks up \(1/|g'|\) at each root; missing it is a common factor-of-something error in scattering and in changes of variable.
- Taking the limit too early. The sequence definitions only make sense under the integral.
- Squaring it. \(\delta(x)^2\) is meaningless. If you need \(|\langle x|\psi\rangle|^2\) that is fine; \(\delta^2\) is not.
Related concepts
- Fourier transform — the delta is the inversion theorem
- Green's functions — response to a delta source
- Convolution — the delta is the identity element
- Dirac notation — continuous completeness
- Special functions
Knowledge graph position
Prerequisites: Fourier series, inner product spaces. Leads to: Fourier transform, Green's functions, distribution theory, quantum mechanics.
Quiz
Q1 (computational). Evaluate \(\int_{-\infty}^{\infty}\delta(2x - 6)\,x^2\,dx\).
Answer
Write \(g(x) = 2x-6\), root at \(x = 3\), \(|g'| = 2\). So \(\delta(2x-6) = \tfrac12\delta(x-3)\) and the integral is \(\tfrac12(3)^2 = 4.5\). Forgetting the \(1/|g'|\) would give 9.
Q2 (conceptual). In what sense is \(\frac{1}{2\pi}\int e^{ik(x-x')}dk = \delta(x-x')\) the same statement as Fourier inversion?
Answer
Substituting it into \(f(x) = \int\delta(x-x')f(x')dx'\) and exchanging the order of integration gives exactly the forward-then-inverse transform. The completeness of the plane waves and the invertibility of the transform are one fact stated two ways.
Q3 (MCQ). The Dirac delta is best understood as:
- (a) a function that is infinite at zero
- (b) a distribution, defined by its action on test functions inside an integral
- (c) a limit of Gaussians, evaluated pointwise
- (d) a discrete Kronecker delta
Answer
(b). (a) is not well-defined; (c) fails because the limit must be taken after integration; (d) is the discrete analogue, not the same object.