Dirac Delta Function

The Dirac delta function $δ (x)$ is a "generalized function" that is zero everywhere except at $x = 0$ , where it is infinitely spiked so that its total integral equals $1$ . Its defining property is the sifting property: $\int_{- \infty}^{\infty} f (x) δ (x - a) d x = f (a)$ . It is used throughout physics and engineering to model point masses, point charges, impulses, and idealized sources, and as the identity element of convolution. Introduced by Paul Dirac (~1927) in quantum mechanics, rigorously founded by Laurent Schwartz (1945) via distribution theory.

The Dirac delta $δ (x)$ is not a function in the ordinary sense — no genuine function can be zero almost everywhere yet integrate to $1$ . Instead it is defined by how it acts inside integrals against well-behaved test functions, extracting a single value (the sifting or sampling property). It matters because it lets us treat discrete, localized, and continuous phenomena in one unified analytical language: a point charge becomes a charge density, an instantaneous kick becomes a force, a sampled signal becomes a sum of deltas. The surprising fact is that this "function" — long used heuristically by physicists and distrusted by mathematicians — turns out to have a completely rigorous home in the theory of distributions, where it is simply a continuous linear functional, and where it can even be differentiated infinitely many times.

The Core Insight

The delta function is best understood not by its pointwise values but by what it does: it is the evaluation functional. Feeding $δ$ a test function $f$ returns $f (0)$ . Everything else — the spike, the unit area, the limit of tall thin bumps — is a picture meant to help intuition. The mathematically honest statement is: $δ$ is the limit (in a weak sense) of unit-area "nascent delta" sequences, and its meaning lives entirely under an integral sign.

Formal Statement

Dirac Delta — Heuristic Form

The Dirac delta function $δ (x)$ is informally characterized by the two conditions

δ (x) = 0 for x \neq = 0, δ (0) = + \infty,

⟳Not a derivation — this is the informal "infinite spike at the origin" picture; it is not literally valid as a Lebesgue function.

\int_{- \infty}^{\infty} δ (x) d x = 1.

⟳Normalization condition; the total area (mass) under the spike is defined to be unity.

Here $x \in R$ is the independent variable, and the integral is taken over the whole real line.

Equations [1]–[2] are mutually inconsistent in classical analysis: a function that is zero except at a single point has Lebesgue integral $0$ , never $1$ . So $δ$ is not a function. The honest definition is the functional one below.

Dirac Delta — Distributional (Rigorous) Form

Let $D (R)$ be the space of test functions: smooth functions $φ : R \to R$ with compact support. The Dirac delta is the linear functional $δ : D (R) \to R$ defined by

⟨ δ, φ ⟩ = φ (0) .

⟳

δ

is defined as the evaluation-at-zero functional;

⟨ \cdot, \cdot ⟩

denotes the action of a distribution on a test function.

More generally, the shifted delta $δ_{a} = δ (x - a)$ satisfies the sifting property

\int_{- \infty}^{\infty} f (x) δ (x - a) d x = f (a),

⟳Apply the evaluation functional at the point

x = a

; the integral notation is formal shorthand for

⟨ δ_{a}, f ⟩

valid for every continuous $f$ . Here $a \in R$ is the location of the spike.

Key Properties of the Delta Function

For constants $a, b$ with $b \neq = 0$ , and suitable test functions, the delta satisfies:

δ (x) = δ (- x) (even), x δ (x) = 0.

⟳Substitute

x \to - x

in the sifting integral; for

x δ (x)

, note

⟨ x δ, φ ⟩ = ⟨ δ, x φ ⟩ = (x φ) ∣_{x = 0} = 0

δ (b x) = \frac{1}{∣ b ∣} δ (x) .

⟳Change variables

u = b x

\int f (x) δ (b x) d x

; the Jacobian

∣ b ∣

and the sign-handling of limits produce the

1/∣ b ∣

factor.

δ (g (x)) = i \sum \frac{δ ( x - x _{i} )}{∣ g ^{'} ( x _{i} ) ∣}, g (x_{i}) = 0, g^{'} (x_{i}) \neq = 0.

⟳Linearize

g

near each simple root

x_{i}

via

g (x) \approx g^{'} (x_{i}) (x - x_{i})

and apply [6] to each isolated zero.

\int_{- \infty}^{x} δ (t) d t = H (x),

⟳Integrating the unit-area spike accumulates from

0

1

x

crosses the origin, giving the Heaviside step function

H

where $H$ is the Heaviside step function, so symbolically $H^{'} (x) = δ (x)$ .

Why It Holds / Derivation

Delta as a Limit of Nascent Deltas

We show that a sequence of ordinary, unit-area functions concentrating at the origin reproduces the sifting property in the limit.

Step 1. Choose a family

δ_{ϵ} (x)

, e.g. the Gaussian

δ_{ϵ} (x) = \frac{1}{ϵ π} e^{- x^{2} / ϵ^{2}} .

⟳A normalized Gaussian of width

\sim ϵ

and height

\sim 1/ ϵ

; its area is

1

for every

ϵ

by the standard Gaussian integral.

Step 2. Verify unit area for all

ϵ > 0

\int_{- \infty}^{\infty} δ_{ϵ} (x) d x = \frac{1}{ϵ π} \int_{- \infty}^{\infty} e^{- x^{2} / ϵ^{2}} d x = \frac{1}{ϵ π} \cdot ϵ π = 1.

⟳Substitute

u = x / ϵ

and use

\int e^{- u^{2}} d u = π

Step 3. Take a continuous

f

. By unit area,

\int_{- \infty}^{\infty} f (x) δ_{ϵ} (x) d x - f (0) = \int_{- \infty}^{\infty} (f (x) - f (0)) δ_{ϵ} (x) d x .

⟳Subtract

f (0) = f (0) \int δ_{ϵ} d x

and merge into one integral.

Step 4. As

ϵ \to 0^{+}

δ_{ϵ}

concentrates near

x = 0

, where

f (x) - f (0) \to 0

by continuity; the right side vanishes. Hence

ϵ \to 0^{+} lim \int_{- \infty}^{\infty} f (x) δ_{ϵ} (x) d x = f (0) = ⟨ δ, f ⟩ .

⟳Split the integral into

∣ x ∣ < η

(small by continuity) and

∣ x ∣ \geq η

(small by rapid decay of

δ_{ϵ}

); both

\to 0

This is exactly the sifting property [4] with $a = 0$ , justifying $δ = lim_{ϵ \to 0^{+}} δ_{ϵ}$ in the weak (distributional) sense. $■$

Paul A. M. Dirac | 1902–1984

Paul Dirac, one of the founders of quantum mechanics, introduced the delta function around 1927 in his work on transformation theory and quantum dynamics. He needed a continuous analogue of the discrete Kronecker delta $δ_{ij}$ to express orthonormality of continuum eigenstates (e.g. position eigenstates) and to write resolutions of the identity. He treated $δ (x)$ operationally — "an improper function" useful as a tool — fully aware it lacked classical rigor, writing that it could "be used as though it were a continuous function." His pragmatic confidence was vindicated two decades later when Laurent Schwartz built the theory of distributions, giving $δ$ a precise mathematical foundation and earning Schwartz the 1950 Fields Medal.

Building Intuition

Picture squeezing a fixed amount of clay (mass $1$ ) into an ever-narrower, ever-taller column. The area never changes, but the shape limits toward an infinitely thin spike. The delta function is what survives in the integral: it remembers only "there is one unit of mass, located at $x = 0$ ." When you multiply by a smooth function $f$ and integrate, the spike samples $f$ at exactly that location — like a probe so sharp it touches the curve at a single point and reports the height there. This is why $δ$ is the identity for convolution: convolving any signal with $δ$ leaves it unchanged, because $δ$ is a "do-nothing" probe that just relays each value back.

A nascent-delta family $\delta_\epsilon(x)$ collapsing to the spike at $x=0$. As $\epsilon$ shrinks, the bumps grow taller and narrower while preserving unit area.

Worked Examples

Sifting with a Shifted Delta

Given. Evaluate

I = \int_{- \infty}^{\infty} cos (π x) δ (x - 2) d x

. Step 1. The delta is centered at

a = 2

. By the sifting property [4],

\int_{- \infty}^{\infty} f (x) δ (x - 2) d x = f (2) .

Step 2. Here

f (x) = cos (π x)

, so

f (2) = cos (2 π) = 1

. Step 3. The point

x = 2

lies inside

(- \infty, \infty)

, so the spike is captured.

I = 1 (dimensionless)

Scaling and Composition — Delta of a Function

Given. Evaluate

J = \int_{- \infty}^{\infty} e^{x} δ (3 x - 6) d x

. Step 1. Write

g (x) = 3 x - 6

, whose only root is

x_{0} = 2

, with

g^{'} (x) = 3

, so

∣ g^{'} (x_{0}) ∣ = 3

. By property [7],

δ (3 x - 6) = \frac{δ ( x - 2 )}{∣3∣} = \frac{1}{3} δ (x - 2) .

Step 2. Substitute and pull out the constant:

J = \frac{1}{3} \int_{- \infty}^{\infty} e^{x} δ (x - 2) d x .

Step 3. Apply sifting at

x = 2

\int e^{x} δ (x - 2) d x = e^{2}

. Step 4. Combine:

J = \frac{1}{3} e^{2} \approx \frac{1}{3} (7.389) \approx 2.463.

J = \frac{1}{3} e^{2} \approx 2.463 (dimensionless)

Common Misconceptions

" $δ (0)$ is a number you can plug in." It is not.

δ

has no well-defined pointwise value; writing

δ (0) = \infty

is shorthand, never an actual quantity to compute with. Any expression that requires evaluating

δ

at a point without integrating is meaningless. Likewise, products of two deltas at the same point — e.g.

δ (x)^{2}

δ (x) δ (x)

— are undefined, since you cannot multiply distributions in general.

A tempting but false manipulation: "Since $δ (x) = 0$ for $x \neq = 0$ and the value at one point doesn't affect an integral, $\int f (x) δ (x) d x = 0$ ."

This reasoning fails because $δ$ is not a Lebesgue-integrable function — it is a distribution. The "value $0$ everywhere except a point" heuristic is exactly the part that cannot be made into a classical integrand. The correct result is $\int f (x) δ (x) d x = f (0)$ , which is generally nonzero. The lesson: never reason about $δ$ as if it were an ordinary function.

Connections

The delta function ties together many areas. Its antiderivative is the Heaviside Step Function, and its derivatives $δ^{'} (x)$ (the "doublet") form a tower of higher-order distributions used to model dipoles. The discrete analogue is the Kronecker Delta, used for orthonormal bases in finite or countable settings. The Fourier Transform of $δ$ is the constant function $1$ , and the Fourier transform of $1$ is $2 π δ$ — making $δ$ central to spectral analysis. In probability, $δ$ is the density of a Dirac Measure (a point mass), connecting it to measure theory. It is the identity element of Convolution, and its rigorous home is the theory of Distributions (Generalized Functions).

Real-World Applications

Point charges and masses (physics): A point charge $q$ at $r_{0}$ is modeled by charge density $ρ (r) = q δ^{3} (r - r_{0})$ , feeding directly into Poisson's equation $\nabla^{2} ϕ = - ρ / ϵ_{0}$ .
Impulse response (signals & systems): Striking a system with an idealized "kick" $δ (t)$ yields its impulse response $h (t)$ ; the output to any input is then the convolution $x * h$ .
Green's functions (PDEs): Solving $L G (r, r^{'}) = δ (r - r^{'})$ gives the Green's function from which general solutions are built by superposition.
Quantum mechanics: Position eigenstates satisfy $⟨ x ∣ x^{'} ⟩ = δ (x - x^{'})$ ; the completeness relation uses $\int ∣ x ⟩ ⟨ x ∣ d x = 1$ .
Sampling theory (digital signal processing): Ideal sampling multiplies a signal by a Dirac comb $\sum_{n} δ (t - n T)$ , the basis of the Nyquist–Shannon theorem.
Structural mechanics: A concentrated (point) load on a beam is written as $w (x) = P δ (x - x_{0})$ in the Euler–Bernoulli equation.
Probability & statistics: A deterministic outcome or a discrete spike in a mixed distribution is represented by a delta in the probability density.

Historical Development

1822 — Fourier: Joseph Fourier writes integral representations equivalent to $δ$ while studying heat conduction, implicitly using the sifting idea.
1828 — Green: George Green introduces what are now called Green's functions in his essay on electricity and magnetism, foreshadowing point-source methods.
1880s — Kirchhoff & Heaviside: Oliver Heaviside develops operational calculus with step and impulse-like objects, used boldly and without rigor in telegraphy.
1927 — Dirac: Paul Dirac formally introduces $δ (x)$ in The Physical Interpretation of the Quantum Dynamics, calling it an "improper function."
1930 — Dirac's textbook: The Principles of Quantum Mechanics spreads the delta function throughout physics.
1935 — Bochner: Salomon Bochner develops generalized Fourier analysis, anticipating elements of distribution theory.
1936 — Sobolev: Sergei Sobolev introduces generalized functions and weak solutions in the study of PDEs.
1945–1950 — Schwartz: Laurent Schwartz publishes the rigorous théorie des distributions, giving $δ$ a firm foundation; awarded the Fields Medal in 1950.
1955 — Gel'fand & Shilov: Their multivolume Generalized Functions systematizes and popularizes the theory for applied mathematics and physics.

Subtleties & Limits

The delta function exposes the boundary of what classical analysis permits, and several subtleties persist:

No multiplication of distributions: There is no consistent general way to define $δ (x) \cdot δ (x)$ or products of distributions with overlapping singularities (Schwartz's impossibility result). This is a genuine obstacle in nonlinear PDEs and quantum field theory, where it appears as ultraviolet divergences requiring renormalization.
Choice of nascent sequence: While any unit-area family converging weakly gives the same $δ$ , regularizing nonlinear expressions can depend on the chosen sequence (the Colombeau algebras were developed to handle such ambiguities).
Higher dimensions and curved supports: Defining $δ$ on surfaces, manifolds, or along curves (the "single-layer/double-layer" potentials) requires careful measure-theoretic and differential-geometric setup.
Physical regularization: In practice every "point" source has finite size; whether $δ$ -idealizations introduce spurious infinities (e.g. the self-energy of a point charge) remains a conceptual issue at the physics frontier.