Relativistic wavefunction normalisation

Every scattering rate and decay width in particle physics begins with a plane wave, and every plane wave begins with a normalisation. In non-relativistic quantum mechanics that step is almost invisible: you demand one particle in a box, $\int_V \psi^*\psi\,d^3x = 1$ , fix the amplitude to $A = V^{-1/2}$ , and move on to Fermi's golden rule. The moment the particle moves relativistically, this innocent choice breaks. "One particle per unit volume" is a statement about a density, and a density is not a Lorentz scalar — the box that holds the particle Lorentz-contracts by $1/\gamma$ under a boost, so the very same single particle is counted at $\gamma = E/m$ per unit volume by a moving observer. A quantity that was supposed to be "1" is frame-dependent. The cure is a principle, not a patch. The relativistic probability density is the time component of a conserved four-current, and for a Klein–Gordon plane wave that current is $j^\mu = 2|N|^2 p^\mu$ — so with unit amplitude the density is $\rho = 2E$ , not $1$ . Adopting $2E$ particles per unit volume as the standard builds the boost factor into the state itself, and it makes two objects Lorentz invariant at once: the matrix element $\mathcal{M}_{fi}$ that replaces the golden-rule $T_{fi}$ , and the phase-space measure $d^3p/[(2\pi)^3\,2E]$ that replaces the naive $d^3p/(2\pi)^3$ . The arbitrary box volume $V$ then cancels from every physical answer, and the leftover $1/2E_a$ in front of a decay rate is exactly the time dilation of a moving particle's lifetime. This article shows where the factor of $2E$ comes from, why the $2$ is forced and not chosen, and how it makes the golden rule manifestly relativistic.

Key takeaways

Non-relativistic plane waves $\psi = A\,e^{i(\vec p\cdot\vec x - Et)}$ normalised to one particle in a box give $A^2 = 1/V$ — but "one particle per volume" holds in only one frame.
Under a boost the normalisation box contracts by $1/\gamma$ , so its density rises by $\gamma = E/m$ : probability density is the time-component of a four-vector, not a scalar.
The Klein–Gordon current $j^\mu = 2|N|^2 p^\mu$ forces the fix: normalise to $2E$ particles per unit volume, $\psi' = \sqrt{2E}\,\psi$ , so $\langle p'|p\rangle = 2E\,(2\pi)^3\delta^3(\vec p'-\vec p)$ .
The invariant amplitude is $\mathcal{M}_{fi} = \prod_k (2E_k)^{1/2}\,T_{fi}$ , and the phase-space element becomes the Lorentz-invariant phase space $d^3p/[(2\pi)^3\,2E]$ .
The box volume $V$ cancels from every rate; the surviving $1/2E_a$ prefactor is the time dilation $\tau = \gamma\tau_0$ — a 1 GeV pion flies 55.9 m before decaying.
Key idea: because a normalisation volume Lorentz-contracts under a boost, fixing the number of particles per volume to a constant only works in one frame; choosing instead $2E$ particles per unit volume builds the boost factor into the normalisation, making the whole golden-rule machinery Lorentz invariant.

Where this lives today

The $2E$ normalisation and the invariant measure $d^3p/[(2\pi)^3\,2E]$ are executed billions of times a second inside every particle-physics computation. Every automated matrix-element generator (MadGraph, WHIZARD, Sherpa) computes an invariant $|\mathcal{M}|^2$ and integrates it against Lorentz-invariant phase space; recursive samplers such as RAMBO exist precisely to generate momenta distributed as $\prod_f d^3p_f/[(2\pi)^3 2E_f]$ subject to the four-momentum delta function. When ATLAS or CMS quotes a cross section in femtobarns, the $1/F$ flux factor and the LIPS measure are baked into the definition. On the accelerator side, the surviving $1/2E_a$ is engineering: secondary beamlines for pions and kaons are built hundreds of metres long because their decay lengths scale as $L = (c\tau_0/m)\,p$ , and the same $\tau = \gamma\tau_0$ dilation lets short-lived muons survive long enough to be stored in the Fermilab $g\!-\!2$ ring, lets multi-GeV cosmic-ray muons reach the ground, and sets the baselines of long-baseline neutrino experiments. The factor of two that once looked like a fatal flaw in the Klein–Gordon equation is now a line of code in every event generator on Earth.

The physical picture

Hold a single particle in a cubical box and count how many there are per unit volume. In the box's rest frame the answer is fixed by the wavefunction's amplitude. Now run past the box: relativity contracts its length along your motion by $1/\gamma$ , the same particle is squeezed into a smaller volume, and you count a higher density. Nothing about the particle changed — only the frame did. A normalisation meant to be a physical fact about the state cannot depend on who is looking, so the density must be allowed to grow with energy. The first figure makes this concrete: the particle number is invariant, the volume contracts, and the density rises by exactly $\gamma$ .

Figure

The same N particles seen from a frame boosted by γ = 2. Their box Lorentz-contracts along the motion to V₀ divided by γ, so the number density rises from ρ₀ = 2m to ρ = γ times 2m = 2E = 4m. Density is the time-component of the four-current jᵘ = 2pᵘ, so it transforms exactly like the energy.

The remedy is to let the amplitude carry the energy. If the density of a unit-amplitude state is $2E$ , then the state itself is the frame-independent object and the density it predicts transforms the right way automatically. This is not a bookkeeping trick — it is what the Klein–Gordon probability current actually delivers, factor of two included. The second figure is the payoff a beamline engineer measures: because the surviving $1/2E_a$ in the decay rate dilates a particle's lifetime to $\tau = \gamma\tau_0$ , an unstable particle's mean flight distance grows linearly with its momentum.

Figure

Mean decay length of a relativistic charged pion, L = cτ₀/m times p = 55.9 m per GeV of momentum, using τ₀ = 26.0 ns and m = 139.6 MeV. The length is linear in p because the 1/2Eₐ normalisation factor dilates the lab lifetime to τ = γτ₀ — a 1 GeV pion flies 55.9 m, a 5 GeV pion nearly 280 m.

The core equations

The starting point is the non-relativistic convention that Fermi's golden rule assumes. A plane wave is normalised to exactly one particle in a cubical box of side $a$ and volume $V = a^3$ , which fixes its amplitude and, through periodic boundary conditions, the density of momentum states.

\int_V \psi^*\psi\,d^3x = 1,\qquad \psi(\vec x,t) = \frac{1}{\sqrt{V}}\,e^{i(\vec p\cdot\vec x - Et)} \qquad (1)

where: $V$ — box volume; $\psi$ — single-particle plane wave; $\vec p, E$ — momentum and energy. Periodic boundaries quantise $p_i = n_i\,2\pi/a$ , so each state fills a cell $(2\pi)^3/V$ and the number of states in $d^3p$ is $V\,d^3p/(2\pi)^3$ . Reading: the density $\rho = \psi^*\psi = 1/V$ is manifestly frame-dependent. The relativistic density is not put in by hand — it is read off the conserved current of the Klein–Gordon equation. For a plane wave $\phi = N\,e^{-ip\cdot x}$ the current collapses to a four-vector proportional to the momentum, whose time component is the density.

j^\mu = i\big(\phi^*\partial^\mu\phi - (\partial^\mu\phi^*)\phi\big) = 2|N|^2\,p^\mu,\qquad \rho = j^0 = 2|N|^2 E \qquad (2)

where: $j^\mu = (\rho,\vec\jmath)$ — probability four-current; $N$ — plane-wave amplitude; $p^\mu = (E,\vec p)$ — four-momentum. Reading: with unit amplitude $N = 1$ the density is $\rho = 2E$ , and because $j^\mu = 2p^\mu$ is a four-vector, $\rho$ transforms as the time component of $p^\mu$ — exactly like the energy. The factor of two is a fact about the current, not a choice. The fix is therefore to normalise every state to $2E$ particles per unit volume. Equivalently, one rescales the box-normalised wave by $\sqrt{2E}$ , which sets the continuum normalisation of momentum eigenstates.

\psi' = \sqrt{2E}\,\psi,\qquad \int_V \psi'^*\psi'\,d^3x = 2E,\qquad \langle p'|p\rangle = 2E\,(2\pi)^3\,\delta^3(\vec p'-\vec p) \qquad (3)

where: $\psi'$ — relativistically normalised state; $\langle p'|p\rangle$ — continuum inner product. Reading: the $2E$ appears in the delta-function normalisation because $2E\,\delta^3(\vec p'-\vec p)$ is Lorentz invariant while $\delta^3$ alone is not. This is the convention of Peskin–Schroeder, Halzen–Martin, Thomson and Schwartz. The scattering amplitude inherits the rescaling. The golden-rule matrix element $T_{fi}$ , built from box-normalised states, is traded for the Lorentz-invariant amplitude $\mathcal{M}_{fi}$ by stripping one factor of $\sqrt{2E_k}$ from every external leg.

\mathcal{M}_{fi} = \Big(\prod_{k\,\in\,\text{ext}} 2E_k\Big)^{1/2}\,T_{fi} = \big(2E_1\,2E_2\cdots 2E_a\,2E_b\cdots\big)^{1/2}\,T_{fi} \qquad (4)

where: the product runs over all initial- and final-state particles. Reading: because each external state carried a frame-dependent $1/\sqrt{2E_k}$ , multiplying them back out leaves $\mathcal{M}_{fi}$ a Lorentz scalar — all the physics of an interaction now lives in the single invariant $|\mathcal{M}_{fi}|^2$ . The same factors of $2E$ convert the phase-space measure into its invariant form. The element $d^3p/[(2\pi)^3\,2E]$ is not merely a definition — it equals a manifestly covariant integral over the positive-energy mass shell.

\frac{d^3p}{(2\pi)^3\,2E} = \frac{d^4p}{(2\pi)^4}\,(2\pi)\,\delta(p^2 - m^2)\,\theta(p^0) \qquad (5)

where: $\delta(p^2-m^2)$ — on-shell condition; $\theta(p^0)$ — positive-energy projection. Reading: every factor on the right is a Lorentz scalar, so the left side is invariant; this is the Lorentz-invariant phase space (LIPS) element that replaces the naive $d^3p/(2\pi)^3$ . Assembling the pieces gives the relativistic golden rule. For a decay $a \to 1 + 2 + \cdots + N$ the box volume cancels completely, leaving an invariant amplitude, an invariant measure, and a single frame-dependent prefactor.

\Gamma = \frac{1}{2E_a}\int |\mathcal{M}_{fi}|^2\,(2\pi)^4\,\delta^4\!\Big(p_a - \sum_{f} p_f\Big)\,\prod_{f=1}^{N}\frac{d^3p_f}{(2\pi)^3\,2E_f} \qquad (6)

where: $\Gamma$ — decay rate; $E_a$ — energy of the decaying particle. Reading: the only non-invariant object is the $1/2E_a$ out front, and it is exactly the time dilation — in the rest frame $\Gamma_0 \propto 1/2m_a$ , in flight $\Gamma = \Gamma_0/\gamma$ . For a $2\to N$ cross section the $1/2E_a$ is replaced by $1/F$ with the invariant flux $F = 4\sqrt{(p_1\cdot p_2)^2 - m_1^2 m_2^2}$ .

Derivation AWhy the relativistic density is exactly 2E

The goal is to show that the factor of two in " $2E$ particles per unit volume" is not a convention chosen for tidiness but is forced by the Klein–Gordon equation itself, and that the resulting density transforms correctly under boosts. We start from the relativistic wave equation, construct its conserved probability current directly from the equation of motion, evaluate that current on a plane wave, and read off both the magnitude $2E$ and the four-vector transformation law that guarantees Lorentz consistency. The whole argument fixes the normalisation with no freedom left except the overall amplitude $N$ .

Step 1. Take the free Klein–Gordon equation $(\Box + m^2)\phi = 0$ , multiply it by $\phi^*$ , subtract the complex conjugate multiplied by $\phi$ , and collect terms into a total divergence: $\partial_\mu j^\mu = 0$ with $j^\mu = i(\phi^*\partial^\mu\phi - (\partial^\mu\phi^*)\phi)$ . Why: a continuity equation is the only way a probability can be conserved locally, and this current is the unique bilinear the equation supplies.

Step 2. Insert the positive-energy plane wave $\phi = N\,e^{-ip\cdot x}$ with $p\cdot x = Et - \vec p\cdot\vec x$ , so that $\partial^\mu\phi = -ip^\mu\phi$ . Then $\phi^*\partial^\mu\phi = -ip^\mu|N|^2$ and $(\partial^\mu\phi^*)\phi = +ip^\mu|N|^2$ , and the current becomes $j^\mu = i(-ip^\mu - ip^\mu)|N|^2 = 2|N|^2 p^\mu$ . Why: the two terms add rather than cancel, and that is precisely where the factor of two is born — it is a consequence of the current's antisymmetric structure, not an external choice.

Step 3. Read off the time component: $\rho = j^0 = 2|N|^2 E$ . Setting the covariant amplitude to $N = 1$ gives $\rho = 2E$ and a spatial current $\vec\jmath = 2|N|^2\,\vec p = \rho\,\vec v$ , consistent with a stream of particles moving at velocity $\vec v = \vec p/E$ . Why: fixing $N = 1$ is the frame-independent statement — the amplitude is the same in every frame — and it produces a density of $2E$ rather than $1$ .

Step 4. Confirm Lorentz consistency. Since $j^\mu = 2p^\mu$ is a four-vector, $\rho = j^0$ transforms as its time component: boosting from rest ( $\rho_0 = 2m$ ) to energy $E = \gamma m$ gives $\rho = \gamma\rho_0 = 2E$ , matching the $1/\gamma$ contraction of the volume so the total number $\rho V$ is invariant. Why: this is the property the non-relativistic " $\rho = 1/V$ " lacked — the density now transforms correctly, and the state built from it is frame-independent.

Go deeper · Derivation B · 1 minThe Lorentz-invariant phase-space measure

Derivation BThe Lorentz-invariant phase-space measure

The aim is to prove that $d^3p/[(2\pi)^3\,2E]$ is a Lorentz scalar, so that the factors of $2E$ from the wavefunction normalisation are exactly what the phase-space integral needs to become frame-independent. We give two complementary arguments: a direct Jacobian computation for a boost, and an identity that writes the three-dimensional measure as a manifestly covariant four-dimensional integral over the mass shell. Together they show that the naive $d^3p$ is not invariant on its own but $d^3p/2E$ is.

Step 1. Boost along the $x$ -axis with velocity $\beta$ : the on-shell components transform as $E' = \gamma(E - \beta p_x)$ and $p'_x = \gamma(p_x - \beta E)$ , while $p_y, p_z$ are unchanged. Differentiate $p'_x$ at fixed transverse momenta, using $dE = (p_x/E)\,dp_x$ from $E^2 = |\vec p|^2 + m^2$ . Why: only the boost direction mixes energy and momentum, so all the non-invariance is concentrated in the single component $p_x$ .

Step 2. The Jacobian collapses: $dp'_x = \gamma\big(dp_x - \beta\,\tfrac{p_x}{E}dp_x\big) = \gamma\,\tfrac{E - \beta p_x}{E}\,dp_x = \tfrac{E'}{E}\,dp_x$ . Dividing by $E'$ gives $dp'_x/E' = dp_x/E$ , and since the transverse pieces are untouched, $d^3p'/E' = d^3p/E$ . Why: the same factor $E'/E$ that stretches the momentum interval also appears in the denominator, so the ratio is invariant — this is the whole content of the theorem.

Step 3. Reach the same result covariantly. Start from $\int d^4p\,\delta(p^2 - m^2)\,\theta(p^0)$ , and use $\delta(p^2 - m^2) = \delta(p_0^2 - |\vec p|^2 - m^2) = \tfrac{1}{2E}[\delta(p^0 - E) + \delta(p^0 + E)]$ . The $\theta(p^0)$ keeps only the positive root, so the $p^0$ integral leaves $\int d^3p/2E$ . Why: $d^4p$ is invariant, a delta function of an invariant is invariant, and the sign of $p^0$ is boost-invariant for timelike momenta, so the surviving $d^3p/2E$ must be invariant too.

Step 4. Verify numerically: a mass-1 particle with $p_x = 0.5$ has $E = 1.1180$ ; a boost $\beta = 0.6$ ( $\gamma = 1.25$ ) sends it to $E' = 1.0225$ , $p'_x = -0.2135$ (still on-shell, $E'^2 - p'^2_x = 1.000$ ). Then $dp_x/E = 0.8944$ and $dp'_x/E' = 1/1.1180 = 0.8944$ — identical. Why: an explicit number confirms the invariance is exact, not approximate.

Go deeper · Derivation C · 1 minFrom box golden rule to the relativistic decay rate

Derivation CFrom box golden rule to the relativistic decay rate

The purpose here is to trace how the arbitrary quantisation volume $V$ enters Fermi's golden rule and then cancels once every external particle is normalised to $2E$ , leaving the manifestly invariant decay formula with its tell-tale $1/2E_a$ prefactor. We track powers of $V$ and $2E$ through the amplitude and the density of states, show that the two exactly compensate, and identify the surviving factor as the relativistic time dilation of the lifetime. This is the step that turns a frame-dependent transition probability into a physical, frame-independent rate.

Step 1. Write the box golden rule for $a \to 1 + \cdots + N$ : the rate is $\int |T_{fi}|^2\,(2\pi)^4\delta^4(p_a - \sum p_f)\prod_f \big[V\,d^3p_f/(2\pi)^3\big]$ , where each final particle contributes one density-of-states factor $V\,d^3p_f/(2\pi)^3$ . Why: this is Fermi's golden rule in its covariant packaging, still written with unit-normalised box states, so every $V$ is visible and nothing has been hidden.

Step 2. Convert to the invariant amplitude. Each external leg's box wave carried amplitude $1/\sqrt{V}$ and must be rescaled by $\sqrt{2E_k}$ , so $T_{fi} = \big[\prod_{k\,\in\,\text{ext}}(2E_k V)^{-1/2}\big]\,\mathcal{M}_{fi}$ . Hence $|T_{fi}|^2$ supplies one factor $1/(2E_k V)$ for every initial and final particle. Why: the wavefunction normalisation is where the $2E$ factors physically live, and equation (4) is just this relation solved for $\mathcal{M}_{fi}$ .

Step 3. Count the volume. Every external leg contributes $V^{-1}$ from $|T_{fi}|^2$ , while each of the $N$ final legs contributes $V^{+1}$ from its density of states; together with the momentum-conserving delta function the powers cancel identically, so no factor of $V$ survives in $\Gamma$ . Why: a physical rate cannot depend on the size of an imaginary box, and the cancellation is the concrete check that the $2E$ normalisation is the self-consistent one.

Step 4. Collect what remains. The initial particle keeps one uncancelled $1/2E_a$ ; each final particle's $1/2E_f$ merges with $d^3p_f/(2\pi)^3$ to build the invariant measure $d^3p_f/[(2\pi)^3 2E_f]$ ; and $|\mathcal{M}_{fi}|^2$ is a scalar. This is equation (6), and since $\Gamma \propto 1/2E_a = 1/2\gamma m_a$ , the lab lifetime is $\tau = \gamma\tau_0$ . Why: the one factor that refuses to cancel is not a blemish but the physics — the time dilation of the moving particle's clock.

Limiting cases

Case	Result	Meaning
Rest frame	$\rho = 2m$ , $j^\mu = (2m,\vec 0)$	density is just twice the mass; no current flows
Non-relativistic limit $E\to m$	$\rho = 2E \to 2m$ (constant)	dividing states by $\sqrt{2m}$ recovers the NR "one per volume" and the Schrödinger normalisation
Ultra-relativistic $E\gg m$	$\rho = 2E \approx 2\|\vec p\|$	density grows without bound with energy; strong time dilation, long decay lengths
Massless particle	$d^3p/2E = d^3p/2\|\vec p\|$	measure stays finite and invariant; the phase space for photons and (approximately) neutrinos
Boost by $\gamma$	$\rho \to \gamma\rho_{\text{rest}}$ , $\Gamma \to \Gamma_0/\gamma$	four-vector transformation of $j^0$ ; lifetime dilates as $\tau = \gamma\tau_0$

Common trapThe factor of 2E is an arbitrary convention with no physical content

It is neither arbitrary nor content-free. The proportionality to $E$ is forced by Lorentz covariance: the probability density is the time component of a four-current, so it must transform like an energy, and any normalisation that fixed the density to a constant would hold in only one frame. The specific factor of $2$ — rather than $1$ or $\gamma$ — is forced by the Klein–Gordon current itself, which evaluates to $j^\mu = 2|N|^2 p^\mu$ on a plane wave; with unit amplitude the density is $2E$ , full stop. That same $2$ makes $2E\,\delta^3(\vec p'-\vec p)$ invariant, makes the measure $d^3p/(2\pi)^3 2E$ fall cleanly out of the on-shell delta function, and dovetails with the Dirac spinor normalisation $\bar u u = 2m$ . Change the $2$ and you must smuggle it back in elsewhere: the convention is bookkeeping, but the energy dependence and the value of the factor are physics. (A related trap: the box volume $V$ is genuinely arbitrary, and precisely because it is, it must cancel from every physical answer — a leftover $V$ is an error, not an effect.)

Worked examples

1 · The invariant measure survives a boost (fully worked)

Show explicitly that $d^3p/2E$ is unchanged by a Lorentz boost, using a mass-1 particle with momentum $p_x = 0.5$ (all in GeV) boosted along $x$ with $\beta = 0.6$ .

Step 1. Fix the initial on-shell state. With $m = 1$ and $p_x = 0.5$ , the energy is $E = \sqrt{m^2 + p_x^2} = \sqrt{1 + 0.25} = \sqrt{1.25} = \mathbf{1.1180}$ GeV. The momentum-space volume factor to be tested is $dp_x/E$ , and its numerical value is $1/1.1180 = \mathbf{0.8944}\,dp_x$ .

Step 2. Apply the boost. With $\beta = 0.6$ , $\gamma = 1/\sqrt{1 - 0.36} = 1.25$ , the transformed energy and momentum are $E' = \gamma(E - \beta p_x) = 1.25(1.1180 - 0.30) = \mathbf{1.0225}$ GeV and $p'_x = \gamma(p_x - \beta E) = 1.25(0.5 - 0.6708) = \mathbf{-0.2135}$ GeV. Check the mass shell: $E'^2 - p'^2_x = 1.0455 - 0.0456 = \mathbf{1.000} = m^2$ , so the boosted state is still a physical particle of mass 1.

Step 3. Compare the measures. The Jacobian of $p'_x$ at fixed transverse momentum is $dp'_x = (E'/E)\,dp_x = (1.0225/1.1180)\,dp_x = 0.9146\,dp_x$ , so $dp'_x/E' = 0.9146/1.0225\,dp_x = \mathbf{0.8944}\,dp_x$ — identical to $dp_x/E$ from Step 1. The naive $d^3p$ changed by the factor $0.9146$ , but $d^3p/2E$ did not change at all: the $2E$ in the denominator is exactly what the boost demands.

2 · The 1/2Eₐ factor is the pion's time dilation

A charged pion ( $m = 139.57$ MeV/c², rest lifetime $\tau_0 = 26.03$ ns) travels with momentum $p = 1.00$ GeV/c in a beamline. Using the relativistic golden rule, find its lab-frame lifetime and mean decay length.

Step 1. The rate carries a $1/2E_a$ prefactor, so $\Gamma_{\text{lab}} = \Gamma_0\,(m/E) = \Gamma_0/\gamma$ , i.e. $\tau_{\text{lab}} = \gamma\tau_0$ . Compute the Lorentz factor from the energy $E = \sqrt{p^2 + m^2} = \sqrt{1000^2 + 139.57^2} = \sqrt{1\,019\,480}\ \text{MeV} = 1009.7$ MeV, giving $\gamma = E/m = 1009.7/139.57 = \mathbf{7.234}$ and $\beta = p/E = 1000/1009.7 = \mathbf{0.9904}$ .

Step 2. Dilate the lifetime: $\tau_{\text{lab}} = \gamma\tau_0 = 7.234 \times 26.03\ \text{ns} = \mathbf{188.3\ ns}$ . Equivalently the lab rate is smaller by $1/\gamma$ : $\Gamma_{\text{lab}}/\Gamma_0 = 1/7.234 = 0.1382$ , so a moving pion decays at 13.8% of its rest-frame rate — a direct consequence of the single uncancelled normalisation factor.

Step 3. Convert to a distance. The mean flight before decay is $L = \gamma\beta\,c\,\tau_0 = (p/m)\,c\tau_0$ . With $c\tau_0 = (2.998\times10^8)(26.03\times10^{-9}) = 7.804$ m and $p/m = 1000/139.57 = 7.165$ , the decay length is $L = 7.165 \times 7.804 = \mathbf{55.9\ m}$ . A 1 GeV pion travels about 56 metres before decaying — the slope of the punchline figure, $cτ_0/m = 55.9$ m per GeV, and the reason secondary beamlines at CERN and Fermilab are hundreds of metres long.

Physics of the portrait

Relativistic normalisation is where a bookkeeping convention meets a genuine crisis in the history of quantum theory. The factor $2E$ traces back to the very first relativistic wave equation and the negative probabilities that nearly killed it.

Origins · 1 minOskar Klein and how this was discovered

Oskar Klein

1894–1977 · relativistic wave equations and the probability-current problem

Oskar Klein was a Swedish theorist whose name is attached to much of modern physics — the Klein–Gordon equation, the Klein–Nishina cross section, the Klein paradox, and the Kaluza–Klein extra dimension. In 1926, working from the relativistic relation $E^2 = p^2 + m^2$ , he (with Walter Gordon, and independently Schrödinger and Fock) wrote down the second-order wave equation $(\Box + m^2)\phi = 0$ . It was the natural relativistic successor to Schrödinger's equation, but it came with a scandal: the conserved density its current supplied was not positive-definite. For a plane wave it equals $2|N|^2 E$ — fine for positive energies, but the equation also admits negative-energy solutions with negative $\rho$ , and a probability cannot be negative. This "negative probability" problem led most of the community to abandon the equation, and it was one of the pressures that drove Dirac in 1928 to seek a first-order equation with a positive-definite density for spin-½. The resolution came in 1934, when Pauli and Weisskopf reinterpreted the Klein–Gordon field not as a single-particle probability amplitude but as a quantised field for charged spin-0 particles: $j^\mu$ is then a charge current, and its density may be positive or negative because charge can be. The equation was rehabilitated, and the very factor of $2E$ that had signalled the crisis became the standard relativistic normalisation — a recurring lesson that mathematics which looks like a fatal flaw is often a signpost to deeper structure.

History

1926

— Klein, Gordon, Schrödinger and Fock write the relativistic wave equation and its conserved current

1928

— Dirac's first-order equation restores a positive-definite density for spin-½ particles

1934

— Pauli and Weisskopf reinterpret the Klein–Gordon density as charge, resolving the negative-probability crisis

1949

— Feynman and Dyson's covariant perturbation theory makes the invariant amplitude and the d³p over 2E measure standard

1964

— Bjorken and Drell codify the 2E normalisation and Lorentz-invariant phase space in the textbook that trained a generation

Going further

Texts. Thomson, Modern Particle Physics, §3.2 — Phase space and wavefunction normalisation, §3.2.1 — Lorentz-invariant phase space, §3.2.2 — Fermi's golden rule revisited · Peskin & Schroeder, An Introduction to Quantum Field Theory, §4.5 (cross sections and the $S$ -matrix; the $2E_p$ state normalisation) · Halzen & Martin, Quarks and Leptons, §4.2–4.3 · Griffiths, Introduction to Elementary Particles, 2nd ed., §6.2 (the Golden Rule) · Schwartz, Quantum Field Theory and the Standard Model, §5.1 (Lorentz-invariant phase space) · Bjorken & Drell, Relativistic Quantum Mechanics, §3.

Provenance. The density $\rho = 2E$ is derived here directly from the Klein–Gordon current of aq-pp-nonrel-qm-review extended to relativistic kinematics; the box density of states and Fermi's golden rule are recalled from aq-pp-fermis-golden-rule; the $1/\gamma$ contraction and $E = \gamma m$ come from aq-pp-special-relativity-kinematics and aq-pp-four-vector-notation; natural units ( $\hbar = c = 1$ ) follow aq-pp-natural-units. A verifier should confirm: (1) the Klein–Gordon current gives $j^\mu = 2|N|^2 p^\mu$ and $\rho = 2E$ ; (2) $d^3p/2E$ is invariant, both by the boost Jacobian and by the on-shell delta-function identity, with the numerical check $m=1$ , $p_x=0.5$ , $\beta=0.6$ giving $dp_x/E = dp'_x/E' = 0.8944$ ; (3) the box volume $V$ cancels from the decay rate; (4) the pion numbers $\gamma = 7.234$ , $\tau_{\text{lab}} = 188$ ns, $L = 55.9$ m at $p = 1$ GeV.

Prerequisites. Non-relativistic QM review · Fermi's golden rule · Special relativity kinematics · Four-vector notation · Natural units. Leads to. Lorentz-invariant phase space · Particle decay rates · Interaction cross sections · Spinor normalisation.

Ask the tutor: Ask me to walk through how the same $2E$ normalisation turns the golden rule into a two-body decay width $\Gamma = |\mathcal{M}|^2\,p^*/(8\pi m_a^2)$ , or to show exactly where the box volume $V$ cancels in a full cross-section calculation.

Why exactly 2E, and not E or the square root of 2E?

The value is fixed by the Klein–Gordon current, not chosen. Evaluating the conserved current $j^\mu = i(\phi^*\partial^\mu\phi - (\partial^\mu\phi^*)\phi)$ on a unit-amplitude plane wave gives $j^\mu = 2p^\mu$ , so the density is $\rho = 2E$ with the factor of two included automatically. Once you accept that the density must be proportional to $E$ (which Lorentz covariance forces, since $\rho$ is the time component of a four-vector), the constant of proportionality is nailed down by the current itself. The same $2$ then makes the state normalisation $\langle p'|p\rangle = 2E(2\pi)^3\delta^3(\vec p'-\vec p)$ Lorentz invariant and makes the phase-space measure fall cleanly out of the on-shell delta function. You could rescale states differently, but only by moving the factor somewhere else — the $2E$ convention is the one where every invariant object stays clean.

Does the arbitrary box volume V ever appear in a physical answer?

Never — and that is the point. The volume $V$ is an artefact of the box-normalisation scaffolding: it labels a fictitious cavity you imagine the particle living in so that plane waves become countable states. Every factor of $V$ from a density of states $V\,d^3p/(2\pi)^3$ is matched by a $1/V$ from a wavefunction normalisation $1/\sqrt{V}$ , and in a correctly assembled decay rate or cross section they cancel identically, leaving the invariant amplitude and phase space. If you ever find a stray $V$ surviving into a physical rate, it is a sign that you have mismatched the number of density-of-states factors and normalisation factors somewhere — not that the answer depends on the box. The cancellation is a useful consistency check to run on any calculation.

Is the negative Klein-Gordon probability density a real problem?

It was historically, and its resolution is one of the reasons quantum field theory exists. The Klein–Gordon density $\rho = 2|N|^2 E$ is positive for positive-energy solutions, but the equation also has negative-energy solutions for which $\rho$ is negative — impossible for a probability. This drove Schrödinger to drop the equation and Dirac to invent his own. The modern resolution, due to Pauli and Weisskopf in 1934, is to reinterpret $\phi$ as a quantum field rather than a single-particle wavefunction: then $j^\mu$ is the charge current, and a negative density simply means antiparticles, whose charge is opposite. So the "negative probability" was really telling us the theory necessarily contains antiparticles. For the purposes of normalisation, we always work with the positive-energy piece, where $\rho = 2E > 0$ is perfectly sensible.

Why is δ³ of p not Lorentz invariant but 2E times δ³ is?

Because a three-dimensional delta function transforms with a Jacobian under a boost. The identity $\int d^3p\,\delta^3(\vec p'-\vec p) = 1$ involves the measure $d^3p$ , which we saw scales by $E'/E$ along the boost direction; to keep the integral equal to one, $\delta^3(\vec p'-\vec p)$ must scale by the inverse, $E/E'$ , so it is not invariant on its own. Multiplying by $2E$ supplies exactly the compensating factor: $2E\,\delta^3(\vec p'-\vec p)$ picks up $E'/E$ from the $2E$ and $E/E'$ from the delta function, and the two cancel. This is the same statement as the invariance of $d^3p/2E$ , viewed through the completeness relation $\int d^3p/[(2\pi)^3 2E]\,|p\rangle\langle p| = 1$ instead of through the measure directly.

How does this connect to the Dirac spinor normalisation?

It is the spin-½ version of the same convention. For the Dirac equation the density $\psi^\dagger\psi$ is already positive-definite, and the spinors are normalised to match the $2E$ convention: $\bar u^{(s)} u^{(r)} = 2m\,\delta^{rs}$ and $u^{\dagger(s)} u^{(r)} = 2E\,\delta^{rs}$ . The $u^\dagger u = 2E$ is exactly " $2E$ per unit volume" for a spinor field, and $\bar u u = 2m$ is its rest-frame value — the direct analogue of $\rho_{\text{rest}} = 2m$ for the scalar. Keeping the same $2E$ across scalars and spinors is what lets one invariant golden rule and one invariant phase space serve particles of any spin, with all the spin information inside $|\mathcal{M}|^2$ . The details are developed in aq-pp-spinor-normalisation.

In the non-relativistic limit, where does the extra factor of 2m go?

It is divided back out. As $E \to m$ the density becomes $\rho = 2E \to 2m$ , a constant, so the relativistically normalised state differs from the Schrödinger-normalised one by an energy-independent factor $\sqrt{2m}$ . To recover ordinary non-relativistic quantum mechanics you absorb $\sqrt{2m}$ into each field, cancelling the $2m$ in the density and restoring the "one particle per unit volume" convention. This is exactly the $\phi = \psi/\sqrt{2m}$ relation between the relativistic field and the non-relativistic wavefunction, and it is why golden-rule cross sections agree with the relativistic formula in the $E \to m$ limit. The $2E$ normalisation contains the non-relativistic one as a smooth limit, not a different theory.