Lorentz-invariant phase space

Fermi's golden rule splits every rate into two clean halves — the dynamics, packed into a matrix element, and the kinematics, the number of final states the process can reach. In ordinary quantum mechanics that state-counting is easy: normalise plane waves to one particle in a box of volume $V$ , quantise the momenta, and each momentum cell carries a weight $d^3p/(2\pi)^3$ . It works, and it is also quietly broken. "One particle per unit volume" holds in only one frame; boost to another and the box Lorentz-contracts, so the same particle now crowds into a smaller volume. A quantity meant to count physical states cannot depend on how fast the observer moves. Lorentz-invariant phase space is the repair. The cure is forced, not chosen: because the normalisation volume contracts by $1/\gamma=m/E$ , you must normalise instead to $2E$ particles per unit volume, which exactly cancels the contraction and pushes the leftover energy factors into a redefined, invariant matrix element $\mathcal{M}$ . What survives for each final particle is the measure $d^3p/(2\pi)^3\,2E$ , and the miracle is that this combination is genuinely frame-independent — because $d^3p/E$ is invariant under a boost. Assemble one factor per final particle, staple on four-momentum conservation, and you have $d\mathrm{LIPS}$ : the covariant density of states that turns a raw amplitude into a decay width or a cross section. This article builds that measure from the box argument, writes it in covariant mass-shell form, integrates the two-body case in closed form, and shows where it does its daily work — from particle widths in the data tables to the Dalitz plots and Monte-Carlo generators of modern experiments.

Key takeaways

The non-relativistic density of states $\prod_i d^3p_i/(2\pi)^3$ (box volume set to one) is not Lorentz invariant, because "one particle per unit volume" holds only in one frame.
Length contraction of the normalisation box by $1/\gamma=m/E$ forces the relativistic choice of $2E$ particles per unit volume, i.e. $\langle p|p'\rangle=2E(2\pi)^3\delta^3(\mathbf{p}-\mathbf{p}')$ .
The surviving per-particle measure $d^3p/(2\pi)^3\,2E$ is invariant because $d^3p/E$ is invariant under a boost — the $2E$ normalisation exactly compensates the contraction.
Covariantly, $d^3p/(2\pi)^3\,2E = d^4p/(2\pi)^4\,(2\pi)\delta(p^2-m^2)\theta(p^0)$ : every final particle is pinned to the forward mass shell.
Two-body phase space integrates in closed form to $d\Gamma = |\mathcal{M}|^2\,|\mathbf{p}^*|\,d\Omega/32\pi^2 M^2$ , with $|\mathbf{p}^*|=\lambda^{1/2}(M^2,m_1^2,m_2^2)/2M$ .
Key idea: normalising wavefunctions to $2E$ particles per unit volume exactly compensates the Lorentz contraction of the normalisation box, turning the naive density of states into the frame-independent measure $d^3p/(2\pi)^3\,2E$ — so that all the physics lives in the invariant matrix element.

The core insight

Normalising wavefunctions to $2E$ particles per unit volume exactly compensates for the Lorentz contraction of the normalisation box, turning the naive density of states into a frame-independent measure $d^3p/(2\pi)^3\,2E$ — so that all the physics lives in the invariant matrix element.

Where this lives today

Lorentz-invariant phase space is executed billions of times a day inside the software that runs particle physics. Every Monte-Carlo event generator — MadGraph, Pythia, Sherpa, POWHEG — produces simulated collisions by sampling points on $d\mathrm{LIPS}$ and weighting them by $|\mathcal{M}|^2$ ; the classic RAMBO algorithm generates flat, unbiased massless $n$ -body phase space in exactly this measure, underpinning the acceptance calculations behind every LHC measurement. The Particle Data Group's tabulated widths and branching ratios are, at bottom, LIPS integrals of squared amplitudes. Three-body decays live on Dalitz plots — because three-body phase space is flat in the plane of invariant-mass-squared pairs $(m_{12}^2,m_{23}^2)$ , any structure there is pure dynamics: resonance bands, interference, and (at LHCb, BESIII and Belle II) the CP-violating asymmetries that probe physics beyond the Standard Model. Phase-space suppression is itself physics: the Higgs decay to two on-shell $Z$ bosons is forbidden below $2m_Z$ , so the "golden channel" $H\to ZZ^*\to4\ell$ proceeds through one off-shell $Z$ , its rate sculpted by the threshold factors above. And every neutrino-mass endpoint experiment, from the Kurie plot to KATRIN's tritium spectrum, reads the shape of final-state phase space near its kinematic edge.

The physical picture

Start with the counting quantum mechanics hands you for free. Confine a particle to a cube of side $a$ with periodic boundary conditions and its momentum components quantise to $p_j = n_j\,2\pi/a$ , so each state occupies a momentum-space cell $(2\pi)^3/V$ with $V=a^3$ . The number of states in a shell $\mathbf{p}\to\mathbf{p}+d\mathbf{p}$ is $dn = V\,d^3p/(2\pi)^3$ , and with the box volume set to one it is just $d^3p/(2\pi)^3$ per particle — the "density of states" that closes Fermi's golden rule. The trouble hides in "box volume set to one": a box measuring one unit of volume to an observer at rest measures only $1/\gamma$ to one flying past, because the dimension along the boost is contracted. One particle in one unit of volume becomes one particle in $1/\gamma$ units — a density of $\gamma=E/m$ . A state count that changes with the observer's speed is no state count at all.

Figure

Why one particle per unit volume is not invariant: a rest-frame box holding one particle per unit volume is Lorentz-contracted by 1/γ under a boost, so the same particle is squeezed into a smaller volume and the density rises to γ = E/m per unit volume. Normalising instead to 2E particles per unit volume cancels the contraction and makes the measure d³p/(2π)³2E frame-independent

The repair writes itself. If the density of a single-particle state scales as $E$ under a boost, then to keep a frame-independent normalisation you must build that same energy factor into the wavefunction from the start — normalise to $2E$ particles per unit volume rather than one. (The factor is $2E$ , not $E$ , for a reason that surfaces below: it is what the mass-shell delta $\delta(p^2-m^2)$ supplies.) The relativistic states obey $\langle p|p'\rangle=2E(2\pi)^3\delta^3(\mathbf{p}-\mathbf{p}')$ , the extra $\sqrt{2E}$ per particle is absorbed into a redefined invariant amplitude $\mathcal{M}$ , and each final particle carries the measure $d^3p/(2\pi)^3\,2E$ . The deeper statement is geometric: every real particle lives on its mass shell $E=\sqrt{\mathbf{p}^2+m^2}$ , and $d^3p/2E$ is precisely the Lorentz-invariant area element on that hyperboloid — shown in the second figure with the on-shell point of a real decay product.

Figure

The invariant measure lives on the forward mass shell E = √(p² + m²), equivalently p² = m² with E greater than zero. Each final particle is pinned to this hyperbola, and d³p/(2π)³2E is the frame-independent surface element on it — shown with the pion daughter of the decay K⁰ → π⁺π⁻, which sits at momentum p = 206 MeV and energy E = 248.8 MeV

Two features of the second figure repay a look. The shell is forward — only $E>0$ , the upper branch of $E^2=\mathbf{p}^2+m^2$ — because a physical final particle carries positive energy; that is the job of $\theta(p^0)$ in the covariant form. And the measure is denser near the vertex (small $\mathbf{p}$ , where $1/2E$ is largest), the invariant way of saying low-momentum final states are "cheaper" to reach. Everything downstream — widths, cross sections, Dalitz plots — is an integral of dynamics against this one surface element, one factor per final particle, constrained by four-momentum conservation.

The core equations

The whole edifice rests on a single per-particle measure and its assembly into a full final state. The invariant phase-space element for an $n$ -body final state staples one factor of $d^3p/(2\pi)^3\,2E$ per particle onto a four-dimensional delta function enforcing energy–momentum conservation.

d\mathrm{LIPS}=(2\pi)^4\,\delta^4\!\Big(P-\sum_{i=1}^{n}p_i\Big)\,\prod_{i=1}^{n}\frac{d^3p_i}{(2\pi)^3\,2E_i} \qquad (1)

where: $P$ — total incoming four-momentum; $p_i=(E_i,\mathbf{p}_i)$ — four-momentum of final particle $i$ , on shell so $E_i=\sqrt{\mathbf{p}_i^2+m_i^2}$ ; the $(2\pi)$ factors are the bookkeeping of the box normalisation. Reading: every factor here is separately Lorentz invariant — each $d^3p_i/2E_i$ by the boost argument, and $\delta^4(\cdots)$ because a four-momentum delta is invariant — so the entire measure is frame-independent. The invariance of the per-particle factor is most transparent when written covariantly. Integrating a four-dimensional momentum against a mass-shell delta function and a positive-energy step function reproduces exactly the three-dimensional measure, which is the cleanest proof that $d^3p/2E$ is invariant.

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

where: $\delta(p^2-m^2)$ — pins the particle to its mass shell; $\theta(p^0)$ — keeps only positive energy; metric signature $(+,-,-,-)$ . Reading: the right-hand side is manifestly invariant — $d^4p$ , the scalar $p^2-m^2$ , and the sign of $p^0$ under orthochronous boosts are all invariant — so the left-hand side must be too, and the factor $2E$ is not arbitrary but is what $\delta(p^2-m^2)$ generates on integrating over $p^0$ . The measure only becomes a rate when contracted with dynamics. For a decay of a particle of mass $M$ at rest, and for a $2\to n$ scattering process, the golden rule in covariant form reads as follows, with the flux factor $F$ supplying the incoming normalisation.

\Gamma=\frac{1}{2M}\!\int|\mathcal{M}|^2\,d\mathrm{LIPS},\qquad \sigma=\frac{1}{F}\!\int|\mathcal{M}|^2\,d\mathrm{LIPS},\qquad F=4\sqrt{(p_a\!\cdot\! p_b)^2-m_a^2m_b^2} \qquad (3)

where: $\mathcal{M}$ — the Lorentz-invariant matrix element (the $2E$ factors already absorbed); $1/2M$ — the decaying particle's own normalisation factor; $F=4E_aE_b|\mathbf{v}_a-\mathbf{v}_b|$ in the collinear case. Reading: all the frame dependence of a rate is now explicit and physical — $\Gamma\propto1/E_a$ in a moving frame reproduces relativistic time dilation, since $E_a=\gamma M$ . Carrying out the two-body integral in closed form gives the single most-used result in the subject: the differential decay rate into two bodies depends on the dynamics only through $|\mathcal{M}|^2$ and on the kinematics only through the common centre-of-mass momentum $|\mathbf{p}^*|$ .

\frac{d\Gamma}{d\Omega}=\frac{|\mathcal{M}|^2}{32\pi^2}\,\frac{|\mathbf{p}^*|}{M^2},\qquad |\mathbf{p}^*|=\frac{\lambda^{1/2}(M^2,m_1^2,m_2^2)}{2M},\qquad \lambda(x,y,z)=x^2+y^2+z^2-2xy-2yz-2zx \qquad (4)

where: $|\mathbf{p}^*|$ — magnitude of either daughter's momentum in the rest frame of $M$ ; $\lambda$ — the Källén (triangle) function; $d\Omega$ — solid angle of one daughter. Reading: near threshold $|\mathbf{p}^*|\to0$ so the rate is phase-space suppressed, while far above threshold $|\mathbf{p}^*|\to M/2$ and the kinematic factor saturates — the decay rate is then set almost entirely by $|\mathcal{M}|^2$ .

Derivation AFrom the density of states to an invariant measure

The goal is to trace, without hand-waving, exactly why the naive quantum-mechanical density of states fails to be Lorentz invariant and how the choice of a $2E$ wavefunction normalisation repairs it, ending with an explicit proof that the combination $d^3p/E$ — and hence $d^3p/2E$ — is unchanged by a boost. We work in the box-quantisation picture, follow the normalisation volume under a boost, and then track a momentum-space cell through the Lorentz transformation of the energy–momentum four-vector.

Step 1. Quantise momenta in a cube of side $a$ with periodic boundary conditions: $p_j=n_j\,2\pi/a$ , so each state fills a momentum cell $(2\pi)^3/V$ and the number of states in $d^3p$ is $dn=V\,d^3p/(2\pi)^3$ . Setting the box to unit volume gives the per-particle weight $d^3p/(2\pi)^3$ that closes Fermi's golden rule. Why: this fixes the starting measure and makes explicit that it was defined relative to a specific normalisation volume $V$ — the object that will not survive a boost.

Step 2. Boost the box. A cube of unit volume in its rest frame has one dimension along the boost contracted by $1/\gamma$ , so its volume in the moving frame is $1/\gamma=m/E$ . "One particle per unit rest-frame volume" therefore reads as $\gamma=E/m$ particles per unit volume to the moving observer. Why: this is the concrete failure — a state count that was supposed to be physical instead scales as $E$ , so the measure must be rescaled to cancel it.

Step 3. Compensate by normalising to $2E$ particles per unit volume, i.e. rescale the plane waves $\psi\to(2E)^{1/2}\psi$ so that $\langle p|p'\rangle=2E(2\pi)^3\delta^3(\mathbf{p}-\mathbf{p}')$ , and absorb the resulting $\sqrt{2E}$ per external leg into a redefined amplitude $\mathcal{M}$ . The leftover kinematic weight per final particle is then $d^3p/(2\pi)^3\,2E$ . Why: the energy factor built into the normalisation is exactly what a frame-dependent density needs to become frame-independent; the price is only a redefinition of the amplitude, which carries no physics of its own.

Step 4. Prove the invariance directly. Under a boost of rapidity along $z$ , $E'=\gamma(E+\beta p_z)$ and $p_z'=\gamma(p_z+\beta E)$ with $p_x,p_y$ fixed. On shell $E=E(\mathbf{p})$ , so differentiating at fixed transverse momentum gives $dp_z'/dp_z=\gamma(1+\beta\,\partial E/\partial p_z)=\gamma(1+\beta p_z/E)=\gamma(E+\beta p_z)/E=E'/E$ . Hence $d^3p'=(E'/E)\,d^3p$ and $d^3p'/E'=d^3p/E$ . Why: this is the beating heart of the whole construction — the Jacobian of the boost on the momentum cell is precisely $E'/E$ , so dividing by $E$ (equivalently by $2E$ ) produces an invariant, with no reference to any frame left over.

Go deeper · Derivation B · 1 minThe covariant mass-shell form of the measure

Derivation BThe covariant mass-shell form of the measure

The aim is to derive identity (2) — that the three-dimensional measure $d^3p/(2\pi)^3\,2E$ equals the four-dimensional integral $d^4p/(2\pi)^4\,(2\pi)\delta(p^2-m^2)\theta(p^0)$ — because that four-dimensional form is manifestly Lorentz invariant and, as a bonus, explains where the specific factor of $2E$ (rather than $E$ , or $\sqrt{2E}$ ) comes from. We use the standard rule for a delta function of a function with simple zeros and then integrate out the energy component.

Step 1. Treat $p^0$ as an independent integration variable, so $p=(p^0,\mathbf{p})$ ranges over all of four-momentum space, and write the invariant $p^2-m^2=(p^0)^2-\mathbf{p}^2-m^2=(p^0)^2-E^2$ with $E\equiv+\sqrt{\mathbf{p}^2+m^2}$ . Why: promoting $p^0$ to a free variable is what makes the measure $d^4p$ and the scalar constraint $\delta(p^2-m^2)$ separately invariant; the on-shell condition is then imposed by the delta rather than assumed.

Step 2. Use $\delta(g(p^0))=\sum_k\delta(p^0-p^0_k)/|g'(p^0_k)|$ for the two roots $p^0=\pm E$ . Here $g(p^0)=(p^0)^2-E^2$ so $g'=2p^0$ , giving $\delta(p^2-m^2)=\dfrac{1}{2E}\big[\delta(p^0-E)+\delta(p^0+E)\big]$ . Why: this is exactly where the factor $2E$ is born — it is the derivative of the mass-shell constraint evaluated at the root, not a convention put in by hand.

Step 3. Impose $\theta(p^0)$ to discard the negative-energy root $p^0=-E$ , keeping only $\delta(p^0-E)/2E$ , and integrate over $p^0$ : $\int dp^0\,(2\pi)\delta(p^2-m^2)\theta(p^0)=2\pi/2E$ . Dividing by $(2\pi)^4$ leaves $d^3p/(2\pi)^3\,2E$ , which is identity (2). Why: a physical final particle has positive energy, so only the forward mass shell contributes; the surviving factor $1/2E$ is precisely the invariant weight of the three-dimensional measure.

Go deeper · Derivation C · 1 minIntegrating two-body phase space in the centre of mass

Derivation CIntegrating two-body phase space in the centre of mass

The purpose here is to reduce the two-particle element $d\mathrm{LIPS}_2$ to a single closed-form kinematic factor, showing that after the delta functions are used up the only survivor is the common centre-of-mass momentum $|\mathbf{p}^*|$ , and thereby to derive the master formula (4) for two-body decays. We work in the rest frame of the total four-momentum $P=(\sqrt{s},\mathbf{0})$ , integrate the three-momentum delta against one particle's momentum, and convert the remaining energy delta into a constraint on $|\mathbf{p}_1|$ .

Step 1. Write $d\mathrm{LIPS}_2=(2\pi)^4\delta^4(P-p_1-p_2)\,\dfrac{d^3p_1}{(2\pi)^3 2E_1}\dfrac{d^3p_2}{(2\pi)^3 2E_2}$ and use the three spatial deltas to do the $d^3p_2$ integral, setting $\mathbf{p}_2=-\mathbf{p}_1$ . What remains is $\displaystyle\int\frac{d^3p_1}{(2\pi)^2\,4E_1E_2}\,\delta(\sqrt{s}-E_1-E_2)$ with $E_1=\sqrt{\mathbf{p}_1^2+m_1^2}$ and $E_2=\sqrt{\mathbf{p}_1^2+m_2^2}$ . Why: momentum conservation in the centre of mass simply makes the two daughters back-to-back, collapsing six integration variables to three and one energy constraint.

Step 2. Split $d^3p_1=p^2\,dp\,d\Omega$ (writing $p\equiv|\mathbf{p}_1|$ ) and change the energy delta into a delta in $p$ . With $W(p)=E_1+E_2$ , $\dfrac{dW}{dp}=\dfrac{p}{E_1}+\dfrac{p}{E_2}=\dfrac{p\,W}{E_1E_2}$ , so $\delta(\sqrt{s}-W)=\dfrac{E_1E_2}{p\,\sqrt{s}}\,\delta(p-p^*)$ , where $p^*$ solves $E_1+E_2=\sqrt{s}$ . Why: the Jacobian of the energy constraint is where the $E_1E_2$ in the denominator gets cancelled, leaving a clean kinematic factor.

Step 3. Perform the $p$ integral: the $p^2/(4E_1E_2)$ times $E_1E_2/(p\sqrt{s})$ collapses to $p^*/4\sqrt{s}$ , giving $\displaystyle\int d\mathrm{LIPS}_2=\frac{1}{(2\pi)^2}\frac{p^*}{4\sqrt{s}}\int d\Omega=\frac{p^*}{4\pi\sqrt{s}}$ once the solid angle is integrated. Solving $E_1+E_2=\sqrt{s}$ for $p^*$ gives $|\mathbf{p}^*|=\lambda^{1/2}(s,m_1^2,m_2^2)/2\sqrt{s}$ . Why: the total two-body phase space is thus a pure number times $p^*/\sqrt{s}$ — all dynamics has been separated out into $\mathcal{M}$ .

Step 4. Assemble the rate. For a decay, $\sqrt{s}=M$ and $d\Gamma=(1/2M)|\mathcal{M}|^2\,d\mathrm{LIPS}_2$ , so the differential form is $d\Gamma/d\Omega=|\mathcal{M}|^2\,p^*/32\pi^2M^2$ , which is equation (4). Why: this is the workhorse formula for every two-body width in the tables — feed it a squared amplitude and a Källén momentum and it returns a rate.

Limiting cases

Case	Result	Meaning
Non-relativistic ( $E\to m$ )	$d^3p/(2\pi)^3\,2E\to d^3p/(2\pi)^3\,2m$	the $2E$ factor becomes a constant $2m$ ; the invariant measure reduces to the ordinary density of states up to a fixed normalisation
Ultra-relativistic / massless	$E=\|\mathbf{p}\|$ , measure $=d^3p/(2\pi)^3\,2\|\mathbf{p}\|$	the shell degenerates to the light cone; still invariant, and $\|\mathbf{p}^*\|\to\sqrt{s}/2$
Two-body at threshold	$\sqrt{s}\to m_1+m_2$ , $\|\mathbf{p}^*\|\to0$	phase space closes linearly in $\|\mathbf{p}^*\|$ ; the rate is phase-space suppressed regardless of $\|\mathcal{M}\|^2$
Two-body well above threshold	$\|\mathbf{p}^*\|\to\sqrt{s}/2$ , $\Phi_2\to1/8\pi$	the kinematic factor saturates; the rate is controlled by the dynamics $\|\mathcal{M}\|^2$
$n$ -body near threshold	$\Phi_n\propto (\sqrt{s}-\sum m_i)^{(3n-5)/2}$	more bodies means a steeper, slower turn-on above threshold — why multi-body channels open gradually
Boosted decaying particle	$\Gamma_{\text{lab}}=\Gamma_{\text{rest}}/\gamma$	the covariant $1/2E$ prefactor reproduces relativistic time dilation of the lifetime

Common trapThe naive density of states d³p/(2π)³ is already Lorentz invariant

It is not — and treating it as if it were is the most common way to scatter spurious energy factors through a rate calculation. The object $d^3p/(2\pi)^3$ was defined relative to one particle per unit volume, true in only one frame; under a boost the transverse dimensions are untouched but the longitudinal one contracts by $1/\gamma$ , so a momentum-space cell transforms with Jacobian $E'/E$ and the bare $d^3p$ is not invariant. Only the ratio $d^3p/E$ is. The factor $1/2E$ — with the matching $2E$ wavefunction normalisation and the invariant amplitude $\mathcal{M}$ that absorbs it — is exactly what restores frame independence. A distinct trap is imagining the split of factors between "measure" and "amplitude" is physical: different textbooks park the $2E$ 's differently, but the product $|\mathcal{M}|^2\,d\mathrm{LIPS}$ , and hence every measurable rate, is convention-independent.

Worked examples

1 · Two-body momentum in a real decay (confidence builder)

The neutral kaon decays $K^0\to\pi^+\pi^-$ . With $m_K=497.6$ MeV and $m_\pi=139.6$ MeV, find the momentum of each pion in the kaon rest frame and the total two-body phase-space factor.

Step 1. Because the two daughters have equal mass, the Källén momentum simplifies. Starting from $|\mathbf{p}^*|=\lambda^{1/2}(M^2,m^2,m^2)/2M$ and expanding the triangle function with $m_1=m_2=m$ gives $|\mathbf{p}^*|=\tfrac12\sqrt{M^2-4m^2}$ — the daughter momentum is set by how far the parent mass sits above the $2m$ threshold.

Step 2. Insert the numbers: $M^2-4m^2=497.6^2-4(139.6)^2=247606-77953=169653\ \mathrm{MeV}^2$ , so $|\mathbf{p}^*|=\tfrac12\sqrt{169653}=\tfrac12(411.9)=\mathbf{206\ MeV}$ . Each pion also has energy $E_\pi=M/2=\mathbf{248.8\ MeV}$ , and one checks $\sqrt{206^2+139.6^2}=248.8$ MeV — the daughter sits exactly on its mass shell, the on-shell point drawn in the second figure.

Step 3. The integrated two-body phase space is $\Phi_2=|\mathbf{p}^*|/4\pi\sqrt{s}=206/(4\pi\times497.6)=\mathbf{0.0330}$ (dimensionless). This is the pure kinematic weight; to get the physical width one multiplies by $|\mathcal{M}|^2/2M$ . The pions emerge back-to-back at 206 MeV each, and the two-body phase-space factor is Φ₂ = 0.033.

2 · Checking that d³p/E is invariant under a boost (fully worked)

Take the same pion, at rest-frame energy $E=248.8$ MeV and momentum $p_z=206$ MeV along $z$ (so $m=139.6$ MeV). Boost along $z$ with $\beta=0.6$ . Verify explicitly that the momentum-space cell transforms with Jacobian $E'/E$ , so that $d^3p/E$ is unchanged.

Step 1. Compute the boost factor: $\gamma=1/\sqrt{1-0.6^2}=1/\sqrt{0.64}=\mathbf{1.25}$ . The transverse components $p_x,p_y$ are untouched, so only the $z$ – $E$ block matters.

Step 2. Transform the four-momentum: $E'=\gamma(E+\beta p_z)=1.25(248.8+0.6\times206)=1.25(372.4)=\mathbf{465.5\ MeV}$ and $p_z'=\gamma(p_z+\beta E)=1.25(206+0.6\times248.8)=1.25(355.3)=\mathbf{444.1\ MeV}$ . The particle stays on shell: $E'^2-p_z'^2=465.5^2-444.1^2=19480\approx139.6^2$ , the invariant mass, as it must.

Step 3. Compute the two quantities that identity requires to be equal. The energy ratio is $E'/E=465.5/248.8=\mathbf{1.871}$ . The Jacobian of the boost on the longitudinal momentum, at fixed transverse momentum and on shell, is $dp_z'/dp_z=\gamma(1+\beta p_z/E)=1.25(1+0.6\times206/248.8)=1.25(1.497)=\mathbf{1.871}$ .

Step 4. They agree to all quoted digits, so $d^3p'=(E'/E)\,d^3p$ and therefore $d^3p'/E'=d^3p/E$ : the cell stretched by $1.871$ in $p_z$ while its energy weight grew by the identical $1.871$ , and the ratio is untouched. The boost Jacobian equals E′/E = 1.871 exactly, confirming that d³p/E — and hence the LIPS measure d³p/(2π)³2E — is Lorentz invariant.

Physics of the portrait

Lorentz-invariant phase space is where three strands of twentieth-century physics braid together: Fermi's rule for transition rates, Wigner's classification of relativistic states, and the covariant amplitudes of post-war field theory. No single person "discovered" it; it crystallised as the natural bookkeeping once amplitudes were made manifestly invariant.

Origins · 1 minLorentz-invariant phase space and how this was discovered

Lorentz-invariant phase space

the covariant measure d³p/(2π)³2E for final states · relativistic decay and scattering rates

The measure has no single author — it is the residue left when several deep ideas are made mutually consistent. Its earliest ancestor is Paul Dirac's 1927 time-dependent perturbation theory, whose transition-rate formula Enrico Fermi taught so effectively that it became "Fermi's golden rule"; Fermi's own 1934 theory of beta decay was the first place the rule was married to an explicit phase-space factor for the outgoing electron and neutrino, whose shape (the "Kurie plot") became a diagnostic of the neutrino's mass. The relativistic normalisation the measure needs was set on firm ground by Eugene Wigner's 1939 classification of the unitary representations of the Poincaré group: a one-particle state is labelled by its mass and spin, and the invariant inner product $\langle p|p'\rangle=2E(2\pi)^3\delta^3(\mathbf{p}-\mathbf{p}')$ is the natural one for it. The final ingredient was the invariant matrix element itself, delivered by the covariant formulation of QED — Tomonaga, Schwinger and Feynman in the late 1940s, unified by Freeman Dyson in 1949 — which showed that once amplitudes are computed covariantly, the leftover kinematics organises itself into exactly $\prod_i d^3p_i/(2\pi)^3 2E_i$ . Gunnar Källén's triangle function $\lambda$ , popularised through his 1964 monograph, gave two-body kinematics its compact closed form. What began as Fermi's practical recipe for counting states thus became, by the 1950s, a manifestly Lorentz-invariant object — the kinematic skeleton on which every modern rate calculation hangs.

History

1927

— Dirac derives the time-dependent transition-rate formula later taught as Fermi's golden rule

1934

— Fermi's theory of β decay pairs the rule with an explicit final-state phase-space factor

1939

— Wigner classifies relativistic one-particle states by mass and spin, fixing the 2E normalisation

1949

— Dyson unifies the Tomonaga, Schwinger and Feynman programmes into a covariant S-matrix and invariant amplitude

1964

— Källén's monograph standardises the triangle function λ for two-body kinematics

Going further

Texts. Thomson, Modern Particle Physics, §3.2 (phase space and wavefunction normalisation) and eqs. (3.6)–(3.12) — the closest match to this article's development · Peskin & Schroeder, An Introduction to Quantum Field Theory, §4.5 (the master formulas for decay rates and cross sections) · Griffiths, Introduction to Elementary Particles, 2nd ed., §6.2 and the "Golden Rules" of Ch. 6 · Halzen & Martin, Quarks and Leptons, §4.3 · Schwartz, Quantum Field Theory and the Standard Model, §5.1.

Provenance. The $2E$ box-normalisation argument follows Thomson §3.2; the covariant mass-shell identity (2) is the standard delta-function manipulation in Peskin–Schroeder §4.5; the two-body reduction (4) and the Källén function follow Griffiths §6.2. A verifier should confirm: (1) the boost Jacobian $dp_z'/dp_z=E'/E$ ; (2) the mass-shell identity $d^3p/(2\pi)^3 2E=d^4p/(2\pi)^4(2\pi)\delta(p^2-m^2)\theta(p^0)$ ; (3) the two-body integral $\int d\mathrm{LIPS}_2=|\mathbf{p}^*|/4\pi\sqrt{s}$ and the formula $|\mathbf{p}^*|=\lambda^{1/2}(s,m_1^2,m_2^2)/2\sqrt{s}$ ; (4) the worked numbers $|\mathbf{p}^*|=206$ MeV and $E'/E=1.871$ at $\beta=0.6$ .

Prerequisites. Fermi's Golden Rule · Special-Relativistic Kinematics · Four-Vector Notation · Natural Units. Leads to. Particle Decay Rates · Interaction Cross Sections · $e^+e^-\to\mu^+\mu^-$ · Matrix Element and Spin Averaging.

Ask the tutor: Ask me to integrate three-body phase space on a Dalitz plot, or derive the collider flux factor F.

Why normalise to 2E particles per unit volume rather than just E?

The Lorentz contraction argument alone tells you the normalisation must scale with $E$ — one factor of energy cancels the $1/\gamma$ contraction of the box. The coefficient $2$ is fixed by consistency with the covariant form: writing the measure as a four-dimensional integral over $d^4p$ constrained to the mass shell by $\delta(p^2-m^2)$ and integrating out the energy component produces $1/2E$ , because the derivative of $(p^0)^2-E^2$ at the root $p^0=E$ is $2E$ . So the $2$ is not a free choice; it makes the three- and four-dimensional pictures agree. Choosing $E$ instead would still be invariant, but it would clash with the standard relativistic state normalisation and force compensating factors elsewhere.

Does the split between the measure and the matrix element have physical meaning?

No — only their product does. The $2E$ factors from wavefunction normalisation can be parked in the phase-space measure, absorbed into the amplitude, or split between them, and different textbooks make different choices: Peskin–Schroeder keep relativistically normalised states with a $1/2E$ in the measure, while some older treatments use a differently normalised amplitude. What is unambiguous and measurable is the combination $|\mathcal{M}|^2\,d\mathrm{LIPS}$ , from which every width and cross section is built. If you compare two calculations and find a stray factor of $2E$ , check whether they use the same convention for where those factors live — not whether one is "wrong".

What is the mass shell, and why does the measure live on it?

The mass shell is the surface in four-momentum space where a particle satisfies $E^2=\mathbf{p}^2+m^2$ , i.e. $p^2=m^2$ — the hyperboloid drawn in the second figure. A real, observable particle is always on shell: it has a definite mass, so its energy and momentum are locked together. The phase-space measure counts configurations of such real particles, so it is supported entirely on the forward ( $E>0$ ) mass shell of each one; the delta $\delta(p^2-m^2)$ in equation (2) enforces this, and $d^3p/2E$ is the invariant area element on the hyperboloid. Virtual particles inside a Feynman diagram are off shell — $p^2\neq m^2$ — but they are internal lines integrated over, not counted by LIPS, which only ever sees the physical external legs.

Why does the two-body rate depend only on a single momentum p*?

Because momentum conservation and the on-shell conditions use up almost all six momentum components of a two-particle final state. In the rest frame of the decaying particle the two daughters go back-to-back with equal and opposite momenta, so their directions are fixed to a single solid angle and their common magnitude is fixed by energy conservation to $|\mathbf{p}^*|=\lambda^{1/2}(M^2,m_1^2,m_2^2)/2M$ . Nothing else is free. The entire kinematic content of two-body phase space collapses to that one number times a solid angle — which is why equation (4) is so compact, and why all the richness of a two-body decay lives in $|\mathcal{M}|^2$ , never in the phase space.

How does the covariant measure reproduce time dilation of a lifetime?

Look at the prefactor of the decay rate: $\Gamma=(1/2E_a)\int|\mathcal{M}|^2 d\mathrm{LIPS}$ , where $E_a$ is the energy of the decaying particle. In its rest frame $E_a=M$ and you get the rest-frame width $\Gamma_0$ . In a frame where the particle moves with Lorentz factor $\gamma$ , its energy is $E_a=\gamma M$ , so the same invariant integral is divided by a larger number: $\Gamma=\Gamma_0/\gamma$ . A smaller rate means a longer lifetime, $\tau=\gamma\tau_0$ — exactly relativistic time dilation of the particle's clock. Because $d\mathrm{LIPS}$ and $|\mathcal{M}|^2$ are invariant, the only frame dependence of the rate sits in that $1/E_a$ , so time dilation drops out automatically rather than being inserted by hand.

Why is three-body phase space "flat" on a Dalitz plot?

For a three-body final state, in coordinates given by the two independent invariant-mass-squared combinations $m_{12}^2$ and $m_{23}^2$ , the element $d\mathrm{LIPS}_3$ is simply proportional to $dm_{12}^2\,dm_{23}^2$ — a constant density over the allowed region. Richard Dalitz noticed this in 1953: a scatter plot of decay events would be uniform if the matrix element were constant. Any departure — bands, holes, peaks — is therefore a direct image of $|\mathcal{M}|^2$ , showing resonances, interference, and CP-violating asymmetries. The Dalitz plot is powerful precisely because LIPS gives it a flat background against which dynamics stands out.

Lorentz-invariant phase space

Where this lives today

The physical picture

The core equations

Limiting cases

Worked examples

1 · Two-body momentum in a real decay (confidence builder)

2 · Checking that d³p/E is invariant under a boost (fully worked)

Physics of the portrait

Going further

Before this

Where this leads