Energy–Momentum Relation

A proton circulating in the LHC at a total energy of $6800\ \mathrm{GeV}$ carries a momentum of $pc = 6799.999935\ \mathrm{GeV}$ . Put those two numbers side by side: they agree to eleven significant figures. Subtract them and you get $E - pc = 6.47\times10^{-5}\ \mathrm{GeV}$ — sixty-five keV adrift in seven thousand GeV, a difference no calorimeter on Earth could resolve. Now subtract the squares instead. $E^2 - (pc)^2 = 0.880354\ \mathrm{GeV^2}$ , and its square root is $0.938272\ \mathrm{GeV}$ — the rest energy of the proton, recovered to six figures from two numbers that looked identical. Nothing was measured more precisely. The information was always there, hiding in a difference of squares — and the difference of squares is the one combination a Lorentz boost cannot touch. That is the whole content of the energy–momentum relation. Boost to another frame and $E$ changes, $\vec{p}$ changes, and they change by mixing into each other — but $E^2 - (pc)^2$ comes out the same number every time, and that number is $(mc^2)^2$ . The relation is not a formula tying three quantities together; it is the statement that a particle's four-momentum has a fixed Minkowski length, and that the length is the mass. Every energy a particle can have and every momentum it can carry is confined to a single curve in the $(pc, E)$ plane — the mass shell — and changing frames only slides it along that curve. It cannot get off.

Key takeaways

Signature $(+,-,-,-)$ throughout: $\eta_{\mu\nu} = \mathrm{diag}(+1,-1,-1,-1)$ , so for $p^\mu = (E/c, \vec{p}\,)$ the invariant norm is $p\cdot p = (E/c)^2 - |\vec{p}\,|^2 = +\,m^2c^2$ . Plus. Timelike — because a massive particle has a rest frame.
Multiply by $c^2$ and the norm wears its familiar dress: $\boxed{E^2 = (pc)^2 + (mc^2)^2}$ — Pythagoras, with $E$ the hypotenuse and $pc$ , $mc^2$ the legs.
$m$ is the invariant rest mass, a Lorentz scalar. It does not grow with speed. What grows is $E = \gamma mc^2$ and $p = \gamma m v$ ; their difference of squares stays pinned.
Velocity is not an independent input but a ratio: $\vec{v} = \vec{p}\,c^2/E$ , so $\beta = pc/E$ and $\gamma = E/mc^2$ . Energy and momentum already know how fast the particle is going.
Owned result: the slope of the mass shell IS the particle's velocity. Differentiating the relation gives $\mathrm{d}E/\mathrm{d}(pc) = pc/E = \beta$ . The asymptote has slope $1$ , and a hyperbola never meets its asymptote — so the speed limit is a geometric fact about a curve, not a prohibition bolted on afterwards.
Key idea: the two famous limits are not two formulas but the two ends of one hyperbola. Near the vertex it is a parabola and you read off $E \approx mc^2 + p^2/2m$ — Newton. Far out it is a straight line and you read off $E \approx pc$ — light. The crossover sits at $pc \approx 7.07\,mc^2$ , where $E \approx pc$ first becomes true to one per cent.

The core insight

Fix the mass and you have fixed a curve, not a point: the set of all $(pc, E)$ a particle of that mass is permitted to have. Every inertial observer finds the particle somewhere on that curve, and a boost merely slides it along. This is why $m$ is the honest label of a particle and $E$ is not — $m$ names the curve, $E$ names a point on it, and which point is a matter of who is looking. A proton at rest on your desk and a proton in the LHC beam are the same particle on the same mass shell.

Where this lives today

Every particle detector ever built is a machine for locating events on mass shells. A tracking chamber in a magnetic field measures $\vec{p}$ from the curvature of a trajectory; a calorimeter measures $E$ by stopping the particle and counting the deposited energy. Neither number identifies the particle. The combination $\sqrt{E^2 - (pc)^2}$ does — and because it is invariant, it does so without anyone needing to know what frame the detector was in, or how fast the particle was going. Particle identification is, in its bones, the evaluation of a Minkowski norm. It is also what makes "GeV $/c^2$ " respectable rather than affected. Because all three sides of the triangle are energies, the arithmetic of a collision becomes ordinary addition and subtraction of numbers of comparable size — which is why the invariant mass of a pair of decay products can be histogrammed straight from detector output, and why a bump in that histogram is a particle. Beyond the laboratory, the relation is the dispersion relation of relativistic quantum theory, and its two limits organise whole subjects. Substituting operators into it gives Klein–Gordon; factorising it gives Dirac. Its ultra-relativistic end is why a white dwarf and a neutron star's crust obey different equations of state: as a degenerate electron gas is compressed it crosses from $E \approx p^2/2m$ to $E \approx pc$ , the pressure's dependence on density softens, and the Chandrasekhar limit follows. A star's fate is set by which end of this hyperbola its electrons sit on.

The physical picture

Draw the plane whose horizontal axis is $pc$ and whose vertical axis is $E$ . Both axes now carry energy units — the first and most useful piece of hygiene in relativistic kinematics. Never plot $p$ against $E$ ; plot $pc$ against $E$ , and the geometry becomes visible. In this plane $E^2 - (pc)^2 = (mc^2)^2$ is a hyperbola, opening upward, with vertex at $(0,\ mc^2)$ and asymptotes $E = \pm\,pc$ . The vertex is the rest frame: no momentum, energy $mc^2$ . The asymptotes are the light cone of momentum space, the locus $E = pc$ where a massless particle lives. A massive particle's curve hugs those asymptotes without ever touching them, and that gap is the mass. This is the invariant hyperbola of spacetime intervals redrawn in momentum space, and the correspondence is exact. There, $c^2t^2 - x^2 = s^2$ was invariant and a boost slid an event along a curve of fixed interval; here $(E/c)^2 - p^2 = m^2c^2$ is invariant and a boost slides a particle along a curve of fixed mass. Same group, same geometry, same reason: a Lorentz transformation is a hyperbolic rotation, and hyperbolic rotations preserve hyperbolae.

Figure

PUNCHLINE — the mass shell, and the fact that its SLOPE is the particle's velocity. Every state of a mass-m particle lies on the hyperbola E² − (pc)² = (mc²)². The tangent at any point has slope dE/d(pc) = pc/E = β. At the vertex it is flat (β = 0, at rest); far out it approaches the asymptote of slope 1 (β → 1, never reached). The proton marks are real: at rest 0.938 GeV, and the LHC beam at 6800 GeV where β = 0.99999999

Read the punchline off the figure. The tangent to the mass shell has slope $\beta$ — flat at the vertex, tending to $1$ far out. Since the hyperbola lies strictly above its asymptote whenever $m > 0$ , the slope is strictly below one, everywhere, forever. The cosmic speed limit is not an extra postulate; it is the statement that a hyperbola does not cross its own asymptote. The figure also shows why the two textbook limits look so different despite being one curve. Near the vertex the hyperbola is locally a parabola, $E \approx mc^2 + p^2/2m$ — the Newtonian dispersion relation. Far out it is locally a straight line of slope one, $E \approx pc$ — the photon's. Newtonian mechanics and light are the two asymptotic ends of one conic section, and the mass is the parameter that keeps them apart.

The core equations

The starting point is not $E^2 = (pc)^2 + (mc^2)^2$ . It is the four-momentum and its norm; the relation is what appears when you write the norm out in components.

p^\mu = m\,U^\mu = \left(\frac{E}{c},\ \vec{p}\,\right), \qquad p\cdot p \;\equiv\; \eta_{\mu\nu}p^\mu p^\nu = \left(\frac{E}{c}\right)^{\!2} - |\vec{p}\,|^2 \;=\; +\,m^2c^2 \qquad (1)

where: $U^\mu = \gamma(c,\vec{v}\,)$ — the four-velocity, of fixed length $U\cdot U = +c^2$ ; $m$ — the invariant rest mass, a Lorentz scalar. Physical reading: $p^\mu$ is a four-vector because it is a scalar times a four-vector, so its full contraction is the same number in every frame. Evaluate that number where the particle sits still — $E = mc^2$ , $\vec{p} = \vec{0}$ — and it is $m^2c^2$ ; being invariant, it is $m^2c^2$ everywhere else too. The sign is $+$ because a massive particle's four-momentum is timelike. In $(-,+,+,+)$ you would write $p\cdot p = -m^2c^2$ ; every intermediate sign here would flip and no physics would change. Multiply (1) by $c^2$ :

\boxed{\,E^2 = (pc)^2 + (mc^2)^2\,} \qquad\Longleftrightarrow\qquad mc^2 = \sqrt{E^2 - (pc)^2} \qquad (2)

where: $p \equiv |\vec{p}\,|$ . Physical reading: a right triangle with legs $pc$ , $mc^2$ and hypotenuse $E$ . All three sides are energies — which is why masses are quoted in eV and momenta in eV $/c$ . Two corollaries follow at once: $E \ge mc^2$ always, with equality only at rest; and $E > pc$ strictly for any $m > 0$ , so $E = pc$ is a statement about massless particles and nothing else. Velocity is not independent data. It is already encoded in the pair $(E, \vec{p}\,)$ :

\vec{v} = \frac{\vec{p}\,c^2}{E}, \qquad\text{equivalently}\qquad \beta = \frac{pc}{E}, \qquad \gamma = \frac{E}{mc^2} \qquad (3)

where: $\beta = v/c$ , $\gamma = (1-\beta^2)^{-1/2}$ . Physical reading: divide $\vec{p} = \gamma m\vec{v}$ by $E = \gamma mc^2$ and the $\gamma m$ cancels — the ratio is $\vec{v}/c^2$ regardless of the mass, which is exactly why it survives the massless limit. A detector that measures energy and momentum has already measured the velocity; Massless Particles and the Photon leans on this line to prove light moves at $c$ for any momentum whatever. Finally the two limits, which are the two ends of the same curve:

E \simeq mc^2 + \frac{p^2}{2m} - \frac{p^4}{8m^3c^2} + \cdots \;\; (pc \ll mc^2), \qquad E \simeq pc + \frac{(mc^2)^2}{2pc} + \cdots \;\; (pc \gg mc^2) \qquad (4)

where: both are binomial expansions of the same square root, about opposite ends. Physical reading: the left one contains Newton — $p^2/2m$ is the kinetic energy, sitting on a constant pedestal $mc^2$ that Newtonian mechanics never sees because constants exert no force. The right one says a fast massive particle is a photon plus a correction suppressed by the square of $mc^2/pc$ — which is why an ultra-relativistic electron and a photon of equal momentum are so hard to tell apart.

The mass shell

Derivation AThe relation is the invariant norm of the four-momentum, evaluated in the rest frame

The cleanest route to $E^2 = (pc)^2 + (mc^2)^2$ does not pass through $E = \gamma mc^2$ and $\vec p = \gamma m \vec v$ at all. It uses one strategic move worth more than the result itself: when you need the value of a Lorentz invariant, compute it in whichever frame makes it trivial, then assert it everywhere. The four-momentum's norm is an invariant; there is a frame in which the particle is at rest; in that frame the norm is embarrassingly easy. Every threshold calculation in particle physics is this trick applied twice.

Step 1. Build the object. $U^\mu = \mathrm{d}x^\mu/\mathrm{d}\tau = \gamma(c, \vec{v}\,)$ is a four-vector, being the coordinate displacement divided by the scalar proper time. Multiply by the scalar $m$ : $p^\mu \equiv mU^\mu = (\gamma mc,\ \gamma m\vec{v}\,) = (E/c,\ \vec{p}\,)$ , recalling $E = \gamma mc^2$ and $\vec{p} = \gamma m\vec{v}$ from Relativistic Energy and Relativistic Momentum. Why: a four-vector times a scalar is a four-vector, so we know before computing anything that $p\cdot p$ is frame-independent. That guarantee is the whole point.

Step 2. Contract it with itself in an arbitrary frame. With $\eta = \mathrm{diag}(+1,-1,-1,-1)$ , $\;p\cdot p = (p^0)^2 - |\vec{p}\,|^2 = (E/c)^2 - |\vec{p}\,|^2$ . Why: this is the only invariant constructible from one four-vector, so whatever it equals is a property of the particle alone — not of the observer.

Step 3. Evaluate it where the work is zero. In the rest frame $\vec{v} = \vec{0}$ , so $\gamma = 1$ , $\vec{p} = \vec{0}$ , $E = mc^2$ , and $p\cdot p = (mc^2/c)^2 - 0 = m^2c^2$ . Why: this step is what $m$ means. The invariant norm equals $m^2c^2$ by construction of the rest frame, and the relation is that fact wearing components.

Step 4. Assert it everywhere, clear the units. Because $p\cdot p$ is a scalar, $(E/c)^2 - |\vec{p}\,|^2 = m^2c^2$ in every frame; multiply by $c^2$ to get $E^2 - (pc)^2 = (mc^2)^2$ . Why: note precisely what is claimed. $E$ and $\vec p$ are frame-dependent and observers will disagree about them. Their combination is not. The relation is a constraint tying two frame-dependent things to one frame-independent thing.

Step 5. Cross-check against the velocity forms. $E^2 - (pc)^2 = \gamma^2m^2c^4 - \gamma^2m^2v^2c^2 = \gamma^2m^2c^4(1 - \beta^2) = m^2c^4$ , since $\gamma^2(1-\beta^2) = 1$ identically. Why: two independent derivations — one geometric, one algebraic — landing on the same equation. The $\gamma$ 's cancel, which is the algebraic shadow of the geometric statement that a boost slid the particle along the hyperbola without changing which hyperbola it was on. *Verified in verify_aq-sr-energy-momentum-relation.py: $p\cdot p = +m^2c^2$ symbolically with $\eta = \mathrm{diag}(+1,-1,-1,-1)$ ; invariance of $p\cdot p$ under an arbitrary boost; and $\gamma^2m^2c^4 - \gamma^2m^2v^2c^2 - m^2c^4 = 0$ identically.*

Go deeper · Derivation B · 2 minThe slope of the mass shell is the particle's velocity, and the asymptote is the speed limit

Derivation BThe slope of the mass shell is the particle's velocity, and the asymptote is the speed limit

This is the article's owned structural result, and it turns a formula into a picture you cannot afterwards un-see. The mass shell is a curve; curves have tangents. The claim is that the tangent's slope at any point is not an auxiliary quantity but the particle's own $\beta$ — so the geometry of the hyperbola and the kinematics of the particle are the same information, and $\beta < 1$ becomes the statement that a hyperbola stays on one side of its asymptote.

Step 1. Differentiate the constraint. Hold $m$ fixed — we move along one particle's curve, not between particles — and differentiate $E^2 = (pc)^2 + (mc^2)^2$ with respect to $pc$ : $\;2E\,\mathrm{d}E/\mathrm{d}(pc) = 2(pc)$ . Why: $m$ is constant along the shell by definition of the shell, so $(mc^2)^2$ contributes nothing. This is the only place the invariance of $m$ is used, and it does all the work.

Step 2. Read off the slope. $\;\mathrm{d}E/\mathrm{d}(pc) = pc/E$ . But equation (3) says $\beta = pc/E$ . Hence $\boxed{\mathrm{d}E/\mathrm{d}(pc) = \beta}$ . Why: no new physics entered. The slope is $pc/E$ by pure calculus; $pc/E$ is $\beta$ by pure kinematics. The content is the coincidence — and the coincidence is that both are shadows of $p^\mu = mU^\mu$ .

Step 3. Locate the endpoints. At the vertex $p = 0$ : slope $0$ , particle at rest. As $pc \to \infty$ the curve approaches its asymptote $E = pc$ , of slope exactly $1$ , so $\beta \to 1$ . Why: and it approaches from above, since $E = \sqrt{(pc)^2 + (mc^2)^2} > pc$ for every finite $pc$ when $m > 0$ . Above the asymptote means slope below one.

Step 4. Conclude the speed limit geometrically. $E > pc$ strictly for $m>0$ , so $\beta = pc/E < 1$ strictly, at every point, for every energy however large. A massive particle cannot reach $c$ because a hyperbola cannot reach its asymptote. Why: this inverts the usual pedagogy. The standard argument says "it takes infinite energy, so you cannot get there." The geometric argument says the light line is simply not on the curve — there is no state of a massive particle with $\beta = 1$ to get to, at any energy at all. Infinite energy is the symptom; the missing point is the disease.

Step 5. Put numbers on it. At $pc = 2mc^2$ : $E = \sqrt{4+1}\,mc^2 = \sqrt5\,mc^2 = 2.236\,mc^2$ , so $\beta = 2/\sqrt5 = 0.8944$ and $\gamma = \sqrt5 = 2.236$ . For the LHC proton, $E = 6800\ \mathrm{GeV}$ and $mc^2 = 0.938272\ \mathrm{GeV}$ give $\gamma = E/mc^2 = 7247.37$ and $\beta = \sqrt{1-\gamma^{-2}} = 0.999999990481$ , so $1 - \beta = 9.52\times10^{-9}$ . Why: the second number shows the asymptote being approached and not reached with brutal literalness — the proton is nine parts in a billion short of $c$ , and the LHC's entire magnet system exists to hold it on that curve. *Verified in verify_aq-sr-energy-momentum-relation.py: $\mathrm{d}E/\mathrm{d}(pc) - pc/E = 0$ symbolically on the shell; $\beta = pc/E$ from $E = \gamma mc^2$ , $p = \gamma mv$ ; $E > pc$ for all $p$ when $m > 0$ ; and the printed numbers $\sqrt5 = 2.236$ , $\beta = 0.8944$ , $\gamma = 7247.37$ , $1-\beta = 9.52\times10^{-9}$ .*

Go deeper · Derivation C · 2 minThe two limits are the two ends of one hyperbola, and they cross over at seven times the rest energy

Derivation CThe two limits are the two ends of one hyperbola, and they cross over at seven times the rest energy

Textbooks present $E \approx mc^2 + p^2/2m$ and $E \approx pc$ as two separate results for two separate regimes. They are one square root expanded about two ends. Doing both in one place makes the structure obvious, tells you the size of the error you are making, and — the part nobody states — tells you where the crossover actually is, which is a number you can use.

Step 1. Factor out the dominant term at the slow end: $E = mc^2\sqrt{1 + (p/mc)^2}$ . With $x = (p/mc)^2 \ll 1$ the binomial series gives $\sqrt{1+x} = 1 + \tfrac12 x - \tfrac18 x^2 + \cdots$ . Why: the expansion parameter is $p/mc = \gamma\beta$ — so "non-relativistic" means $\gamma\beta \ll 1$ , and it is $\gamma\beta$ , not $\beta$ , that must be small. At $\beta = 0.9$ , $\gamma\beta = 2.06$ and the expansion is already useless.

Step 2. Collect: $E = mc^2 + \dfrac{p^2}{2m} - \dfrac{p^4}{8m^3c^2} + \cdots$ — a constant pedestal, Newton's kinetic energy, then the leading relativistic correction. Why: Newtonian mechanics is the second term. The first is invisible to it because a constant energy exerts no force; the third because it is $O(\beta^2)$ relative to the second. Relativistic Kinetic Energy takes $K = E - mc^2$ from here.

Step 3. Now factor out the dominant term at the fast end instead: $E = pc\sqrt{1 + (mc^2/pc)^2}$ , and with $y = (mc^2/pc)^2 \ll 1$ the same series gives $E = pc + \dfrac{(mc^2)^2}{2pc} - \cdots$ . Why: identical algebra, opposite end. The mass has demoted itself from leading term to correction — the precise sense in which a very fast particle "behaves like light".

Step 4. Find the crossover. Demand the correction be one per cent of the leading term: $\dfrac{E}{pc} - 1 \approx \dfrac{1}{2}\left(\dfrac{mc^2}{pc}\right)^2 = 0.01 \;\Rightarrow\; \dfrac{pc}{mc^2} = \dfrac{1}{\sqrt{0.02}} = 7.071$ . Why: $E = pc$ is good to $1\%$ once the momentum exceeds about seven times the rest energy, i.e. $\gamma \gtrsim 7.1$ — which is why a 50 GeV electron ( $\gamma \approx 10^5$ ) is treated as massless without apology.

Step 5. Check the number honestly. At $pc = 7.071\,mc^2$ exactly: $E = \sqrt{7.071^2 + 1}\;mc^2 = 7.1414\,mc^2$ , so $E/pc = 1.00995$ — a $0.995\%$ error, confirming the estimate. Why: the estimate kept only the first correction, so landing within half a per cent of its own target is the sign that the neglected $y^2$ term really was negligible. *Verified in verify_aq-sr-energy-momentum-relation.py: both binomial series against a sympy expansion; the coefficients $1, \tfrac12, -\tfrac18$ ; the crossover $pc/mc^2 = 1/\sqrt{0.02} = 7.0711$ ; and the exact check $E/pc - 1 = 0.00995$ there.*

Figure

The relation is Pythagoras in energy units. Legs pc and mc², hypotenuse E — and the two limits are the two ways the triangle can degenerate. LEFT the 3–4–5 case: pc = 4 GeV, mc² = 3 GeV, E = 5 GeV, so β = 4/5 and γ = 5/3 are just side ratios. CENTRE the slow limit, where the pc leg shrinks and E → mc². RIGHT the fast limit, where the mc² leg becomes negligible and E → pc

Limiting cases

Case	Result	Meaning
$p = 0$	$E = mc^2$	the vertex of the shell — rest energy, and the definition of $m$
$m = 0$	$E = pc$ , $p\cdot p = 0$	the shell degenerates onto its own asymptote: a null four-momentum, $\beta = 1$
$pc \ll mc^2$	$E \approx mc^2 + \dfrac{p^2}{2m}$	Newton, sitting on a pedestal he could not see
$pc \gg mc^2$	$E \approx pc + \dfrac{(mc^2)^2}{2pc}$	ultra-relativistic: a photon plus a quadratically small correction
$pc = 7.071\,mc^2$	$E/pc = 1.00995$	the crossover: $E \approx pc$ becomes true to $1\%$
$c \to \infty$	$E - mc^2 \to p^2/2m$	every relativistic correction carries $1/c^2$ and dies
any $E, p$	$\beta = pc/E$ , $\;\gamma = E/mc^2$	velocity is a ratio, never an independent input

Common trapE = pc is not a shortcut for fast particles. It is a statement that the mass is zero.

The most common wrong move in relativistic kinematics is to reach for $E = pc$ because a particle is fast. Speed does not entitle you to it. It is an exact identity for $m = 0$ and a controlled approximation for $m > 0$ whose error you must be able to name. From equation (2), $E > pc$ strictly whenever $m > 0$ — the hypotenuse of a right triangle is longer than either leg, always. The fractional error is $\tfrac12(mc^2/pc)^2$ : one per cent at $pc = 7.07\,mc^2$ , and one part in $10^{6}$ only by $pc \approx 707\,mc^2$ . For a 50 GeV electron that is fine to parts per billion; for a 5 GeV proton it is a $1.8\%$ error that will wreck a threshold calculation. Compute $\gamma$ ; do not eyeball "fast".

Common trapThe mass does not increase. The mass names the curve.

Nothing on this page grows with speed except $E$ and $p$ . The symbol $m$ is the invariant rest mass — a Lorentz scalar, the same number for every observer, defined by $p\cdot p = m^2c^2$ . There is no $m_0$ here because there is no other $m$ to distinguish it from. The old device of writing $E = m_{\text{rel}}c^2$ with $m_{\text{rel}} = \gamma m$ makes $E = mc^2$ true by definition and thereby says nothing; worse, it tempts you into $\vec F = m_{\text{rel}}\vec a$ , which is false — Relativistic Force shows the response coefficients are $\gamma^3m$ and $\gamma m$ , direction-dependent, so no single scalar can absorb them. Lev Okun's 1989 Physics Today essay is the standard demolition. One real caveat: a composite system's invariant mass is not the sum of its constituents' masses, and it genuinely does change in an inelastic collision. That is binding energy — see Conservation of Four-Momentum — not relativistic mass.

Worked examples

1 · Recover a proton's mass from an LHC beam, where E and pc look identical

The LHC beam carries protons at total energy $E = 6800\ \mathrm{GeV}$ ; take $m_pc^2 = 0.938272\ \mathrm{GeV}$ . Find $pc$ , $\beta$ , $\gamma$ , then recover $mc^2$ from $E$ and $pc$ — and see why the difference of squares works where the plain difference fails.

Step 1 — identify. Two sides of the right triangle are known: the hypotenuse $E$ and the leg $mc^2$ . Everything else follows from (2) and (3). No dynamics, no fields — pure kinematics on the mass shell.

Step 2 — apply. $\gamma = E/mc^2 = 6800/0.938272 = \mathbf{7247.37}$ . Then $\beta = \sqrt{1-\gamma^{-2}} = \mathbf{0.999999990481}$ , so $1-\beta = 9.52\times10^{-9}$ . And $pc = \beta E = \sqrt{E^2 - (mc^2)^2} = \mathbf{6799.999935\ \mathrm{GeV}}$ .

Step 3 — the naive subtraction fails. $E - pc = 6800 - 6799.999935 = \mathbf{6.47\times10^{-5}\ \mathrm{GeV}} = 64.7\ \mathrm{keV}$ . The number is real, but it is the difference of two eleven-figure quantities: extracting it would need $E$ and $p$ each measured to a part in $10^{11}$ . No calorimeter comes within nine orders of magnitude of that.

Step 4 — the difference of squares succeeds. $E^2 - (pc)^2 = (E-pc)(E+pc) = (6.47\times10^{-5})(13599.99994) = \mathbf{0.880354\ \mathrm{GeV^2}}$ , and $\sqrt{0.880354} = \mathbf{0.938272\ \mathrm{GeV}}$ — the proton, to six figures. The factorisation is the trick: the tiny difference is multiplied by the huge sum, and the product has ordinary size.

Step 5 — the boundary number. What would a $1\ \mathrm{GeV}$ error in the beam energy do? It shifts $\gamma$ by $0.015\%$ and shifts $\beta$ in the twelfth decimal place — but leaves $mc^2 = \sqrt{E^2-(pc)^2}$ exactly unchanged, because $E$ and $pc$ move together along the shell. That insensitivity is not luck; it is invariance. It is why a collider reports $\sqrt{E^2 - (pc)^2}$ and not $E$ .

2 · The 3–4–5 triangle, and reading a particle's identity off two numbers

A tracker measures $p = 4\ \mathrm{GeV}/c$ ; a calorimeter measures $E = 5\ \mathrm{GeV}$ . Identify the mass, speed and Lorentz factor — then decide whether treating the particle as massless would have been acceptable.

Step 1 — identify. The right triangle of (2), with the hypotenuse and one leg given. Both measured quantities are already energies: $pc = 4\ \mathrm{GeV}$ , $E = 5\ \mathrm{GeV}$ .

Step 2 — apply. $mc^2 = \sqrt{E^2 - (pc)^2} = \sqrt{25-16} = \sqrt9 = \mathbf{3\ \mathrm{GeV}}$ . The 3–4–5 triangle, in momentum space: legs $3$ and $4$ , hypotenuse $5$ .

Step 3 — read off the kinematics. $\beta = pc/E = 4/5 = \mathbf{0.8}$ and $\gamma = E/mc^2 = 5/3 = \mathbf{1.667}$ . Cross-check: $\gamma = 1/\sqrt{1-0.64} = 1/0.6 = 5/3$ ✓. Note $\gamma\beta = pc/mc^2 = 4/3$ — so $\beta$ , $\gamma$ and $\gamma\beta$ are all just side-ratios. The triangle is the kinematics.

Step 4 — would "massless" have been safe? Not remotely. Here $pc/mc^2 = 1.333$ , far below the $7.07$ crossover. Setting $E = pc$ would claim $E = 4\ \mathrm{GeV}$ when the truth is $5$ — a $\mathbf{20\%}$ error. The rule earns its keep: the massless approximation is a statement about $\gamma$ , and $\gamma = 1.67$ is not "fast".

Step 5 — the boundary number. How fast must this same $3\ \mathrm{GeV}$ particle be for $E \approx pc$ to hold to $1\%$ ? By the crossover, $pc \ge 7.071 \times 3 = \mathbf{21.2\ \mathrm{GeV}/c}$ , hence $E \ge 21.4\ \mathrm{GeV}$ and $\gamma \ge 7.14$ . Below that its mass is not a correction — it is a leg of the triangle.

Physics of the portrait

The mass shell is one of the few objects in physics named for what it forbids. It is the constraint surface on which a real, freely propagating particle must lie, and off which nothing observable is found.

Origins · 2 minMass shell and how this was discovered

Mass shell

1908–present · Minkowski's momentum space · the Klein–Gordon operator · every Feynman diagram ever drawn

The surface $p^\mu p_\mu = m^2c^2$ enters physics with Minkowski's four-dimensional reformulation of 1908, in which energy and momentum are welded into one four-vector whose invariant length is the mass. In that language $E^2 = (pc)^2 + (mc^2)^2$ is not a law to be derived but a definition unpacked: the mass shell is the set of four-momenta of length $mc$ , a two-sheeted hyperboloid whose upper sheet holds the particles of positive energy. The name earns its keep in quantum field theory. Substituting $E \to i\hbar\,\partial_t$ and $\vec{p} \to -i\hbar\nabla$ turns the relation into the Klein–Gordon operator, whose plane-wave solutions are exactly the on-shell momenta; a wave that fails the mass-shell condition is not a solution and cannot propagate to a detector. Inside a Feynman diagram, by contrast, internal lines carry four-momenta that conserve at every vertex but violate $p\cdot p = m^2c^2$ — they are virtual, off shell, and the further off shell they are the shorter they live. So the hyperbola of this article draws the boundary between what is real and what is merely intermediate, and "on shell" and "off shell" are heard in every corridor at CERN.

History

1905

— Einstein's September sequel to the relativity paper, "Ist die Trägheit eines Körpers von seinem Energieinhalt abhängig?", obtains the rest energy of a body as E = mc² — the vertex of the shell, long before anyone drew the curve.

1906

— Planck founds relativistic dynamics by keeping Newton's law as F = dp/dt with p = γmv, fixing the momentum leg of the triangle.

1908

— Minkowski's Cologne lecture "Raum und Zeit" makes spacetime a geometry; energy and momentum become one four-vector whose invariant length is the mass.

1909

— Lewis and Tolman rebuild relativistic mechanics from conservation principles rather than from electrodynamics (Phil. Mag. 18, 510), reaching the same energy and momentum by a route Newton would have recognised.

1926

— Klein and Gordon substitute quantum operators into E² = (pc)² + (mc²)², and the relation becomes a wave equation; the mass shell becomes the condition for a solution to exist.

1928

— Dirac takes the square root of the same relation instead of using it as it stands, and spin and antimatter fall out of the factorisation.

Going further

Texts. Rindler, Introduction to Special Relativity, 2nd ed., §§35–38 · Griffiths, Introduction to Elementary Particles, 2nd ed., §3.2 · Taylor & Wheeler, Spacetime Physics, 2nd ed., ch. 7 · Morin, Introduction to Classical Mechanics, ch. 12 · Hartle, Gravity, ch. 5 · Okun, "The concept of mass", Physics Today 42(6), 31 (1989).

Provenance. Signature $(+,-,-,-)$ , so $p\cdot p = +m^2c^2$ and the mass shell is timelike. Every equation and number above is recomputed in verify_aq-sr-energy-momentum-relation.py, which parses this article's own printed digits and compares them against an independent recomputation: the invariant norm and its invariance under a boost; $\beta = pc/E$ , $\gamma = E/mc^2$ ; the owned result $\mathrm{d}E/\mathrm{d}(pc) = \beta$ and the strict inequality $E > pc$ for $m > 0$ ; both binomial expansions and the crossover $pc/mc^2 = 7.071$ ; and the worked numbers $\gamma = 7247.37$ , $\beta = 0.999999990481$ , $pc = 6799.999935$ GeV, $E-pc = 6.47\times10^{-5}$ GeV, $E^2-(pc)^2 = 0.880354\ \mathrm{GeV^2}$ , $mc^2 = 0.938272$ GeV, and the 3–4–5 triangle.

Scope. The massless case is stated here and developed in Massless Particles and the Photon, which owns $E = pc$ , the null four-momentum and the photon's quantum relations. Threshold energies, the centre-of-momentum frame and the Mandelstam invariant $s$ are not derived here — they belong to Threshold Energy, The Centre-of-Momentum Frame and Mandelstam Invariants. $E = \gamma mc^2$ and $\vec p = \gamma m\vec v$ are recalled from Relativistic Energy and Relativistic Momentum, not re-derived.

Builds on. Four-Momentum · Relativistic Energy · Relativistic Momentum · Invariant Mass. Leads to. Massless Particles and the Photon · Conservation of Four-Momentum · Relativistic Kinetic Energy · Mass–Energy Equivalence.

Is this a definition of m, or a law of physics?

Both. Read left to right with $E$ and $\vec p$ given, it defines the invariant mass as $mc^2 = \sqrt{E^2 - (pc)^2}$ — the Minkowski length of the four-momentum, computable from any frame's measurements. Read as a constraint, it is a genuine law: a free particle's energy and momentum are not independent, only a one-parameter family of $(E, \vec p\,)$ is realisable, and the parameter is $|\vec p\,|$ . The empirical content is the constraint. You could imagine a universe in which a particle could be found anywhere in the $(pc, E)$ plane; in ours it is confined to a curve, and it is the same curve in every frame.

Why is the sign +m²c² and not −m²c², and how do I tell which convention a book is using?

The sign is a convention; the physics is not. In $(+,-,-,-)$ — used here, and by Rindler and most of particle physics — timelike vectors have positive norm, so $p\cdot p = +m^2c^2$ and $U\cdot U = +c^2$ . In $(-,+,+,+)$ — Misner–Thorne–Wheeler and much of the relativity literature — all of those flip. The reliable tell is to find a formula whose answer you already know: if a text writes $U\cdot U = +c^2$ , or says a timelike interval has $s^2 > 0$ , it is in our convention; if it writes $\mathrm{d}s^2 = -c^2\mathrm{d}t^2 + \mathrm{d}x^2$ , it is in the other. The observable form $E^2 = (pc)^2 + (mc^2)^2$ is identical in both — which is exactly why you can copy it safely and copy $p\cdot p = \pm m^2c^2$ fatally.

If a fast particle has more energy, does it have more mass?

No. It has more energy, and $m$ is untouched — that is what "invariant" means. The temptation comes from writing $E = \gamma mc^2$ and sliding the $\gamma$ onto the $m$ , a re-labelling that buys nothing and costs a great deal: it makes $\vec F = m\vec a$ look plausible when it is false, it makes $m$ frame-dependent when its entire value is that it is not, and it hides the fact that $\sqrt{E^2-(pc)^2}$ is what a detector reconstructs. There is exactly one place a mass legitimately grows, and it is not this one: the invariant mass of a composite system exceeds the sum of its parts' masses whenever those parts carry kinetic energy in the system's rest frame. Two protons that collide and stick really do form an object of invariant mass greater than $2m_p$ . That is binding-energy accounting — see Conservation of Four-Momentum.

The relation has E² on the left. Are negative-energy solutions allowed?

Algebraically, yes — $E = \pm\sqrt{(pc)^2 + (mc^2)^2}$ , so the mass shell is a two-sheeted hyperboloid, upper sheet at $E \ge +mc^2$ and lower at $E \le -mc^2$ . Classically the lower sheet is discarded by fiat: a free particle cannot cross the gap of width $2mc^2$ , so positive energy stays positive. In quantum field theory the gap can be crossed, by pair creation, and the lower sheet cannot be thrown away. Dirac's response in 1930 was the hole theory; the modern reading is that those solutions describe antiparticles of positive energy moving forward in time. The second sheet is not an artefact — it is where the positron lives.

Can I use E = pc for a very fast proton at, say, 5 GeV?

That is exactly the trap. "Very fast" is not the criterion; $\gamma$ is. A 5 GeV proton has $\gamma = 5/0.938 = 5.33$ and $pc = \sqrt{25 - 0.880} = 4.911\ \mathrm{GeV}$ , so $E = pc$ would be a $1.8\%$ error — enough to move a threshold calculation outside its error bars. Contrast a 5 GeV electron: $\gamma = 9785$ , and $E = pc$ holds to five parts in $10^{9}$ . Same energy, same casual "fastness", utterly different licences. Compute $pc/mc^2$ and compare it to $7.07$ ; that is the only test.

Why plot pc rather than p on the horizontal axis?

So the geometry is true rather than merely suggestive. With $pc$ and $E$ both in energy units the relation is literally Pythagoras: the asymptote has slope exactly $1$ , the tangent's slope is exactly $\beta$ , and distances on the plot mean something. Plot $p$ against $E$ and the asymptote acquires a slope of $c$ , the light line points almost straight up, and the two legs cannot be compared at all. Same hygiene that puts $ct$ rather than $t$ on a spacetime diagram, and for the same reason: a Minkowski norm can only be seen when its terms carry the same units.