Massless Particles and the Photon

Hand a physicist an electron and tell her its energy. She can tell you its speed: $\beta = pc/E$ , and the two are locked together so tightly that doubling the energy of a slow electron measurably changes how fast it moves. Hand her a photon and tell her its energy. She can tell you its colour — and nothing whatever about its speed. A $2.00\ \mathrm{eV}$ photon of orange light and a $1\ \mathrm{GeV}$ gamma ray from a pulsar differ in energy by a factor of five hundred million. They travel at exactly the same speed. Not approximately, not to within measurement error — exactly, and in every inertial frame, by the same argument. The energy of a photon has been completely severed from its velocity, and the severing is the whole content of $m = 0$ . You can see why in one line, and the line is not a new law but an old one used carefully. For any particle, massive or not, $\vec{v} = \vec{p}\,c^2/E$ . Put $E = pc$ into it and the magnitudes cancel: $|\vec{v}| = pc^2/(pc) = c$ , for every value of $p$ whatsoever. The momentum has dropped out. That degeneracy — one speed, all energies — is what it means for a four-momentum to be null: to lie exactly on the light cone rather than inside it. And because $p\cdot p = 0$ is a Lorentz-invariant statement, no observer, however fast, can disagree.

Key takeaways

Signature $(+,-,-,-)$ : $p\cdot p = (E/c)^2 - |\vec{p}\,|^2 = +\,m^2c^2$ . Setting $m = 0$ makes the four-momentum null: $\boxed{p\cdot p = 0}$ — it sits on the light cone, not inside it.
Equivalently $\boxed{E = pc}$ . This is exact for $m = 0$ and false for every $m > 0$ , where $E > pc$ strictly.
Speed is degenerate. From $\vec v = \vec p\,c^2/E$ , a massless particle moves at $c$ for any momentum at all. Energy tells you the colour, never the speed.
No rest frame, no proper time. $p\cdot p = 0$ is invariant, so no boost can bring a photon to rest; and $\mathrm{d}\tau = 0$ on a null worldline, so a photon's four-velocity does not exist. Use $p^\mu$ , never $U^\mu$ .
Quantum labels: $E = h\nu = \hbar\omega$ , $p = h/\lambda = \hbar k$ , so $p^\mu = (E/c)(1, \hat{n}) = \hbar(\omega/c, \vec{k}\,)$ — null because $\omega = ck$ .
Owned result: for any two future-pointing four-momenta, $p_1\cdot p_2 \ge m_1m_2c^2$ , with equality only if they are parallel. A timelike vector and a null one can never be parallel — so $p_e\cdot p_\gamma > 0$ strictly, and two no-go theorems fall out at once.
Key idea: a free electron cannot absorb a photon, and a photon cannot decay into an electron–positron pair in vacuum. Both are forbidden by nothing but the nullity of $p_\gamma$ — you cannot build a timelike vector out of a null one, and no amount of energy will fix that. This is why the photoelectric effect needs a bound electron and why pair production needs a nucleus.

The core insight

"Massless" is not "very light". It is a statement that the four-momentum lies exactly on the light cone, and being on the cone is qualitatively different from being anywhere inside it — the way being on a boundary is different from being near one. A particle inside the cone has a rest frame; a particle on it does not, at any energy, for any observer. No continuous process takes you from one to the other, which is why the massless limit is singular and why "just set $m = 0$ " fails on every formula that has a $\gamma$ in it.

Where this lives today

Masslessness is not a photon-only story, and its consequences are structural. In the Standard Model the photon is massless because the unbroken $U(1)$ gauge symmetry of electromagnetism forbids a mass term; the $W$ and $Z$ , whose symmetry is broken, are correspondingly heavy. So $m_\gamma = 0$ is a statement about a symmetry, and the experimental bound of $10^{-18}\ \mathrm{eV}/c^2$ is a test of that symmetry — which is why the limit is worth pushing. A massive photon would make Coulomb's law fall off exponentially, give light a frequency-dependent speed, and add a longitudinal polarisation; each of those is searched for, and each comes back null. The gluon is likewise massless, and the graviton would be if it exists. The neutrino, famously, was believed massless for decades and is not — oscillation requires non-zero masses, and that discovery forced a real amendment to the Standard Model. The episode is the best possible argument for stating masslessness as a bound. Practically, $E = pc$ is what makes a photon's momentum useful: radiation pressure and solar sails; laser cooling, which slows an atomic beam one $h/\lambda$ at a time; optical tweezers; and the radiation pressure that helps hold a massive star against its own gravity.

The physical picture

Return to the $(pc, E)$ plane and the mass shell $E^2 - (pc)^2 = (mc^2)^2$ . Each mass draws its own hyperbola, with vertex at $E = mc^2$ and asymptotes $E = \pm\,pc$ . Now shrink $m$ . The vertex slides down toward the origin, the curve tightens against its asymptotes — and at $m = 0$ exactly, the hyperbola degenerates into its own asymptote. The curve becomes the straight line $E = pc$ : the light cone of momentum space. That degeneration explains the speed degeneracy immediately. The slope of the mass shell is the particle's velocity — the owned result of the Energy–Momentum Relation. On a hyperbola the slope varies from point to point, so a massive particle's speed depends on where it sits: more momentum, more speed, always short of $c$ . But the degenerate "curve" is a straight line of slope exactly 1, and a straight line has the same slope everywhere. Every point on it has $\beta = 1$ . The photon's velocity does not depend on its momentum because the graph it lives on has no curvature left to encode the dependence.

Figure

PUNCHLINE — the mass shell degenerating into the light cone. As m shrinks the hyperbola tightens against its asymptote; at m = 0 exactly it BECOMES that asymptote, the straight line E = pc. The slope of the shell is β, and a straight line has the same slope everywhere — so β = 1 for EVERY momentum. That is the speed degeneracy. A 2 eV photon and a 1 GeV gamma ray sit at wildly different points on the same line, at identical slope

The line is the boundary of the region a massive particle can occupy, and boundaries behave differently from interiors. An LHC proton misses it by nine parts in a billion — but "nearly null" is not null. The proton still has a rest frame; the photon has none, at any energy, for any observer. That is the sense in which the massless limit is singular rather than smooth, and it is why formulas containing $\gamma$ cannot simply have $m = 0$ substituted into them.

The core equations

Everything begins with the norm, and with the observation that setting $m = 0$ in it is legitimate while setting $m = 0$ in $E = \gamma mc^2$ is not.

p\cdot p \;=\; \left(\frac{E}{c}\right)^{\!2} - |\vec{p}\,|^2 \;=\; m^2c^2 \;\xrightarrow{\;m\,=\,0\;}\; \boxed{\,p\cdot p = 0\,} \qquad\Longleftrightarrow\qquad \boxed{\,E = pc\,} \qquad (1)

where: $p^\mu = (E/c, \vec{p}\,)$ , $\;p \equiv |\vec{p}\,|$ . Physical reading: a four-vector of zero Minkowski length that is not itself zero — impossible in Euclidean space, where only the zero vector has zero length. Nullity is Lorentz-invariant, since $p\cdot p$ is a scalar. This is the safe route to $E = pc$ : it uses only the invariant, and the invariant has no $\gamma$ in it to blow up. The speed follows, and it follows without reference to the mass:

\vec{v} \;=\; \frac{\vec{p}\,c^2}{E} \;\xrightarrow{\;E\,=\,pc\;}\; |\vec{v}| \;=\; \frac{pc^2}{pc} \;=\; c \qquad \textbf{for every } p \qquad (2)

where: the relation $\vec v = \vec p\,c^2/E$ holds for any particle — it is the ratio $\gamma m\vec v / \gamma mc^2$ , and the $\gamma m$ cancels before the massless limit is ever taken. Physical reading: the momentum has dropped out. A massive particle's speed is a function of its momentum; a massless particle's is a constant. That degeneracy is what "on the light cone" means dynamically, and it is why $c$ can be an invariant speed at all — an object whose speed depends on its energy could not have the same speed for everybody. Because there is no rest frame, the quantum relations are how a photon is actually labelled:

E = h\nu = \hbar\omega, \qquad p = \frac{h}{\lambda} = \hbar k, \qquad p^\mu = \frac{E}{c}\big(1,\ \hat{n}\big) = \hbar\left(\frac{\omega}{c},\ \vec{k}\,\right) \qquad (3)

where: $\hat{n}$ — unit propagation direction; $\vec k$ — wave three-vector, $k = \omega/c$ . Physical reading: $E = pc$ and $\omega = ck$ are the same statement, which is why the wave four-vector is null exactly when $p^\mu$ is. A photon carries no mass to be labelled by, so it is labelled by frequency — and frequency is frame-dependent, which is the Doppler effect (owned by Doppler Shift from the Four-Momentum, not derived here). Finally, the inequality that does the article's real work:

\boxed{\,p_1\cdot p_2 \;\ge\; m_1 m_2 c^2\,} \quad\text{for future-pointing } p_1, p_2, \qquad\text{equality iff } p_1 \parallel p_2 \qquad (4)

where: "future-pointing" means $E_i > 0$ ; $m_i \ge 0$ allows either particle to be massless. Physical reading: evaluate it in particle 1's rest frame (if it has one) and it reads $m_1E_2 \ge m_1m_2c^2$ , i.e. $E_2 \ge m_2c^2$ — which is true, and is an equality only when particle 2 is also at rest there, i.e. when the two four-momenta point the same way. A timelike vector and a null vector can never be parallel, so for an electron and a photon the inequality is strict: $p_e\cdot p_\gamma > 0$ , always. Two famous no-go theorems follow, and they are the next section.

What nullity forbids

Derivation AThe massless limit is singular, and the safe way to take it

The pitfall this derivation exists to kill is the commonest error about photons: writing $E = \gamma mc^2$ , setting $m = 0$ , and concluding $E = 0$ — reasoning that would prove light does not exist. The error is that $\gamma \to \infty$ as $v \to c$ , so $\gamma m$ is an indeterminate $\infty \times 0$ , not a zero. Taking the limit properly shows what survives, what diverges, and why the invariant route of equation (1) was the honest one all along.

Step 1. Expose the offending factor. $E = \gamma mc^2$ and $\vec{p} = \gamma m\vec{v}$ : the same combination $\gamma m$ appears in each. Why: that is the structural fact that makes a limit exist at all. If $\gamma m$ appeared in one and $\gamma^2m$ in the other, no consistent massless limit would be available and there would be no such thing as a massless particle.

Step 2. Take the limit at fixed energy, not at fixed velocity. Hold $E$ fixed and let $m \to 0$ . Then $\gamma = E/(mc^2) \to \infty$ , which forces $v \to c$ ; but the product is pinned: $\gamma m = E/c^2$ , a perfectly finite number. Why: this is the whole trick. The factors are individually singular and their product is not. Physically: to keep a shrinking mass at a fixed energy you must drive it ever closer to $c$ , and the two divergences cancel.

Step 3. Read off the momentum. $\;\vec{p} = (\gamma m)\,\vec{v} \to \left(\dfrac{E}{c^2}\right)(c\,\hat{n}) = \dfrac{E}{c}\,\hat{n}$ , so $p = E/c$ , i.e. $\boxed{E = pc}$ . Why: the same answer equation (1) got from the invariant, now obtained by an explicit limit — so the two routes agree, and the limit is not a fudge.

Step 4. Confirm the degeneracy. Nothing in Step 3 fixed $E$ : any $E$ works, and every one gives $|\vec v| = c$ . Why: the limit has destroyed the map from energy to speed. For $m > 0$ that map is a bijection; at $m = 0$ it collapses to a constant. That collapse is the singularity — you cannot reach the massless case by making $m$ small, you have to be there.

Step 5. Note what does not survive. $\mathrm{d}\tau^2 = \mathrm{d}t^2(1 - v^2/c^2) \to 0$ along a null worldline, so $\mathrm{d}\tau = 0$ and the four-velocity $U^\mu = \mathrm{d}x^\mu/\mathrm{d}\tau$ is undefined. A photon has a four-momentum but no four-velocity. Why: not a technicality — it is why every formula built on $U^\mu$ (the four-force $K^\mu = mA^\mu$ , the orthogonality $K\cdot U = 0$ ) is silent about photons, and why $p^\mu$ , which needs no $\mathrm{d}\tau$ , is the only object that survives. *Verified in verify_aq-sr-massless-particles.py: $\lim_{m\to0}\gamma m = E/c^2$ at fixed $E$ (sympy limit); $\lim_{m\to0} p = E/c$ ; $\lim_{m\to0} v = c$ ; $E/p = c^2/v$ exactly for all $m$ , so $v = c \iff E = pc$ ; and $\mathrm{d}\tau = 0$ on a null worldline.*

Go deeper · Derivation B · 2 minA free electron cannot absorb a photon, and a photon cannot decay into a pair

Derivation BA free electron cannot absorb a photon, and a photon cannot decay into a pair

Here is the article's owned structural result. Both theorems are usually presented as separate curiosities with separate algebra. They are the same theorem: a null four-vector and a timelike one live on opposite sides of a boundary that four-momentum conservation cannot cross. One inequality — equation (4) — proves both, and it needs no energies, no angles, and no choice of frame.

Step 1. Prove the inequality. Let $p_1, p_2$ be future-pointing four-momenta with masses $m_1, m_2 \ge 0$ . If $m_1 > 0$ , evaluate $p_1\cdot p_2$ in particle 1's rest frame, where $p_1 = (m_1c,\ \vec 0\,)$ : then $p_1\cdot p_2 = m_1c\cdot(E_2/c) = m_1E_2$ , and since $E_2 \ge m_2c^2$ always, $\;p_1\cdot p_2 \ge m_1m_2c^2$ . Why: the invariant is frame-independent, so a result proved in one convenient frame holds everywhere. Equality needs $E_2 = m_2c^2$ — particle 2 at rest in particle 1's frame — i.e. the two four-momenta parallel.

Step 2. Note the strict case. If $p_1$ is timelike ( $m_1 > 0$ ) and $p_2$ is null and non-zero ( $m_2 = 0$ , $E_2 > 0$ ), the bound reads $p_1\cdot p_2 \ge 0$ — but equality would require the two to be parallel, and a timelike vector is never parallel to a null one (their norms, $m_1^2c^2 > 0$ and $0$ , would have to agree up to a positive factor). Hence $\boxed{p_e\cdot p_\gamma > 0 \text{ strictly}}$ . Why: in the electron's rest frame this is just $p_e\cdot p_\gamma = m_e E_\gamma > 0$ — a real photon has positive energy, so the product cannot vanish. The strictness is the whole theorem.

Step 3. Forbid absorption. Suppose a free electron absorbed a photon: $p_e + p_\gamma = p_e'$ , with the final object still an electron, so $p_e'\cdot p_e' = m_e^2c^2$ . Square the left side: $(p_e + p_\gamma)\cdot(p_e + p_\gamma) = \underbrace{p_e\cdot p_e}_{m_e^2c^2} + \underbrace{p_\gamma\cdot p_\gamma}_{0} + 2\,p_e\cdot p_\gamma$ . Setting the two sides equal gives $2\,p_e\cdot p_\gamma = 0$ — contradicting Step 2. Why: no energies were specified, no angles chosen, no frame preferred. The process is forbidden for every photon energy and every geometry. Turning up the intensity or the frequency cannot help: the obstruction is a sign, not a shortfall.

Step 4. Forbid pair creation in vacuum. Suppose $\gamma \to e^-e^+$ : $p_\gamma = p_- + p_+$ . Square the right side using Step 1 with $m_1 = m_2 = m_e$ : $(p_-+p_+)^2 = 2m_e^2c^2 + 2\,p_-\!\cdot p_+ \ge 2m_e^2c^2 + 2m_e^2c^2 = 4m_e^2c^2$ . But the left side is $p_\gamma\cdot p_\gamma = 0$ . So $0 \ge 4m_e^2c^2 > 0$ — false. Why: the pair's total four-momentum is timelike with invariant mass at least $2m_e$ , i.e. at least $1.022\ \mathrm{MeV}/c^2$ , while the photon's is null. Conservation demands they be the same four-vector, and a null vector cannot equal a timelike one. Again: true at any photon energy.

Step 5. Extract the physics. Both processes become allowed the moment a third body is present to absorb four-momentum: a nucleus, a lattice, a binding potential. That is not a loophole — it is the reason the photoelectric effect requires a bound electron and pair production requires matter (which is why its cross-section is zero in empty space and rises as $Z^2$ ). Why: a null vector plus a timelike vector can perfectly well equal a timelike vector. The no-go theorem is not a nuisance but a prediction — confirmed every time a gamma ray converts in a detector's lead plate rather than in the vacuum in front of it. *Verified in verify_aq-sr-massless-particles.py: the inequality $p_1\cdot p_2 \ge m_1m_2c^2$ checked symbolically in particle 1's rest frame and numerically at 500 random future-pointing pairs; $p_e\cdot p_\gamma = m_eE_\gamma > 0$ strictly; the absorption contradiction $2p_e\cdot p_\gamma = 0$ ; and $(p_-+p_+)^2 \ge 4m_e^2c^2 = (1.022\ \mathrm{MeV}/c)^2 > 0 = p_\gamma\cdot p_\gamma$ .*

Go deeper · Derivation C · 2 minNullity is invariant, so no observer can ever catch a photon

Derivation CNullity is invariant, so no observer can ever catch a photon

"Light moves at $c$ in every frame" is usually presented as Einstein's second postulate — an assumption. Once the four-momentum is in hand it becomes a theorem, in two lines. Worth doing, because it shows the postulate and the geometry are the same statement, and it settles what a photon "looks like" to an observer chasing it at $0.999c$ .

Step 1. Nullity is a Lorentz scalar statement. $p\cdot p$ is invariant, so if $p\cdot p = 0$ in one inertial frame it is $0$ in every inertial frame. Why: there is no boost that changes the norm, because $\Lambda^{\mathsf T}\eta\,\Lambda = \eta$ . The photon's four-momentum stays exactly on the cone no matter who looks.

Step 2. Therefore $E' = p'c$ in the new frame too, and by equation (2), $|\vec v'| = c$ . The speed is $c$ for every observer — not as a postulate but as a consequence of the norm being zero and staying zero. Why: the chase fails at the level of geometry. Run faster and you do change the photon's energy — you redshift it, $E' = \gamma E(1 - \beta\cos\theta)$ , which can be made arbitrarily small — but you never change its speed, because you cannot move a point off the light cone.

Step 3. Rule out the rest frame explicitly. A rest frame needs $\vec p' = \vec 0$ with $E' > 0$ ; then $p\cdot p = (E'/c)^2 > 0$ , contradicting $p\cdot p = 0$ . And $E' = 0$ with $\vec p' = \vec 0$ is not a photon — it is nothing. So no rest frame exists. Why: sharper than "you would have to travel at $c$ ." There is no frame — not an unreachable one, none — in which the photon's momentum vanishes while it still exists.

Step 4. Conclude that proper time does not advance. $c^2\mathrm{d}\tau^2 = c^2\mathrm{d}t^2 - |\mathrm{d}\vec x|^2 = 0$ on a null worldline, so a photon emitted at the Big Bang and absorbed in your eye has $\Delta\tau = 0$ between the two events. Why: this is not "time stops for the photon" — that phrasing smuggles in a photon's-eye frame, which Step 3 has just proved does not exist. The honest statement is that the interval along the worldline is zero: a fact about the path, not about anyone's experience of it. *Verified in verify_aq-sr-massless-particles.py: $p\cdot p = 0$ preserved by an arbitrary boost $\Lambda$ ; the boosted null four-momentum still satisfies $E' = p'c$ ; no solution exists to $\vec p' = 0$ with $E' > 0$ under the null constraint; and $\mathrm{d}\tau = 0$ identically on the null worldline.*

Figure

The two no-go theorems, in one picture. The photon's four-momentum lies ON the light cone. Any electron-positron pair has a total four-momentum INSIDE the cone, with invariant mass at least 2mₑc² = 1.022 MeV — there is a gap, and it does not close at high energy. Likewise an electron that absorbed a photon would have to stay on its own mass shell, which the strict inequality pₑ·p𝛾 > 0 forbids. Conservation cannot turn a null vector into a timelike one

Limiting cases

Case	Result	Meaning
$m = 0$ exactly	$p\cdot p = 0$ , $\;E = pc$	null four-momentum: on the light cone, not inside it
any $p$ , $m = 0$	$\beta = pc/E = 1$	speed is degenerate — energy says nothing about velocity
$m \to 0$ at fixed $E$	$\gamma \to \infty$ , $\;\gamma m \to E/c^2$	a finite $\infty\times0$ : the limit exists, but only as a product
$m \to 0$ in $E = \gamma mc^2$	undefined	the classic error. $\gamma m$ is indeterminate; use the invariant instead
massive, $pc \gg mc^2$	$E \approx pc$	approaches the cone, never reaches it. Still has a rest frame
null worldline	$\mathrm{d}\tau = 0$	no proper time, hence no four-velocity $U^\mu = \mathrm{d}x^\mu/\mathrm{d}\tau$
any boost	$p\cdot p = 0$ preserved	no rest frame exists — for any observer, not merely an unreachable one
boost toward/away	$E' = \gamma E(1-\beta\cos\theta)$	energy changes freely; speed does not change at all

Common trapNever set m = 0 in a formula that contains a gamma.

$E = \gamma mc^2$ and $\vec p = \gamma m\vec v$ are correct for massive particles and useless for massless ones. Substituting $m = 0$ gives $E = 0$ and $\vec p = \vec 0$ — which would prove that light carries neither energy nor momentum, contradicting every sunburn ever suffered. The error is that $v = c$ forces $\gamma = \infty$ , so $\gamma m$ is an indeterminate $\infty\times0$ , not a zero. The safe formulas are the ones with no $\gamma$ in them: the invariant $p\cdot p = m^2c^2$ , the relation $E^2 = (pc)^2 + (mc^2)^2$ , and the velocity $\vec v = \vec p\,c^2/E$ . Each of these accepts $m = 0$ without complaint, because each was derived from the four-momentum rather than from the four-velocity — and it is the four-velocity that does not exist for a photon.

Worked examples

1 · Weigh the sunlight — radiation pressure from E = pc alone

The solar constant at Earth's orbit is $S = 1361\ \mathrm{W\,m^{-2}}$ . Find the radiation pressure on a perfectly absorbing surface facing the Sun, and the force on a $1\ \mathrm{km^2}$ solar sail. Compare to atmospheric pressure.

Step 1 — identify. Each photon delivers energy $E$ and momentum $p = E/c$ — that is the entire input, and it is equation (1). A flux of energy is therefore automatically a flux of momentum, smaller by exactly $c$ . Momentum delivered per second per unit area is pressure.

Step 2 — apply. Energy flux $S$ carries momentum flux $S/c$ . So $P_{\text{abs}} = S/c = 1361 / (2.998\times10^{8}) = \mathbf{4.54\times10^{-6}\ \mathrm{Pa}} = 4.54\ \mu\mathrm{Pa}$ . For a perfect reflector the photons reverse, so the momentum transfer doubles: $P_{\text{refl}} = 2S/c = 9.08\ \mu\mathrm{Pa}$ .

Step 3 — the force on a sail. A $1\ \mathrm{km^2} = 10^{6}\ \mathrm{m^2}$ absorbing sail feels $F = 4.54\times10^{-6} \times 10^{6} = \mathbf{4.54\ \mathrm{N}}$ ; a perfectly reflecting one, $\mathbf{9.08\ \mathrm{N}}$ . Small — but it never stops and needs no fuel, which is why solar sails are flown.

Step 4 — the boundary number. Compare sea-level atmospheric pressure, $1.013\times10^{5}\ \mathrm{Pa}$ : sunlight's pressure is smaller by a factor of $\mathbf{2.2\times10^{10}}$ . That ratio is why measuring radiation pressure required a torsion balance in a vacuum.

Step 5 — what was not needed. No mass, no $\gamma$ , no velocity. The whole calculation used $E = pc$ and nothing else. A massless particle carries momentum, and the Sun pushes.

2 · Prove a free electron cannot absorb a photon — then explain the photoelectric effect

Show that $e^- + \gamma \to e^-$ is kinematically forbidden for a free electron at any photon energy. Then say what changes when the electron is bound in an atom.

Step 1 — identify. Conservation of four-momentum plus the mass-shell condition for the electron, before and after. Nothing else. If the process is forbidden it is forbidden invariantly, so we work with invariants.

Step 2 — apply conservation and square. $p_e + p_\gamma = p_e'$ . Take the Minkowski square of both sides. Left: $p_e\cdot p_e + 2\,p_e\cdot p_\gamma + p_\gamma\cdot p_\gamma = m_e^2c^2 + 2\,p_e\cdot p_\gamma + 0$ . Right: $p_e'\cdot p_e' = m_e^2c^2$ , because the final particle is still an electron and its mass is invariant.

Step 3 — the contradiction. Equating, $\;2\,p_e\cdot p_\gamma = 0$ , so $p_e\cdot p_\gamma = 0$ . But evaluate that invariant in the electron's rest frame, where $p_e = (m_ec,\ \vec 0\,)$ and $p_\gamma = (E_\gamma/c)(1,\hat n)$ : $\;p_e\cdot p_\gamma = m_ec \times E_\gamma/c = \mathbf{m_eE_\gamma}$ , which is strictly positive for any real photon ( $E_\gamma > 0$ ). Zero and positive cannot both be true. The process is forbidden — at every photon energy, every angle, and in every frame.

Step 4 — the boundary number, or rather its absence. There is no threshold to exceed. A $1\ \mathrm{eV}$ photon cannot be absorbed by a free electron, and neither can a $1\ \mathrm{TeV}$ one. The obstruction is a sign, not a shortfall, so no amount of energy relieves it — a far stronger statement than "it takes too much energy", and one only an invariant argument can make.

Step 5 — so how does the photoelectric effect work? The electron is not free. It is bound to an atom, and the atom recoils, contributing its own timelike four-momentum: $\gamma + \mathrm{atom} \to e^- + \mathrm{ion}$ . Now the sums can match. The recoil momentum is real but the recoil energy is tiny (the ion is thousands of times heavier), which is why Einstein's $K_{\max} = h\nu - \phi$ works so well while quietly relying on a third body being there at all. The same argument explains why pair production has zero cross-section in vacuum and rises as $Z^2$ near a nucleus: both processes are buying a timelike four-vector, and something has to sell it.

Physics of the portrait

The photon was inferred before it was named, named by someone who meant something else by it, and its defining property — exactly zero mass — remains, strictly speaking, an experimental upper bound rather than a proven fact.

Origins · 2 minPhoton and how this was discovered

Photon

1905–present · Einstein's light quantum · Compton's X-ray table · Lewis's letter to Nature

Einstein's March 1905 paper proposed that light is emitted and absorbed in localised quanta of energy $h\nu$ — the "heuristic viewpoint" that won him the Nobel Prize, and the only one of his own results he ever called revolutionary. It was not immediately believed, since it appeared to contradict a century of successful wave optics. The decisive step came in 1916, when Einstein showed the quantum must also carry a directed momentum $h\nu/c$ : energy alone could have been an artefact of the emission process, but directed momentum meant a genuinely localised carrier with a four-momentum. Arthur Compton confirmed it in 1923 by scattering X-rays off electrons and measuring a wavelength shift that made sense only if the quantum recoiled like a billiard ball with $p = h/\lambda$ . The name arrived last, and by accident: in a letter to Nature on 18 December 1926 the chemist Gilbert N. Lewis coined "photon" for a hypothetical conserved atom of radiant energy — a theory that was wrong and is now forgotten. Physicists kept the word and attached it to Einstein's quantum. Today the photon's masslessness is the best-tested zero in physics: the Particle Data Group's upper limit is $1\times10^{-18}\ \mathrm{eV}/c^2$ , some $5\times10^{23}$ times below the electron's mass.

History

1900

— Planck introduces the quantum of action h and the energy element E = hν to fit the black-body spectrum, without believing light itself is quantised.

1905

— Einstein's March paper, "Über einen die Erzeugung und Verwandlung des Lichtes betreffenden heuristischen Gesichtspunkt", proposes that light itself comes in localised quanta of energy hν.

1916

— Einstein shows in "Zur Quantentheorie der Strahlung" that the light quantum must also carry a directed momentum hν/c, making it a genuine four-momentum carrier rather than a bookkeeping device.

1923

— Compton scatters X-rays off electrons and measures the wavelength shift (Phys. Rev. 21, 483), confirming that the quantum carries momentum h/λ and recoils like a particle.

1926

— Gilbert N. Lewis coins the word "photon" in a letter to Nature (118, 874) for a different and now-forgotten concept; the name is adopted for Einstein's quantum instead.

2025

— the Particle Data Group's Review of Particle Physics carries the photon-mass upper limit m < 1e-18 eV/c², so masslessness remains an experimental bound rather than a theorem.

Going further

Texts. Rindler, Introduction to Special Relativity, 2nd ed., §§6.6, 35 · Griffiths, Introduction to Elementary Particles, 2nd ed., §3.4 · Taylor & Wheeler, Spacetime Physics, 2nd ed., ch. 7 · Jackson, Classical Electrodynamics, 3rd ed., §11.5 · Einstein, "Zur Quantentheorie der Strahlung", Phys. Z. 18, 121 (1917) · Compton, Phys. Rev. 21, 483 (1923) · Lewis, "The Conservation of Photons", Nature 118, 874 (1926).

Provenance. Signature $(+,-,-,-)$ , so $p\cdot p = +m^2c^2$ and a massless particle has $p\cdot p = 0$ — null, not negative. Everything above is recomputed in verify_aq-sr-massless-particles.py, which parses this article's own printed digits: the singular limit $\gamma m \to E/c^2$ ; $v = c$ for all $p$ ; invariance of nullity under a boost; the owned inequality $p_1\cdot p_2 \ge m_1m_2c^2$ (symbolically, and numerically at 500 random future-pointing pairs); both no-go theorems; and the printed $\lambda = 619.9$ nm, $p = 1.069\times10^{-27}\ \mathrm{kg\,m\,s^{-1}}$ , $S/c = 4.54\ \mu\mathrm{Pa}$ , sail forces $4.54$ N and $9.08$ N, the ratio $2.2\times10^{10}$ , $2m_ec^2 = 1.022$ MeV, and the photon-mass bound $10^{-18}\ \mathrm{eV}/c^2$ .

Scope. The Compton shift is not derived here — that kinematics belongs to Compton Scattering. The relativistic Doppler effect, obtained by boosting the null $p^\mu$ , belongs to Doppler Shift from the Four-Momentum. This article owns only what follows from $m = 0$ itself.

Builds on. Four-Momentum · Energy–Momentum Relation · Invariant Mass · Relativistic Energy. Leads to. Conservation of Four-Momentum · Compton Scattering · Doppler Shift from the Four-Momentum · Relativistic Kinetic Energy.

If a photon has no rest frame, what is its "rest mass" a statement about?

About the norm of its four-momentum, and nothing else. $m$ is defined by $p\cdot p = m^2c^2$ — a Lorentz-invariant number computable from any frame's measurements, with no need for a rest frame to exist. For a photon that norm is zero, so $m = 0$ . "Rest mass" is a historical accident of a name; invariant mass is the honest one, and it never mentions rest.

A photon has energy, and energy gravitates. Doesn't that make it massive?

No, and the distinction matters. What gravitates in general relativity is the stress–energy tensor, not "mass". A photon contributes energy, momentum and pressure to that tensor, so light does bend spacetime and is bent by it — but its invariant mass, the norm of its four-momentum, is exactly zero. Conflating the two is the relativistic-mass error again: it takes the frame-dependent $E/c^2$ and calls it mass, which makes "the photon is massless" sound like a contradiction rather than the precise statement it is. Two photons flying in different directions do jointly have a non-zero invariant mass — but that is a property of the system, not of either photon.

Does a redshifted photon slow down?

Never. Boosting changes a photon's energy by any factor you like — arbitrarily large blueshift, arbitrarily small redshift, and $E' \to 0$ in the limit — but it cannot change its speed, because it cannot move the four-momentum off the light cone. This is exactly the split the article is about: for a massive particle, energy and speed are two views of the same fact; for a photon they are independent. A photon redshifted by cosmological expansion has lost energy, has a longer wavelength, and travels at precisely $c$ , as it always did.

Photons are massless, so why do they carry momentum?

Because momentum in relativity is not $m\vec v$ . It is the spatial part of the four-momentum $p^\mu = (E/c, \vec p\,)$ , and $E^2 = (pc)^2 + (mc^2)^2$ permits $p \ne 0$ with $m = 0$ perfectly happily — that is precisely the case $E = pc$ . The Newtonian intuition that momentum requires mass is an artefact of the low-speed limit, where $p = \gamma m\vec v \approx m\vec v$ . It fails on the light cone, and it fails in a way you can feel: a solar sail is pushed by objects with no mass at all.

Is the photon's masslessness proved, or assumed?

Assumed by the theory, bounded by experiment. In the Standard Model an unbroken $U(1)$ gauge symmetry forbids a photon mass term, so $m_\gamma = 0$ is a consequence of a symmetry, not an independent postulate. Experimentally it is an upper limit: the Particle Data Group quotes $m_\gamma < 1\times10^{-18}\ \mathrm{eV}/c^2$ , some $5\times10^{23}$ times below the electron mass. A non-zero value would give Coulomb's law an exponential cutoff, make the speed of light frequency-dependent, and add a third polarisation state; all three are searched for, none is seen. The neutrino's history — assumed massless for decades, then found not to be — is why the bound is stated as a bound.

Can a photon decay into two lower-energy photons?

No, for the same reason as the other no-gos. If $p_\gamma = p_1 + p_2$ with all three null, squaring gives $0 = 2\,p_1\cdot p_2$ , so $p_1\cdot p_2 = 0$ — which by the article's inequality requires $p_1 \parallel p_2$ . The "decay" would have to send both daughters in exactly the same direction with no change to the total: not a decay, merely a relabelling. Anything genuinely branching — two photons at a real angle — has a timelike total four-momentum with non-zero invariant mass, and cannot equal the null $p_\gamma$ . (Read forwards, that is how the $\pi^0$ and the Higgs are found in the diphoton channel: a real angle between two photons means a real mass.)