Does Light Have a Clock?

§1 Prologue: A Sentence Worth Taking Apart

I was listening to a clip from a popular physics conversation. The argument goes like this. The faster you move, the slower your time runs as seen by others. As you approach the speed of light, time keeps slowing. At the speed of light, time stops. Therefore a photon from the Big Bang, absorbed in your retina tonight, is absorbed at the same instant it was emitted. We measure a delay of 13.8 billion years. The photon experiences none.

It is a beautiful sentence, and people repeat it because it feels like the deepest possible statement about light. I want to take it apart with the same tools this notebook uses everywhere else: derive every step, flag every idealization, and question every premise, including mine.

The argument makes three moves.

1. Moving clocks run slow as seen by others.

2. In the limit $v \to c$, time stops, so there is a sense in which the photon’s clock reads zero.

3. Therefore emission and absorption are, for the photon, the same instant.

The first move is correct. The second takes a limit that does not exist. The third confuses two geometric ideas that Lorentzian geometry keeps strictly apart. And beneath all three lies an assumption that is not a theorem but a measurement: that the photon’s mass is exactly zero. Whether light has a clock is not a question for philosophy. It is a question about one number, and that number has not been measured to be zero. It has only been bounded.

I will also show that the true content hidden inside the sentence is more interesting than the sentence. A massless particle carries no internal clock, and the reason is the same symmetry that makes light blind to the scale of spacetime. That is the thread I want to pull.

Conventions. Metric signature $(-,+,+,+)$. I keep $c$ and $\hbar$ explicit, and restore SI units in every numerical estimate. Cosmological numbers use a spatially flat Friedmann-Lemaitre-Robertson-Walker model with Planck 2018 parameters ($H_0 = 67.66\ \mathrm{km\,s^{-1}\,Mpc^{-1}}$, $\Omega_m = 0.3111$, radiation included, $\Omega_\Lambda$ fixed by flatness), integrated numerically.

§2 Proper Time, and What Is Actually Invariant

For any worldline $x^\mu(\lambda)$ with arbitrary parameter $\lambda$, the proper time is the length functional

Two properties matter. First, it is independent of the parametrization. Under $\lambda \to \tilde\lambda(\lambda)$ with $d\tilde\lambda/d\lambda > 0$, the square root picks up a factor $d\lambda/d\tilde\lambda$ that exactly cancels the change of measure. Second, it is a scalar: every observer computes the same number for the same worldline.

In an inertial frame, parametrize by coordinate time $t$ and write $\mathbf v = d\mathbf x/dt$. Then $-g_{\mu\nu}\dot x^\mu\dot x^\nu = c^2 – v^2$ and

This is time dilation. Two facts the clip passes over deserve stating. The effect is reciprocal: each of two inertial observers computes the other’s clock as running slow, because $t$ in the formula is the coordinate time of whoever is computing. Only proper time along a specified worldline is observer-independent, and asymmetries such as the twin effect come from comparing worldlines that are not both inertial. And the traveller never feels her time slow down. Her own clock ticks one second per second in her own frame. “Slower time” is always a statement about someone else’s clock, read in your coordinates.

Now the null case. A light ray is a curve with $g_{\mu\nu}\dot x^\mu\dot x^\nu = 0$ everywhere. The integrand vanishes identically, so

This is exact, invariant, and true. It is the solid ground under the clip. Everything that follows is about what may and may not be built on it.

§3 There Is No Frame Moving with Light

The clip’s second move needs an observer riding with the photon, whose clock we could consult. I will show in four independent ways that no such observer exists.

3.1 The four-velocity cannot be normalized

A rest frame for a particle is built from its four-velocity $u^\mu = dx^\mu/d\tau$, which by construction satisfies $u^\mu u_\mu = -c^2$. The rest frame is the orthonormal frame whose time axis is $u^\mu/c$. For a null worldline $d\tau = 0$, so $u^\mu$ is undefined. Nor can any rescaling rescue it: the tangent $k^\mu$ satisfies $k^\mu k_\mu = 0$, and for any real $\alpha$, $(\alpha k)^\mu(\alpha k)_\mu = 0$ as well. No multiple of a null vector is a unit timelike vector. A null curve has no time axis to offer.

3.2 The boost has no limit

Work in $1+1$ dimensions with light-cone coordinates

A boost with velocity $\beta c$ along $+x$ is $ct’ = \gamma(ct – \beta x)$, $x’ = \gamma(x – \beta ct)$. Then

With rapidity $\varphi$ defined by $\beta = \tanh\varphi$, we have $\gamma = \cosh\varphi$ and $\gamma\beta = \sinh\varphi$, so $\gamma(1\pm\beta) = e^{\pm\varphi}$ and

The boost is a hyperbolic rotation: it stretches one null direction and squeezes the other, preserving the product $uv$ and hence the interval. A photon moving in $+x$ travels along a line of constant $u$, and its emission and absorption events differ only in $v$. In the boosted frame their separation is $\Delta v’ = e^{-\varphi}\Delta v$, so both $\Delta t’$ and $\Delta x’$ shrink toward zero as $\varphi$ grows. That is the “zero time, zero distance” of the clip.

Now look at everything else. Any separation with $\Delta u \neq 0$ is stretched by $e^{\varphi}$ and diverges. The limit $v \to c$ is $\varphi \to \infty$, and the family of maps $(u,v) \mapsto (e^{\varphi}u, e^{-\varphi}v)$ has no limit in the Lorentz group. If one rescales to force a limit, the result is a rank-one projection that sends every event on a line of constant $u$ to one point. It is not invertible and assigns identical coordinates to distinct events. It is not a frame. The “photon’s point of view” is the limit of a sequence of frames that does not converge to a frame.

3.3 Chasing light redshifts it to nothing

The phase of a right-moving plane wave is $\omega(t – x/c) = (\omega/c)\,u$. Phase is a scalar: every observer agrees on the number of wave crests between two events. So $(\omega/c)\,u = (\omega’/c)\,u’ = (\omega’/c)\,e^{\varphi}u$, which gives

the longitudinal Doppler shift, derived here without any reference to clocks. The field amplitude transforms the same way: for a plane wave with $\mathbf E = E\hat{\mathbf y}$ and $\mathbf B = (E/c)\hat{\mathbf z}$, the boost gives $E’_y = \gamma(E_y – \beta c B_z) = \gamma(1-\beta)E = e^{-\varphi}E$.

As $\varphi \to \infty$, the frequency, the photon energy $\hbar\omega’$, and the field amplitude all go to zero. An observer who chases light harder and harder sees it redden, weaken, and vanish. The limiting “frame of the photon,” if it existed, would contain no photon.

3.4 Einstein’s frozen wave, from Maxwell

In his Autobiographical Notes Einstein recalled asking at sixteen what a light beam would look like to someone keeping pace with it. The answer is a field that oscillates in space but not in time, for instance

Test it against Faraday’s law. The curl is

which is nonzero, while $-\partial_t\mathbf B = 0$. Faraday’s law fails. (Ampere-Maxwell fails too, unless $\nabla \times \mathbf B = 0$, which removes the magnetic partner of the wave.) A frozen electromagnetic wave is not a solution of the source-free Maxwell equations. The wave equation derived in the first essay of this notebook requires every such disturbance to move at $c$ in every inertial frame. The same fact that makes $c$ a structural constant denies light a rest frame.

§4 Zero Interval Is Not the Same Instant

The clip’s third move reads $\Delta s^2 = 0$ as “no time passes, so emission and absorption are one moment.” This is where the argument actually fails, and the failure is instructive.

In Euclidean geometry, and in any Riemannian geometry, distance is positive definite: $d(A,B) = 0$ if and only if $A = B$. Lorentzian geometry is different. The quadratic form $\eta_{\mu\nu}\Delta x^\mu\Delta x^\nu$ is indefinite, and the set of events at zero interval from a given event is not a point. It is a three-dimensional cone.

Stand outside tonight. Every event you see sits on your past light cone: the Sun eight minutes ago, Andromeda 2.5 million years ago, the last-scattering surface of the cosmic microwave background. Every one of them is at zero interval from your eye. If zero interval meant “the same instant,” all of these would be the same instant as each other. They cannot be, and here is the proof.

Two events on the same past light cone are spacelike separated unless they lie on the same light ray. Put the observer at the origin and set $c = 1$ for this calculation. An event on the past light cone at distance $r$ in direction $\hat{\mathbf n}$ is $x = (-r,\ r\hat{\mathbf n})$. For two such events $A$ and $B$,

where $\theta$ is the angle between the two directions. This is strictly positive unless $\theta = 0$. The Sun and Andromeda lie in different directions, so the emission events you see from them are spacelike separated from each other. They are not even causally connected, yet both are null-separated from your retina. (Idealization flagged: flat spacetime. In curved spacetime the qualitative statement survives locally, and the global structure of light cones is exactly what the time-machines essay in this notebook was about.)

So null separation is not transitive. “Being at the same instant” has to be an equivalence relation: reflexive, symmetric, and transitive. Zero interval fails transitivity, so it cannot mean simultaneity in any sense, not even a private sense reserved for photons.

What zero interval does mean is precise and narrower. The emission event and the absorption event are distinct events, related by the boundary of causality: a signal travelling at exactly $c$ can connect them, and nothing slower can. That is a statement about the geometry between two points. It is not a statement that the two points merge.

§5 The Clock a Particle Carries

Having taken the sentence apart, I want to rebuild its honest core. There is a precise sense in which light carries no clock, and it comes from quantum mechanics, not from kinematics.

5.1 Proper time is a phase

Take a free particle of mass $m$ in the plane-wave state

Evaluate the phase along the particle’s classical worldline, starting at the origin. For a free massive particle, $p^\mu = m u^\mu$ and the worldline is $x^\mu(\tau) = u^\mu\tau$. Then

It is worth checking this in components, because it is surprising. With $\mathbf x = \mathbf v t$, $\mathbf p = \gamma m\mathbf v$ and $E = \gamma mc^2$:

So the quantum phase a particle accumulates along its own worldline is

Every massive particle carries an internal oscillator at its Compton frequency, and the number of radians that oscillator has turned is its proper time, in units of $1/\omega_C$. Proper time is not an abstraction laid on top of the particle. It is the particle’s own phase. (This is the de Broglie clock, which de Broglie himself used in his 1924 thesis to motivate matter waves.)

5.2 A light ray’s phase does not advance along the ray

For a photon, $p^\mu = \hbar k^\mu$ with $k^\mu k_\mu = 0$, and the ray is $x^\mu(\lambda) = k^\mu\lambda$. Then

The phase of the wave is constant along its own ray. The same result holds for a general wave in curved spacetime. Write the field as a rapidly varying phase $S(x)$ times a slowly varying amplitude, and define $k_\mu = \partial_\mu S$. The wave equation forces $k^\mu k_\mu = 0$ at leading order (§6.2 derives this). Along a ray, $dx^\mu/d\lambda = k^\mu$, and

The surfaces of constant phase are null hypersurfaces, and the rays lie inside them. A crest of light rides its own ray without ever becoming the next crest.

This is the honest content of “time stops for the photon.” A massless particle carries no internal clock: it has no Compton frequency, so its phase does not advance along its own trajectory. The statement is exact. It says nothing about experience, and it does not say emission and absorption are the same event.

5.3 Why light has no scale: conformal invariance

The deeper reason is a symmetry. In the time-machines essay I showed that the source-free Maxwell action,

is invariant under $g_{\mu\nu} \to \Omega^2(x)g_{\mu\nu}$, because $\sqrt{-g} \to \Omega^4\sqrt{-g}$, each inverse metric contributes $\Omega^{-2}$, and $F_{\mu\nu} = \partial_\mu A_\nu – \partial_\nu A_\mu$ involves no metric. A theory invariant under local rescaling of lengths and times contains no preferred length or time. A clock is a preferred time. So Maxwell theory cannot contain one.

Now add a mass. The Proca action is

Under the same rescaling, $\sqrt{-g}\,g^{\mu\nu} \to \Omega^4\,\Omega^{-2}\sqrt{-g}\,g^{\mu\nu} = \Omega^2\sqrt{-g}\,g^{\mu\nu}$, and $A_\mu$ with a lower index does not transform. The mass term is not invariant. It breaks conformal symmetry, and in breaking it introduces exactly one scale: the Compton length $\hbar/(mc)$, equivalently the Compton frequency $mc^2/\hbar$ of §5.1. Mass is a clock. Masslessness is the absence of one. The same symmetry that makes light blind to the one component of geometry that light cones do not determine also denies light a way to tell time.

(Idealizations flagged: conformal invariance of electromagnetism is classical and source-free. It fails with charges present, and at the quantum level through the trace anomaly, where vacuum polarization introduces the electron mass scale.)

5.4 The parameter along a ray is pure gauge

The point-particle action $S = -mc^2\int d\tau$ collapses at $m = 0$, so it cannot describe light. The standard fix introduces an auxiliary field $e(\lambda)$ along the worldline, the einbein:

Under reparametrization $\lambda \to \tilde\lambda$, require $e\,d\lambda = \tilde e\,d\tilde\lambda$. Then both terms are invariant, so $e\,d\lambda$ is the physical measure of “how far along the worldline.” Varying $e$ gives

Substituting back, $S = -mc\int\sqrt{-\dot x^2}\,d\lambda = -mc^2\int d\tau$, the usual action, and $e\,d\lambda = d\tau/m$. For a massive particle, the invariant length along the worldline is fixed by the dynamics, and it is proper time.

At $m = 0$ the variation of $e$ gives only the constraint $\dot x^2 = 0$. The einbein itself is left undetermined. It is a gauge degree of freedom. One may set $e = 1$, which makes $\lambda$ an affine parameter, but affine parameters are defined only up to $\lambda \to a\lambda + b$, and nothing in the theory fixes $a$. The photon’s worldline has an ordering of its points but no duration. It is not that the photon’s clock reads zero. It is that the question “how long did it take, according to the photon” has no gauge-invariant answer.

§6 A Ray with No Clock Still Has a History

The sentence I started from carries an unspoken corollary: if the photon experiences no time, nothing happens to it on the way. That corollary is false. Zero proper time does not mean an uneventful journey.

6.1 Redshift: the ray loses energy relative to the world it crosses

In a flat FLRW universe, $ds^2 = -c^2dt^2 + a^2(t)\,\delta_{ij}dx^idx^j$. The metric does not depend on $x^i$, so each $\partial_i$ is a Killing vector, and for any geodesic with tangent $p^\mu$ the covariant component $p_i = g_{i\mu}p^\mu = a^2 p^i$ is conserved. A comoving observer measures momentum in an orthonormal spatial frame $e_{\hat\imath} = a^{-1}\partial_i$, so the physical momentum is

This holds for massive and massless particles alike. For a photon, $E = p_{\text{phys}}c \propto 1/a$, so the ratio of emitted to received frequency is $1 + z = a_0/a_{\text{emit}}$.

For the cosmic microwave background, $z \approx 1090$. The mean photon energy today is $2.701\,k_BT_0 = 6.34\times10^{-4}$ eV at $T_0 = 2.7255$ K. At emission it was 1090 times larger, about $0.69$ eV, in a gas at about $2970$ K glowing near the infrared edge of the visible. The same quantum that left as a warm glow arrives as a microwave. It has aged in the only sense available to it: relative to the geometry it crossed.

6.2 Amplitude and polarization: geometric optics in curved spacetime

In vacuum, in Lorenz gauge $\nabla^\mu A_\mu = 0$, Maxwell’s equations in curved spacetime read

(The Ricci term arises from commuting covariant derivatives in $\nabla^\nu\nabla_\mu A_\nu$.) Insert the eikonal ansatz

and collect powers of the small parameter $\epsilon$, which measures wavelength against the scale on which amplitude and curvature vary.

At order $\epsilon^{-2}$: $k^\nu k_\nu\,a_\mu = 0$, so $k^\mu k_\mu = 0$. The wave vector is null.

The rays are geodesics. Since $k_\mu = \nabla_\mu S$ and the connection is torsion-free, $\nabla_\nu k_\mu = \nabla_\nu\nabla_\mu S = \nabla_\mu\nabla_\nu S = \nabla_\mu k_\nu$. Hence

At order $\epsilon^{-1}$:

Write $a_\mu = \mathcal A\,f_\mu$ with scalar amplitude $\mathcal A$ and unit polarization $f_\mu$, $f^\mu \bar f_\mu = 1$. Contract the transport equation with $\bar f^\mu$ and take the real part to get

conservation of photon number along the bundle. The amplitude falls as the bundle’s cross-section grows, which is the expansion $\theta = \nabla_\nu k^\nu$ of the Raychaudhuri equation from the time-machines essay. Substituting back gives

the polarization is parallel-transported along the ray, so it records the geometry the ray has crossed. (Idealization flagged: geometric optics holds when the wavelength is small compared with every other length in the problem. It fails near caustics and at wavelengths comparable to the curvature scale.)

So the ray redshifts, its amplitude evolves, and its polarization carries the imprint of the curvature it passed through. The ray has no clock of its own. The universe keeps time around it, and the ray arrives changed.

6.3 A caveat from quantum field theory

Even “the photon travels along a null line” is an idealization. In quantum electrodynamics the Feynman propagator, in Feynman gauge,

(overall sign depending on convention) is nonzero at spacelike and timelike separations as well as on the light cone. Causality is protected by a different object: the commutator $[A_\mu(x), A_\nu(y)]$, which vanishes for spacelike separation. What the theory contains are emission and absorption events and the amplitude connecting them. A little particle with a worldline is the geometric-optics limit of that picture. Asking what the photon experiences between the two events assumes a trajectory that the full theory does not assign.

§7 What the Clip Got Wrong About the Sky

Two factual errors in the clip are worth correcting, because the corrections are physics.

The cosmic microwave background was not emitted at the Big Bang. For its first few hundred thousand years the universe was a plasma, opaque because free electrons Thomson-scatter photons efficiently. Only when the temperature fell to about 3000 K could electrons and protons bind into neutral hydrogen, and the photons decoupled. Integrating the Friedmann equation with the parameters above, I find that last scattering at $z = 1090$ occurred about $380{,}000$ years after the Big Bang, so the light-travel time to us is $13.8$ billion years. No photon reaches us from the Big Bang itself.

“13.8 billion years” is a time, not a distance. The same integration gives the comoving distance to the last-scattering surface,

while the proper distance at the moment of emission was $a_{\text{rec}}\chi \approx 41.5$ million light-years. The light left a region only about forty million light-years away and has been fighting the expansion ever since.

Your retina cannot absorb a CMB photon. Rhodopsin, the photopigment in rod cells, needs about $1.8$ eV to isomerize, corresponding to the long-wavelength edge of vision near 700 nm. The mean CMB photon carries $6.34\times10^{-4}$ eV, short by a factor of about 2800. The spectrum peaks near $1.06$ mm in wavelength. CMB photons that reach you are absorbed as heat in tissue. You have heard them, though: a small fraction of the static on an old analog television came from the microwave background. The oldest light your eye can actually detect comes from Andromeda, about 2.5 million years old, which is where the previous essay in this notebook began.

§8 The Clock We Did Find: Neutrinos

There is a historical precedent that should make anyone careful about saying a fast particle “experiences no time.” For decades the Standard Model treated neutrinos as exactly massless. On that assumption, every sentence in the clip would have applied to them. A neutrino from the Sun would reach your body at the same instant it left the Sun, as far as it was concerned.

Then neutrinos were caught changing identity in flight: Super-Kamiokande’s atmospheric neutrinos in 1998, and the Sudbury Neutrino Observatory’s solar neutrinos in 2002. Flavour oscillation is impossible without mass differences, because oscillation is literally a disagreement between clocks.

8.1 Oscillation as a comparison of clocks

A neutrino produced in flavour state $|\nu_\alpha\rangle$ is a superposition of mass eigenstates $|\nu_i\rangle$ with masses $m_i$. Take two eigenstates with mixing angle $\vartheta$, and assume they share the same energy $E$ (a simplification that wave-packet treatments justify for this purpose). Each has momentum

At the detection event $(t, L)$, each component has phase $\phi_i = (Et – p_iL)/\hbar$. The phase difference is

and the survival probability is $P(\nu_\alpha\to\nu_\alpha) = 1 – \sin^2 2\vartheta\,\sin^2(\Delta\phi/2)$, the standard result.

Now I want to question my own §5.1. Suppose I naively compute each eigenstate’s Compton phase along its own worldline: $\phi_i = m_ic^2\tau_i/\hbar$ with $\tau_i = L/(\gamma_iv_i) = m_iL/p_i \approx m_icL/E$. Then $\phi_i \approx m_i^2c^3L/(\hbar E)$ and

twice the correct value. This factor of two has a literature of its own. The resolution is that interference compares the two components at the same detection event, while their classical worldlines reach $x = L$ at different times $t_i = L/v_i$. Split the exact phase:

Using $v_i = p_ic^2/E$, the second term is $\dfrac{m_i^2c^4}{E}\,\dfrac{L}{v_i} = m_ic^2\tau_i$, the Compton clock. The first term is an arrival-time offset. Its difference between the two eigenstates is $-\Delta(Et_i)$, and since $Et_i = E^2L/(p_ic^2) \approx (EL/c)\left(1 + m_i^2c^4/2E^2\right)$, it contributes exactly $-\tfrac12\Delta\phi_{\text{naive}}$. The clocks are real, and each reads its proper time exactly. But comparing two clocks at one event requires correcting for the fact that they did not arrive together.

8.2 How old is a supernova neutrino?

The neutrinos detected from Supernova 1987A in the Large Magellanic Cloud travelled about $51$ kpc, or $168{,}000$ light-years, with energies around $10$ MeV. The atmospheric mass splitting $|\Delta m^2| \approx 2.5\times10^{-3}\ \mathrm{eV}^2$ implies that the heaviest mass eigenstate has $m \geq 0.05$ eV. For that eigenstate,

At least one component of those neutrinos aged by at least seven hours on a journey of 168,000 years. The lightest eigenstate could still be massless: its mass has not been measured. So a single neutrino arriving from that supernova may be a superposition of a component that aged by hours and a component that did not age at all. We know neutrinos have clocks because we watched those clocks disagree.

§9 Does Light Age? The Number That Decides

Everything so far assumed $m_\gamma = 0$ exactly. That is consistent with every measurement, and it is not a theorem. The Particle Data Group quotes an upper bound $m_\gamma < 10^{-18}$ eV, from magnetohydrodynamic arguments about the solar wind, and even that bound rests on modelling assumptions that have been debated. If the photon has any mass at all, it has a rest frame, a Compton clock, and a finite proper time. The clip’s claim then becomes quantitatively false.

9.1 The proper time of a relic photon

A photon of mass $m$ obeys $E^2 = p^2c^2 + m^2c^4$, and by §6.1 its momentum measured by comoving observers is $p = p_0/a$, with $a_0 = 1$ today and $E_0 = p_0c$ its present energy. Then

For every relic photon, $E_0/mc^2 \geq 6\times10^{14} \gg 1 \geq a$, so

The integral weights each epoch by the scale factor, because earlier the photon was more energetic and its clock ran slower. With the Planck 2018 parameters (radiation included), I find

(I also evaluated the exact integrand with $a/\sqrt{(E_0/mc^2)^2 + a^2}$. It agrees with the approximation to the precision quoted below.) With $E_0 = 6.34\times10^{-4}$ eV and the Compton phase $\omega_C\tau = mc^2\tau/\hbar$:

At the present experimental bound, a microwave-background photon could have aged about six minutes over 13.8 billion years, turning its internal clock through a little more than half a radian. Nothing we have measured rules that out.

For comparison, consider the fastest massive particle ever recorded: the cosmic-ray proton detected by the Fly’s Eye array in 1991, at $3.2\times10^{20}$ eV. Its Lorentz factor was $E/m_pc^2 \approx 3.4\times10^{11}$. Crossing the Milky Way, about $10^5$ light-years, would take it

That is extraordinarily close to zero, and it is not zero. The difference between “almost no time” and “no time” is the difference between a particle that carries a clock and one that does not, and the only thing that separates light from the first category is the claim that $m_\gamma$ vanishes exactly.

9.2 Two different ways light could carry a clock

The table assumes an intrinsic Proca mass. There is a second possibility, and it is the one experiments can reach most easily. Suppose the photon is exactly massless but mixes weakly with a massive “hidden photon” of mass $m’$, through a kinetic-mixing term $-\tfrac{\chi}{2}F_{\mu\nu}F’^{\mu\nu}$ (Holdom 1986; the idea goes back to Okun’s paraphotons). Then the state we call a photon is, to first order in $\chi$, a superposition of a massless eigenstate and a massive one, exactly as a flavour neutrino is. By the argument of §8.1, the relative phase after distance $L$ at energy $E$ is $\Delta\phi = m’^2c^3L/(2\hbar E)$, and the conversion probability is

In this scenario the photon has no clock of its own, but its state carries a relative clock, one component ageing against another. Laboratory “light-shining-through-a-wall” experiments and distortions of the microwave background spectrum search for exactly this phase. The question “does light have a clock?” is therefore not rhetorical. It is being asked, with lasers and cavities, right now.

9.3 A question worth one calculation

The last row of the table is not arbitrary. If the photon mass is of order $\hbar H_0/c^2$, its Compton length is of order the Hubble radius. Elsewhere in this research programme I have asked whether an infrared cutoff at that scale could connect the photon mass to the observed vacuum energy, the same $\Lambda$ that set the size of the reachable universe in the previous essay. At that mass, the photon’s clock would have turned through about $10^{-30}$ of a radian since recombination. Whether light has a clock at all, and whether the vacuum’s energy is what it is, may turn out to be questions about the same number.

I want to be exact about the status of the Proca mass itself. A Proca photon has three polarization states, not two. The longitudinal mode couples to conserved currents with a strength suppressed by $m_\gamma/\omega$, which for a CMB photon at the PDG bound is about $10^{-18}/(6\times10^{-4}) \approx 2\times10^{-15}$. That suppression is why a tiny mass is so hard to detect, and why “consistent with zero” and “equal to zero” remain different statements.

§10 Questioning the Clip, and Myself

10.1 The clip

The clip deserves credit for its first move, which is correct and which most people still find astonishing. Its instinct is also sound: there is a deep fact about light hidden in the zero proper time of a null ray. The trouble is in the words chosen to express it.

– “At the speed of light, time stops.” No observer travels at the speed of light, and the limit of observers approaching it is not an observer (§3). What is true is that proper time along a null curve is zero (§2). – “It is the same instant it was emitted.” Zero interval is not identity and is not transitive (§4). Emission and absorption are distinct events on the boundary of causality. – “The photon does not experience that time delay.” A massless particle has no clock (§5), but it still redshifts, dims, and carries the parallel-transported record of the geometry it crossed (§6). – “Emitted at the Big Bang, absorbed in your retina.” Emitted about 380,000 years later, and invisible to the eye (§7). – The unstated premise: $m_\gamma = 0$ exactly. That is a bound, not a measurement (§9).

10.2 Myself

Does §5.1 prove that proper time “is” a phase? It proves that for a free particle in a momentum eigenstate, evaluated along its classical worldline, the quantum phase equals $-mc^2\tau/\hbar$. Section 8.1 shows that using that clock naively in an interference problem gives the wrong answer by a factor of two. The clock is real. Its operational meaning depends on how clocks are compared.

Is “a massless particle has no clock” a law? It is a consequence of conformal invariance of the free Maxwell field. That invariance is broken by sources and by the trace anomaly. A photon propagating through the quantum vacuum has self-energy corrections that could in principle induce an effective dispersion. Gauge invariance protects $m_\gamma = 0$ in quantum electrodynamics, but gauge invariance is itself an assumption whose origin this notebook has repeatedly questioned. The Stueckelberg mechanism allows a photon mass without breaking gauge invariance at all.

Is geometric optics the full story? No. It fails near caustics, at long wavelengths, and wherever the photon is not a localized packet. Section 6.3 shows that the full quantum theory does not assign a trajectory. “What the photon experiences in between” may simply be a question without a referent.

What does “experience” mean for a particle? For a massive particle, the only physically meaningful content is proper time along its worldline, which is also its accumulated Compton phase. For a massless particle, even that is undefined up to an affine rescaling (§5.4). I do not claim anything about experience beyond this. Claims about what a photon “sees” add nothing that the invariant statements do not already contain.

§11 Coda: Light Has an Order but Not a Duration

I began with a sentence that says light experiences its whole history as one instant. What survives careful questioning is stranger and more precise.

Light has no rest frame, because the only way to catch it is to redshift it to nothing. Its emission and absorption are not one event but two, joined by the edge of causality, and zero interval is not the same thing as coincidence. Along its own path its phase does not turn, because it carries no internal oscillator, and it carries none because the equations of light have no preferred scale. The parameter along its ray is a gauge choice: light has an order but not a duration. And yet it does not travel untouched. It arrives reddened, dimmed, and carrying the geometry of every stretch of space it crossed.

Whether even this is exactly true depends on one number. If the photon’s mass is zero, light has no clock at all. If it is anything else, however small, then somewhere inside every photon from the edge of the observable universe a clock has been turning very slowly, perhaps for six minutes in fourteen billion years, perhaps for a trillionth of a trillionth of a radian. We have not yet read it. We do not know if it is there.

References

DOIs verified against Crossref, September 2026.

A. Relativity and the foundations of the argument

1. Einstein, A. (1905). Zur Elektrodynamik bewegter Körper. Ann. Phys. 322, 891. doi:10.1002/andp.19053221004

2. Einstein, A. (1949). Autobiographical notes. In P. A. Schilpp (Ed.), Albert Einstein: Philosopher-Scientist. Library of Living Philosophers.

3. Wald, R. M. (1984). General Relativity. University of Chicago Press. (Conformal invariance of Maxwell theory, Appendix D; causal structure.)

4. Misner, C. W., Thorne, K. S., and Wheeler, J. A. (1973). Gravitation. Freeman. (Geometric optics in curved spacetime, §22.5.)

5. Hawking, S. W., King, A. R., and McCarthy, P. J. (1976). A new topology for curved space-time which incorporates the causal, differential, and conformal structures. J. Math. Phys. 17, 174. doi:10.1063/1.522874

6. Malament, D. B. (1977). The class of continuous timelike curves determines the topology of spacetime. J. Math. Phys. 18, 1399. doi:10.1063/1.523436

B. Matter waves, clocks, and worldline actions

7. de Broglie, L. (1925). Recherches sur la théorie des quanta. Ann. Phys. (Paris) 10(3), 22. doi:10.1051/anphys/192510030022

8. Brink, L., Di Vecchia, P., and Howe, P. (1977). A Lagrangian formulation of the classical and quantum dynamics of spinning particles. Nucl. Phys. B 118, 76. doi:10.1016/0550-3213(77)90364-9

9. Peskin, M. E., and Schroeder, D. V. (1995). An Introduction to Quantum Field Theory. Addison-Wesley. (Photon propagator and microcausality.)

C. Cosmology and the microwave background

10. Penzias, A. A., and Wilson, R. W. (1965). A measurement of excess antenna temperature at 4080 Mc/s. Astrophys. J. 142, 419. doi:10.1086/148307

11. Peebles, P. J. E. (1968). Recombination of the primeval plasma. Astrophys. J. 153, 1. doi:10.1086/149628

12. Fixsen, D. J. (2009). The temperature of the cosmic microwave background. Astrophys. J. 707, 916. doi:10.1088/0004-637X/707/2/916

13. Planck Collaboration (2020). Planck 2018 results. VI. Cosmological parameters. Astron. Astrophys. 641, A6. doi:10.1051/0004-6361/201833910

D. Neutrino clocks

14. Hirata, K., et al. (Kamiokande-II) (1987). Observation of a neutrino burst from the supernova SN1987A. Phys. Rev. Lett. 58, 1490. doi:10.1103/PhysRevLett.58.1490

15. Bionta, R. M., et al. (IMB) (1987). Observation of a neutrino burst in coincidence with supernova 1987A in the Large Magellanic Cloud. Phys. Rev. Lett. 58, 1494. doi:10.1103/PhysRevLett.58.1494

16. Fukuda, Y., et al. (Super-Kamiokande) (1998). Evidence for oscillation of atmospheric neutrinos. Phys. Rev. Lett. 81, 1562. doi:10.1103/PhysRevLett.81.1562

17. Ahmad, Q. R., et al. (SNO) (2002). Direct evidence for neutrino flavor transformation from neutral-current interactions in the Sudbury Neutrino Observatory. Phys. Rev. Lett. 89, 011301. doi:10.1103/PhysRevLett.89.011301

18. Akhmedov, E. K., and Smirnov, A. Yu. (2009). Paradoxes of neutrino oscillations. Phys. Atom. Nucl. 72, 1363. doi:10.1134/S1063778809080122

E. Photon mass and hidden photons

19. Proca, A. (1936). Sur la théorie ondulatoire des électrons positifs et négatifs. J. Phys. Radium 7, 347. doi:10.1051/jphysrad:0193600708034700

20. Goldhaber, A. S., and Nieto, M. M. (2010). Photon and graviton mass limits. Rev. Mod. Phys. 82, 939. doi:10.1103/RevModPhys.82.939

21. Ryutov, D. D. (2007). Using plasma physics to weigh the photon. Plasma Phys. Control. Fusion 49, B429. doi:10.1088/0741-3335/49/12B/S40

22. Navas, S., et al. (Particle Data Group) (2024). Review of particle physics. Phys. Rev. D 110, 030001. doi:10.1103/PhysRevD.110.030001

23. Okun, L. B. (1982). Limits of electrodynamics: paraphotons? Sov. Phys. JETP 56, 502.

24. Holdom, B. (1986). Two U(1)’s and ε charge shifts. Phys. Lett. B 166, 196. doi:10.1016/0370-2693(86)91377-8

25. Jaeckel, J., and Ringwald, A. (2010). The low-energy frontier of particle physics. Annu. Rev. Nucl. Part. Sci. 60, 405. doi:10.1146/annurev.nucl.012809.104433

F. The fastest massive particle

26. Bird, D. J., et al. (1995). Detection of a cosmic ray with measured energy well beyond the expected spectral cutoff due to cosmic microwave radiation. Astrophys. J. 441, 144. doi:10.1086/175344