Gravitational Redshift and the Pound–Rebka Experiment

In 1960 two physicists at Harvard weighed a photon. They did it by dropping one 74 feet down a lift shaft and asking whether it arrived heavier. It did, by $2.46\times10^{-15}$ of itself. `aq-sr-rindler-metric` had the number waiting; `aq-sr-gravitational-time-dilation-heuristic` gives three ways to get it. That is not what this article is about. This article is about the fact that the measurement should have been impossible. Their gamma ray has a natural linewidth of $3.24\times10^{-13}$ — one of the sharpest features in physics. The shift they were hunting is one hundred and thirty-two times narrower than that line. You are not measuring a peak that moved; you are measuring a peak that moved by less than a percent of its own width — in a building, over days, while the building's temperature changes. And here is the trap that nearly ended it: > A one-degree temperature difference between the source and the absorber produces a frequency shift of $-2.44\times10^{-15}$ — the same size as the entire gravitational effect they were looking for, with the opposite sign. Pound and Rebka say so themselves, in the paper: "A difference of 1 °C would produce a shift as large as that sought." The systematic that could have counterfeited their result is itself relativistic time dilation — the second-order Doppler shift of nuclei jiggling in a hot crystal. The noise and the signal were the same physics. They beat it. Here is how.

Key takeaways

Signature $(+,-,-,-)$ : $\mathrm{d}s^2 = c^2\mathrm{d}t^2 - \mathrm{d}x^2$ , timelike positive. No metric is derived here; this article is about the measurement.
Scope, up front. The prediction $\Delta\nu/\nu = -gh/c^2 = -2.46\times10^{-15}$ is owned by `aq-sr-rindler-metric`, which also owns the analysis of what Pound and Rebka could and could not have seen. The three routes to it are owned by `aq-sr-gravitational-time-dilation-heuristic`. This article owns the experiment: why a shift of $10^{-15}$ is measurable at all.
Owned result: without the Mössbauer effect the experiment is not hard, it is impossible. A free $^{57}$ Fe nucleus emitting a $14.4$ keV gamma recoils with $E_R = \mathbf{1.96\ meV}$ against a natural linewidth $\Gamma = \mathbf{4.67\ neV}$ — pushing emission and absorption lines $2E_R = \mathbf{8.4\times10^{5}}$ linewidths apart. There is no resonance. None. Mössbauer's 1958 discovery — the recoil goes to the whole crystal — collapses that gap to zero.
Owned result: the shift is 1/132 of a linewidth. $\Gamma/E = 3.24\times10^{-13}$ against a shift of $2.46\times10^{-15}$ . In velocity units the whole effect is $\mathbf{0.74\ \mu m/s}$ against a line $0.194$ mm/s wide — one part in 459 of the line they actually observed.
Owned result — THE TRAP: the dominant systematic IS special relativity. Thermal motion gives every nucleus a second-order Doppler shift $-\langle v^2\rangle/2c^2$ , so the line moves with temperature: $\mathrm{d}z/\mathrm{d}T = -3k_B/2Mc^2 = \mathbf{-2.44\times10^{-15}\ K^{-1}}$ (classical limit) — $\mathbf{0.99}$ times the gravitational shift, opposite in sign. 1.01 K cancels gravity exactly.
Owned result: the raw observable is the TWO-WAY difference, $2\times$ the textbook number. Swapping source and absorber and taking the difference kills every shift that does not reverse with direction. Predicted $-4.92\times10^{-15}$ ; measured $-(5.13 \pm 0.51)\times10^{-15}$ ; headline $1.05 \pm 0.10$ .
⚠ The literature trap. Several databases list this paper's "abstract" as reporting $(-17.6 \pm 0.6)\times10^{-15}$ . That is not the gravitational result — it is the mean of the two orientations, which measures the source–absorber pair's intrinsic offset.
Key idea: the Mössbauer effect did not make this experiment easier. It made it exist. And the systematic that almost killed it was the same relativity it was built to test.

The core insight

Every precision experiment is a fight between one number you want and a dozen you do not. What makes Pound–Rebka a masterpiece is that its worst enemy was its own subject: thermal jiggling produces a time-dilation shift of exactly the size of the gravitational one, so a badly-insulated tower would have returned a beautiful, self-consistent, completely wrong answer. The two-way protocol — run it up, run it down, subtract — is what saves it, because gravity reverses when you turn the apparatus over and a warm basement does not.

Where this lives today

Mössbauer spectroscopy outgrew the physics that spawned it, immediately. The same $10^{-13}$ resolving power that sees a gravitational shift sees the hyperfine structure of the nucleus's environment — oxidation state, magnetic field, lattice site. It became a routine tool in chemistry, mineralogy and materials science, and it is why the Mars rovers Spirit and Opportunity each carried a miniature Mössbauer spectrometer. The technique built to weigh a photon now identifies rust on another planet.

The redshift measurement moved to clocks and never came back. Pound and Rebka reached $10\%$ ; Pound and Snider $0.8\%$ ; Vessot and Levine's hydrogen maser on a rocket reached $7\times10^{-5}$ in 1976. Optical lattice clocks have since resolved the redshift across a millimetre of atomic sample (Bothwell et al., Nature 602, 420, 2022).

And the temperature systematic never went away. Every atomic clock carries a second-order Doppler correction — now called the thermal or time-dilation shift — and it is a leading term in the uncertainty budget of the best of them, because trapped atoms still move and moving clocks still run slow. The nuisance Pound and Rebka had to beat in 1960 is the one the optical-clock community is still beating, with the same formula.

The physical picture

Nuclear resonance is the sharpest kind of resonance there is. The $^{57}$ Fe nucleus has an excited state $14.4$ keV above its ground state, living for a mean time $\tau = 141$ ns, so by the uncertainty relation the level has an energy width

\Gamma = \frac{\hbar}{\tau} = \mathbf{4.67\times10^{-9}\ eV} \qquad\Longrightarrow\qquad \frac{\Gamma}{E} = \mathbf{3.24\times10^{-13}} .

Three parts in ten trillion. Emit that gamma from one nucleus and absorb it in another and you have a frequency comparator good to $10^{-13}$ — exactly what is needed to see $10^{-15}$ after averaging.

You cannot. The reason is embarrassingly simple. A nucleus emitting a photon of energy $E$ recoils with momentum $E/c$ , carrying away kinetic energy

E_R = \frac{E^2}{2Mc^2} = \mathbf{1.96\times10^{-3}\ eV} .

The emitted photon is short by $E_R$ ; the absorbing nucleus, which must also recoil, needs one $E_R$ too energetic. So the two lines sit $2E_R$ apart — in units of the linewidth,

\frac{2E_R}{\Gamma} = \mathbf{8.4\times10^{5}} .

Eight hundred and forty thousand linewidths. The lines do not merely fail to overlap; they are not in the same postcode. Free-nucleus gamma resonance is not weak. It is zero. That was the state of play until 1958, when a graduate student in Heidelberg noticed that a nucleus locked into a crystal does not have to recoil alone.

Figure

WHY THIS EXPERIMENT SHOULD NOT HAVE BEEN POSSIBLE — three scales, and they span ten orders of magnitude. Top bar: the FREE NUCLEUS. Emitting a 14.4 keV gamma, an Fe-57 nucleus recoils with 1.96 meV of kinetic energy, and the absorber needs the same again — so the emission and absorption lines sit 2E_R apart, which is 840,000 natural linewidths. They do not overlap. There is NO nuclear resonance with free nuclei, and there never can be. Middle bar: MÖSSBAUER, 1958. Lock the nucleus into a crystal and the recoil momentum is taken up by the entire lattice — some 10^20 nuclei — so the recoil ENERGY, which goes as 1/M, collapses to nothing. The gap closes. Emission and absorption lines land on top of one another and the full 3.24 × 10⁻¹³ resolving power of the nuclear line becomes available. That single fact is the whole experiment. Bottom bar: WHAT THEY WERE LOOKING FOR. The gravitational shift over 74 feet is 2.46 × 10⁻¹⁵ — one part in 132 of the natural linewidth they had just unlocked, and one part in 459 of the line they actually observed. In velocity units the entire weight of the photon is 0.74 MICROMETRES PER SECOND. That is the speed of a growing fingernail

The core equations

\boxed{\;E_R \;=\; \frac{E^{2}}{2Mc^{2}}\;} \qquad\Longrightarrow\qquad \frac{2E_R}{\Gamma} \;=\; 8.4\times10^{5} \quad\text{for }^{57}\text{Fe}

where: $E = 14.4$ keV is the transition energy, $M$ the recoiling mass, $\Gamma = \hbar/\tau$ the natural width. Physical reading: the recoil energy goes as $1/M$ — and that is the entire Mössbauer effect. Bind the nucleus into a lattice and the momentum goes to the whole crystal, so $M \to M_{\text{crystal}} \sim 10^{20}M_{\text{nuc}}$ and $E_R \to 10^{-20}$ of its former self, twenty orders of magnitude below the linewidth. The gap does not shrink. It vanishes.

\frac{\Gamma}{E} \;=\; \frac{\hbar}{\tau E} \;=\; \mathbf{3.24\times10^{-13}} \qquad\Longleftrightarrow\qquad 2\Gamma \;\hat{=}\; \mathbf{0.194\ mm/s}\ \text{ of Doppler velocity}

where: $\tau = 141$ ns is the mean life of the $14.4$ keV state. Physical reading: the instrument's resolution, as a speed. A Mössbauer spectrometer scans the line by moving the source, so every frequency here is quoted as a velocity — millimetres per second is the natural unit of the field.

\boxed{\;\left(\frac{\Delta\nu}{\nu}\right)_{\text{grav}} = -\frac{gh}{c^{2}} = -2.46\times10^{-15}\;}\qquad (h = 74\ \text{ft} = 22.56\ \text{m})

where: the sign is negative for light going up (redshift). Physical reading: the prediction, quoted — not derived here. `aq-sr-rindler-metric` derives it, owns it, and shows the difference between Earth's real $1/r^2$ field and a uniform one is $8.7\times10^{-21}$ over this tower, $2.9\times10^{4}$ times below the error bar. Pound and Rebka tested the equivalence principle; they could not have tested curvature. In velocity units the shift is $\mathbf{0.74\ \mu m/s}$ ; on the $3.49\times10^{18}$ Hz gamma ray, $\mathbf{8.6\ kHz}$ .

\boxed{\;\frac{\mathrm{d}z}{\mathrm{d}T} \;=\; -\frac{3k_B}{2Mc^{2}} \;=\; \mathbf{-2.44\times10^{-15}\ K^{-1}}\;}

where: $z = \Delta\nu/\nu$ , $M$ the $^{57}$ Fe nuclear mass. Physical reading: the second-order Doppler shift — the villain. A nucleus vibrating in a hot lattice has $\langle v^2\rangle = 3k_BT/M$ (equipartition), and time dilation slows its internal clock by $\langle v^2\rangle/2c^2$ . It is $0.99$ times the gravitational shift, with the opposite sign — so a source one degree warmer than its absorber fakes the entire experiment, backwards.

How you measure one part in a thousand trillion

Derivation AWhy free nuclei cannot resonate, and how a crystal fixes it

Nuclear resonance fluorescence — one nucleus emits a gamma ray, an identical nucleus absorbs it — was hunted for decades and never seen. The goal here is to compute exactly why; to show the failure is not marginal but catastrophic by five orders of magnitude; and then to show that Mössbauer's fix does not chip away at the problem, it annihilates it. Everything Pound–Rebka is rests on this one number.

Step 1. Conserve momentum. A nucleus at rest emitting a photon of energy $E$ acquires momentum $p = E/c$ , hence kinetic energy

E_R \;=\; \frac{p^{2}}{2M} \;=\; \frac{E^{2}}{2Mc^{2}} .

Why: the photon is short by $E_R$ — the recoil took a cut. The nucleus is heavy, so the recoil is non-relativistic and $p^2/2M$ is exact enough.

Step 2. Put in the numbers for $^{57}$ Fe: $E = 14.41$ keV, $Mc^2 = 57 \times 931.5\ \mathrm{MeV} = 5.30\times10^{10}$ eV.

E_R = \frac{(1.441\times10^{4})^{2}}{2 \times 5.30\times10^{10}}\ \mathrm{eV} = \mathbf{1.96\times10^{-3}\ eV} .

Why: two millielectronvolts. It sounds like nothing. Hold that thought.

Step 3. Compare with the linewidth. Mean life $\tau = 141$ ns gives

\Gamma = \frac{\hbar}{\tau} = \frac{6.582\times10^{-16}\ \mathrm{eV\,s}}{1.411\times10^{-7}\ \mathrm{s}} = \mathbf{4.67\times10^{-9}\ eV} .

Why: and now the two millielectronvolts are enormous. $E_R$ is $4.2\times10^{5}$ linewidths.

Step 4. Close the trap. The emitter loses $E_R$ ; the absorber needs an extra $E_R$ for its own recoil. The lines are separated by

\frac{2E_R}{\Gamma} = \mathbf{8.4\times10^{5}} \quad\text{linewidths.}

Why: the overlap is not small — it is nil. A Lorentzian falls as $1/\Delta^2$ , so the overlap integral is suppressed by $\sim(\Gamma/2E_R)^2 \sim 10^{-12}$ . You could run the experiment for the age of the universe. Nuclear resonance fluorescence was a graveyard.

Step 5. Mössbauer's escape. In a crystal the nucleus is bound in a lattice with quantised vibrational modes. If the recoil energy falls below the lowest phonon, the lattice cannot accept it — there is no fractional phonon — and with probability $f$ (the Lamb–Mössbauer factor) the entire crystal recoils as one body. Then $M \to M_{\text{crystal}} \sim 10^{20}M_{\text{nucleus}}$ , and since $E_R \propto 1/M$ ,

E_R \;\longrightarrow\; \sim 10^{-23}\ \mathrm{eV} \;\lll\; \Gamma .

Why: this is the whole thing. The photon now carries the full transition energy, the two lines land on top of one another, and the natural width $\Gamma/E = 3.24\times10^{-13}$ becomes an instrument. Mössbauer found it in $^{191}$ Ir in 1958 and had the Nobel Prize three years later. Pound and Rebka's experiment is younger than the effect that makes it possible by eighteen months. *Verified in verify_aq-sr-gravitational-redshift.py: $E_R$ , $\Gamma$ , $\Gamma/E$ , $E_R/\Gamma$ and $2E_R/\Gamma$ are recomputed at fifty digits from $E = 14.41295$ keV, $M(^{57}\mathrm{Fe}) = 56.9354\ u$ and $t_{1/2} = 97.8$ ns, and each is grepped out of this file. The $1/M$ scaling is asserted by perturbing $M$ .*

Go deeper · Derivation B · 2 minThe temperature trap — the systematic and the signal are the same physics

Derivation BThe temperature trap — the systematic and the signal are the same physics

This is what makes Pound–Rebka more than an arithmetic exercise, and it is why this article exists separately from the prediction `aq-sr-rindler-metric` owns. The goal is to compute the shift produced by nothing more sinister than a warm source — and to discover it is the same size as gravity. An experimenter who did not know this would have published a number, the number would have been wrong, and it would have looked beautiful.

Step 1. A nucleus in a crystal vibrates even at the bottom of its well, so its emission frequency is Doppler-shifted. The first-order shift averages to zero — it goes as often one way as the other — but the second-order shift does not:

\frac{\Delta\nu}{\nu} \;=\; -\frac{\langle v^{2}\rangle}{2c^{2}} \;+\; O(v^4/c^4) .

Why: this is time dilation — the transverse Doppler shift, the $1/\gamma$ of special relativity expanded — and it is negative for any motion, because a moving clock always runs slow. The vibrating nucleus is a moving clock.

Step 2. Feed it equipartition. In the classical limit $\tfrac12 M\langle v^2\rangle = \tfrac32 k_BT$ , so $\langle v^2\rangle = 3k_BT/M$ and

\frac{\Delta\nu}{\nu} = -\frac{3k_BT}{2Mc^{2}} \qquad\Longrightarrow\qquad \boxed{\;\frac{\mathrm{d}z}{\mathrm{d}T} = -\frac{3k_B}{2Mc^{2}}\;}

Why: the constant contains nothing but $k_B$ , the nuclear mass and $c^2$ . It is not adjustable, not a property of the apparatus. Every Mössbauer source in the world has it.

Step 3. Evaluate it, and flinch.

\frac{\mathrm{d}z}{\mathrm{d}T} = -\frac{3 \times 1.381\times10^{-23}}{2 \times (9.465\times10^{-26})(8.988\times10^{16})}\ \mathrm{K^{-1}} = \mathbf{-2.44\times10^{-15}\ K^{-1}} .

Compare with the thing being measured: $gh/c^2 = +2.46\times10^{-15}$ .

\frac{|\mathrm{d}z/\mathrm{d}T|}{gh/c^{2}} \;=\; \mathbf{0.99} \qquad\Longrightarrow\qquad \Delta T = \mathbf{1.01\ K}\ \text{ counterfeits the entire result.}

Why: one degree. Not of absolute temperature control — of difference between the top of the tower and the bottom. A warm basement and a cold roof produce a signal indistinguishable from gravity, in the opposite direction. Pound and Rebka print the warning themselves: "A difference of 1 °C would produce a shift as large as that sought."

Step 4. What they did about it. They measured the coefficient in a companion paper (Phys. Rev. Lett. 4, 274, two weeks earlier), monitored both ends of the tower continuously, averaged the temperatures to $\mathbf{0.03\ ^\circ C}$ , and corrected every run — the correction column of their Table I runs $-5$ to $-11 \times10^{-15}$ , larger than the effect itself. For the theory they used the Debye heat capacity at $\theta_D = 420$ K, about $92\%$ of the classical limit at room temperature. Why: the number in this card is the classical (Dulong–Petit) limit, honest as such — do not attribute it to their paper. Theirs is the same physics with a better heat capacity, and the same conclusion: a degree of drift is fatal.

Step 5. Name the trap. The thermal shift is $-\langle v^2\rangle/2c^2$ ; the gravitational shift is $\Delta\Phi/c^2$ . Both are relativistic clock shifts — the two terms of the same expression, the same pair that appears, with the same relative sign, in the GPS budget (`aq-sr-equivalence-principle`: kinematic $-7.2$ , gravitational $+45.8$ , net $+38.6$ µs/day). Why: the noise in this experiment is not an imperfection of the apparatus. It is the other half of the theory being tested. That is the strongest possible argument that this article, and not the prediction, is where Pound–Rebka lives. *Verified in verify_aq-sr-gravitational-redshift.py: $\mathrm{d}z/\mathrm{d}T = -3k_B/2Mc^2$ is derived symbolically from $-\langle v^2\rangle/2c^2$ with $\langle v^2\rangle = 3k_BT/M$ ; the value $-2.437\times10^{-15}\ \mathrm{K^{-1}}$ is computed at fifty digits; the ratio to $gh/c^2$ is asserted to be $0.99$ and the crossover temperature to be $1.01$ K; and an artefact guard perturbs $M$ , $k_B$ and $c$ and asserts every one of them moves.*

Go deeper · Derivation C · 2 minThe two-way protocol, and why the published number is twice the textbook one

Derivation CThe two-way protocol, and why the published number is twice the textbook one

Almost every account of this experiment quotes $2.46\times10^{-15}$ and stops. Open the paper and that number is not in it. The goal here is to explain what Pound and Rebka actually published and why it is twice as large — because one-way and two-way are different quantities, and confusing them is the same error as confusing a summit's elevation with a flight path.

Step 1. State the problem the protocol solves. Source and absorber are two physical lumps of iron with their own intrinsic frequency offset $\delta_0$ — different temperatures, lattice environments, isomer shifts. A single measurement returns $\delta_0 + (\Delta\nu/\nu)_{\text{grav}}$ , and $\delta_0$ is not small: their data show a mean offset of $\sim-17.6\times10^{-15}$ , seven times the signal. Why: you cannot calibrate $\delta_0$ away — you cannot make two identical lumps of iron to one part in $10^{15}$ . It must be cancelled, not measured.

Step 2. Turn the apparatus over. Run with the source at the bottom (light climbs, redshift, $-gh/c^2$ ), then swap them so the source is at the top (light falls, blueshift, $+gh/c^2$ ). The intrinsic offset does not care which way up the tower is. Gravity does.

\text{bottom}: \;\delta_0 - \frac{gh}{c^{2}}, \qquad\qquad \text{top}: \;\delta_0 + \frac{gh}{c^{2}} .

Why: the oldest trick in metrology — modulate the signal, leave the systematic alone. Everything that fails to reverse with direction is now dead.

Step 3. Subtract.

\boxed{\;\Delta_{\text{two-way}} \;=\; \left(\text{top}\right) - \left(\text{bottom}\right) \;=\; \frac{2gh}{c^{2}}\;}

\frac{2gh}{c^{2}} = \frac{2 \times 9.80665 \times 22.5552}{8.98755\times10^{16}} = \mathbf{4.92\times10^{-15}} .

Why: the raw observable is twice the textbook number, and it is what the paper prints: "The shift observed agrees with $-4.92\times10^{-15}$ , the predicted gravitational shift for this 'two-way' height difference." Note that $4.92$ requires $h = 74$ feet $= 22.5552$ m. The popular $22.5$ m gives $4.91\times10^{-15}$ — a sign that a source has rounded the tower rather than read the paper.

Step 4. Read out the answer, from all data taken after they understood the temperature systematic:

\Delta_{\text{measured}} = -(5.13 \pm 0.51)\times10^{-15}, \qquad \frac{(\Delta\nu)_{\text{exp}}}{(\Delta\nu)_{\text{theor}}} = \mathbf{+1.05 \pm 0.10} .

Why: that ratio is the published result — agreement with the equivalence principle at the $10\%$ level. The plus sign, says the paper, "indicates that the frequency increases in falling, as expected". Pound and Snider repeated it over a $75$ -ft path, announcing in 1964 (Phys. Rev. Lett. 13, 539) and publishing in full in 1965 (Phys. Rev. 140, B788), with a ratio of $(0.9990 \pm 0.0076)$ — agreement to $0.8\%$ , again quoted against $2gh/c^2$ .

Step 5. Post the warning. Several databases list this paper's "abstract" as reporting $(-17.6 \pm 0.6)\times10^{-15}$ . That is not the gravitational result: it is the mean of the two orientations, which by Step 2 measures $\delta_0$ — and the paper says so, noting the mean "should measure an effective shift of other origin". Gravity is in the difference. (The paper carries no abstract at all; PRL did not print them in 1960.) Why: not pedantry. Anyone who looks the paper up and copies that number publishes a figure seven times too large with nothing to do with gravity — and it is sitting there, in a reputable index, waiting. *Verified in verify_aq-sr-gravitational-redshift.py: $2gh/c^2$ is recomputed at fifty digits for $h = 74$ ft and asserted to be $4.922\times10^{-15}$ , matching the paper's $4.92$ ; the $22.5$ m rounding is shown to give $4.91\times10^{-15}$ instead; the measured/predicted ratio $5.13/4.92 = 1.04$ and its uncertainty $0.51/4.92 = 0.10$ are checked against the published $1.05 \pm 0.10$ ; and the script asserts that $-17.6\times10^{-15}$ appears in this file ONLY inside a passage that names it as the non-gravitational mean.*

Figure

THE ONE-WAY / TWO-WAY LADDER, AND THE SYSTEMATIC SITTING ON TOP OF IT. Left: the four numbers that all describe this single experiment, and which are constantly confused with one another. The PREDICTED ONE-WAY shift gh/c² = 2.46 × 10⁻¹⁵ is the textbook number and appears NOWHERE in the paper. The PREDICTED TWO-WAY shift, 2gh/c² = 4.92 × 10⁻¹⁵, IS the paper's number, because source and absorber were swapped top-for-bottom and the observable is the DIFFERENCE — which cancels every shift that does not reverse when you turn the tower over. The MEASURED two-way value is −(5.13 ± 0.51) × 10⁻¹⁵, and the PUBLISHED HEADLINE is the ratio, 1.05 ± 0.10. Right: the enemy. The second-order Doppler shift from thermal vibration is −2.44 × 10⁻¹⁵ per kelvin — plot it against the gravitational shift and the two bars are the same height and point in opposite directions. One kelvin of temperature difference between the ends of the tower exactly cancels gravity. The dashed line marks the 0.03 °C to which Pound and Rebka held their tower, which is what bought them their 10% result. Note also the intrinsic source–absorber offset of −17.6 × 10⁻¹⁵ — SEVEN TIMES the signal — which some databases mistakenly list as the experiment's result

Limiting cases

Case	Result	Meaning
$M \to M_{\text{crystal}}$	$E_R \to 0$	the Mössbauer effect. $E_R \propto 1/M$ : a $10^{20}$ -fold heavier recoiling body kills the recoil dead
$M \to M_{\text{nucleus}}$	$2E_R/\Gamma = 8.4\times10^5$	free nuclei: lines $840{,}000$ linewidths apart. No resonance exists
$T \to 0$	zero-point residue	the classical formula fails; below $\theta_D$ the Debye heat capacity takes over and the shift saturates
$\Delta T = 1.01$ K	shift $= -gh/c^2$	the counterfeit. A one-degree gradient exactly cancels the signal
swap source/absorber	signal $\to 2gh/c^2$	gravity reverses, systematics do not. The two-way protocol
average the two	signal $\to \delta_0$	the mean measures the intrinsic offset ( $-17.6\times10^{-15}$ ), not gravity

Common trapBase defects repaired, and one live defect in canon

In this article's base. (i) The base dates the measurement to "1959"; the result was published 1 April 1960 (PRL 4, 337–341, received 9 March 1960). The 1959 Letter by the same authors (PRL 3, 439) is the feasibility study. (ii) The base prints the predicted shift as " $\approx 2.5\times10^{-15}$ " — that comes from $g = 9.8$ , $c = 3\times10^8$ ; with proper constants it is $2.46\times10^{-15}$ . (iii) The base's eq[1] gives $-gh/c^2$ and its eq[7] gives $+gh/c^2$ for the same quantity, four cards apart. It is negative for light going up. (iv) The base credits the $\sim1\%$ result to "a refined 1964 experiment"; 1964 is a Letter, and the 1965 Phys. Rev. paper is the same experiment in full.

In shipped canon. `aq-sr-rindler-metric` — which owns this prediction and is otherwise excellent — prints the shift as "about $0.035$ Hz on a $3.5\times10^{18}$ Hz gamma ray". That is wrong. $2.46\times10^{-15} \times 3.485\times10^{18} = \mathbf{8.6\ kHz}$ , not $0.035$ Hz — a factor of $2.5\times10^{5}$ . The physics around it is right and the conclusion unaffected; the number is a transcription error. This article prints 8.6 kHz, and the discrepancy is logged for the canon-fix queue.

Worked examples

1 · How far do you have to move the source to cancel gravity?

Step 1 — identify. A Mössbauer spectrometer tunes the line by moving the source at velocity $v$ , a first-order Doppler shift $\Delta\nu/\nu = v/c$ . To null a gravitational shift you need $v = gh/c$ .

Step 2 — apply. With $h = 74$ ft $= 22.5552$ m,

v = \frac{gh}{c} = \frac{9.80665 \times 22.5552}{2.99792458\times10^{8}} = \mathbf{7.38\times10^{-7}\ m/s} = \mathbf{0.74\ \mu m/s} .

Step 3 — the two-way version. The observable is the difference between orientations, so the scan must span twice that: $\mathbf{1.48\ \mu m/s}$ .

Step 4 — put it in context. The natural line is $2\Gamma = 0.194$ mm/s wide, so the entire gravitational effect is $0.74/194 = \mathbf{1/262}$ of it — and against the line they actually observed (Lorentzian, fractional FWHM $1.13\times10^{-12} = 0.339$ mm/s, $1.75\times$ the natural width) it is one part in 459.

Step 5 — the boundary number. They drove the source on a loudspeaker cone at $0.74$ micrometres per second and found the centre of a line four hundred and fifty times wider. That is the experiment — not the physics but the statistics, fitting a centroid to a fraction of a percent of its width over ten days. The whole result stands or falls on whether the line's centre stayed put for reasons other than gravity. Which is why the previous derivation is the important one.

2 · Could you do it in your building?

Step 1 — identify. Scale to a domestic tower: $h = 10$ m gives $gh/c^2 = 1.09\times10^{-15}$ .

Step 2 — the resolution demand. As a fraction of the natural linewidth ( $3.24\times10^{-13}$ ) that is $1/297$ : you must locate a line centroid to $0.3\%$ of its width.

Step 3 — the temperature demand. From $\mathrm{d}z/\mathrm{d}T = -2.44\times10^{-15}\ \mathrm{K^{-1}}$ , that same shift is produced by

\Delta T = \frac{1.09\times10^{-15}}{2.44\times10^{-15}} = \mathbf{0.45\ K} .

So to measure gravity to $10\%$ you must control the two ends of your building to $0.045$ K — not the absolute temperature, the difference, continuously, for the whole run.

Step 4 — the verdict. The physics is free; the thermometry is the experiment. That is why Pound and Rebka spent an entire companion paper on the temperature coefficient before they dared publish, and why they held the tower to $0.03\ ^\circ$ C.

Step 5 — the boundary number. $0.045$ K over $10$ m of building. A person standing next to the source raises its temperature by more than that. This experiment is famous not because gravity is subtle, but because everything else is louder.

Origins · 2 minMössbauer effect and how this was discovered

Mössbauer effect

1958–1961 · A graduate student in Heidelberg finds the resonance nobody could find · and Harvard weighs a photon eighteen months later

The problem. Nuclear resonance fluorescence — a nucleus emitting a gamma ray another identical nucleus can absorb — should have been the sharpest spectroscopy in existence. It was not: it did not work at all, because the recoil pushes the two lines nearly a million linewidths apart. Physicists spent years closing the gap by brute force — heating the source so Doppler broadening smeared the lines into partial overlap, or spinning it on a centrifuge. These worked feebly, and everyone knew they were fighting the problem rather than solving it. The discovery. Rudolf Mössbauer, a doctoral student in Heidelberg, was measuring the resonance in $^{191}$ Ir and cooled his source expecting the effect to weaken — thermal smearing was supposed to be what made the lines overlap at all. It got stronger. The explanation, published in 1958 (Z. Physik 151, 124), is that a nucleus in a cold lattice can emit without exciting a phonon at all: the recoil momentum goes to the entire crystal, $\sim10^{20}$ times heavier, so the recoil energy — which falls as $1/M$ — simply vanishes. The line snaps to its natural width. Nobel Prize 1961, three years later, at the age of 32. The consequence, immediately. Recoil-free emission handed physics a frequency standard good to $3\times10^{-13}$ , off the shelf, in a lump of iron. Pound and Rebka saw what it was for and had the gravitational redshift within eighteen months. The effect is not a technique that made a hard experiment easier — it is the reason the experiment exists, and that is why the portrait here is the effect rather than either of the men who used it.

History

1958

— Rudolf Mössbauer, in Heidelberg, discovers recoil-free nuclear resonance absorption in iridium-191 (Zeitschrift für Physik 151, 124). The nuclear line becomes a frequency standard good to three parts in ten trillion.

1959

— Pound and Rebka publish the feasibility study, "Gravitational Red-Shift in Nuclear Resonance", Phys. Rev. Lett. 3, 439. It is not the result.

1960

— Pound and Rebka, "Variation with Temperature of the Energy of Recoil-Free Gamma Rays from Solids", Phys. Rev. Lett. 4, 274, on 15 March: the second-order Doppler systematic, measured and understood, two weeks before the result depends on it. Josephson gives the theory of the same shift in the next issue, Phys. Rev. Lett. 4, 341.

1960

— Pound and Rebka, "Apparent Weight of Photons", Phys. Rev. Lett. 4, 337–341, on 1 April: the 74-foot Jefferson tower, the two-way protocol, and 1.05 ± 0.10 times the predicted shift.

1961

— Mössbauer receives the Nobel Prize in Physics, shared with Robert Hofstadter, three years after the discovery.

1964

— Pound and Snider announce the improved measurement, Phys. Rev. Lett. 13, 539.

1965

— Pound and Snider publish it in full, "Effect of Gravity on Gamma Radiation", Phys. Rev. 140, B788: a 75-foot path, a ratio of 0.9990 ± 0.0076 against 2gh/c², agreement to 0.8 per cent.

1976

— Vessot and Levine fly a hydrogen maser to 10,000 km and confirm the redshift to 7 parts in 100,000 (Phys. Rev. Lett. 45, 2081, 1980) — a thousandfold improvement on the tower.

Going further

Texts. Pound & Rebka, Phys. Rev. Lett. 4, 337–341 (1960) — read it; it is four pages · Pound & Rebka, Phys. Rev. Lett. 4, 274 (1960), the temperature shift · Josephson, Phys. Rev. Lett. 4, 341 (1960) · Pound & Snider, Phys. Rev. 140, B788 (1965) · Mössbauer, Z. Physik 151, 124 (1958) · Vessot et al., Phys. Rev. Lett. 45, 2081 (1980) · Bothwell et al., Nature 602, 420 (2022) · Frauenfelder, The Mössbauer Effect (1962).

Provenance. Signature $(+,-,-,-)$ . Every number is recomputed in verify_aq-sr-gravitational-redshift.py, which reads this file, regexes the printed values out and asserts them at fifty digits, with artefact guards on $g$ , $c$ , $M$ and $k_B$ . Base defects repaired: the date (1959 → 1960); the predicted shift ( $2.5 \to 2.46\times10^{-15}$ ); the sign contradiction between eq[1] and eq[7]; the Pound–Snider citation (1964 Letter, 1965 full paper, one experiment). Live canon defect logged: `aq-sr-rindler-metric` prints "0.035 Hz" for the shift on the $14.4$ keV line; the correct figure is 8.6 kHz.

Scope. The prediction $gh/c^2$ , the exact clock-rate law, the crest-counting argument, and the demonstration that this experiment tested equivalence and not curvature, are all owned by `aq-sr-rindler-metric` — recalled by name, never re-derived. The three routes to the prediction are owned by `aq-sr-gravitational-time-dilation-heuristic`. The nuclear physics of the Mössbauer transition belongs to the atomic/nuclear track; derived here is only the recoil energy and the linewidth, which is what the redshift measurement needs.

Builds on. The Equivalence Principle · Gravitational Time Dilation from the Equivalence Principle · Relativistic Uniform Acceleration. Leads to. The Rindler Metric · Tidal Forces and the Limits of the Equivalence Principle · The Unruh Effect.

If the Mössbauer effect makes the recoil vanish, where does the momentum go?

Into the crystal, all of it. Momentum is exactly conserved — the recoil momentum $p = E/c$ is real and the lattice takes it as a whole. What vanishes is the recoil energy, because kinetic energy at fixed momentum is $p^2/2M$ : it falls as $1/M$ while the momentum stays put. A crystal of $10^{20}$ nuclei absorbs the same momentum with $10^{20}$ times less energy — $10^{-23}$ eV, twenty orders of magnitude below the linewidth. The nucleus does not fail to recoil. It recoils attached to $10^{20}$ friends, and the energy cost of that is nothing. The quantum ingredient: the lattice takes energy only in phonon-sized lumps, so if the would-be recoil energy is below the lowest phonon, it must take zero.

Why is everything in this experiment quoted in millimetres per second?

Because that is how the instrument works: to scan a nuclear line you Doppler-tune the source by physically moving it, so the natural unit is a velocity. A fractional shift $z$ corresponds to a source speed $v = zc$ , so $\Gamma/E = 3.24\times10^{-13}$ becomes $0.097$ mm/s and the full width $2\Gamma = 0.194$ mm/s — the number every Mössbauer spectroscopist has memorised. On the same scale the entire gravitational redshift over a 74-foot tower is $0.74$ micrometres per second. This is a mechanical instrument being asked to resolve one part in $10^{15}$ by driving a loudspeaker cone very, very slowly.

Did Pound and Rebka test general relativity?

No, and `aq-sr-rindler-metric` makes the point precisely: the uniform-field prediction and Earth's real $1/r^2$ field differ, over 22.6 m, by $8.7\times10^{-21}$ — about $2.9\times10^{4}$ times smaller than their error bar. There is no possibility whatever that they distinguished a real gravitational field from a uniformly accelerating rocket. What they tested is the equivalence principle, which `aq-sr-gravitational-time-dilation-heuristic` shows is forced by energy conservation alone. It is not a test of curvature — and curvature is what separates general relativity from special relativity plus equivalence. The curvature tests are light bending, Shapiro delay and perihelion precession, and they belong to the general-relativity track.

How can the "noise" and the "signal" be the same physics?

Because both are relativistic clock shifts, and relativity does not care whether a clock is slowed by sitting in a well or by moving. The gravitational shift is $\Delta\Phi/c^2$ ; the thermal shift is $-\langle v^2\rangle/2c^2$ . These are precisely the two terms of the relativistic clock-rate formula — the same two that appear, with the same relative sign, in the GPS budget ( $+45.8$ and $-7.2$ µs/day). Pound and Rebka's nuclei sat in a gravitational potential and jiggled at a few hundred metres per second, so both terms were present — and at 300 K they come out the same size. That is a coincidence of scale, not of principle, but it is a devastating one, and it is why the experiment needed a companion paper about heat capacity before it could say anything about gravity.

Is the −17.6 × 10⁻¹⁵ figure in the databases simply wrong?

The number is real; the label is wrong, which is worse. $-17.6\times10^{-15}$ is what Table I calls the "mean shift" — the average of the source-at-top and source-at-bottom runs. Averaging the two orientations cancels the gravitational term (which reverses) and leaves everything that does not: the intrinsic offset between that particular source and that particular absorber. The paper is explicit that the mean "should measure an effective shift of other origin". So it is a genuine measurement of a genuine quantity — just not gravity, and seven times bigger. If you find it presented as the result, the presenter has taken the wrong row out of the table.

Why is the portrait card the Mössbauer effect rather than Pound or Rebka?

Because without the effect there is no experiment — not a harder experiment, no experiment. The recoil calculation is not an obstacle to be engineered around: it puts the two lines $8.4\times10^{5}$ linewidths apart, and no cleverness with towers and loudspeakers closes a gap like that. Mössbauer's discovery created a frequency standard sharp enough to see $10^{-15}$ , and Pound and Rebka used it within eighteen months. (There is a housekeeping reason too: the concept Gravitational redshift is already the portrait of `aq-sr-rindler-metric`, which owns the prediction.)