1919 eclipse test of general relativity investigation

01 · The briefing

Begin with the question.

General relativity predicted that starlight passing the edge of the Sun would be bent by about 1.75 arcseconds, twice the amount predicted by Newtonian gravity. The practical test in 1919 was to photograph stars close to the Sun during a total solar eclipse and compare their positions with plates taken when the Sun was absent. Frank Watson Dyson sent Charles Davidson and Andrew Crommelin to Sobral in northern Brazil, and Arthur Eddington went to the island of Príncipe off West Africa. On 29 May 1919 both expeditions observed the eclipse. Heavy clouds left few usable images at Príncipe, while at Sobral the main instrument lost focus. Afterwards the plates had to be measured and reduced, and that stage involved choices. Three outcomes were possible: no deflection, a Newtonian half-deflection, or the full general relativistic value. The investigation separates three questions: what the instruments recorded, which data the analysts chose to use, and whether the later charge of selection is well founded. Each has a different kind of evidence. Earman and Glymour note that the expedition report made Einstein an international celebrity and drove scientific opinion in his favor, so the stakes of the analysis were high.

A tall stone column topped by a statue stands in a paved plaza in front of a blue-and-white church with a bell tower, beside a white modern building under a clear blue sky. — Where it happenedPhotographed 2014 The Praça do Patrocínio in Sobral, Ceará, photographed in 2014 as a later view of the square. Photo: GLandovsky · CC BY-SA 4.0 · Wikimedia Commons.

The timeline

1911
Einstein's first paper predicts that light passing near the Sun is deflected.
1919-03
The British expeditions leave for Sobral in Brazil and for Príncipe in West Africa.
1919-05-29
The total eclipse is observed: clouds at Príncipe thin only enough for a few usable images, and the Sobral astrographic lens is out of focus.
1919-10-03
Eddington writes to Dyson, glad that the 4-inch-lens plates give the full deflection but worried about his own Príncipe plates.
1920
Dyson, Eddington and Davidson publish their full report in the Philosophical Transactions of the Royal Society.
1979
The 1919 Sobral plates are remeasured with modern equipment; their 4-inch-lens result is close to the 1919 value.
2009-03
Daniel Kennefick argues in Physics Today that the 1919 team had reasonable grounds for its conclusions.

A researched reconstruction for learning. These reasoning steps simplify a complex record; no score can capture the full history.

02 · The evidence

What does the record show?

Black-and-white image of a dark disc of the Sun surrounded by a bright corona, with a bright prominence at upper right and star labels in the constellation Taurus. — On the evidence table1919 A photograph of the 29 May 1919 eclipse from the Dyson, Eddington and Davidson expeditions, digitised and released by ESO in 2019 with star labels added. Photo: ESO / Landessternwarte Heidelberg-Königstuhl · CC BY 4.0 · Wikimedia Commons.

Read each exhibit before settling on an explanation. The source links below every exhibit lead to the record behind it.

Exhibit 01

A small shift and a careful method

General relativity predicted a shift of about 1.75 arcseconds at the Sun's limb, twice the Newtonian value. Because the shift is so small, the 1919 teams compared eclipse photographs of stars near the Sun with plates of the same star fields taken when the Sun was absent. At Sobral the team waited about two months for the star field to return to the same height in the sky.

Source record s1doi.org

Exhibit 02

Clouds at Príncipe

At Príncipe heavy clouds on the eclipse day thinned enough for Eddington to obtain images of the brightest stars only on the last few exposures. The team could not take comparison plates on site, so comparison exposures were made in Britain before departure. Those conditions limited the checks that could be made at the station.

Source record s1doi.org

Exhibit 03

The astrographic lens and the choice of data

At Sobral the astrographic lens had lost focus during the eclipse, and its streaky images were hard to measure. Kennefick reports that Dyson's team measured only one coordinate on the astrographic plates and obtained 0.93 arcseconds, and he argues that the decision to rely on the 4-inch-lens exposures was Dyson's, made once the two instruments were found to disagree.

Source record s1doi.org

Exhibit 04

The 1979 reanalysis

Kennefick's table compares the 1979 remeasurement with the 1919 results: 1.90 plus or minus 0.11 arcseconds for the 4-inch lens, against 1.98 plus or minus 0.18 in 1919, and 1.55 plus or minus 0.34 for the astrographic lens, against 0.93 in 1919. The reanalysis broadly supports the 4-inch conclusion and suggests the astrographic data were hard to reduce in 1919. It does not show that the original choices were free of judgment.

Source record s1doi.org

Exhibit 05

The selection charge

The Pesquisa FAPESP feature reports that some critics argued the measurements were not accurate enough to decide the question, while others contended that Eddington deliberately discarded Sobral data that supported Newton. Luiz Nunes de Oliveira says there is no evidence the data were fudged. Kennefick argues that Eddington was not at Sobral and had no hand in analyzing its data, which Dyson's team handled.

Source record s2revistapesquisa.fapesp.br

03 · Your reasoning

Follow the strongest explanation.

Cut-out portrait of Arthur Stanley Eddington. — In the roomArthur Stanley EddingtonCambridge astronomer who led the Príncipe expedition and reduced his own plates.1900 Photo: Bain News Service (Library of Congress) · Public domain · Wikimedia Commons.

Choose an interpretation, then compare it with the evidence. You can revise your choice. The goal is understanding, not a perfect score.

Reasoning step 01 of 04

Which feature of the 1919 observations most directly made the results hard to interpret?The eclipse occurred in the morning at one site and near midday at the other.The timing mattered for comparison plates, but it does not explain the disagreement between instruments.The two instruments at Sobral gave measurements that disagreed sharply once the plates were reduced.The disagreement forced a choice about which data to use, which is where judgment entered the result.The results were published only after the 1979 reanalysis.The results were published in 1920, and the reanalysis came much later, so this is false.

Evidence-based explanation

The Sobral instruments disagreed once reduced, so the analysis had to choose between them. The astrographic images were streaky and out of focus, which made the disagreement hard to resolve on the evidence alone.

Weigh it against

Exhibit 03The astrographic lens and the choice of data

Reasoning step 02 of 04

Which conclusion from the 1979 comparison is best supported?The 1919 astrographic figure differs from the 1979 value, while the 4-inch-lens figure agrees closely with its modern remeasurement.This is what the comparison shows; it does not show which theory the astrographic data were meant to support.Both 1919 figures agree with the 1979 values, so the choice between them did not matter.The comparison shows that the astrographic figures differ, so the choice did matter.The 1979 reanalysis showed that Eddington chose the Newtonian value.The reanalysis concerns the measurement of the plates and does not support this claim.

Evidence-based explanation

The 1979 remeasurement supports the 4-inch figure and not the 1919 astrographic figure, which fits the view that the astrographic data were hard to reduce. The reanalysis is of the same plates, so it cannot answer every question about the 1919 method.

Weigh it against

Exhibit 04The 1979 reanalysis

Reasoning step 03 of 04

What is the most useful question to ask when weighing the charge that Eddington discarded Sobral data?Whether Einstein approved of the expedition's conclusions.Approval by a theorist does not bear on how the data were reduced.Whether the result was popular with the public in 1919.Public reception does not show whether data were chosen to fit a prediction.Who chose which plates to reduce, and what reasons were recorded for that choice.The charge concerns decisions in the analysis, so records of those decisions are the relevant evidence.

Evidence-based explanation

The selection charge concerns who decided what to use and why. Kennefick argues that Dyson made the decision at Greenwich and that Eddington was not part of the Sobral analysis; critics dispute that reading. The question turns on records of decisions, not on the outcome alone.

Weigh it against

Exhibit 03The astrographic lens and the choice of data Exhibit 05The selection charge

Reasoning step 04 of 04

Which conclusion is best supported by the evidence as a whole?The 1919 plates showed no deflection at all, and the expeditions reported otherwise.The sources describe measured deflections, so this misreads the observations.The 1919 results were fabricated to support Einstein, and no later analysis has checked them.The sources report no evidence of fabrication, and the 1979 reanalysis did check the plates.The 1919 measurements were too imprecise to decide between Newton and Einstein, and no later analysis changed that.This is a partial truth: the measurements were uncertain, but the 1979 reanalysis changed the picture.The 1919 results were consistent with general relativity within large uncertainties, a 1979 reanalysis broadly supports the Sobral 4-inch result, and the selection question remains debated.This keeps the measurements, the reanalysis and the historical dispute separate, which is what the sources support.

Evidence-based explanation

The sources support a qualified conclusion. The measurements were consistent with general relativity within large uncertainties, a later reanalysis broadly supports the 4-inch result, and the selection charge is still a question rather than a settled finding. Claims of fabrication have no support in these sources.

Weigh it against

Exhibit 01A small shift and a careful method Exhibit 04The 1979 reanalysis Exhibit 05The selection charge

04 · The findings

The chain behind the outcome.

General relativity predicted a shift of about 1.75 arcseconds for starlight grazing the Sun, twice the Newtonian figure. The 1919 expeditions had to decide from photographic plates whether the displacement was nearer one prediction or the other, and the margins were small. The Príncipe observations were limited by cloud, and comparison plates had to be taken in Britain before departure. The Sobral observations were complicated by an astrographic lens that had lost focus during the eclipse. Once the plates were reduced, the astrographic and 4-inch-lens results disagreed sharply, so the analysis had to choose between them. Kennefick reports that Dyson, at Greenwich, decided to rely on the 4-inch-lens exposures, and that Eddington was not part of the Sobral analysis. Critics have argued that Eddington discarded Sobral data that supported Newton. Defenders reply that there is no evidence the data were fudged, and that the analysis was done by others. The 1979 reanalysis of the Sobral plates with modern measuring equipment gave a 4-inch-lens value close to the 1919 figure, while the astrographic value moved from 0.93 to 1.55 arcseconds. That supports the 4-inch conclusion and suggests the astrographic data were hard to reduce, but it does not remove every question about the method. The best-supported account is narrower than either a triumph or a fraud. The measurements were difficult, the reduction involved decisions, and the result was consistent with general relativity within large uncertainties. The debate matters because such decisions can be examined in records, and because a famous result can be shaped by the stories told about it.

Contributing causes and conditions

Weather at Príncipe, which left few usable images and forced comparison plates to be taken in Britain.
Instrument problems at Sobral: an astrographic lens that lost focus and a backup 4-inch lens.
Reduction decisions after the instruments disagreed, including measuring only one coordinate on the astrographic plates and choosing the 4-inch data.
Expectations on both sides, which critics argue may have shaped choices: Eddington was an enthusiast for the theory, and Dyson was moderately skeptical of it.

Outcome1919 eclipse test of general relativity

Take the method with you

A very small predicted effect depends on the method and on how the measurements are reduced, not only on the photograph.
When a selection is alleged, look for records of the decision and of who made it before judging motive.
A reanalysis can support one result while leaving other choices open, so check which parts it actually tests.

You can read findings at any time. Exploring every reasoning step marks this case as complete in your notebook.

Open another investigation

Starlight bent by the Sun