Is the moon there when nobody looks? Reality and the

Is the moon therewhen nobody looks?Reality and the quantum theoryEinstein maintained that quantum metaphysics entails spooky actionsat a distance; experiments have now shown that what bothered Einsteinis not a debatable point but the observed behavior of the real world.

N. David Mermin

Quantum mechanics is magic1

In May 1935, Albert Einstein, BorisPodolsky and Nathan Rosen published2

an argument that quantum mechanicsfails to provide a complete descriptionof physical reality. Today, 50 yearslater, the EPR paper and the theoreti-cal and experimental work it inspiredremain remarkable for the vivid illus-tration they provide of one of the mostbizarre aspects of the world revealed tous by the quantum theory.

Einstein's talent for saying memora-ble things did him a disservice when hedeclared "God does not play dice," for ithas been held ever since that the basisfor his opposition to quantum mechan-ics was the claim that a fundamentalunderstanding of the world can only bestatistical. But the EPR paper, hismost powerful attack on the quantumtheory, focuses on quite a differentaspect: the doctrine that physical prop-erties have in general no objectivereality independent of the act of obser-vation. As Pascual Jordan put it3

Observations not only disturbwhat has to be measured, theyproduce it. . . . We compel [the elec-tron] to assume a definite posi-tion. . . . We ourselves produce theresults of measurement.

Jordan's statement is something of atruism for contemporary physicists.Underlying it, we have all beentaught, is the disruption of what isbeing measured by the act of measure-ment, made unavoidable by the exis-tence of the quantum of action, whichgenerally makes it impossible even inprinciple to construct probes that canyield the information classical intu-

38 PHYSICS TODAY / APRIL 1985

ition expects to be there.Einstein didn't like this. He wanted

things out there to have properties,whether or not they were measured4:

We often discussed his notions onobjective reality. I recall that dur-ing one walk Einstein suddenlystopped, turned to me and askedwhether I really believed that themoon exists only when I look at it.The EPR paper describes a situation

ingeniously contrived to force the quan-tum theory into asserting that proper-ties in a space-time region B are theresult of an act of measurement inanother space-time region A, so farfrom B that there is no possibility of themeasurement in A exerting an influ-ence on region B by any known dynami-cal mechanism. Under these condi-tions, Einstein maintained that theproperties in A must have existed allalong.

Spooky actions at a distanceMany of his simplest and most explic-

it statements of this position can befound in Einstein's correspondencewith Max Born.5 Throughout the book(which sometimes reads like a Nabokovnovel), Born, pained by Einstein's dis-taste for the statistical character of thequantum theory, repeatedly fails, bothin his letters and in his later commen-tary on the correspondence, to under-stand what is really bothering Ein-stein. Einstein tries over and overagain, without success, to make himselfclear. In March 1948, for example, hewrites:

That which really exists in Bshould . . . not depend on whatkind of measurement is carried out

in part of space A; it should also beindependent of whether or not anymeasurement at all is carried outin space A. If one adheres to thisprogram, one can hardly considerthe quantum-theoretical descrip-tion as a complete representationof the physically real. If one triesto do so in spite of this, one has toassume that the physically real inB suffers a sudden change as aresult of a measurement in A. Myinstinct for physics bristles at this.

Or, in March 1947,I cannot seriously believe in [thequantum theory] because it cannotbe reconciled with the idea thatphysics should represent a realityin time and space, free from spookyactions at a distance.The "spooky actions at a distance"

(spukhafte Fernwirkungen) are the ac-quisition of a definite value of a proper-ty by the system in region B by virtue ofthe measurement carried out in regionA. The EPR paper presents a wave-function that describes two correlatedparticles, localized in regions A and B,far apart. In this particular two-parti-cle state one can learn (in the sense ofbeing able to predict with certainty the

David Mermin is director of the Laboratory ofAtomic and Solid State Physics at CornellUniversity. A solid-state theorist, he hasrecently come up with some quasithoughtsabout quasicrystals. He is known to PHYSICSTODAY readers as the person who made"boojum" an internationally accepted scienti-fic term. With N. W. Ashcroft, he is about tostart updating the world's funniest solid-statephysics text. He says he is bothered by Bell'stheorem, but may have rocks in his headanyway.

0031-9228 / 85 / 0400 38-10/$0,1.Q0 © 1985American Institute of Physios

An EPR apparatus. The experimental setup consists of two detectors, A and B, and a source of something ("particles"or whatever) C. To start a run, the experimenter pushes the button on C; something passes from C to both detectors.Shortly after the button is pushed each detector flashes one of its lights. Putting a brick between the source and one ofthe detectors prevents that detector from flashing, and moving the detectors farther away from the source increases thedelay between when the button is pushed and when the lights flash. The switch settings on the detectors vary randomlyfrom one run to another. Note that there are no connections between the three parts of the apparatus, other than viawhatever it is that passes from C to A and B. The photo below shows a realization of such an experiment in thelaboratory of Alain Aspect in Orsay, France. In the center of the lab is a vacuum chamber where individual calciumatoms are excited by the two lasers visible in the picture. The re-emitted photons travel 6 meters through the pipes tobe detected by a two-channel polarizer. Figure 1

PHYSICS TODAY / APRIL 1985 39

The result of a run. Shortly after theexperimenter pushed the button on thesource in figure 1, the detectors flash onelamp each. The experimenter records theswitch settings and the colors of the lampsand then repeats the experiment.Here, for example, the record reads32RG—the switches are in positions 3 and2 and the lamps flashed R and G,respectively. Figure 2

result of a subsequent measurement)either the position or the momentum ofthe particle in region B as a result ofmeasuring the corresponding propertyof the particle in region A. If "thatwhich really exists" in region B doesnot depend on what kind of measure-ment is carried out in region A, then theparticle in region B must have had botha definite position and a definite mo-mentum all along.

Because the quantum theory is in-trinsically incapable of assigning val-ues to both quantities at once, it mustprovide an incomplete description ofthe physically real. Unless, of course,one asserts that it is only by virtue ofthe position (or momentum) measure-ment in A that the particle in Bacquires its position (or momentum):spooky actions at a distance.

At a dramatic moment Pauli appearsin the Born-Einstein Letters, writingBorn from Princeton in 1954 with hisfamous tact on display:

Einstein gave me your manuscriptto read; he was not at all annoyedwith you, but only said you were aperson who will not listen. Thisagrees with the impression I haveformed myself insofar as I wasunable to recognize Einstein when-ever you talked about him in eitheryour letter or your manuscript. Itseemed to me as if you had erectedsome dummy Einstein for yourself,which you then knocked down withgreat pomp. In particular, Ein-stein does not consider the conceptof "determinism" to be as funda-mental as it is frequently held to be(as he told me emphatically manytimes).... In the same way, hedisputes that he uses as criterionfor the admissibility of a theory the

question: "Is it rigorously deter-ministic?"

Pauli goes on to state the real nature ofEinstein's "philosophical prejudice" toBorn, emphasizing that "Einstein'spoint of departure is 'realistic' ratherthan 'deterministic'." According toPauli the proper grounds for challeng-ing Einstein's view are simply that

One should no more rack one'sbrain about the problem ofwhether something one cannotknow anything about exists all thesame, than about the ancient ques-tion of how many angels are able tosit on the point of a needle. But itseems to me that Einstein's ques-tions are ultimately always of thiskind.Faced with spooky actions at a dis-

tance, Einstein preferred to believethat things one cannot know anythingabout (such as the momentum of aparticle with a definite position) doexist all the same. In April 1948 hewrote to Born:

Those physicists who regard thedescriptive methods of quantummechanics as definitive in princi-ple would . . . drop the requirementfor the independent existence ofthe physical reality present indifferent parts of space; they wouldbe justified in pointing out that thequantum theory nowhere makesexplicit use of this requirement. Iadmit this, but would point out:when I consider the physical phe-nomena known to me, and espe-cially those which are being sosuccessfully encompassed by quan-tum mechanics, I still cannot findany fact anywhere which wouldmake it appear likely that [the]requirement will have to be aban-

doned. I am therefore inclined tobelieve that the description ofquantum mechanics . . . has to beregarded as an incomplete andindirect description of reality

A fact is foundThe theoretical answer to this chal-

lenge to provide "any fact anywhere"was given in 1964 by John S. Bell, in afamous paper6 in the short-lived jour-nal Physics. Using a gedanken experi-ment invented7 by David Bohm, inwhich "properties one cannot knowanything about" (the simultaneousvalues of the spin of a particle alongseveral distinct directions) are requiredto exist by the EPR line of reasoning,Bell showed ("Bell's theorem") that thenonexistence of these properties is adirect consequence of the quantitativenumerical predictions of the quantumtheory. The conclusion is quite inde-pendent of whether or not one believesthat the quantum theory offers a com-plete description of physical reality. Ifthe data in such an experiment are inagreement with the numerical predic-tions of the quantum theory, thenEinstein's philosophical position has tobe wrong.

In the last few years, in a beautifulseries of experiments, Alain Aspect andhis collaborators at the University ofParis's Institute of Theoretical andApplied Optics in Orsay provided8 theexperimental answer to Einstein'schallenge by performing a version ofthe EPR experiment under conditionsin which Bell's type of analysis applied.They showed that the quantum-theore-tic predictions were indeed obeyed.Thirty years after Einstein's challenge,a fact—not a metaphysical doctrine—was provided to refute him.


Attitudes toward this particular 50-year sequence of intellectual historyand scientific discovery vary widely.9From the very start Bohr certainly tookit seriously. Leon Rosenfeld describes10

the impact of the EPR argument:This onslaught came down upon usas a bolt from the blue. Its effecton Bohr was remarkable. . . . Anew worry could not have come ata less propitious time. Yet, as soonas Bohr had heard my report ofEinstein's argument, everythingelse was abandoned.Bell's contribution has become cele-

brated in what might be called semi-popular culture. We read, for example,in The Dancing Wu Li Masters that11

Some physicists are convinced that[Bell's theorem] is the most impor-tant single work, perhaps, in thehistory of physics.

And indeed, Henry Stapp, a particletheorist at Berkeley, writes that12

Bell's theorem is the most pro-found discovery of science.At the other end of the spectrum,

Abraham Pais, in his recent biographyof Einstein, writes13 of the EPR arti-cle—that "bolt from the blue," thebasis for "the most profound discoveryof science":

The only part of this article whichwill ultimately survive, I believe,is . . . a phrase ["No reasonable de-finition of reality could be expectedto permit this"] which so poignant-ly summarizes Einstein's views onquantum mechanics in his lateryears.I think it is fair to say that more

physicists would side with Pais thanwith Stapp, but between the majorityposition of near indifference and theminority position of wild extravagance

.31 rCK33 QQ33 RR188R33 GQeiQR81 RRS2RR33 BB11QQ23 RR32GR12BR12 RB116631 RB12 R613GR226612 RG12GR2266236R33 RR336631 R631 RR33 RR32 R631 R611RR236R12GG11GG13RG31 RG23GR316R23 R622 RR126R32 6R22 RR126633 RR11RR2366236633RR236R2166136R3366116612RR1266316G32RG216R226622RR13RR2166

22 6G11 RR33 6G31GR126621GR33 G621 RR12BR22 RR13 RB12 R623 66116613 RG21 RG33 RR32 6R3266336621 RR12 R6226611G623 6R22 RR116632 6R13RG13GR23 GG33 RR31 GR13RG23 RR12GR31 R632 RG216R226622 RR13RR21G6236R226622663166136R216R33RR23RR22 RR12RR23 R623 RG326R31RG22GG116611GG21RG11RR12RQ

33 66216R136R236R22 RR11 RR216R21 RR23 GG32GR33 RR33 GG33 GG23GR21 GR12 RR32GR32 6R33 GG31 RG13 RR13 RG32 R6316R23 RR33 RR136R11GG31 6R31 RG136R23 R6316623 RG21 RR23 RG11662266116611BG21 RB11RR12 RG83 SR32 BRS1B621R613R613R613R613BR23RB22 BB11RR31RB83RR83R611RR32RG32BR13BB238R32BR

Data produced by the apparatus of figure1. This is a fragment of an enormous setof data generated by many, many runs:Each entry shows the switch settings andthe colors of the lights that flashed for arun. The switch settings are changedrandomly from run to run. Figure 3

is an attitude I would characterize asbalanced. This was expressed to memost succinctly by a distinguishedPrinceton physicist on the occasion ofmy asking how he thought Einsteinwould have reacted to Bell's theorem.He said that Einstein would have gonehome and thought about it hard forseveral weeks—that he couldn't guesswhat he would then have said, exceptthat it would have been extremelyinteresting. He was sure that Einsteinwould have been very bothered byBell's theorem. Then he added,

Anybody who's not bothered byBell's theorem has to have rocks inhis head.To this moderate point of view I

would only add the observation thatcontemporary physicists come in twovarieties. Type 1 physicists are both-ered by EPR and Bell's theorem. Type2 (the majority) are not, but one has todistinguish two subvarieties. Type 2aphysicists explain why they are notbothered. Their explanations tend ei-ther to miss the point entirely (likeBom's to Einstein) or to contain phys-ical assertions that can be shown to befalse. Type 2b are not bothered andrefuse to explain why. Their position isunassailable. (There is a variant oftype 2b who say that Bohr straightenedout14 the whole business, but refuse toexplain how.)

A gedanken demonstrationTo enable you to test which category

you belong to, I shall describe, in black-box terms, a very simple version ofBell's gedanken experiment, deferringto the very end any reference whatevereither to the underlying mechanismthat makes the gadget work or to thequantum-theoretic analysis that ac-


counts for the data. Perhaps thisbackwards way of proceeding will makeit easier for you to lay aside yourquantum theoretic prejudices and de-cide afresh whether what I describe isor is not strange.15

What I have in mind is a simplegedanken demonstration. The appara-tus comes in three pieces. Two of them(A and B) function as detectors. Theyare far apart from each other (in theanalogous Aspect experiments over 10meters apart). Each detector has aswitch that can be set to one of threepositions; each detector responds to anevent by flashing either a red light or agreen one. The third piece (C), midwaybetween A and B, functions as a source.(See figure 1.)

There are no connections betweenthe pieces—no mechanical connec-tions, no electromagnetic connections,nor any other known kinds of relevantconnections. (I promise that when youlearn what is inside the black boxes youwill agree that there are no connec-tions.) The detectors are thus incapa-ble of signaling to each other or to thesource via any known mechanism, andwith the exception of the "particles"described below, the source has no wayof signaling to the detectors. Thedemonstration proceeds as follows:

The switch of each detector is inde-pendently and randomly set to one ofits three positions, and a button ispushed on the source; a little after that,each detector flashes either red orgreen. The setting of the switches andthe colors that flash are recorded, andthen the whole thing is repeated overand over again.

The data consist of a pair of numbersand a pair of colors for each run. A run,for example, in which A was set to 3, Bwas set to 2, A flashed red, and B flashedgreen, would be recorded as "32RG," asshown in figure 2.

Because there are no built-in connec-tions between the source C and thedetectors A and B, the link between the


Switches set the same: the data of figure3, but highlighted to pick out those runs in

which both detectors had the sameswitch settings as they flashed. Note that

in such runs the lights always flash thesame colors. Figure 4

pressing of the button and the flashingof the light on a detector can only beprovided by the passage of something(which we shall call a "particle,"though you can call it anything youlike) between the source and thatdetector. This can easily be tested; forexample, by putting a brick betweenthe source and a detector. In subse-quent runs, that detector will not flash.When the brick is removed, everythingworks as before.

Typical data from a large number ofruns are shown in figure 3. There arejust two relevant features:• If one examines only those runs inwhich the switches have the samesetting (figure 4), then one finds thatthe lights always flash the same colors.• If one examines all runs, withoutany regard to how the switches are set(figure 5), then one finds that thepattern of flashing is completely ran-dom. In particular, half the time thelights flash the same colors, and halfthe time different colors.

That is all there is to the gedankendemonstration.

Should you be bothered by these dataunless you have rocks in your head?How could it work?

Consider only those runs in whichthe switches had the same setting whenthe particles went through the detec-tors. In all such runs the detectorsflash the same colors. If they couldcommunicate, it would be child's playto make the detectors flash the samecolors when their switches had thesame setting, but they are completelyunconnected. Nor can they have beenpreprogrammed always to flash thesame colors, regardless of what is goingon, because the detectors are observedto flash different colors in at least someof those runs in which their switchesare differently set, and the switchsettings are independent randomevents.

How, then, are we to account for the

31RR33 QQ33 RR1£GR33 QQ£1GR21RR22 RR33 GGUGG£3RR3£GR1£GR1£RGUGG31 RG1£RG133R22 QQ1£RG1£GR22 QQ£3GR33RR33 QQ31RG31RR33RR3£RG31RGURR£3GR1£GG11GB13RG31RG£3GR31GR£3RG22RR1£GR3£GR22RR1£GG33RRURR£3GG£3GG33RR£3GR£1GG13GR33GGUGG1£RR1£GG31GG3£RG£1GR2&GGS2RR13RR£1GG

22QQURR33QQ31GR1£GG£1GR33QQ£1 RR1£GR22RR13RG1£RG£3GG11GB13RG£1 RG33RR32GR3£GG33GQ£1RR1£RG22 QQUSB£3GR22 RR11BG3£GR13RG13GR£3GG33RR31GR13RG£3RR1£GR31RG3£RG£1GR22QG22RR13RR£1GG£3GR22QG22QQ31GG13GR£1GR33RR£3RR22RR1£RR£3RG£3RG3£GR31RG22GQUGGUGG£1RGURR1£RG

33 QQ£1GR13GR£3GR22 RRURR£1GR£1 RR£3GG3£GR33RR33 GG33 GG£3GR£1GR1£RR3£GR3£GR33 GQ31RG13RR13RG3£RG31GR£3RR33 RR13GRUGG31GR31RG13GR£3RG31GG£3RG£1RR£3RGUGG22QQUGGUGG£1RGURR1£RG£3GR3£GR£1GG£1RG13RG13RG13RG13GR£3RG22GGURR31RG£3RR£3RGURR3£RG3£GR1366£3GR3£GR

Switches set any way: the data of figure3, but highlighted to emphasize only the

colors of the lights that flashed in each run,no matter how the switches were set whenthe lights flashed. Note that the pattern of

colors is completely random. Figure 5

first feature of the data? No problem atall. Born, in fact, in a letter of May1948, offers5 such an explanation toEinstein:,

It seems to me that your axiom ofthe "independence of spatiallyseparated objects A and B" is notas convincing as you make out. Itdoes not take into account thefact of coherence; objects farapart in space which have a com-mon origin need not be indepen-dent. I believe that this cannot bedenied and simply has to be ac-cepted. Dirac has based his wholebook on this.

In our case the detectors are triggeredby particles that have a common originat the source C. It is then easy to dreamup any number of explanations for thefirst feature of the data.

Suppose, for example, that what eachparticle encounters as it enters itsdetector is a target (figure 6) dividedinto eight regions, labeled RRR, RRG,RGR, RGG, GRR, GRG, GGR, andGGG. Suppose each detector is wiredso that if a particle lands in the GRGbin, the detector flips into a mode inwhich the light flashes G if the switch isset to 1, R if it is set to 2, and G if it is setto 3; RGG leads to a mode with R for 1and G for 2 and 3, and so on. We canthen easily account for the fact that thelights always flash the same colorswhen the switches have the samesettings by assuming that in each runthe source always fires its particles intobins with the same labels.

Evidently this is not the only way.One could imagine that particles comein eight varieties: cubes, spheres, te-trahedra,.... All settings produce Rwhen a cube is detected, a sphereresults in R for settings 1 and 2, G forsetting 3, and so forth. The firstfeature of the data is then accountedfor if the two particles produced by thesource in each run are always both ofthe same variety.

Common to all such explanations is

31RR33QQ33RR1£GR33BQ£1QR£1RR££RR33BB1166£3RR3£GR1£6R1£R611663 IRQ1£R6136R££661£R61£QR££66£38R33RR336631R631RR33RR3£RB3 IRQ11RR£36R1£66116613R631R6£36R316R£3R6££RR1£6R3£6R££RR1£6633RR11RR£366£36633RR£36R£166136R336611661£RR1£6631663£R6£16R££66££RR13RR£1BB

££6611RR3366316R1£66£16R3366£1RR1£6R££RR13R61£R6£366116613R6£1R6

3£6R3£663366£1RR1£R6££661166£36R££RR11663£6R13R6136R£36633RR316R13R6£3RR1£6R31R63£R6£16R££66££RR13RR£166£36R££66££663166136R£16R33RRci3RR££RR1£RR£3R6£3R63£6R31R6££661 1661166£1R811RR1£R6=• T ra D

33BB£1BR13BR£3QR££RR11RR£1QR£1RR£3Q63£6R

33663366£36R£16R1£RR3£6R3£6R336631R613RR13R63£R8316R£3RR33RR136R1166316R31R6136R£3R63166£3R6£1RR£3R61166££6611661166£1R611RR1£R6£36R3£6R£166£1R613R613R613R6136R£3R6££6611RR31R6£3RR£3R611RR3£R33£QR13BQ£3QR3£QR::DD

the requirement that each particleshould, in one way or another, carry toits detector a set of instructions for howit is to flash for each of the threepossible switch settings, and that in anyrun of the experiment both particlesshould carry the same instruction sets:• A set of instructions that covers eachof the three possible settings is requiredbecause there is no communicationbetween the source and the detectorsother than the particles themselves. Inruns in which the switches have thesame setting, the particles cannot knowwhether that setting will be 11, 22, or33. For the detectors always to flashthe same colors when the switches havethe same setting, the particles mustcarry instructions that specify colorsfor each of the three possibilities.• The absence of communicationbetween source and detectors also re-quires that the particles carry suchinstruction sets in every run of theexperiment—even those in which theswitches end up with different set-tings—because the particles alwayshave to be prepared: Any run mayturn out to be one in which the switchesend up with the same settings.This generic explanation is picturedschematically in figure 7.

Alas, this explanation—the only one,I maintain, that someone not steeped inquantum mechanics will ever be able tocome up with (though it is an entertain-ing game to challenge people to try)—isuntenable. It is inconsistent with thesecond feature of the data: There is noconceivable way to assign such instruc-tion sets to the particles from one runto the next that can account for the factthat in all runs taken together, withoutregard to how the switches are set, thesame colors flash half the time.

Pause to note that we are about toshow that "something one cannot knowanything about"—the third entry in aninstruction set—cannot exist. For evenif instruction sets did exist, one couldnever learn more than two of the three


Model of a detector to produce data likethose in figure 4. Particles from the sourcefall with equal probability into any of theeight bins; for each bin the color flasheddepends on the switch as indicated on theback of the box. Figure 6

entries (revealed in those runs wherethe switches ended up with two differ-ent settings). Here is the argument.

Consider a particular instruction set,for example, RRG. Should both parti-cles be issued the instruction set RRG,then the detectors will flash the samecolors when the switches are set to 11,22, 33, 12, or 21; they will flashdifferent colors for 13, 31, 23, or 32.Because the switches at each detectorare set randomly and independently,each of these nine cases is equallylikely, so the instruction set RRG willresult in the same colors flashing % ofthe time.

Evidently the same conclusion holdsfor the sets RGR, GRR, GGR, GRG andRGG, because the argument uses onlythe fact that one color appears twiceand the other once. All six suchinstructions sets also result in the samecolors flashing % of the time.

But the only instruction sets left areRRR and GGG, and these each result inthe same colors flashing all of the time.

Therefore if instruction sets exist,the same colors will flash in at least %of all the runs, regardless of how theinstruction sets are distributed fromone run of the demonstration to thenext. This is Bell's theorem (alsoknown as Bell's inequality) for thegedanken demonstration.

But in the actual gedanken demon-stration the same colors flash only \

the time. The data described aboveviolate this Bell's inequality, and there-fore there can be no instruction sets.

If you don't already know how thetrick is done, may I urge you, beforereading how the gedanken demonstra-tion works, to try to invent some otherexplanation for the first feature of thedata that does not introduce connec-tions between the three parts of theapparatus or prove to be incompatiblewith the second feature.

One way to do itHere is one way to make such a

device:Let the source produce two particles of

spin V2 in the singlet state, flying aparttoward the two detectors. (Granted, thisis not all that easy to do, but in the Orsayexperiments described below, the sameeffect is achieved with correlated pho-tons.) Each detector contains a Stern-Gerlach magnet, oriented along one ofthree directions (a(1), a(2), or a(3)), perpen-dicular to the line of flight of theparticles, and separated by 120°, asindicated in figure 8. The three settingsof the switch determine which orienta-tion is used. The light on one detectorflashes red or green, depending onwhether the particle is deflected towardthe north (spin up) or south (spin down)pole of the magnet as it passes betweenthem; the other detector uses the oppo-site color convention.

That's it. Clearly there are no con-nections between the source and thedetectors or between the two detectors.We can nevertheless account for thedata as follows:

When the switches have the samesetting, the spins of both particles aremeasured along the same direction, sothe lights will always flash the samecolors if the measurements along thesame direction always yield oppositevalues. But this is an immediate conse-quence of the structure of the spinsinglet state, which has the form

(1)

independent of the direction of the spinquantization axis, and therefore yields+ — or — 4- with equal probability,

but never + + or — — , whenever thetwo spins are measured along anycommon direction.

To establish the second feature of thedata, note that the product mlm2 of theresults of the two spin measurements(each of which can have the values+ V2 or - V2) will have the value - XU

when the lights flash the same colorsand + V4 when they flash differentcolors. We must therefore show thatthe product vanishes when averagedover all the nine distinct pairs oforientations the two Stern-Gerlachmagnets can have. For a given pair oforientations, a<0 and a ° \ the mean


' 2 3

R6R

Instruction sets. To guarantee that the detectors of figure 6 flash the same colorwhen the switches are set the same, the two particles must in one way or anothercarry instruction sets specifying how their detectors are to flash for each possibleswitch setting. The results of any one run reveal nothing about the instructionsbeyond the actual data; so in this case, for example, the first instruction (1R) is"something one cannot know anything about," and I've only guessed at it,assuming that "it exists all the same." Figure 7

value of this product is just the expecta-tion value in the state ip of the corre-sponding product of (commuting) her-

iti b b l ( 0 (1)g p ( g

mitian observables a(0 • S(1) andiJ) • S(2). Thus the second feature of&

the data requires:

(2)

But equation 2 is an immediate conse-quence of the linearity of quantummechanics, which lets one take thesums inside the matrix element, andthe fact that the three unit vectorsaround an equilateral triangle sum tozero:

=2Ja(J) = 0 (3)

This completely accounts for thedata. It also unmasks the gedankendemonstration as a simple embellish-ment of Bohm's version of the EPRexperiment. If we kept only runs inwhich the switches had the same set-ting, we would have precisely theBohm-EPR experiment. The assertionthat instruction sets exist is then bla-tant quantum-theoretic nonsense, for itamounts to the insistence that eachparticle has stamped on it in advancethe outcome of the measurements ofthree different spin components corre-sponding to noncommuting observablesS-a('\ t = l,2, 3. According to EPR,this is merely a limitation of thequantum-theoretic formalism, because

instruction sets are the only way toaccount for the first feature of the data.

Bell's analysis adds to the discussionthose runs in which the switches havedifferent settings, extracts the secondfeature of the data as a further elemen-tary prediction of quantum mechanics,and demonstrates that any set of dataexhibiting this feature is incompatiblewith the existence of the instructionsets apparently required by the firstfeature, quite independently of theformalism used to explain the data, andquite independently of any doctrines ofquantum theology.

The experimentsThe experiments of Aspect and his

colleagues at Orsay confirm that thequantum-theoretic predictions for thisexperiment are in fact realized, andthat the conditions for observing theresults of the experiment can in fact beachieved. (A distinguished colleagueonce told me that the answer to theEPR paradox was that correlations inthe singlet state could never be main-tained over macroscopic distances—that anything, even the passage of acosmic ray in the next room, woulddisrupt the correlations enough to des-troy the effect.)

In these experiments the two spin-V2particles are replaced by a pair ofphotons and the spin measurementsbecome polarization measurements.

The photon pairs are emitted by cal-cium atoms in a radiative cascade aftersuitable pumping by lasers. Becausethe initial and final atomic states haveJ=0, quantum theory predicts (andexperiment confirms) that the photonswill be found to have the same polariza-tions (lights flashing the same colors inthe analogous gedanken experiment) ifthey are measured along the samedirection—feature number 1. But ifthe polarizations are measured at 120°angles, then theory predicts (and exper-iment confirms) that they will be thesame only a quarter of the time[V4 = cos2( 120°)]. This is precisely whatis needed to produce the statistics offeature number 2 of the gedankendemonstration: The randomly setswitches end up with the same setting(same polarizations measured) % of thetime, so in all runs the same colors willflash V3xl +2/3X(V4) =V2 the time.

The people in Orsay were interestedin a somewhat modified version ofBell's argument in which the angles ofgreatest interest were multiples of22.5°, but they collected data for manydifferent angles, and, except for EPRspecialists, the conceptual differencesbetween the two cases are minor.16

There are some remarkable featuresto these experiments. The two polar-ization analyzers were placed as far as13 meters apart without producing anynoticeable change in the results, there-


by closing the loophole that the strangequantum correlations might somehowdiminish as the distance between re-gions A and B grew to macroscopicproportions. At such separations it ishard to imagine that a polarizationmeasurement of photon # 1 could, inany ordinary sense of the term, "dis-turb" photon #2. Indeed, at theselarge separations, a hypothetical dis-turbance originating when one photonpassed through its analyzer could onlyreach the other analyzer in time toaffect the outcome of the second polar-ization measurement if it traveled at asuperluminal velocity.

In the third paper of the Orsaygroup's series, bizarre conspiracy the-ories are dealt a blow by an ingeniousmechanism for rapidly switching thedirections along which the polariza-tions of each photon are measured.Each photon passes to its detectorthrough a volume of water that sup-ports an ultrasonic standing wave.Depending on the instantaneous ampli-tude of the wave, the photon eitherpasses directly into a polarizer with oneorientation or is Bragg reflected intoanother with a different orientation.The standing waves that determine thechoice of orientation at each detectorare independently driven and havefrequencies so high that several cyclestake place during the light travel time

from one detector to the other. (Thiscorresponds to a refinement of thegedanken demonstration in which, tobe absolutely safe, the switches are notgiven their random settings until afterthe particles have departed from theircommon source.)

What does it mean?What is one to make of all this? Are

there "spooky actions at a distance"?A few years ago I received the text of aletter from the executive director of aCalifornia thinktank to the Under-secretary of Defense for Research andEngineering, alerting him to the EPRcorrelations:

If in fact we can control the faster-than-light nonlocal effect, it wouldbe possible... to make an untap-pable and unjammable command-control-communication system atvery high bit rates for use in thesubmarine fleet. The importantpoint is that since there is noordinary electromagnetic signallinking the encoder with the de-coder in such a hypothetical sys-tem, there is nothing for the enemyto tap or jam. The enemy wouldhave to have actual possession ofthe "black box" decoder to inter-cept the message, whose reliabilitywould not depend on separationfrom the encoder nor on ocean or

weather conditions....Heady stuff indeed! But just what is

this nonlocal effect? Using the lan-guage of the gedanken demonstration,let us talk about the 'W-color" of aparticle (iVcan be 1, 2, or 3) as the color(red or green) of the light that flasheswhen the particle passes through adetector with its switch set to N.Because instruction sets cannot exist,we know that a particle cannot at thesame time carry a definite 1-color, 2-color and 3-color to its detector. On theother hand, for any particular N(say 3),we can determine the 3-color of theparticle heading for detector A before itgets there by arranging things so thatthe other particle first reaches detectorB, where its 3-color is measured. If theparticle at B was 3-colored red, theparticle at A will turn out to be 3-colored red, and green at B means greenat A.

Three questions now arise:• Did the particle at A have its 3-colorprior to the measurement of the 3-colorof the particle at B? The answer cannotbe yes, because, prior to the measure-ment of the 3-color at B, it is altogetherpossible that the roll of the dice at B orthe whim of the B-operator will resultin the 2-color or the 1-color beingmeasured at B instead. Barring themost paranoid of conspiracy theories,"prior to the measurement of the 3-


A realization of the detector to produce the data of figure 3. Theparticles have a magnetic moment and can be separated into "spinup" and "spin down" particles by the Stern-Gerlach magnet insidethe detector. Setting the switch to positions 1, 2, or 3 rotates thenorth pole of the magnet along the coplanar unit vectors a(1), a(2), ora(31, separated by 120°. The vector sum of the three unit vectors is, ofcourse, zero. The switch positions on the two detectors correspondto the same orientations of the magnetic field. One detector flashesred for spin up, green for spin down; the other uses the opposite colorconvention. Figure 8

color at B" is indistinguishable from"prior to the measurement of the 2- (or1-) color at B." If the 3-color alreadyexisted, so also must the 2- and 1-colorshave existed. But instruction sets(which consist of a specification of the1-, 2-, and 3-colors) do not exist.• Is the particle at A 3-colored red afterthe measurement at B shows the colorred? The answer is surely yes, becauseunder these circumstances it is invaria-bly a particle that will cause thedetector at A to flash red.• Was something (the value of its 3-color) transmitted to the particle at A asa result of the measurement at B?

Orthodox quantum metaphysicianswould, I believe, say no, nothing haschanged at A as the result of themeasurement at B; what has changed isour knowledge of the particle at A.(Somewhat more spookily, they mightobject to the naive classical assumptionof localizability or separability implicitin the phrases "at A" and "at B.") Thisseems very sensible and very reassur-ing: iV-color does not characterize theparticle at all, but only what we knowabout the particle. But does that lastsentence sound as good when "particle"is changed to "photon" and' W-color" to"polarization"? And does it really helpyou to stop wondering why the lightsalways flash the same colors when theswitches have the same settings?

What is clear is that if there is spookyaction at a distance, then, like otherspooks, it is absolutely useless exceptfor its effect, benign or otherwise, onour state of mind. For the statisticalpattern of red and green flashes atdetector A is entirely random, howeverthe switch is set at detector B. Whetherthe particles arriving at A all come withdefinite 3-colors (because the switch atB was stuck at 3) or definite 2-colors(because the switch was stuck at 2) orno colors at all (because there was abrick in front of the detector at B)—allthis has absolutely no effect on thestatistical distribution of colors ob-served at A. The manifestation of this"action at a distance" is revealed onlythrough a comparison of the dataindependently gathered at A and at B.

This is a most curious state of affairs,and while it is wrong to suggest thatEPR correlations will replace sonar, itseems to me something is lost byignoring them or shrugging them off.The EPR experiment is as close tomagic as any physical phenomenon Iknow of, and magic should be enjoyed.Whether there is physics to be learnedby pondering it is less clear. The mostelegant answer I have found17 to thislast question comes from one of thegreat philosophers of our time, whoseview of the matter I have taken theliberty of quoting in the form of thepoetry it surely is:We have always had a great deal of difficultyunderstanding the world viewthat quantum mechanics represents.At least I do,because I'm an old enough manthat I haven't got to the pointthat this stuff is obvious to me.Okay, I still get nervous with it. .. .You know how it always is,every new idea,it takes a generation or twountil it becomes obviousthat there's no real problem.I cannot define the real problem,therefore I suspect there's no real problem,but I'm not surethere's no real problem.

Nobody in the 50 years since Ein-stein, Podolsky and Rosen has ever putit better than that.

Some of the views expressed above weredeveloped in the course of occasional techni-cal studies of EPR correlations supported bythe National Science Foundation undergrant No. DMR 83-14625.

References1. Daniel Greenberger, discussion remarks

at the Symposium on FundamentalQuestions in Quantum Mechanics,SUNY, Albany, April 1984.

2. A. Einstein, B. Podolsky, N. Rosen,Phys. Rev. 47, 777 (1935).

3. Quoted by M. Jammer, The Philosophyof Quantum Mechanics, Wiley, NewYork (1974) p. 151.

4. A. Pais, Rev. Mod. Phys. 51, 863 (1979).5. The Born-Einstein Letters, with com-

ments by M. Born, Walker, New York(1971).

6. J. S. Bell, Physics 1, 195 (1964).7. D. Bohm, Quantum Theory, Prentice-

Hall, Englewood Cliffs, N.J. (1951) pp.614-619.

8. A. Aspect, P. Grangier, G. Roger, Phys.Rev. Lett. 47, 460 (1981). A. Aspect, P.Grangier, G. Roger, Phys. Rev. Lett. 49,91 (1982). A. Aspect, J. Dalibard, G.Roger, Phys. Rev. Lett. 49, 1804 (1982).

9. For a discussion of the views of today'sphysicists toward the meaning of thequantum theory, see the interesting andprovocative essay "Cognitive Repressionin Contemporary Physics" by E. F. Kel-ler, Am. J. Phys. 47, 718 (1977).

10. L. Rosenfeld in Niels Bohr, His Life andWork as Seen by His Friends and Collea-gues, S. Rozental, ed., North Holland,Amsterdam (1967) pp. 114-36.

11. G. Zukav, The Dancing Wu-Li Masters—An Overview of the New Physics, Mor-row, New York (1979) p. 282. On thesame page it is also said that "Bell'stheorem is a mathematical constructwhich as such is indecipherable to thenon-mathematician," a view that I hopethe rest of this article will dispel.

12. H. Stapp, Nuovo Cimento 40B, 191(1977).

13. A. Pais, "Subtle is the Lord..." TheScience and the Life of Albert Einstein,Oxford U. P., New York (1982) p. 456.

14. N. Bohr, Phys. Rev. 48, 696 (1935).15. What follows is a somewhat refined ver-

sion of an argument I published a fewyears ago in Am. J. Phys. 49, 940 (1981),incorporating some improvements sug-gested by Richard Friedberg. For otherelementary treatments see J. S. Bell'sbeautiful essay, "Bertlemann's Socksand the Nature of Reality," J. Phys.(Paris) 42, C2-41 (1981), B. d'Espagnat'sarticle in the November 1979 ScientificAmerican, or d'Espagnat's recent book,In Search of Reality, Springer-Verlag,New York (1983).

16. For a survey of other attempts to realizethe EPR experiment, and the variants ofBell's original argument used to inter-pret experimental tests, see J. F.Clauser, A. Shimony, Repts. Prog. Phys.41, 1881 (1978).

17. R. P. Feynman, Int. J. Theor. Phys. 21,471 (1982). •


Documents

Is the moon there when nobody looks? Reality and the