26 June 1964, Volume 144, Number 3626

Reversals of the Earth'sMagnetic Field

Recent paleomagnetic and geochronologic data provideinformation on time and frequency of field reversals.

Allan Cox, Richard R. Doell, G. Brent Dalrymple

The idea that the earth's main mag-netic field changes its polarity wasfirst advanced during the early decadesof this century by geophysicists whowere investigating the remanent mag-netization of volcanic rocks and bakedearth (1-3). They found that thesematerials, when heated to their Curietemperatures and cooled in the presentweak field of the earth, acquire a weakbut extremely stable remanent magneti-zation parallel to the present directionof the geomagnetic field. They furtherfound that bricks and baked earthfrom archeological sites, as well asrock samples from young lava flows,possess a similar natural magnetization,undoubtedly acquired when the bakedearth and lava flows originally cooled.The direction of this magnetizationwas found to be close to that of thepresent geomagnetic field at the sam-pling site, with differences of the orderof 10 degrees attributable to secularvariation of the geomagnetic field.However, when the early paleomagnet-ists studied rocks of early Pleistoceneage or older, they discovered that asubstantial proportion were magnetizedin a direction more nearly 180 degreesfrom that of the present field. In ex-planation of these reversed remanentmagnetizations they proposed 1 80-de-

The authors are on the staff of the U.S.Geological Survey, Menlo Park, Calif.

26 JUNE 1964

SCIENCE

ment in techniques of determining thestability and reliability of rock magnet-ism for determining past geomagneticfield directions, and partly as a resultof the vast increase in paleomagneticdata available. Although many of thesepaleomagnetic studies have focused onthe problems of continental drift andpolar wandering, interest in the rever-sal problem has also continued up tothe present, and the observations ofthe early workers have been amplyconfirmed. The directions of remanentmagnetization in many tens of thou-sands of rock samples, with ages in therange 0 to 30 million years, show astrikingly bimodal distribution; the di-rections are grouped about either thepresent geomagnetic field directions atthe sampling sites or about antiparallel'(reversed) directions. Remarkably fewintermediate directions have been ob-served.

Solid-state physicists have also madeimportant contributions to the study ofreversals. In 1950 Graham (5) con-cluded that reversals he had observedin sediments were due not to reversalsof the earth's field but, rather, to miner-alogically controlled self-reversal alongthe lines earlier reported by Smith,Dee, and Mayneord (6). Informed ofthis result, Neel (7) was able to pre-dict, on the basis of contemporary ad-vances in ferromagnetism and ferri-magnetism, that rock-forming mineralsmight acquire remanent magnetizationsat an angle 180 degrees from the di-rection of the ambient magnetic fieldin which they cool. This prediction wasalmost immediately confirmed by Naga-ta and Uyeda's (8) discovery of a vol-canic rock from Japan that possessedthis predicted property of self-reversal.As we discuss more fully later on, thisdiscovery of mineralogically controlledself-reversal, in offering a possible al-ternative explanation for reversed mag-netization, considerably clouded the ex-perimental evidence for geomagneticfield reversal.

What, then, is the present state ofknowledge about geomagnetic field re-versals? Our conclusion, based upon

1537

gree reversals of the direction of thegeomagnetic field.

During the two decades (1928 to1948) that followed this early workthere was surprisingly little scientificreaction to the hypothesis of geomag-netic field reversal. This silence reflectsthe embarrassing lack, even at so latea date, of a theory adequate to ac-count for the present geomagneticfield, let alone reversed magnetic fieldswhich may or may not have existedearlier in the earth's history. The prob-lem of field reversals was taken upagain around 1950 by scientists whobecame involved in problems of geo-magnetism from theoretical as well asobservational starting points. Thetheory of magnetohydrodynamic mo-tions in an incompressible fluid wasinterpreted to show how magnetic fieldsmay be generated in rotating spheresof electrically conducting fluid. It wasfurther shown that reversals in polarityare a possible, if not a necessary,property of simplified dynamo modelsfor the magnetohydrodynamic motions.Moreover, conditions necessary for themaintenance of such a magnetic sys-tem were shown to be consistent withconditions thought to exist in the earth'score (4).The observational basis for field re-

versals has been greatly extended since1950, partly as a result of improve-

on

Janu

ary

8, 2

012

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

recent paleomagnetic-geochronometricevidence presented in this article, isthat after a period of considerable un-certainty the balance has finally tippedin favor of geomagnetic field reversalsas the cause of the reversals in mostrocks. Recent experimental evidencenot only confirms this interpretationbut also tells us exactly when someof the most recent switches in polarityoccurred.

Early Discoveries

Some of the most compelling evi-dence favoring the field-reversal hy-pothesis appears in the earliest investi-gations. In 1906 Brunhes (1) made theexperimental observations that are sum-marized in Fig. 1. Faced with theseresults, and the experimentally verifiablefact that bricks and other baked earthsacquire remanent magnetization parallelto the ambient magnetic field in whichthey cool, Brunhes concluded that thereversed natural remanent magnetiza-tion had been acquired in a reversedmagnetic field which existed during theepoch when the lava flow originallycooled.

Although Brunhes's classic work wasdone decades prior to the discovery ofmineralogically controlled self-reversals,his results from baked sediments havea direct bearing on this problem. Whenlava flows and other igneous bodiesare emplaced, the sediments or otherrocks in contact with them are heatedand also commonly are altered chemi-cally. The ferromagnetic minerals in theigneous body are generally differentfrom those in the surrounding bakedcountry rock. As the igneous body andthe baked rocks cool in the same ambi-ent geomagnetic field, they becomemagnetized in the same direction un-less magnetization of the ferromagneticminerals in one of them is self-revers-

NORTH

ing. If the fraction of self-reversingferromagnetic minerals in all rocksis x and if these self-reversing min-erals are randomly distributed be-tween igneous and baked rocks, thenthe fraction of pairs of igneous rockand baked rock with opposing polaritieswill be 2(x - x2). In a recent reviewof paleomagnetic studies of igneousand baked rocks, Wilson (9) found50 pairs with the same polarity andthree pairs for which the original paleo-magnetic data were ambiguous andwhich may or may not have opposingpolarities. On the basis of these data,it appears that, at most, 3 percent ofthe reversals in igneous rocks andbaked sediments are due to mineralogi-cally controlled self-reversals.The next important step in the study

of reversals came with the stratigraphicinvestigations of Mercanton and Ma-tuyama (3), in which an attemptwas made to delineate the times whenthe geomagnetic field was reversed.Matuyama's study of volcanic rocksfrom Japan and Korea is especially im-portant because it was the first suc-cessful attempt to determine the ageof the most recent switch from areversed-polarity epoch to the presentepoch of normal polarity. He foundthat "there is a group of specimenswhose directions of magnetization fallsaround the present earth's field. A num-ber of other specimens forms anothergroup almost antipodal to the former.[Although] the ages of the collectedbasalts are not always clearly known,. . . we may consider that in theearlier part of the Quaternary Periodthe earth's magnetic field in the areaunder consideration was probably in thestate represented by the second group,which gradually changed to the staterepresented by the first group." This-result, like that of Brunhes, has beenamply confirmed by subsequent investi-gations.

Worldwide Extent of Reversals

If we accept for the moment thehypothesis that most reversely mag-netized rocks were formed in reversedmagnetic fields, the question arises,Were such reversed fields anomalies re-stricted to the paleomagnetic samplingsites or were they worldwide phenom-ena? A theory of field reversals dueto local concentrations of ferromag-netic minerals in the earth's crust maybe eliminated at the outset on severalgrounds, one of them magnitude: themagnetizations of rock formations aregenerally too small by a factor of 10to produce a reversed field. From thetheoretical point of view, if field re-versals occur at all, they should beworldwide events. This follows fromthe very nature of the earth's magneticfield. Since the time of Gauss we haveknown from spherical harmonic analy-sis of the field that its source lies withinthe earth, and from physical reasoningit is equally clear that this source iswithin the earth's fluid core. The onlyproposed physical mechanism for pro-ducing the field which at present ap-pears tenable is the magnetohydrody-namic mechanism, in which generationof the field is attributed to the inter-action between fluid motions and elec-trical currents in the core.The shape of the geomagnetic field

is approximately that of a dipole atthe earth's center, the direction of thefield observed at any locality today dif-fering from that of the dipole field by5 to 25 degrees. The worldwide fit tothe dipole-field configuration is vastlyimproved if average rather than instan-taneous field directions at each localityare considered. With paleomagnetictechniques it is possible to obtain thedirection of the earth's field at a givenlocality averaged over thousands ofyears, and when this is done it is foundthat young rocks from all over the

SOUTH

Present geomagneticfield in France

TRM of brick cooledin present field

/ ormal flow

s D ~ ~~IA\or baked zone

NRM of bricks from NRM of Tertiaryarcheologicdl sites sediments bake

/ reversed flow

reversed baked zone

lava flows anded by them

Fig. 1. Brunhes's observations of experimentally produced thermoremanent magnetization (TRM) and of natural remanent mag-netization (NRM), which led him to propose geomagnetic field reversals.

1538 SCIENCE, VOL. 144

on

Janu

ary

8, 2

012

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

world fit a dipole-field configurationwith the dipole axis parallel to theearth's present axis of rotation (10).Two such rotationally symmetrical di-pole configurations are possible, onenormal (Fig. 2, a) and one reversed(Fig. 2, b). Only two processes ap-pear capable of explaining why theremanent magnetizations of rocks fromall over the world are tightly groupedabout these two field configurations.Either 180-degree self-reversals con-trolled mineralogically are responsible,or else the fluid and current systemsin the earth's core have changed insuch a way as to produce 180-degreereversals in the direction of the maindipolar component of the earth's mag-netic field.

Self-Reversal

Self-reversal is defined as the ac-quisition of thermoremanent magneti-zation in a direction 180 degrees fromthat of the ambient magnetic field inwhich a rock or mineral specimencools. Knowing as we do that somerocks, such as the Haruna dacite (8),are reproducibly self-reversing in thelaboratory, the crucial question iswhether all reversely magnetized rockshave undergone self-reversal. If theyhave, the experimental basis for seek-ing evidence of geomagnetic field re-versal vanishes, and the hypothesismust be discarded. If not, we are facedwith the difficult task of sorting outreversals due to mineralogy from thosedue to reversals of the geomagneticfield.The obvious experiment might seem

to be to heat and cool rocks in knownmagnetic fields and determine whetherthese rocks are self-reversing. This ex-periment has been performed by nu-merous investigators on thousands ofrocks from many localities; in the lab-oratory, only a small fraction of 1 per-cent of the rocks have been found tobe self-reversing. However these experi-ments do not tell the entire story orcompletely rule out the possibility ofself-reversal. The inconclusiveness ofsimple heating and cooling experimentsis illustrated in the studies by Uyeda(8) and Ishikawa and Syono (11) onminerals of the ilmeno-hematite series.Minerals of this series were found tobe self-reversing only over a limitedrange of chemical composition and onlyfor certain rates of cooling. Thus, a26 JUNE 1964

GEOGRAPHIC GEOGRAPHICNORTH NORTH

I IIIf t(a) NORMAL

GEOMAGNETIC FIELD(b) REVERSED

GEOMAGNETIC FIELD

Fig. 2. Directions of an axial dipole geo-magnetic field over the earth for (a) nor-mal polarity (present configuration) and(b) reversed polarity.

natural self-reversal in a rock contain-ing these minerals might not be re-produced if the rate of cooling in thelaboratory were different from thatwhen the rock originally formed.

Self-reversal requires the coexistenceand interaction of two ferromagneticconstituents. Of the many types ofinteraction that may produce self-re-versal, the simplest involves the mag-netostatic energy of interaction be-tween two closely intergrown ferro-magnetic minerals, A and B, havingdifferent Curie temperatures. If thegeometry of the mineral intergrowthsis appropriate, and if the intensity ofthe magnetization of the A constituentis sufficiently large when the rockhas cooled to the Curie temperatureof the B constituent (the Curie tem-perature of A being greater than thatof B), B will become reversely mag-netized. The rock will undergo repro-ducible self-reversal if, after completecooling, the remanent magnetization ofB is greater than that of A. Even ifthe remanent magnetization of B isinitially less than that of A, the rockmay still become self-reversed if A isselectively dissolved during the geologichistory of the rock, or if the remanentmagnetization of A undergoes spon-taneous decay to a value less than thatof B.The two constituents, A and B, need

not be different ferromagnetic minerals.In ferrites they are the two interpene-trating cubic sublattices which make upthe crystal structure. Because the ex-change interaction between the cationsites in the two sublattices is negative,the B sublattice acquires a spontane-ous magnetization exactly antiparallelto that of the A sublattice when theferrite cools through its Curie tempera-ture. Given appropriate temperature co-

efficients for the two spontaneous mag-netizations, reproducible self-reversalwill occur on further cooling. Self-reversing minerals of this type havebeen synthesized (12) but have notbeen discovered in rocks.

Ferrites may also undergo self-re-versal through cation migration be-tween adjacent lattice sites. At temper-atures above the Curie temperature theequilibrium distribution of cations onlattice sites is disordered, but at lowtemperatures it is more highly or-dered. On rapid quenching, however,such as may occur when a lava flowcools, a ferrite could temporarily re-tain a disordered cation distributionand later undergo a self-reversal throughslow diffusion of cations toward themore ordered low-temperature distribu-tion. Verhoogen (13) has shown thata self-reversal of this type would nottake place in magnetites containing im-purities or vacancies on certain latticesites until 10' or 106 years after cooling.These self-reversals would not be re-producible in the laboratory.

In view of the complexity and non-reproducibility of many types of self-reversal, what approaches are open tothe investigator attempting to assess theprevalence of self-reversal in rocks?One is to look for correlation betweenmineralogic properties and magneticpolarity. If all the reversely polarizedmembers of a suite of rocks possess aunique mineralogy, self-reversal is prob-able even if it cannot be reproduced inthe laboratory.

Balsley and Buddington (14) havediscovered such a correlation in theirstudy of metamorphic rocks in the Adi-rondack Mountains. Rocks of reversedpolarity invariably contain ilmeno-hem-atite; normal rocks invariably containmagnetite. These mineralogical differ-ences indicate self-reversal and also in-dicate that the geomagnetic field wasnot reversing its polarity during thelong interval when these rocks wereacquiring their remanent magnetiza-tions.

Significant as these results from theAdirondack Mountains are, they ap-pear to be the exception rather thanthe rule. In our own petrologic andpaleomagnetic studies of many hun-dreds of samples from normal and re-versely polarized lava flows, no signifi-cant correlation between polarity andpetrography has emerged. However,some uncertainty lingers even after themost careful search fails to reveal a

1539

on

Janu

ary

8, 2

012

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

physical property correlated with re-versals. New theoretical mechanismswhich might produce reversals are stillbeing advanced, and it is possible thatthe significant variables controlling self-reversals have not been isolated andidentified.A much stronger and more direct ap-

proach to the problem of self-reversalis available where two different ferro-magnetic minerals have become mag-netized at the same place and the sametime. The magnetization of igneousbodies and sedimentary deposits bakedby them was discussed earlier. Recentlywe have been able to extend this tech-nique to some igneous rocks not as-sociated with baked sedimentary rocksby demonstrating that the natural rem-anent magnetization of these rocks re-sides in two distinct ferromagnetic min-erals-hematite, with a Curie tempera-ture of 680°C, and magnetite or titano-magnetite, with a Curie temperaturebelow 580°C. The remanent magneti-zation of the two minerals has thusfar always been found to have thesame polarity. These results, like thosefrom the baked sedimentary rocks, sug-gest that self-reversals in nature arerare.

Stratigraphic Relations of Reversals

If the field-reversal theory is cor-rect and if self-reversals are rare, thenthe same magnetic polarity should befound in rocks of the same age allover the globe. Conversely, a demon-stration that groups of strata of nor-mal and reversed polarity are exactlycorrelative in age all over the earthwould constitute the strongest possibledemonstration of the validity of thetheory of geomagnetic field reversal.The success of the stratigraphic ap-proach is contingent, however, on oneprincipal condition: the duration ofepochs in which magnetic polarity isconstant must be sufficiently long tobe resolved by means of the availablegeologic techniques for establishing glo-bal contemporaneity of events. For ex-ample, if polarity epochs lasted only50,000 years, the classic techniques ofpaleontology could hardly be used to es-tablish contemporaneity. The pace ofevolution and the rates of dispersal oforganisms are not sufficiently rapid toproduce significant changes in widelyseparated fossil assemblages in so shorta time.

1540

Matuyama's discovery that theyoungest reversely magnetized rocks areearly Quaternary in age has been con-firmed by paleomagnetic studies inmany parts of the world. Thus, in ageneral way, these stratigraphic rela-tions lend support to the theory ofgeomagnetic field reversal. If, however,a substantial proportion of the rocksstudied contain minerals which under-go self-reversal after an interval aboutequal to the duration of the Pleisto-cene, as suggested by Verhoogen (13)for impure magnetite, then a similarstratigraphic distribution of rocks ofnormal and reversed polarity is to beexpected without a field reversal's hav-ing occurred. If this explanation is cor-rect, one might expect variations, fromplace to place, in the age of the young-est reversely polarized rocks, reflectingvariations in magnetite composition,cooling rate, and other factors whichmight control the time required for self-reversal. However, if field reversal isthe correct explanation, the transitionfrom reversed to normal polarizationmust occur at exactly the same strati-graphic horizon everywhere in theworld.

For the past 5 years we have at-tempted to determine whether the mostrecent transitions from reversed to nor-mal polarization in volcanic rocks fromthe Snake River Plain of Idaho, fromAlaska, from California, from Hawaii,and from New Mexico are contempo-raneous. In a general way, our resultsagree with those from other continents.Reversals are lacking in rocks from theupper Pleistocene and appear first instrata designated middle or lower Pleis-tocene. However, our attempts to de-termine whether the most recent transi-tions at all these localities are exactlycontemporaneous have proved incon-clusive. Much of the difficulty lies inthe fact that this last transition occursduring the Pleistocene Epoch. Due toboth the slow rate at which evolutionproceeds and the time required forplant and animal migrations, paleonto-logical techniques are not capable ofprecisely resolving separate events oc-curring within such a short time in-terval as the Pleistocene. Independentindicators of difficulties encountered inattempts to obtain early Pleistocenecorrelations are the well-known prob-lems of correlating the Pliocene-Pleis-tocene boundary from place to place,and the equally great difficulty of cor-relating glaciations over even moderate

distances. In working with volcanicrocks, as we do, the difficulties arecompounded by the paucity of inter-calated sedimentary deposits bearingfossils or indicators of glacial climaticfluctuations.Our stratigraphic studies of rever-

sals indicate that the adjustments onemust make in local stratigraphic as-signments to make all data consistentwith the hypothesis that a worldwidetransition from reversed to normal po-larity occurred during the Pleistocenefall well within the uncertainties of theoriginal assignments. However, this isfar from proving stratigraphically thatthe transition occurred everywhere atthe same time. Indeed, if the most fa-vored stratigraphic assignments at alllocalities are taken at face value, thenit must be concluded that the most re-cent transitions were not contemporane-ous.

Paleomagnetic-GeochronometricStudies

An obvious way to resolve this ambi-guity is to obtain dates by radiometricmethods from groups of rocks of nor-mal and reversed polarity. On the basisof general stratigraphic considerations,the duration of geomagnetic polarityepochs is estimated to have been be-tween 0.25 and 0.5 million years (15).Until recently, none of the radiometricmethods used for dating carbonaceousmaterials or for dating minerals couldbe used in this time range. The de-velopment, by J. Reynolds at the Uni-versity of California, of a gas massspectrometer with improved sensitivityand low background has made it pos-sible to date young rocks by meansof the decay of potassium-40 to argon-40, despite the low concentration ofradiogenic argon (a 1 0-g sample of ahalf-million-year-old rock containing 2percent of potassium contains only 1.78X 1 011 mole of radiogenic argon-40).The group at the University of Cali-fornia led by G. H. Curtis and J. F.Evernden has been particularly success-ful in dating young volcanic rocks byusing refined extraction techniques toreduce contamination from atmosphericargon and using the Reynolds spec-trometer for argon analysis. For severalyears the time has been ripe for apply-ing the potassium-argon geochrono-metric method to the study of reversalsof the magnetic field.

SCIENCE, VOL. 144

on

Janu

ary

8, 2

012

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

The polarities of volcanic rockswhich have been dated by radiometryare shown in Fig. 3. Both the paleo-'magnetic and the geochronometric datarepresent the results of independentinvestigations by several laboratories(16). We obtained the paleomagneticdata for rocks from North America;Tarling obtained the results for Hawaii;Rutten, the results for Europe; andGromme and Hay, the results forAfrica. The radiometric analyses weremade by Curtis and Evernden at theUniversity of California, by McDougallat the Australian National University,and by Dalrymple at the Universityof California and later at the U.S.Geological Survey.

Except for the samples from Eu-rope, all the samples in these studieswere treated in alternating magneticfields to test stability of the remanentmagnetization, and in several instancesother reliability tests were made. Forexample, in our paleomagnetic investi-gations of radiometrically dated volcan-ic rocks we have been especially con-cerned with the problem of detecting

z

0

0

;90 -

L&J

possible self-reversals, because even afew undetected self-reversals couldgreatly obscure the age relations ofrocks of normal and reversed polarity.In an attempt to detect reversals de-pendent on cooling rate, we have col-lected from each outcrop multiple sam-ples that cooled at different rates.Where available, baked zones weresampled, as were xenolithic inclusionscontaining ferromagnetic minerals dif-ferent from those in the parent lavaflow. Never in the course of collectingseveral thousand samples from severalhundred lava flows and intrusive bodieshave we encountered both normal andreversed polarity in the same igneouscooling unit. [A trivial exception is theoccasional sample that has been struckby lightning; the magnetization effectsof lightning are easily detected and re-moved (17).] When heated andcooled in the laboratory, none of thesesamples is self-reversing-a findingwhich suggests, as does the field pvi-dence, that self-reversals dependent oncooling rate have not occurred.

In addition, a variety of thermomag-

netic, crystallographic, and chemical ex-periments and observations were madein an attempt to find a correlation be-tween magnetic polarity and other phys-ical properties. None was found. It istherefore very unlikely that lavas ofself-reversed polarity contributed to thedata for North America cited in Fig. 3.Moreover, the agreement of all the datafrom the diverse sources is remark-ably good. The ten volcanic rocks withages between 0 and 1.0 million yearsall have normal polarization, whereasover 30 reversely magnetized volcanicunits have ages in the range 1.0 to 2.5million years. The only normally mag-netized rocks in this interval are twonormally magnetized lava flows, bothwith ages of 1.9 million years; theirsignificance is discussed later. Some ofthe dates are for mineral separates ofsanidine, biotite, and plagioclase; twoare for glass; and some are for totalrock basalt, this being by far the larg-est category. The ferromagnetic min-erals have a wide range of composition,including magnetite, titanomagnetiteand oxidized titanomagnetite, and hem-

2.0TIME IN MILLIONS OF YEARS

o NORTH AMERICA

o EUROPE

7 AFRICA

A HAWAII

CLOSED SYMBOLS: NORMAL POLARITY

OPEN SYMBOLS: REVERSED POLARITY

Fig. 3. Magnetic polarities of 64 volcanic rocks and their potassium-argon ages (16). Geomagnetic declination for moderatelatitudes is indicated schematically.

A 0

- I

Brunhes Matuyama Gaussnormal epoch reversed epoch normal epoch reversed epoch?

Olduvai event Mammoth event

l i lt t

0 1.0 3.0 4.0-I I

26 JUNED 1964 1541

on

Janu

ary

8, 2

012

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

Reversed Normal Reversedepoch epoch epoch

Mixed L;... Reversedepoch I epoch

3 (+)0'-

c -

L-)

1AWqM P

II - i-i--

~ WVmMI

(+)0a'-41-c4(I..

C-

H-

time -lTime :o

b. C.

Fig. 4. (a) Two-disc dynamo model with two solutions (b and c) for current flow. [After Mathews and Gardner (21)]

atite, and they also show a widerange of intergrowth textures. In thesamples we have studied, ferromagneticminerals of all types occur in both thenormal and the reversely polarizedgroups. Faced with these data, onemust conclude (i) that the geomagneticfield has reversed polarity repeatedly;(ii) that self-reversals in young vol-canic rocks are rare; and (iii) that thepotassium-argon technique for datingyoung volcanic rocks is remarkably pre-cise.

Discussion

The succession of young volcanicrocks at any one locality rarely if everis sufficiently continuous to give a com-plete record of the history of the geo-magnetic field. Nonetheless, because thepolarity of the field is a global phenom-enon, data from all over the worldmay be synthesized, as has been donein Fig. 3, to indicate the declinationthat would be observed at any samplingsite that possessed a continuous record.The short-period deviations shown sche-matically about the north and southdirections are typical of the amplitudeof secular variation in moderate lati-tudes; little is known as yet about theactual periods of the variation. Thegeneral character of the geomagneticfield appears to change abruptly at theboundaries of certain time intervalswhich we have termed polarity epochs.Epochs of two types may be recognized-those represented wholly or predomi-nantly by rocks of normal polarity andthose represented wholly or predomi-nantly by rocks of reversed polarity.

1542

Within the youngest reversed-polari-ty epoch there occurs a short intervalof normal polarity, identified in Fig. 3as the Olduvai event. This event is nowsupported by data from both Africaand North America and may be con-sidered well established. Less securelydocumented is the Mammoth event(Fig. 3), based on two points of re-versal near 3.0 million years ago; thetime interval between these points isabout equal to the precision of thepotassium-argon dating method; thus,additional data will be needed to deter-mine whether these two points repre-sent one or two events. At several otherplaces in the reversal time scale thereare gaps in the data sufficiently broadto suggest the possibility of additionalevents, but the pattern of epochs ap-pears well established.The accuracy with which the boun-

daries of epochs and events may bedrawn depends on the accuracy of in-dividual datum points, on their densityand distribution, and on the time re-quired for transitions between polarityepochs. For this time interval, it isestimated that the standard deviation ofages obtained by potassium-argon dat-ing averages about 5 percent, althoughthe precision of individual dates mayvary considerably. We estimate thatthe boundary shown at 1.0 millionyears is precise to within 0.05 millionyears; that at 2.5 million years, towithin 0.2 million years; and that at3.4 million years, to within 0.1 millionyears.The time required for transitions be-

tween polarity epochs is not knownfrom direct measurement. Estimates ofthe order of 10' years have been made

on the basis of the proportion of allstrata that are intermediate stratigraph-ically between rock sequences of nor-mal polarity and sequences of reversedpolarity and that also have intermedi-ate directions of magnetization (18).Another estimate may be made on thebasis of the observation that of the 64flows for which we have radiometrical-ly determined dates, none has an inter-mediate direction of magnetization. Un-der the assumption that each of theseven transitions shown in Fig. 3 lasted50,000 years, the probability that 64randomly distributed dates would all falloutside all of the transition zones is31/2 percent, suggesting that the tran-sitions are shorter than 50,000 years.The corresponding probability for tran-sitions lasting 10,000 years or less is28 percent or more; thus such transi-tions are more compatible with theradiometric and paleomagnetic data.The rapidity of the transitions makesit difficult to study the important prob-lem of the global shape of the fieldand the intensity of the field duringthe transitions, both of which factorscontrol the magnetic shielding of theearth from cosmic rays and hence mayhave significant biological effects (19).On the other hand, the sharpness of theepoch boundaries makes them especial-ly useful for worldwide stratigraphic cor-relation and also for assessing the pre-cision of radiometrically determineddates. For example, the overlap, nearthe 2.5-million-year transition, of nor-mal and reversed polarity, althoughconceivably due to rapid oscillation ofthe geomagnetic field, more probablyreflects the precision of the individualdates. Reversals may thus provide a

SCIENCE, VOL. 144

a.

4events

H

on

Janu

ary

8, 2

012

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

convenient method for directly compar-ing radiometric measurements on dif-ferent types of rocks from all parts ofthe world.We have given names rather than

numerical or sequential designations tothe polarity epochs for the followingreasons. The previously used numericalsystems count back sequentially fromthe present at each change in polarity(first normal, first reversed, secondnormal, and so on). However, if evena short-lived polarity interval is missedwhen a numerical system is set up, allolder designations must be changedwhen the new polarity interval is iden-tified. This would undoubtedly introduceconsiderable confusion into attempts touse the epochs for purposes of strati-graphic correlation. For example, in thefirst four articles linking paleomagneticand radiometric results, most of thepolarity data in the period 1.0 to 2.5million years ago were reported, yetthe significance of the short normal-polarity event at 1.9 million years agowas missed. The numerical systems alsopreclude designation of any given in-terval between polarity changes untilall later intervals have been recognized.The magnetohydrodynamic theory

for the origin of the geomagnetic field,generally considered the only reason-able theory yet proposed, seems capa-ble of the accommodating reversals (4).Complete hydrodynamic analyses of thehomogeneous-type dynamos that couldbe operating within the fluid core ofthe earth have not yet been made. Infact, the very formidable mathematicsrequired has precluded all but the mostsimple analyses. However, analogies,

suggested by Bullard and Rikitake (20),to the self-excited single-disc dynamoand the mutually excited two-disc dy-namo may be analyzed more complete-ly, and several solutions have beenobtained for certain configurations inwhich the oscillating currents in themodels change sign.An axially symmetric two-disc dyna-

mo recently analyzed by Mathews andGardner (21) is depicted in Fig. 4, a.One of their solutions for this model(Fig. 4, b) suggests the features shownin Fig. 3 for the geomagnetic field.Another of their solutions (Fig. 4, c)infers a third type of epoch in whichthe field rapidly and repeatedly changespolarity. Epochs of this type, whichmay be termed mixed, have not yetbeen recognized.

Striking as these similarities are, thetwo-disc dynamo model is vastly dif-ferent from a more realistic homogene-ous dynamo model, and care should beused in interpreting the solutions forcurrent flow as indicative of geomag-netic-field behavior. Nonetheless, thesolutions do suggest that the magneto-hydrodynamic theory is capable of ex-plaining polarity reversals of the typesobserved, as well as the origin of thegeomagnetic field itself. Conversely,any complete model or theory for thegeomagnetic field must be capable ofproducing reversals of irregular dura-tion.

References and Notes

1. B. Brunhes, J. Phys. Radium 5, 705 (1906).2. R. Chevallier, Ann. Phys. Paris 4, 5 (1925).3. M. Matuyama, Proc. Imp. Acad. Japan 5,

203 (1929); P. Mercanton, Compt. Rend. 182,859 (1926); , ibid., p. 1231.

4. R. Hide and P. Roberts, in Phys. Chem.Earth 4, 27 (1961); J. Jacobs, The Earth'sCore and Geomagnetism (Pergamon, Oxford,1963), p. 137.

5. M. Tuve, "Report, Department of TerrestrialMagnetism," in Carnegie Inst. Wash. Year-book 50 (1950-51), 65 (1951).

6. S. Smith, A. Dee, W. Mayneord, Proc. Phys.Soc. London 37, 1 (1924).

7. L. N6el, Ann. Geophys. 7, 90 (1951).8. T. Nagata, S. Akimoto, S. Uyeda, Proc.

Japan Acad. 27, 643 (1951); S. Uyeda, Japan.J. Geophys. 2, 1 (1958).

9. R. Wilson, Geophys. J. 7, 194 (1962).10. A. Cox and R. R. Doell, Bull. Geol. Soc.

Am. 71, 645 (1960).11. Y. Ishikawa and Y. Syono, Phys. Chem.

Solids 24, 517 (1963).12. E. Gorter and J. Schulkes, Phys. Rev. 90,

487 (1953).13. J. Verhoogen, J. Geophys. Res. 61, 201 (1956).14. J. Balsley and A. Buddington, J. Geomagnet-

ism Geoelectricity 6, 176 (1954).15. J. Hospers, Koninkl. Ned. Akad. Wetenschap.

Proc. B57, 112 (1954); A. Khramov, Dokl.Akad. Nauk SSSR Geol. 112, 849 (1957)(English translation, Consultants Bureau,New York, 1958); , Paleomagnetismand Stratigraphic Correlation (Gostoptechiz-dat, Leningrad, 1958), p. 204 (English trans-lation, Australian National University, Can-berra, 1960).

16. A. Cox, R. Doell, G. Dalrymple, Nature 198,1049 (1963); , Science 142, 382 (1963);

, ibid. 143, 351 (1964); I. McDougalland D. Tarling, Nature 200, 54 (1963); M.Rutten, Geol. Mijnbouw 21, 373 (1959); C.Gromm6 and R. Hay, Nature 200, 560 (1963);J. Evernden, G. Curtis, J. Lipson, Bull. Am.Assoc. Petrol. Geologists 41, 2120 (1957); J.Evernden, D. Savage, G. Curtis, G. James,Am. J. Sci. 262, 145 (1964); G. Dalrymple,Bull. Geol. Soc. Am. 74, 379 (1963);- thesis, University of California,

Berkeley (1963). Also included are 13 newdata recently determined in our laboratory.The dates are preliminary and subject torevision of ±5 percent; paleomagnetic analysisfor self-reversal was made on all samples,with the procedure described in the text andin our earlier publications, cited above.

17. A. Cox, U.S. Geol. Surv. Bull. 1083-E (1961),p. 131; K. Graham, Geophys. J. 6, 85 (1961).

18. R. Doell and A. Cox, Advan. Geophys. 8,221 (1961); M. Picard, Bul. Am. Assoc.Petrol. Geologists 48, 269 (1964).

19. R. Uffen, Nature 198, 143 (1963).20. E. Bullard, Proc. Cambridge Phil. Soc. 51,

744 (1955); T. Rikitake, ibid. 54, 89 (1958);D. Allan, Nature 182, 469 (1958).

21. J. Mathews and W. Gardner, U.S. NavalRes. Lab. Rept. 5886 (1963).

22. Publication of this article has been author-ized by the director of the U.S. GeologicalSurvey.

26 JUNE 1964 1543

on

Janu

ary

8, 2

012

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om