VOL. 78, NO.

11 •OURNAL OF GEOPHYSICAL RESEARCH APRIL 10, 1973

Superparamagnetic
andSingle-Domain
Threshold
Sizesin Magnetite
D. J. DUNLOP

GeophysicsLaboratory, Department o! Physicsand Erindale College

University o] Toronto, Toronto, Ontario, Canada

Particle size distributions have been obtained by grain counts on electron micrographs
of four samples containing submicroscopicequant magnetite particles. Blocking temperature
data indicate thermally unstable (superparamagnetic,SPM) magnetic behavior in one sample
whose grains range from 100 to 650 A in size. The SPM is confirmed by a large increase
of saturation remshencewhen the hysteresisof the sample is measured at 77øK. Quantitative
estimates of the SPM, single-domain (SD), and multidomain (MD) magnetization fractions
in this sample are made by comparing theoretical and observed values of the reduced
saturation remanenceat 77ø and 300øK. The particle size distribution then yields a critical
size d, -- (290 -- 360) _--*-50 A for the onset of SPM in magnetite at room temperature,
in close agreement with N•.el's SD theory. The critical size do at which domain structure
developsin equidimensionalgrains of magnetite is lesswell defined.It is certainly lessthan
480 _--*-
50 A and may even be lessthan d,; that is, a direct MD-SPM transition in magnetite
is a distinct possibility.The extreme narrownessof the stable SD range in equidimensional
magnetite particles suggeststhat either elongated SD grains or else small MD grains with
pseudo-SDcharacteristicsmust be the major carriersof the high-stability natural remshence
of igneousrocks.

Although the paleomagneticmethod is now
widely felt to be reliable, the essentialmecha-
nisms by which ferromagneticgrains in a rock
acquireand preservea remanentmagnetization
that is stable over millions of years in the field
and equally stable against demagnetizingfields
of hundredsof oerstedsin the laboratory remain
far from clear. Ferromagnetic particles that
lack domain structure (so-calledsingle-domain
grains) possess the requisitestability. However,
below a threshold size d, (which increaseswith
increasingtemperature) thermal agitation de-
stroys this high stability, and the particles
become superparamagnetic (SPM), whereas
above a second threshold size do domain struc-
ture beginsto develop,and the stability again
decreases[Kneller and Luborsky, 1963; Koba-
yashi, 1962]. The width of the stable single-
domain (SD) range (the range of grain sizes
between d, and do) thus essentiallydetermines
whether a particular ferromagneticsubstance
is likely to carry high-stability remanence.
Where the fossil remanence of a rock is car-
ried by mineralsof the hematite-ilmeniteseries, an accurateknowledgeof the SD thresholddo.
Copyright ¸ 1973 by the American Geophysical Union. theoriesof TRM are viable only if the SD range
1780
which occur in SD form over a wide range of
grainsizes[Banerjee,1971], it is clearthat this
stabilityreflectsthe high coercivityof SD hema-
tite [Dunlop, 1971]. In most igneousrocks,
however,the fossilremanenceis a thermorema-
nentmagnetization (TRM) carriedby members
of the magnetite-ulvospinel series.Thesemin-
erals have a much more restricted SD range
[Stacey, 1963], and the bulk of the titanomag-
netiresin rocks are simply too large to occurin
SD form. On the other hand, multidomain
(MD) magnetite is incapable of supplying
coercivities of several hundred oersteds.
The numerousexplanationsproposedto ex-
plainthis seemingparadoxfall into two general
categories.(1) The predominantremanencecar-
riers are MD grains,just aboveSD size,having
pseudo-SDbehavior[Stacey,1963; Dicksonet
a/., 1966]. (2) The predominant remanence
carriersare SD grainsor subgrainregionsof SD
dimen•ons[Verhoogen,1959; Ozimaand Ozima,
1965; Strangwayet al., 1968]. Choosingobjec-
tively betweenthesepossibilities
clearlyrequires
A reliablevalue of d, is alsoimportant, for SD
 DUNLOP.' THRESHOLD SIZES IN MAGNETITE 1751

in magnetiteis wide enoughto encompass

a conjunctionwith size distributions determined
reasonablefraction of the grain size spectrum by electronmicroscopyto deduceexperimental
of igneousrocks. estimates of do and ds. Values of do and d•
Unfortunately, theoretical estimates of do determined in this way are reported here and
vary widely.A detaileddiscussion of the prob- provide the first clear experimental check of
lem has recently been given by Evans [1972]. theories of SPM and SD behavior in equi-
For pure magnetite,dohasbeenvariouslyesti- dimensional magnetite particles. In addition,
mated as 300 A [Stacey,1963], 500 A [Morrish the dovalue is relevant to theoriesof elongated
and Yu, 1955], 700 A [Evans and McElhinny, particles [Strangway et al., 1968; Evans and
1970], and 1000 A [Strangwayet al., 1968] in McElhinny, 1970] since it fixes the end point
equidimensionalgrains.Experimentaldovalues of the do versus axial ratio curve.
[Berkowitzand Schuele,1959] of 300-700A in
EXPERIMENTAL DESIGN
the related ferrites NiFe20,, CoFe20,, and
CuFe•O, agree in a general way with these The detection through magnetic measure-
estimateswithout strongly favoring any one of ments of the relative proportions of SPM,
them. The SPM threshold d8 is calculated to be SD, and MD particlesin a sampleis basedon
about 300 A at room temperature by N•el's two well-known facts. (1) The SD particles
[1949] theory of magnetizationblockingin SD have a high saturation remaneneeJrs that is
particles;thereis no experimental confirmation grain sizeindependent;SPM and MD particles
of this size estimate. have low Jrs. (2) The SPM thresholdis strongly
The extremelyfine particle size of SD mag- temperaturedependent.The SPM fraction can
netite has severelylimited attempts to deter- be inferred from the difference between Jr•
minedoexperimentally. In elongated particlesdo measured at room temperature and J• mea-
is expectedon theoreticalgroundsto increase sured at a temperature low enoughfor all par-
considerably,and indeed Motfish and Yu ticles to be magneticallystable.The MD frac-
[1956] have measuredsquarehysteresisloops tion can then be deduced from the residual
for individual Fe30, particles about 0.25 X differencebetweenJ• as measuredand as pre-
1.5 t•m in size.Thesemeasurements and others dicted by SD theory.
of a similarnaturemadeby Muirhead[1963] The central experimentsin this paper there-
are the only direct confirmationof SD be- fore involve measurementof room temperature
haviorin magnetite.Studiesof the moretypical and low-temperaturesaturationhysteresis.Each
caseof roughly equidimensional particleshave of the four magnetite powderswas packed in
servedto definedoonly within very broadlimits. a small (3 mm in diameter) nonmagnetic
Dunlop and West [1969] determinedby mag- bucket. Magnetization curves were measured
netic granulometryof a basalt an approximate at 77ø and 300øK with a commercialvibrating
SD range of 200-800 A. Domain observations sample magnetometer(VSM) in the field (up
made near the optical limit by So#el [1969, to 6000 oe) of an electromagnet.
1971] gave compatible results which sug- High-temperature hysteresiswas also mea-
gestedthat SD TRM in basaltsis associated sured by using large cylindrical clay samples,
with grainsbelow about 0.7 t•m in size.Below 10 cm long by 1.4 cm in diameter, in which
the optical limit Evans and Wayman [1970] the magnetite was dispersed (about 1% by
have used electron microscopyto determine a volume). The measuringinstrumentwas a sensi-
size distributionof magnetiteparticles,at least tive ballistic magnetometer incorporating a
someof which were SD, peakingbetween1250 2400-oe solenoid [West and Dunlop, 1971].
and 2500 A. Although the packing fraction of magnetite in
The study reported here grew out of an in- samplesmeasuredballistically was much less
vestigationof TRM and hysteresis in four mag- than that used in VSM measurements, room
netite powderscontainingequidimensional par- temperaturehysteresisdata measuredby the
ticles 300-2500 A in size [Dunlop, 1972]. two methods did not differ greatly. Thermal
Because both SPM and SD thresholds are ex- demagnetizationof weak-fieldTRM was mea-
pectedto fall in this range,it was anticipated sured by using the large samples and the
that the magnetic propertiescould be used in ballistic magnetometer.
1782 I)UNLOP: THRESHOLD SIZES IN •VIAGNETITE
CHEMICAL AND THERMOMAGNETIC ANALYSIS
Electron micrographsof the experimental
magnetite powdersare shownin Figure 1. The
powderswerekindly suppliedby Dr. T. Takada
of Kyoto University,who describes their prep-
aration as follows.ColloidalFe(OH),. particles
were precipitated from an aqueoussolution of
FeSO, and oxidizedby bubblingair throughthe
suspensionat moderate temperatures. The
particle size obtained depends on both tem-
perature and initial Fe•'* c(incentration. The
Fe30, particles prepared in this way are ex-
pectedto be monocrystalline and cubicin shape.
The shapesobservedin Figure 1 are indeed
approximately cubic (the shapesseen in the
micrographs are, of course, two-dimensional
projectionsof three-dimensionalshapes), and
X ray line broadeningis of about the amount
expectedfor crystallitesof a size equal to the
observed
particle
sizes[Bartram,'
1967].The
meanparticlesize(d) of eachpowder
wasde-
terminedin two ways: first, by averagingover
particle size distributions obtained from the
electronmicrographs(to be discussedlater),
and second,by measuringthe total surface
area of powdersamplesof knownmassby the
standard BET nitrogen adsorptiontechnique
[Adamson, 1967], assumingcubic particle
shape.The BET size is lesssubjectto observa-
tional bias and shouldbe more reliable,but, as
Table 1 shows,the two estimatesagreequite
well.
Table 1 alsodemonstrates that the magnetites
are deficientin the cation Fe•'*. That is, they
have been partially oxidized during the low-
temperature preparation processto a spinel
phase(no aFe•O3linesappear in X ray diffrac-
tion profiles) whose compositionlies between

Fig. 1. Electron micrographs of the experimental magnetite powders. Note the pre-
dominantly cubic particle shape and the relatively narrow dispersionin particle size in each
sample.
 DUNLOP' THRESHOLDSIZES IN I•IAGNETITE 1783

that of Fe•O, and the metastablephase7Fe•O8 dsat(normalized)

(maghemite).This property of ultrafine-grain
magnetites is apparently universal [Colombo
et al., 1965], particularly when they are pre-
pared in the presenceof water [Elder, 1965;
Sakamoto et al., 1968]. Similar cation-deficient
magnetitesoccur commonlyin nature [Sanver
and O'Reilly, 1970; Ozima and Larson, 1970].
The work of Gallagher et al. [1968] sug-
gests a high concentrationgradient of cation
holes in submicron Fes_•O,crystals. In other
words, such magnetites are not chemically
homogeneous but consistof a core of relatively
.'6•
I.SAMPLE
2 • ß

pure Fe•O, and an oxidizedsurfacelayer. This

fact would readily account for the systematic
decrease of Fe•* with decreasing grain size
recorded in Table 1.
.4
Further corroborationfor Gallagher et al.'s
conclusions is provided by the saturation mag-
.2
netization-temperature curves in Figure 2
(measuredin a moderate vacuum (•10-' torr)
o
in a 5000-oe field). The cooling curves fall
below the heating curves and thus indicate
that 7Fe•O• has inverted to practically non-
magnetic aFe•O• (hematite) during the heat- .2

ings. The cooling curves and the subsequent

OI I I i
heating curves follow closelythe spontaneous o 200 4oo 600
magnetizationcurve of Fe•O, [Pauthenet and Temperature
(øC)
Bochirol, 1951]. The Curie points of samples
1-4 fall in a narrow range, 582ø-588øC,about Fig. 2. Saturation magnetization-temperature
585øC,the Curie point of pure magnetite. curves of the experimental samples. Curie points
are all near 585øC, but the curves are irreversible
Evidently, the experimental'magnetites'be- becauseof the presenceinitially of •Fe•O•.
fore heating contain a core of pure Fe•O, sur-
rounded by a layer consistingin part or in
whole of 7Fe•Os.However, the irreversiblede- entire surfacelayer consistedof 7Fe•O• vary-
creasein J,at is less than that expected if the ing from 3% in sample1 to 16% in sample4.
It is possiblethat aFe•O• as well as 7Fe•Oa
was presentbefore heating,althoughthe X ray
TABLE 1. Some Physical and Chemical Properties
of Experimental Magnetite Powders data and the conclusionsof Elder [1965],
Colombo et al. [1965], and Gallagher et al.
Sample Sample Sample Sample
1 2 3 4 [1968] make this explanation unlikely. It is
even conceivablethat not all the 7Fe•O•inverts
(d) from BET surface
area, A
2210 1000 760 370
to aFe•O• at 600øC (R. T. Merrill, private
(d) from observed size 2320 1080 765 395 communication,1972).
distribution, A
Standard deviation of 420 290 250 150 It seemsmore likely, however,that the non-
size distribution, A
Fe2+ content,* wt % 21.9 17.4 13.7 12.7 stoichiometryis simply less extreme than the
Fes0q content, wt % 91 72 57 53
AJsat in first heating- -3 -6 -10 -16 chemicalanalyseswould indicate, the analyses
cooling cycle, % being very approximate. Unfortunately, post-
Agreement is good between nitrogen adsorption and electron
heating Fe•O,/Fe•O8ratios cannotbe checked
microscope determinations of (d). There is conspicuous disa- becauseaFe•O• is so weakly magnetic that it
greement between Fe203 content as indicated by chemical anal-
ysis and as indicated by •7sa t during thermomagnetic analysis. makes no appreciable contribution to the
*24.1% for Fe30q.
thermomagneticcurvesin Figure 2. Compara-
1784 DUNLOP: THRESHOLD SIZES IN MAGNETITE

rive preheating and postheating hystere•s of small grains. This variation is probably
curveslikewisegive no unambiguousindication minor in comparisonwith that produced by
of the nature and amount of the Fe'*-deficient variations in titanium content,however.
phase.The main effect observedis an increase
of 7-20% in the coerciveforceafter heating, PARTICLE
SIZEDISTRIBUTIONS
theamount
of theincrease
beingroughly
cor- Particlesizedistributions
determined
from
related
withcation
deficiency.
Thisincrease graincountsontheelectronmicrographsin
couldbeexplained
by manymodels(e.g.,
in FigureI andothersimilar micrographsare
termsoftheunknown,
butpresumably
differ- shownin Figure
3.Becauseofthedifficulty
of
ing,anisotropies
of FEB0,
andFeB_•O,
or by dispersing veryfineparticles,
it wasdifficult
appealing to possible exchange coupling of to measure more than 50 separate grains of
Fe•O,to a shellof high-coercivity
aFe•O•).samples
1-3.In thecase
ofsample
4,125grains
Muchmorelikelyis a lesselegant
andpurely weremeasured.
The distributions
are there-
physical
mechanism'
thechange
of effectiveforeonlyapproximate.
Thesmallgrainsize
grain size during heating. Before heating,each part of each spectrum is especiallyuncertain
individual particle acts as a unit. After heat-
ing, only the Fe•O, core is magnetically sig-
nificant, and the coercive force, which is
1
stronglygrain size dependentin theseparticles SAMPLœ •
(cf. Figure 6), increasesaccordingly.
Some
experiments
reported
inthis
paper,
in ]
particular high-temperature hysteresis and
thermal demagnetizationof TRM, were neces- • • I I
I
1600 2C•00
• 2400 2800 3200 3600
sarily made with heated samples.In fact, these d(A)
sampleswere preannealedat 700øC for several
i
hours to stabilize their properties. However, SAMPLœ
2 !
i

,,
in the critical 77ø and 300øK hysteresismea-
surements,from which doand d8 were indirectly ! I !

deduced,it wasimportant
toavoid
thepossi- ]
bilityof aneffective
magnetic
grain
sizeless •, , ,,
1400
,I
thanthephysical
grainsizeobserved
under
the 6oo d(A)
electron
microscope.
Thesemeasurements,
ac- SAMPLœ
3
cordingly,were made on unheatedsamples.As
a result do and d8 values in this paper refer
to cation-deficientmagnetite rather than stoi-
chiometric magnetite. However, the differences
are expectedto be small, since Morrish and
600 800 K)00 1200
Yu's [1955]theoretical
valueof dofor Fe•O• 4oo
d(A)
does not differ greatly from the corresponding SAMPLœ
4
value for 7Fe•OB,which is, in effect, 100%
Feø'*-deficient
magnetite. In reality the critical
sizes
forFe•O•
and7Fe•O•
maybemore
dis- i
similar
than
Morrish
and
Yu's
calculation
wouldI zbo'
suggest,sincemagnetocrystallineanisotropyis •
' 'oo' ebo
neglectedin their calculation.
Even so, the do •oo
and d• values determinedin this paper are by Particle diameter,d (A)
no means irrelevant, since natural titanomag- Fig. 3. Grain size distributions determined
netitesof grainsizesimilarto that of the pres- from graincountson electronmicrographs
like
entpowders
areprobably
alsocation
deficient.of
those in Figure 1. Theordinate is thenumber
grains. The mean of each distribution is indi-
Somevariation
of theappropriate
value
of do cated by a dashed line.Notethattheoverlap
from rock to rock is to be expectedbecause betweendistributions
is not large,exceptfor
of variation in the degreeof maghemitization samples2 and 3.
 DUNLOP: THRESHOLD SIZES IN MAGNETITE 1785

becausesmall grains tend to be obscuredby temperature,and not after coolingto room

their largerneighbors. temperature, and sothe resultsshownin Figure
Nevertheless,it is apparent that these 4 includea thermal decreasedue to the tem-
powdersare suitablefor determiningreasonably peraturedependence of spontaneous magnetiza-
precisemagneticthresholdsizes.For onething, tion J..
the size distributionof each powder is rela- There is a distinctdifferencein demagnetiza-
tively narrow.The standarddeviations(Table tion characteristics betweensamples1-3 on
1) rangefrom 20 to 40% of the meanparticle the one hand and sample4 on the other. Up
size, and only samples2 and 3 have signifi- to about 540øCthe demagnetization curvesof
cantly overlapping distributions.
Secondly,the samples1-3 match the experimentalJ,t-T
particlesare essentially
equidimensional.
There curves(Figure 2) within experimentalerror.
is verylittle variation
of particleshapeand In otherwords,
all the blocking
temperature•
hencelittle variationof shapedependentmag- in these sampleslie above 540øC. Sample4,
netic parameters, suchas doand coerciveforce however,evidentlyhas blockingtemperatures
H•. distributed over a very broad range, from
560øC to room temperature.
BLOCKING TEMPERATURE MEASUREMENTS
These low blocking temperatures were the
first indication that sample 4 might contain a
A weak-field TRM was induced in each significantfraction of superparamagneticpar-
sampleby coolingfrom 600øC to room tem- ticles with blockingtemperaturesbelow room
peraturein a 1-oe field. The TRM were then temperature.This possibilitywas investigated
thermally demagnetized by stepwiseheatingin quantitatively by the hysteresisexperiments
vacuum. Magnetizationwas measuredat high describedin the next section.

J/Jo
1.0

•o
0.8 "'•

06

04 x•\,,,
O.

øo ' ' ' ' '

Temperature
("C)
•i•. •. Stepwise •herm•] dems•ae•i3s•ioa of l•e WRY. Because ms•ae•i3atioas were
measured • high tempers•ure, •he reversible therms] decreaseof •, is iac]uded ia •he resairs.
Whis •empers•ure depeadeace wss determiaed from Yi•ure 2 •ad is p]o•ed as the dashed
cur•e. 8amp]ee 1-3 hs•e ao b]ockia• •emper•ures below 5•0øC, bu• ssmp]e • hss b]ockia•
temperstures distributed dowa •o room tempers•ure.
1786 DUNLOP: THRESHOLD SIZES IN MAGNETITE

HYSTERESIS
AT 77øK AND300øK talline anisotropywith [111] easy axes.Dunlop
Interpretation
o/ thermal
changes
in satura- [1970]hasshown
experimentally
that the
tion remanence.
Determination
of the SPM crystalline
contribution
to a mixedanisotropy
fraction in a sample, as was discussedpre- increasesin importance at low temperature,
viously,
followsfromthedecrease
ofsaturation
withtheresultthatj,, alsoincreases.
Radhak-
remanenceJ,, between
somelowtemperaturerishnamurthyet al. [1970]haveobserved
the
(here77øK)at whichall particles
havea same effect
in low-temperaturehysteresis
mea-
common thermallystable
domainstate(SD surements on several basalts.
The effectis
orpossibly
MD)androom temperature.
Under readily explained.Thecrystallineanisotropy
these
idealconditions
theSPMmagnetizationconstantK, varies withT likeJ,', wheren
fractionis is about8 or 9 for magnetite
[Fletcher
and
Banerjee, 1969], whereas the equivalent con-
stant Ku for shape anisotropy is proportional
],,m
= jr,(77)
--j,,(300) (1) toJ,•'.
j,,(77) Because
thecoercive
force
H• varies
as
wherej,, -- J,•/J, is the reducedsaturation K/J,, valuesof H• measuredat varioustem-
remanence. peratures can usually be fitted to a relation
In practicej, may changewith temperature [Dunlop,1970]
for a number of reasonsquite unrelated to the

presence
Five orabsence
ofthe ofparticles
principal in
aSPM
reasons
are state.
H•(T)
(1)transi- - aIj•(T)
H•(To) ]•-1
J•(To)J J•(T)(2)
d-bJ•(To)
tions in crystallinestructureor anisotropybe-
tween 77ø and 300øK; (2) changein the rela-
rive proportions of shape and crystalline
anisotropyin SD particles; (3) changein j,,
in MD particlesin responseto a thermal change
in coercivity; (4) changein do with tempera-
ture resultingin different proportionsof SD
and MD particlesin a mixture; and (5) non-
saturationhysteresismeasurements.
Coarse-grained pure magnetite,with a crys-
tallographic transition at 119øK and an
'isotropic'point at about 130øK, is obviously
not amenableto analysisby (1). Fortunately,
both transitions.are reportedlyshifted to tem-
peratures below 77øK if the grain size is
below about l•m [Nagata et al., 1964] or if
the magnetiteis cation deficient(C. Radhak-
rishnamurthy,personalcommunication,1971).
Measured J,,-T curves (not shown) confirm
the absenceof any transitionsbetween77ø and
300øK for the present samples.Dr. S. K.
Banerice has kindly extended these measure-
ments to still lower temperaturesand reports
that no transitions appear between 77ø and
4.2øK, either. This suppressionof the transi-
tionsis thoughtto be a resultof the fine particle
sizerather than cationdeficiency,sincesample1
containspractically stoichiometricmagnetite.
The value of j,, for SD particlesdependson
the dominanttype of anisotropy,being 0.5 for
uniaxial shape anisotropyand 0.87 for crys-
where To is somereferencetemperature (e.g.,
77ø or 300øK). Equation2 then yields (through
the constantsa and b) the relative contribu-
tions of shape and crystallineanisotropiesat
any temperature.
Hysteresismeasurementson samples1-4 at
temperaturesbetween300ø and 850øK (to be
reported elsewhere)do not give a goodfit to
(2). Instead, H•(T) varies approximately as
J,'(T), where 1.5 • m • 2. This behavior
is not yet understood,but it is clear that
crystallineanisotropyis much le• important
than might have been expected for roughly
equidimensional particles.
The valuesof H• listed in Table 2 support
this conclusion. The fractionaldecreasein H•
between77ø and 300øK is about the sameas
the corresponding thermal decreasein J,, ex-
cept in the caseof sample 4, where the larger
decreaseis presumably due to a substantial
SPM fraction.
The apparent predominanceof shape ani-
sotropy in these sampleshas some theoretical
support. Tonge and Wohl)•arth [1958] have
shown that a small admixture of shape ani-
sotropy has a disproportionateeffect on the
value of j,,. For this reason,and on the basis
of the experimentalevidence,the appropriate
j,, value for the SD fraction of the present
samplesis taken to be 0.5. This value will be
usedthroughoutthe calculations.
 DUNLOP' THRESHOLD SIZES IN MAGNETITE 1787
TABLE 2. High-Field Hysteretic Properties of Experimental Samplesat 77øand 300øK

Jrs/J (1500) Jrs/J (6000) Jrs/J (•) * •,ñ oe

Sample A 77øK 300•K 77øK 300øK 77øK 300øK 77øK 300øK

1 2210 0.128 0.124 0.105 0.103 0.100 0.098 110 95

2 1000 0.204 0.201 0.169 0.170 0.161 0.162 165 145
3 760 0.240 0.225 0.196 0.188 0.185 0.180 200 170
4 370 0.313 0.251 0.259 0.212 0.243 0.201 260 175

Number in parentheses represents field in oersteds in which magnetization J was measured.

*J(•) is an extrapolated value of saturation magnetization (method of extrapolation is explain-
ed in text).
ñTabulated values of coercive force He are from 6000-oe hysteresis curve.

Changesin jr, due to coercivity changesin of 1500 and 6000 oe are compared with true
MD particles are more difficult to take into saturation j, (Table 2). About 95% saturation
account. The effect does not occur in SD is reachedat 6000 oe, and only about 79% at
particles,where jr, dependson the number of 1500 oe, the bulk of the nonsaturationbeing
easy axes but (except in the case of mixed attributable to J, rather than Jr,. Saturation
anisotropiesdiscussedabove) not on the mag- valuesof J, were obtainedby the standardpro-
nitude of the anisotropy constants. But, in cedure of plotting J versus l/H, for H in the
MD particles, range 3000-6000 oe, and extrapolating to 1/H
-- 0 [Bean and Jacobs,1960]. The plots were
j•., = H•/NJ., excellent straight lines; no 1/H •' terms were
to a first approximation[N•el, 1955; Stacey, evident, perhaps becauseof the rather limited
1963]. BecauseH•(T) is not in general pro- range of fieldsused.
portional to J,(T), jr, is temperature depen- Discussion of experimental hysteresis data.
dent. In the present samplesHdJ, varies as Representative 77ø and 300øK hysteresiscurves
J,•-•(T), where 0.5 ( m -- 1 ( 1, and so are comparedin Figure 5. Samples 1, 2, and 3
the effect is not a major one. Two extreme behavein a similar manner' H c decreases, and
models for the behavior of the MD fraction the shape of the hysteresiscurve changesin
in samples 1-4 will be discussedin a later responseto the softening of the coercivity
section. spectrum between 77ø and 300øK, but, within
Changein the SD/MD remaneneeratio with experimental error, jr, remains constant. In
temperature is a potentially serious effect in sample 4, on the other hand, jr, decreasesby
sampleswhere grains of both types occur. Be- more than 15% betweenthe sametwo tempera-
cause little is known about the thermal varia- tures, and thus the presenceof a substantial
tion of do at low temperature, the effect will SPM fraction in this sampleis confirmed.
be ignoredin the one two-component(i.e., SD A striking illustration of the unique behavior
and MD) model consideredin this paper. of sample4 is providedby Figure 6, a log-log
Finally, jr, may show a thermal variation un- plot of Hc and j, as functionsof mean particle
related to SPM particles if hysteresismeasure- size (d). The error bars on (d) are equal in
ments are made in fields insufficient to saturate length to the standard deviations in Table 1
jr,, since the approach to saturation of both and are intended to reflect the considerable un-
Jr, and J, is temperature dependent. (This de- certainties inherent in the size distributions of
pendenceis seenin the differing shapesof the Figure 3. Nonetheless,at 77øK all the samples
descendingbranches of the 77ø and 300øK obviously follow the usual MD relations
hysteresiscurves in Figure 5.) A field of [Parry, 1965]
1500-2000 oe may be sufficient to effectively
saturate Jr,, but J, approachestrue saturation Jr, o• (d)-p H• • (d)-• (3)
only in fields of 50,000 oe or more. Values of with p • 0.50 and q • 0.52. The implication
jr, for hysteresis measured in limiting fields is that all the samples,sample4 included,con-
1788 DUNLOP: THRESHOLDSIZES IN MAGNETITE
rain major MD fractions (a conclusionsub- The 300øK data are very different. Samples
stantiatedby the low intrinsic valuesof j,B). 1-3 follow the samepowerlaws as they do at
The amount of SPM material in sample4 at 77øK, with essentiallyunchangedp and q
77øK is evidentlyminor. values.The Hc and j, values of sample4 fall

d/dsat

-2000 -I000 • I/ I000 2000

- ,

-0.4

-0.6

b d/Jsat __.=

/ /'/

,,'/
/,"1 //
' _ ' I/ I /I
-•oo ,o• /,' I ,7 '. .. !oo__o
,•x•
.... ' •'c;oo

//•••"
X Applied
field
(øersteds)
•/• I-0-6 -0.8

Fig. 5. Hysteresis curves measured in vacuum in a maximum field of 6000 oe at 77øK

(solid curves) and 300øK (dashed curves). The magnetization values have been normalized
to true saturation magnetization,J,at, obtained by a 1/H extrapolation. The hysteresiscurves
have different shapesat the two temperatures becauseof thermal differencesin the approach
to saturation of J and J,. (a) Sample 2 shows a small decreaseof Hc between 77ø and 300øK
consistent with the thermal change in shape anisotropy and no change in J,,/J,. Samples
1 and 3 behave similarly. (b) Sample 4 has a larger change in H• and a decrease of 17% in
J,,/JB between the same two temperatures owing to the presenceof SPM particles.
 DUNLOP.' THRESHOLD SIZES IN •/•AGNETITE 1789

Hc (oersteds)

250

2OO

150

I00

Jrs/Js

0'25f:
0'20
f'
0.15

0.10 L

Mecnpcrticlesize,<d)(•m)
Fig. 6. Confirmation of SPM behavior in sample 4. At 77øK (solid circles) Ho and
Jr,/J. of all samples have the power law dependenceon mean particle size typical of MD
behavior.At 300øK (open circles)samples1-3 obey the samepowerlaws as they do at 77øK,
but Ho and Jr,/J. are lower than expected for sample 4 because the SPM grains no 16nger
carry a remanence. Error bars are equal in length to the standard deviation of the size
distribution of each sample.

far below these relations, however. A sub-
stantial SPM fraction in sample 4 at 300øK
is again indicated.
CALCULATION0r ds AND do
Amongthe samplesexamined,sample4, with
grain sizesrangingfrom 100 to 650 A, is unique
in containing both MD and SPM particles.
It is therefore suitable for determining
and, if the sample contains in addition SD
particles, for determining do as well. Because
the magnetic evidence from which SPM and
MD magnetization fractions will be derived
is basedon measurementsof remanentmagnetic
moment, an appropriate first step is to convert of ds and minimufti estimates of do.
the grain sizenumberdistributionn(d) (Figure
2) into the correspondingdistribution m, (d)
of saturation remanent moment by using the
fact that
m,.(d)= VJr,(d)n(d)ecd•jr.(d)n(d) (4)
where v is particle volume.
Three simple theoretical models for jr•(d)
have been usedin calculatingm,.(d) (Figure 7).
Model A is basedon the very real possibility,
indicated by the 77øK data in Figure 6, of a
MD-SPM transition in equant magnetite with-
out an intervening stable SD range. For sim-
plicity, jr• of MD magnetite in the range
100-650 A has been taken to be grain size
independentand equal to 0.243, the 77øK
jr• value for the sample as a whole.
Model B is a more realistic version of
model A, in which jr• has the grain size de-
pendencefound experimentallyin Figure 6 and
(3). ModelsA and B give maximumestimates
Model C is based on a conventional mixture
of SD and MD particles in sample 4. For
ease of calculationsjr• is assumedto be 0.5
for the SD fraction and 0 for both SPM and
MD fractions. Obviously, this choice for j,8
of MD grains is extreme. It is not clear, how-
 1790 DUNLOP' T•m•S•O•,D Szz]•s zN MAGNETITE

shown in Figure 8. These would be the ap-

Jrs/Js ds(77OK)ds(500OK
) MODEL A
03 ,• ,• ,
propriate remanencedistributions if sample 4
contained exclusivelyMD (models A and B)
o.I or SD (model C) particles. Under these con-
i i

øl 0 200 300 400 5OO •oo ditionsthe expected

j,, of sample4 wouldbe
d(A) 0.243 (the observedvalue at 77øK) for models
A and B or 0.5 for model C.
The fact that these values are not actually
drs/Js d•
s(77øK) MODEL B
observedat 300øK (or, in the caseof modelC,
03
•r/•, ds
(300øK) not observedat all) can be attributed to the
O.2 presence of SPM particles, which lose their
remanence between 77ø and 300øK, and, in
o0, I00 200 300 400 100 i model C, also to MD particles with j,, -- 0,
d(A)
which likewise merely subtract from the ex-
pected remanence. If the values j•(77) --
0.243 and j,•(300) -- 0.201 (Table 2) are used,
Jrs/Js
•s(77øK)
d,•(500•<) d•oMODEL
C the SPM fraction accordingto modelsA and B
O4 is
03

0.2

0.1 Saturation remanent

moment,
mr (d)
I00 200 300 O0 500 600 (norrnahzed)

Particle diameter,d (A)

0.15
Fig. 7. Models of the dependenceof J,•/J• on
grain size for use in calculating the mr(d) dis-
tribution of sample 4. Model A assumes MD 0.10
particles with grain size independent remanence.
Model B assumesMD particles with J,, ccd -ø'5,
asis foundexperimentally
in Figure6. ModelC 0.05
assumes a mixture of SD particles (j,• -- 0.5)
and MD particles (j,• -- 0). The solid lines in-
dicate the j,•(d) function at room temperature. o
The dashed lines and hatched areas indicate how I0o
d(A)
the functionis extendedat 77øK, where many m (d) (normolized)
of theSPMparticles
arestabilized.
Thenumerical o.15b
values of do, d, (300), and d, (77) shown result
from the calculations (see text).
o IO

ever, what value of j,• is appropriate to these

small MD particles, and this choice at least 0.05
has the advantage of yielding a maximum value
for do. c • I I I I I

Thevalues
ofd,and
doshown
inFigure
7 • 2oo •o 4•o 5•o 6•o
Particlediameter,d (A)
were of course not known a priori but were
the result of the calculationsthat follow. Fig. 8. Normalized mr(d) distributionsob-
tained from the n(d) distribution of sample 4
Thefirststep
inthecalculation
was.
togen- (Figure
3) ontheassumption
thatj,, is(a)in-
eratethemrdistributions
of Figure8 fromthe dependent
of d or (b) varies
asd-ø'5
forallgrain
n(d) data for sample 4 by using (4) as a sizes.In practicethesedistributions
are truncated
prescriptionand substitutingeither j•(d) -- at the lowerendbecause
of the presence
of SPM
const(models
A andC) or j,•(d) o: d-ø'• because
particlesand,in model C, alsoat theupper end
of MD particles. The do and d• are the
(modelB). The resulting
distributions
were truncation sizes
required to givetheSPMand
then summedover the range100 A < d < MD magnetization
fractionsindicatedby hys-
650 A and normalized and gave the results teresisdata.
 DUNLOP' THRESHOLD SIZES IN MAGNETITE 1791

jr•(77) -- jr•(ao0) smallerthan the estimateduncertainty of ñ50 A

]sv = = 0.173
0.243
whereas,accordingto model C, it is
jr..(77) -- jr.•(ao0)
fspm= = 0.084
0.5
By similar reasoningthe MD fraction accord-
ing to modelC is
0.5 -- jr,(77)
]ma
= 0.5 = 0.514
Finally, d, was obtained by subtractingthe
computedSPM fractions from the small grain
size end of the distributionsof Figure 8. For
example, when model A is used, 17.3% of the
distributionin Figure 8a lies belowd - 360 A;
thus d, - 360 A. In a similar way, a maximum
value for do was obtained by subtractingthe
upper 51.4% of the distributionin Figure 8a.
Results are summarized in Table 3. The same
values are incorporatedin the modelsin Fig-
ure 7.
It is now possibleto justify, a fortiori, the
assumptionimplicit in these calculationsthat
none of the particles.seenin the electronmicro-
graphs of sample 4 are SPM at 77øK. Values
of d, at 77øK are indicated in Figure 7. These
were computed, with measured values of Ha
and J.. at 77ø and 300øK, from the standard
N•el [1949] null field relaxation equation
a2(Y)Z•(Y)•(Y)
- Q-]- loget
2kT
in which k is Boltzmann's constant,Q - 25,
and t is a typical experimentaltime. Only 1-2%
of the m.r distributions(Figure 8) lie below
d, (77). Strictly speaking, all d, (300) values
should be increasedslightly as a result, but
the correction is a small one, certainly much
TABLE3. Values of SPMTransition Size dz and
Critical SD Size do for Equant Particles
of Nonstoichiometric Magnetite
dz,
Model A A
A 360 ñ 50
B 330 ñ 50
C 290 ñ 50
in all d, and do values.The possiblepresence
of particles below 100 A in size, which are
poorly resolvedby the electron microscope,is
another potential sourceof error but probably
a minor one, since Figure 8 suggeststhat the
volume fraction of such particlesis very small.
DISCUSSION
Despite the rather dissimilarmodelsused to
interpret the experimental results, the com-
puted values of the room temperature SPM
threshold of magnetite fall in a rather narrow
range, from 290 to 360 A. These figures are in
good agreementwith the value of about 300 A
predicted by the N•el [1949] theory of mag-
netization relaxationin SD particles (equation
5).
The value of the SD-MD transition is less
well defined by the experimentalevidence.A
direct MD-SPM transition (i.e., do < d,) is a
distinctpossibilityin equantmagnetiteparticles.
If a stable SD range existsin such particles,
it is certainly very narrow; the maximum value
of doconsistentwith the experimentaldata was
480 ñ 50 A, and the corresponding value of
was 290 ñ 50 A.
The necessityof using true saturation
values in calculations of do and d, should be
emphasized.As Table 2 illustrates,J,dJ changes
by about 20%, and the fractional difference
between Jr•/J values at 77ø and 300øK (which
is attributed to SPM particles in the calcula-
(5)
tions) by as muchas'2%, between1500oe and
saturation hysteresismeasurements.A prelimi-
nary report [Dunlop, 1972] of model C calcula-
tions in which nonsaturation J•.8/J data were
used gave do - 570 A and d• = 350 A. These
values are significantlyhigher than those ob-
tained in the final calculations.
Given the restrictedrange of grain sizesover
which equidimensional magnetite particles can
exhibit stableSD behavior,the role played by
elongatedmagnetiteparticles.takes on particu-
lar importance.If the high-stabilitycomponent
of natural remanencein igneousrocksis carried
do, by true SD particles (and there is evidence
that it is in some rocks at least [Larson et al.,
1969; Evans and McElhinny, 1970; Hatgraves
<360
<330 and Young, 1969; Dunlop and West, 1969]),
480 ñ 50 elongatedparticlesappear to be the only likely
candidate.There is virtually no experimental
1792 DUNLOP' THRESHOLD SIZES IN MAGNETITE

indication of the appropriate do values for The do value falls in the range 3000-7000 A
magnetite particles with axial ratios different calculatedby Sorrel[1971] for titanomagnetite
from 1, but the do value reported here for of this composition.The d, estimateis a maxi-
equant particlesof magnetiteat least servesas mum value, which would decreaseradically in
an anchor point for theoretical curves of do the likely event that crystallineanisotropy
versus axial ratio. rather than shape anisotropy predominatesin
Finally, the doand d8valuesreportedherefor SD particleswith J, as low as 100 emu/cm•.
Fe80, can be usedin conjunctionwith theoreti- Whatever the precise anisotropy may be,
cal ideas to predict d8 and dovalues for Ti-rich it is apparent that the stable SD range is
titanomagnetites,which occur frequently in broaderand occursa.t largergrainsizein mag-
rocks. Stacey [1963], followingKittel [1949], netites with an appreciableTi content. On the
derivesthe equation other hand, it should be kept in mind that
Ti-rich natural titanomagnetitesare not neces-
do02 A1/2K1/2/J•:• sarily, or even typically, single phase. Inter-
whereA and K are the exchangeand crystalline growthsof a Ti-poor phase,near Fe•O, in com-
anisotropyconstants,respectively.BecauseA is position,and an only slightly magneticTi-rich
proportional to Curie temperature, which de- phase are common, and in this case do and d,
creaseswith Ti content in roughly the same as reported in this paper are the relevant
way as J., Av• 02 J? approximately.Stacey thresholdparameters.
further assumes that K 02 J.• and thus predicts Acknowledgments. I am grateful to Dr. H.
that do 02 J?•. Now K 02J.• is certainlytrue Kinoshita, Tokyo University, for the electron
for shape anisotropy,but it is not true for micrographs,to Dr. T. Takada, Kyoto University,
crystalline anisotropy, which is the relevant for the chemical analyses and BET particle size
data, and to Dr. S. K. Banerice, University of
anisotropyin calculatingenergiesof postulated Minnesota, for extending the remanence measure-
domain structures above the SD-MD transition. ments to 4.2'K. At the University of Toronto
If K• remainsrelatively constantas Ti content A. C. Dubey measured the X ray diffraction pro-
varies, we find that do 02 J.-•/•. (Syono [1965] files, and Dr. D. W. Strangway offered the use
of his vibrating sample magnetometer for the
has shownthat in fact the room temperature low-temperature measurements.Dr. R. T. Merrill
value of K• has a complexdependenceon Ti of the University of Washington offered useful
content and decreases from about -1.3 X 10• criticism.
ergs/cm• for pure magnetite to a minimum of This research was supported by the National
Research Council of Canada through grant A7709.
about -2.3 X 10• ergs/cm• for 10 mole % Ti
and subsequentlyincreases.
The assumption of ]•EFERENCES
constantK• is not grosslyin error for many Adamson, A. W., Physical Chemistry o] Surfaces,
compositions.) pp. 586-588, Interscience, New York, 1967.
The SPM thresholdvolume,accordingto (5), Banerice, S. K., New grain size limits for palaeo-
variesas K•-• (i.e.,asJ,-•), providedthat shape magnetic stability in haematite, Nature Phys.
anisotropypredominates in Ti-rich SD particles Sci., 232, 15-16, 1971.
Bartram, S. F., Crystallite-size determination from
as it does in the magnetite particles used in line broadening and spotty patterns, in Hand-
the samplesdescribedin this paper. Thus d, book o] X-Rays, edited by E. F. Kaelble, chap.
02 Zs-•/a. 17, pp. 17.1-17.9,McGraw-Hill, New York, 1967.
Consider,for example,a titanomagnetiteof Bean, C. P., and I. S. Jacobs, Magnetization of a
dilute suspensionof a multidomain ferromag-
compositionFe•.•sTio.•O•,which would have J, netic, J. Appl. Phys., 31, 1228-1230, 1960.
• 100 emu/cm8 at room temperaturein com- Berkowitz, A. E., and W. J. Schuele, Magnetic
parisonwith about 500 emu/cm8 for Fe30•.The propertiesof someferrite micropowders,J. Appl.
data of Syo• [1965] indicate that K• for this Phys., 30, 134S-135S, 1959.
compositionis approximatelyequal to K• for Colombo,U., F. Gazzarrini, G. Lanzavecchia,and
G. Sironi, Magnetite oxidation: A proposed
pure Fe304.We find for sucha composition mechanism,Science,147, 1033, 1965.
Dickson, G. O., C. W. F. Everitt, L. G. Parry,
do---•(500/100)s/2480 = 5300A and F. D. Stacey, Origin of thermoremanent
magnetization, Earth Planet. Sci. Lett., 1, 222-
d,- (500/100)2/s290 = 1400A 224, 1966.
 DUNLOP: THRESHOLD SIZES IN i•AGNETITE
Dunlop, D. J., Coercive force and remanence of
monodomain grains with mixed crystalline and
shape anisotropy, Phil. Mag., 21, 385-397, 1970.
Dunlop, D. J., Magnetic properties of fine-particle
hematite, Ann. Geophys., 27, 269-293, 1971.
Dunlop, D. J., Magnetite: Behavior near the
single-domain threshold, Science, 176, 41-43,
1972.
Dunlop, D. J., and G. F. West, An experimental
evaluation of single-domain theories, Rev.
Geophys. Space Phys., 7, 709-757, 1969.
Elder, T., Particle-size effect in oxidation of nat-
ural magnetite, J. Appl. Phys., 36, 1012-1013,
1965.
Evans, M. E., Single-domain particles and TRM
in rocks, Comments Earth Sci. Geophys., 2,
139-148, 1972.
Evans, M. E., and M. W. McElhinny, An investi-
gation of the origin of stable remanence in
magnetite-bearing igneous rocks, J. Geomagn.
Geoelec.,21, 757-773, 1970.
Evans, M. E., and M. L. Wayman, An investiga-
tion of small magnetic particles by means of
electron microscopy,Earth Planet. Sci. Lett., 9,
365-370, 1970.
Fletcher, E. J., and S. K. Banerjee, High tem-
perature dependenceof single-crystalanisotropy
constants of titanomagnetites. (abstract), Eos
Trans. A GU, 50, 132,1969.
Gallagher, K. J., W. Feitknecht, and U. Mann-
weiler, Mechanism of oxidation of magnetite to
.•-Fe20s, Nature, 217, 1118-1121, 1968.
Hargraves, R. B., and W. M. Young, Source of
stable remanent magnetization in Lambertville
diabase,Amer. J. Sci., 267, 1161-1177,1969.
Kittel, C., Physical theory of ferromagnetic do-
mains, Rev. Mod. Phys., 21, 541-583, 1949.
Kneller, E. F., and F. E. Luborsky, Particle size
dependence of coercivity and remanence of
single-domainparticles, J. Appl. Phys., 34, 656-
658, 1963.
Kobayashi, K., Magnetization-blocking processby
volume development of ferromagnetic fine
particles, J. Phys. Soc. Japan, 17(B1), 695-698,
1962.
Larson, E., M. Ozima, M. Ozima, T. Nagata,
and D. Strangway, Stability of remanent mag- Syono, Y., Magnetocrystalline anisotropy and
netization of igneous rocks, Geophys. J., 17,
263-292, 1969.
Morrish, A. H., and S. P. Yu, Dependenceof the
coercive force on the density of some iron Tonge, D. G., and E. P. Wohlfarth, The remanent
oxide powders, J. Appl. Phys., 26, 1049-1055,
1955.
Morrish, A. H., and S. P. Yu, Magnetic measure-
ments on individual microscopicferrite particles Verhoogen, J., The origin of thermoremanent
near the single-domain size, Phys. Rev., 102,
670-673, 1956.
Muirhead, F. R., Measurements of the remanent
magnetizationof someindividual magneticpar-
ticles near the single-domainsize, Phil. Mag.,
8, 1361-1368, 1963.
Nagata, T., K. Kobayashi, and M.D. Fuller,
Identification of magnetite and hematite in
] 793
rocks by magnetic observation at low tempera-
ture, J. Geophys. Res., 69, 2111-2120, 1964.
N.•el, L., Th.•orie du trainage magn.•tique des
ferromagn•tiques en grains fins avec applica-
tions aux terres cuites, Ann. Geophys.,5, 99-136,
1949.
N•el, L., Some theoretical aspects of rock mag-
netism, Advan. Phys., 4, 191-255, 1955.
Ozima, M., and E. E. Larson, Low- and high-
temperature oxidation of titanomagnetite in
relation to irreversible changesin the magnetic
properties of submarine basalts, J. Geophys.
Res., 75, 1003-1017,1970.
Ozima, M., and M. Ozima, Origin of thermo-
remanent magnetization, J. Ge'ophys.Res., 70,
1363-1369, 1965.
Parry, L. G., Magnetic properties of dispersed
magnetite powders, Phil. Mag., 11, 303-312,
1965.
Pauthenet, R., and L. Bochirol, Aimantation
spontan•e des ferrites, J. Phys., 12, 249-251,
1951.
Radhakrishnamurthy,C., S. D. Likhite, and N. P.
Sastry, Low temperature magnetic hysteresisof
fine particle aggregatesoccurring in some nat-
ural samples,Phil. Mag., 23, 503-507, 1970.
Sakamoto, N., P. I. Ince, and W. O'Reilly, The
effect of wet-grinding on the oxidation of
titanomagnetites,Geophys.J., 15, 509-515, 1968.
Sanver, M., and W. O'Reilly, Identification of
naturally occurring non-stoichiometric titano-
magnetites, Phys. Earth Planet. Interiors, 2,
166-174, 1970.
Softel,H. C., The origin of thermoremanentmag-
netization of two basalts containing homo-
geneous single phase titanomagnetite, Earth
Planet. Sci. Lett., 7, 201-208, 1969.
Softel, H. C., Single domain-multidomain transi-
tion in natural intermediate titanomagnetites,
Z. Geophys.,37, 45.1-470,1971.
Stacey, F. D., The physicaltheory of rock mag-
netism, Advan.. Phys., 12, 45-133, 1963.
Strangway, D. W., E. E. Larson, and M. Gold-
stein, A possiblecauseof high magnetic stability
in volcanic rocks, J. Geophys. Res., 73, 3787-
3795, 1968.
magnetostriction of Fe•O4-Fe2TiO• series with
special application to rock magnetism,Jap. J.
Geophys.,4, 71-143, 1965.
magnetization of single-domain ferromagnetic
particles, 2, Mixed uniaxial and cubic ani-
sotropies.Phil. Mag., 3, 536-537, 1958.
magnetization,J. Geophys.Res., 64, 2441-2449,
1959.
West, G. F., and D. J. Dunlop, An improved
ballistic magnetometer for rock magnetic ex-
periments,J. Sci. Instrum., 4, 37-40, 1971.
(Received July 11, 1972;
revisedNovember 13, 1972.)