SPWLA 40thAnnual Logging Symposium, May 30-June 3,1999

RESTRICTED DIFFUSION AND INTERNAL FIELD GRADIENTS

Matthias Appel, J. Justin Freeman, Rod B. Perkins
Shell E&P Technology Company, Houston, TX, USA
Jan P. Hofman
Shell International E&P RTS, Rijswijk, The Netherlands -

ABSTRACT gradients.Our measurementsconfirm recent theories

publishedin the open literature and may be of
Nuclear Magnetic Resonance(NMR) is typically significant importancefor the interpretationof
assumedto be lithology-independentsince the pore
wireline NMR data that is basedon the diffusion of
fluids are the only sourceof the measuredsignal.
pore fluids over extendeddiffusion times.
However, due to interactionsof the pore fluids with
rock surfaces,the rock matrix can significantly
influence the fluid response.Although these
interactionsmay seriouslycomplicatethe INTRODUCTION
interpretationof the data, they also provide useful Nuclear Magnetic Resonance(NMR) has developed
information about the pore structureof the rock. into a powerful petrophysicaltool for reservoir
characterization(Kenyon, 1997).In particular,
Two primary mechanismsthat complicate
diffusion-basedNMR techniqueshave beenfound
interpretationof NMR data (wireline logs or core
useful for pore fluid identification due to the inherent
analyses)are: (1) strong internal gradients;and (2)
contrastin the molecular mobility of gas, oil and brine
restricteddiffusion
(Akkurt et al., 1989).
Under laboratory conditions, measuredin a
NMR experimentsbecomesensitiveto the molecular
homogeneousbackgroundmagneticfield, internal
diffusion of the pore fluid if the applied magnetic
magnetic field gradientscan be calculatedfrom the
field is not homogeneousbut possessesa gradient
increaseof the transverserelaxationrate of pore fluids
over length scalesthat are comparablewith the
with increasinginter-echotime. Whereasa linear
dephasinglength of the spins, i.e., the typical
relation betweenthe increaseof the relaxationrate
displacementlength of fluid moleculesbefore they
and the squaredinter-echotime exists for the caseof
dephaseduring the time of the NMR experiment.
unrestrictedmolecular diffusion of the pore fluid,
deviationsfrom this linear relation can be expected Interpretationof NMR data relies on knowledge of
due to effects of restricteddiffusion and a pore size the gradientin which the pore fluids reside.Typically,
distribution causing a distribution of magneticfield this gradientis that applied by the wireline tool or
gradients. laboratoryspectrometerand henceis well
characterized.However, in porous media, magnetic
We report results of laboratoryNMR experimentson
susceptibility differencesbetweenthe solid phaseand Ff=
core samplestaken from the Gulf of Mexico and from
the fluid filling the pore spacemight lead to
the Far East. For thesesamples,the calculation of
significant magneticfield inhomogeneities,the so-
internal magneticfield gradientswas complicatedby
called internal magneticfield gradients.It is
the effects of restricted diffusion. Whereasthe
commonly assumedthat thesefield gradientsare
reduction of molecular mobility due to restricted
diffusion was sufficient to explain the measureddata mainly causedby paramagneticions such as iron,
in the range of inter-echotimes interestingfor nickel or manganesewhich are frequently found in
clays (Kleinberg et al., 1994). Internal magneticfield
standardNMR logging procedures,an additional
gradientsare proportional to the applied magnetic
reduction of the effective internal magneticfield
field. When thesegradientsare larger than the
gradienthad to be assumedto interpret the measured
data for a wider rangeof inter-echotimes. This gradientapplied by logging tools, interpretationof
wireline NMR data is considerablymore complicated.
observationwas supportedby separatePulsedField
GradientNMR diffusion measurementsusing pulse Restricteddiffusion occurs when the size of the pores
sequencesthat minimize the effects of internal becomessmaller than the Brownian motion of the

1
SPWLA 40Annual Logging Symposium, May 3OJune 3,1999
fluid molecules.In a homogeneous,unconfined
system,the random motion of moleculesgives rise to
a probability distribution of molecular displacements
which is determinedby a Gaussian.Deviations from
this Gaussianprobability function can be expectedif
subsequentmolecular displacementsare not
independentof eachother but correlatedwith regard
to length or direction of the elementaryjump process.
Such a memory effect may be brought about by the
moleculesthemselves,e.g., in polymers, or by
physical confinementsof the diffusion processin
porous system with length scalesless than about 100
pm, such as in sedimentaryrocks. In this situation,
the maximum displacementof the diffusing molecules
will be determinedby the length scaleof the matrix
rather than by the time allowed for the moleculesto
propagate.This processis known as anomalous
diffusion and is characterizedby a diffusion
coefficient that is smaller than that of the bulk pore
fluid and is a function of the experimentdiffusion
time.
Since knowledge of the diffusion coefficient is
required to analyzewireline NMR data,restricted
diffusion may significantly complicate evaluation.
However, given the fact that propagatingmolecules
scanthe entire pore space,restricteddiffusion may
also representa powerful tool to derive information
about the structureof the confining matrix.
In the following paper we will presentresults of
laboratoryNMR experimentson core samplestaken
from the Gulf of Mexico and from the Far East. For
thesesamples,we found significant internal magnetic
field gradientsthat exceedthose applied by wireline
tools or standardbenchtopNMR analyzers.The
interpretationof the measureddata was complicated
by effects of restricteddiffusion of the pore fluids.
We will support our conclusionswith additional
PulsedField Gradient (PFG) NMR experimentsusing
a pulse sequencethat minimizes the effects of internal
magneticfield gradients.
EXPERIMENTAL
Sevensandstonesfrom the Far East and one siltstone
from the Gulf of Mexico were investigatedusing
commercial low-field benchtopNMR analyzers
working at proton resonancefrequenciesof 1 and 2
MHz, respectively.Prior to the NMR measurements,
the samplesfrom the Far East were vacuum- and
subsequentlypressure-saturated at 30bar with 50,000
ppm NaCl brine. The Gulf of Mexico samplewas
saturatedin a similar way with 220,000ppm NaCl
brine. To further investigatethe producibility of the
2
formation, movablebrine was later displacedby a
light mineral oil (Soltrol) at 100 psi capillary
pressureequivalent.For this sample,only the
relaxation-and diffusion propertiesof the mineral oil
will be discussedin this paper.
All measurementswere performedat room
temperatureand normal atmosphericpressure.The
core plugs were wrappedwith NMR-silent material
using a testedprocedureto avoid any loss of
saturationduring the experiment.The sample
nomenclatureand standardpetrophysicalparameters
are summarizedin Table I.
The spectraof the transverserelaxation time, T2, of
the sampleswere determinedusing a Carr-Purcell-
Meiboom-Gill (CPMG) pulse sequencewith varying
echo spacingsbetween200 ps and 4 msec.The
experimentswere done without external magnetic
field gradients.For eachsample,the delay time
betweensubsequentpulse sequenceswas chosento be
at least five times longer than the longest component
of its longitudinal (Ti) relaxationtime spectrumto
assurea completepolarization for eachexperiment.
CPMG signal decaycurves were then inverted to
distributions of relaxation times using a distributed
exponentialfitting procedurewith an optimization of
the regularizationparameteraccordingto the signal-
to-noise of the experiment.To obtain the value of the
peak position independentof any bias due to the
incrementingTz-valuescalculatedby the fitting
routine, ten data points aroundthe centerof the
investigatedrelaxationtime peak were fitted to a log-
normal distribution and the position of this log-
normal peak usedin the further evaluationof the data.
As a comparisonto the samplesinvestigatedand to
assurethat the observedshifts of the relaxation time
peaks were not causedby any inhomogeneityof the
backgroundmagneticfield, a similar set of CPMG
experimentswas also performedon a cleanedand
water-saturatedBerea sandstone.
For both the Far East samplesand the Gulf of Mexico
core plug, the self-diffusion coefficient of the brine
and oil, respectively,in the pore spacewas
determinedusing PulsedField Gradient (PFG) NMR.
Becauseof the suspectedlarge internal magneticfield
gradients,standardPFG NMR approachessuch as
primary or stimulatedecho were not useful, and a
modified pulse sequence,the 13-interval,condition I
pulse sequence(Cotts et al, 1989)was applied. Figure
1 showsthe sequenceof RF- and magneticfield
gradientpulsesusedfor theseexperiments.
 SPWLA 40AMU~
Qualitatively, the effects of internal field gradientson
the echo amplitude are reducedby use of 180
refocusingpulsesand oppositelypolarized applied
field gradients,which lead to an accumulationof the
phaseencodingcausedby the applied gradients.In
contrastto this, the phaseencodingcausedby a
constantlypolarized internal magneticfield gradient
is refocused,and its effect is canceledout.
Experimentally, during the PFG NMR experimentthe
intensity of the NMR signal, il4, was monitored as a
function of the areaof the applied field gradients,Sg.
The PFG NMR experimentswere performedfor a
seriesof diffusion times, A, to monitor effects of
restricted diffusion which would be revealedby a
reduction of the measureddiffusion coefficient, D,
with increasingdiffusion time. The diffusion
coefficients were calculatedfrom the acquiredsignal
attenuationplots using the following relation (Cotts
et al, 1989):
M(g)
= M(g=O)
Normalization of the measuredsignal intensity, M(g),
to the signal intensity without applied field gradients,
M(g=O), eliminatesany effects of signal attenuation
due to relaxation processes.y denotesthe magneto-
gyric ratio. The pulse sequenceand the meaningof
the other timing constantsusedin Eq. (1) are
illustrated in Figure 1.
DISCUSSION
Shiji of Relaxation Peaks. Figure 2 showsthe Tz
spectraof the cleanedand water-saturatedBerea
sandstonesample.The inter-echospacing,TE, has
beenincreasedfrom 200 psec to 4 msecto test
whether any inhomogeneitiesof the magnetic
backgroundfield are present.Since the cleanedBerea
sandstoneis unlikely to producesignificant internal
magneticfield gradientsat low resonancefrequency,
any shift of the peak positions to shorterrelaxation
times with increasinginter-echospacingwould have
to be causedby diffusion within gradientsof the
external backgroundfield, BO.The inset in Figure 2
shows the position of the major peak around300
msec as a function of inter-echospacing.The Tz
spectraof the Berea sandstoneare not TE-dependent;
the slightly increasedrelaxationrate at very long
inter-echospacingsis not significant and is an artifact
of the inversion routine. The only significant
difference betweenthe spectraacquiredwith different
3
Logging Symposium, May 30-June 3,1999
inter-echospacingsis a decreasedsensitivity to fast
relaxationprocessesat long TEs due to the delayed
start of signal acquisition. The experimentson the
cleanedBereasandstoneexclude external field
inhomogeneitiesto be the reasonfor the observed
relaxationpeak shifts of the Far East and Gulf of
Mexico samples.
Figure 3 showsthree examplesof Tz-spectra
measuredfor a seriesof inter-echospacingsfor a Far
East sampleand the Gulf of Mexico core plug. In
contrastto the Berea sample,significant shifts of the
transverserelaxationpeaksare observed.Figure 3a
representsa set of experimentson a Far East sample
(sampleNo. I), measuredat 2 MHz. The inter-echo
spacinghas beenincreasedfrom 256 psec to 4 msec,
which covers the full rangeof inter-echospacings
typically usedfor the acquisitionof wireline NMR
data,including more recent approachessuch as
enhanceddiffusion techniques(Akkurt et al., 1998).
Figure 3b shows the peak shift of the Gulf of Mexico
samplewhere the movablebrine had beenreplacedby
mineral oil. Figure 3c illustrates the peak shift for the
samesampleshown in Figure 3a but measuredat 1
MHz.
For all samplesdiscussedhere, the NMR responseof
capillary boundbrine could be resolvedfrom that of
the movablefluids even at the longest measuredinter-
echo spacings.The responseof the capillary bound
brine appearsat longer relaxation times due to the
delayedstart of signal sampling.
In Figure 3b, the short Tz peak in the relaxation
spectracorrespondsto the NMR signal from capillary
boundbrine. The long Tz peak representsthe signal
from the mineral oil. The clear separationof the two
peaksallows us to establisha BVI cut-off value that
can be usedin the interpretationof wireline NMR
data.Note that only the signal from the movable FF
mineral oil is shifted to shorterrelaxation times with
increasinginter-echospacing,TE. The peak of the
capillary bound brine remainsalmost unchanged
exceptfor effects due to delayedsignal sampling at
the longestinter-echospacing.That meansthat the
TE-dependentdiffusion term does not significantly
contributeto the overal relaxation rate of the capillary
boundbrine. Since the strongestinternal magnetic
field gradientscan be expectedat the rock-fluid
interface,the stationarybrine peak is a clear
indication of restricteddiffusion: the molecular
diffusion of the brine moleculesin the bound water
layer is strongly restricteddue to chemical
interactions(first molecular water layers) and due to
SPWLA 40A~ual Logging Symposium, May 30-June 3,1999
structuralconfinement.Theseconstraintsresult in a
strongly reducedeffective diffusion coefficient of the
capillary bound brine which explains its static peak
position despitethe strong internal field gradientsthat
the capillary bound brine is expectedto experience.
Evaluation of the Observed Peak Shifts. The overall
rate of the transverserelaxation, Tz, of the pore fluid
in a porous system is the sum of the bulk, surface,and
diffusion relaxation rates,
T; = T<&, + +$y2g2TE2D.
(2)
For the surfaceterm, p denotesthe surfacerelaxivity
and S/V the surface-to-volumeratio of the pore
structure.
To calculatethe intensity of the internal field
gradients,for each inter-echospacingthe shift of the
position of the movable peak (either brine or mineral
oil) relative to its position at the shortestmeasuredTE
was plotted as function of the difference betweenthe
squaredrespectiveinter-echospacingand the squared
shortestmeasuredTEL Since only the contribution due
to diffusion dependson TE (Eq. 2), plotting the
experimentaldata in this way separatesthe relaxation
due to diffusion from the bulk- and surface
relaxations.However, including bulk- and surface
relaxation into the evaluation of the data would
simply causea vertical shift in a T2-1versusTE plot.
Figure 4 shows the shift of the position of the
movable brine T2 peak of sample#l when the inter-
echo spacingwas increasedfrom 0.256 msecto 4
msec.This plot is a representativeof all data
evaluated;plots of similar shapebut different slopes
accordingto the respectivefield gradientsof the
sampleswere obtainedfor all samples.The error bars
representan estimatedexperimentaluncertaintyof
10%.
In this diagram, a peak shift causedby fluid
moleculesdiffusing freely within a constantmagnetic
field gradient would be representedby a straight line.
Clearly, the measureddata significantly deviate from
this relation, which, accordingto Eq. 2, has to be
brought about by a decreasingproduct of the diffusion
coefficient and squaredgradient strengthwith
increasinginter-echotime. However, the first six data
points in Figure 4 appearto follow the linear relation
expectedfor free diffusion within a constantfield
gradient.This behavior was found for all samples
measured.For sample#I illustrated in Figure 4, a
4
linear fit of the first 6 data points to a free diffusion
model results in an internal field gradientof 90 G/cm
(dotted line).
Free diffusion for very short diffusion times, i.e.,
inter-echospacingsin a CPMG experiment,is not
unexpectedsince most of the moleculesof the
movable fluid fraction will not initially experiencethe
surroundingbarriersinitially. Only if the majority of
moleculeshas interactedwith the walls of the matrix,
significant deviationsfrom the Gaussianprobability
distribution of moleculardisplacementswill occur. As
a result, deviationsfrom unrestricteddiffusion will be
detected.
DiSfirsion Models. The reduction of the diffusion
coefficient of the pore fluids due to effects of
restricteddiffusion dependson the pore size and is, in
principle, unknown. Mitra et al. have developeda
model to describethe short-timebehavior of restricted
diffusion of fluids in porous media with smooth
boundaries(Mitra et al., 1993):
In Eq. 3, S/V denotesthe surface-to-volumeratio of
the pore structure,Do the diffusion coefficient of the
bulk fluid, and A the diffusion time. This model is
consistentwith previous experiments;however, the
dam measuredat 2 MHz show that a & -dependence
of the restricteddiffusion coefficient is not sufficient
to explain the observedpeak shifts in the investigated
time interval.
Although a straightforwardderivation of the internal
field gradientsfrom the experimentaldata was not
possible,an upper limit of the gradientscan be
estimatedusing a generalizedmodel for the
diffusional behavior of the movablepore fluids.
Following the conceptof diffusion under the
influence of confinementwithin a fractal structure,
the diffusion of the pore fluids may be expectedto
follow a relation of the type
(4)
with < Y2 > denotingthe molecularmean-square
displacementduring the diffusion time, A. The time
exponent,K, representsthe fractal dimensionof
propagation;if x=1, unrestricted,or free diffusion
will occur, whereastime exponents~1 are causedby
restrictions to the diffusion process(Avenir, 1989).

Eq. 4 can be transformedinto the time-dependenceof
the diffusion coefficient,
D(A)- AK- , (5)
which meansthat free diffusion is characterizedby a
time-independentdiffusion coefficient whereas
restricteddiffusion is indicatedby a diffusion
coefficient that decreaseswith increasingdiffusion
time.
Obvioulsy, any diffusion processof fluids in a porous
structure,will start out as free diffusion given
sufficiently small diffusion times. This behaviouris
indicated by the first six data points in Figure 4 as
discussedabove.However, a diffusion coefficient that
is independentof diffusion time will also be found at
very long diffusion times when the diffusing
moleculeshave propagatedover displacementsthat
are large comparedto the typical length scalesof the
pore structure.In a statistical sense,molecular
diffusion in subsequenttime intervals is then
subjectedto identical conditions so that a reducedbut
time-independentdiffusion coefficient will be
observed.This processis known as long-range
diffusion and allows us to calculatethe tortuosity of
the diffusion path (and hence,of the pore structure)
from the reduction of the diffusion coefficient
comparedto its bulk value (Coateset al., 1993).
Between thesetwo limits of time-independent
diffusion, the measureddiffusion coefficient will be a
function of diffusion time. The strongestreduction of
the measureddiffusivity can be expectedfor the case
of completely restricted diffusion within closed pores.
In this situation, the maximum mean-square
displacementwill be only deteterminedby the pore
structurebut not by the diffusion time. Subsequently,
accordingto Eq. 4, the time exponent K has to be
equal to zero, which translatesinto a diffusion
coefficient that is inversely proportional to the
diffusion time.
Combining the measuredshifts of the relaxation
peaksand the theory describedabove,the following
diffusion model can be madeto estimatean upper
limit of the internal field gradientsfrom the
experimentaldata:
1. The diffusion processof the movablepore fluid
startsout unrestricted.The time-independent
diffusion coefficient, in combinationwith an
averaged,constantmagneticfield gradientresults
SPWLA 40thAnnual Logging Symposium, May 30-June 3,1999
in a linear relation betweenthe peak shift and the
differenceof the squaredinter-echospacing.
2. At longer diffusion times, restricteddiffusion will
occur. The plot of peak shift versusdifference of
squaredinter-echospacingswill have a smaller
and constantlydecreasingslope due to the TE-
dependenceof the restricteddiffusion coefftcient.
Assuming the caseof completely restricted
diffusion, the diffusion coefficient scaleswith
TE*in this interval.
3. At very long diffusion times, time-independent
diffusion will again take place. If only one
(averaged)field gradientgovernsthe shift of the
relaxationspectra,the plot of peak shift versus
squaredTE-increasewill in this interval keep the
smallestslope of interval 2 constant.The cross-
. over from time-dependentrestricteddiffusion to
time-independentlong-rangediffusion depends
on the tortuosity (and, hence,permeability) of the
rock matrix.
This diffusion model in principal only estimatesan
upper limit for the internal field gradient.Becauseof
the finite permeability of the samples,completely
restricteddiffusion for all fluid moleculesis not likely
to occur.Therefore,the coefficient of restricted
diffusion will be somewhatless dependenton
diffusion time (i.e., less reducedfrom the bulk fluid
value) than the theoreticallimit assumedabove. Since
the product g2D determinesthe magnitudeof
relaxationpeak shift at a given TE, an
underestimationof the diffusion coefficient, D, results
in an overestimationof the gradient,g.
For all samplesinvestigatedat 2 MHz resonance
frequency,such a diffusion model was applied to
calculatethe internal field gradients.For the
evaluationof the 1 MHz data, insufficient data points
were availableto establishthe crossoverfrom free to FF
restricteddiffusion at short inter-echospacingsand a
model basedon the reduction of the restricted
diffusion coefficient accordingto Eq. (3) was applied
to allow comparisonof the data. The difference
betweenthe results obtainedwith thesetwo models
will be discussedlater.
The coefficient of free diffusion was measured
seperatelyby PFG NMR for both brine and mineral
oil. The cross-overfrom free diffusion at short times
to restricteddiffusion at longer TEs was determined
from the measuredpeak shifts as the points at which
deviationsfrom the linear relation betweenthe peak
shift and the difference of the squaredinter-echo
SPWLA 40Aunual Logging Symposium, May 30-June 3,1999
spacingfirst occurred.From this point onward,
compIetelyrestricted diffusion was assumedfor the
remaining range of inter-echospacings.
Figure 5 illustrates the diffusion model applied to
sampleNo. 1. This diagram also contains PFG NMR
dam that were made to supportthe chosendiffusion
model. Unfortunately, for the samplesfrom the Far
East it was not possible to detect~atime-dependent
coefficient of restricteddiffusion since the shortest
diffusion time that could be realized experimentally
using the 13-intervalpulse sequenceof Cotts et al.
was 9 msec. Obviously, the crossoverto restricted
diffusion occurs at earlier diffusion times and only the
long-rangediffusion coefficient could be measured.
Nevertheless,the experimentaldata fit well into the
diffusion model since a smoothtransition from the
experimentallydeterminedvalues for free and long-
rangediffusion can be achievedassumingcompletely
restricteddiffusion in the time interval not accessible
by the experiment.
The result of applying the diffusion model outlined
aboveto evaluatethe internal field gradientsfrom the
measuredpeak shifts is illustrated as the thin solid
line in Figure 4. The initial slope of the peak shift of
this samplecorrespondsto an internal field gradient
of 90 G/cm. We found that even the maximum
theoretically possiblereduction of the diffusion
coefficient that was assumedin our diffusion model
was insufficient to describeover the entire rangeof
measuredinter-echospacingsthe strong deviation of
the measureddata from a linear relation. Since the
initial value of 90 G/cm for the internal field gradient
is required to fit the first half of the data points, the
only possible way to explain the measuredpeak shift
over the entire TE-rangeis to assumea further
reduction of the internal field gradientat longer inter-
echo spacings.For sample#l, a gradualdecreaseof
the gradientfrom 90 to 77 G/cm was sufficient to fit
the measuredpeak shifts. This behavior was found for
both the Far East and the Gulf of Mexico samples.
The determinedgradientintensitiesrangefrom about
40 to 90 G/cm with a lo-15% decreaseat long times.
A summaryof the determinedintensity rangeof the
internal field gradientsof all samplesmeasuredis
given in Table I.
An even strongerreduction of the field gradients
would be necessaryto explain the measureddata if
completely restricted diffusion was not assumedin
the evaluation of the experiments,i.e., the diffusion
model usedprovides an upper limit of the field
gradientsand a lower limit of the gradientreduction
at longer TEs.
The maximum calculatedvalues of about 100 G/cm
are surprisingly high for internal field gradients,
especiallyconsideringthe low applied magneticfields
usedin the experiments.Hirasaki et al. reported
valuesof internal field gradientsfor pure chlorite /
brine systemsof about 300 G/cm (Hirasaki et al.,
1998).To understandthe existenceof this
susceptibility contrastbetweenthe fluids and rock
grains,the crystalline componentsof the Gulf of
Mexico samplewere determinedby X-ray diffraction
analysis.
Table II showsthe mineralogicalcomposition as
weight percentin crystalline portion for this sample.
The given uncertaintyof the data representsthe
estimatedstandarddeviation at the 95% confidence
level. Additional X-ray diffraction analysisof the clay
mineralsprovided the clay composition listed in
Table III. However, this analysiscovers only clay
minerals in the particle size rangebetween0.26 to 4.7
microns. Since clay minerals in the samplemay fall
outsidethis range,the results of the clay analysisare
not necessarilycomparableto the analysisof the bulk
samples.
The clay content of 11+5% of the Gulf of Mexico
sampleis relatively high. Internal field gradientsare
most likely causedby chlorites becauseof its pore-
lining morphology and its high iron content after
isomorphoussubstitutionof Mg2by Fe3.The fact
that only minor amountsof Fe-chlorite were found in
the Gulf of Mexico sampleis, however, consistent
with the young age of this reservoir which has not
beendeeply buried or subjectedto high temperature.
It is known that the chlorite content increasesin
sandstonesthat have undergoneburial diagenesisat
temperaturesgreaterthan lOO-120C.Nevertheless,
the internal gradientscan be explainedby the
presenceof a significant amountof mixed-layer clays
(smectite/illite) with a large smectitecomponent.
Smectitesare extremely variable in composition and
readily adsorband exchangecations.Therefore,the
large gradientsobservedin the Gulf of Mexico
sampleare most likely causedby a combination of
small pore geometriesand high content of mixed-
layer clays.
Table I also comparesthe results of the gradient
determinationfrom the data measuredat 1 and 2 MHz
proton resonancefrequency.Theoretically, a linear
relation betweenthe internal magneticfield gradients
and the applied backgroundmagneticfield should
 SPWLA 40AMWI
exist. Comparingour data measuredat 1 and 2 MHz,
respectively,it should be rememberedthat the 2 MHz
data were evaluatedusing a model that tendsto
overestimatethe field gradients.In contrastto this, the
model usedto calculatethe gradientsfrom the 1 MHz
data will underestimatethe gradientssince it assumed
a constant fi - dependenceof the restricteddiffusion
coefficient over the entire rangeof inter-echo
spacings,Additionally, the 1 MHz data should be
comparedto the lower limit of gradientsmeasuredat
2 MHz since this value better representsthe 1 MHz
model than the maximum 2 MHz value that mainly
correspondsto the fluid fraction diffusing freely.
Consideringthesedifferences,the typically observed
factor of three insteadof two betweenthe data sets
might representthe uncertaintiesof both the
experimentand the evaluationproceduresutilized.
Figure 6 illustrates the dependenceof the calculated
internal magneticfield gradientas function of
porosity and permeability for the samplesfrom the
Far East. Clearly, the highest internal field gradients
can be expectedfor the sampleswith the lowest
porosity and permeability becauseof more
pronouncedinteraction of the rock matrix with the
pore fluids. The internal field gradientsare more
strongly correlatedwith samplepermeability than by
sampleporosity, which confirms the perceptionthat
the strongestfield gradientswill be presentat the pore
throats.
Comparison to Recent Theories. Problemsrelated to
internal magneticfield gradientsin inhomogeneous
samplesinvestigatedby NMR have beendiscussedin
the open literature for natural and artificial porous
materials,polymers and biological samples
(Htirlimann, 1998; Geschkeet al., 1991;Packer,
1973).
Glasel et al. have shown that the magneticfield
gradient,G, outside of a spherewith radius r. can be
approximatedfor a region close to the sphereas
G = kHoAX
(6)
4ro
where ,uodenotesthe magneticpermeability of
vacuum,HO the applied magneticfield strengthand
Ax the difference betweenthe magnetic
susceptibilitiesof the sphereand the external medium
(Glasel et al., 1974).
7
Logging Symposium, May 30-June 3,1999
In their discussionof transverserelaxation under the
influence of magneticfield gradients,Hiirlimann et al.
describedthe NMR responseof fluids in porous
structures.It was proposedthat the decay of the NMR
spin echo is governedby the interplay of three length
scaleswhich were referredto as:
1. diffusion length, ld Id = fi,
2. size of pore structure,lS,
3. dephasinglength, &, 1, = m.
For eachratio of combinationsof theselength scales,
different attenuationsof the NMR signal as function
of diffusion time and gradientintensity can be
expected.Additionally, Htirliman introducedthree
asymptoticregimes,thefree difJsion, the motional
averaging, and the localization regime. In the
motional averagingregime, the pore size is the
shortestlength scale,so that the diffusing spins
averageall field inhomogeneitieswithin one pore and
the NMR signal decaydependson the local geometry
only but not on diffusion time. In contrastto this, the
localization regime appliesin large enoughpores
when the dephasinglength is shorterthan the pore
structure(Htirlimann, 1998).
In the analysisof our data,the cross-overfrom the
free diffusion to the motional averagingregime has
beenincorporatedinto the time-dependenceof the
restricteddiffusion coefficient. Our experiments
clearly show that severaldiffusion mechanismshave
to be consideredin the evaluationof the measured
data.It is important to note that for any given
diffusion time, there must be a different distribution
of effective gradients:at long times, spins in smaller
poreshave lost their phasecoherencedue to surface
relaxationand only the spins in larger pores survive.
Therefore,the distribution of gradientsat longer times
should be dominatedby smaller gradients.In contrast
to this, spins in both small and large pores contribute
to the NMR signal at shortertimes, which results in a
larger effective field gradientthat governsthe
relaxationof the spins due to their diffusion
Implication to the Interpretation of Wireline NMR
Data. Fluid identification that relies on diffusion-
basedNMR techniquesis only robust if the gradient
applied by the logging tool dominatesthe dephasing
of the spins, i.e., the internal gradientsdue to
susceptibility contrastsare negligible. Although
carbonatesand pure quartz are diamagnetic,
sandstonesmight be paramagneticdue to
paramagneticimpurities in clays and may possess
strong internal field gradientseven at low resonance
SPWLA 401hAMUZ~ Logging Symposium, May 3OJune 3,1999
frequencies.Especially in formations where small
pore sizes and high clay content is expected,fluid
identification derived from diffusion coefficients may
be problematic.
Our dam show that for the samplesinvestigated,
significant reductionsof the magneticfield gradients
are observedfor inter-echospacingslonger than 1.2
msec.In this case,only restricteddiffusion within a
constantfield gradientwould have to be consideredin
the evaluation of the dam for most standardwireline
NMR applications.Deviations from a constant
gradientmight be of significance if Tz relaxation
spectraare obtainedwith longer inter-echospacings,
such as in the enhanceddiffusion method.
The magnetic susceptibility contrastand, hence,the
correspondinginternal magneticfield gradientswill
be smaller at reservoir temperaturethan at ambient
conditions. Since water is diamagentic,and
diamagnetismis not temperature-dependent, only the
changeof the magneticsusceptibility of the
paramagneticgrain and clay materialswith
temperaturehas to be considered.For low magnetic
fields and high temperatures,the temperature
dependenceof paramagneticmaterialscan be
describedby a Curie-law,
where C denotesthe material-specificCurie constant
and T the absolutetemperature.
We have estimatedthe temperaturedependenceof the
magnetic susceptibility from data publishedfor
sandstones(Htirlimann, 1998) and for pure clays
(Mattesonet al., 1998).We found on averagea 20%
reduction of the magnetic susceptibility at a
temperatureof 200 F comparedto 75 F, which would
lead to the samereduction of the internal magnetic
field gradientsfor this temperatureincrease.
Additionally, the linear dependenceof the magnetic
field gradientson the external magneticfield of the
NMR logging tool has to be consideredwhen scaling
laboratory data measuredat ambient temperaturesto
reservoir conditions.
CONCLUSIONS
We have presentedexperimentalresults on transverse
relaxation time and diffusion measurementson core
samplestaken from the Far East and from the Gulf of
8
Mexico. For all samples,we found significant internal
magneticfield gradientsthat exceedthose applied by
NMR logging tools, and were able to characterize
thesegradientsusing laboratoryNMR measurements.
Since fluid identification using diffusion-basedNMR
techniquesrelies on knowledgeof the applied field
gradients,laboratory measurementsmay be required
to supportinterpretationof wireline NMR data.
Effects of restricteddiffusion complicatedthe
interpretationof our data. The experimentshave
shown that it might be sufficient to assumerestricted
diffusion within an effective, but constantmagnetic
field gradientfor the rangeof inter-echospacings
typically applied in standardwireline NMR programs.
However, if the rangeof inter-echospacingsis
extended,more complicateddiffusion- and gradient
models may be required.
ACKNOWLEDGEMENTS
The authorswish to thank Wim J. Looyestijn, Walter
F.J. Slijkerman and Yakov Volokitin, Shell
InternationalE&P for helpful discussionsand
reviewing the findings of this paper.We also wish to
thank Wim J. Looyestijn for supplying the samples
from the Far East.
REFERENCES
Akkurt, R., Mardon, D., Gardner,J.S., Marshall,
D.M., Solanet,F., 1998,Enhanceddiffusion:
expandingthe rangeof NMR direct hydrocarbon-
typing applications,Transactions of the SPWLA 39h
Annzeal Logging Symposium, Keystone,Colorado,
May 26-29, 1998,paperGG.
Avenir, D. (Editor), 1989, Thefractal approach to
heterogeneous chemistry: su$aces, colloids,
polymers, Wiley, Chichester.
Coates,G.R., Vinegar, H.J., Tutunjian, P.N., Gardner,
J.S., 1993,Restricteddiffusion from uniform
gradientNMR well logging, Transactions of the SPE
68 Annual Technical Conference, October 3-6, 1993,
paper26472.
Cotts, R.M., Hoch, M.J.R., Sun, T., Markert, J.T.,
1989,Pulsed field gradientstimulatedecho methods
for imoroved diffusion measurementsin
heterogeneoussystems,Journal of Magnetic
Resonance, 83, pp. 252-266.
 SPWLA 40& Annuai Logging Symposium, May SO-June 3,1999

Geschke,D., Fleischer, G., 1991, Dynamics of ABOUT THE AUTHORS

polyethylene melts as studied by nuclear magnetic
relaxation including effects of inner field-
inhomogeneities, Acta Polymerica, 42, No. 8, pp. Matthias Appel received a Ph.D. degreein Physics
from Leipzig University, Germany in 1995. Before
362-366.
joining Shell E&P Technology Company in Houston
Glasel, J.A., Lee, K.H., 1974, On the interpretation in 1997 he was a postdoctoral researcherand later
principal investigator at the Exxon Research&
of water nuclear magnetic resonancerelaxation times
Engineering Company in New Jersey.Matthias
in heterogeneoussystems,Journal of the American
primary interest and responsibility is the application
Chemical Society,.96, No. 4, pp. 970-978.
of NMR techniquesto investigate the interplay
Hirasaki, G.J., House, W.V., Zhang, G.Q., 1998, betweenthe geometry of porous media and the
confined pore fluids as well as the development of
Diffusion in internal field gradients, Transactions
laboratory NMR capabilities to support Shells NMR
of the I998 International Symposium of the Society of
Core Analysts, September14-16, 1998, The Hague,
logging program.
paper SCA 9823.
J. Justin Freeman is a Staff ResearchPhysicist for
Shell E&P Technology Company in Houston, Texas.
Htirlimann, M.D., 1998, Effective gradientsin
Sincejoining Shell in 1988 he has worked in the
porous media due to susceptibility differences,
fields of geophysics,acoustic rock properties,
Journal of Magnetic Resonance, 131, pp. 232-240.
compaction and permeability estimation. He has also
spenttime in operationsworking as lead
Kenyon, W.E., 1997, Petrophysical principles of
petrophysicist in Shell Offshore Inc.s Mars and
applications of NMR logging, The Log Analyst, 38,
No. 2, pp.21-43. Auger projects. He currently headsShell Oil Co.s
NMR researchactivities. Justin has BSc (1982) and
Kleinberg, R.L., Kenyon, W.E., Mitra, P.P., 1994, Ph.D. (1985) degreesin Physics from the University
of Leeds,England
Mechanisms of NMR relaxation of fluids in rock,
Journal of Magnetic Resonance, Series A, 108, pp.
Roderic B. Perkins received a B.S. degree in
206-214.
Electronics Technology from the University of
Houston in 1978. He joined Shell Development
Matteson, A., Tomanic, J.P., Herron, M-M., Allen,
Company in Houston in 1982. He initially worked in
D.F., Kenyon, W.E., T\TMRrelaxation of clay-brine
the marine seismic data acquisition section and later
mixtures, Transactions of the SPE Annual Technical
on several acoustics,laser and rock properties
Conference, SeptemberU-30, 1998, paper 49008.
projects. Rod is currently on the NMR team, and is
primarily involved with making laboratory
Mitra, P.P., Latour, L.L., Kleinberg, R.L., Sotak,
measurementsof rocks and fluids.
Ch.H., 1993, Time-dependent diffusion coefficient
of fluids in porous media as a probe of surface-to-
Jan P. Hofman joined Shell Researchin Rijswijk in
volume ratio, Journal of Magnetic Resonance, Series
1971. From 1980 to 1985 he worked at the Technical
A, 101, pp. 342-346. FF
University in Delft. He has worked in the
developmentof many core analysis techniqueswith
Packer, K.J., 1973, The effects of diffusion through
an emphasison electrical properties. Jan is currently
locally inhomogeneousmagnetic fields on transverse
responsiblefor the analysis and interpretation of
nuclear spin relaxation in heterogeneoussystems.
NMR core and fluid measurements.
Proton transverserelaxation in striated muscle tissue,
Journal of Magnetic Resonance, 9, pp. 438-443.

9
SPWLA 40Annual LoggingSymposium,May 30-June3,1999

I I Internal Field Gradient 1 Int. Field I

Sample Location Porosity Permeability Mealsured @ 2 MHz Gradient
No. WI Gmax GIllhI Ranae
--i_~ @ lMHz
bu.) bm (G/cm) (G/cm) (%I I (G/cm)
1 Far East 21.0 75 90 77 14.4 I 28
I 2 Far East 24.0 240 I 73 63 1 13.7 1 26

17.0 1 16

Table I: Nomenclature,basic petrophysicalpropertiesand calculatedinternal magneticfield gradientsof the

samplesinvestigated.

I I I
I Crystalline Component

70% Smektite/ 30% Illite

I
Weight%

48410
I

Illite 40fll
Chlorite-Fe 8&3
Kaolonite 3+1

Table III: M ineralogic compositionof the clay in the

Gulf of Mexico sample(sample8) as determinedby
X-ray diffraction. The uncertaintyof the data
representsthe estimatedstandarddeviation of the data
at a 95% confidencelevel. Only clay m ineralsin the
Illite 6*3 particle size rangebetween0.26 and 4.7 pm were
SmektiteAite 3+1 coveredby the analysis.

Table II: M ineralogic compositionof the Gulf of

Mexico sample(sample8) as determinedby X-ray
diffraction. The uncertaintyof the data representsthe
estimatedstandarddeviation of the data at a 95%
confidencelevel.

9o 1so0 90 9o 180 90 9o

field
gradient gradient I
Figure 1: Pulse sequenceapplied to measurethe self diffusion coefficients of brine
space.By use of bipolar magneticfield gradientsand 180refocussingRF pulses,this
condition I sequencedevelopedby Cotts et al. m inimizesthe effects of internal magneticfield
the signal attenuation.The crushergradientssupportthe balanceof the field gradients.
 SPWLA 40thAnnual Logging Symposium, May 30-June 3,1999

m
.-8 0.8
7

E OS4
= 0.2

0
1 10 100 1000 10000
T2 / msec

Figure 2: Spectraof the transverserelaxation time, T2, for a cleanedand water-saturatedBerea

sandstone.The inter-echospacing,TE, has beenvaried between0.2 msecand 4.0 msec.The
inset shows the position of the relaxationpeak relative to its position at the shortestTE. No
significant shift of the relaxation peak was observed.

100.
90
80
70

g 60 --
w

2 50
* 40
30
2c
q
FF
I(
4
1
22
- 23

Figure 6: Calculatedinternal magneticfield gradientsof the Far East samples.

11
SPWLA 40uIAmmal Logging Symposium, May 30-June 3,1999

Wis
z 1

2 iI8
cn
2 a6
.I
TE4.0 Ins

p
TE=36ms
co.2
TlS3.2 ms
0
TlSk2.8 ms
o.ctl a1 1 10 100 1mJ lam
T&2.4 m T2lmec

TE=ZOm

TFA8m

TFd.6 Ins

TE=lAm -+-TM.Oms
+TES?.Oms
TFA 2x11s

TFFl.0 Ins

TEM-J.8 Ins

TFA.6 m

TFGo.5 ms

Tk0.4 117s

0.01 0.1 1 10 im imo iowo

TJITEC

Figure 3: Examplesof transverserelaxation time, T2, spectrafor the samplesinvestigated.

Figure 3a: Far East sample#l, measuredat 2 MHz proton resonancefrequency.The inter-echo
spacinghas been increasedfrom 0.256 msecto 4 msec.

Figure 3b: Tzspectra for the Gulf of Mexico sample,measuredat 2 MHz resonancefrequency.The
inter-echospacingwas varied between0.28 msec and 2.048 msec.Note that for this sample,
movable brine has beenreplacedby mineral oil. The short T2 peaksin thesespectracorrespondto
the signal from capillary bound brine, whereasthe long T2 peaksrepresentthe NMR responseof the
mineral oil. The short T2 peaksdo not changedue to restricteddiffusion effects. The shift of the
mineral oil peakswas evaluatedto calculatethe internal magneticfield gradientfor this sample.

Figure 3c: T2 relaxation peaksfor the samesampleas plotted in Figure 3a but measuredat 1 MHz
proton resonancefrequency.The inter-echospacinghas beenincreasedfrom 0.32 msecto 4 msec.

12
 SPWLA 401hAnnual Logging Symposium, May 30-June 3,1999

Completelyrestricteddiffusion
Gradient gradually decreasing
from !?OGkm to 77 G/cm

0 2 4 6 6 10 12 14 16 II

TE* - TEo* / msec*

Figure 4: Shift of the position of the T2-relaxationpeak of sample#l. The peak shift has been
normalizedto the position of the relaxationpeak at 0.256 msec.This changein peak position is
causedby diffusion of the pore fluid within internal magneticfield gradients.Note the constant
slope of the first data points which reflects free diffusion within a constantfield gradientof 90
G/cm (dotted line). At longer inter-echospacings,effects of restricteddiffusion (thin solid line)
and, additionally, changesin the gradientintensity haveto be consideredin the evaluationof
the measureddata (bold solid line).

8
@ l.OE-09 -
N\
E
\
r3
FF

l.OE-10 -I I
0.1 1 10 loo

TE / msec

Figure 5: Diffusion model applied to interpretthe measuredpeak shift of sample#l. For inter-
echo spacings,TE, shorterthan 1 msec,free diffusion of the movable brine was assumed,
supportedby a linear slope of the measuredpeak shift in this interval (seeFigure 4). For
longer inter-echospacings,completely restricteddiffusion within closed pores was assumedto
estimatean upper limit of the internal field gradients.The experimentaldata have been
measuredwith PFG NMR and representthe long-rangediffusion behavior of brine through the
rock matrix. For all samplesmeasured,similar diffusion models have beenused.

13