GEOPHYSICS, VOL. 64, NO.2 (MARCH-APRIL 1999); P.470-478,11 FIGS., 3 TABLES.

Downloaded 02/18/13 to 141.117.79.62. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Permeability relation with other petrophysical parameters

for periodic porous media

Keh-Jim Dunn *, Gerald A. LaTorraca *, and David J. Bergman$

of the transport properties of rocks, the primary goal has been

ABSTRACT to estimate permeability from logs (Brown, 1972). NMR log
measurements have been used extensively to predict perme-
We modeled permeability (k) estimation based on ability, assuming that the dominant NMR relaxation time of
porosity (0), electrical formation factor (F), and nuclear fluid-saturated rocks is related to the surface-to-volume ra-
magnetic resonance (NMR) relaxation time (T), using tio of the pore space. It is the purpose of the present pa-
periodic structures of touching and overlapping spheres. per to re-examine these correlations for permeability using
The formation factors for these systems were calculated periodic structures where all the necessary parameters can be
using the theory of bounds of bulk effective conductiv- calculated.
ity for a two-component composite. The model allowed Two commonly used correlation schemes are k = aT l .P and
variations in grain consolidation (degree of overlap), k = aT'0c, where k is the permeability, T the NMR relaxation
scaling (grain size), and NMR surface relaxivity. The cor- time, ¢ the porosity, and Y the formation factor. The fitting con-
relation of the permeability (k) with the predictor aTIF` stants (a, b, and c) are often empirically determined from core
was slightly higher than aTbçb` (i.e., a correlation coeffi- measurements. However, few attempts have been made to un-
cient of 0.98 versus 0.95). The exponent b ranged from derstand the physical meaning of these fitting constants, what
1.4 for a pure grain consolidation system to 2 for a pure properties of the rocks affect them, or if there is a fundamen-
scaling system. Variations in surface relaxivity are shown tal relationship between these parameters. The periodic arrays
to cause significant scatter in the correlations. of identical spheres offer a unique opportunity for such an in-
vestigation, because all the petrophysical quantities in the cor-
relations (i.e., permeability, porosity, formation factor, NMR
INTRODUCTION relaxation times, etc.) can be computed with high accuracy.
Thus, these empirical formulas can be studied theoretically
The permeability of a subsurface formation is a very im- with controlled variation of the affecting factors. We varied
portant parameter in the petroleum industry for estimating the size scale, the surface relaxation strength, and the degree
producible hydrocarbon. Continuous direct measurements of of consolidation, and computed their effects on the empirical
permeability in boreholes is currently impractical. Conse- formulas for the permeability.
quently, well log measurements of other petrophysical param- We considered periodic porous systems of simple cubic (sc),
eters (porosity, formation factor, nuclear magnetic resonance body-centered cubic (bcc), and face-centered cubic (fcc) ar-
(NMR) relaxation times, mineral compositions, Stoneley wave rays of touching and overlapping spheres. The computation of
attenuation, etc.) are used to estimate permeability (Timur, porosity for these structures is based on a simple geometric con-
1969; Herron,1987; Kenyon et al., 1988; Sen et al., 1990; Tang sideration. The formation factor is calculated using a method
et al., 1998). These estimations are based largely on empirical proposed by Bergman and Dunn (1992). The NMR relaxation
correlations. time for most clastics is given by Vp /(pS), where VP is the vol-
Permeability has been empirically correlated with porosity, ume of the pore space, S is the surface area of the pore-matrix
formation factor, and surface-to-volume ratio, assuming that interface, and p is the surface relaxation strength. The values
permeability is controlled by pore volume, pore throat size, of permeability are taken from Larson and Higdon (1988).
and surface area (Walsh and Brace, 1984; Johnson et al., 1986; Our paper is organized as follows. The next section reviews
Katz and Thompson, 1986; Blair et al., 1996). Although vari- the calculation of a formation factor. The third section discusses
ous aspects have been studied to help understand the physics various schemes of correlating these theoretically computed

Manuscript received by the Editor September 15, 1997; revised manuscript received September 22, 1998.
*Chevron Petroleum Technology Co., P.O. Box 446, La Habra, California 90633-0446. E-mail: kedu@chevron.com; latg@chevron.com.
$Tel Aviv University, School of Physics and Astronomy, Raymond and Beverly Sackler Faculty of Exact Sciences, Tel Aviv 69978, Israel.
© 1999 Society of Exploration Geophysicists. All rights reserved.
470
 Permeability for Periodic Porous Media 471
petrophysical parameters. Section four uses real rock data for method with only N =1 in the reciprocal space (N is a measure
the analysis. Section five summarizes our present findings. of the size of the reciprocal space over which the Fourier expan-
sion is carried out; see the Appendix). The convergence of the
COMPUTATION OF FORMATION FACTOR Fourier expansion method has been shown to be reasonably
fast for porosities where spheres just touch or slightly overlap
We used the method of Bergman and Dunn (1992) to com-
with each other, but becomes quite slow for lower porosities
pute the formation factor for periodic porous media. We calcu-
where many higher order Fourier expansion terms are needed
lated moments of the pole spectrum of a two-component com-
Downloaded 02/18/13 to 141.117.79.62. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

(Bergman and Dunn, 1995a). Our present calculation used a

posite to produce a continued fraction expansion for the bulk
Fourier expansion with N up to 21 in the reciprocal space.
effective dielectric constant or electrical conductivity. Then, we
used the rigorous bounds for the bulk effective conductivity to PERMEABILITY FOR PERIODIC POROUS MEDIA
determine the formation factor for the composite. A more de-
tailed discussion of the computation method is given in the Theoretical computation of the permeability of periodic
Appendix. porous media has been studied extensively in the past using
The upper and lower bounds of the bulk effective conduc- spherical harmonics (Hasimoto, 1959; Zick and Homsy, 1982;
tivity for the composite usually converge very quickly, as the Sangani and Acrivos, 1982; Larson and Higdon, 1988). Here,
number of moments used in the calculation increases, to nearly we use Larson and Higdon's (1988) results for our analysis.
the same value when the conductivity contrast between the Table 1 shows the results from Larson and Higdon, along with
two components of the composite is small. When the conducti- our calculation of formation factor and NMR relaxation times.
vity contrast becomes large, such as the case for computing The friction coefficient K, given by Larson and Higdon,
the formation factor, the two bounds diverge from each other.
However, the lower bound converges to the true bulk effec- K = f )

tive conductivity for the composite. This is corroborated by 6irpRU^ 1

computing the bulk effective conductivity of the same com- is a dimensionless quantity, where f is the force on a single
posite by switching the values of conductivity of the matrix inclusion, µ is the fluid viscosity, R is the radius of the sphere,
and pore space, as discussed in the Appendix. Recent work and U is the superficial or average velocity through the lattice.
of Helsing (1996) and Cheng and Greengard (1997) have also The permeability k is defined by
dealt with similar problems of computing electrostatic fields of
two-component composites, but with different approaches.
µUuo V0 )
The formation factors we computed for three different pe- kf 6 rRK' 2
riodic structures (simple cubic, body-centered cubic, and face-
where Vo is the total volume per inclusion. The permeability
centered cubic arrays for identical touching and overlapping
defined in this way has dimensions of (length) 2 . In Table 1, we
spheres) are shown in Figure 1. We found results of our
show values of the dimensionless permeability k/d 2 , where d
calculation are quite consistent with those of Schwartz and
is the edge length of the cubic unit cell of the Bravais lattice.
Kimminau (1987), who used the network simulation and mini-
The values quoted from Larson and Higdon include only those
mum area approximation approaches. The formation factors
for touching and overlapping spheres and exclude those with
calculated by Shen et al. (1990) tend to be smaller at low
very small porosities where the corresponding permeabilities
porosity values. This is because they used a Fourier expansion
are less than 0.1 and for d =10 µm.
We calculated NMR relaxation times assuming that surface
relaxation is the dominant mechanism for polarization de-
103
cay and that bulk relaxation can be ignored. The dominant
o FF(sc) NMR relaxation rate in the porous medium has been shown
❑ FF(bcc)
• FF(fcc) (Brownstein, and Tarr, 1979; Mitra and Sen, 1992; Bergman
and Dunn, 1995a) to be equal to pS/VP when pd/D is small
0 (i.e., the fast diffusion regime), and D is the diffusion constant
10 2
U for water. Hence, the dominant NMR relaxation time is cal-
ea
I culated as V,,/pS and shown in Table 1 as T1 . We computed
o ❑

❑
these Tl values (in milliseconds) assuming a surface relaxation
strength p =10 -3 cm/s and evaluating V, and S directly for the
case of d =10 µm. Here, we do not make the distinction be-
I 10' Sao tween the longitudinal relaxation time T l (for which p i should
be used) and the transverse relaxation time T2 (for which p2
mo o
should be used) (Kleinberg et al., 1994). In general, as the Lar-
mor frequency gets lower, T2 is quite close to T1, except that T2
may be diffusionally shortened by induced internal field gradi-
10 0
0.01 0.1 ents or an externally applied field gradient. The strength of the
induced internal gradients depends on the magnetic suscep-
tibility contrast between solid and pore fluid (Bergman and
FIG. 1. Formation factor as functions of porosity for Sc, bcc, and Dunn, 1995b).
fcc periodic structures of identical touching and overlapping Figure 2 shows the permeability as a function of porosity for
spheres. sc, bcc, and fcc sphere arrays, where we have also included large
 472 Dunn at al.

porosity values for which the spheres do not touch. Evidently, dation and scale change. In the functional form aT b 0c, T is the
porosity alone is not a very good parameter for characterizing only parameter that depends on the length scale (T is simply
the permeability because the points arising from the different proportional to that scale). If the permeability changes were
structures do not lie on a single curve, even in regions of high entirely due to the scale change, the exponent for T would be
porosity, where they exhibit the same power law dependence exactly 2. Naturally, the b value for the pure consolidation case
on 0. depends on our model for the periodic structures. The b value
Figure 3 shows the result of using the form aT 1 ç/Y to rep- for the real rock systems may be different. However, this exer-
Downloaded 02/18/13 to 141.117.79.62. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

resent permeability. Here, we have included only porosities cise elucidates how the grain consolidation and the size scaling
for which the spheres are touching or overlapping. The per- affect the value of the exponent b.
meability is computed using an edge length for the Bravais Kenyon et al. (1988) and Banavar and Schwartz (1987) used
lattice of 10 µm. The fitting constants are a =1.024, b =1.404, a =1.0, b = 2.0, and c = 4.0. Figure 5 shows that for these con-
and c = 2.139, with a correlation coefficient r 2 = 0.942. Note stants, the correlation coefficient r 2 becomes 0.913, and the
that the data sets we used are different packings of identi- correlation is not as good as that shown in Figure 4, especially
cal spheres where we achieved the porosity reduction by in- at the higher permeabilities. Comparing Figure 5 with Figure 4,
creasing the amount of overlap between neighboring spheres. the change in the exponent c seems quite significant. In fact,
This grain consolidation process reduces permeability and pro- such a large change is tolerated because the correlation coeffi-
vides the dynamic range for the permeability shown in Figure 3. cient is only decreased very slightly: the reduction in the factor
However, such a dynamic range for permeability can also be OC by increasing c is mostly offset by the modest increase of the
achieved at a fixed porosity by simply varying the overall size exponent b in T b . This shows that T and 0 are not independent
scale. In the latter case, the permeability would simply vary as parameters in characterizing the permeability.
the square of the length scale. So far, we have assumed a constant surface relaxivity
Figure 4 shows the results of such a change of size scale p = 10 µm/s in our calculation. For real rocks, values of p range
where the data sets include those of Figure 3 and the scaled from 10 to 1 µm/s (Kenyon et al., 1988; Kleinberg et al., 1994).
data sets for edge lengths 30, 60, and 100 µm. Using the Figure 6 shows the results of varying p where we use the result
same functional form, aTb¢C, the fitting constants now become of Figure 3 for a unit cell edge length of 10 µm but with surface
a = 0.120, b =1.853, and c =1.833, with a correlation coeffi- relaxation strength ranging from 10 to 1 µm/sec. Since there
cient r 2 = 0.952. Notice that the constant b changes from 1.404 is no scale change, the permeability value remains the same
to 1.853, from pure grain consolidation to both grain consoli- as the surface relaxivity changes. This broadens the range of

Table 1. Theoretically computed permeability from Larson and Higdon (1988) with the Tl and the formation factor F for simple
cubic, body-centered, and face-centered arrays of identical touching and overlapping spheres for different porosities from this work.

¢ Concentration Friction Coefficient Permeability* k/d2Tl F

Simple Cubic
0.4726 0.5236 41.99 2.5269 x 10 -3 151.64 2.907
0.47 0.53 43.60 2.4237 x 10 -3 150.22 2.98
0.45 0.55 48.80 2.1380 x 10 -3 145.80 3.21
0.40 0.60 66.10 1.5288 x 10 -3 134.73 3.88
0.35 0.65 93.36 1.0480 x 10 -3 123.61 4.75
0.30 0.70 139.8 6.7718 x 10 -4 112.34 5.96
0.25 0.75 228.3 4.0080 x 10 -4 100.79 7.76
0.20 0.80 426.9 2.0681 x 10 -4 88.82 10.73
0.15 0.85 1.020 x 103 8.3276 x 10 -4 76.18 16.65
0.10 0.90 4.29 x 10 3 1.8951 x 10 -5 62.57 32.73
0.08 0.92 1.20 x 10 4 6.6396 x 10 -6 56.81 51.09
0.06 0.94 5.726 x 10 4 1.3599 x 10 -6 51.02 109
Body-Centered Cubic
0.31982 0.68018 162.3 3.7675 x 10 -4 67.87 4.60
0.30 0.70 191.0 3.1761 x 10 -4 64.96 5.07
0.25 0.75 299.2 1.9772 x 10 -4 57.34 6.52
0.20 0.80 520.5 1.1072 x 10 -4 49.22 8.67
0.15 0.85 1.073 x 10 3 5.2253 x 10 -5 40.34 12.29
0.10 0.90 3.05 x 10 3 1.7846 x 10 -5 30.23 19.80
0.08 0.92 5.47 x 10 3 9.8253 x 10 -6 25.66 25.62
0.06 0.94 1.14 x 104 4.6519 x 10 -6 20.67 37.22
0.04 0.96 3.2 x 10 4 1.6325 x 10 -6 16.91 86.42
Face-Centered Cubic
0.25952 0.74048 431.0 8.7038 x 10 -5 41.30 6.25
0.25 0.75 480.0 7.7816 x 10 -5 40.49 6.67
0.20 0.80 426.9 3.9916 x 10 -5 36.08 9.67
0.15 0.85 913.8 1.6077 x 10 -5 31.37 15.28
0.10 0.90 2.21 x 10 3 3.6879 x 10 -6 26.27 31.33
0.08 0.92 9.34 x 10 3 1.3694 x 10 -6 24.17 52.49
*The permeability is in dimensionless units. In order to get its value in and (millidarcys), the dimensionless value should be multiplied
by d 2 , where d is the unit cell size in µm (1 and = 0.987 x 10-3µm2).
 Permeability for Periodic Porous Media 473

scatter even in the high permeability case, where our model exponents b and c in Figures 7 and 8 are quite consistent with
showed very good data collapse when p is constant and small. those calculated in Figures 3 and 4 (where aT b 4 is used) if we
Hence, we propose that in clastic sediments the breadth of scat- apply the relation F 0 -15 (as observed in Figure 1 for large
ter of the permeability data in the high-permeability regime porosities). However, the use of formation factor F instead
is caused mainly by the wide range of values for the surface of porosity 0 significantly increases the correlation with per-
relaxivity. meability data. This is quite evident when comparing Figures 4
Figure 7 shows the results when we take the data set of and 8 in the low permeability region, where data scatter is much
Downloaded 02/18/13 to 141.117.79.62. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Figure 3 (i.e., the pure grain consolidation system) and plot less when aT b .F, is used.
permeability versus aT b Fc instead of versus aT b q , . Figure 8 The concept of using the formation factor to correlate per-
shows the results when we also include the size scaling effect, meability is not new. Walsh and Brace's (1984) simple model of
with the edge lengths of 10, 30, 60, and 100 µm. The calculated considering flow in a rock as a flow through a bundle of tubes

10 3 10 5

10 2 • perm(sc) •o
❑ perm(bcc)
1
•t• °• 10 °
• 10 um
❑ 30um
• perm(fcc) •❑ • 60 um QA
10' • ° 100 urn 4
G
❑•
6 103
1o° E
° A

10 1 °S A
10 2
❑
CO
••
r, 10 2
a • a 10'
•, •° • a=0.120
10 3
b = 1.853
10 ° • •o ❑ c=1.833
10' A
• •, 2 =0.952
•

10 5 I IS .I .I
10'
10310-2 10, 100 10'' 10° 101 102 103 104 105
aT"O'

FIG, 2. Permeability as functions of porosity for periodic FIG. 4. Permeability versus aT b ¢` for periodic structures of sc,
structures of sc, bcc, and fcc where we have included very bcc, and fcc using the data from Figure 3 (i.e., 10 µm) and the
high porosities where the spheres are not touching. The edge scaled data sets for 30, 60, and 100µm. The length scale refers to
length of the Bravais lattice is 10 µm. Permeability is in the edge length of the Bravais lattice used in the computation.
millidarcys, where 1 and = 0.987 x 10 -11 cm2 . The fitting constants are a = 0.120, b = 1.853, and c = 1.833,
with a correlation coefficient rz = 0.95.
10 3
10 °
• perm(sc)
❑ perm(bec) •0 • 10 um °A&
• perm(fcc) • 10° ❑ 30 um
102 •
• 60um °•
° 100um •
E ❑Q • o ^
4, a E 1 o'
10' • AA
m a 10z ° •
E • to
a
d (0
E °• 5 ❑ ?
° •• 0=1024 ° • ❑ .R• •
10 ° ❑ b=1.404 a 10'
• ❑ •ice 0=l
• c=2.139 ° ❑• b=2
• r2=0.942 ❑ • mti c=4
10 ° •
• t=0.913
10'
10 1100 10' 102 103
10 3101 10' 103 105
aT"O
T 2 Ø4
FIG. 3. Permeability versus aT b o , for periodic structures of
Sc, bcc, and fcc using only porosities where the spheres are FIG. 5. Permeability versus aT b o` for periodic structures of sc,
touching or overlapping. The edge length of the Bravais lattice bcc, and fcc using the data from Figure 4 but with fixed constants
is 10 µm. The fitting constants are a = 1.024, b = 1.404, and a = 1.0, b = 2.0, and c = 4.0. The correlation coefficient now
c = 2.139, with a correlation coefficient r 2 = 0.94. becomes r 2 = 0.91.
 474 Dunn et al.

yielded the permeability We found that using the formation factor in computing the
above described correlations does more than merely replac-
k = ( 1 / b)O 3 (V / AS) 2 ( 1 / f), (3) ing 0"` by 1/F. From Figure 3, different values of m must be
where b is a shape factor, A s / V is the surface area per unit used for different structures when 0 decreases below 0.1. This
volume, and F is the formation factor. Katz and Thompson divergence of values of F corresponds to a similar divergence
(1986), on the basis of a percolation argument, and Johnson of values of the permeability at low permeabilities. Using .F as
et al. (1986), on the basis of the analogy between electrical and the characteristic parameter improves collapse of the data in
Downloaded 02/18/13 to 141.117.79.62. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

hydraulic conductions in porous media, proposed the following the low porosity—low permeability regime, as shown in Figures
relationship:
4 and 8. Although our model demonstrated that in principle
aTI.F is a better choice than aTb4,`, in reality, other heterge-
k — L 2 /,F = L2cbm, (4) neous factors obscure the slight superiority in its correlation
with the permeability, as discussed in the following section.
where L is either the critical pore diameter in the mercury
porosimetry measurements or the characteristic length in the DISCUSSION OF REAL ROCK DATA
electrical conduction for the porous medium. Blair et al. (1996)
summarize the discussion on this subject in great detail. The real rock data present many heterogeneities and uncon-
trollable factors. The model systems we considered used iden-
10 3 tical spheres. To appy the results to real rock data, we need
to determine significant and insignificant differences between
• 0.001 cm/s real rocks and our models for the permeability correlations.
❑ 0.0006 cm/s
A 0.0003 cm/s •IA We can think of the following:
10 2 A 0.0001 cm/s O°
1) Real rock systems are not identical spheres. They come in
c
E many different shapes and sizes. The packing is random,
mm disordered, and certainly not periodic. This will signifi-
a 10' m O° ^ cantly affect the formation factor, and hence the Archie's
m m exponent m. However, we expect that the general rela-
E
tion such as k L 2 /.F = L 2 cbm holds for both the real
4W m rock systems and the periodic arrays of spheres, only the
m a=661.8
10° •
mb=0.169 formation factor will be different. Thus, we expect our
m c=3.098 conclusion on the scaling effect on the fitting constant,
= 0.920 exponent b, should hold for real rocks excluding other
influencing factors.
10'
102 101 100 101 102 103 2) The real rock systems certainly have different values of
surface relaxation strength p. Within the zone probed by
aT b 0c the NMR logging tool, variations in p can often be ex-
FIG. 6. Permeability versus aT b o` for periodic structures of sc, pected to be minimal. However, at larger scales, p can
bcc, and fcc with the surface relaxation strength p ranging from not be assumed to be constant. Large surface relaxiv-
0.001 to 0.0001 cm/s. ity caused by transition metal ions has been known to

10 3 10 5

• pe-(sc) • 10 um A
° pem'(bcc) f ❑ 30um
• pemi(fcc) •
10`
• 60 um P^
10 2 • ° 10 um
•

E E 103
4
10'
°Wj
m n 102
m Cl d
E A E A
a 0. 10' ° a=0.111
DA A ^ At b=7.883
a=0.934 o
10° ° b = 1.438 c=-I.310
❑ • c=-1.509
10 0
•
° •• ?=0.983
A •
/=0.982

10' 10' I • •♦ I I 1
102 101 100 10 102 103 10-210 110 010 110 2 10 3 10 4 10 5

aT I F` aT b F`
FiG. 7. Permeability versus aT'F'c for periodic structures of Sc, FIG. 8. Permeability versus aT b .T'` for periodic structures of Sc,
bcc, and fcc using the data from Figure 3. bcc, and fcc where using the data from Figure 4.
 Permeability for Periodic Porous Media 475

cause fast NMR relaxation and have very short relax- The large scatter of data points in these figures is, in our
ation times. Our conclusion on the variation of surface opinion, due to a large variation in surface relaxation strength
relaxavity causing the scattering in permeability cross- among different samples.
plots is independent of our specific model. Hence, we As indicated by the correlation coefficients r 2 in Figures
believe that in real rock systems the scattering of data in 9 and 10, the correlation scheme of aT 6 .F is slightly better
the crossplots of core permeability with the permeability than aT b 4', similar to our finding from model calculations in
estimators is in large part due to the variability of p. Figures 4 and 8.
Downloaded 02/18/13 to 141.117.79.62. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

3) In our model, we have assumed a single dominant re-

laxation time given by Vp /pS. In real rock systems, we SUMMARY
expect more than one dominant relaxation time. Usually, We have studied the correlation between theoretically com-
a geometric average of these relaxation times is used for puted permeability, porosity, formation factor, and NMR relax-
the permeability correlation. However, depending on the ation times for periodic porous media of identical touching and
diffusion coupling between large and small pores, two dif- overlapping spheres. Although the systems that we consider
ferent length scales can show up either as two dominant are extremely idealized, this approach has the advantage of
relaxation times or as merged into one relaxation time isolating each physical parameter, such as grain consolidation
(Ramakrishnan et al., 1998). This effect is in fact con-
trolled by the factor pd/D and, therefore, related to the
variability of p. When the p becomes too small and the
relaxation time T =VP /pS becomes large enough that 10 °
the bulk relaxation of water is not negligible, the NMR
relaxation time is no longer inversely proportional to the 10 3 4+ ^ I'
'•d
surface-to-volume ratio. All these uncertainties disrupt •Z !v -bt •
the correlation of permeability with the predictors a T" 102
and aT t F` C
E
T 101
Figures 9 and 10 show the results for the two correlation
schemes (aT b / and aT b .Fc) for permeability of sandstones of E 10 °
260 core samples from different parts of the world. The NMR C)
relaxation times are T1G, the geometric averages of T1 over the •f • a= 1.501
10' • b=2.053
distribution of relaxation times of each sample between 1 ms • c=-2.395
and 10 s measured at 10 MHz (LaTorraca et al., 1993). The r 2 = 0.795
10 2
fact that the exponent b is close to 2 shows that the data are
mainly from rocks of different grain sizes. The large dynamic
range of permeability is caused by size scaling rather than grain 10a
102 101
10' 100 102 10' 104 10 5
consolidation. This is indeed the case as shown in Figure 11
where, out of 260 data points, only 10 samples have porosity aT b F`
values below 0.1. Most of the samples have porosity values FIG. 10. Permeability versus aT b F` for sandstones of real rock
between 0.20 and 0.25. systems using the same set of data as in Figure 9.

10 °
10 °
10, s»• •
• 103 M •
102 N ti •
10 2
E G
10' E
a
• ••f•^
• •• 10'
E 10 ° i
••• £ 10 °
•
S

a • % l•
•*•• o=0.214
10' b=2.163 a
• • c=3.424 10^ I.
•
10 2 • • r2=0.778
102 •
103
101 102 10 10010' 10 210 310410 5 10-3
10210' 10°
aT b o`

FIG. 9. Permeability versus aT b çt` for sandstones of real rock FIG. 11. Permeability versus 0 for sandstones of real rock sys-
systems. tems using the same set of data as in Figure 9.
 476 Dunn et al.

(porosity reduction), size scaling (size of the grain), and surface of electrostatic fields in dense random dispersions of cylinders: J.
relaxation strength, and allowing us to study its implication and Comput. Phys. 136, 629-639.
Hasimoto, H., 1959, On the periodic fundamental solutions of the
effects individually in the various correlation schemes. We have Stokes equations and their application to viscous flow past a cubic
arrived at the following conclusions: array of spheres: J. Fluid Mech., 5, 317-328.
Helsing, J., 1996, Thin bridges in isotropic electroctatics: J. Comput.
1) The form aT ) Fc generally provides slightly better corre- Phys. 127, 142-151.
Downloaded 02/18/13 to 141.117.79.62. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Herron, M., 1987, Estimating the intrinsic permeability of clastic sedi-

lation to permeability than aT b c` ments from geochemical data: 28th Ann. Logging Symp., Soc. Prof.
2) The exponent b for the relaxation time T ranges from 1.4 Well Log Analysts, paper HH.
for pure grain consolidation systems to 2 for pure scaling Johnson, D. L., Koplik, J., and Schwartz, L. M., 1986, New pore-size pa-
rameter characterizing transport in porous media: Phys. Rev. Lett.,
change systems. Reservoir rocks often have a value near 57,2564-2567.
2, indicating that the changes are due mainly to grain size Katz, A. J., and Thompson, A. H., 1986, Quantitative prediction of
change. permeability in porous rock: Phys. Rev., B 34, 8179-8181.
Kenyon, W. E., Day, P. I, Straley, C., and Willemson, J. F, 1988, A
3) The scattering of data in the correlation of the permeabil- three-part study of NMR longitudinal relaxation properties of water-
ity with the predictors aT b 0' and aT b .FC appears to be due saturated sandstones: SPE Formation Evaluation, 3, 622-636.
mainly to the variation in surface relaxation strength. Kleinberg, R. L., Kenyon, W. E., and Mitra, P. P., 1994, Mechanism of
NMR relaxation of fluids in rock: J. Mag. Reson., A 108, 206-214.
4) The form aTbcbc is relatively tolerant, especially for some- Korringa, J., 1984, The influence of pore geometry on the dielectric
what large variation of c. This happens because a large dispersion of clean sandstones: Geophysics, 49, 1760-1762.
Korringa, J., and LaTorraca, G. A., 1986, Application of the Bergman-
increase in c can be easily offset by a modest increase in b. Milton theory of bounds to the permittivity of rocks: J. Appl. Phys.,
60, 2966-2976.
Larson, R. E., and Higdon, J. J. L., 1988, Periodic grain consolidation
ACKNOWLEDGMENTS model of porous media: Phys. Fluids, A 1, 38-46.
LaTorraca, G. A., Dunn, K. J., and Brown, R. J. S., 1993, Predicting per-
The authors acknowledge a helpful discussion with R. L. meability from nuclear magnetic resonance and electrical properties
measurements: 1993 Soc. Core Analysts Conf., paper 9312.
Kleinberg on the range of values for the surface relaxation Mitra, P. P., and Sen, E N.,1992, Effects of microgeometry and surface
strength p for sandstones. Timely and constructive reviews by relaxation on NMR pulsed-field-gradient experiments: simple pore
K. D. Mahrer and other Geophysics reviewers are greatly ap- geometry: Phys. Rev., B 45,143-156.
Ramakrishnan, T. S., Schwartz, L. W., Fordham, E. J., Kenyon, W. E.,
preciated. and Wilkinson, D. J., 1998, Forward models for nuclear magnetic
resonance in carbonate rocks: 39th Ann. Logging Symp., Soc. Prof.
Well Log Analysts, paper SS.
REFERENCES Sangani, A. S., and Acrivos, A., 1982, Slow flow through a periodic
array of spheres: Internat. J. Multiphase Flow, 8, 343-360.
Banavar, J. R., and Schwartz, L. M., 1987, Magnetic resonance as a Schwartz, L. W., and Kimminau, S., 1987, Analysis of electrical conduc-
probe of permeability in porous media: Phys. Rev. Lett., 58, 1411- tion in the grain consolidation model: Geophysics, 52, 1402-1411.
1414. Sen, E N., Straley, C., Kenyon, W. E., and Whittingham, M. S., 1990,
Bergman, D. J., and Dunn, K. J., 1992, Bulk effective dielectric con- Surface-to-volume ratio, charge density, nuclear magnetic relax-
stant of a composite with a periodic microgeometry: Phys. Rev., B ation, and permeability in clay-bearing sandstones: Geophysics, 55,
45, 13262-13271. 61-69.
1995a, Self-diffusion in a periodic porous medium with inter- Shen, L. C., Liu, C., Korringa, J., and Dunn, K. J., 1990, Computation
face absorption: Phys. Rev., E 51, 3401-3416. of conductivity and dielectric constant of periodic porous media: J.
1995b, NMR of diffusing atoms in a periodic porous medium Appl. Phys., 67, 7071-7081.
in the presence of a nonuniform magnetic field: Phys. Rev., E 52, Tang, X. M., Atunbay, M., and Shorey, D., 1998, Joint interpretation of
6516-6535. formation permeability from wireline acoustic, NMR, and image log
Blair, S. C., Berge, P. A., and Berryman, J. M., 1996, Using two-point data: 39th Ann. Logging Symp., Soc. Prof. Well Log Analysts, paper
correlation functions to characterize microgeometry and estimate KK.
permeabilities of sandstones and porous glass: J. Geophys. Res., 101, Timur, A., 1969, Pulsed nuclear magnetic resonance studies of poros-
20359-20375. ity, movable fluid, and permeability of sandstones: J. Petr. Tech., 21,
Brown, R. J. S., 1972, Nuclear magnetism well logging method and 775-786.
apparatus: U. S. Patent 3 775 671. Walsh, J. B., and Brace, W. F, 1984, The effect of pressure on porosity
Brownstein, K. R., and Tarr, C. E., 1979, Importance of classical dif- and the transport properties of rocks: J. Geophys. Res., 89, 9425-
fusion in NMR studies of water in biological cells: Phys. Rev., A 19, 9431.
2446-2453. Zick, A. A., and Homsy, G. M., 1982, Stokes flow through periodic
Cheng, H., and Greengard, L., 1997, On the numerical evaluation arrays of spheres: J. Fluid Mech., 115, 13-26.

APPENDIX
MATHEMATICAL PROCEDURE FOR COMPUTING FORMATION FACTOR

We briefly describe below the mathematical procedure for conductivity a,, we first use an integral equation for the electric
computing the formation factor of a fluid-saturated porous potential 0(r):
medium. For a more detailed description, please refer to
Bergman and Dunn (1992) and related references. ^(r) = z + 1 Po, (A - 1)
s
Consider a sample of periodic porous structure made of
where
solid spheres where the pore space is filled with water. The
-
sample fills up the entire volume in between the plates of an S =o w , (A-2)
infinite parallel-plate capacitor of thickness L with plates per- Qw - Gm

pendicular to the z-axis. In order to calculate the bulk effective

dielectric constant of the composite E or the bulk effective
Q
r0 - f dV'9m (r')V'G(r, r') . V'O(r'), (A-3)
 Permeability for Periodic Porous Media 477

and where a. and am are the conductivities of water and the This permits us to include a very large number of Fourier com-
solid matrix, respectively, G(r, r') is a Green's function for the ponents in the calculation, leading to very large matrices t.
Laplace equation of the electric potential, 0,n (r') is the charac- This is necessary in order to ensure good convergence of the
teristic function which defines the domain of solid matrix, and calculation and accurate results for the formation factor.
P is a linear operator which depends on the microgeometry Using these computed moments as input information, we can
of the porous medium through 0,,(r'). We have assumed the construct a continued fraction expansion for F(s) and obtain a
average electric field is equal to 1. Hence, the first term for the sequence of successively tighter, converging bounds for F(s).
Downloaded 02/18/13 to 141.117.79.62. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

electric potential is simply equal to z, the coordinate. The t With the formation factor .F defined as
operator is made self-adjoint by defining the scalar product of /
two (scalar) functions as follows: alim 0 tee) 1 - F 1 (A-11)
\ w ( )'

(øI) - J dV6m V^i* • V , (A-4) similar converging bounds are calculated fora', and the results
are shown in Table A-1.
where V is the total volume under consideration. The bulk
The calculation involved computation of 20 moments of
effective conductivity of the composite a, is calculated from
F(s), which was carried out using reciprocal lattice vectors
g= (2n/d)(n x , n^, n^), where d is the lattice constant and
F(s)=l- 6e = (Z I ^) =( Z1. (A-5) n X , n y , n, are integers ranging from -N to +N. We used N = 19
6wS S - 1:
and 21 (two values of N are always used, and the results are then
Because r has a complete set of eigenfunctions 0,, with real extrapolated to N = oo assuming a linear dependence on 1/N).
eigenvalues s,,, Upper and lower bounds (bound 2 and bound 1 in Table A-1)
were calculated using r lowest moments of F(s). Thus the
r0n = S,4n, (A-6) bounds obtained for r =0 used only the zero moment E F n
Equation (A-5) can be transformed into as input information. As more and more moments are used in
the calculation (i.e., as r increases), the bounds get progres-
sively tighter. When r exceeded 11, we usually encountered a
F (s) _ F,, = l (Z I ^n) I 2 , (A-7)
S Fn
- Sn , situation where small errors in the lower moments caused the
where the following inequalities are also satisfied: continued fraction expansion to terminate (this is discussed
in detail in Bergman and Dunn, 1992). Even before that hap-
0<s<1, F(1)<1, Fn < 1. (A - 8) pened, the lower bound (bound 1) already seemed to be well
converged with increasing r. A separate calculation carried out
Expanding F(s) in powers of 1/s from equation (A-7), we get using N = 25 and 27 did not result in significantly better results.
n Ys,,F.E-1 Sn F,n Therefore, in order to reduce computation time, all subsequent
F(s) = EF
s + s2+ ... + n + .... (A-9) ,

calculations involved only 11 moments using N =19 and 21.

Notice that the gap between the upper and lower bounds on
When comparing this to a similar expansion of equation (A-5), the formation factor in Table A-1, obtained using 11 moments,
we find that are within 1% of each other when s =1.01, which corresponds
^snFn = (zlt"iZ). (A - 10) to a conductivity ratio Q./am 0.01. This means that, at this
porosity and this conductivity ratio, the formation factor can
For a periodic porous medium, we can expand the eigenfunc- be accurately determined to within 1%. However, as s gets
tions 0„ in Fourier series, and the operator r is then represented closer to 1, the convergence of bound 2 (the upper bound) be-
by an infinite matrix. The right-hand side of equation (A-10) comes poorer. Already from this behavior, we may conclude
can then be evaluated by a sequence of matrix-vector prod- that bound 1 (the lower bound) is closer to the exact result.
ucts without ever having to evaluate a matrix-matrix product. The poor convergence of the upper bounds on F(s) and .F can

Table A-1. Results for the bounds for the formation factor for a simple cubic array of identical, slightly overlapping spheres with
a porosity c5 = 0.47.
s = 1.0001 S = 1.001 s = 1.01
r Bound 1 Bound 2 Bound 1 Bound 2 Bound 1 Bound 2
0 2.13 5301.00 2.13 531.00 2.10 54.00
1 2.69 2733.36 2.69 274.59 2.64 28.71
2 2.85 1026.63 2.85 104.55 2.79 12.32
3 2.92 410.23 2.92 43.29 2.85 6.56
4 2.95 173.20 2.94 19.78 2.88 4.39
5 2.97 73.20 2.96 9.89 2.89 3.50
6 2.97 42.49 2.97 6.86 2.90 3.23
7 2.98 17.70 2.97 4.41 2.90 3.02
8 2.98 12.34 2.97 3.89 2.90 2.97
9 2.98 7.89 2.97 3.45 2.90 2.94
10 2.98 6.60 2.97 3.32 2.90 2.93
11 2.98 5.00 2.97 3.17 2.90 2.92
 478 Dunn at al.

be understood as follows. When s is close to 1, it means that which we switch the roles of solid matrix and pore space and
the conductivity ratio Q,„/v„ is very small. In that case, if the calculate the function
pore space is connected, then the bulk effective conductivity
H(t) = 1-
a is finite and of the same order of magnitude as ow , even
^e = A + H" , (A-12)
Umt tn/=Ot -tn
when o. = 0. Consequently, F(1) will be strictly less than 1.
However, by a small change in the microstructure, it is pos- where
t sible to disconnect
= the pore space in neighboring unit cells, in a(A-13)
Downloaded 02/18/13 to 141.117.79.62. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

which case we would get ue = 0 and F(1) =1. This could happen 6m - 6w
even though the changes in the series expansion coefficients of and
equation (A-9) are very small. That is why the upper bound
on F(1) will always be 1, and the upper bound on F(1.0001) -wlimtH(t). (A-14)
converges to the correct value very slowly with increasing r. L=0
Because we typically have Q„,/aw N 10 -6 , we will always try The residue of the pole at t = 0 is just the reciprocal of the for-
to approximate F by using the limiting behavior of its lower mation factor (Korringa, 1984; Korringa and LaTorraca, 1986).
bound for r oo and s ->. 1. Thus, for 4=0.47, we deduce This is a property H(t) must have because the pore space is
from Table A-1 that F = 2.98. connected. Instead of computing tH(t) for t = -0.0001, we cal-
When the porosity gets smaller, the convergence becomes culate the entire pole spectrum for H(t) (not shown here). We
progressively poorer and extremely sensitive to the conductiv- used 11 moments and N = 27 and 29 for the calculation. The
ity ratio. Table A-2 shows the result of the formation factor resulting approximant to H(t) has five poles between 0 and 1,
calculation for a simple cubic array of identical overlapping in addition to the percolation (or connectivity) related pole at
spheres with a porosity of ¢ = 0.06. As before, we again choose t = 0. The formation factor estimated from the pole amplitude
the lower bound as being closer to the correct value for the at t =0 (i.e., 107) is very close to the lower bound shown in
system. This is also corroborated by the following exercise in Table A-2.

Table A-2. Results for the formation factor for a simple cubic array of identical overlapping spheres with a porosity ¢ = 0.06
calculated at finite reciprocal lattice sizes (N = 19 and 21).
S = 1.0001 s=1.001 S = 1.01
r Bound 1 Bound 2 Bound 1 Bound 2 Bound 1 Bound 2
0 16.64 9401.00 16.41 941.00 14.43 95.00
1 24.44 8394.13 23.93 840.55 19.80 85.19
2 48.44 8112.86 46.30 812.58 32.22 82.54
3 57.71 7468.02 54.63 748.78 35.80 76.79
4 77.87 7206.04 72.17 723.02 41.93 74.61
5 87.83 6730.79 80.54 676.65 44.31 70.94
6 94.74 6148.49 86.22 620.56 45.74 66.99
7 100.04 5529.18 90.51 561.88 46.73 63.40
8 101.80 4244.28 91.91 443.54 47.03 57.57
9 109.06 4021.10 97.62 423.45 48.12 56.74
10 111.74 3523.57 99.68 379.30 48.47 55.09