PHYSICAL REVIE%' B VOLUME 34, NUMBER 11 1 DECEMBER 1986

Quantitative prediction of permeability in porous rock

A. 3. Katz and A. H. Thompson

Exxon Production Research Company, P. O. Box 2189, Houston, Texas 77252-2189
(Received 3 March 1986; revised manuscript received 28 April 1986)
%e show that percolation concepts lead to the definition of a characteristic length for the per-
meability in random porous media. Application of the model to sandstone and carbonate rocks
yields quantitative agreement between theory and experiment ~ith no adjustable parameters.

The transport properties of fluids in porous media are of simulations — and support the conjecture that mercury in-
broad interest in science and technology. They belong to jection has a percolation geometry. s
the general class of transport in random media including It is well established9 that pore spaces of many reservoir
electronic transport in amorphous conductors. Transport rocks are characterized by a broad distribution of pore
in porous media is central to technological problems in sizes. A broad distribution of pore sizes suggests that
porous electrochemical electrodes, filters, and gels and to transport through the pore space must be understood in
establishing the producibility of petroleum reservoirs. terms of a broad distribution of local conductances. For
There have been numerous attempts over the last sixty each point r in the pore space, we assign a value X(r),
years to establish a relationship between the permeability where X(r) represents the diameter of the largest sphere
of a random porous material and other characteristic ma- that contains the point r and is fully contained in the pore
terial properties. ' Early work' expressed the permeability, space. We next group the points into continuous patches
which has units of length squared, as a function of hy- of pore space with common value X. We assume a patch
draulic radius (the pore volume divided by the pore sur- where assigned value A, is also roughly of linear dimension
face area) and such macroscopic properties as the total X [we can relax this constraint without invalidating our ar-
porosity. The hydraulic radius has been used as the gument, but it is important that no long-range correlations
characteristic length up to the present time. 23 Recent in the function X(r) persist].
work attempts to justify the Carmen-Kozeny equation on AHL suggest that transport in a random system with a
the basis of network or effective-medium models. 3 These broad distribution of conductances is dominated by those
attempts fail to give satisfactory results for two reasons. conductances with magnitudes greater than some charac-
First, they attempt to relate the transport properties to the teristic value g, . The characteristic conductance g,
macroscopic geometric properties. Second, they do not represents the largest conductance, such that the set of
recognize the fundamental physical significance and sensi- conductances Ig g & g, ] still forms an infinite, connected
~

tivity of the length scale in the permeability prediction. cluster. Hence, transport in such a system reduces to a per-
We propose and present experimental evidence support- colation problem with threshold value g, . Kirkpatrick'
ing the following relationship for the permeability of rocks and Shante" have carried these ideas further. They assign
saturated with a single liquid phase: all local conductances with values g ~g, the value g„and
set all conductances with values g & g, to zero; they arrive
k -cl2(cr/ere) . at a trial solution for the sample conductance of the form

The permeability k is defined by the Darcy relation:

g -kg, [p(g, ) —p, ]',
v (k/rl)Vp, where v is the velocity of fluid flow through which can be maximized with respect to the variable g, .
the rock, rl is the fluid viscosity, and Vp is the pressure Here, p(g, ) denotes the probability that a given conduc-
gradient imposed across the rock sample. Here l, is some tance is greater than or equal to g„and k is an appropriate
characteristic length of the pore space, o is the conductivi- constant. In three dimensions, the percolation exponent
ty of the rock saturated with a brine solution of conductivi- t -19"
ty rrn, and c is a constant on the order of +, , The conduc- %e interpret transport within rock pore spaces in terms
tivity term reflects the connectedness of the pore space, of these percolation ideas. Since the local hydraulic con-
whereas the length term sets the scale for the permeability. ductance is a function of the length l, the threshold con-
We show that Eq. (1) follows from the percolation argu- ductance g, defines a characteristic length I„which is the
ments of Ambegaokar, Halperin, and Langer (AHL), length that appears in Eq. (1). Moreover, the length that
which lead to a definition of the characteristic length 1,. marks the percolation threshold in the hydraulic conduc-
The length l, is a unique transport length scale and dom- tance problem also defines the threshold in the electrical
inates the magnitude of the permeability. New contribu- conductance case. We can, therefore, study Eq. (2) as a
tions of this work include recognizing the importance of function of the length parameter I:
this length to the physics of permeability and directly
measuring I, from mercury injection experiments. Our
g(l) -yg, (l)[p(l) —p, l' . (2')
results verify the quantitative accuracy of the AHL The porosity p appears here to ensure a proper normaliza-
approximation — in agreement with recent computer tion of the fluid or the electric charge density. As we de-

34 8179 1986 The American Physical Society

 A. J. KATZ AND A. H. THOMPSON 34

crease I past the threshold value I„ the function g, (l) I ORE OXAVE, ER (pm)

(which we take equal to cl for the fluid flow problem and 1.24 0. 62 Q. 41
equal to c'I for the analogous electrical problem) also de-
creases, but the power-law expression [p(l ) — p, l' will in- 60
crease as more and more of the pore space is included in
the largest cluster that spans the sample. For appropriate 5Q
choices of the function p(l), the conductance g(l) as-
sumes a maximum value for some l,
„~l,. Since Eq. (2')
should always yield at least a lower bound to the true rock
conductance, we take l~,„ to be the "best" choice to use in
the trial solution [Eq. (2')]. We observe that, in general,
I~,„ for the hydraulic conductance will be different from
I~,„ for the electrical conductance problem, since the con-
duction pathways have different weights for the two cases.
We can derive some general results concerning the rela-
0
tionship between permeability and conductivity without 0 LG 1000 2000 3000 4000
specifying the function p (I ). These relationships hold as PORE PRESSURE (kPa)
long as p(l) allows for a maximum in the conductance
FIG. 1. Typical mercury injection curve taken on a quartz ce-
and the maximum occurs for I~,„&I, . If these conditions
mented sandstone from eastern Utah. The characteristic length
are satisfied, we find to first order in Al, and ltlr, is labeled l, . The pore diameters are calculated from the pore
pressures using the Washburn equation.

tension and contact angle yield errors that are small com-
For very broad pore size distributions [such that pared to the expected uncertainty in measuring the per-
I, tp"(I, )/p'(I, ) «1] Eqs. (3) reduce to meability. ) The Washburn equation assumes a local
cylindrical geometry.
Figure 1 shows a typical mercury injection curve where
the volume of mercury intruded (normalized by the total
pore volume) is plotted versus the applied pressure.
Using the above results for broad distributions of conduc- Several authors have interpreted the rapid rise in the
tances, we establish a relationship between the electrical curve to occur when the intruded mercury initially forms a
conductivity connected cluster that spans the sample. The intitial por-
tion of the intrusion curve with positive curvature is associ-
a-an't [p(I',„)—p, ]' ated with surface defects and a broad variation in sample
and the permeability thickness found in the rock chips used in the experiment.

k -pqO". .„)'[pO".„)-p, ]' .

We take the inflection point's of the rapidly rising portion
of the curve to mark the threshold pressure p, for the for-
mation of the "infinite" cluster. From the Washburn
To first order in ill, or in Alr„

To interpret the constants a and p, we assume that locally

the rock conductivity is simply pro, the conductivity of the
brine solution, and that the local pore geometry is cylindri-
cal. These assumptions imply that a cro and p
Combining the above relationships, ~e obtain O
k +~ I, cr/pro, which is of the same form as Eq. (1). o
E
To determine the characteristic length I, of actual rock o
aD
1. 0
samples, me performed mercury injection' experiments on LJ
CP

a set of 50 rocks that range in permeability from 50 micro- Ch

0
darcies to 5 darcies. Mercury, which is assumed to be an O

ideal nonwetting fluid, is forced into the evacuated pore

space under quasistatic conditions. For each externally
applied pressure, the diameter of the mercury-pore space
inter- face is determined by the Washburn equation 2. 0
—4ycos&/d, where p, is the capillary pressure, which p,is log [: k (rndarcy})

the difference in pressures on the two sides of the men- FIG. 2. Calculated permeability k~, vs measured permeabili-
iscus, y is the surface tension ( 485 dyn/cm), 8 is the ty k for various sandstones and carbonates. The dashed lines
contact angle (130'), and d is the local diameter of the indicate a factor of 2 deviation. Note that the unit of permeabil-
pore space. (The experimental uncertainties in the surface ity is the millidarcy (md) 10 "cm .
 QUANTITATIVE PREDICTION OF PERMEABILITY IN POROUS ROCK 8181

equation, we conclude that the pore widths l included in permeabilities is within experimental error.
the infinite cluster satisfy the relation —4ycose/p, ;
i~ The length scale I, determined from the inflection point
moreover, it is not possible to form an infinite cluster such on a capillary pressure curve, is accurate to + 15% of its
that the included pore widths are strictly greater than value. The inflection point is a unique, well-defined exper-
—4cos8/p, . This expression then defines the characteris- imental point and corresponds to the formation of electri-
tic length 1,. cal continuity across the sample. In contrast, the choice of
Figure 2 presents the gas permeabilities extrapolated to a length in the neighborhood of I, would alter the value of
infinite pressure' of sandstone samples and a few porous the constant 226 and would not be directly relevant to the
carbonate rocks versus the permeabilities calculated from critical path for percolation.
Eq. (2). The gas permeabilities extrapolated to infinite In conclusion, we cite several phenomenological expres-
hydrostatic pressure are customarily assumed to represent sions in the geophysics literature relating permeability and
the geometrically determined absolute permeabilities. The parameters derived from mercury intrusion data. A rela-
conductivities were measured at 20 kHz (these measure- tionship established by Swanson' comes close quantita-
ments agree well with dc values) on rock samples saturat- tively to our Eq. (1). Swanson's derivation, however, is
ed with 10000 and 100000 ppm NaC1 brine. The per- very different from ours. The excellent predictability he
meabilities, conductivities, and I, s were all measured on obtains reflects his choice of a characteristic length close
the same piece of rock sample. The experimental errors
are dominated by the measurements of /, and k. The cu-
to I, or I, „. There are also numerous expressions for k
that use the porosity raised to some large power. 's'9
mulative errors are estimated to be plus or minus a factor Large exponents are required to give the observed range of
of 2. The absolute values of the permeabilities of the least k values in the absence of an I, measurement. In our ex-
permeable samples are systematically high because of the periment l, varies by three orders of magnitude while o/cro
difficulty in perfectly sealing rock samples within the mea- varies by a factor of 20. Expressions involving large ex-
surement rig. The line drawn in Fig. 2 is the prediction of ponents on porosity are reasonable only when p tracks l, .
Eq. (1). Agreement between the measured and predicted

'A. E. Scheidegger, The Physics of F/ow Through Porous R. Maynard, and G. Toulouse (North-Holland, Amsterdam,
AAdia (Univ. Toronto Press, Toronto, 1974). 1979).
2J. B. Walsh and W. F. Brace, J. Geophys. Res. 89, 9425 (1984). "V. K. S. Shante, Phys. Rev. B 16, 2597 (1977).
3P. Wong, J. Koplik, and J. P. Tomanic, Phys. Rev. 8 30, 6606 '2R. Fisch and A. B. Harris, Phys. Rev. B 18, 416 (1978).
(1984). 3See, for example, L. O. Landau and E. M. Lifshitz, Fluid
4J. Koplik, C. Lin, and M. Vernette, J. Appl. Phys. 56, 3127 Mechanics (Addison-Wesley, Reading, MA, 1959).
(1984). '4See, for example, J. Van Brakel et al. , Powder Technol. 29, 1
5V. Ambegaokar, 8. I. Halperin, and J. S. Langer, Phys. Rev, 8 (1981).
4, 2612 (1971). '5We have demonstrated that the inflection point corresponds to
6See, for example, L. Moscou and S. Lub, Powder Technol. 29, the point at which the mercury first forms a cluster spanning
45 (1981). the sample by simultaneously measuring the electrical resis-
~D. Berman, 8. G. Orr, H. M. Jaeger, and A. M. Goldman, tance of the sample.
Phys. Rev. B 33, 4301 (1986). F. A. L. Dullien, Porous Media: Fluid Transport and Pore
sP. G. de Gennes and E. Guyon, J. Mec. 17, 403 (1978); Structure (Academic, New York, 1979).
J. Chatzis and F. Dullien, J. Can. Pet. Technol. 16, 97 (1977). ~7B. F. Swanson, J. Pet. Technol. 2498 (December, 1981).
9T. R. Madden, Geophysics 41, 1104 (1976); D. A. Seeburger 'SSee, for example, M. R. J. Wyllie and W. D. Rose, Pet. Trans.
and A. Nur, J. Geophys. Res. 89, 527 (1984). AIME 189, 105 (1950).
' S. Kirkpatrick, in Ill-Condensed Matter, edited by R. Balian, '9A. Timur, Log Anal. 9, 8 (1968).