SPE 107954
Improved Permeability Prediction Relations for Low-Permeability Sands
F.A. Florence, SPE, Texas A&M U.; J.A. Rushing, SPE, Anadarko Petroleum Corp.; K.E. Newsham, SPE, Apache Corp.;
and T.A. Blasingame, SPE, Texas A&M U.
Copyright 2007, Society of Petroleum Engineers
This paper was prepared for presentation at the 2007 SPE Rocky Mountain Oil & Gas
Technology Symposium held in Denver, Colorado, U.S.A., 16–18 April 2007.
This paper was selected for presentation by an SPE Program Committee following review of
information contained in an abstract submitted by the author(s). Contents of the paper, as
presented, have not been reviewed by the Society of Petroleum Engineers and are subject to
correction by the author(s). The material, as presented, does not necessarily reflect any
position of the Society of Petroleum Engineers, its officers, or members. Papers presented at
SPE meetings are subject to publication review by Editorial Committees of the Society of
Petroleum Engineers. Electronic reproduction, distribution, or storage of any part of this paper
for commercial purposes without the written consent of the Society of Petroleum Engineers is
prohibited. Permission to reproduce in print is restricted to an abstract of not more than
300 words; illustrations may not be copied. The abstract must contain conspicuous
acknowledgment of where and by whom the paper was presented. Write Librarian, SPE, P.O.
Box 833836, Richardson, TX 75083-3836, U.S.A., fax 01-972-952-9435.
Abstract
This work addresses the problem of estimating Klinkenberg-
corrected permeability from single-point, steady-state
measurements on samples from low permeability sands. The
"original" problem of predicting the corrected or "liquid
equivalent" permeability (i.e., referred to as the Klinkenberg-
corrected permeability) has been under investigation since the
early 1940s — in particular, using the application of "gas
slippage" theory to petrophysics by Klinkenberg.1
In the first part of our work, the applicability of the Jones-
Owens4 and Sampath-Keighin5 correlations for estimating the
Klinkenberg-corrected (absolute) permeability from single-
point, steady-state measurements is investigated. We also
provide an update to these correlations using modern petro-
physical data.
In the second part of our work, we propose and validate a new
"microflow" model for the evaluation of an equivalent liquid
permeability from gas flow measurements. This work is based
on a more detailed application of similar concepts employed
by Klinkenberg. In fact, we can obtain the Klinkenberg result
as an approximate form of our result. Our theoretical "micro-
flow" result is given as a rational polynomial in terms of the
Knudsen number (the ratio of the mean free path of the gas
molecules to the characteristic flow length (typically the
radius of the capillary)).
The following contributions are derived from this work:
● Validation and extension of the correlations proposed by Jones-
Owens and Sampath-Keighin for low permeability samples.
● Development and validation of a new "microflow" model which
correctly represents gas flow in low permeability core samples.
This model is also applied as a correlation for prediction of the
equivalent liquid permeability in much the same fashion as the
Klinkenberg model, although our new model is substantially
more theoretical (and robust) as compared to the Klinkenberg
correction model.
Introduction
The gas slippage phenomenon typically occurs in the labora-
tory when gas flow experiments are conducted at low
pressures. Gas slippage is defined as the condition where the
mean free path of the gas molecules is no longer negligible
compared to the average effective rock pore throat radius —
i.e., the gas molecules tend to "slip" on the surfaces of the
porous media. This effect yields an overestimation of the
measured gas permeability compared to the true absolute
permeability if it were measured using a liquid.
For flow in tubes, the gas slippage phenomenon has been
investigated since the end of the nineteenth century. The first
study of gas slippage in porous media was conducted by
Klinkenberg.1 The Klinkenberg model approximates a linear
relationship between the measured gas permeability and the
reciprocal absolute mean core pressure.2 This model has been
a consistent basis for the development of methods computing
the absolute liquid permeability of a core sample based on a
single data point — i.e., single-point steady-state permeability
measurement methods.
Subsequent work focused on correlating the parameters of the
Klinkenberg model (i.e, the Klinkenberg-corrected permeabil-
ity or equivalent liquid permeability (k∞) and the Klinkenberg
gas slippage factor (bK). Heid et al3 and Jones and Owens4
proposed two correlations similar in form between bK and k∞,
while Sampath and Keighin5 proposed a different form of
correlation using effective porosity (φ) as a third parameter.
At a more theoretical level, the gas slippage phenomenon is
part of a larger area of study — the theory of rarefied gases.
This theory has developed substantially in the last fifty years,
leading to a better understanding and more accurate modeling
of the gas slippage effect. The primary objectives of this work
are:
● To compare and evaluate existing correlations for single-point,
steady-state measurement of permeability, and
● To develop the concept of a new model for gas slippage in low
permeability sands.
2 SPE 107594
Overview of Existing Correlations
The "Klinkenberg" model was developed to account for the
discrepancies between permeabilities measured with gases and
liquids as flowing fluids. This model is based on the argument
developed by Kundt and Warburg6 that, at low pressures, or
more specifically, when the mean free path of the gas
molecules is within two orders of magnitude of the capillary
radius, the velocity of the gas molecules at the surface of the
porous medium is non-zero.
Klinkenberg proved that there exists an approximately linear
relationship between the measured gas permeability (ka) and
the reciprocal mean pressure ( 1 / p ), which is given as:
⎡ b ⎤
k a = k∞ ⎢1 + K ⎥ ................................................................ (1)
⎣ p ⎦
Where bK is the "gas slippage factor" which is a constant that
relates the mean free path ( λ ) of the molecules at the mean
pressure ( p ) and the effective pore radius (r).
The bK-term is defined by:
bK 4cλ
= , where c ≈ 1 .................................................... (2)
p r
In Eq. 1, k∞ is the "equivalent liquid permeability" (which is
also called Klinkenberg-corrected permeability). Klinkenberg
has shown that k∞ is the true, absolute permeability of the
sample. Derivations of the Klinkenberg relations are given in
Appendix A.
Taking the Klinkenberg equation as fact, later work in this
area focused on the determination of the gas slippage factor.
In the late 1940s, Heid et al3 conducted a study including the
effects of pressure, pore size, and the type of porous medium
on permeability. Heid et al determined the gas slippage factor
(bK) and the (extrapolated) permeability (k∞) of 11 synthetic
cores and 164 natural core samples of representative sands
from different producing areas of the United States.
Based on the results of their core sample experiments shown
in Fig. 1, Heid et al3 developed a relationship between the gas
slippage factor (bK) and the corresponding equivalent liquid
permeability (k∞) which is written as:
bK = 11.419(k ∞ ) −0.39 ........................................................ (3)
Fig. 1 Heid et al data and correlation (ref. 3).
Additional research in the 1970s focused on the gas slippage
effect because of the emergence of "unconventional gas
resources" as an important energy source. In particular, much
of the laboratory research in the 1970s/early 1980s addressed
permeability measurements in tight gas sands — which typi-
cally exhibit permeabilities much lower than that of the core
samples Heid et al. studied.
As a case in point, Jones and Owens4 studied the effects of gas
slippage for more than one hundred samples from tight gas
sands. Although the database used by Jones and Owens is not
available, they did provide a correlation similar in form to the
one presented by Heid et al. for the relationship between the
gas slippage factor (bK) and the equivalent liquid permeability
(k∞). The Jones-Owens formula also confirms that the gas
slippage effect is more significant for lower permeability
rocks. The Jones-Owen result is given as:
bK = 12.639(k ∞ ) −0.33 ......................................................... (4)
In 1981 (two years after the Jones and Owens work), Sampath
and Keighin5 studied 10 core samples from a tight gas sand
field in Uinta County, Utah, and published a formula relating
the gas slippage factor (bK) to the ratio of Klinkenberg-
corrected permeability to effective porosity (i.e., k∞/φ). The
Sampath-Keighin correlation is given by:
− 0.53
⎡k ⎤
bK = 13.851 ⎢ ∞ ⎥ ...................................................... (5)
⎣ φ ⎦
The data and the correlation relation developed by Sampath-
Keighin are shown in Fig. 2.
Fig. 2 Sampath-Keighin data and correlation (ref. 5).
Although based on a reduced number of samples, the Sam-
path-Keighin correlation (Eq. 5) is interesting at a theoretical
level since it is noted in the literature7 that the square root of
the Klinkenberg-corrected permeability/porosity ratio is con-
sidered to be a "characteristic length" (recall that permeability
has the dimension of length-squared). Eq. 2 shows that the
Klinkenberg gas slippage factor, bK, is inversely proportional
to the capillary radius — where the capillary radius is related
to the square root of k∞/φ by:
r = 8.886 ×10 −6 k ∞ / φ ...................................................... (6)
SPE 107954
Eq. 6 is derived analytically in Appendix B — we note that r
is the capillary radius (expressed in centimeters) and k∞ is the
Klinkenberg-corrected permeability (in md). Assuming stan-
dard conditions, it is possible to derive a general but rigorous
"square-root" k∞/φ correlation of the form:
−0.5
⎡k ⎤
bK = β ⎢ ∞ ⎥ ....................................................................... (7)
⎣ φ ⎦
The "β-term" in Eq. 7 is a parameter that depends on the type
of gas used in the core flow experiment — we present Eq. 7
computed for different gases on various data correlation plots
later in this paper.
Core Data Used in Current Research
The core data used in our work consists of data gathered in the
literature and from industry sources. The "literature database"
is comprised of data from Heid et al and Sampath and
Keighin. The "industry database" is composed of two datasets
obtained from the Lower Cotton Valley sandstones in North
Louisiana.8,9 The first "Cotton Valley" dataset includes 12
core samples; the second has 18 core samples. We note that
both datasets utilized multi-point, steady-state permeability
measurements. For each sample, we have effective porosity as
well as several gas permeabilities (ka) measured at various
mean core pressures ( p ). Klinkenberg-corrected permeability
(k∞) and gas slippage factor (bK) are estimated using the
classical Klinkenberg plotting method.
We also obtained "older" datasets from a Gas Research
Institute (GRI) project conducted in the 1980s and early 1990s
in the Travis Peak (East Texas) and Frontier (Wyoming)
formations (refs. 10-12). We should note that all of the GRI
Fig. 3 Comparison of the "bK vs. k∞" correlations by Heid et al (ref. 3) and Jones-Owen (ref. 4) with various
field and literature data, acquired with unsteady-state (USS) or steady-state (SS) techniques.
3
datasets were derived from unsteady-state permeability
measurements; however, the new correlations and models
derived in our work are based on the multi-point, steady-state
data. We have included the unsteady-state data for (visual)
reference only — the data are not used in the correlations.
Comparison of Core Data and Correlations
Fig. 3 presents a comparison of available core data with the
Jones and Owens and Heid et al correlations in the bK versus
k∞ format. In Fig. 4 we present a comparison of the available
data with the Sampath-Keighin correlation and our generalized
square-root model given by Eq. 7 (bK versus k∞/φ format plot).
Figs. 3 and 4 serve to highlight trends in the data — as well as
to establish existence of a significant vertical spread in the
data sets — which can lead to errors in permeability estimates
for low permeability reservoirs.
Next, we compare the Jones-Owens correlation, the Sampath-
Keighin correlation, and our new proposed square-root
correlation with the data from the Lower Cotton Valley
formation (or modern "standard" database). The comparison
of the Heid et al and Jones-Owens correlations with the
(Cotton Valley) steady-state data is shown on Fig 5 (bK vs.
k∞). The Sampath-Keighin correlation and our proposed
"square-root" correlation are compared to the steady state data
in Fig 6 (bK vs. k∞/φ).
For both of the Cotton Valley datasets, the steady-state
permeability measurements were performed using nitrogen as
the flowing fluid — and we note that the nitrogen curve on
Fig. 6 overestimates the gas slippage factor (bK), whereas the
carbon dioxide curve provides a good fit. Regardless of the
"intercept" issue, our model matches the data trends well.
4 SPE 107594

Fig. 4 Comparison of the "bK vs. k∞/φ" correlations by Sampath and Keighin (ref. 5) and this work with various field
and literature data, acquired with unsteady-state (USS) or steady-state (SS) techniques.

Fig. 5 Comparison of the steady-state data with the Jones-
Owens and Heid et al correlations. The Heid et al
correlation seems well adapted for these datasets,
whereas the Jones-Owens correlation generally
underestimates the gas slippage factor, especially for
the first Cotton Valley sample set.
Fig. 6 Comparison of the Cotton Valley steady-state data with
the Sampath-Keighin and proposed square-root correla-
tions — although measurements were performed with
nitrogen as the flowing gas, we note the square-root
correlation for carbon dioxide gives the best match (the
best fit is actually obtained for a β-value of 34).
SPE 107954
In Figs 7 and 8 we present the computed equivalent liquid
permeability plotted against the extrapolated Klinkenberg-
corrected permeability on log-log scales for both "Cotton
Valley" datasets. We should note that the "Klinkenberg-
corrected permeability" refers only to the value of permeabil-
ity actually extrapolated from a Klinkenberg plot (k∞ measured),
while the permeability values computed from correlations (k∞
computed) are referred as "equivalent liquid permeabilities."
However, we will use the same symbol (k∞) for either case.
Fig. 7 Computed equivalent liquid permeability versus
extrapolated Klinkenberg-corrected permeability, for the
Lower Cotton Valley Sample No. 1 — the Jones-Owens
and Sampath-Keighin correlations overestimate k∞ and
the theoretical square-root model for nitrogen under-
estimates k∞.
Fig. 8 Computed equivalent liquid permeability versus
extrapolated Klinkenberg-corrected permeability, for the
Lower Cotton Valley Sample No. 2 — the Jones-Owens
and Sampath-Keighin correlations provide fair results
for k∞ > 1 µd. The theoretical square-root model for
nitrogen underestimates k∞, but performs well for low
permeability samples.
5
In Figs. 9 and 10 we present the average absolute relative
errors for the computed equivalent liquid permeability plotted
against the extrapolated Klinkenberg-corrected permeability.
This format helps us to assess the accuracy of the correlation
on a "point-by-point" basis for a particular dataset.
For Cotton Valley Sample No. 1, the Jones-Owens and Sam-
path-Keighin models generally overestimate the permeability.
In both cases (Cotton Valley samples 1 and 2), the theoretical
square-root model (nitrogen curve) underestimates the
Klinkenberg-corrected permeability — but the overall per-
formance of the square-root relation is comparable to the other
methods.
Fig. 9 Average absolute relative errors for equivalent liquid
permeability versus measured Klinkenberg-corrected
permeability, for Lower Cotton Valley Sample No. 1.
Errors are generally less than 35 percent.
Fig. 10 Average absolute relative errors for equivalent liquid
permeability versus measured Klinkenberg-corrected
permeability, for Lower Cotton Valley Sample No. 2.
Errors are generally less than 30 percent, although there
are numerous outliers which exhibit greater than 50
percent error.
As shown in Figs. 9 and 10, the three models for k∞ exhibit a
fairly high absolute relative error (>50 percent for some cases)
— however; the "clustering" for both datasets suggests that
6 SPE 107594

most samples exhibit less than 30 percent error. Given the

data and the relatively approximate nature of the correlations,
we consider the performance to be good, perhaps very good.
While it is difficult to discriminate the "best" of the three
correlations based on our results, we believe the Sampath-
Keighin correlation and the generalized square-root
correlations are more "consistent" with the theory — and as
such, should be favored for this application.
Similar to previous work, the objective of our research is to
estimate the bK-term accurately, and then utilize this estimate
in Eq. 1 to yield the Klinkenberg-corrected permeability. A
general procedure for "single-point data" is to use any given
pair of ka and φ values to estimate bK, and then to use the bK
and p values to estimate k∞. In our case, we will use a model
(i.e., correlation) for bK and substitute this result into Eq. 1 to
estimate k∞.
As an alternative, one could simply establish a correlation of
k∞ = f(ka) — this process was proposed by Jones and Owens
based on the gas permeability (ka) estimated at 100 psig. This
correlation is given as:
2
k ∞ = 10 ( −0.398 log ka +1.067 log ka −0.0825) , for 0.1 µd < k a < 1 md
........................................................................................... (8)
The purpose of our work is to establish a theoretically robust
correlation of k ∞ = f ( p, φ ,..., k a ) — so we will not pursue a
result of the form of Eq. 8.
Klinkenberg Model — Revisited
We begin our discussion of the Klinkenberg concept by first
returning to the fundamentals — we will utilize the same
measurements as with the correlations presented in the
previous section (i.e., ka, p , φ, and backpressure (if used)) —
but we now consider the rigorous mechanisms of flow at the
microscopic (and smaller) scales. As a first step, we accept
the Klinkenberg approximation as a "first order" type of
estimate — Eq. 1, repeated below for clarity:
⎡ b ⎤
k a = k∞ ⎢1 + K ⎥ ................................................................ (1)
⎣ p ⎦
We "define" k∞ as the true, equivalent liquid permeability of
the system, but we will also use estimates of k∞ estimated
using the Klinkenberg correction as a standard to correlate
against. That is, we will use the Klinkenberg-corrected per-
meability (k∞) as reported (or calculated) from our data
sources as the reference permeability. Obviously, we would
prefer to use a more rigorous estimate, but for the present
study, we must prove the concept model (i.e., the "microflow"
model presented in this section) against some standard — and
we believe that the Klinkenberg-corrected estimates are
appropriate for that purpose.
In Fig. 11, we present the Klinkenberg-corrected (equivalent
liquid) permeability (k∞) plotted against the measured gas
permeability (ka) on log-log coordinates as a "correlation" to
ensure that, at least directionally, Eq. 1 is valid.
Fig. 11 Klinkenberg-corrected (extrapolated) permeability (k∞)
versus measured gas permeability (ka).
As a primer for the development of our microflow model, we
consider the phenomenon of gas slippage at a fundamental
level — as a phenomenon which occurs as a subset of a much
larger area of study known as the theory of rarefied gases.
With the development of the common interest for aeronautics
and aerospace, as well as more recently, the development of
MEMS (Micro Electro-Mechanical Systems), the flow of
gases at very low pressure — or, more aptly, the flow of gas
molecules with a high mean free path (the mean free path
being inversely proportional to the pressure) has been
thoroughly studied in the last half century.
In the pursuit of a new microflow model, it then seems natural
to revisit the Klinkenberg model, if for no other reason than to
confirm that our new microflow model reverts to the
Klinkenberg model under certain conditions (which it does).
The flow regime for a gas flowing in a micro-channel is
typically determined by the value of the Knudsen number (Kn)
— which is defined as:13
λ
Kn = ......................................................................... (9)
lchar
In Eq. 9, λ is the mean free path of the gas molecules (i.e., the
average distance (length) between two consecutive molecular
interactions) and lchar is the characteristic length of the flow
geometry (e.g., channel height, pipe radius).
For our purposes, the Knudsen number is difficult to define
rigorously for a porous medium — but for the sake of
argument, we presume that some "characteristic length" can be
estimated for a porous medium. More generally, we assume
that we can define the Knudsen number based on the pro-
perties of the porous medium.
SPE 107954 7

The classical definition of the mean free path from thermo- The volumetric gas flowrate (q) flowing through a capillary of
dynamics is: radius r and length L is given by:13

λ ( p, T ) = π / 2
1
µ
RT
[ µ ≡ µ ( p, T )] ......................... (10) π r4 ⎡ 4 Kn ⎤ ∆p
q= (1 + α ( Kn) Kn) ⎢1 + ⎥ ........................... (11)
p M 8 µ ⎣ 1 + Kn ⎦ L
where: The α(Kn)-term is defined by Karniadakis and Beskok13 as:
p = mean pressure (absolute)
T = absolute temperature α ( Kn) =
128
15π 2
[ ]
tan -1 4 Kn 0.4 ............................................. (12)
µ = viscosity of the gas at T and p
R = universal gas constant Using the same capillary model as Klinkenberg,1 we derived
M = molecular weight of the gas the following "microflow" model which relates the measured
permeability to gas (ka), the equivalent liquid permeability
This particular definition (Eq. 10) is based on the kinetic
(k∞), and the Knudsen number (Kn):
theory for a perfect gas.14
⎡ 4 Kn ⎤
In addition to the definitions above, we also must consider k a = k ∞ [1 + α ( Kn) Kn] ⎢1 + ..................................... (13)
⎣ 1 + Kn ⎥⎦
various other flow regimes as follows (see Fig. 12):
● Continuum Flow Regime: For Kn < 0.01, the mean free Eq. 13 is rigorously derived and should be valid for all low
path of the gas molecules is negligible compared to the pressure/low velocity flow regimes which exist for gas flow in
characteristic dimension of the flow geometry (i.e., the lchar porous media. We have not investigated the application of Eq.
-parameter). In this case the continuum hypothesis of fluid 13 for high pressure and/or high velocity flow.
mechanics is applicable (i.e., the system is described by the
Navier-Stokes equations). Correlation of the Knudsen Number with Porosity,
● "Slip-Flow" Regime: For 0.01 < Kn < 0.1, the mean free Permeability and Mean Pressure
path is no longer negligible, and the slippage phenomenon The Knudsen number (Kn), unlike the mean pressure ( p ),
appears in the "Knudsen" layer (layer of gas molecules cannot be measured by direct laboratory measurements. Our
immediately adjacent to the wall). objective is to design a model for estimating the equivalent
● "Transition" Regime: For 0.1 < Kn < 10. liquid permeability (k∞) using single-point, steady-state mea-
● Free Molecular Flow Regime: For Kn > 10, the flow is surements (i.e., using a single pair of ka and p values). To
dominated by diffusive effects. remedy this issue, we propose a "pseudo" Knudsen number
(Knp) — which is defined as a function of the "typically"
measured parameters (i.e., p , φ, ka).
Assuming that the core flow experiments are conducted at
isothermal conditions (i.e. there is no temperature gradient in
the core) and that variations in viscosity are negligible over
the range of pressures considered, then the Knudsen number
(defined by Eq. 9) is inversely proportional to both the mean
pressure ( p ) and the characteristic flow geometry length
(lchar). Typically the pore throat radius (defined by Eq. 6 as a
function of the equivalent liquid permeability (k∞) and the
porosity (φ)) will be used to estimate lchar.
Substituting Eq. 12 into Eq. 13 yields
⎡
k a = k ∞ ⎢1 +
⎣ 15π
128
2
[ ⎤⎡
⎦ ⎣⎢
]
tan -1 4 Kn p 0.4 Kn p ⎥ ⎢1 +
1
4 Kn p ⎤
⎥
+ Kn p ⎦⎥
........................................................................................ (14)
where Knp is the "pseudo" Knudsen number. In order to
Fig. 12 Limits of the different flow regimes, as a function of utilize Eq. 14, we must assume that the mean pressures,
characteristic length lchar, and reciprocal mean free
path normalized (1 atm, 300K). Lines defining various porosities, and the gas and equivalent liquid permeabilities are
flow regimes are based on flow of air at isothermal available. To solve for the "pseudo" Knudsen number (Knp)
conditions (Modified from ref. 13). for a particular case, we then rearrange Eq. 14 into the
Karniadakis and Beskok13 developed a unified model for gas following "root solution" form:
flow in micro-tubes (see Appendix C) where this model is
valid over the entire range of flow regimes.
⎡
k a − k ∞ ⎢1 +
⎣ 15π
128
2
[ ⎤⎡
⎦ ⎣⎢
]
tan -1 4 Kn p 0.4 Kn p ⎥ ⎢1 +
1
4 Kn p ⎤
⎥=0
+ Kn p ⎦⎥
........................................................................................ (15)
8 SPE 107594

Applying the Pseudo-Knudsen Number Concept Substituting Eq. 18 into Eq. 15 and rearranging yields:
To demonstrate an application of the "pseudo" Knudsen
k a ⎡⎢ ⎡ ⎤ ⎤⎡ 1 ⎤⎤
0.4
128 -1 ⎡ 1
number (Knp), we only use the Lower Cotton Valley data (No. = 1+ tan ⎢4⎢a0 k ∞ a1φ a2 ⎥ ⎥ ⎢a0 k ∞ a1φ a2 ⎥ ⎥ ×
2) (ref. 9), from which we derive two separate correlations for k ∞ ⎢ 15π 2 ⎢ ⎣ p ⎦ ⎥⎦ ⎣ p ⎦ ⎥⎦
⎣ ⎣
Knp (see Appendix D). Our first correlation relates Knp to the
reciprocal mean pressure, the porosity, and the gas per- ⎡ ⎤
⎢ ⎥
meability as follows: ⎢1 + 4 ⎥
1 − 0.598 − 0.352 ⎢ ⎡ 1 a1 a2 ⎤ ⎥
Kn p = 0.4 ka φ ............................................ (16) ⎢ 1 + 1 / ⎢a0 k ∞ φ ⎥ ⎥
p
⎣⎢ ⎣ p ⎦ ⎦⎥
As implied, Eq. 16 was derived for the specific case of the ............................................................................................ (19)
Cotton Valley No. 2 data, in a process where Knp (Eq. 15) was
estimated using known data for ka, p , and k∞. After Knp was
obtained as the root of Eq. 15, we then correlated these Knp
values with the ka, p , and φ data to yield Eq. 16.
At this point, Eq. 16 can be substituted into Eq. 15 — and this
result can be solved directly for the equivalent liquid
permeability (k∞). Obviously, k∞ data must be available to
"calibrate" Eq. 15 (i.e., to estimate Knp). However, we believe
that the calibration of Eq. 16 may (in future work) be reduced
to the determination of the intercept term (the ka exponent is
approximately -0.5 and the φ exponent may be imposed as 0.5
— similar to the "square-root" correlations).
As an alternative, we can follow the exact same calibration
procedure, but this time substitute k∞ for ka in Eq. 16 (i.e., the
Knp = f(ka, p ,φ ) correlation). For this case, we obtain:
1 − 0.553 0.3897
Kn p = 2.62 k∞ φ .......................................... (17)
p
The equivalent liquid permeability values computed using Eq.
15 (based on the values of Knp from Eq. 16 or 17) are Fig. 13 Computed (Klinkenberg) equivalent liquid permeability
(k∞) versus Klinkenberg-corrected (extrapolated)
correlated with the "given" Klinkenberg-corrected permea- permeability (k∞). (pseudo-Knudsen number approach,
bility data for this case in Fig. 13. implicit and explicit relations for Knp)
We recognize that Klinkenberg-corrected permeability data The "trick" is to obtain the "calibration" coefficients (a0, a1,
may not always be available in practice — and the process of and a2) in Eq. 19. As noted earlier, Eq. 19 can be formulated
solving simultaneously for the pseudo-Knudsen number (Knp) as an "equation-of-state" and the coefficients (a0, a1, and a2)
and the equivalent liquid permeability (k∞) will require an can be tuned using non-linear regression — provided that
implicit formulation (an analog in the field of phase behavior there are sufficient data measurements, and that such measure-
would be an equation-of-state (EOS) which is implicit in fluid ments are taken for samples from the same depositional
density). sequence.
In such a case, (i.e., an implicit relation) we would have to We must note that we do not in any manner propose that Eq.
determine coefficients for the Knp = f(k∞, p ,φ ) simul- 18 is "universal" (i.e., one set of coefficients (a0, a1, and a2)
taneously with the estimation of k∞ using Eq. 15. Writing the does not apply to all possible data cases, Eq. 18 must be
correlation for the pseudo-Knudsen number in general form, calibrated for each dataset). For this case, we did formulate
we have: Eq. 19 as a "fully implicit" solution using non-linear regres-
1 sion and we achieved following results:
Kn p = a0 k ∞ a1φ a2 ........................................................ (18)
p a0 = 0.93
where Eq. 18 provides a "correlation" for the variables in this a1 = -0.49
problem — and this relation allows us, in theory, to solve Eq. a2 = 0.13
15 for the equivalent liquid permeability (k∞) in an implicit The results of our "fully implicit" correlation are shown in
fashion. Fig. 14.
SPE 107954 9

will provide significant improvement for the estimation of

equivalent liquid permeability from typical steady-state (gas)
permeability measurements.

Nomenclature
Variables
b = General slip coefficient
bK = Klinkenberg gas slippage factor, psi
c = Proportionality constant
Kn = Knudsen number, dimensionless
Knp = Pseudo-Knudsen number, dimensionless
ka = Apparent gas permeability, md
Equivalent liquid permeability or Klinkenberg-
k∞ =
corrected permeability, md
lchar = Characteristic length of the flow geometry, cm
p = Mean core pressure, psia
∆p = Differential pressure, psi
q = Gas flowrate, cc/sec
r = Effective pore radius, cm
α = Rarefaction coefficient parameter, dimensionless
Fig. 14 Computed (Klinkenberg) equivalent liquid permeability µ = Gas viscosity, cp.
(k∞) versus Klinkenberg-corrected (extrapolated) λ = Mean free path of the gas molecules, cm
permeability (k∞). (fully implicit formulation for the φ = Porosity, fraction
pseudo-Knudsen number).
References
Summary and Conclusions
1. Klinkenberg, L.J.: "The Permeability of Porous Media to Liquid
Summary and Gases," paper presented at the API 11th Mid Year Meeting,
The development of the microflow model for permeability Tulsa, Oklahoma (May 1941); in API Drilling and Production
prediction in low permeability gas sands offers an alternative Practice (1941), 200-213
to the classical Klinkenberg model, which is the long-time 2. RP40, Recommended Practices for Core Analysis, 2nd edition,
reference for measuring permeability. Although the Klinken- API, Washington, DC (1998).
3. Heid, J.G., et al: "Study of the Permeability of Rocks to
berg model is the universally accepted standard, we believe it
Homogeneous Fluids," in API Drilling and Production Practice
may yield poor results for very or ultra low permeability core (1950), 230-246
samples. 4. Jones, F.O. and Owens, W.W.: "A laboratory Study of Low
Our new model has been validated using field data from a low Permeability Gas Sands," paper SPE 7551 presented at the 1979
permeability reservoir case, but this "proof" is by no means SPE Symposium on Low-Permeability Gas Reservoirs, May 20-
22, 1979, Denver, Colorado.
exhaustive — additional validation is warranted.
5. Sampath, K. and Keighin, C.W.: "Factors Affecting Gas Slippage
Conclusions in Tight Sandstones," paper SPE 9872 presented at the 1981
1. The Jones-Owens, Sampath-Keighin, and our "Square-Root" SPE/DOE Low Permeability Symposium, May 27-29, 1981,
correlations may be used satisfactorily for single point, steady- Denver, Colorado.
state measurements as mechanisms to estimate the equivalent 6. Kundt, A. and Warburg, E.: "Über Reibung und Wärmelei-tung
liquid permeability. The Sampath-Keighin and Square-Root verdünnter Gase," Poggendorfs Annalen der Physik und Chemie
correlations should be preferred based on the theoretical (1875), 155, 337.
formulations for these models. 7. Jones, S.C.: "Using the Inertial Coefficient, β, to Characterize
2. The microflow model presented in this work is promising Heterogeneity in Reservoir Rock," paper SPE 16949 presented at
as it provides a second-order correction for gas slippage the 1987 SPE Annual Technical Conference and Exhibition,
(beyond the "first-order" Klinkenberg formulation. The use September 27-30, 1987, Dallas, Texas
of such a correction should be especially relevant for low to 8. Lower Cotton Valley Formation Core Report — Anadarko
ultra low permeability core samples. Petroleum Corp., (2005)
9. Lower Cotton Valley Formation Core Report — Anadarko
Recommendations/Comment Petroleum Corp., (2006)
We note that the theoretical square-root correlation (Appendix 10. Travis Peak Formation Core Report — Well Howell No. 5, S.A.
B) has a limited accuracy in this work. However, the square- Holditch, (1986)
11. Travis Peak Formation Core Report — Well S.F.E. No. 2, S.A.
root model validates use of the permeability/porosity ratio in
Holditch, (1987).
modeling the gas slippage factor and serves to connect the 12. Frontier Formation Core Report — Well S.F.E. 4-24, S.A.
classical Klinkenberg model with our new "microflow model." Holditch, (1991).
Upon further validation, we believe that our microflow model
10 SPE 107594

13. Karniadakis, G.E. and Beskok, A.: Micro-flows, Fundamentals yields:

and Simulation, Springer-Verlag, New-York (2002).
dv
14. Loeb, L.B.: The Kinetic Theory of Gases, second edition, d (v + δr )
McGraw-Hill Co. Inc., New York City (1934). 2πµL(r + δr ) dr =
15. NIST Chemistry Webbook, NIST, http:\\webbook.nist.gov. dr
⎡ dv d 2v dv d 2v ⎤
Appendix A: Derivation of Klinkenberg Equation 2πµL ⎢ r + δr (r + ) + δr 2 ⎥
⎢⎣ dr dr 2 dr dr 2 ⎥⎦
This appendix presents our derivation of the Klinkenberg
equation (ref. 1). Note that our derivation uses components ...................................................................................... (A-6)
that Klinkenberg did not employ originally (i.e., elements of In Eq. A-6, the term δr2(d2v/dr2) is negligible compared to the
the literature for the kinetic theory of gases (ref. 14)). other terms — neglecting this term and substituting Eq. A-6
into Eq. A-5 yields:
Gas Flow Through a Straight Capillary. We consider a
capillary tube of radius (R0) and length (L) with its axis dv
2π∆prδr = 2πµLr −
coincident with the x-axis. Gas is flowing through the dr
capillary as a result of a pressure drop (∆p = (p1 - p2)) across ⎡ dv d 2v dv ⎤
the length of the capillary tube. Assume that the velocity of 2πµL ⎢r + δr (r + )⎥
⎢⎣ dr dr 2 dr ⎥⎦
the flowing gas (v) is a function only of the distance r from the
x-axis. Consider now a cylindrical shell of length (L) between ...................................................................................... (A-7)
the cylinders of radii r and r+δr (see Fig. A.1). The force (F1) Expanding and simplifying Eq. A-7 gives:
acting on the cross-section of this shell is the normal pressure ⎡ d 2 v dv ⎤
due to the flowing gas. In derivative form, this force is: r∆p = − µL ⎢r + ⎥ .................................................. (A-8)
dF1 = 2π∆prδr ................................................................ (A-1) ⎢⎣ dr 2 dr ⎥⎦

Assuming steady-state fluid motion, this force is balanced by Dividing Eq. A-8 by (-rµL) yields the following differential
the viscous drag exerted by the flowing gas on the outer and equation:
the inner surfaces (along the x-axis) of the cylindrical shell. d 2v 1 dv ∆p
The viscous drag is defined by: + =− ........................................................ (A-9)
dr 2 r dr µL
dv
F=µS .................................................................... (A-2) A particular solution of Eq. A-9 (i.e., a non-linear differential
dr
equation) is:
where: ∆p 2
µ = Coefficient of viscosity of the gas. v = A0 − r ........................................................... (A-10)
4µL
S = Surface considered (2πrL for the inner surface,
2π (r+δr)L for the outer surface). where A0 is a constant fixed by the boundary conditions. Two
different boundary conditions can be used — slip-flow at the
The gas velocity reaches its maximum on the axis of the cylin- wall or no slip-flow at the wall.
der (i.e., r=0) and decreases radially towards the wall to a
velocity of zero — therefore the velocity gradient dv/dr is If no slip-flow is assumed at the wall, the condition is v = 0
∆p
negative. The viscous drag exerted on the inner surfaces is: for r = R0; this imposes A0 = R0 2 . The velocity solution
4 µL
dv
dFinner Surface = −2πµLr ............................................. (A-3) to Eq. A-10 is then:
dr
∆p
The component of the viscous drag on the outer surface of the v= ( R0 2 − r 2 ) ...................................................... (A-11)
cylindrical shell, where the gas velocity is lower, is: 4 µL
⎡ dv ⎤ The gas flowrate (qg) is obtained by integrating the gas flow-
d ⎢v + δr ⎥
dr ⎦ rate through the considered cylinder shell section (v(2πrdr))
dFOuter Surface = −2πµL(r + δr ) ⎣ ................... (A-4) over the cross-section of the capillary tube, which yields the
dr
Poiseuille equation:
The resulting viscous drag exerted on the entire shell is given
R0 ∆p π ∆p 4
by:
dF2 = dFOuterSurface − dFInnerSurface
qg =
∫0 2πr
4 µL
( R0 2 − r 2 ) dr =
8 µL
R0 ................ (A-12)

The equilibrium yields dF1 = dF2, hence: If the gas slippage is considered at the wall, the condition can
be written v = v0 for r= R0 (where v0 is non-zero). Kundt and
⎡ dv ⎤
d ⎢v + δr ⎥ Warburg (ref. 6) proved that the velocity at the wall (v0) is
dv dr ⎦
2π∆prδr = 2πµLr − 2πµL(r + δr ) ⎣ ............... (A-5) proportional to the velocity gradient at the wall ( [dv / dr ]r = R0 ),
dr dr
as given by:
Expanding the second term on the right-hand side of Eq. A-5
SPE 107954 11

v0 = −cλ [dv / dr ]r = R ................................................... (A-13) Rewriting Eq. A-20 in terms of superficial velocity (vl):
0
ql
In Eq. A-13, c is a constant with a value slightly less than 1 (as vl =
Atube
given by Kundt and Warburg) and λ is the average mean free ( πR0 2 = Atube) .............................. (A-21)
path of the gas molecules (i.e., the mean free path of the gas 1 R 2 1 ∆p
= N 0
molecules at the mean pressure p (see ref. 14)), defined by 3 8 µ L
p = [ p1 + p2 ] / 2 ). The velocity gradient at the wall is obtained
by differentiating Eq. A-10:
dv ∆p
=− r ............................................................... (A-14)
dr 2 µL
Applying the condition given by Eq. A-13 to Eq. A-14 yields:
∆p
v0 = cλ R0 ............................................................ (A-15)
2 µL Fig. A.1 System schematic for a tube (Loeb (ref. 14)).

Using Eq. A-12 in Eq. A-9, at r = R0, and rearranging yields: Darcy's law gives:
A0 =
∆p
4µL
[ ]
R0 2 + 2cλ R0 ............................................... (A-16)
ql = kl ( L2 )
1 ∆p
(L2 = Acube (i.e., rock sample))..... (A-22)
µ L
Substituting Eq. A-16 for A0 in Eq. A-10 gives: Rewriting Eq. A-22 in terms of superficial velocity (vl):
v=
∆p
4 µL
[ ]
R0 2 − r 2 + 2cλ R0 .......................................... (A-17) vl =
ql
Acube
(L2 = Acube (i.e., rock sample))..... (A-23)
As previously done for Eq. A-12, the gas flowrate (qg) is 1 ∆p
= kl
obtained by integrating v(2πrdr) over the whole radius of the µ L
capillary:
where kl is the permeability to liquid of the porous medium —
qg =
∫0
R0 ∆p
4 µL
[R 2 2
0 − r + 2cλ R0 dr ] which is also the absolute permeability of the porous medium
(provided that the saturation is 100 percent — i.e., single-
π ∆p 4 ⎡ 4cλ ⎤ phase flow). Equating Eqs. A-19 and A-20 and simplifying
= R0 ⎢1 + ⎥ yields:
8 µL ⎣ R0 ⎦
R 2 1 R0 2
1 ∆p ⎡ 4cλ ⎤ kl = N .............................................................. (A-24)
= 0 πR0 2 ⎢1 + ⎥ 3 8
8 µ L ⎣ R0 ⎦
If we assume that that the flowing fluid is a gas, with a gas
..................................................................................... (A-18)
slippage condition at the wall (v0 ≠ 0), the gas flowrate (qg) is
Re-writing Eq. A-18 in terms of superficial velocity (vl): given by Eq. A-18:
ql
vg = qg R 2 1 ∆p ⎡ 4cλ ⎤
Atube vg = = 0 ⎢1 + ⎥ ................................ (A-19)
πR0 2 8 µ L ⎣ R0 ⎦
ql R 2 1 ∆p ⎡ 4cλ ⎤
= = 0 ⎢1 + ⎥
πR0 8 µ L ⎣
2 R0 ⎦ 1 R0 2
Substituting kl = N into Eq. A-19 for yields:
..................................................................................... (A-19) 3 8
1 ∆p ⎡ 4cλ ⎤
Gas Flow Through an Idealized Porous Medium. The v g = kl ⎢1 + ⎥ ................................................ (A-25)
"Klinkenberg" idealized porous medium is composed of solid µ L ⎣ R0 ⎦
material through which the capillaries are oriented randomly For the same gas flow rate, Darcy's law gives:
and have all the same radius, r. The direction of flow is 1 ∆p
parallel to one of the planes of the cube, and let there be N vg = ka .............................................................. (A-26)
µ L
capillaries. The system for Klinkenberg's idealized porous
medium is shown in Fig. A-1. Where ka is the apparent permeability to gas. Setting Eq. A-
The liquid flow rate (ql) through the capillary is given by 25 equal to Eq. A-26 and simplifying, yields:
Poiseuille's law as: (3-dimensional flow, so Klinkenberg used ⎡ 4cλ ⎤
k a = kl ⎢1 + ⎥ ......................................................... (A-27)
1/3 of the total flow for a particular direction). ⎣ R0 ⎦
1 R0 2 1 ∆p As the mean free path ( λ ) is proportional to the reciprocal
ql = N πR0 2 ( πR02 = Atube) ................... (A-20)
3 8 µ L mean pressure ( p ), we can define:
12 SPE 107594

4cλ bK ⎡ ⎤
≡ ................................................................... (A-28) r2 ⎥ 2
r p ⎢n × r = 8k → φ r 2 = 8k ........................................ (B-7)
⎢ 2⎥
R0 ⎦
⎣
where bK is a coefficient of proportionality, known as the
Klinkenberg gas slippage factor. Substituting Eq. A-28 for Or, solving Eq. B-7 for the "equivalent capillary" radius, r, we
obtain:
4cλ / r in Eq. A-24 gives the Klinkenberg equation:
r = 2 2 × k / φ ............................................................. (B-8)
⎡ b ⎤
k a = kl ⎢1 + K ⎥ (kl ≡ k ∞ for our definition) ................... (A-26)
⎣ p ⎦ In Eq. B-8, the "units" of permeability depend on the units
system. For consistency, we must apply a conversion factor,
Appendix B: Derivation of a Theoretical Square- C0, as a multiplier (e.g., when k is in md and r is in cm — C0 =
Root Correlation 3.1415×10-6 cm / md , recall that 1 Darcy = 9.86923×10-9
Theoretical Capillary Radius. Considering the flow of a cm2). The general form of Eq. B-8 with the units conversion
liquid through a capillary tube of inner radius r and length L, factor (C0) is:
the fluid flow rate (q) is given by Poiseuille’s law as: r = 2 2 × C0 × k / φ .................................................. (B-9)
r 2 1 ∆p π 1 ∆p 4
q = πr 2 = r ....................................... (B-1) Using C0 = 3.1415×10-6, Eq. B-9 becomes:
8 µ L 8 µ L
r = 8.886 × 10 −6 k / φ .................................................. (B-10)
where µ is the viscosity of the fluid and ∆p is the pressure
drop across the length of the tube. Considering a cylinder of Theoretical Square-Root Correlation. To correct for the
idealized porous medium with a radius R0 and a length L, effects of gas slippage in permeability measurements, Klin-
composed of n identical capillary tubes (such as described kenberg (ref. 1) derived an approximate linear relationship
above), with the same orientation — parallel to the axis of the
between the measured gas permeability (ka) and the reciprocal
cylinder. We note that for this discussion (as opposed to our
mean pressure ( p ). This result is given as:
previous work in this section), that R0 is now defined as the
outer radius of the bulk core sample. By definition, the ⎡ b ⎤
k a = k ∞ ⎢1 + K ⎥ ......................................................... (B-11)
porosity φ of a porous medium is given by: ⎣ p ⎦
Void Volume πr 2 L r2 In Eq. B-11, k∞ is the Klinkenberg-corrected permeability and
φ= =n =n ................................. (B-2)
Bulk Volume πR02 L R02 bK is the Klinkenberg gas slippage factor. The gas slippage
The total volumetric flow rate (qtot) of a fluid flowing through factor (bK) is a "constant" which relates the mean free path of
this cylinder is defined by multiplying Eq. B-1 by the number the gas molecules ( λ ) at the mean (absolute) pressure ( p )
of tubes, n. This gives: and the effective pore-throat radius (r), as given by:
r2 1 ∆p π 1 ∆p 4 4cλ bK
qtot = n (πr 2 ) =n r ............................ (B-3) = ................................................................... (B-12)
8 µ L 8 µ L r p

For flow in a porous medium, Darcy's law is defined as: In Eq. B-12, c is a constant very close to unity (ref. 6). We
will consider c = 1 for the remainder of this derivation.
k ∆p
qtot = A ................................................................ (B-4) Rearranging Eq. B-12 yields:
µ L
4
In Eq. B-4, k is the absolute permeability of the porous bK = λ p ................................................................... (B-13)
r
medium, and A is the cross-sectional area of the cylinder.
Since A = π R02, Eq. B-4 yields: Substituting r in Eq. B-13 with Eq. B-10 gives:
− 0.5
k ∆p 4 ⎡k ⎤
qtot = (πR02 ) .......................................................... (B-5) bK = λ p⎢ ⎥ .................................... (B-14)
µ L 8.886 × 10 − 6 ⎣φ ⎦
Equating Eqs. B-3 and B-5, and rearranging yields: In Eq. B-14, the mean free path of the gas molecules is
r4 defined by: (ref. 14)
n = k R02 ................................................................... (B-6)
8 1 RT
λ ( p, T ) = π / 2 µ [ µ ≡ µ ( p, T )] ...................... (B-15)
Rearranging Eq. B-6 gives: p M
⎡ ⎤ In this definition, R is the universal gas constant, M is the
⎢n r2 ⎥ 2 molecular weight of the gas and µ is the viscosity of the gas at
× r = 8k
⎢ R02 ⎥⎦ the corresponding thermodynamic state (i.e., µ is a function of
⎣
the absolute pressure, p , absolute temperature, T, and
Substituting Eq. B-2 into the above result, we have:
composition of the gas). For clarity, the subsequent formulae
are expressed in S.I. units, unless otherwise specified.
SPE 107954 13

Table B.1 Intercept for Eq. B-20, various gases..

The square-root correlation developed in this work conforms 4

Molecular Gas Viscosity at Eq. B-20

to the following assumptions: Weight 1atm and 298 K Intercept
● The temperature in the core sample is uniform during Flowing Gas (kg/kg-mole) (Pa.s) (psi)
steady-state flow (assumed: T = 298 K). hydrogen 2.0159 8.845x10-6 80.236
● The gas used for the experiments is nitrogen (M = helium 4.0026 1.985x10-5 127.802
28.01348 kg/kg-mole). air 28.9586 1.842x10-5 44.106
nitrogen 28.01348 1.781x10-5 43.345
● Assuming that, in the range of pressure applied in the carbon dioxide 44.0095 1.503x10-5 29.181
laboratory, nitrogen behaves as an ideal gas, the variation
of the nitrogen viscosity with pressure is negligible; hence These estimates are based on the fluid properties of ref.2 and
µ N2 (T , p ) ≈ µ N2 ,1 atm (T ) . the NIST correlations (ref.15) as appropriate.
Nitrogen viscosity as a function of temperature ( µ N 2 ,1 atm (T ) ,
expressed in Pa.s) is computed using Sutherland's equation:
(ref. 2)
13.85T 1.5
µ N2 ,1 atm (T ) = × 10 −7 ................................... (B-16)
T + 102
The value of the universal gas constant, R = 8,314 J/K/kg-
mole. The product λ p in Eq. B-15 can be calculated using
Eq. B-15 and B-16:
1 8314 × 298 13.85 × 2981.5
λp = π /2 × × 10 − 7 × p = 0.00664
p 28.01348 298 + 102
........................................................................................ (B-17)
To obtain the radius, r, in meters (recall that the base unit of
the mean free path λ is the meter), Eq. B-10 (expressed in
traditional units) becomes;
r = 8.886 × 10 −8 k / φ (B-18)
Fig. B.1 Intercept of the square root model versus molecular
Using the result from Eq. B-17 and Eq. B-18 in Eq. B-14 weight of the gas used in the measurements — the
yields: viscosity of the different gases except H2 is similar
(see Table B.1), hence the λp-term (Eq. B-17) varies
− 0.5 − 0.5 approximately with the reciprocal square-root of the
4 × 0.00664 ⎡ k ⎤ ⎡k ⎤
bK = ⎢ ⎥ = 2.9885 × 105 ⎢ ⎥ ........ (B-19) molecular weight.
8.886 × 10 −8 ⎣ φ ⎦ ⎣φ ⎦
Where bK is expressed in Pa. Converting bK to psi yields: Appendix C: A Rigorous Microflow Model Applied to
− 0.5 the Problem of Gas Flow through Porous Media
⎡k ⎤
bK = 43.345 ⎢ ⎥ .................................................... (B-20) Unified Flow Model Gas Flow in Pipes. Karniadakis and
⎣φ ⎦
Beskok (ref. 13) developed a unified model that predicts volu-
The significance of Eq. B-20 is that this result establishes the metric and mass flow rates for gas flow in channels and pipes
basis for the bK versus k / φ correlation given by Sampath over the entire Knudsen regime (i.e., all flow regimes). The
and Keighin (ref. 5). The exponent (-0.5) is established Karniadakis-Beskok "microflow" model is given (without
rigorously from theory as shown above. derivation) as:
π 1 ∆p ⎡ 4 Kn ⎤
The intercept (43.345) is based on the following assumptions: q= lchar 4 [1 + α ( Kn) Kn] ⎢1 + ⎥ ................ (C-1)
8 µ L ⎣ 1 − bKn ⎦
● ambient temperature (T = 298 K) and
● ambient pressure (1 atm). where:
However, these assumptions are probably not significant q = Volumetric flow rate in the conduit, cc/sec
compared to the assumption of nitrogen as the reference gas. lchar = Characteristic length of the flow geometry (e.g.,
In Table B.1 we summarize our estimates of the intercept channel height, pipe radius), cm
coefficient for hydrogen, helium, air, nitrogen, and carbon L = Length of conduit, cm
dioxide. We have also shown the intercept of the square-root ∆p = Pressure drop across the length of the conduit, atm
model plotted against molecular weight for hydrogen, helium, µ = Gas viscosity at temperature and pressure, cp
air, nitrogen, and carbon dioxide in Fig. B.1. b = Dimensionless slip coefficient, (b is defined as -1)
α(Kn) = Dimensionless term in the rarefaction coefficient
In Eq. C-1, the Knudsen number (Kn) is defined by:
14 SPE 107594

λ terms in S.I. units):

Kn = ..................................................................... (C-2)
lchar
where λ is the mean free path of the gas molecules (i.e., the
average distance (length) between two consecutive molecular
interactions).
We use a value of -1 for the general slip coefficient (b) as
recommended by Karniadakis and Beskok. The role of the
rarefaction coefficient [1+α(Kn) Kn] is to account for the
transition between the "slip-flow" regime (for which Klin-
kenberg model was developed) and the "free molecular flow"
regime. In the "slip-flow" regime (i.e., 0.01 < Kn < 0.1), the
rarefaction coefficient is equal to 1 (i.e., α = 0); in the "free
molecular flow" regime (i.e., Kn > 10), the volumetric flow-
rate is independent of the Knudsen number and the parameter
α tends toward a constant value (for Kn → ∞ ).
For reference, the "Knudsen" flow regimes are defined as
follows below and are illustrated graphically in Fig. 12 in the
text. .
● Continuum Flow Regime: For Kn < 0.01, the mean free
path of the gas molecules is negligible compared to the
characteristic dimension of the flow geometry (i.e., the lchar
-parameter). In this case the continuum hypothesis of fluid
mechanics is applicable (i.e., the system is described by the
Navier-Stokes equations).
● "Slip-Flow" Regime: For 0.01 < Kn < 0.1, the mean free
path is no longer negligible, and the slippage phenomenon
appears in the "Knudsen" layer (layer of gas molecules
immediately adjacent to the wall)
● "Transition" Regime: For 0.1 < Kn < 10.
● Free Molecular Flow Regime: For Kn > 10, the flow is
dominated by diffusive effects.
The variation of the α-parameter as a function of Kn is repre-
sented using: (see ref. 13)
2
α ( Kn) = α 0 tan -1 ⎡c1Kn c2 ⎤ .......................................... (C-3)
π ⎢⎣ ⎥⎦
where α0, c1 and c2 are constants. The values for the constants
c1 and c2 are respectively 4.0 and 0.4, and the parameter α0 is
given by:
64 1 64
α0 = = ................................................... (C-4)
3π ⎡ 4 ⎤ 15π
1 −
⎢ b⎥
● The core flow experiments were performed at a constant
room temperature (T = 298 K).
● The gas used for the experiments was nitrogen (M =
28.01348 lb/lbmole).
● The nitrogen viscosity was computed using Sutherland's
equation (ref. 2):
13.85T 1.5
µ N 2 ,1 atm (T ) =
T + 102
µ N 2 (T , p ) = µ N 2 ,1 atm (T ) − 0.12474 + 0.123688 p
+ 1.05452 × 10 − 3 p 2 − 1.5052 × 10 − 6 p 3
.................................................................................................(C-6)
The characteristic length of the conduit (lchar) is required in
order to estimate the required Knudsen numbers. The
"typical" characteristic length is estimated from capillary pres-
sure data or an equivalent single capillary concept — and
these "lengths" are typically the capillary radii. We use this
exercise simply to evaluate one definition over another —
conceptually, there is no perfect definition because we are
investigating flow in porous media, not uniform capillaries.
The definitions we consider for lchar are:
● The computed average pore throat radius estimated from
capillary pressure data (these data are given in the various
reports from which the sample data were extracted).
● The theoretical "capillary radius" given by (see Appendix
B):
lchar = 8.886 × 10 −6 k∞ / φ ................................................... (C-7)
The minimum and maximum computed Knudsen numbers
(based on mean pressures) for each sample are presented in
Table C-1. Since most of the Knudsen numbers are greater
than 0.1, the "Transition" Regime (0.1 < Kn < 10) is the
dominant type of flow regime — therefore, the rarefaction
coefficient defined in the previous section plays a major role
for modeling the volumetric flow rate. In concept, this obser-
vation validates the application of the microflow model for
(steady-state) gas flow in porous media.
Table C.1 Minimum and maximum Knudsen numbers corres-
ponding to the models used to represent the
characteristic length lchar.
4

⎣ ⎦ Kn Estimated from the Kn Estimated from

Average Pore Throat Radius Eq. C-7
Validation of the Microflow Model Concept. In order to Sample ID min Kn max Kn min Kn max Kn
validate the applicability of the concept of the unified flow 1 0.15 0.34 0.64 1.51
2 0.15 0.36 0.32 0.77
model (Eq. C-1) for the problem of steady-state gas flow
3 0.43 0.97 0.40 0.91
through a porous medium, we computed the Knudsen number 4 0.06 0.15 0.72 1.67
for a given data set of 11 core samples tested at various mean 5 0.15 0.34 0.12 0.27
pressures. 6 0.37 0.72 0.41 0.80
The mean free path of gas molecules is defined by: 7 0.15 0.34 0.22 0.50
8 0.05 0.09 0.16 0.26
1 RT
λ ( p , T ) = 2 /π µ [ µ ≡ µ ( p, T )] ........................ (C-5) 9 0.15 0.33 0.28 0.63
p M 10 0.18 0.34 0.28 0.52
12 0.18 0.34 0.31 0.57
The mean free path ( λ ) is computed using the following (all
SPE 107954 15

A Rigorous Microflow Model for Gas Flow in an Idealized 3), which yields a direct relation for α(Kn) of the form of:
Porous Medium. The "microflow" model is defined in the
previous section as: α ( Kn) =
128
15π 2
[ ]
tan -1 4 Kn 0.4 ........................................ (C-14)

π 1 ∆p ⎡ 4 Kn ⎤
q = lchar 4 [1 + α ( Kn) Kn] ⎢1 + ................ (C-1) We now substitute Eq. C-14 into Eq. C-13 to yield our formal
8 µ L ⎣ 1 − bKn ⎥⎦ (or complete) result for this work:
Poiseuille's law for fluid flow in a pipe (or tube) is given by:
π 4 1 ∆p
⎡ ⎡ 128
k a = k ∞ ⎢1 + ⎢
2
[ ⎤ ⎤ ⎡
]
tan -1 4 Kn 0.4 ⎥ Kn ⎥ ⎢1 +
4 Kn ⎤
⎥
⎥⎦ ⎣ 1 + Kn ⎦
q= lchar ......................................................... (C-8) ⎣⎢ ⎣ 15π ⎦
8 µ L
.................................................................................... (C-15)
while Darcy's law for fluid flow in a porous media is:
1 ∆p Klinkenberg Model as a Simplified Microflow Model for
q = k Acore ............................................................ (C-9) Slip-Flow Regime. We can prove "mathematically" (using
µ L
rigorous assumptions) that the model developed by
In Eq. C-9, A is the cross-sectional area of the porous medium Klinkenberg is actually a simplification of the microflow
(perpendicular to the direction of the flow). Following the model for the case of the slip-flow regime (0.01 < Kn < 0.1).
procedure given by Klinkenberg (ref. 1), we can equate Recalling the Klinkenberg equation:
Poiseuille's and Darcy's laws to yield an expression for the
permeability (k). This procedure requires us to equate Eqs. C- ⎡ b ⎤
k a = k ∞ ⎢1 + K ⎥ ......................................................... (C-16)
8 and C-9, which yields: ⎣ p ⎦

π lchar 4 where:
k= ................................................................ (C-10)
8 Acore k∞ = Klinkenberg-corrected permeability, D
Substituting Eq. C-10 into Eq. C-1, we have: ka = Apparent gas permeability, D
k ∆p ⎡ 4 Kn ⎤ bK = Klinkenberg gas slippage factor, atm
q= Acore [1 + α ( Kn) Kn] ⎢1 + 1 − bKn ⎥ .................. (C-11) p = Mean core pressure, atm
µ L ⎣ ⎦
where: The Klinkenberg gas slippage factor (bK) is defined by
k = Permeability, D Klinkenberg as a coefficient of proportionality:
q = Volumetric flow rate in the conduit, cc/sec 4cλ bK
lchar = The characteristic length of the flow geometry (e.g., = ................................................................... (C-17)
r p
channel height, pipe radius), cm
L = Length of conduit, cm where c is a constant close to 1 (see ref. 6) and r is the
∆p = Pressure drop across the length of the conduit, atm effective pore throat radius (in our case r is the characteristic
µ = Gas viscosity at temperature and pressure, cp length lchar).
b = Dimensionless slip coefficient (b is defined as -1) Applying Eq. C-17 to Eq. C-16 yields:
α(Kn) = Dimensionless rarefaction coefficient ⎡ λ ⎤
Kn = Knudsen number, dimensionless k a = k ∞ ⎢1 + 4c ⎥ ........................................................ (C-18)
⎣ r ⎦
µ L
Multiplying through Eq. C-11 by and using b=-1, we Recall the base form of our microflow model (Eq. C-13):
A ∆p
⎡ 4 Kn ⎤
obtain: k a = k ∞ [1 + α ( Kn) Kn] ⎢1 + ............................... (C-13)
⎣ 1 + Kn ⎥⎦
µ L ⎡ 4 Kn ⎤
q = k∞ [1 + α ( Kn) Kn] ⎢1 + ...................... (C-12) We assume that we are in a slip-flow regime, then by defini-
A ∆p ⎣ 1 − bKn ⎥⎦
tion, α(Kn) = 0 (see ref. 13). Hence Eq. C-13 becomes:
where we note that the left-hand-side of Eq. C-12 is simply the
⎡ 4 Kn ⎤
"gas" permeability (ka) as defined by Darcy's law (i.e., the k a = k∞ ⎢1 + ⎥ .................................................... (C-19)
"uncorrected" permeability). Making this reduction, we have ⎣ 1 + Kn ⎦
our base form: Also, for 0.01 < Kn < 0.1, we have:
⎡ 4 Kn ⎤ 1
k a = k ∞ [1 + α ( Kn) Kn] ⎢1 + ⎥ ............................... (C-13) ≈ 1 − Kn + o( Kn) ................................................ (C-20)
⎣ 1 + Kn ⎦ 1 + Kn
Eq. C-13 provides an independent relation between the ap- Hence:
parent gas permeability (ka), the slip-corrected permeability 4 Kn
≈ 4 Kn − 4 Kn 2 + o( Kn 2 )
(or Klinkenberg-corrected permeability) (k∞), and the 1 + Kn ..................................... (C-21)
Knudsen number (Kn). We now need to finalize Eq. C-13 by ≈ 4 Kn
substitution of the relations for α(Kn). We substitute Eq. C-3 By definition of the Knudsen number (Eq. C-2):
into Eq. C-1 (and assume that c1 = 4.0 and c2 = 0.4 in Eq. C-
16 SPE 107594

Kn =
λ
lchar
..................................................................... (C-2)
⎡ ⎡ 128
k a − k ∞ ⎢1 + ⎢
2
[ ⎤ ⎤ ⎡
⎦ ⎥⎦ ⎣
]
tan -1 4 Kn 0.4 ⎥ Kn ⎥ ⎢1 +
1
4 Kn ⎤
+ Kn ⎥⎦
=0
⎣⎢ ⎣ 15π
Eq. C-19 now yields: ...................................................................................... (D-1)
k a ≈ k ∞ [1 + 4 Kn ] It is easily proven that for any pair of values for ka and k∞, Eq.
⎡ λ ⎤ .................................................... (C-22) D-1 has a unique positive solution: we refer to this solution as
≈ k ∞ ⎢1 + 4 ⎥ the "pseudo Knudsen" number.
⎣ lchar ⎦
Eq. C-22 is very similar (and almost identical) to Eq. C-18, Correlation of the Pseudo-Knudsen Number with Pres-
provided that the constant c in Eq. C-18 is equal to one and the sure, Permeability and Porosity. We have developed two
mean free path of the gas molecules used in the microflow correlations to relate the "pseudo-Knudsen" numbers (Knp)
model ( λ ) is defined as the average mean free path of the gas and the available data (ka, k∞, φ, p). The first correlation given
below relates the "pseudo-Knudsen" number (Knp) to the
molecules ( λ in the Klinkenberg model, evaluated at the
reciprocal mean pressure, the porosity (φ), and the gas
mean core pressure). We can consider that the Klinkenberg permeability (ka):
model is an approximation of the microflow model.
1 − 0.598 − 0.352
Kn p = 0.4 ka φ .......................................... (D-2)
Appendix D: Correlations for the "Pseudo-Knudsen" p
Number — Application to a Tight Gas Example Applying Eq. D-2 to Eq. D-1 yields:
(North Louisiana, USA) ⎡ ⎡ ⎤ ⎤
128 1 1
k a = k ∞ ⎢1 + tan −1 ⎢4[0.4 k a −0.598φ −0.352 ] 0.4 ⎥ 0.4 k a −0.598φ −0.352 ⎥ ×
2 p p
Determination of the Pseudo-Knudsen Numbers. This data ⎣⎢ 15π ⎣ ⎦ ⎦⎥
set (ref. 9) includes 18 core samples where the Klinkenberg ⎡ ⎡ 1 −0.598 −0.352 ⎤ ⎤
⎢ 4 ⎢ 0 .4 k a φ ⎥⎥
analysis (ref. 1) was performed in order to estimate the liquid ⎢1 + ⎣ p ⎦⎥
⎢ 1 −0.598 −0.352 ⎥
equivalent permeability (k∞). The measured mean pressure, ⎢ 1 + 0 .4 k a
p
φ ⎥
the permeability to gas, and the porosity are given for each ⎣⎢ ⎦⎥

sample — and using the mean pressure and permeability to ...................................................................................... (D-3)
gas data, a Klinkenberg plot is constructed, and the gas The equivalent liquid permeability can then be easily
slippage factor (bK) and the Klinkenberg-corrected computed by:
permeability (k∞) are estimated. ⎡ 128 ⎡ 1 ⎤ 1 ⎤
k ∞ = k a / ⎢1 + tan −1 ⎢4[0.4 k a −0.598φ −0.352 ] 0.4 ⎥ 0.4 k a −0.598φ −0.352 ⎥
As noted, the porosity of the sample is also given, but no ⎢⎣ 15π 2 ⎣ p ⎦ p ⎥⎦
reference conditions were given, so we assumed 300 K (80 ⎡ ⎡ 1 −0.598 −0.352 ⎤ ⎤
⎢ 4 ⎢0.4 k a φ ⎥⎥
Deg F) and 1 atm. The data summary is given in Table D.1. ⎢1 + ⎣ p ⎦⎥
⎢ 1 −0.598 −0.352 ⎥
Table D.1 Summary of the core data (given data and com- ⎢ 1 + 0.4 k a φ ⎥
⎢⎣ p ⎥⎦
4
puted results).
Sample k∞ bK φ ...................................................................................... (D-4)
ID (md) (psi) (fraction) For the second correlation we replaced the gas permeability
1-8 0.0001 565.3 0.023 with the equivalent liquid permeability:
2-10 0.1322 14.2 0.063
1 − 0.553 0.3897
2-22 0.0168 32.5 0.056 Kn p = 2.62 k∞ φ ........................................ (D-5)
3-8 0.0439 32.8 0.078 p
3-48 0.0035 83.6 0.07
1-1 0.00007 753.5 0.029 Substitution of Eq. D-5 into Eq. D-1 yields:
1-5 0.00004 801.0 0.02 ⎡ 128 ⎡ 1 ⎤ 1 ⎤
k a = k ∞ ⎢1 + tan −1 ⎢4[ 2.62 k ∞ −0.553φ 0.3897 ]0.4 ⎥ 2.62 k ∞ −0.553φ 0.3897 ⎥
1-7 0.00003 771.0 0.017 2 p p
⎣⎢ 15π ⎣ ⎦ ⎦⎥
2-2 0.0004 165.1 0.036
⎡ ⎡ 1 −0.553 0.3897 ⎤ ⎤
2-7 0.0124 26.8 0.06 ⎢ 4⎢2.62 k ∞ φ ⎥⎥
2-8 0.0025 77.1 0.051 ⎢1 + ⎣ p ⎦⎥
2-12 0.0025 215.1 0.055 ⎢ 1 − 0 . 553 0.3897 ⎥
⎢ 1 + 2.62 k ∞ φ ⎥
2-28 0.0052 58.6 0.037 p
⎣⎢ ⎦⎥
3-4 0.0002 303.1 0.04
3-6 0.0999 26.4 0.066 ...................................................................................... (D-6)
3-10 0.0067 84.8 0.071 Rearranging Eq. D-6 yields:
3-36 0.0168 39.2 0.057
3-55 0.0013 183.7 0.066

Since we cannot compute the Knudsen number explicitly, the

goal of this step is to evaluate a "pseudo Knudsen" number
(Knp) using the relation below: (see Appendix C and ref. 13)
SPE 107954 17

⎡ 128 ⎡ 1 ⎤ 1 ⎤
k a − k ∞ ⎢1 + tan −1 ⎢ 4[2.62 k ∞ −0.553φ 0.3897 ] 0.4 ⎥ 2.62 k ∞ −0.553φ 0.3897 ⎥
2 p p
⎣⎢ 15π ⎣ ⎦ ⎦⎥
⎡ ⎡ 1 −0.553 0.3897 ⎤ ⎤
⎢ 4⎢2.62 k ∞ φ ⎥⎥
⎢1 + ⎣ p ⎦⎥
⎢ =0
1 −0.553 0.3897 ⎥
⎢ 1 + 2.62 k ∞ φ ⎥
⎢⎣ p ⎥⎦

....................................................................................... (D-7)
Eq. D-7 is an implicit relation where k∞ is solved as a root of
this relation. Fig. D.1 presents the "pseudo-Knudsen" num-
bers obtained with Eqs. D-2 and D-5 plotted against the refer-
ence "pseudo-Knudsen" numbers (recall that we defined the
reference "pseudo-Knudsen" number as solution of Eq. D-1).

Fig. D.2 Computed equivalent liquid permeability versus

reference Klinkenberg-corrected permeability (extra-
polated from a Klinkenberg plot (ka versus reciprocal
mean pressure)) — both models present reasonably
good results.

Fig. D.1 Correlated "pseudo-Knudsen" number (Eq. D-2 and

Eq. D-5) versus reference "pseudo-Knudsen" number
(solution of Eq. D-1) — the implicit model (Eq. D-5)
achieves a better estimation of Knp than the explicit
model (Eq. D-2).

In Fig. D.2 we present the equivalent liquid permeabilities

computed using Eqs. D-4 and D-7 plotted against the
reference Klinkenberg-corrected permeability. Figs. D.3 and
D.4 show the behavior of the following function of k∞ for a
Fig. D.3 Plot of residuals computed with Eq. D-8 against
given data point (e.g., Sample 1-8, ka = 0.0012699 md, p = equivalent liquid permeability (k∞) for an example
45.27 psig, φ = 0.023). point from our data set — Sample 1-8, ka = 0.0012699
md, p = 45.27 psig, φ = 0.023. The Klinkenberg-
corrected permeability for this sample is k∞ =
0.000125 md. The root solution for this case yields k∞
= 0.00018 md, which implies an absolute relative error
of 30.56 percent (compared to the "Klinkenberg"
permeability).
18 SPE 107594

Fig. D.4 Plot of absolute value of residuals computed with Eq.

D-8 against equivalent liquid permeability (k∞) for the
same point as in Fig. D.3 — Sample 1-8, ka =
0.0012699 md, p = 45.27 psig, φ = 0.023. The Klin-
kenberg-corrected permeability for this sample is k∞ =
0.000125 md. The root solution for this case yields k∞
= 0.00018 md, which implies an absolute relative error
of 30.56 percent (compared to the "Klinkenberg"
permeability).