Journal of NUCLEAR SCIENCE and TECHNOLOGY, Vol. 44, No. 9, p.

1233–1247 (2007)

ARTICLE

Mathematica Programs for the Analysis of Three-Dimensional Pore Connectivity

and Anisotropic Tortuosity of Porous Rocks using X-ray
Computed Tomography Image Data
Yoshito NAKASHIMA1;
and Susumu KAMIYA1
1
Exploration Geophysics Research Group, National Institute of Advanced Industrial Science and Technology (AIST),
Central 7, Higashi 1-1-1, Tsukuba, Ibaraki 305-8567, Japan
(Received January 12, 2007 and accepted in revised form April 18, 2007)

Understanding of the transport properties of porous rocks is important for safe nuclear waste disposal
because harmful contaminated groundwater can migrate along pore spaces over long distances. We devel-
oped three original Mathematica
version 5.2 programs to calculate the transport properties (porosity,
pore connectivity, surface-to-volume ratio of the pore space, and anisotropic tortuosity of the pore struc-
ture) of porous rocks using three-dimensional (3-D) 8-bit TIFF or BMP X-ray computed tomography (CT)
images. The pre-processing program Itrimming.nb extracts a 3-D rectangular region of interest (ROI) from
the raw CT images. The program Clabel.nb performs cluster-labeling processing of the pore voxels in the
ROI to export volume, surface area, and the center of gravity of each pore cluster, which are essential for
the analysis of pore connectivity. The random walk program Rwalk.nb simulates diffusion of non-sorbing
species by performing discrete lattice walks on the largest (i.e., percolated) pore cluster in the ROI and
exports the mean-square displacement of the non-sorbing walkers, which is needed to estimate the geo-
metrical tortuosity and surface-to-volume ratio of the pore. We applied the programs to microfocus X-
ray CT images of a rhyolitic lava sample having an anisotropic pore structure. The programs are available
at http://www.jstage.jst.go.jp/browse/jnst/44/9/ and http://staff.aist.go.jp/nakashima.yoshito/progeng.htm
to facilitate the X-ray CT approach to groundwater hydrology.
KEYWORDS: anisotropy, diffusometry, diffusion tensor, MRI, NMR, permeability, percolation
cluster, pore size, porous media, self-diffusion coefficient, X-ray CT

flow rate strongly depends on the pore size.5) Because the

I. Introduction
pore structure is complex and three-dimensional (3-D), a
For safe nuclear waste disposal, understanding of the mi- two-dimensional (2-D) approach such as photomicroscopy
croscopic aspects of groundwater migration through natural of a thin section is inadequate and a system capable of meas-
porous rocks is essential because harmful contaminated uring the 3-D pore structure in porous geological samples is
groundwater can migrate along sub-millimeter pore spaces needed.
over long distances.1,2) The microscopic aspect of groundwa- Micro-focus or synchrotron X-ray Computed Tomography
ter transport in the geosphere depends on the pore structure. (CT) is a powerful tool to obtain the 3-D images of sub-mil-
Examples of the influence of the pore structure on macro- limeter pores nondestructively.6,7) A computer program can
scopic transport properties are shown in Fig. 1. The geomet- then be used to quantitatively analyze the transport proper-
rical tortuosity of the pore structure is important because the ties using the digital images.8–12) To the best knowledge of
diffusivity and permeability decrease with increasing tor- the authors, however, few such programs have been made
tuosity.3) A percolated pore cluster enables long distance mi- publicly available at little or no cost. Thus, in the present
gration of pore fluid molecules by diffusion and the Darcy study, we developed Mathematica
version 5.2 programs
flow, while isolated pores cannot contribute to long distance to calculate the transport properties (porosity, pore connec-
material transport.4) Thus, a pore connectivity analysis or a tivity, surface-to-volume ratio of the pore space, and aniso-
cluster-labeling analysis of the pores is needed to evaluate tropic tortuosity of the pore structure) of the porous rocks.
the transport properties of the rocks. The pore size or the re- Although the programs are not intended for the Darcy flow
ciprocal of the surface-to-volume ratio of the pore space is simulations, it is possible to estimate the macroscopic per-
also important for the groundwater flow because the Darcy meability (k) in Fig. 1 using the porosity, surface-to-volume
ratio and pore tortuosity values obtained by the programs.12)


We, then, applied the programs to a CT image set of a nat-
Corresponding author, E-mail: [email protected] ural rock sample (vesicular rhyolitic lava having an aniso-
Atomic Energy Society of Japan
1233
1234 Y. NAKASHIMA and S. KAMIYA

of porous media. However, these programs had the follow-

ing limitations: (1) The programs (DMAP.m, RW3D.m,
and Kai3D.m) read the 3-D CT images as text files and can-
not import binary files such as the Tagged Image File Format
(TIFF) and Bit MaP (BMP) files. (2) DMAP.m and RW3D.m
assume an isotropic pore structure and, thus, are not applica-
ble to the anisotropic porous media. (3) It is not possible to
apply RW3D.m to the CT images for which the pore size is
as large as the total system size. (4) Kai3D.m cannot export
the cluster-labeled 3-D pore images, which are essential for
performing computer simulations of a random walk on a per-
colated pore cluster. These limitations no longer exist in the
new programs.

II. Descriptions of the Mathematica
Programs

1. General
All programs developed in the present study are of the
notebook type and are for the Mathematica
version 5.2 or
later. It should be noted that, although there are some 2-D
illustrations below for simplicity and pedagogical purposes,
all the programs are for the 3-D image analysis. Thus, users
should prepare 3-D 8-bit (not 16-bit) CT images as a set of
contiguous 2-D slices. The dimensions of the voxel (a vol-
ume element) of each image should be cubic. If they are
not cubic, Clabel.nb cannot calculate the correct surface area
value of each pore cluster and Rwalk.nb cannot calculate the
correct value of the mean-square displacement of random
Fig. 1 The effects of pore geometry on the macroscopic permea- walkers. The programs, user manuals, and an example of
bility (k) and pore fluid diffusivity (D) in porous rocks. (a) Same 3-D CT images of a rhyolitic lava sample are available at
pore diameter but with different tortuosity. The straight or less http://www.jstage.jst.go.jp/browse/jnst/44/9/ and http://staff.
tortuous pipe yields higher permeability and diffusivity. (b) Per- aist.go.jp/nakashima.yoshito/progeng.htm. The programs
colated pore crossing the whole system compared with isolated are outlined briefly below and summarized in Table 1. For
pores. The former gives higher permeability and diffusivity. (c) further information, such as details about data preparation
Same porosity but with different pore diameter. The former
and program execution, readers should refer to the ‘‘readme’’
yields higher permeability.
text file available at the URLs above.

2. Itrimming.nb
tropic pore structure) to calculate the pore connectivity and The function of the Itrimming.nb program is to trim the
anisotropic tortuosity and to discuss the reliability of the pro- raw CT images and to export the trimmed rectangular im-
grams’ performance. We offer the programs on the Internet ages in TIFF, BMP, Comma-Separated Values (CSV), or
(http://www.jstage.jst.go.jp/browse/jnst/44/9/ and http://staff. Tab-Separated Values (TSV) format. This program should
aist.go.jp/nakashima.yoshito/progeng.htm) to facilitate the be run before using Clabel.nb and Rwalk.nb to extract a 3-
X-ray CT approach to groundwater hydrology. D rectangular region of interest (ROI) from a set of the
We previously made available on the Internet free pro- raw CT images. Both pore connectivity analysis (i.e., clus-
grams for calculation of the tortuosity (i.e., DMAP.m13) ter-labeling analysis) and random walk simulations will be
and RW3D.m14)) and pore connectivity (i.e., Kai3D.m15)) performed on the extracted 3-D rectangular image system.

Table 1 Outline of the three notebook-type Mathematica
programs

Program name Function Input Output

Itrimming.nb Image trimming of ROI Raw CT image Trimmed CT image
Voxel intensity histogram of the 3-D image
Clabel.nb Cluster labeling of pores Trimmed CT image Labeled CT image
Volume and surface area of each cluster
Rwalk.nb Random walk in a pore cluster Labeled CT image
Mean-square displacement of walkers
3-D trajectories of some walkers


Pre-processing of the labeled CT image by pre Rwalk image.nb or pre Rwalk csv.nb is required to convert into an internal binary format file.

JOURNAL OF NUCLEAR SCIENCE AND TECHNOLOGY

Mathematica Programs for the Analysis of Three-Dimensional Pore Connectivity and Anisotropic Tortuosity 1235

3-D image set of the labeled pore clusters.18) All pore voxels
in the porous media are colored cluster by cluster and are as-
signed to one of the pore clusters by this processing. This
cluster-labeling analysis is important for understanding the
contribution of pores to groundwater migration. Some pores
in the porous media are three-dimensionally connected to
form a single large percolation cluster responsible for the
macroscopic transport of materials; other pores are isolated
and do not contribute to macroscopic diffusion and the Dar-
cy flow. The Clabel.nb program allows us to characterize
such pore clusters.
Voxels are arranged like a simple cubic lattice in the 3-D
digital CT image set and the program scans the pore voxel
connectivity systematically voxel by voxel. When a pore
voxel is in full contact with another voxel, the two voxels
are judged to be connected. When the two pore voxels are
in contact only at a vertex or an edge, the clusters are con-
sidered to be disconnected. This cluster neighborhood rule
is commonly used in the connectivity analysis,18,19) and
shown in Fig. 3.
The fast algorithm of Hoshen and Kopelman (1976)18,20)
Fig. 2 Example of CT image trimming to extract a region of in-
was employed for Clabel.nb, and an example of the algo-
terest (ROI). This 2-D image of a cylindrical andesitic lava sam-
ple (effective porosity  22 vol.%) has an image dimension of
rithm for a 2-D case is shown in Fig. 4. This algorithm re-
5122 voxels = 7:82 mm2 . Pores and ambient air are dark in the quires only two scans of the whole image system. The first
CT image. The trimmed ROI inscribed within the cylindrical scan is carried out following the criterion of Fig. 3. The di-
rock image is indicated by the open square. rection of the first scan is shown in Fig. 4. This line scan
starts from the origin (the left top corner) and checks the
pore connectivity voxel by voxel along the arrow indicated.
An example of selecting an ROI is shown in Fig. 2. The If a pore voxel is not face-adjacent to any of the surrounding
sample is porous andesite lava obtained from Sumikawa, pore voxels, the voxel is labeled with a cluster color denot-
Akita, Japan.16) An X-ray CT image visualizes the spatial ing a new voxel intensity. If a pore voxel is face-adjacent to
distribution of the X-ray linear absorption coefficient a surrounding pore voxel, the voxel is labeled with the same
(LAC) within the sample.17) Thus, it is straightforward to cluster color as the adjacent pore voxel. The number of ad-
distinguish between solid areas with high LAC and air-filled jacent voxels to be checked during the line scan is two in the
pores with a low LAC. The ambient air and pores with the 2-D case of Fig. 4 (three in the 3-D case). Unfortunately, the
low LAC are shown by the dark voxels while groundmass cluster color of the two (or three) adjacent voxels is not al-
and phenocrysts appear as the light areas (in particular, ways common. As a result, the first scan occasionally yields
iron-rich high-density phenocrysts such as pyroxene are very mislabeling in which two or more colors are labeled to a sin-
light). Because ambient air is not needed to estimate the gle cluster. An example of this mislabeling is shown for a U-
transport properties of the andesite sample, the ambient air shaped cluster in Fig. 4. During the first scan, this mislabel-
voxels located at the four corners of the image system should ing is recorded in a temporary errata file stating that the
be eliminated. As a result, an ROI inscribed within the cylin- cluster colors ‘‘1’’ and ‘‘2’’ should be identical. The second
drical sample was chosen. scan is performed to correct the mislabeling. The direction
In the program, users are requested to specify the coordi- of the second scan is the same as that of the first scan. By
nates of the upper left corner of the rectangular ROI and the referencing the errata file, the program changes the cluster
dimensions of the rectangle in a right-handed coordinate sys- colors and exports a labeled 3-D image set in which each
tem. The program extracts the rectangles from all slices (i.e., pore cluster is labeled with a single unique color.
2-D images) and saves them as TIFF, BMP, CSV, or TSV Clabel.nb exports a labeled 3-D image set as TIFF, BMP,
files. A histogram of the 8-bit voxel intensity of the trimmed CSV, or TSV files. In the files, each pore cluster is colored
3-D image dataset is calculated and saved as a CSV text file. according to a color table (the file name: color.txt) provided
This histogram is useful for specifying the threshold for dis- by the user. This labeled pore image set is essential to prob-
criminating between air-filled pores with a low LAC and ing the tortuosity by the long-distance random walk simula-
solid areas with a high LAC, which is required to run the tion of pore fluid molecules along a percolated pore cluster
Clabel.nb program. using Rwalk.nb. The program also exports a record of the
volume, surface area, and 3-D coordinates of the center of
3. Clabel.nb gravity for each pore cluster as a text file. The surface-to-
Clabel.nb is a cluster-labeling program. Pore clusters are volume ratio of each pore cluster is obtained by dividing
connected pore voxels and cluster labeling refers to the ex- the cluster surface area by the cluster volume. This ratio is
amination of the 3-D pore connectivity in order to export a an important transport property because its reciprocal is

VOL. 44, NO. 9, SEPTEMBER 2007

1236 Y. NAKASHIMA and S. KAMIYA

Fig. 3 Criterion of pore voxel connectivity in a 3-D simple cubic Fig. 4 Two-dimensional example of the cluster-labeling algo-
image system. Adjacent voxels are judged as connected when rithm by Hoshen and Kopelman (1976).20) The 1-bit gray-levels
they share a face. Voxels are not connected when they make con- of the pore voxels and solid voxels are 1 and 0, respectively, in
tact by an edge or vertex. The connected pore voxels form a sin- the CT image before labeling. The first scan starts from the origin
gle pore cluster, and pore fluid molecules can migrate within the (left top corner) and proceeds along the first row of the 2-D ma-
pore cluster by Darcy flow and/or diffusion. trix as indicated by an arrow. The line scan along the second row
follows in the same direction of the arrow. After the completion
of the 2-D matrix scan of a single CT slice image, the program
nearly equal to the pore diameter.5) The output about the continues the scan of the adjacent 2-D slice to accomplish the
scan of the whole image system. After the first scan, the U-shap-
center of gravity is useful for analyzing the 3-D positions
ed gray pore cluster contains two different cluster colors (i.e.,
of small pore clusters.
voxel intensities), namely 1 and 2. This contradiction is corrected
There are two possible methods for calculating the surface using the errata file after the second scan. The color of each pore
area of a pore cluster; which one of these is used depends on cluster after the second scan obeys the color table provided by
the judgment of whether or not the rim of the image system the user.
is a real pore–solid interface (Fig. 5). If one considers that
the blue rim of Fig. 5 is not real and disappears when a larg-
er ROI is selected, then, the surface area (perimeter in the 2- known. It was confirmed that Clabel.nb outputted the correct
D case) of the orange cluster is counted using only the green volume and surface data, demonstrating good performance
interface. On the other hand, if one considers the blue rim to reliability. The cluster labeling of pore images of a pack of
be real, the total surface area of the orange pore cluster is the glass beads15) and natural sand grains26) has been carried
sum of the green and blue lines. Clabel.nb exports both sur- out using Kai3D.m. Clabel.nb was applied to these images
face area values. to confirm that the Clabel.nb output results were identical
Clabel.nb was applied to a synthetic 3-D test image of to those by Kai3D.m, supporting again the reliability of
which the volume and surface data of each pore cluster is the program.

JOURNAL OF NUCLEAR SCIENCE AND TECHNOLOGY

Mathematica Programs for the Analysis of Three-Dimensional Pore Connectivity and Anisotropic Tortuosity 1237

Fig. 5 Two possible definitions, A and B, of the surface area of Fig. 6 Example of a 2-D lattice-walk trajectory in a porous medi-
the pore cluster. In this example, the orange pore cluster reaches um over 200 time steps. The image system consists of 30  30
the rim of a 2-D image system of 7  7 voxels. The dimension of discrete voxels. The initial and final positions of the walker are
each voxel is unity. A: If the edge of the pore voxels at the rim marked by solid and open circles, respectively.
(blue) is not considered to be a real pore–solid interface, the total
surface area (total perimeter in the 2-D case) of the orange cluster
is the sum of the green lines, namely 22. B: If the blue rim is ac- where n is the number of the random walkers and xi ð
Þ, yi ð
Þ,
cepted as a real pore–solid interface, the total perimeter is the and zi ð
Þ are the 3-D coordinates of the walker’s position at
sum of the green and blue lines, namely 22 þ 10 ¼ 32. time
for the ith walker. The x-y plane is embedded within
the 2-D CT slice and z is the stacking direction of the slices
based on a right-handed coordinate system. The exact solu-
4. Rwalk.nb tion of the mean-square displacement for a lattice walk in
Rwalk.nb is a 3-D random walk program to simulate the a free space (i.e., porosity = 100 vol.%), hr 2 ifree , is given
diffusion of non-sorbing species (e.g., H2 O, Br , and I ) by15,18)
in the pores. The random walk should be non-sorbing be-
hr 2 ifree ¼ 6D0 t ¼ a2
ð2Þ
cause the purpose is to calculate the geometrical tortuosity
of the pore structure and the undesirable effects of the sorp- where t is the time, D0 is the diffusion coefficient of the
tion of walkers on the solid surface should be eliminated. walker in the free space without solids (e.g., H2 O self-diffu-
The random walk performed by Rwalk.nb is a discrete lattice sivity in bulk water), and a is the lattice constant of the sim-
walk in a simple cubic lattice18) (not an off-lattice walk). An ple cubic lattice (i.e., the dimension of a cubic CT voxel).
example of a random walk trajectory is shown in Fig. 6 for a For diffusion in rock pores, hr 2 i is reduced compared with
2-D case. The walker migrates on discrete voxels whose hr 2 ifree owing to the obstruction effects of solids. The degree
gray-levels correspond to the pore space. A pore voxel is of the reduction is measured quantitatively by the tortuosity
chosen randomly from among the whole image system as as follows. The mean-square displacement is important be-
the start position of the lattice walk trial at
¼ 0, where
cause the (scalar) diffusion coefficient, D, of the non-sorbing
is the dimensionless integer time. The walker executes a ran- species in the three-dimensionally isotropic porous media is
dom jump to one of the nearest pore voxels (the maximum related to the time-derivative of hr 2 i:
number of the nearest pore voxels is six for a 3-D simple cu-
1 dhr 2 i
bic lattice);
is incremented by a unit time after the jump so DðtÞ ¼ ð3Þ
that the time becomes
þ 1. If the randomly selected voxel 6 dt
is a solid voxel, the jump is not performed, but, the time still The tortuosity of the pore structure is a key transport prop-
becomes
þ 1. erty for the systems with small Péclet numbers and is defined
The main output of Rwalk.nb is the mean-square displace- as the limiting value of the ratio of D in the free space to D
ment, hr 2 i, of the walkers as a function of
(the file name: in the porous media:
Rwalk.csv). D0 a2
Tortuosity ¼ ¼ as t and
! 1 ð4Þ
1X n
DðtÞ dhrð
Þ2 i
hrð
Þ2 i ¼ ½ðxi ð
Þ  xi ð0ÞÞ2 þ ðyi ð
Þ  yi ð0ÞÞ2
n i¼1 d

þ ðzi ð
Þ  zi ð0ÞÞ2  ð1Þ Although the tortuosity is defined as the square root of
Eq. (4), namely, ðD0 =DÞ1=2 in some literatures,3,21) we obey

VOL. 44, NO. 9, SEPTEMBER 2007

1238 Y. NAKASHIMA and S. KAMIYA

the definition of Eq. (4), which is commonly used in the nu-

clear magnetic resonance diffusometry.22)
It should be noted that, while the diffusivity in the free
space is time-independent, it in the porous media depends
on the time or the diffusion distance because the diffusion
is restricted by obstacles (i.e., solids).23,24) For unrestricted
diffusion, for example, H2 O self-diffusion in bulk water,
hrðtÞ2 i, is linear with respect to t, and, thus, D is constant be-
cause of the homogeneity of the space. On the other hand,
because local heterogeneity (an finite pore size) exists, D
is time-dependent for a random walk in the porous media
(Fig. 7). Solid parts of the porous media are obstacles to
the diffusing material and, thus, the random walk trajectory
is restricted by the obstacles, which reduces the diffusion co-
efficient in the porous media compared to that in bulk fluid.
The degree of the diffusivity reduction is governed by the
average pore size and the characteristic diffusion distance
(root-mean-square displacement) of walkers. In the limit of
t ! 0, the root-mean-square displacement becomes smaller
than the pore size. The walkers rarely collide with solid
walls and the obstruction effects of solids are weak. As a re-
sult, the diffusion coefficient in the porous media normalized
to that in bulk fluid is slightly smaller than unity. This slight
decrease in the diffusion coefficient is proportional to the
surface-to-volume ratio of the porous media with negligible
solid surface relaxivity of the nuclear spin:23–25)
 
D 4 S pffiffiffiffiffiffiffi
¼ 1  pffiffiffi D0 t þ c1 t as t ! 0 ð5Þ
D0 9  V pore
where ðS=VÞpore is the surface-to-volume ratio of the pore
space and c1 is a constant. Nakashima and Yamaguchi
(2004) converted t into
using Eq. (2) and integrated
Eq. (5) to obtain a useful expression for the simulation data
fitting robustly against random noise:26)
 
hrð
Þ2 i 8a S pffiffiffi
2
¼ 1  p ffiffiffiffiffi
ffi
þ c2
as
! 0 ð6Þ
hrð
Þ ifree 27 6 V pore
where c2 is a constant. Equation (6) allows us to calculate
ðS=VÞpore by performing the random walk simulations in
the short-time limit if the 3-D pore structure is isotropic.
The random walk approach for the estimation of ðS=VÞpore Fig. 7 Schematics of a random walk of pore fluid molecules in a
mentioned above is time-consuming and less accurate owing fluid-saturated tortuous pipe of diameter d. (a) Example of a tra-
jectory of a random walker (solid circle) diffusing in the pore
to the stochastic nature of the random walk simulations com-
space. At t  d2 =ð6D0 Þ where D0 is the self-diffusion coefficient
pared with the deterministic cluster-labeling approach. How- of bulk fluid, the walker rarely collides with grains, so the diffu-
ever, this diffusometry-based method is, in principle, appli- sion is nearly equal to that in free space. The collision frequency
cable to the in situ nuclear magnetic resonance (NMR) well increases with time, reducing the mean-square displacement in
logging27) of the water-saturated porous strata of a nuclear the porous rock. The walker eventually negotiates many solid
waste disposal site. Magnets of a special design (i.e., one- walls and experiences the full geometrical tortuosity of the po-
sided magnets) are equipped on the NMR logging sonde25) rous media at t  d 2 =ð6D0 Þ. (b) Mean-square displacement of
to enable the measurement of self-diffusion coefficients of the random walk in (a). There is a transition from the unrestricted
the pore fluid molecules several centimeters inside the bore- diffusion regime to the restricted diffusion regime. The transition
hole wall. The obtained diffusion data are used to character- occurs at about t ¼ d 2 =ð6D0 Þ when the root-mean-square dis-
ize the pore structure and fluid species, which are difficult to placement is nearly equal to the pore diameter, d.
perform by other logging methods. Thus, diffusometry using
the CT images and Eq. (6) is useful for interpreting the phys-
ical background revealed by the NMR logging data. high tortuosity of the porous media and the slope of hr 2 i
As time elapses, a random walker in the porous rocks mi- reaches a constant value.22) The geometrical tortuosity of
grates further than the average pore size (Fig. 7). In this the porous rocks can, then, be calculated by using this slope
long-time limit, the random walkers fully experience the in Eq. (4). Hence, it is possible to calculate the tortuosity by

JOURNAL OF NUCLEAR SCIENCE AND TECHNOLOGY

Mathematica Programs for the Analysis of Three-Dimensional Pore Connectivity and Anisotropic Tortuosity 1239

(a) (b)

Fig. 8 Boundary conditions for the random walk simulation for a 2-D example. The green square denotes the original
image system, and the red square is a super system made by a mirror operation. (a) A straightforward periodic boundary
condition with parallel arrangement of the green square. Note that the undesirable discontinuity of the pore structure
occurs, through which pore fluid molecules cannot jump into an adjacent image system. (b) A modified periodic boun-
dary condition using a super system (red) containing mirror copies. The parallel arrangement of the red square yields a
continuous and percolated pore network through which random walkers can travel the long distance essential for the
correct estimation of the tortuosity value.

performing the long-time random walk simulations using the needed to correctly compute the tortuosity defined by
3-D CT image data. Eq. (4). However, as time elapses, the random walkers even-
Natural rocks often possess an anisotropic pore struc- tually go out of a 3-D CT image system of a finite size. This
ture.28–30) If the pore is anisotropic, D is a tensor31) (not a out-leaching is undesirable because the lattice walk (e.g.,
scalar) and Eqs. (3) to (6) break down. Directional mean- Fig. 6) cannot be carried out for the walkers outside the sys-
square displacement, hx2 i, hy2 i, and hz2 i, is needed to discuss tem. A periodic boundary condition is useful to avoid this
the tortuosity of anisotropic porous rocks: difficulty. It should be noted, however, that a simplistic pe-
1X n riodic boundary condition (Fig. 8a) is useless because the
hxð
Þ2 i ¼ ðxi ð
Þ  xi ð0ÞÞ2 ð7Þ pore connectivity breaks down at the boundary and the walk-
n i¼1
ers cannot migrate beyond the boundary. A modified boun-
1X n
dary condition using a mirror operation on the original 3-D
hyð
Þ2 i ¼ ðyi ð
Þ  yi ð0ÞÞ2 ð8Þ
n i¼1 image was employed in the present study to solve the discon-
tinuity problem (Fig. 8b).
1X n
hzð
Þ2 i ¼ ðzi ð
Þ  zi ð0ÞÞ2 ð9Þ Some programming techniques were implemented in
n i¼1 Rwalk.nb to conserve a memory and to reduce the CPU cal-
Rwalk.nb exports Eqs. (7) to (9) as well as Eq. (1) as a func- culation time. In the 3-D case, the mirror operation of
tion of
. Their exact solutions for the lattice walk in the free Fig. 8b requires a memory allocation eight times larger than
space (i.e., porosity = 100 vol.%), hx2 ifree , hy2 ifree , and the method of Fig. 8a. As this could sometimes exceed the
hz2 ifree , are given by: installed RAM limit of a user’s computer, we load only
the original image set (the green frame in Fig. 8b) into
1 2 1
hx2 ifree ¼ hy2 ifree ¼ hz2 ifree ¼ hr ifree ¼ a2
ð10Þ RAM and calculate the position in the red frame using the
3 3 mirror symmetry, reducing the memory use by 7/8. The cal-
The directional tortuosity can be calculated using Eqs. (7) to culation of the mean-square displacement at every time step
(10). For example, the tortuosity in the x-direction is the is one of the most time-consuming processes of Rwalk.nb.
time-derivative of Eq. (10) (i.e., a2 =3) divided by that of To save the CPU time, we implemented vectorization in
Eq. (7). the process of adding hr 2 ð
Þi data for the latest ith walker
Long-time data on the mean-square displacement are to hr 2 ð
Þi summed for the first to i  1th walkers. This vec-

VOL. 44, NO. 9, SEPTEMBER 2007

1240 Y. NAKASHIMA and S. KAMIYA

Fig. 9 Photomicrograph of a thin section of Niijima lava (open

nicol). Pores are filled with blue resin. The volcanic glass is
white. The brown mineral at left is biotite. The thin section
and cylindrical sample for CT were made from the same rock
sample.

Fig. 10 Example of CT image trimming for the cylindrical sam-

ple images of Niijima rhyolitic lava. Pores and ambient air are
torization drastically reduced the CPU time (e.g., from 34 dark, and phenocrysts and groundmass are light in the CT image.
hours to 15 hours for the Niijima lava sample case described The original image dimensions are 5122 voxels = 7:82 mm2 .
below). When running Rwalk.nb, users should import a la- The trimmed ROI indicated by the open square is 2562
beled pore-cluster image data set and perform the lattice voxels = 3:92 mm2 .
walk repeatedly to find the optimum simulation parameters
(e.g., number of walkers and number of time steps) by trial
and error. Importing a labeled 3-D CT image set as text, was 0.76 g/cm3 and the true density of the solid21) was
TIFF, or BMP files is another time-consuming step. To 2.39 g/cm3 .
save time, we made a pre-processor, pre Rwalk csv.nb (or A cylindrical sample of the lava (7.5 mm in diameter,
pre Rwalk image.nb), which converts the CSV text files 8.1 mm in length) was prepared and scanned by a cone-beam
(or the TIFF/BMP image files) into an internal binary format micro-focus X-ray CT scanner (Nittetsu Elex Co., Ltd.,
file (the file name: NT.mx). Rwalk.nb imports the NT.mx file Tokyo, Japan). The imaging conditions were as follows: ac-
(not the labeled raw CSV/TIFF/BMP files) as a 3-D perco- celeration voltage, 50 kV; tube current, 0.035 mA; number of
lated pore-cluster image data set. As NT.mx is optimized for projections, 1,800; time required for 360 projection, 39 min;
fast input/output by Mathematica
, we confirmed that im- field of view of the 2-D slice, 7:82 mm2 ; cubic voxel dimen-
porting the NT.mx file is about six times faster than import- sions, 15:23 mm3 (i.e., a ¼ 15:2 mm); reconstruction filter,
ing the labeled raw CSV/TIFF/BMP files. Shepp Logan; reconstructed 3-D image system, 512  512 
256 voxels.
First, the original 3-D image system consisting of
III. Application to Rhyolitic Lava Sample Images
512  512  256 voxels was trimmed using Itrimming.nb
The programs Itrimming.nb, Clabel.nb, and Rwalk.nb to extract a cubic ROI of 2563 voxels = 3:93 mm3 . The
were applied to a CT image set of a rhyolitic lava sample CPU time required was 9 min for the PC used. An example
to demonstrate their performance. A personal computer of a trimmed 2-D square region is shown in Fig. 10. Because
(PC) with an Intel Core2
Duo T7600 CPU (2.33 GHz) a lava sample is highly porous and mechanically weak, the
and 2 GB RAM running Windows XP
was used for the pore structure very near the surface of the cylindrical sample
demonstration. A CT image set of rhyolitic lava from the may be destroyed during cutting. To avoid including such a
Mukaiyama volcano,32) Niijima Island, Japan was used in destroyed pore structure within the ROI, a smaller square re-
the present study. This sample was chosen because (1) pores gion (not inscribed within the cylinder) compared to that
in rhyolitic lava are as large as several hundred micrometers shown in Fig. 2 was chosen. The trimmed 3-D TIFF images
in dimension (Fig. 9) and can be readily imaged by a con- are available at the previously mentioned URLs to enable
ventional micro-focus X-ray CT apparatus, and (2) rhyolitic readers to reproduce the cluster-labeling and tortuosity anal-
lava has a strong pore anisotropy21) suitable for examining ysis described below.
the diffusion anisotropy using the Rwalk.nb program. The Itrimming.nb exports a histogram of the 8-bit (i.e., 0 to
total porosity of the lava sample was measured by an con- 255) voxel intensity of the trimmed 2563 voxels (Fig. 11).
ventional laboratory method; it was 68 vol.%, calculated An analysis of the histogram is essential for the best choice
by 1 minus (bulk density of the porous rock)/(true density of the threshold for discriminating between solid and air-fill-
of the solid), where the bulk density of the porous rock ed pores. Figure 11 shows a bimodal distribution having a

JOURNAL OF NUCLEAR SCIENCE AND TECHNOLOGY

Mathematica Programs for the Analysis of Three-Dimensional Pore Connectivity and Anisotropic Tortuosity 1241

3
500000 10
pore solid largest pore cluster
2
400000 10

pore cluster surface area (mm2)

number of voxels

1
10
300000
upper limit
lower limit

midpoint (122)
0
10
200000
–1
10

100000
–2
10

0
–3
10
0 50 100 150 200 250 300 –6 –5 –4 –3 –2 –1 0 1 2
10 10 10 10 10 10 10 10 10
voxel intensity 3
pore cluster volume (mm )
Fig. 11 Histogram of the 8-bit (i.e., 0 to 255) voxel intensity of (a)
the 3-D ROI of 2563 voxels = 3:93 mm3 . The peaks of the pore
and solid voxels are 39 and 205, respectively, yielding a mid-
point (threshold) of 122. This threshold value yields a total po-
rosity of 69 vol.%.

peak (intensity 39) of air-filled pore voxels with a low LAC

and a peak (intensity 205) of solids with a high LAC. The
boundary between solids and pores is blurred owing to the
finite spatial resolution of the CT system. This is the undesir-
able partial-volume effect,42) which is responsible for the
voxels located between the two peaks, namely 39 and 205
in Fig. 11. It is reasonable to assume that the probability
of occupying a voxel at a solid-pore boundary is equal for
both solids and pores.15) This assumption leads to the choice
of the midpoint (i.e., intensity 122) of the peaks as the
threshold, implying that the voxels with an intensity equal
to or smaller than 122 are pores. Based on this threshold val-
ue, the number of pore voxels is 11,554,125 and, thus, the
total porosity of the ROI is 11;554;125=2563  69 vol.%.
The choice of the threshold value is critical to the results
of the pore-connectivity analysis and tortuosity calculation.
Thus, the validity of the choice should be cross-checked
by other methods. The total porosity of 69 vol.% is consis-
tent with that of 68 vol.%, measured by the conventional lab-
oratory method, implying that the choice of the midpoint as
the threshold is reasonable.
With a threshold of 122, Clabel.nb was applied to a cubic (b)
ROI of 2563 voxels = 3:93 mm3 , the output image file of
Itrimming.nb. The results are shown graphically in Fig. 12 Statistics of the labeled pore clusters. (a) Cross-plot of the
volume, V, and surface area, S, of each pore cluster; 923 pore
Figs. 12–14 and are summarized in Table 2. The CPU time
clusters were identified. The surface area was calculated accord-
required was 29 min for the PC used. Clabel.nb outputs the
ing to definition A of the caption of Fig. 5. The theoretical upper
volume and surface area of each pore cluster, and the output and lower limits (Eqs. (11) and (12)) of the data points are shown
is plotted in Fig. 12a. The volume of the largest or percolat- by dotted lines. (b) 3-D pore structures for the theoretical upper
ed pore cluster is 11,507,114 voxels = 40.4 mm3 . The sur- and lower limits plotted in (a). A lattice-like fine pore network
face areas of the largest cluster calculated according to def- contact at a single face to the adjacent pore voxel is an example
initions A and B in the caption of Fig. 5 are 683 and of the upper limit. An isolated blocky sphere yields the lower
743 mm2 , respectively. Thus, ðS=VÞpore is 683/40.4 limit of the surface-to-volume ratio for a specified pore volume.

VOL. 44, NO. 9, SEPTEMBER 2007

1242 Y. NAKASHIMA and S. KAMIYA

(a)
(a)

(b)

(b) Fig. 14 Results of 3-D pore-cluster labeling. A commercially

available 3-D viewer, T3D (ITT Visual Information Solutions,
Fig. 13 Cluster-labeling processing. (a) Example of a 2-D slice Colorado, USA), was used to visualize the 3-D image. The x-y
(8-bit gray scale) at z ¼ 200 before cluster labeling for the plane is embedded within the 2-D CT slice, and z is the stacking
ROI shown in Fig. 10; 2562 voxels = 3:92 mm2 . A 3-D Cartesi- direction of the slices based on the right-handed coordinate sys-
an coordinate system is indicated. (b) Labeled color image of (a) tem. (a) 2563 voxels = 3:93 mm3 ; coloring as in Fig. 13b and
after the 3-D cluster labeling for the cubic system of 2563 Table 2. (b) Shaded part (slab of 31  2562 voxels =
voxels = 3:93 mm3 . Yellow, the largest or percolated pore clus- 0:47  3:92 mm3 ) extracted from (a). Solid and isolated pore
ter; green, solid; purple, isolated pore clusters. voxels are made transparent in (b). Pores are elongated in the
x-direction.

mm1 = 1:69  104 m1 or 743/40.4 mm1 = 1:84 

104 m1 . The value of (the largest pore-cluster volume)/ thin section (e.g., Fig. 9) cannot perform the 3-D pore con-
(total pore-cluster volume) is as high as 11,507,114/ nectivity analysis mentioned above. This is the advantage
11,554,125 = 99.6%, implying that almost all of the pores of the X-ray CT method over the photomicroscopy in the
in the lava sample are connected to form a single percolated 3-D pore connectivity analysis.
pore cluster responsible for the long-distance material trans- Although isolated pores occupy only 100  99:6 ¼ 0:4%
port in the rock by diffusion and the Darcy flow. It should be of the pore space of the ROI, their number is as large as
noted that the conventional 2-D photomicroscopy of a single 922 (Table 2); this remarkable statistic is plotted in Fig. 12a.

JOURNAL OF NUCLEAR SCIENCE AND TECHNOLOGY

Mathematica Programs for the Analysis of Three-Dimensional Pore Connectivity and Anisotropic Tortuosity 1243

Table 2 Results of cluster-labeling analysis of pore voxels in the 3-D 2563 system of Fig. 14a

Color of voxels Number of pore clusters Number of voxels Total surface area
(mm2 )
Percolated pore Yellow 1 11,507,114 683
Isolated pores Purple 922 47,011 15
Solid Green — 5,223,091 —
Total 923 2563 698


The definition A of Fig. 5 was employed.

This log-log plot of the isolated pore cluster data follows a of Clabel.nb. The results (the volume and surface area of
slope (exponent) of 0.76. This is slightly larger than 2=3  each cluster) by Kai3D.m were completely identical to those
0:67, the exponent of completely similar 3-D objects. It is by Clabel.nb of Fig. 12a, demonstrating that Clabel.nb was
useful to note here the theoretical upper and lower limits programmed correctly.
for sufficiently large clusters in Fig. 12a. For large clusters Because of the large porosity and high pore connectivity
consisting of cubic voxels of dimension a, the upper limit (Fig. 14), the percolated pore occupies a significant portion
is given by of the ROI, implying a small tortuosity. Figure 14 also
shows a pore anisotropy that indicates that the pores are
4
S¼ V as V ! 1 ð11Þ somewhat oblate (compressed in the y-direction), suggesting
a a relatively large tortuosity in the y-direction. These points
where S and V are the surface area and volume of the pore were confirmed quantitatively by the random walk simula-
cluster, respectively. The 3-D shape of the pore cluster for tions described below.
the upper limit is not unique; an example is shown in Before running Rwalk.nb, pre-processing with
Fig. 12b. This is a fine pipe network characterized by the pre Rwalk csv.nb was performed to import the labeled CT
cross-sectional area of each pipe being as small as a2 . On images of Fig. 14a in a CSV format and to export an internal
the other hand, the 3-D shape for the lower limit is unique; binary format file, NT.mx. The CPU time required was 1 min
it is an isolated spherical pore: for the PC used. Then, the main program, Rwalk.nb, was run
 2=3 by importing NT.mx to output the mean-square displace-
3V
S ¼ 6 as V ! 1 ð12Þ ment (text files) and 3-D trajectories (on the PC display).
4 The number of the walkers was 10,000 and the maximum
It should be noted that, in a simple cubic lattice system, the time step was 400,000 for the tortuosity estimation. The
surface of a sphere is not smooth but blocky (Fig. 12b). This walkers should travel a sufficiently long distance to probe
rough surface yields an inevitable overestimation of the the tortuosity according to Eq. (4). This condition is satisfied
ðS=VÞpore value15) and the overestimation factor, 1.5, was if the walkers travel a distance larger than the characteristic
considered in Eq. (12). The data points for the lava sample pore size. Because the pore size is typically 30 voxels
fall within the theoretical upper (S / V 1 ) and lower (Fig. 13), the root-mean-square displacement should be larg-
(S / V 2=3 ) limits, suggesting that cluster labeling was per- er than 30 voxels. The maximum time step value, 400,000,
formed in a reliable manner. was chosen to allow the walkers to migrate a distance much
A detailed analysis of the Clabel.nb output revealed that larger than 30 voxels. The CPU time required was 15 hours
there are two types of isolated pore clusters in Figs. 13b, for the PC used. If the vectorization technique mentioned
14a, and Table 2. The first type is pore clusters completely above was not used, the time increased to 34 hours. An ex-
embedded within the 2563 image system. The number of ample of a long-distance random walk is shown in Fig. 15.
clusters of this type was 784 and the total volume was The 3-D system size indicated by the wire frame in the figure
24,347 voxels. Figure 13b depicts these pores, which are was expanded 3  3  3 times the original 2563 image sys-
small sub-spherical vesicles probably formed during degass- tem by the mirror operation described in Fig. 8b. It is evi-
ing from a cooling magma. The second type is pore clusters dent that the walkers leached out of the original system
that reached the surface of the 2563 image system. The num- and traveled a long distance much larger than the character-
ber of clusters of this type was 138 and the total volume was istic pore size of 30 voxels. Therefore, the simulation is
22,664 voxels. Figure 14a includes examples of the isolated completely under the restricted diffusion regime (Fig. 7)
pore clusters that are connected to the surface of the 2563 for which the tortuosity can be calculated using Eq. (4).
image system. This suggests that, if a larger ROI (e.g., The mean-square displacement of 10,000 walkers is
5123 voxels) was employed, some of the 138 clusters would shown in Fig. 16. Equations (2), (4), and (10) become sim-
possibly be judged to be connected to the percolated clus- ple if a is unity (dimensionless). Thus, a ¼ 1 was assumed in
ter.12) the figure, yielding a dimensionless mean-square displace-
Although the program Kai3D.m15) cannot export the la- ment. A random walk in the free space without solids was
beled 3-D images, it can output the volume and surface area also performed and plotted. The fitted slopes are 0.334,
of each pore cluster. Thus, Kai3D.m was applied to the same 0.334, 0.335, and 1.003 for hx2 i, hy2 i, hz2 i, and hr 2 i, respec-
3-D data set of the Niijima sample to check the performance tively. These agree well with theoretical predictions, namely,

VOL. 44, NO. 9, SEPTEMBER 2007

1244 Y. NAKASHIMA and S. KAMIYA

(a) 2
<x > percolated pore
2
<y > percolated pore
2
<z > percolated pore
2
<x > free space
2
150000 <y > free space
2
<z > free space

dimensionless mean-square displacement

100000

50000

0
0 100000 200000 300000 400000
τ
(a)

<r2> percolated pore

<r2> free space

400000
dimensionless mean-square displacement

300000

200000

(b)

100000

0
0 100000 200000 300000 400000
τ
(b)

Fig. 16 Dimensionless mean-square displacement of random

walks in the Niijima percolated pore cluster averaged over
10,000 walkers. Results for a random walk in free space (i.e., po-
Fig. 15 Example of a 3-D trajectory of a single random walk trial rosity = 100 vol.%) are also shown. (a) Dimensionless mean-
through the percolated pore space in the Niijima lava sample square displacement in the orthogonal directions, hx2 i, hy2 i,
with the boundary condition of Fig. 8b. The total time step is and hz2 i calculated by Eqs. (7) to (9). The quantities, hx2 i, hy2 i,
400,000. The initial (
¼ 0) and final (
¼ 400;000) positions and hz2 i in free space are indistinguishable, giving a common
of the walker are marked by open circles. (a) Projected trajectory. slope value of 1=3. (b) Dimensionless mean-square displace-
(b) Bird’s-eye-view trajectory. ment, hr 2 i, calculated by Eq. (1).

JOURNAL OF NUCLEAR SCIENCE AND TECHNOLOGY

Mathematica Programs for the Analysis of Three-Dimensional Pore Connectivity and Anisotropic Tortuosity 1245

3000 0.152, 0.114, 0.177, and 0.442 for hx2 i, hy2 i, hz2 i, and hr 2 i,
respectively. Therefore, the tortuosity averaged over all di-
rections is 1=0:442  2:3. This value is significantly smaller
dimensionless mean-square displacement

<x2>
2500 than that of the typical sedimentary rocks12,22,33) probably
<y2> due to the large porosity and high pore connectivity of
<z2> the lava sample. The directional tortuosity is 1=ð3 
2000 0:152Þ  2:2, 1=ð3  0:114Þ  2:9, 1=ð3  0:177Þ  1:9,
<r2> for the x-, y-, and z-directions, respectively. The results indi-
cate that the pore structure is most tortuous in the y-direc-
1500
tion. This is consistent with the oblate pore structure of
Fig. 14a. Similar diffusion anisotropy has been observed
1000 for rhyolitic lava by the conventional laboratory diffusion
experiments,21) suggesting that the Rwalk.nb simulation per-
formed reliably.
500 A Rwalk.nb simulation for a very short travel distance was
performed to estimate the surface-to-volume ratio of the per-
colated pore. The number of the walkers was 50,000 and the
0 maximum time step was 5,000. The root-mean-square dis-
0 1000 2000 3000 4000 5000 placement should be smaller than the characteristic pore size
τ of 30 voxels to calculate the surface-to-volume ratio (Fig. 7).
The maximum time step, 5,000, was chosen to allow the
(a)
walkers to travel a distance as short as 30 voxels. According
to the algorithm of Rwalk.nb, the start position of the 50,000
1 walkers was chosen randomly from among the percolated
pore clusters consisting of 11,507,114 voxels. Thus, it should
be noted that the calculated mean-square displacement and
derived ðS=VÞpore are quantities averaged over the whole im-
age system of 2563 voxels. This is essential for the quantita-
0.9
tive comparison of the ðS=VÞpore value by Eq. (6) and that by
Clabel.nb (Table 2 and Fig. 12a) which was obtained by the
<r2>/<r2>free

pore-connectivity scan of the whole image system. The

simulation data NT.mx file used for the long-distance random walk of
0.8 Figs. 15 and 16 was used again for this short-distance ran-
dom walk. The CPU time required was 55 min for the PC
used.
The obtained dimensionless mean-square displacement,
assuming a ¼ 1, is shown in Fig. 17a; a convex mean-
0.7 Equation (6) square displacement curve can be seen. This is evidence
for a transition from the unrestricted diffusion regime to
the restricted diffusion regime as described in Fig. 7. The di-
mensionless mean-square displacement is mostly less than
1,000 and, thus, the root-mean-square displacement is small-
0.6 er than 30 voxels, satisfying the short-distance random walk
0 200 400 600 800 1000 condition of Fig. 7. Figure 17a shows that hx2 i  hy2 i  hz2 i

τ
for
1;000, ensuring that Eq. (6), which was developed
for the isotropic porous media, is applicable to the anisotrop-
(b) ic lava sample if
1;000. Equation (6) was fitted to the
normalized hr 2 i data to obtain ðS=VÞpore ¼ 1:70  104 m1
Fig. 17 Very early time stage of a random walk in the Niijima (Fig. 17b). The modified periodic boundary condition with
percolated pore cluster. (a) Dimensionless mean-square displace- the mirror operation (Fig. 8b) adopted for Rwalk.nb suggests
ment averaged over 50,000 walkers. Note that hx2 i  hy2 i  hz2 i
that definition A in the caption of Fig. 5 should be used for
for
1000. (b) Mean-square displacement, hr2 i, of (a) normal-
ized by hr 2 ifree ¼ a2
(solid curve). A dotted curve, Eq. (6), was
the calculation of ðS=VÞpore . According to the results of
fitted to the data points for
1000 to obtain ðS=VÞpore ¼ Clabel.nb (Table 2 and Fig. 12a), ðS=VÞpore for definition
1:70  104 m1 . A is 1:69  104 m1 , nearly equal to 1:70  104 m1 . This
good agreement supports the reliability of (1) the perform-
ance of Clabel.nb and Rwalk.nb programs and also (2) the
1/3 and 1 (see Eqs. (2) and (10)), suggesting that the simu- methodology of the surface-to-volume ratio estimation by
lation performed reliably. The fitted slopes for the random diffusometry and cluster labeling.
walk in the percolated pore cluster in the lava sample are

VOL. 44, NO. 9, SEPTEMBER 2007

1246 Y. NAKASHIMA and S. KAMIYA

ity in complex lithologies using high-resolution X-ray comput-

IV. Conclusions ed tomography,’’ J. Hydrol., 302, 92 (2005).
We have developed three original Mathematica
pro- 8) P. Spanne, J. F. Thovert, C. J. Jacquin, et al., ‘‘Synchrotron
grams for the analysis of the 3-D pore connectivity and tor- computed microtomography of porous media: Topology and
transports,’’ Phys. Rev. Lett., 73, 2001 (1994).
tuosity of anisotropic porous rocks. These programs were
9) L. M. Schwartz, F. Auzerais, J. Dunsmuir, et al., ‘‘Transport
successfully applied to the conventional micro-focus X-ray
and diffusion in three-dimensional composite media,’’ Physica
CT images of a rhyolitic lava sample having the anisotropic A, 207, 28 (1994).
pore structure with 15.2 mm voxel dimension. The use of 10) N. S. Martys, H. D. Chen, ‘‘Simulation of multicomponent flu-
more advanced CT apparatus systems will allow a wider ids in complex three-dimensional geometries by the lattice
range of porous samples to be analyzed. For example, a su- Boltzmann method,’’ Phys. Rev. E, 53, 743 (1996).
per high-resolution synchrotron-based microtomographic 11) F. M. Auzerais, J. Dunsmuir, B. B. Ferreol, et al., ‘‘Transport
system with sub-micrometer voxel dimensions is being de- in sandstone: A study based on three dimensional microtomog-
veloped,37) with which it will be possible to probe the pore raphy,’’ Geophys. Res. Lett., 23, 705 (1996).
structure of fine-grained bentonite38–41) by X-ray CT. Our 12) Y. Nakashima, T. Nakano, K. Nakamura, et al., ‘‘Three-di-
programs run on the Mathematica
version 5.2 installed in mensional diffusion of non-sorbing species in porous sand-
stone: Computer simulation based on X-ray microtomography
the various operating systems (Windows, Macintosh, Unix,
using synchrotron radiation,’’ J. Contam. Hydrol., 74, 253
and Linux). Three-dimensional pore images obtained by nu-
(2004).
clear magnetic resonance imaging34,35) and neutron CT36) 13) Y. Nakashima, T. Yamaguchi, ‘‘DMAP.m: A Mathematica
(not X-ray CT images) are also acceptable. Thus, our pro- program for three-dimensional mapping of tortuosity and po-
grams will be useful for a microscopic approach using the rosity of porous media,’’ Bull. Geol. Survey Japan, 55, 93
3-D pore images for diverse studies on the transport of (2004). (Available at http://www.gsj.jp/Pub/Bull new/vol 55/
groundwater and contaminants through the natural and arti- 55 03/55 03 03.pdf).
ficial barriers at radioactive waste disposal sites. 14) Y. Watanabe, Y. Nakashima, ‘‘RW3D.m: Three-dimensional
random walk program for the calculation of the diffusivities
in porous media,’’ Comp. Geosci., 28, 583 (2002).
Acknowledgements 15) Y. Nakashima, Y. Watanabe, ‘‘Estimate of transport properties
of porous media by micro-focus X-ray computed tomography
The authors are grateful to Dr. T. Nakano for pre-process-
and random walk simulation,’’ Water Resour. Res., 38, article
ing the raw CT images used in this study and also wish to
number 1272, DOI number 10.1029/2001WR000937, (2002).
thank others who provided helpful comments during the 16) Y. Nakashima, T. Kikuchi, ‘‘Estimation of the apertures of wa-
preparation of this paper. The financial support was provided ter-saturated fractures by nuclear magnetic resonance well log-
by the Budget for Nuclear Research, Ministry of Education, ging,’’ Geophys. Prospect., 55, 235 (2007).
Culture, Sports, Science and Technology of Japan with 17) O. G. Duliu, ‘‘Computer axial tomography in geosciences: An
screening and counseling provided by the Atomic Energy overview,’’ Earth-Sci. Rev., 48, 265 (1999).
Commission. 18) D. Stauffer, A. Aharony, Introduction to Percolation Theory,
Revised 2nd Edition, Taylor & Francis, London, 192 (1994).
19) S. Ikeda, T. Nakano, Y. Nakashima, ‘‘Three-dimensional study
References
on the interconnection and shape of crystals in a graphic gran-
1) Japan Nuclear Cycle Development Institute, ‘‘Second Progress ite by X-ray CT and image analysis,’’ Miner. Mag., 64, 945
Report on Research and Development for the Geological Dis- (2000).
posal of HLW in Japan H12: Project to Establish the Scientific 20) J. Hoshen, R. Kopelman, ‘‘Percolation and cluster distribution.
and Technical Basis for HLW Disposal in Japan, Supporting I. Cluster multiple labeling technique and critical concentra-
Report 2 Repository Design and Engineering Technology,’’ tion algorithm,’’ Phys. Rev. B, 14, 3438 (1976).
JNC Technical Report, JNC TN1410 2000-003 (2000). 21) T. Yokoyama, S. Nakashima, ‘‘Diffusivity anisotropy in a
2) Japan Nuclear Cycle Development Institute, ‘‘Second Progress rhyolite and its relation to pore structure,’’ Eng. Geol., 80,
Report on Research and Development for the Geological Dis- 328 (2005).
posal of HLW in Japan H12: Project to Establish the Scientific 22) L. L. Latour, R. L. Kleinberg, P. P. Mitra, et al., ‘‘Pore-size
and Technical Basis for HLW Disposal in Japan, Supporting distributions and tortuosity in heterogeneous porous media,’’
Report 3 Safety Assessment of the Geological Disposal Sys- J. Magn. Reson., A 112, 83 (1995).
tem,’’ JNC Technical Report, JNC TN1410 2000-004 (2000). 23) P. P. Mitra, P. N. Sen, L. M. Schwartz, ‘‘Short-time behavior
3) J. H. Schön, Physical Properties of Rocks: Fundamentals and of the diffusion coefficient as a geometrical probe of porous
Principles of Petrophysics, Pergamon, Amsterdam, 583 media,’’ Phys. Rev. B, 47, 8565 (1993).
(2004). 24) L. L. Latour, P. P. Mitra, R. L. Kleinberg, et al., ‘‘Time-de-
4) D. L. Turcotte, Fractals and Chaos in Geology and Geophy- pendent diffusion coefficient of fluids in porous media as a
sics, Cambridge Univ. Press, Cambridge, 231 (1993). probe of surface-to-volume ratio,’’ J. Magn. Reson., A 101,
5) Y. Guéguen, V. Palciauskas, Introduction to the Physics of 342 (1993).
Rocks, Princeton University Press, Princeton, 294 (1994). 25) K.-J. Dunn, D. J. Bergman, G. A. Latorraca, Nuclear Magnetic
6) D. Wildenschild, J. W. Hopmans, C. M. P. Vaz, et al., ‘‘Using Resonance Petrophysical and Logging Applications, Perga-
X-ray computed tomography in hydrology: systems, resolu- mon, Amsterdam, 293 (2002).
tions, and limitations,’’ J. Hydrol., 267, 285 (2002). 26) Y. Nakashima, T. Yamaguchi, ‘‘Original free Mathematica
7) R. A. Ketcham, G. J. Iturrino, ‘‘Nondestructive high-resolution programs for the calculation of transport properties of porous
visualization and measurement of anisotropic effective poros- media,’’ In: X-ray CT for Geomaterials; Soils, Concrete,

JOURNAL OF NUCLEAR SCIENCE AND TECHNOLOGY

Mathematica Programs for the Analysis of Three-Dimensional Pore Connectivity and Anisotropic Tortuosity 1247

Rocks. (Edited by J. Otani and Y. Obara) A. A. Balkema, 35) M. Yamaguchi, K. Kobori, K. Suzuki, et al., ‘‘Imaging of
Lisse, 103 (2004). water distribution in thermally fractured granites by SPRITE,’’
27) R. Freedman, ‘‘Advances in NMR logging,’’ J. Petrol. Tech- Magn. Reson. Imaging, 23, 325 (2005).
nol., 58, 60 (2006). 36) B. Winkler, K. Knorr, A. Kahle, et al., ‘‘Neutron imaging and
28) S. Pfleiderer, H. C. Halls, ‘‘Magnetic susceptibility anisotropy neutron tomography as non-destructive tools to study bulk-
of rocks saturated with ferrofluid: A new method to study pore rock samples,’’ Euro. J. Mineral., 14, 349 (2002).
fabric,’’ Phys. Earth Planet. Inter., 65, 158 (1990). 37) H. Yasuda, I. Ohnaka, S. Fujimoto, et al., ‘‘Fabrication of
29) S. Siegesmund, H. Kern, A. Vollbrecht, ‘‘The effect of orient- aligned pores in aluminum by electrochemical dissolution of
ed microcracks on seismic velocities in an ultramylonite,’’ monotectic alloys solidified under a magnetic field,’’ Scr.
Tectonophys., 186, 241 (1991). Mater., 54, 527 (2006).
30) H. Dürrast, S. Siegesmund, ‘‘Correlation between rock fabrics 38) T. Kozaki, S. Suzuki, N. Kozai, et al., ‘‘Observation of micro-
and physical properties of carbonate reservoir rocks,’’ Int. J. structures of compacted bentonite by microfocus X-ray com-
Earth Sci., 88, 392 (1999). puterized tomography (micro-CT),’’ J. Nucl. Sci. Technol.,
31) S. Capuani, C. Rossi, M. Alesiani, et al., ‘‘Diffusion tensor 38, 697 (2001).
imaging to study anisotropy in a particular porous system: 39) Y. Nakashima, ‘‘Diffusivity measurement of heavy ions in
The trabecular bone network,’’ Solid State Nucl. Magn. Reson., Wyoming montmorillonite gels by X-ray computed tomogra-
28, 266 (2005). phy,’’ J. Contam. Hydrol., 61, 147 (2003).
32) N. Isshiki, Geology of the Nii Jima district. Geological sheet 40) S. Suzuki, H. Sato, T. Ishidera, et al., ‘‘Study on anisotropy of
map scale 1:50 000, Geological Survey of Japan [in Japanese effective diffusion coefficient and activation energy for deuter-
with English abstract] (1987). ated water in compacted sodium bentonite,’’ J. Contam. Hy-
33) S. Nakashima, ‘‘Diffusivity of ions in pore water as a quanti- drol., 68, 23 (2004).
tative basis for rock deformation rate estimates,’’ Tectono- 41) Y. Nakashima, ‘‘H2 O self-diffusion coefficient of water-rich
phys., 245, 185 (1995). MX-80 bentonite gels,’’ Clay Miner., 41, 659 (2006).
34) D. A. Doughty, L. Tomutsa, ‘‘Multinuclear NMR microscopy 42) R. A. Ketcham, W. D. Carlson, ‘‘Acquisition, optimization and
of two-phase fluid systems in porous rock,’’ Magn. Reson. interpretation of X-ray computed tomographic imagery: appli-
Imaging, 14, 869 (1996). cations to the geosciences,’’ Comp. Geosci., 27, 381 (2001).

VOL. 44, NO. 9, SEPTEMBER 2007