Professional Documents
Culture Documents
Fry 1979
Fry 1979
N. FRY
(Received February 13, 1978; revised version accepted October 24, 1978)
ABSTRACT
Fry, N., 1979. Random point distributions and strain measurement in rocks. Tectono-
physics, 60: 89-105.
INTRODUCTION
1 * ‘ *
. . * *. .
. . ‘.
.
I . .
.
. . .
. * . .
. . . .
.
0‘ * * *
l* * * l .
*
* . * * L
Fig. lc. This papers considers why, how and in what circumstances finite
strains can be elucidated in this way.
Comparisons are drawn between the method producing Fig. Id and meth-
ods dealing with nearest neighbours, such as that producing Fig. le.
The problems of dealing with randomly distributed objects have been
mentioned in passing by Ramsay (1976, p. 8), and a somewhat similar
approach to that employed here was taken by Stone et al. (1976). See also
Barbier and Leymarie (1972) and Leymarie (1968).
THEORETICAL CONSIDERATIONS
ther away, then the full pattern contains more and closer clusters than in
the Poisson case, and there is said to be “clustering”. If the probability that
a second individual item exists is lower at small distances from the position
of a known one than further away, then the full pattern contains fewer and
more open clusters, and also fewer intervening gaps, than in the Poisson case.
So, spacing-out or “anticlustering” is said to exist.
Reality and mathematical exactitude are more complicated than made out
in the above paragraphs, but it is necessary to introduce as a starting point
the idea of “clustering”, together with the idea that both clustered and anti-
clustered patterns, because t,hey have some constraint on positions of indi-
vidual objects, are more ordered than Poisson ones in which positions are
unconstrained.
Fundamental theory tells us that in the absence of ordering a thing can-
not record or retain information. So with distributions of objects in rocks
which are Poisson type, we are unable to make a distinction between config-
urations which have been strained and those which have not. Patterns with
a degree of ordering can carry information, and the more order the more
exact the information. It is tempting at this stage to try to make these ideas
more palatable by saying there is a possibility of measuring the strain suf-
suffered by clusters. However, anticlustering is common in rocks, can be
used equally well, and is in danger of being neglected on account of the
doubtful acceptability of a concept of straining anticlusters.
There is no general agreement on terminology amongst the various disci-
plines which concern themselves with these concepts. Sometimes use of
the term “random” is restricted and applied only to the Poisson case. The
term “aggregated” may be used in place of “clustered”, and distributions
which are “anticlustered” may also be referred to as “uniform” or “overdis-
persed”.
‘
. . . *
. .
. . . * - .
* :
. . . . . .
: .
: * :
. . . .
. . . .
. . . .
. . .
. .:. .
.
. * *
* * I . *
* *. * - *
. .
. . :
. . .
. .
. . -.
. . L .
. .
‘. .
. * . .
. .
. .
* .
** . “. . . . . . . ..I. I.
. . * W” . . . . ” ” ““” . .
. I) . . . **,:,*..: “..“. ”
. w..““”
. . . . “..“” . . . . . . “. .I**
. . . . . .UI “. . . . . .”
. . . . ” . . . . . .““. . . .
. . . ” VU”. . ..” .”
. . . . I). . . . . I. . ..“““..:’
.
.
l . ..“..““.” .W”..”
* ** . * “... l. . f a” *
. .
. . -
. ‘ * “.“.W ” .“.“zz
a*.
. * .* “I ”
*...*.**“.*
I) . . ..* * ” .I
***I)
* .
..” . .“. . . . ._ . . ”
. *. -. 0 . “. . “.” . I.“...._”
. .._“I . . . . ” . *. .
. . . . . . . . “.. . .” . . . ”
. .
_: *. . *. . ..I. . . “. ” .
* ‘.
.
. . ..” U.. .
. . .“” . . . . . * “.“Y
. * . ” . . . . . “. . *...a.. 1)
’
* 0
. * ’ .. .._” ..”
. *.
. ““I... “” * . z”” :*
.
. .
. .
. .. .*1.* “. * .” .
. . . . a. b
. ... .. . . . . ’ cd
. -
.. .. . .
..“.
“. . ” . .l
““a.
7 ”
.
1’2”
. . .”
:.
”
. .
0..
a . .“. .“......U” “...“.
S..“.“., . * “. . . “..
* .
.
* *
. .
“.
.
*z-z.* .
. “a.”
.“I “.
.
*.t
”
“.
.z
I”” .I. - *“I *a
. * .” .I.. *. . . ..I” . .
* *. . . . . ..“..“..” *” .
. *. .“” . . . “. ” .
.
.
..
. . l
. .
.
. , . . ‘” *” “V
.., .‘.‘z” ,-z l..
. ’ . ** * ‘* ’ . . “.“”
” .“.““““”
. . .“.. “.
l .
.
. . . l ., : - “.” . :.. *** I”%.*
” ” .” a... *.a.. .
0 .” .” “..” * ““0 *.
l * * * f ..I ‘. .” . . a..._**
.s. l ..” *.-*-Be.
c * ..* ” *. l . . . *..a ”
.
. . . ...” . ” . ““:““r
. *. :
. . . . .
. * “. . . . . . . .
. . - l . . .
. . .
. . . . .
. *
l e
.
:
. ”
.
. .
.
* *
.
: f
. .
l .
: .
. .
. . *. . . *. *
. .
. .. : . *.
. . . . .
. l * “: “.. :
. ” “.“...
l..11”,“.
*
*
.
: . .
* . . ** .
. l . *
.
‘ * ‘ * l *
. . . *
. . . -
. .
” . .
. . . -
. : . . ‘2.
derived plot, and the resulting positions of all others are added to the derived
plot. This procedure is further repeated until all points of the original plot
have been used as the new centre. The accumulated pattern which results
shows the frequency with which points in the original pattern have neigh-
bouring points at different distances in different directions. Though the
method is fundamentally simple, there are modifications to avoid problems
such as edge effects while maximising the number of points used, as men-
tioned in the section on “constraints”.
The patterns of Figs. lc and 2c are those of Figs. la and 2a subjected to
a homogeneous strain (in this case a sinistral simple shear in a vertical direc-
tion with an “engineering” shear strain value, y, of 1, and any points moving
1
1 1 1
ii’ ;1
1 1 1 1 1 l1 2 l1 1
11 1
: 11
112 2fl 1 1i 1 1 1 2y1 1 1! l1 ll 1
1 1 1 11 1 2
1 1 111
1 1 1 1 1
; 1 1 111 )ll 1
15 1 11 11 1 A1
11 1 li : 1! 113 1
1 1 1v 11 1 12 1 221 1 1 1
1 1I 11 1 ;I 1 1
1 i ll jll 1 21r
l1 I
ll ! 1 2 11 1
i 111 1 1 1 1 11, I
1
1 11 111 l1
111 11 11 lf l1 1
I L
1 1
I1 111
2 1 11 li
1 i 1
I
1
1
1 i
11
1
113
11
1
1
12 1
11
I1
1 1 1 2 1
1 1
1 2 1 1
111
1 1,ll 1
If1 ll; 2 1 1
1 p1 111 l1 1
1
2 I1
11
i 1 q 1 11
1 1 15 11 1
11 5 1 1
11 8 1 ,ll 11 11 P . 1. 42
1 1
11 1 1
1
1
12 1 l1
11
1
t 15 2 i 11 11
121 1
1 1’ 1 3 1
2 2
2 1 11 11
1 A1
1 1% --1 1 1
1
1 5 l1 211 ll; 12 1 11 1 1 L II1 1
li 2 2 1’1
1 I’ll1
ll. 1 1 !’ i_ 1 1 l1 :
2
11 l1 i I1
Fig. 3. Object-bject separations in a strained, generated, extremely anticlustered dis-
tribution.
95
over the top edge reappearing at the bottom, as if on the surface of a hori-
zontal cylinder). Figures Id and 2d are derived from Figs. lc and 2c by the
same method as used to obtain Figs. lb and 2b from Figs. la and 2a. Fig-
ure 3 is generated similarly to Fig. Id, but more detail is displayed by enlarg-
ing the plot and dividing it into a rectangular array of small boxes. The num-
ber of points falling into each box is recorded and then printed at the cen-
tral point of the box, empty boxes remaining blank. This is the method
adopted for real distributions of objects on rock surfaces (Figs. 4 and 5).
It can be seen that the strain, though not visible clearly in Fig. lc, is dis-
played by Figs. Id and 3. The shape of the girdle of highest or lowest or, as
in this case, sharpest change in neighbour frequency is that of the strain
ellipse. The practicability and accuracy of measuring strain this way is depen-
dent on the sharpness of definition of the girdle, which in turn is dependent
on two factors, the degree of ordering (clustering or anticlustering) and the
number of points in the sample. For the Poisson case (Fig. 2) there’is no
ordering, and so the properties of the strained and unstrained distribution
are indistinguishable.
This method gives results for Figs. la and 3, because the original generated
patterns were homogeneous and isotropic. Homogeneity can be tested for in
a strained distribution. Original isotropy cannot. However, it seems intui-
tively likely that an original distribution lacking isotropy would possess a
non-elliptical pattern of neighbour distances. On finding a real case of an
elliptical girdle, the reasonable question to ask is at what stage there could
have been an isotropic distribution. If that question can reasonably be an-
swered, then the ellipticity of the girdle probably represents the strain ratio
of the finite strain accumulated since that stage.
EXAMPLES
Figures 4 and 5 are the results of applying the method described to real
distributions of objects visible on rock surfaces. Both show anticlustering.
In Fig. 4 the points used are the centres of 382 phenocrysts in a porphyritic
lava. The distribution illustrates the degree of definition commonly obtain-
able with this number of points. It would be interesting to know whether
with a larger sample the visible anisotropy would remain or diminish in this
plot. Were this rock to be subjected to a homogeneous deformation it would
be possible to make an approximate strain estimate from the strained equiv-
alent of the pattern in Fig. 4.
Figure 5 is the result of using this method on a deformed oolitic lime-
stone. The points used are the original centres of only 126 ooids. Deforma-
tion could be seen to be by pressure solution at contacts of originally packed
ooids of almost constant size. In this case, there was originally a very sharp
cut-off preventing ooids centres from coming closer than one ooid diameter,
and that is why far fewer objects than normal are needed to define the anti-
clustering strain ellipse. Although given here as an example of this method,
96
the strain ellipse of this rock could easily be determined by established pro-
cedures, as discussed later.
2 1 4 1 1 1 1 122 2 2 1 2 1 1
2 3 2 1 111312 1 2 1 1 13 1 21
2 3 311 33 1 21211111333
1 2 2 1 2 1 321 2 1 4 1 2 1 3
1 1 1 1 1 3 2 3 1 2 12312212
4 3 31111 2 2 1 2 2 2
1242332 2 11111 1 1 1
1 2243122232212 12221221
3 1 1 3 1 1 3 2 3 2 1 2 3 2 3 1 3
1 1 2 2 2 1 1 3211 11 4
1 2 2 3 1 111 4 2 1 1
1 22 2 4112 21
13121 21 1 2 1 2 1 3 1
1 2 2 1 1 4 2 2 2 1
1 1 2 4 1 1 1 1 3 2 2 1
4 1 1 1 1 2 3 1 1 2221 1
3 1 3 2 3 2 1 2 3 2 3 1 1 311 3
12212221 2122322213422 1
1 1 1 11111 2 2332421
2 2 2 1 2 2 11113 3 4
21221321 2 1 3 2 3 1 1 1 1 1
3 12141 2 1 2 3 121 2 2 1
33311111212 1 33 113 3 2
12 1 31 1 1 2 1 213111 1 2 3 2
1 1 2 1 2 2 2 2 1 1 1 1 1 4 1 2
Our consideration of the Poisson process starts with two definitions. The
mean rate at which the number of objects inside a boundary is increased by
increasing the area within that boundary is designated h. This may also be
thought of as the overall density of objects. The probability that an area con-
tains no objects whatsoever is a function of the area’s extent a, and is here
designated f(a) (or for another area of extent b, f(b), etc.).
In the Poisson process, h and f(a) do not vary either with individual char-
acteristics of areas other than their extent, or with position in the field.
Therefore their values at any point cannot be changed by any amount of
deformation of the field. So what follows applies to both strained and un-
strained cases.
From the above, the following relationship applies:
f(c + 6) = f(c) * f(b)
Functions behaving in this way are exponential.
The rate at which increasing an area will decrease that area’s probability
of being object-free - i.e. -(df/da); is given by the product of the probabil-
ity that the area is object-free in the first place (f(a)) and the rate at which
98
Edge effects
Certain constraints become apparent during the process of using each ob-
ject position on a limited field as origin of a derived plot. Edge effects have
to be avoided. This requires a decision not to consider possible interaction
between objects which are more than a certain distance apart, i.e., to put a
limit on the size of the derived plot. Once this distance is chosen, it is possi-
ble to use as origins for the derived plot only objects lying further than this
distance from the edge of the original field. This prevents the apparent den-
sity of neighbours falling away with distance simply from failure to find
objects which lie outside the field of the sample.
If the exercise is undertaken manually, it is possible to use a circular
derived plot, with its radius the chosen distance limit, and to use original
sample fields with irregular boundaries. However, there are advantages in
digitising the rectangular coordinates of each object, and calculating the
coordinates on the derived plot as coordinate-differences on the original.
In this case it is easiest to put a limit on the coordinate-differences in each
direction, i.e., create a rectangular derived plot. In this case, avoiding edge
effects becomes far simpler if rectangular boundaries are imposed on the
original sample field.
With a rectangular field it is easy, without edge effects, to use more ob-
jects as origin of the derived plot than suggested above. For any pair of
objects, each lies, as a neighbour of the other, the same distance away,
but in exactly the opposite direction. Therefore the derived plot is an
attempts to define a pattern which is necessarily centro-symmetric. So, each
101
such pair need only be recorded once, by limiting the derived plot to neigh-
bours which lie within a 180” sector (say, to the right side) of each new
origin. This allows origins to be used which lie right up to one edge (left in
this case) of the original field. If it is desired to view a full 360” derived
plot, instead of the half that is now produced, all that is necessary is to oper-
ate the centre of symmetry.
It is not possible to choose the optimum limits of the derived plot in the
first instance. For isotropic distributions, the Poisson nearest neighbour peak
at dm gives an idea of the required size. A suitable limit would be just
over twice this distance, say, for convenience, a However, this limit
itself needs to be extended in one direction and may be contracted in the
other in accordance with the strain ellipse it is desired to identify. The
best procedure is therefore to create a series of derived plots, the first of
which is large. It can then be seen to what extent dimensions may be
reduced, so as to increase the number of usable origins and therefore the
density of points which gives the new plot “definition”. This repetition
makes computer handling of digitised coordinates worthwhile. It should
also be clear at this point that fitting the dimensions of the new plot to an
identified ellipse is much easier if the coordinate axes parallel the ellipse
axes, as well as lying parallel to the boundary of the sample field. Therefore
care should be taken in original sampling to use lithological indicators such
as cleavage trace, shape fabric, mineral lineations, etc., to obtain an orig-
inal sample field with rectangular boundaries paralleling likely principal
strain directions.
Sample size
objects in rocks. With less than 300 objects it is unlikely that quantitative
results can be obtained. With over 1000 objects, definition may be good
enough to simply measure up dimensions on a mildly ordered concentric
ellipse pattern to an accuracy competitive with other methods of strain
measurement. This requirement of a planar, homogeneous rock surface,
bearing such large numbers of objects, is a very severe practical limitation
on this method of strain measurement.
Visual interpretation
When looking for concentric elliptical girdles on the new plot, we are
looking for a shape which relies for definition on density variations which
arise from contributions from all parts of an original sample field. It is possi-
ble that strain will vary a little across the field. If strain is low, the slight
blurring of definition of the pattern will not be detectable above the blur of
statistical variations in densiby. If the strain ellipse semiaxis ratio is 6 : 1 or
more, a small number of sampled object--object separations are relied on to
define ellipse ends, Small variations in strain, particularly in the direction of
principal extension, can cause ellipse girdle ends to become indefinable. This
problem increases with strain, and although avoidable by substituting a more
homogeneous sample, cannot be compensated for by sample size. There are
mathematical methods of overcoming this difficulty and evaluating strain.
However, while we rely upon visual appreciation, an elliptical girdle pattern
must be identifiable. This requires either than the strain be represented by
an ellipse with less than 6 : 1 semiaxis ratio, or that the strain be remark-
ably homogeneous.
There is a general problem of visual interpretation. It is the inability of the
human eye to identify concentric elliptical girdles of even quite startling dif-
ferences in object densities (e.g., 2 : 1) when girdle widths are small com-
pared to the average spacing of objects at the densities concerned. Again, this
difficulty is susceptible to mathematical treatment, but it can also be allevi-
ated in visual interpretations by contouring the density on the new plot.
Counting statistics are too bad for neatly concentric contours to be the
result of this process. Nevertheless it can assist the human eye in recognising
girdles of differing density.
To those worried about statistical significance, all that can be said at this
stage is that the human eye is so inept at seeing artifically created patterns
of high significance, that indentification of insignificant girdles seems highly
improbable.
Emphasis here has been on visual interpretation. If it could be assumed
that all object distributions in rocks started as perfectly isotropic clustering
or anticlustering, and that all strains were perfectly homogeneous, then we
would be justified in using mathematical methods which assume elliptical
girdles. However, we know that there will be many types of object distribu-
tion which were not originally isotropic and which did not deform homo-
gcneously. Until some understanding is accumulated of the nature of differ-
ent distribution types in different rocks, visual interpretation of results
remains necessary for judging whether or not each sample warrants use as
a strain measure.
This method belongs clearly in the realm of those types of strain deter-
mination which in the first instance define relative dimensions of a two-
dimensional strain ellipse by accumulating large numbers of individual
measurements (such as the methods given by Ramsay, 1967, pp. 202-221,
for dealing with deformation of originally non-spherical objects). Their
general applicability is either where individual strain markers, e.g., reduc-
tion spots, do not exist, or to provide independent corroboration of their
results.
Amongst such methods, the one described here falls into the same cate-
gory as the established centre-to-centre method (method 3, pp. 195-197,
of Ramsay 1967, and variations), in that it deals only with object position,
and so is not dependent directly on assumptions or evidence of relative com-
petence of the objects and their surroundings. The centre-to-centre method
assumes that each object used has a group of neighbours which are identifi-
able as having been its closest neighbours before deformation, and that the
original pattern was isotropic. In the terminology of this paper, these neigh-
bours would, if all plotted out from a common origin, have defined a circu-
lar girdle. Strain of the rock results in this girdle taking on the relative
dimensions of the strain ellipse. However, new neighbours, not originally
in the closest set, may now lie in the compression direction just as close as
the original ones now do in the extension direction. Identifying the partic-
ular girdle representing the strain ellipse relies upon the ability to omit new
close neighbours from centre-to-centre measurement while retaining those
previously close neighbours moved apart in the extension direction. If this
cannot be done unambiguously the resulting girdle will be unreliable, depen-
dent either on subjective criteria in the choice of which neighbours to use
(and so in resulting ellipticity), or on the objective criterion of present near-
est neighbours discussed earlier. Therefore the centre-to-centre method
should only be used where there is no ambiguity as to the originally nearest
neighbours. By comparison, the method of this paper is theoretically usable
whether or not original neighbours are identifiable. In practice it is found in
cases where both methods have been applied that the centre-tocentre
method gives greater accuracy from a small sample, but that the method of
this paper is nevertheless quicker in cases of large sample size. Where objects
are widely dispersed in a nearly homogeneously deforming matrix, a study
of object to object distances may show the strain experienced by the matrix.
If the objects remain rigid, this could give a finite strain value not dependent
on the strain path, as are orientation methods involving theories of rigid
104
body rotation. If the objects deform, this study may provide an alternative
strain measure to that of object shape. As such, it may be useful as a test of
homogeneity of deformation, and it may give the better measure of overall
strain. It may suggest two phases of deformation, in only the first of which
the objects remained rigid. In this case it provides a value of total finite
strain where object shapes give that of the later phase alone. It is even possi-
ble that metamorphic reaction has completely obliterated all previous fabric,
except for the grain positions of just one metamorphic mineral species which
has nucleated on the site of an earlier mineral. Study of the mineral positions
may give the finite strain since a stage earlier than any other object or com-
positional heterogeneity now in the rock. Because of the separation of ob-
jects, identification of earlier neighbours is not possible without ambiguity.
Therefore the method of this paper is the one which provides the possibility
of making such strain measurements.
If a rock consisting mainly of packed objects has deformed homogene-
ously, there should be no difference in strain as measured by object shape
and as measured by a centre-to-centre method. Because the criterion of
touching in this case identifies the original neighbours, the established cen-
tre-to-centre method gives good results from fewer objects than the method
described here. If deformation is by pressure solution at object boundaries,
the two methods should again give similar results, though in this case the
problem of original neighbours arises again.
In a rock composed of packed objects, which has deformed even partly
by sliding at object boundaries, a similar strain value will be arrived at by
studying object shape, centre-to-centre distance of touching neighbours, or
the method described here. Each method is essentially measuring the vari-
ation with direction of the distances from centres of objects to their edges,
whether this be as the edge-to-edge length through the centre of individual
objects, or the centre-to-edge-to-centre lengths between pairs of objects.
Such measurements do not take the boundary sliding component of strain
into account. Each method will underestimate strain by the amount of this
unknown component and therefore be equally unsatisfactory.
CONCLUSION
REFERENCES
Barbier, J. and Leymarie, P., 1972. Disposition reguliere de certaines mineralisations ura-
niferes. Bull. Bur. Rech. Geol. Min. (Fr.), Sect. 2, 1: 1 l-18.
Leymarie, P., 1968. Une methode permettant de mettre en evidence le caractere ordonne
de la distribution de certains gites mineraux. Miner. Deposita, 3: 334-343.
Matheron, G., 1975. Random Sets and Integral Geometry. Wiley, New York, N.Y.,
288 pp.
Meyer, P.L., 1970. Introductory Probability and Statistical Applications. Addison-
Wesley, Reading, Mass., 2nd ed., 367 pp.
Ramsay, J.G., 1967. Folding and Fracturing of Rocks. McGraw-Hill, New York, 568 pp.
Ramsay, J.G., 1976. Displacement and Strain. Philos. Trans. R. Sot. London, Ser. A,
283 : 3-25.
Stone, D., Dales, G.R. and Schwerdtner, W.M., 1976. Strain analysis by means of undis-
rupted inclusions with unknown primary shapes. Geol. Miner. Assoc. Can., Progr.
Abstr., 1: 52.