Joumalo,fStructural Geology, Vol. 12, No. 8, pp. 1047 to 1059, 1990 0191--8141/90 $03.00+0.

00
Printed in Great Britain © 1990 Pergamon Press plc

Least-squares center-to-center and mean object ellipse fabric analysis

ERIC A. ERSLEV and HONGXINGGE

Department of Earth Resources, Colorado State University,Fort Collins, CO 80523, U.S.A.

(Received 13 June 1989;accepted in revised form 28 February 1990)

Abstract--The subjectivityof ellipse fittingin many strain techniques has hindered the determinationof fabric
anisotropyand tectonicstrain. However,manysets of x, y co-ordinatescan be approximatedas an ellipseusinga
least-squares algorithmto calculatea best-fitellipse and associatedaverage radial error. For instance, the two
dimensionalshape of many objects can be approximatedas an ellipse by entering digitizedco-ordinatesof the
object margin into the ellipse algorithm.
The rim of maximum point density in a normalized Fry diagram is defined by normalizedcenter-to-center
distances between touching or nearly touching objects. The enhanced normalized Fry (ENFry) method
automates ellipse fittingby entering center-to-center distances between these 'touching' objects into the least-
squares ellipse algorithm.For homogeneouslydeformed populationsof 200 objects, the ENFry methodgivesan
accurate and precise measure of whole-rock fabric anisotropy, particularlyfor low ellipticities. When matrix
strain exceedsclast strain, manualellipsefittingof normalizedFry plotsgivesmore accuratematrixanisotropies,
The mean object ellipse (MOE) method calculatesthe best-fitellipse from the geometryof the objects. Three
points fromthe marginof each object ellipse, centered at the origin and expandedor reducedto unit volume,are
used to calculatethe best-fitfabricellipse. The MOE method is very precise for smalldata sets, makingit a good
method for mappingheterogeneousobjectstrain. However,least-squarescalculationsmaximizethe influenceof
distaland spuriousellipticities,causingthe MOE methodto overestimatethe fabricellipticityof mostaggregates.

INTRODUCTION through an aggregate with perfect three-dimensional

anticlustering (e.g. packed, equi-volume spheres) con-
QUANTITATIVE analysis of rock fabric can provide im- tains variably sized cross-sections through both the mid-
portant constraints on many petrogenetic problems, die and ends of the objects (Bhattacharyya & Longiaru
Structural geologists have traditionally used fabric 1986, Erslev 1988). This effect, when combined with
anisotropy to quantify deformation histories of tecto- normal sorting, limits anticlustering and obscures the
niteR (e.g. Cloos 1947, Ramsay 1967, Schmid et al. fabric ellipse.
1987). Whilst these strain methods are not inherently In packed aggregates, plotting the normalized dis-
limited to the measurement of deformation, quantit- tances between object centers (actual distance divided
ative evaluation of subtle fabric anisotropy formed by by the sum of the average radii of the two objects) in the
mild deformation, compaction and depositional align- normalized Fry diagram (Fig. lb) (Erslev 1988) provides
ment is hindered by the coarse resolution and subjec- a clearer definition of the fabric ellipse by reducing the
tivity of many strain methods, number of points inside the rim of maximum point
In addition, the lack of automated ellipse calculation density. In packed objects, the rim of maximum point
and quantitativeerrordeterminationpreventstheestab- density is defined by touching (or nearly touching)
lishment of accuracy and precision standards. For in- objects irrespective of their two-dimensional size. Since
stance, evaluation of object strain by plotting object the normalized Fry method is not dependent on two-
ellipticity against long axis inclination (Rf/~ technique: dimensional anticlustering, it should be independent of
Ramsay 1967, Lisle 1985) typically involves determining sorting, which can further decrease the anticlustering of
the top and bottom of an onion-like shape encompassing packed aggregates.
most of the data in an Rf/q~ diagram. The selection of However, the normalized Fry method still requires
objects to include and objects to exclude can be very manual ellipse fitting, introducing a subjective, irrepro-
arbitrary, ducible step to the analysis. We have found that students
Measuring center-to-center strain from scatter plots in structural geology classes are often difficult to con-
of center-to-center distances is usually even more sub- vince that the 'answer' ellipse is significantly better than
jective. Determining the level of maximum point density theirs. In complex natural aggregates, variable initial
in the Ramsay (1967) technique, where inclinations of and tectonic ellipticities broaden the rim of maximum
center-to-center distances are plotted against the actual point density, making ellipse fitting even more difficult.
distances, is matched in difficulty with determining the The fabric methods proposed by this paper allow
rim of maximum point density in a Fry (1979) plot (Fig. analytical, unbiased determination of individual object
la). Both methods require an unrealistic degree of two- and overall fabric anisotropy. The analyses are imple-
dimensional anticlustering, which is inherently limited mented using a least-squares ellipse-fitting algorithm in
in packed aggregates. Even a two-dimensional plane specially modified versions of INSTRAIN 2.5, an inte-
Rn 12.,-~ 1047
1048 E . A . ERSLEV and H. GE

(a) (b)
Fry (1979) Method Normalized Fry Method
-~-..~:'~,
•~.... :. =': "-...;.-
c.:~.. ;-.:.,..:..~:..,;..s..~..-:..::
". • ,... ~-',." F'z,:';~i,L~.:2"-:.;;,~I..:,:..,"
L".. "..:.~j: "" :']
"" ~.-,~,'.-'." ~,.:'..,.'..,"-;"'"-= ,a
•: . . . ~ ," "'." .-~-'~."...¢,:~.., ..~ ,,- ~
i 111
L'.'.. ¢; .-u~s.'.,.......¢,,.~. ,-.,..,
"b" • ~lb • 1• i 1 . • , -]
*e ~ "•1~1 • * ib ~

• , " ~.-~,~
• 0".~., ,.,
t t ...
- , " . . , ' - .-~...~
- • " " ..r~,-,
~' "" ? , - . " s ,r B ~
. . . . . .:.- ..,
~. .;,.,
.. ~
.'~" . -;" " ' . . - . ' ; . . o °. ~ . . : " : ' . ' ' ~ , .
1;i~.., . ~ - . . "..."
r' -,.-~.',', ~ ,..~'.~'~...~..,:.
".'~,.-;~x.~-, -'v.-
"" , b ' . ~'~'I'~" " . " "", ~'~"
,::1
-.~
•" =.:, 7 " : " " ~ ( " ..~.~'¢~.~.".'¢cy" I" ~" " . " ~ : . ~ ' ~ , . ~ ,~." "" "" " ~ . " , ' " "1

': ~" ";~'-~;~" : " " :":",':~:;:"~'." g;':.:."':"6~"" ~'-?."' :':i

.':.:6."~% ' : . ~ "~ . ."'"
. " . "• ". . ' . ; . . 4 '~ . :.. .. . ,. . E',
, . "~,-,,~'~'.
......-.. . ". . . .'~
. . ~ . : ' -" -:J
.~

[: . . . . ':', ° " ...., '. * ,'1

I":'::
~'":?'::'~
,.:.: "~ " ". "" - . ' . .':'.',~.~a..~','.~:"
,~:-.'-.... t'~..~.." :~. "':. . . ."~"? ~~, . ."
[i...: " " .,,,,.:
,.:~:~:: . . . . : .". - '::t
•~ - ~ , ~ , ~ . . . _~.~'~__L'-..".j:.~ : """'~" " '~-~" " : " .... "
[-I~'::.,; .... • ,. , . ',.'.
. , i t . ~ ;;'...':.,, . , . . ~ .:..o~.-.": ;, . . ~ . . . . .c.
*~ [:'""~,,:-.~,..~ : . ." ~". , "", ~~ , . , ""~" '..'-'..':-';...
- ~" ;" . "
"""
" .~"' ~';., ~" ..• ~~t.~,-:
~ - , . . ' ~ . . •" . ' . ';¢..,~
",,~'.'.':"
• ' "-', •
,~L"
", . ,¢=,~
. .' , ' - • I ~~ ; ,t ,. . . •o , . .• - . ~ . L ~. ~ - . ... • .~.=l~,=, . . . .
.,.,....-..,.,.. :.'-..I
; -...,'."I,~_
".'.'
" ~ ' : . " , ' : ' "'" , ' ~ ." . . . .
.... s.'.:.;.~.- ::, ~ . . : ~.,:'.~. "~. •
." '"." h'... . . . . .
I..,~ "'-'-.- :.~, ".". '"..:,'.:"
",-~, ":7,'. " , .-. ",'~.~': ~ . '
~ .',.:~'..~"" " :,.
":,:1
• "~ • ', . ":" .,";' ".. :,','.,- ,:;.. I'_: .. • ..' ~ ' 1 ~ . ~. • ~."~". " ,-;',~:'l:'..;.'-,l

(c) (d)
Enhanced Fry Enhanced Normalized Fry
• • • ••

! ,. • :. "...
• ,,. ,, . . .•,~'• ;.•
.- .:~ .- . ..-.,. ~ .;.''
: ..• ~ ,~, • !
• ~ • • • • • .
• .. " • .. .I,:,;' " ...'-

•. •..." ..~" . . . . " ~" • : .,:;..'. i;';:

•: " • •...'.. : :;,7 - :" •
• • • .:~,
""~" "" + "" "" " , ".d'. + :" "
• o • . ,, .;
-...;'. ;.". .-:,.-.
. . . . ,: . "..*." :
:..,~ . . . . .',.. .. : ~,. • .-.~:'.
• .. " . . .. .~,. :... . ~.-¢
,.-. ~.~..: .¢ • ... . . . . .
• • ... : "v.':.-.'.~',.:, "
• . ,:. ..:'. . ~ ".'...
.- ",.. , )

.. * .•. " •

162 pairs of nearest-neighbor objects (selection factor - 1.07)

LEAST SQUARES BEST-FIT ELLIPSES

Conventional Fry method: X/Y - 1.484 Phi - 45.17
Average error: 2 0 . 6 1 %
Normalized Fry method: X / Y . 1.567 Phi - 47•77
Average error: 7.86 %

Fig• I. (a) Fry (1979) all--object--object separations, (b) normalized Fry (Erslcv 1988), (c) enhanced Fry (EFry) and (d)
enhanced normalized Fry (ENFry) plots with least-squares best-fit ellipses and their associated errors for the aggregate from
fig• 5.7 in Ramsay & Hubcr (1983).

grated fabric analysis program for IBM-PC compatible lished. Current line strain (Panozzo 1984, Schmid et al.
and Macintosh computers. This paper will derive a least- 1987), mean ellipticity (Ramsay 1967) and mean ellipse
squares algorithm for a best-fit ellipse centered at the (Shimamoto & Ikeda 1976) methods give unique values
origin, propose new methods of fabric analysis using of ellipticity independent of analyst bias but lack direct
this algorithm and test the precision and accuracy of the error calculations.
methods• A n o t h e r approach is to minimize an error equation
quantifying the deviation of the observations from a
best-fit ellipse• The standard least-squares method was
LEAST-SQUARES FABRIC ANALYSIS M E T H O D S used in this analysis because the solution of the least-
squares equations gives a unique, unbiased result. The
The classical approach to quantifying the two- least-squares derivation for an ellipse centered at the
dimensional geometry of textural elements in a material origin is given in the Appendix• Some of the short-
(e.g. grain dimension or aggregate geometry) is to comings of least-squares methods are illustrated and
calculate a fabric ellipse measuring the deviation from discussed later in this paper. Other options involving the
an ideal initial state approximated by a circular geo- iterative minimization of error equations are currently
metry (Ramsay 1967)• This requires the graphical or being evaluated.
mathematical estimation of a best-fit ellipse• Calculation In order to measure the closeness of the data to the
of a best-fit ellipse allows the reproducibility necessary if calculated best-fit ellipse, the average radial error is
standards of precision and accuracy are to be estab- calculated for each least-squares ellipse• In this pro-
 Least-squares center-to-center methods 1049

cedure, the distance between the ellipse center (at the each co-ordinate. The translated co-ordinates are
origin) and data point is compared with the predicted entered into the least-squares ellipse algorithm (Appen-
distance between the ellipse center and the ellipse mar- dix) to define the best-fit ellipse.
gin on the line including the data point. The predicted " One motivation for incorporating the least-squares
distance is calculated by combining the ellipse equation ellipse algorithm in a digitizer program was to check the
with the equation for the line from the center to the data algorithm for errors and rotational invariance, which
point. The absolute value of the difference between was a problem with an earlier least-squares formulation.
these distances is divided by the predicted distance, Tests, including digitizing ellipse templates, deforming
yielding the error for a specific data point. The errors are points by synthetic simple shear and rotating points in a
summed for all points and divided by the number of spreadsheet, gave the expected ellipticities, indicating
points, giving the average radial error, no significant deviations or rotational variability (Ge
The following methods of fabric analysis use least- 1990). The comparison of results from least-squares
squares ellipse (Appendix)and error algorithms, imple- digitizing with earlier results from manual four-point
mented in PASCAL subroutines, to calculate the size, and five-point digitizing shows increased consistency,
ellipticity and inclination of the best-fit ellipse, suggesting improved accuracy.
The number of points required for accurate esti-
Least-squares object digitizing mation of the best-fit ellipse depends on the angularity
of the object. Three objects, a slightly pear-shaped
Accurate fabric analysis presumes the collection of ooid, a six-sided quartz polygon and a triangular
accurate, reproducible object shape data. The definition quartz grain were digitized 20 times using different
of elliptical objects using four points at the end-points of increments between data points (Fig. 2). These incre-
the principal axes of an ellipse requires the manual ments gave between five and 44 points on the margin
determination of an average or inscribed ellipse for each of the objects, with smaller increments giving larger
object, adding subjectivity and time. Digitizing five numbers of points. For each increment and grain
points on the margin of the object, with the computer shape, the standard deviation of 20 least-squares ellip-
calculating the conic uniquely defined by the points, is ticities was calculated to evaluate the effect of different
faster, particularly for elliptical aggregates like strained numbers of points on the determination of the best-fit
ooids, but assumes that the selected five points are ellipse. For all three grain shapes, the standard devi-
representative. The selection of representative pointsis ation of ellipticity increased dramatically for larger
particularly problematical for objects with planar or increments, which gave fewer points per grain margin.
cuspate grain boundaries. The difference in precision between the angular, tri-
To increase the accuracy of individual object descrip- angular grain and the more equant grains reflects the
tions, the least-squares algorithm (Appendix) was used greater deviations from an ideal ellipse at grain asperi-
in a digitizing program which approximates object out- ties, which may not be represented in large increment
lines as ellipses. The program collects points on the grain digitizing. This analysis suggests that the digitiz-
margin of an object as it is traced in increment mode. ing increment should be set to allow 25 points per
These points are translated toward the origin by sub- grain to allow accurate approximation of irregular
tracting the centroid co-ordinates (average x, y) from grains as ellipses.

0.3
"l
\ Precisionof Least-SquaresObject Digitizing
k,.,~

~. 0.2 t.l

~ , . . .. . . . . . . . . ,

~; -.. ~ Triangle
0.1 " ""

qP Polygon

i l Ooid
o =s ,s

Points Per Object

Fig. 2. The reproducibility of least-squares ellipse definitions based on different numbers of digitized points on the margins
of three grain shapes.
1050 E . A . ERSLEVand H. GE

Enhanced Fry (EFry) and enhanced normalized Fry normalized Fry plot starts at a distance D, = 1,0, which
(ENFry) methods is defined by touching circular cross-sections. For de-
formed, elliptical objects, object radii are variable so the
The Fry (1979) and normalized Fry (Erslev 1988) average radius for each object (R -- (X/2 * }'/2) lr2, where
methods offer a graphical approach to center-to-center X and Y are the long and short axes of the ellipse) must
fabric analysis (Figs. la & b). Fry (1979) plots are an be defined. The deformed case is simply related to the
elegant way to analyze anisotropy in anticlustered popu- undeformed case by a stretch factor, which does not
lations but the lack of sufficient two-dimensional anti- affect the validity of this argument (see equations 4 in
clustering in most aggregates makes ellipse selection Erslev 1988).
difficult. The increased point density contrast in normal- Thus, for packed aggregates, the rim of maximum
ized Fry diagrams (Erslev 1988) facilitates the selection point density in a normalized Fry diagram is defined by
of a fabric ellipse. However, manual ellipse fitting to the objects in or nearly in contact with each other. These
rim of a maximum point density is still required, pairs of 'touching' objects can be manually selected by
Ideally, we would like to eliminate, in an unbiased entering pairs of approximate center locations. How-
way, all points except those contributing to the rim of ever, this adds a subjective step which might bias the
maximum point density. The problem with quantifying analysis. Alternatively, touching pairs can be computer
the Fry technique is the lack of logical, non-subjective selected by comparing the center-to-center distance (D)
criteria to eliminate distances which do not contribute to between each pair of objects with the sum of object radii
the rim of maximum point density. Restricting the size for each pair (ra + rb) measured parallel to D. These
range of object cross-sections used to define centers does radii are elliptical radii measured on the line between the
clean out the central region of standard Fry plots, but two centers in question, not average radii used to norma-
this also reduces the number of nearest neighbors for a lize the distances. Dividing the center-to-center distance
given data set. In an attempt to clarify the rim of by the sum of these elliptical radii ( D / ( r a + rb) ) gives an
maximum point density, Crespi (1986) eliminated dis- object-pair selection factor (or selection factor for short)
tances below a minimum center-to-center distance, but which allows an adjustable criterion for selecting object
this makes the inner void more circular. In addition, no pairs. For example, a selection factor of 1.0 will only
unbiased criteria exist for removing center-to-center accept pairs whose D/(ra + rb) -< 1.0. These pairs include
distances which plot outside the rim of maximum point elliptical objects in contact with each other and objects
density, whose defining ellipses overlap. This is common in
For homogeneously deformed, packed aggregates, polygonal aggregates with interpenetrating grain boun-
the rim of maximum point density in normalized Fry daries.
plots consists of distances between touching or nearly The normalized center-to-center distances between
touching neighbors. The following, synthetic example of 'touching' pairs, as defined by the object pair selection
a two-dimensional, undeformed aggregate of circular factor or by manual identification of nearest-neighbors,
objects illustrates this relationship. The normalized dis- can be plotted in an enhanced normalized Fry (or
tance (Dn) between objects a and b equals the actual ENFry) plot (Fig. ld). The true center-to-center dis-
center-to-center distance D divided by the sum of their tances can also be plotted in an enhanced Fry (or EFry)
radii. Since these objects cannot overlap and each object plot (Fig. lc). However, the rim of maximum point
has perfectly circular cross-sections, the minimum Dn density in a Fry plot is not uniquely defined by touching
(1.0) is defined by touching pairs of objects. For this objects so this diagram must be interpreted carefully.
undeformed case, the rim of maximum point density in a The selected center-to-center and normalized center-to-

(a) (b) (c)

Fig. 3. Tracings of representative portions from thin sections with at least 1500 objects used for tests of precision and
method variables. These samplescome from (a) oolitic Ing|esid¢ Formation of Permian age from northwest of Fort Collins,
Colorado, (b) Cambrian Flathead Sandstone from the northern Teton Range and (c) upper amphibolite facies quartzite of
Archean age from the southwestern Beartooth Mountains.
 (a) Oolite (b) Sandstone (c) Quartz Polygons

, 2.6 -v. 2, i ,,o ° o° .

"~2, ~o.
:!tl
-e,• 2 ,
2.2
e,
,.~
24
2.2"
,"
0 o
l.
° . 2,
e,
.6
2,
2.2" II
i , .
2- ! ~ 2-
o- ~ to ! ! + ,8- ~ O°
[ t.6- O :+ t.6- 0
1.4- 0
,+,. ; -. ,R ,
,+ , ~ ~. ,2 o <,
1 .l* ~. t- S*
p+ o.8-
' i i i~ :+ o.,~ I] ' p o.e- e ,-,+
06-
~, o.+- ni ~ o.4- ,i ~ o.4 li ,',
~- 0.2 I 0- 0.2
0.2-
o ; ''"neuile*" 0 0
o.8 ~ 1:2 o.s ~ t'.2 o.e '~ t'.2 +~
(..p

40. 40 40- i ~"

35" , 35- Enhanced Fry 3.5- *0 * , ° g ~ II ~) i o~ ~ ~ B ea

25
0o
° Enhanced Fry ~ 2.5- o 9° g 9
"o* °o
e e
~"
~
2.5- e 8 o , e .oo, ° Enhanced Fry =
,-I
' * **. E, 20- 00 ~ 2o. ". • ~
~o ° ~ , ° o. ,:~,,l!:++:_o,~[~. o o i - . --.+ +,: ; ' i + ; + ; i i i i i i i i i ; ; ; ; ' ; o
-~
+
o~+° t.5 o :~..egOS* 0~: ,5- ::''';~l!!:;illllllllill':... 0: t.5. . . . .-q+
+0. Enhanced Normalized Fry
,o
• •
,..
°
: ] i ~ :ii
~ "
| | | | | I ! I I I ! I 10" Enhanced Normalized Fry

5- 8 : s Enhanced Normalized Fry 5. 5-

0¢:" 0 0
o.8 '~ ai2 o.e q ;.2 0.0 '~ ~'.2
Object Pair Selection Factor Object Pair Selection Factor Object Pair Selection F'octo~

Fig. 4. Tile number of touching pairs and average radial error plotted vs selection factor for the aggregates illustrated in Fig. 3.
1052 E . A . ERSLEVand H. GE

center distances from 'touching' objects can now be Sorting

entered into the least-squares ellipse algorithm which 3s
provides an unbiased estimation of ellipticity, long axis EFry _ . . . . . . . - -
inclination and average radial error for both the EFry 3o ~ ~ \ /
and ENFry methods.
In order to determine the range of optimum selection ~-
2s
=" ~ /
2 "~ - ~_
factors, data sets of 1500 objects from minimally de- °S ~ ~ ~ ~
formed samples (Fig. 3) were digitized using the least- ~ ,.~ ~, ~ ~
squares program and the smallest digitizing increment, " 10
ensuring at least 25 points per object. These aggregates ENFn/ ~ ~
were selected to represent the spectrum of packed 5
aggregates, ranging from well-rounded, undeformed 0
0.2 0.4 0.8 o.s
oolites to polygonal quartz in an upper amphibolite Standardt~,iationofPhl
facies quartzite. The samples show no obvious fabric
heterogeneity in thin section. Data files were checked Fig. 5. Sorting, as measured on the analyzed plane by the standard
deviation of ¢ ( Folk 1974), plotted vs average radial error given by the
for anomalous data using Rf/¢ plots. Objects with ellip- enhancedFry (EFry) and enhancednormalizedFrymethods(ENFry).
ticities greater than 1 above the next largest ellipticity
were eliminated from the data sets.
The number of 'touching' pairs selected is a function the number of touching pairs visually and adjust the
of the selection factor and the aggregate type (Fig. 4). selection factor to give that number of pairs.
Increasing the selection factor increases the number of One advantage of normalized Fry techniques over Fry
'touching' pairs chosen for the calculation of the best-fit techniques is that they appear to be independent of
ellipse and average radial error. Elliptical objects like sorting. This hypothesis was tested using the aggregates
the ooids in Fig. 3(a) require higher selection factors of spheres from Erslev (1988) and three aggregates of
than the polygonal quartz grains in Fig. 3(c) to get variably-sized lead shot. Sorting wascalculatedusingthe
equivalent numbers of 'touching' pairs because ooids standard deviation of the object ¢ derived from object
have less object interpenetration and resulting overlap areas on the two-dimensional surface (Folk 1974, Ge
of elliptical object approximations. In many cases, each 1990). Figure 5 shows that the average radial error of the
aggregate must be evaluated separately for the optimum enhanced Fry method increases dramatically with de-
selection factor, creased sorting. The average radial error of the
For ooids, the average radial error from the enhanced enhanced normalized Fry method remains roughly the
normalized Fry method is highly variable at selection same regardless of sorting, with the aggregates of
factors less than 1.0, stays levelbetween 1.0and 1.1 and slightly elliptical lead shot giving higher errors than
then increases (Fig. 4a). The scatter at low selection aggregates of spheres. This experiment confirms that
factors is the result of the small number of object pairs normalized Fry techniques are independent of sorting.
selected. Increasing errors at selection factors above 1.1
result from decreased point density beyond the rim of Meanobject ellipse (MOE) method
maximum point density in a normalized Fry plot, result-
inginalargerspreadofx, yco-ordinates. The same basic The calculation of mean ellipticities is one of the
patterns occur in the sandstone and polygonal quartz oldest techniques for quantifying strain. Ramsay (1967),
aggregates, but at lower selection factors because of Dunnet (1969) and Lisle (1977) have shown the utilityof
more object interpenetration (Figs. 4b & c). arithmetic, geometric and harmonic mean ellipticities
Another approach to choosing the optimal selection despite their tendency to overestimate the ellipticity,
factor is to determine how many touching pairs should particularly for minimal fabric anisotropies.
be expected for the objects if they originated as a close- The reason for this error is illustrated in Figs. 6(a) &
packed aggregate. Synthetic aggregates of equi-volume (b). If an aggregate of undeformed objects has equival-
spheres in Erslev (1988) had 0.60 touching pairs per entyet randomly orientedinitialellipticity, mean ellipti-
object. Packed spheres in Chilingarian & Wolf (1975) cities will be closer to the initial ellipticity (1.5) than the
had 0.63 touching pairs per grain. Samples of lead shot tectonic ellipticity (1.0). Because mean ellipticities ig-
with variable sorting had 0.63, 0.57 and 0.55 touching nore the effect of long axis orientation (¢), they do not
pairs per object. Thus, for an aggregate of originally fully characterize the geometry of the fabric anisotropy.
close-packed objects, approximately0.6 touching pairs An alternative method, the mean object ellipse
per object should be selected by varying the selection (MOE) method, is illustrated in Fig. 6(c). MOE calcu-
factor. In general, concave-inward objects like ooids or lares the least-squares best-fit ellipse from the shape of
poorly cemented sand grains require selection factors objects, which are input as ellipses, by combining their
between 1.0 and 1.1 whereas polygonal aggregates re- shapes into a single, average ellipse. First, the center of
quire selection factors between 0.9 and 1.0. the ellipse approximating each object is translated to the
Non-close packed aggregates with abundant matrix origin. In order to remove size effects, the object area is
material will yield fewer 'touching' pairs for a given normalized to a constant value, effectively expanding or
selection factor. In this case, the analyst should estimate shrinking the object without distorting its shape. Since a
 Least-squares center-to-center methods 1053

(a) Given randomly oriented ellipses (b] Mean Ellipticities (X/Y)

with Rt = 1.5. R s = 1.0
Arithmetic Mean = 1.5

Bath means = RI, not R s

(c) Mean Object Ellipse

Step 1: Center each object at Step 2: Calculate l e a s t - s q u a r e s

0,0 and normalize to equal area b e s t - f i t conic ( A x 2 + B x y + C y 2 = l )
from 3 points per object

X / Y = 1.0 = R s

Fig. 6. Graphical illustration of the difference between (b) mean ellipticity and (c) mean object ellipse calculations. For
clarity, the diagram in Step 2 of (c) plots three points per object plus their symmetric equivalents, giving a total of six points
per object.

centered ellipse is uniquely characterized by three points the three points back to their original position. The three
on its margin, the program (1) rotates the principal non-centrosymmetric points for each object are input
ellipse axis parallel to the x co-ordinate axis, (2) calcu- into the least-squares and error procedures to determine
lates the end-points of the maximum and minimum the mean object ellipse. These routines output the best-
ellipse axes and an intermediate point on the ellipse at x fit ellipticity, long axis angle (~) and associated average
equal to one half the maximum stretch, and (3) rotates radial error for the entire aggregate. This method is
similar to the mean ellipse calculations of Shimamoto &
Strain Versus Distance Ikeda (1976) who calculated the ellipse equation for
6
o = = = ( a ) ObjectElliptieity each object and then averaged the coefficients.
s The M O E method is ideally suited for the creation of
° smoothed strain maps by calculating the mean object
4 ellipse for the nearest neighbors of each object. Figure 7
*
OoO O OOO
° shows two ellipticity profiles through a quartzite cut by
"~ 3 o =
g~o ° 0° 0 . ° . ° o • . ° three cleavage zones (Powell 1982). Both individual
z" ~ o° a o o, d , ,gs ° . .~ ° ~ o object ellipticity (Fig. 7a) and M O E (calculated for the
O 2 o= on ~ ~ ~0 o oO s a =_~0¢~ a %
0_~.*~=~-o ~ ~ , _ ~ . ~ =_= = =e = = nearest five objects and plotted in Fig. 7b) show higher
1 e~~, ° ==~'~'#00 as*.~d~=¢~="=~0 0~ r ~ ellipticities in the three cleavage zones. The variability
of the raw ellipticity values is smoothed out by the
0 . . . . . . . . . . . . . . . averaging effect of the M O E ellipticity.
0.6 1 1.4 1.8 2.2 2.6 3 3.4 3.8

6
( b ) MeanObjectEllipse (n-S) PRECISION AND ACCURACY OF E F R Y , E N F R Y
s AND M O E M E T H O D S

_ 4
~. The precision (reproducibility) of the new fabric tech-
"'
1~ 3 °%o = % niques was tested on natural aggregates using four data
~" * = =° *** sets of 1500 objects from the minimally deformed
J ~ l _ * .= - *=~ ~ ~*= * = samples illustrated in Fig. 3. The selection factor was
° • O ~.~
~--'s.~*~*~0~, o ~ - ~ g ° =~Jl,,= ~ 0 ~ 8 . ~ . , , ~ _ _ chosen to allow approximately 0.6 'touching' pairs per
1 °°°**~*~= ~*~f°~ °*'~ ~ object. The digitized object data were entered into a
modified version of I N S T R A I N which outputs fabric
00.e' ~ ' 1'.,' ~'.e a'.a' a'.e' ~ ~'.,' ~.~ data to a disk file for non-overlapping subsets of objects.
The number of objects per subset was incremented by 25
Distance Perpendicular to Cleavage (ram) from 25 to 600, the maximum allowed by the program.
Fig. 7. Evaluation of the heterogeneous strain in spaced cleavage Figures 8 and 9 show the variations of average radial
cutting a quartz arenite in Powell (1982). Distance is measured from
the bottom of the photomicrograph in Powell (1982), with cleavage error, long axis inclination and ellipticity for the EFry,
zones at 1.2, 2.1 and3.2 ram. M O E and ENFry methods with different data subset
1054 E . A . ERSLEVand H. GE

Enhanced Fry Ellipse: Oolite Enhanced Normalized Fry Ellipse: Oolite Mean Object Ellipse: Oolite
40. 40- 40-
35" 35" 35-
30 30" 30-
25 25. 25-

1.5, ! ill: ......... ' ~ 20.

.. 15. Is- llllllitl,,,,,,,,=:i,,
~o ;. ,o. " !:::::':":: ~o.
5' 5" = 5-

o 260 400 600 o 2oo 460 600 o 260 460 5oo

9o I : ': ,! •

30: " "" 30: m • :l,m ''" " .

• | .... ' " " " i " "-' .30:. I ~|lill'''''''l:
o ::.." • •
~. o ::,
"-
~ o: !1:"
- ;I . •

-~o: • .' , • -3o: . ; i . : :., -3o: , : ' ,

-60: . ' ' -60: • -60: I

'I .. • 'I : •.
-90 "; " -90' "" " -90 •
260 460 600 260 460 600 260 460 600
3" 3" 3

2.6: 2.6: 2.6:

2.2: 2.2: 2.2:

,,, iI | 1.8: × 1.8:

• " I li~w
260 460 600 0 260 400 600 260 460 600
Number of Objects Number of Objects Number of Objects

Fig. 8. Scatter plots of average radial error, long axis inclination and cllipticity for EFry, ENFry and M O E ellipse
dctcrrninations using non-overlapping subsets of the oolitic aggregate illustrated in Fig. 3(a).

Enhanced Fry Ellipse: Qtz Polygons Enhanced Normalized Fry Ellipse: Qtz Poly~ ~ns Mean Object Ellipse: Qtz Polygons
40. ii 40. 40.--

2,5. I!!~ 25. 25-

Is • ~ :s
.1:; 20. I ,~ 20. I 20,

,o ,o illi'""'"'"'"'"" li,,,,,',,,,,"
........
5' 5' 5"

0 280 460 600 0 260 480 600 0 260 400 600

90 90" 90.
ii ° •
601 "," ' 60: * 601 •

'°:' if!" "" "" '°:- ~o:. .

•o.~ o.l ,:,; ' ; : " : ........... -~. o' '"I . :' ~ I : . . , ", . , , . , , "~ o' i,.: ........ ' ........

-301. • """ : : "" " ..."" -30:. .!"

I:=;:: ' • "" -301 ilJ; :;
"': . . . . . . ". . . .
-6o :l: -5o: I -6o: '

-90' ' 260 460 600 -90' : i 260 460 600 -9o u 260 460 600
3. 3: 3-

2.6: ' 2.6: 2.6:

~22::;: ... ~22i ~22:

"iiii
,:200i:":
"
i ,00 .. . .
600
. .
'i
. .
i200;lii:::::::::,,
460 600
:
u
;;;,:.._,..:::.::.::..___...
260 . . . 460
. . 600
Number of Objects Number of Objects Number of Objects
Fig. 9. Scatter plots of average radial error, long axis inclination and cl]ipticity for EFry, ENFry and M O E ellipse
determinations using non.overlapping subsets of the polygonal quartz aggregate illustrated in Fig. 3(¢).
 Least-squares center-to-center methods 1055

sizes for an oolite and a quartz polygon subset. The other the MOE least-squares algorithm whereas each object
two subsets (another oolite and the quartz arenite in Fig. only contributes 0.6 points to the ENFry least-squares
3b) are analysed in Ge (1990) and summarized in Fig. 10. algorithm.
For all the methods, the range of values narrows con- Figure 10 summarizes these precision measurements
siderably for subsets of 200 objects, giving a good esti- by plotting the standard deviation of the ellipticity vs the
mate of the minimum number of objects needed for number of objects in the data set. MOE and ENFry
reasonable precision. The EFry method, which is depen- methods are approximately twice as precise as the EFry
denton the two-dimensional anticlustering of the aggre- method, which is clearly not the best choice for the
gate, gives larger errors and results in more variable analysis of packed aggregates. For large (>200 object)
eUipticity and ¢ values. The MOE method is more aggregates, MOE and ENFry give similar levels of
precise than the ENFry method for smaller numbers of precision, with all data sets giving standard deviations of
objects because each object contributes three points to ellipticity <0.1. For these methods, the oolitic lime-
stones gave the lowest standard deviations of ellipticity,
averaging <0.05 for aggregates with more than 200
(a ) objects.
0.4 Precision of Enhanced Fry Method The accuracy of fabric techniques is difficult to quan-
~ tify from natural samples since we lack independent
.', methods of determining the true fabric anisotropy. An
"6 0.3 ~x QuartzPolygon=
j ~\ , ,, ~ . alternate approach is to create an ideal, undeformed
\ \ , , " ", ~ aggregate with a given range of initial ellipticity and then
0.a ~ ,.~".._- - - ' .\. . . . . . . . . " , deform the aggregate by applying a stretching factor.
i " ~"x~_._~'-'-~-'-'--~ Sand=tor~ ", Figure 11 shows two synthetic aggregates created using
~ ~ " ~ . _ . < . ~ ~ - ' - /"'. ', an ellipse template. These aggregates were copied by
0.1 "~"-..,.,
~ . ~ ~ . ~ ~ ' _ five rotations at 30° increments to increase the total
number of objects and to assure random initial ellipse
0 . . . . . . . . . . . . . orientation. The samples were defined by five points on
20 60 100 140 180 220 260 300
Numberof Object8 the margin of the ellipses, allowing incremental defor-
mation of the objects by multiplying a stretch to the x co-
( b ) 0.2 ordinate of each point. Since the samples are not realisti-
Precision of Enhanced Normalized Fry Method caily anticlustered in two dimensions, the accuracy of
the EFry method, which is dependent on the degree of
o.ls • anticlustering, cannot be evaluated.

i 0.1

0.05
~
Ou~uPo~gon=
",.~-_--'-- I'-, ~
"• ~ , . _._.,.- . . . . . ~
s~a=to~

" " , , .. .-. ~. .,.,

', ,.
,..~'_.
The ratios of the strain values given by MOE, arith-
metic mean ellipticity, harmonic mean ellipticity and
ENFry methods to the synthetic deformation are plotted
versus the synthetic deformation in Fig. 12 and summar-
ized in Table 1. The harmonic and arithmetic mean
• ...... " " ellipticities overestimate the strain, particularly at low
synthetic strains where most of the object ellipticity is
o contributed by initial eilipticity. The MOE method cor-
20 ' ~ ' 16o ' 1~o - 18o ' 22o ' ~o ' 30o rectly calculates the undeformed case yet overestimates
Num~ofO¢~== the deformed case. This appears to be the result of the
(c) 0.2 asymmetry of R e values relative to the tectonic ellip-
ticity. Since the tectonic eilipticity is closer to the mini-
Precision of Mean Object Ellipse Method mum ellipticity than the maximum ellipticity, the strain
0.1s is overestimated. The same effect causes some of the
.~ , overestimation of strain by the arithmetic mean.
I ,,. ~ QuartzPolygon=
0.1 ~'~ " ' ~
~-~.,, "~. ~,~ ~Nlu'tdltone (i) Synthetic Acjgreejate 1 (b) Synthetic Aggregote 2

O.Ofl "" " ......

0 . . . . . . . . . . . . ,
20 60 100 140 180 220 260 300
Number of ObJectl
Fig. 10. Summary of the tests of precision (reproducibility) in Figs. 8
and 9, and Ge (1990). The standard deviation of each measure of
ellipticity for subsets of 25-300 objects is plotted vs the n u m b e r of Fig. 11. Synthetic aggregates used to test the accuracy of the fabric
objects in the data subset, methods.
1056 E . A . ERSLEVand H. GE

a) gram was modified so that the x-component of the

1.r center-to-center distance increased more than the actual
Synthetic 1
1.6 deformation of initially circular objects. The resulting
1.5
MeanInitialEIIipticity= 1,3 Fry and normalized Fry plots gave correct results for
~ matrix deformation whereas MOE and mean ellipticities
c 1.4
gave correct results for object deformation. However,
1.a the EFry and ENFry methods, implemented with auto-
1.2
~Mean
N,/
Object Ellipse
matic selection of touching pairs, gave least-squares
1,1 ~ ~ Harmonic Mean ellipses intermediate between the object and matrix
~I strains.
1 -- An analogous difference between matrix and object
Enhanced Normalized Fry J
o.9 . . . . . . . . . . . . . . . . . . . strains occurs in the ironstone oolites from fig. 7.7 of
1.2 1.4 1.s 1.a 2 2.2 2.4 2.e 2.a Ramsay & Huber (1983). Figure 13 shows two pairs of
.. Synthetic Deformation (Rs) EFry and ENFry plots for this aggregate with different
1.7
b) object-pair selection factors. The lower selection factor
1,6 ~ Synthetic 2 preferentially picks object pairs aligned parallel the the
MeanInitial Ellipticity 1.7 = Y axis of the strain ellipse. Object pairs parallel to the X
=~ 1.s ~ axis are not as readily selected since the larger matrix
1.4
\\Arithmetic Mean deformation has pulled them apart so they no longer
\ \ 'touch' in the x direction. Thus, the EFry and ENFry
Mean Object Ellipse
1"3 ~ / MeanObje~l,ElllpSeu_" . ..___ least-squares ellipses calculated from object pairs
monlc
1.2 selected by the computer do not give accurate esti-
mations of the true matrix strain. If a sample population
~= 1.1 shows the radially biased selection of pairs seen in Fig.
1 Enhanced Norrnalizedgry J 13(a), a better estimate of matrix strain is given by
0.9 . . . . . . . . . . . . . . . . . . . manually fitting an ellipse to a normalized Fry plot.
1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 3 The effect of spurious data on the strain methods is
Synthetic Deformation (Rs)
illustrated in Table 2 using a data set from fig. 5.7 of
Fig. 12. The r a t i o o f t h e measured ellipticity over the actual, synthetic Ramsay & Huber (1983). It was acquired using an older
deformation (R,) plotted vs the synthetic deformation of the x dimen- version of the least-squares digitizing program which
sions. A perfect fabric method should always give a ratio of 1.0.
lacked comprehensive error checking. As a result, two
spurious objects were defined with ellipticities of 38 and
The ENFry method gives a close approximation of the 11. Each spurious object contributed one additional
actual strain in both cases. Of all the automated tech- center-to-center distance between 'touching' objects.
niques, only the ENFry method gives errors which do MOE is particularly affected indicating that any spuri-
not vary systematically with deformation. A simple ous data will overwhelm the technique. The arithmetic
manual implementation of the Re/~ method gives accu- mean ellipse is also strongly affected. The harmonic
rate results for synthetic 1 but shows large errors for the mean ellipticity and enhanced normalized Fry method
more variable synthetic 2. This error (note the large show the least distortion. Since the harmonic mean sums
standard deviations)reflects observer errors (E.A.E.) the inverse ellipticities, the effect of large values is
in the determination of Rfmax and R f m i n . automatically minimized. The enhanced normalized Fry
This accuracy test assumes homogeneous defer- method requires two 'touching' objects with aligned
mation with identical object and matrix strain. Ideally, spurious ellipticities to create an equally anomalous
we would like to evaluate these strains separately. To center-to-center distance. Thus, the effect of one spuri-
test the effects of heterogeneous deformation, the pro- ous object is mitigated by the surrounding objects.

Table I. Summary of accuracy tests

Synthetic 1 (382 objects)* Synthetic 2 (408 objects)-t

Average Total Average Total
measured/strain error:~ measured/strain error¢
Method (-+1 SD) (%) (_1 SD) (%)

Enhanced normalized Fry 0.9949 __. 0.0028 0.79 1.0015 __. 0.0025 0.40
Mean object ellipse 1.0475 _ 0.0269 7.44 1.1194 _ 0.0364 15.6
Harmonic mean 1.0525 __. 0.0840 13.6 1.1386 + 0.1478 28.6
Arithmetic mean 1.0885 -4" 0.0739 16.2 1.2636 --- 0.1293 39.3
Rfl¢p§ 1.0059 __. 0.0141 2.00 1.2018 _ 0.1380 34.0
* 14 measurements at 0.15 R, increments from 1.0 to 2.95 (Fig. 1 l a).
"t21 measurements at 0.10 Re increments from 1.0 to 3.0 (Fig. l i b ) .
eTotal error = 100 * (Absolute value (1 - average measured strain) + 1 SD).
§From manually measured Rf m=~ and Rf ml. by E . A . E . on nine (synthetic 1) and seven (synthetic 2) Rf/~p
plots.
 Least-squares center-to-center methods 1057

(a)
Enhanced Fry Enhanced Normalized Fry

.. '; j .
°. .'..
: • "." . .. :.. ., . . .. . • ,2 . ' ~ . ; . ~
• -...- :. :':.

.: - - •

.. 4- .'.. 4-
..... • .: ;•:
"..":; ~ : ..,,. . . . . ..
" . ". ... • ...:.. ..
...

48 pairs of nearest-neighbor objects (selection factor = 1.01)

LEAST SQUARES BEST-FIT ELLIPSES

Conventional Fry method: X / Y , 1.643 Phi = -23.82
Average error: 14.87 %
Normalized Fry method: X/Y = 1.598 Phi = -23.93
Average error: 5.35 %

(b)
Enhanced Fry Enhanced Normalized Fry

• • "...~..., :. .

•" • .', "• , ' " " ~ ' L , ' :". I."" "": , . . : . q.e.. .
. . ..,.... C.•~¢,.....
•" . • "." .. " . . " ~'; " t',,.
r'. • ":;.

"~.," . ... • + • . ".: .:,~. + .:,"

• • .....* t ';';" • .t .
.. "..." ...:..
• .. :'.. • . :.'.' ..:. ..
, .;.,, ,.: .. • • • : "'.l . .;.'r
• j . • . , ". t .. " .. "~-~.;',..:.:,: .... s

141 pairs of nearest-neighbor objects (selection factor ,, 1.07)

LF_AST SQUARES BEST-FIT ELLIPSES

Conventional Fry method: X/Y = 1.672 Phi - -21.94

Average error: 12.56 %
Normalized Fry method: X/Y = 1.641 Phi ,, -22.87
Average error: 5.79 %

Fig. 13. Enhanced Fry (EFry) and enhanced normalized Fry (ENFry) plots of fig. 7.7 in Ramsay & Huber (1983) showing
the effect of heterogeneous strain on the EFry and ENFry methods. Note the unequal radial distribution of points in ENFD'
plots, particularly at low selection factors.

Table 2. Effect of spurious data on quantitative fabric methods

Files with spurious data

Method 276 objects* 277 Objects* 278 Objects*

Maximum ellipticity 3.11 11.00 38.36

Rogue values none 11.00 38.36, 11.00
Mean object ellipse 1.599 1.657 3.177
Arithmetic mean ellipticity 1.634 1.667 1.799
Harmonic mean ellipticity 1.568 1.573 1.578
Enhanced normalized Fry';" 1.556 1.559 1.561
*Digitized ironstone ooids from fig. 5.7 of Ramsay & Huber (1983).
~'Using an object-pair selection factor of 1.05.
1058 E.A. ERSLEV a n d H. GE

CONCLUSIONS Erslev, E. A. 1988. Normalized center-to-center strain analysis of

packed aggregates. J. Struct. Geol. 10, 201-209.
Folk, R. L. 1974. Petrology of Sedimentary Rocks. Hemphill Publish-
Best-fit ellipse c a l c u l a t i o n s can g r e a t l y i n c r e a s e t h e ing Co., Austin, Texas.
a c c u r a c y a n d p r e c i s i o n o f f a b r i c a n a l y s e s while r e d u c i n g Fry, N. 1979. Random point distributions and strain measurement in
rocks. Tectonophysics 60, 806-807.
subjectivity. Accurate, unbiased approximation of an Ge, H. 1990. Quantitative center-to-center fabric analysis of homo-
o r i g i n a l l y e q u a n t o b j e c t o u t l i n e as an ellipse is f a c i l i t a t e d geneous deformation. Unpublished M.S. thesis, Colorado State
b y l e a s t - s q u a r e s ellipse fitting o f d i g i t i z e d p o i n t s o n its University.
margin. Angular grains require the input of more points Graton, L. C. & Fraser, H. J. 1935. Systematic packing of spheres--
with particular reference to porosity and permeability. J. Geol. 43,
t h a n r o u n d e d grains for e q u i v a l e n t levels o f p r e c i s i o n . 785-9o9.
The enhanced normalized Fry (ENFry) method calcu- Lisle, R. J. 1977. Estimation of tectonic strain ratio from the mean
lates t h e best-fit ellipse t h r o u g h the rim of m a x i m u m shape of deformed elliptical markers. Geol. Mijnb. 56, 140-144.
Lisle, R. J. 1985. Geological Strain Analysis: A Manual for the Rt/~
p o i n t d e n s i t y o f a n o r m a l i z e d F r y d i a g r a m by o n l y Technique.Pergamon Press, Oxford.
plotting distances between 'touching' pairs. The method Panozzo, R. 1984. Two-dimensional strain from the orientation of
is extremely accurate and precise for aggregates contain- lines in a plane. J. Struct. Geol. 6,215-221.
Powell, C. MeA. 1982. Reduction of elastic grain size within cleavage
ing m o r e t h a n 200 o b j e c t s . T h e E N F r y m e t h o d is also zones. In: Atlas of Deformational and Metamorphic Rock Fabrics
independent of variations in object size due to two- (edited by Borradaile, G. J., Bayly, M. B. & Powell, C. McA.).
d i m e n s i o n a l effects a n d sorting. T h e e q u i v a l e n t Springer-Verlag, Berlin, 302-303.
Ramsay, J. G. 1967. Folding and Fracturing of Rocks. McGraw-Hill,
e n h a n c e d F r y ( E F r y ) m e t h o d gives at least twice t h e NewYork.
average radial error for packed aggregates and should Ramsay, J. G. & Huber, M. I. 1983. Techniques of Modern Structural
o n l y be u s e d for p o p u l a t i o n s with m o r e highly d e v e l o p e d Geology, Volume 1: Strain Analysis. Academic Press, London.
Schmid, S. M., Panozzo, R. & Bauer, S. 1987. Simple shear experi-
two-dimensional anticlustering, ments on calcite rocks: rheology and microfabric. J. Struct. Geol. 9,
The mean object ellipse (MOE) method provides a 747-778.
high precision estimate of object shape. However, the Shimamoto, T. & lkeda, Y. 1976. A simple algebraic method for strain
estimation from deformed ellipsoidal objects. Tectonophysics 36,
use o f a l e a s t - s q u a r e s a l g o r i t h m c a u s e s o v e r e s t i m a t i o n o f 315-337.
object eilipticity due to the asymmetric distribution of Thomas, G. B. 1967. Elements of Calculus and Analytical Geometry.
o b j e c t ellipses a r o u n d t h e t e c t o n i c ellipticity. I d e a l l y , a Addison-Wesley, Menlo Park, California.
l i n e a r o r l e a s t - s q u a r e - r o o t best-fit a l g o r i t h m w o u l d les-
sen t h e influence o f s p u r i o u s a n d distal v a l u e s o f ellip-
ticity. H o w e v e r , t h e l e a s t - s q u a r e s M O E m e t h o d d o e s APPENDIX
p r o v i d e a g o o d tool for f a b r i c m a p p i n g b e c a u s e o f its
high p r e c i s i o n in small d a t a subsets. LEAST-SQUARES ALGORITHM FOR A
CENTERED ELLIPSE
T h e c o m b i n a t i o n o f a u t o m a t e d , best-fit ellipse d e t e r -
r u i n a t i o n with q u a n t i t a t i v e m e t h o d s o f fabric analysis A general equation for conics, including ellipses, can be written as:
promises to provide n e w t o o l s for a wide range of A x 2 + Bxy + Cy2 + Dx + Ey + F = 0. (A1)
scientific investigations. The current concentration on If (B2 - 4AC) -< 0, the equation (A1) describes an ellipse (Thomas
strain analysis should be replaced with more generalized 1967).If the elliptical pattern is centered on the origin, equation (A1)
studies of fabric anisotropy, can be simplified to that of a centered conic:
Ax 2 + B x y + C y 2 - 1=0. (A2)
Acknowledgements--We would like to thank William Dunne, Ana
Vargo, Alfred Barnes, Roy Kligfield and Declan De Paor for their For a given A, B and C, apoint(xa,ya)willsatisfyequation(A2) with
comments and insights. Reviews from David Sanderson, Bob Ratliff the residual ka defined by
and an anonymous reviewer helped condense and clarify the manu- k a = A ~ + Bx~a + Cy2 - 1. (A3)
script. Acknowledgement is made to the Donors of the Petroleum
Research Fund, administered by the American Chemical Society, for If ka = 0 and (B2 - 4A C) -< 0, the observed point (xa,ya) plots on the
the support of this research, ellipse defined by constants A, B and C.
For an observed set of points (x,,yi), the sum of the squares of the
deviations of the points from the best-fit ellipse can be minimized by
minimizing the sum of the squares of the residual ki (equation A4):
REFERENCES
Minimum ~ (k~) = Minimum Z (A~ + Bxiyi + C~ - 1)2. (g4)
Bartlett, D. P. 1915. General Principlesof the Method of Least Squares
With Applications. Rumford Press, Concord, New Hampshire. The conventional least-squares method (Bartlett 1915) can be
Bhattacharyya, T. & Longiaru, S. 1986. Ability of the Fry method to applied to determine the coefficients A, B and C in the following way.
characterize pressure-solution deformation--Discussion. Tectono- First, equation (A4) is partially differentiated with respect to A, B and
physics 131. 199-200. C and these equations are set to zero (equation A5):
Chilingarian, 1. G. V. & Wolf, K. H. 1975. Diagenesis in sandstones
and compaction. In: Compaction of Coarse-grained Sediments d(~'~ (k2i))/dA = ~ (2 * (A~ + BxiYi + Cy2 - 1) * ~ ) = 0
(edited by Chilingarian, I. G. V. & Wolf, K. H.). Developments in
Sedimentology, 18B. Elsevier, New York, 69 AAA.
Cloos, E. 1947. Oolite deformation in the South Mountain fold, Z X
Maryland. Bull. geol. Soc. Am. 58,843-918. d( (k~i))ldB= (2 * ( A ~ + BxiYi + C~ - 1) * xiYi) = 0 (A5)
Crespi, J. M. 1986. Some guidelines for practical application of Fry's
method of strain analysis. J. Struct. Geol. 8, 799-808. ~ Z
Dunnet, D. 1969. A technique of finite strain analysis using elliptical d( (k~))/dC = (2 * (Ax~ + Bx~y;+ C~ - 1) * ~ ) = 0.
particles. Tectonophysics 7, 117-136.
 Least-squares center-to-center methods 1059

These simplify to (A6): and solve equations (A6)

= (cef+ cdg + bdh - c2h - d2f - beg)
A * ~ (x~) + B* Z (x~iyi) + C* Z (x~iy~i) = Z (x~i) m
(ace + 2bcd c ~ ad 2 bZe)

A*~(~ri)+B*Z(~y2)+C*Z(x.oPi)=~(xiYi) (A6) B= (aeg+bch+cdf-c2g-adh-be't) (AS)

(ace + 2bcd - cs - ad z - b2e)
(ach + bdf + cbg b2h adg)
A* ~ ( ~ ) + B* Z (x/y3) + C* ~ (y~')= ~ (.v/2). C= (ace + 2bcd - ~ - ad z - bZe) "

Let The resulting A, B and C are the coefficients of the least-squares

best-fit ellipse. The ellipticity and inclination of the long axis can be
a = ~ (x)), b = Z (~Yi), c = Z (~)' solved using analytical geometry (Thomas 1967).

d= Z (x/y~), e= Z (y~)' f= Z (~)' (A7)