1978 - Jelínek-Kropáček1978 Article StatisticalProcessingOfAnisotr

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

STATISTICAL PROCESSING OF ANISOTROPY OF MAGNETIC

SUSCEPTIBILITY MEASURED ON GROUPS OF SPECIMENS

D e d i c a t e d to R N D r . J a n P i c h a , CSc., on his 60th B i r t h d a y

WIT JELiNEK

Geofyzika, N.C., Brno*)

S u m m a r y : The theory of multivariate statistical processing of the anisotropy of magnetic


susceptibility, measured on a group of specimens, originating from a single geological body (outcrop,
locality, etc.), is described. The result of the processing is an estimate of the mean normalized tensor
and the estimates of the principal susceptibilities, derived from it, together with the respective
intervals of confidence, and the estimates of the principal directions with the respective regions of
confidence. An anisotropy test for a group of specimens is proposed. The function of the A N S 2 I
computer program employed is briefly described and an example of its output plot is presented.

1. I N T R O D U C T I O N

The study of the anisotropy of susceptibility creates several specific problems resulting from the
tensorial nature of the quantity being investigated. This applies to the process of measuring itself,
as well as the mathematical processing of the results, particularly as regards the application of
methods of mathematical statistics.
The fundamental method for determining the anisotropy of susceptibility in weak magnetic
fields is the A.C. bridge method, see, e.g., [1, 3]. The so-called directional susceptibilities along
suitably chosen directions in the specimen are measured. These are then used to compute the
elements of the tensor of susceptibility, the principal susceptibilities and the principal directions
[I, 2]. For the purposes of the present study it will be sufficient if we restrict ourselves in our
considerations to this fundamental method.
In connection with the development of the bridge for measuring the anisotropy of susceptibility
[3] a detailed theory and method of processing the results of the measurement were elaborated,
encompassing statistical estimates of the accuracy and some statistical tests [4]; the theory is
founded on Hext's statistics of the second-order tensor [6]. In [4] there is also described the
A N I S O 10 computer program which serves the purpose of complex processing of the results of
the anisotropy measurements on the individual specimens.
The solution of the outline problems represents a significant, however, only initial step towards
practical applications. The principal object of interest are not the individual specimens, but rather
groups of specimens from certain geological bodies, localities, etc.
In this paper we shall show how a group of specimens -- and thus simultaneously the geological
object being considered - - can be assigned a certain " m e a n " tensor and how one is able to
assess the accuracy with which it has been determinedi). Apart from this, we shall derive the
anisotropy test for a group of specimens. The theory is again founded on the statistics of a second-
order tensor, however, it has been expanded considerably by using some methods of multivariate

*) Address: Je6n~i 29a, 612 46 Brno.


1) Fisher [5] also solved an analogous problem from the field of vector statistics, see als0 [8].
His method found broad application in geophysics, particularly in palaeomagnetism.

50 Studia geoph, et geod. 22 [1978]


Statistical Processin9 o f Anisotropy o f Magnetic Susceptibility ...

analysis. The function of the ANS21 computer program, which was written for this purpose,
will also be briefly described.

2. SOME F U N D A M E N T A L T E R M S A N D NOTES C O N C E R N I N G T H E S Y M B O L S

The relation between the intensity of the magnetic field /-/ and the magnetic polarization ]
induced in a linear anisotropic medium can be expressed, in a Cartesian co-ordinate system, by
the matrix equation

Ja Lk3t k32 k33J H3


briefly ] - - ¢zqkH.
The symmetric matrix k is the susceptibility tensor expressed in matrix form; this matrix is
also sometimes briefly called the susceptibility tensor. Alternatively we shall also express the
susceptibility in the form of the column matrix

(2) k = [kll k22 k33 k12 k2a k31]'.


We shall denote the e igenvalues of the matrix k by ~1, ~2, x3, and the eigenvectors 0t, P2, Pa;
we call them the principal susceptibilities and principal directions (of the tensor) of susceptibility.
As is known, these quantities satisfy the equation

(3) kpi ~ XiPi, i = 1, 2, 3 .


The mean susceptibility is then x,
l 1
(4) ~t ~ ~-(x1 -~- ~'2 -~ x3) ~ 3-(kll 2y k22 ~_ k33) .
This brief review already indicates some of the general principles governing the symbols. All
vectors (in the geometric sense of the word) will be expressed by column matrice s . All column
matrices, regardless of whether they express a vector, will be printed in semibold italics, all other
matrices in semibold stand-up type.
Unless stated otherwise, we shall assume that all the quantities are expressed in the selected
fundamental co,ordinate system, either' the geographic or tectonic system, ref. Section 8. If we
require the quantities to be expressed in the Cartesian co-ordinate system determined b y the
vectors of the principal directions, we shall use the samesymbol to which we shah add the superior
letter P to the right. For example, the tensor of susceptibility expressed by a column matrix in this
co-ordinate system will read kl':
(5) kP ~ [~$ ~2 ~3 0 0 0l t ,
Finally, we should like to draw attention t o the very important convention which is being
introduced in the subsequent Section 3.

3. M E A N N O R M A L I Z E D T E N S O R

W e a r e c o n s i d e r i n g a c e r t a i n g e o l o g i c a l o b j e c t ( b o d y , locality, etc'). T h e f a m i l y
o f all s p e c i m e n s , w h i c h we c o u l d c o l l e c t f r o m this object, m a y be c o n s i d e r e d , f o r p r a c -
tical p u r p o s e s , to be infinitely large. O u r p r o b l e m ,will be to c h o o s e s u i t a b l e statistical

Studia geoph, et geod. 22 [1978] 5|


V. Jelinek

parameters which would characterize this family and which could be estimated from
the measurements of the anisotropy of magnetic susceptibility on N randomly
selected specimens, i.e. on a statistical sample 1) of size N.
We shall use the symbol qk to denote the tensor of susceptibility of the randomly
chosen sample, qxi and qPl its principal susceptibilities and unit vectors of the prin-
cipal directions, respectively. In this connection the superfix q only has a distin-
guishing function and cannot be assigned a numerical value.
One possibility is to consider the mean values of the principal susceptibilities
E(q~i) and mean values of the principal directions E(qpi)to be the resultant character-
istics. Fisher's statistics [5, 8] may be applied immediately to estimating the principal
directions.
The outlined ]grocedure has a certain amount of attraction, particularly for those
who deal with applications. It is illustrative and easy to effect. Nevertheless, one can
have a number of objections to it: the resultant characteristics do not have a simple
physical meaning; the estimated resultan t principal directions are not orthogonal;
the regions of confidence of the resultant principal directions on a unit sphere are
circles and, therefore, cannot express the directional character of the uncertainty in
determining the principal directions. If the sample being investigated contains an
approximately isotropic specimen, the principal directions of this specimen will be
poorly defined; one such specimen may then severely affect the estimates "of the
resultant principal directions.
It is much more natural to consider the mean tensor to be the resultant charac-
teristic,
(6) k = E(qk),

i.e. the tensor, the components of which are

(7) kij = E(qk,j); i,j = 1,2,3,


and the principal susceptibilities and directions, derived from this tensor. The
physical meaning of the tensor k is evident. Roughly speaking, it is the tensor which
we would obtain, if we were able to measure the geological object as a whole.
In studying anisotropy for the purposes of structural geology we have to obtain
the resultant principal directions a n d ratios of the principal susceptibilities defined
as best as possible. In this connection we do not study the fluctuation of the mean
susceptibility of the specimens. If we use the method of the mean tensor, the varia-
bility of the mean susceptibility will affect the variability of the results negatively.
This phenomenon can be observed particularly well in cases where the geological
object being considered contains several rock types which do not differ very much
in their principal directions and ratios of principal susceptibilities, which, however,
do differ substantially in their mean susceptibilities. The same situation may also

1) Under the term "specimen" a piece of rock suitable for measurement is understood. The
term "sample" means a group of specimens taken randomly.

52 $tudia geoph, et geod. 22 [1978]


Statistical Processing o f Anisotropy o f Magnetic Susceptibility ...

occur within a single rock type, if its susceptibility is caused by ferromagnetic grains
distributed sparcely. In this case the mean susceptibility will fluctuate severely
depending on the number of grains contained in the individual specimens, whereas
the variability of the principal directions and of the ratios of the principal suscepti-
bilities may be considerably lower.
The method of eliminating the effect of the fluctuating mean susceptibility is
founded on normalizing the susceptibility tensors of the individual specimens by
means of their mean susceptibilities. The normalized tensor of a general specimen
will then read
(8) qk . . . . = q ~ - i qk.

The mean normalized tensor,


(9) k ..... = E( qk . . . . )
will then be considered to be the resulting characteristic of the fundamental set, as
well as the principal susceptibilities and principal directions derived from it.
Since this paper is the result of the requirements of structurally geological applica-
tions, we decided to prefer the mean normalized tensor. However, we should like to
point out that most considerations may also be applied to unnormalized tensors.
In this respect it would not be very difficult to amend the computer program ac-
cordingly.
In connection with this selection we shall now establish an important convention
concerning the symbols. Since we are going to operate exclusively with normalized
tensors, we shall, as a rule, omit the index "norm", as well as the adjective "normal-
ized". The simplified record of Eq. (9) for the normalized mean tensor will then read

(10) k = E(qk).

The principal susceptibilities and the principal directions, derived from the normal-
ized mean tensor, will be denoted by n,, n=, Za and p,, Pa, Pa, respectively.

4. ESTIMATE OF THE N O R M A L I Z E D M E A N TENSOR A N D ITS VARIABILITY

Consider qk to represent a column matrix expressing the normalized tensor of


susceptibility of the specimen, and consider k to represent the column matrix expres-
sing tl~e normalized mean tensor. With a view to Eq. (10) it is evidently true that
(11) k = E(qk).

We shall assume that qk has a normal distribution with the covariance matrix V;
we shall briefly record its distribution as N ( k , V). The covariance matrix is of the
(6 x 6)-type, its elements are determined by equation
(12) Vis = E[( q k , - k i ) ( q k j - kj)]; i , j = 1,2 . . . . . 6 .

Studia geoph, et geod. 22 [1978] 53


V. Jelinek

We shall assume that we have at our disposal a sample formed by N specimens,


k, 2k, ..., Nk being the column matrices, expressing the susceptibility of the individual
specimens. The unbiased estimate ~¢ of the covariance matrix V is given by
N
(13) l?,j = (N - 1)-' E(qk,- ~,)(qki - [~j).
q=l

The unbiased estimate k of the mean matrix k can be computed as the arithmetic
mean
N
(14) l{ = N - ' Z " k "
q=l

This estimate will display a normal distribution N(k,N-1V), the estimate o f the
covariance matrix N - 1 V can be determined according to Eq. (13).
It should be noted that as a result of the normalizing the elements of the matrix qk
are subject to the linear condition

(15) qkx + qk2 + ~ka = 3.

The matrices k and/{ then have analogous properties. This fact will not be reflected
particularly in the subsequent considerations. An exception, however, is Section 7.
As regards matrix V, we assume that its elements are not limited by any relations
except those which follow from the symmetry and conditions (15).
That which h a s s 0 far been said implies that the problem of estimating the mean
tensor and its variability belongs to the field of multivariate analysis. However, our
original endeavour was to make do with the means of univariate analysis. We
assumed that the covariance matrix can be expressed in the form aV o, where V o is
a constant matrix of a special type, invariant with respect to the orthogonat trans-
formation of co-ordinates, and a is the only quantity which has to be estimated, in
the same way as in measuring specimens i n t h e rotatable system of directions [4, 6].
Practical tests have shown, however, (see also Section 6) that this simplified approach
in many cases does not allow the complicated behaviour of the actual rock material
to be expressed.

5. V A R I A B I L I T Y O F T H E P R I N C I P A L S U S C E P T I B I L I T I E S
AND PRINCIPAL DIRECTIONS

The estimate of the mean tensor, expressed by the column matrix, will read

(16) /c = k + d k .

The column matrix dk represents the variation of the estimated matrix k. The cova-
riance matrix N - ' V of the elements of the matrix k is simultaneously the covariance
matrix of the elements of the matrix dk.

54 Studia geoph, et geod. 22 [1978]


Statistical Processing of .4nisotropy of Magnetic Susceptibility ...

We shall express the estimates of the principal susceptibilities and principal direc-
tions analogously:
(17) ~i=n~+dni, b,=P'+ dp,.
We shall now try to express the variations dxi and dpl as functions of dk; however,
we shall restrict ourselves to linear approximation.
In order to simplify the issue as far as possible, we shall go over to a Cartesian
system of co-ordinates determined by the vectors of the principal directions P1, P2, P3-
The quantities referred to this system will be denoted by the superfix P. Equation
(16) will then become
(18) itP = k P + dk P,
where
(19) k ' ~- [~1 X2 X3 0 0 0 ] ' .
The covariance matrix N - xVP of the elements of the column matrix k P and simul-
taneously of the column matrix dk P can be determined by means of the transfor-
mation relation for VP:
(20) VP: TVT',
(21) T=
=- P~a Pz2t P~I 2pttP21 2pElP31 2p31Pll -
p22 p22 p22 2p,2P2z 2p22P32
2p32Pt2
p23 p223 pa2a 2p13P23 2p23P33 2p33P13
PllPI2 PzlP22 P31P32 PllP22 + P21P12 P21P32 + P31P22 P31P12 + PllP32
Pt2P;3 P22P23 P32P33 P12P23 + Pz2P~3 P22Pa3 + P32P23 Pa2PI3 + P12Pa3
_PlaP11 P23P21 P33P31 Pl,3P21 + P23Pll P23P31 + Pa3P21 P33Pll + P13P31
In Eq. (21) Pij represent the i-th component of the j-th principal direction. See also
[6].
For formal reasons we shall now express the matrix dk P in the form d k P =
= [ d k ~ dk~z dk~3 dk~2 dk~a dkP1]. For the time being let us assume that only
one of its elements is non-zero. Assume that this is one of the elements dk~ (i = 1, 2,
3). In this case clearly
(22) dx, = dk~i
and all the other variations will be zero. Now, assume that one of the elements
dkij (i, j = 1, 2, 3; i * k) differs from zero and that Idk, t < Under this -

assumption we can prove, see also [6], that the non-zero element will cause the system
of principal directions to rotate around the vector Pk in the sense from vector p~
to vector pj by a small angle dk~i/(,~ , - ~:i) (The indices i,j, k differ from one
another). This rotation can be interpreted as the variation of the vectors p~ and pj:
(23) dPi = (x,- ,nj)-' P j , dpj
d.k,.i# (~j _ Xi) -1 dk,eij#i, .

Studia geoph, et geod. 22 [1978] 55


V. Jelinek

The variation of the remaining vector Pk will be zero and the variations of the prin-
cipal susceptibilities will be zero under the approximation being considered.
Now assume that the elements of the matrix dk P can be non-zero simultaneously.
If these variations are sufficiently small, their effects will become linearly super-
imposed. We emphasize that this is a question of the variations of the estimate of the
mean tensor and not of the variations of the individual tensors being considered.
A more detailed analysis would show that the variations can be considered very small,
provided they are much smaller than the differences between the principal suscep-
tibilities. For our further considerations we shall assume that this condition is satis-
fied.
Equation (22) remains valid, however, Eq. (23) has to be supplemented as follows:

(24) dpi = (zi - xj) -1 dk~spj + ( ~ - ~k) -1 dk~kPk"

The vector dp~ lies in plane ¢i which is tangential to a unit sphere at the terminating
point Oi of the vector Pi (Fig. 1). We shall introduce a Cartesian system of co-ordi-

l , , ltz
Joi do:, 1
Fig. 1. Concerning the interpretation of the variability of the principal directions.

nates with its origin at point Oi and with its axes parallel to the vectors pj and ~k
in the plane ¢i. We shall take dpji and dPk i to represent the components of the
vector dPi in this system. Clearly,
(25) dpji = (~i - xj) -1 d k i j , dpkl = (xi - ~¢k)-1 dkik.
As regards the variance of the estimates of t h e principal susceptibilities, Eq. (22)
implies that
(26) a2(~i) = 0"2(d~¢i) = C o v (kPi, k~i) = N -1V~ .

56 Studia geoph, et geod. 22 [1978]


Statistical Processing of Anisotropy of Magnetic Susceptibility ...

Drawing on Eq. (25) we are able to determine the covariance matrix N - a W i of the
random variables dpj, and dpki,

(27) N xW i =
Cov (dk/j, dkij) Cov (dkii, dkik) 1 =
-

Cov (dkij, dkik) Cov (dk,k, dkik) I

=N-1 i( VP+i'3~j)2
+i
~'3+,,3+k l
(~,-
V3P+ i,3 +k
,P
V~+k,3+k

assuming that the indices i, j, k form a cyclic permutation of the numbers 1, 2, 3.


1
We shall now determine the limits of the variabilities of the principal susceptibilities
and principal directions.
We shall construct the statistic
(28) u = (~, - ,,,)/~(~,).
This statistic has a standard normal distribution, so that the interval which will
eliminate 100~ ~o of the least probable estimates of the principal susceptibilities
will read
(29) ~, ___tr(~,)uo_,/=),
where uta-~t]2) is a 100(1 - e/2) ~ quantile of the standard normal distribution.
The quadratic form
(30) Q2 = N[dpji dpk,] W - ' rdpjil
Ldp~d
has a cbi-squared distribution on two degrees of freedom [6]. Assume that •2;(12 --~)
represents the 100(1 - ~) ~o quantile of this distribution. Drawing on (30) an ellipse
~i can be constructed in the plane Qi, which will eliminate (100 - ~) %o of the least
probable positions of the vector p,. The ellipse is defined by the equation
(31) Q2 2
= ~2;(100-~t) "

Denote by 2,., and [Ct,.,C2mi]' the eigenvalues and eigenvectors (column matrices),
respectively, of the matrix W i (m = 1, 2). The lengths of the semi-axes of the
ellipse e~ will be
(32) /(N-'2
N/~ 2 --~)),
miZ2;(lOO
and they will be located along the directions

(33) ClmlPj ~- C2miPk .

Stt~dia geoph, et geod. 22 [1978] 57


V. Jelinek

We shall project the ellipse d°s centrally onto a u n i t sphere; in this way we arrive at
the region in which (100 - e) % of the terminating points of the vector ~ , the esti-
mate of the vector of the principal direction p~, are located.

6. T H E I N T E R V A L S O F C O N F I D E N C E O F T H E P R I N C I P A L
SUSCEPTIBILITIES AND THE REGIONS OF CONFIDENCE
OF THE PRINCIPAL DIRECTIONS

Under practical conditions we have no knowledge of the susceptibility matrix k


or of the covariance matrix of its elements ¥. We are, therefore, only able to draw
o n the corresponding estimates k and V.
The estimates of the principal susceptibilities ~ , ~2, ~3 and the estimates of the
principal directions Pl, P2, P3 are determined as the eigenvalues and eigenvectors
of the estimated square matrix ~, expressing the tensor of susceptibility. In a first
approximation these estimates are unbiased and they have the statistical properties
mentioned in Section 5.
The estimate of the variance of the principal susceptibilities is defined by an equa-
tion analogous to Eq. (26):
(34) s2(~,) = N-'^PV,,.
The statistic t = (~, - ,,)/s(~i)will have a t,distribution on N - 1 degrees of
freedom. This implies that the (100 - ~) % interval of confidence for the principal
susceptibility xi will be
(35) z, ± s(~,)tN_,;(,oo_,/2).
For the time being let us assume that we know the real principal directions p~,
P2, P3 and that we only have the estimate W available for the matrix W. We shall
establish the quadratic form

(36) T2 = N[dpji dpki ] W - 1 Fdpjil


Ldpkd "

The quantity T 2 has the so-called Hotelling distribution (T2-distribution) [7]. The
expression ½ ( N - 2)(N 1) "1 T 2 then has an F-distribution on 2 and N - 2
degrees of freedom. The ellipse gl in plane ~ which will eliminate ~% of the least
probable positions of the vector Pi will be determined by the equation

(37) T ~ = Tz ,

where Toz = 2(N-- 1)(N -- 2 ) - ' Fz,/q_2;(l_~). We shall now denote by 2ml the
eigenvalues of the matrix W , , and by [Clm, C2mJ' its eigenvectors (m = 1, 2). The
lengths of the semi-axes of the ellipse 8~ will then be

(38) erm, = 4{2(N- 1)[N(N- 2 ) ] - ' Fz,N_z;(,_,)2=, }

58 Studia geoph, et geod. 22 [!978]


Statistical Processing o f Anisotropy o f Magnetic Susceptibility ...

and they will be located along the directions

(39) ClmiPi + C2md~k.


NOW consider the ellipse gi, which will lie in plane 0~, perpendicular to the vector p~,
and will pass through its end point. Assume the ellipse to have its centre at this end
point, the lengths of its semi-axes being determined by (33) and their directions being

(40) C,m,b.i + C2,.iPk.


The ellipse gl can be considered to be an approximate (100 - ~) ~o ellipse of con-
fidence of the i-th principal direction. If we project it centrally onto a unit sphere,
we arrive at the approximate 100(1 - ~) 70 region of confidence of the end point
of vector p~. The lengths of the semi-axes of the region of confidence will be denoted
by eml and called confidence angles:

(41) emi = tan -1 eTm i .

It should be pointed out that, as a result of the adopted simplifications, the intervals
and regions of confidence are not exact. Their determination will be the more accurate,
the better the condition of small variations, formulated in Section 5, is satisfied. One
can also say that the interval and region of confidence for the i-th principal direction
will be determined the better, the smaller the corresponding angles of confidence. In
order to arrive at usable estimates, the angles of confidence must not exceed a certain
limit, say about 25 °.

7. ANISOTROPY TEST

The anisotropy test is designed to answer the question whether we are justified
in considering the mean normalized tensor anisotropic on the evidence of the sample
of specimens measured. In other words the test is to prove whether the estimated
normalized mean tensor differs significantly from the isotropic tensor under the
determined variability of the sample. Since the variability is described by the
covariance matrix and that it cannot be fully characterized by a single number, the
test must be designed as multivariate.
Consider the susceptibility tensor qk of a specimen selected at random. As a result
of the normalization, the diagonal components are subject to the linear condition
~k~l + qk22 -~- qk33 = 3; the mean tensor k and its estimate ft display corresponding
properties. We, therefore, have to adopt a five-dimensional distribution, such that
its variables will not be subject to any conditions. This can be done by omitting the
variable kaa as superfluous from our considerations, and the remaining x~ariables are
then trasformed in a way suitable for the test. We shall seek to determine a linear
transformation which will transform the matrix k = [ k l l k22 k12 k23 k31 ] ' into
matrix x = Ix1 x2 x3 x4 xs]', the matrix x having the property that it will become

Studia geoph, et geod. 22 [1978] 59


11. J e l i n e k

zero if the tensor is isotropic. There is an infinite number of such transformations;


here, we shall adopt the transformation

(44) xl = k 1 2 , x2 = k 2 3 , x 3 = k31,

x, = (k, + - 2), x5 = ½(k,, - k2 ).

The covariance matrix of the random variables qxl, q x 2 . . . . , ax 5 will be denoted by U


and its estimate by 0 .
Transformation (44) was not chosen arbitrarily. If the covariance matrix had the
"rotatable" property [6], matrix 13 would be a numerical multiple of the unit matrix.
It was indeed the estimates of matrix 13 that showed the assumption of the "rotatable"
properties not to be even approximately satisfied (ref. note at the end of Section 3).
We shall now assume that we have at our disposal a sample of size N _> 6. As the
zero hypothesis we shall consider the statement that the tensor k is isotropic, i.e.
that the mean value of the matrix q× is zero.
We shall again use Hotelling's statistic T 2 for testing [7]:

(45) T 2 = Nx'13-1x .

We shall reject the zero hypothesis at the level of significance e, provided


(46) T 2 > Toz ,

where To = 5(N - 1)(N - 5) -1 Fs,N_5;(~_,), and we shall adopt the alternative


hypothesis that the mean tensor is anisotropic.
The expression T2/T~ represents a certain index of the precision of determination
of the tensor of anisotropy and it can be used as an auxiliary parameter in interpreting
the results.
The anisotropy test described, as opposed to the considerations in Section 6, does
not require any simplifying assumptions and, therefore, is exact.

8. C O M P U T E R P R O G R A M

To facilitate the application of the described statistical theory, the ANS21 computer program
was written. The program was written in F O R T R A N IV and its implementation is designed for
the medium E M R 6070 type computer. A detailed description of the program would exceed the
scope of this paper and, therefore, we shall restrict ourselves just to general information about
its function and give an example of its graphical output.
The punch cards with the input data for the ANS2I program are obtained from the processing
of the initial results of the measurements on the individual specimens by means of the A N I S O 10
program. Each punch card contains data on one specimen, i.e. (1) specimen label, (2) mean sus-
ceptibility, (3) measuring error, (4) components of the normalized susceptibility tensor, (5)
co-ordinate system Syfnbol in which the tensor is expressed.
In this connection an explanation is required~ regarding the co-ordinafe systems being used
(ref. also to [4]). The basic computation of the anisotropy of the individual specimen by means
of the A N I S O 10 program is carried out in the co-ordinate system of the specimen, i.e. in the

60 Studia geoph, et geod. 22 [1978]


Statistical Processing of Anisotropy of Magnetic Susceptibility ...

system the axes of which are parallel with the edges of the specimen if this is of cubic shape. In
order to be able to process the results obtained from the individual specimen simultaneously,
the tensors are expressed in a suitable c o m m o n co-ordinate system, i.e. either the geographic, or,
in some cases, the tectonic system. The axes of the geographic system correspond to the north,
east and vertical directions, and the axes of the tectonic system are identical at each point of the
geological body with the tectonic axes cross. The A N I S O 10 program is capable of punching
cards in all the systems mentioned, however, if the ANS21 program is to be used only the geo-
graphic and tectonic systems are considered.
The ANS21 program reprints the input data, computes the estimate k of the mean normalized
tensor, the estimates V, 0 , ~l, of the covariance matrices, the estimates "~i of the principal sus-
ceptibilities and their standard deviations, the estimates p~ of the principal directions, the angles
of confidence and the regions of confidence of the principal directions. The estimates of the
principal directions are determined by spherical co-ordinates (declination and inclination), and
the regions of confidence by the spherical co-ordinates of the end points of their semi-axes.
A digital output has the disadvantage of not being very illustrative. For this reason the program
also has a graphical output, an example of which is shown in Fig. 2. The principal directions of
the individual specimens (the smaller symbols) and the resulting principal directions (larger

SSV 2 2N

a • • ee

Fig. 2. Example of the graphical output of the ANS21 computer Program: treatment of the
andesite locality in the Stiavnick6 pohorie (Mrs.).

symbols) are plotted in Lambert's equal-area projection, the principal direction corresponding
to the maximum, intermediate and m i n i m u m susceptibilities being indicated by a triangle,
square a n d circle, respectively. The boundaries of the regions of confidence are also plotted;
the end points of the semi-axes of the regions of confidence are marked with small crosses.
The example shown presents the treatment of an andesite locality in the Stiavnick6 p0horie
(Mts.). We intentionally chose an example with highly scattered principal directions of the in-
dividual specimen in order to demonstrate that, even in such difficult cases, assuming a suf-
ficient n u m b e r of specimens is available, acceptably defined principal directions can be obtained.

Studia geoph, et geod. 22 [1978] 6]


V. Jelinek: Statistical Processing o f Anisotropy of Magnetic Susceptibility...

Since the regions of confidence are relatively large, they only have an approximate character.
The mean direction of the maximum susceptibility agrees with the orientation of the lava flow
structures at this locality, which were determined geologically.

9, CONCLUSION

We have presented the fundamental features of the method of statistical treatment


of the anisotropy of magnetic susceptibility measured on a group of specimen
originating from a single geological object. We have shown how the anisotropy of
magnetic susceptibility of the object being considered can be characterized as a whole.
The proposed method respresents a first attempt at solving this problem and one
cannot, of course, exclude the possibility that more suitable methods will be found
in future. Nevertheless, our present experience indicates that the method, together
with the computer program, have proved themselves and considerably facilitated the
structurally geological applications of the anisotropy of magnetic susceptibility.
We expect that this method will also be useful in other fields, in which symmetric
tensors of the 2nd order are employed.
Acknowledgements: The author would like to thank Mrs. M. S k o r k o v s k ~ i for her co-operation
in writing the computer program and Dr. F. H r o u d a for the example of the processing which
was used to illustrate the function of the computer program.

Received 22. 10. 1976 Reviewer: V. Kropd~ek

References

[1] R. W. G i r d l e r : The Measurements and Computation of Anisotropy of Magnetic Suscepti-


bility of Rocks. Geoph. journ. R.A.S., 5 (1961), 34.
[2] F. Jan~.k: Determination of Anisotropy of Magnetic Susceptibility of Rocks. Studia geoph.
et geod., 9 (1965), 290.
[3] V. J e l i n e k : Precision AC Bridge Set for Measuring Magnetic Susceptibility of Rocks and
]its Anisotropy. Studia geoph, et geod., 17 (1973), 36.
[4] V. J e l i n e k : The Statistical Theory of Measuring Anisotropy of Magnetic Susceptibility
of Rocks and Its Application. Geofyzika, n. p., Brno 1976. (Not published -- can be obtained
from Geofyzika, N. C.).
[5] R. A. F i s h e r : Dispersion on a Sphere. Proc. Roy. Soc. Lond., A217 (1953), 295.
[6] G. R. H e x t : The Estimation of Second-order Tensors, with Related Tests and Designs.
Biometrika, 50 (1963), 353.
[7] T. W. A n d e r s o n : An Introduction to Multivariate Statistical Analysis. J. Wiley and Sons,
New York, Chapman and Hall, London 1958.
[8] K. V. M a r d i a : Statistics of Directional Data. Acad. Press, London and New York 1972.
[9] F. H r o u d a , O. K o l o f l k o v ~ i , V. K0ne~n3~" Magnetic Fabric in Rocks. To be published
in the bull. ,,Z~ipadn6 Karpaty" (edited by the Slovak Academy of Sciences).

62 Studia geoph, et geod, 22 [1978]

You might also like