APPROXIMATION OF THE ORIENTATION DISTRIBUTION OF GRAINS IN POLYCRYSTALLINE SAMPLES BY MEANS OF GAUSSIANS

The orientation distribution function (ODF) obtained by classical spherical harmonics analysis may be falsified by ghost influences as well as series truncation effects. The ghosts are a consequence of the 
inversion symmetry of experimental pole figures which leads to the loss of information on the “odd” 
part of ODF.


INTRODUCTION
The most widely applied method to solve the reproduction problem of the ODF from experimental pole figures is the series expansion formalism proposed by Bunge (1965).In its originally used version it was not able to avoid regions of negative values and ghost effects in the ODF, caused by series termination and the missing of information on the odd part of the ODF (Matthies, 1979).Now, the zero range method (Bunge and Esling, 1979) and more recently the positivity method (Dahms and Bunge, 1989) permit to take into account the odd part of the ODF also.
Another method to eliminate the mentioned phenomena has been proposed by Luecke et al., 1981.They determine the even part of the ODF using the series expansion technique.The true ODF is assumed to be constructed by the minimum number of Gaussians fitted to the spherical harmonics function and takes into account the characteristics of the experimental pole figures.In the present paper a new method for ODF determination from experimental data is proposed.The ODF is represented by Gaussians the parameters of which are fitted in the pole figure space directly using the known analytical relationships between ODFs and pole figures formed by such type of distributions.
The way of calculations is illustrated for cubic lattice symmetry.It can be generalized to other symmetries as well.
The application of the proposed method is demonstrated for the example of the rolling texture of copper.The results are compared with the corresponding data obtained from spherical harmonics analysis.
ODF REPRESENTATION BY 6-FUNCTIONS (DISCRETE CASE)   The solution of the reproduction problem from pole figures of one single crystal is discussed by Matthies (1981) taking into account various symmetries.In this paper the corresponding mathematical formalism has been developed for a polycrystalline sample consisting of a finite number of grains with cubic, hexagonal or orthorhombic lattice symmetry, respectively (T.I. Bukharova and   T. I. Savyolova, 1985; T. I. Savyolova, 1985; D. I. Nikolaev and T. I. Savyolova

1987).
If there is a limited number N of grains in the specimen the corresponding ODF can be represented as a linear combination of 6-functions: N f (g) , aifi(g, gi) (1) i=1 where ai >0 is the size-induced weight factor of the given grain and f is a 6-function with the mass at point gi SO(3).In the case of cubic lattice symmetry the function f(g) in Eq. ( 1) can be shown to be determinable from one pole figure in an unique way.
Let GB {gBi}, 1, 2,..., N be the subgroup of the single crystal symmetry in the SO(3)-space.For cubic lattices Gn consists of 24 elements.
The set of all unit vectors being normal to one of the lattice planes of a selected family {hkl} is named as the vector star {hkl}.The number of unit vectors of a given vector star is equal to the multiplicity.In this way, the {001} vector star consists of 6 vectors (the {111} one of 8 and the {110} vector star of 12 vectors, respectively).Three of the {001} vectors are orthogonal to each other, the others are oppositely directed to the first ones.This vector star corresponds to the {001} pole figure of the single crystal.
Problem A: Let the vector star I be given.Then, all transformations g e SO(3) have to be found in a way, that {g hi} I, i= 1, 2,..., k.
(2) Solution: The transformation g* is determined unambiguously, if the vectors ]/1, ]/2 are chosen from the vector star I as an ordered pair performing the condition (Yl, Y2)= 0 in the case of {001} with g*-h Yl and g*-hE Y2-All other rotations being solutions of problem A are obtained from g* by multiplica- tion from the left side with gBi -GB.
The effect of any rotating operation g* SO(3) on the vectors hi, h2, [hi, h2] is well-known.The mapping of the vectors as a result of the rotating operation g* is: hi (100) g', Yl (yl, y2, y31); h2 (010)g---:-> Y2 (Y21, y2, y3); Y213) These vectors correspond to the traces from equivalent lattice plane normals of a cubic single crystal in the pole figure.Now the desired rotation g* can be written as an orthogonal matrix fYl Y [Y, y2]l g*=lY Yz [Yl, Y2]2J (4) 0' Yz 3 lYe, y2]3/ The determination of g* for other pole figures can be performed in an analogous manner (see appendix).
Problem B: There is a finite set of unit vectors Z corresponding to the chosen kind of vector stars of all grains under discussion and a function r with the property Z R + (5) where R / represents the set of non-negative numbers.The function r corresponds to one of the pole figures related to an ODF as described by Eq. (1).Now a finite set of rotations {gl, g2, gN} e SO(3) and a function R have to be found with {gl, gE,---,gN} g ,R + (6) possessing the following properties: -U{gj.hi} {gj-hib=l,2 N;i=l,2 k} Z, i.e. the set Z represents the con- junction of vector stars corresponding to the N grains of the polycrystalline sample.
-the following relationship is valid for any z e Z R(gi) r(z) (7) {gj-z {gi-hi, i= l,2 k}} Eq. ( 7) permits to determine the weights ai in an ODF as described in Eq. ( 1) by means of the function r(z).This is equivalent to the representation of rotation gj to within the transformation gBi GBi corresponding to the given pole figures for hi, i=1, 2 k.
Solution of problem B: The set Z has to be split up into vector stars/.. From each vector star // an ordered pair of vectors Yli and YEi is selected which corresponds to the vectors hi and h E. The desired rotation gj is unambiguously determined by the conditions yj=gj-!11 and y2j=g !12.The function R is uniquely found from the equation system (7).Every matrix g has the form of formula (4) and is determined in an unambiguous way to within the multiplication from the left to one of the matrices from GB.
Model investigations have been performed concerning the direct and the reverse problem in order to verify the described formalism.From an ODF of type Eq. ( 1) the {001} pole figure has been determined.Then the ODF was reproduced as a superposition of single orientations on the base of the calculated pole figure (D.I. Nikolaev and T. I. Savyolov, 1987).
If more than ten grains have to be taken into account the occurrence of experimental errors can cause any ambiguity of the ODF reproduction task by di-function approximation.In this case f(g) must be constructed by Gaussians in the SO(3)-space as described below.For more detail see Savyolova, 1984 and   1985, Nikolayev and Savyolova, 1986, Parthasarathy, 1964 and Perrin, 1928.APPROXIMATION OF THE ODF BY GAUSSIANS IN THE

SO(3)-SPACE
The main problem of the ODF reproduction from experimental pole figures raises from the lack of information on the odd part of the ODF.Here, the ODF is represented as a superposition of Gaussians.The even part of the desired function should be near to the ODF from the classical harmonics method.The odd component is "added" by the Gaussians.Pole figures which are constructed by such kind of distributions do not contain any information on the odd part.They depend, however, on parameters like weight, peak position and width in the same way as the Gaussian distributions of the rotation group.Therefore, the parameters of the standard distributions can be derived from experimental pole figures directly.
The following algorithm is obtained to determine an ODF fexp(g) from exp experimental pole figures Ph, (Y)- Step 1" The number of maxima in feXp(g) and their coordinates go , k exp 1 2, N has to be found from the maxima of the function Ph, (y) and their spherical coordinates yo(k 1, 2,..., N).
Step 2: The ODF f(g) has to be represented as a superposition of spherical Gaussians (Eq.( 8)) whose positions are at go N f(g) AkG,(g, ek, g) ( 12) where Gk(gko, ek, g) is given by Eq. ( 8) substituting e, by e, t.Taking into account Eq. ( 10) pole figures may be written as N G Pn,(y) AkP(y, ek, go) (13) k=l where P(y, ek, go ) is a function of type Eq. ( 10) and cos(zk) is the scalar product of hi and'g6-Y.
Step 3" The Gaussian parameters A,, e have to be determined (fit to the experimental data).
If there are isolated maxima in the pole figures the number of distributions and their positions can be found using the method described above.Otherwise, additional conditions have to be taken into account such as a minimum of Gaussians or other considerations.Moreover, the weights of single distributions Ak and the widths e are required.
The criterion for the selection of these parameters is the degree of coincidence of experimental and fitted pole figures.In the following two possible approxima- tion methods are discussed.

APPROXIMATION IN EVERY POINT
The Gaussian parameters are determined minimizing the expression N exp max P, (y) , AkP(y, ek, gko) min ( 23) 'es2 k=l The solution of the problem has been found using the coordinate minimization method.The choice of initial values depends on the structure of the given pole figure.The start approximation for the parameters g0 k is found solving the equation [Ph,(Y) Ptt(Y2)] G pG [Phi(Yl, e, go k) hi(Y2, e, gok)] 0 (24) The coordinates Yl and Y2 havi.ng a distance of I or 2 steps of the measuring mesh are taken in the range of a pole figure maximum.The input for A k is the reciprocal number of pole figure maxima.
Thus, the ODF can be reproduced from experimental pole figures in an unique way using the described methods.If there is a small number of grains (N < 8-10).f(g)can be fitted by 6-functions.For larger N spherical Gaussians have to be used to determine an unambiguous ODF.

HARMONICS STUDIES
The application of the proposed method is demonstrated at the rolling texture of copper.Figure 1 shows the pole figures {100}, {111} and {110} being measured at the RFR reactor of the CINR Rossendorf (in the case of {100} and {110} the second order pole figures were recorded).The pole figures were scanned by an equiangular point mesh (steps of 6 degrees).There are no obviously isolated maxima.Therefore the number and position of peaks is not directly recognizable.
For the parameter determination of spherical Gaussians the { 100} pole figure has been analyzed.This pole figure is characterized by a girdle of high pole density denoted by the "skeleton" line in Figure 2a.The circles represent the traces of possible vector stars (ideal orientations) without consideration of specimen symmetry.Figure 2b is derived from Figure 2a taking into account the orthorhombic symmetry of experimental data.A sufficient pole figure fit was attainable assuming N 5 grains as shown in Figure 2c (without specimen symmetry) and in Figure 5d respectively.The orientations g (k 1,..., 5) are determined from the vector stars by the methods described above.Besides the girdle there are maxima (denoted by A in Figure 2d) at ;t :r/2 and tp 0, :r and somewhat increased pole density at q9 :r/2, 3:r/2 refering to the existence of an axial component which is denoted by a dotted line in Figure 2d.This axial component is not visuable in Figure 1.Its intensity is higher than the pole figure background but less than unity.Therefore, the ODF can be written as 5 1 NB =24 fn(g) Ao + AfA(g, flo, Yo, eA) + E A, G,(g, gn, go, e,) (25) k=!

B j=!
The calculation of the pole figures corresponding to the ODF of Eq. ( 25) can be performed by R pA llPG Ph,(Y) Ao + A h,(g, flo, "Yo, eA) + A, h,(Y, e,, yo ) k--!

B
The parameters Q {A0, A, A1,.. As; eA, el e5} are found using the minimum condition for Eq. ( 23).In the given case they were assumed to be constant for all orientations A A2 =...= A5 and El, e2 =...= es.The max- ima have been determined taking into account 30 points situated in the high pole density range.The calculations were performed in an iterative way.Starting with orientations go a parameter set Q is found permitting to correct orientations go k.
The iteration procedure has been stopped when the maximum relative deviation between experimental and fitted pole figures was less than 15%.
The pole figures {100}, {111} and {110} constructed by means of Eq. ( 26) from the fitted parameter set are shown in Figure 3.Because only the { 100} pole figure  ' 2 Scheme of the proposed arrangement of vector stars (ideal orientations) in the {100} pole figure.The solid line describes the maximum pole density range where the intensity is nearly constant in any perpendicular section.The dotted line represents the assumed position of the axial component.Figare 2 (Continued) Scheme of the proposed arrangement of vector stars (ideal orientations) in the {100} pole figure.The solid line describes the maximum pole density range where the intensity is nearly constant in any perpendicular section.The dotted line represents the assumed position of the axial component.The circles label the exit points of the vector stars.(a)some possible vector stars (triclinic sample symmetry); (b)representation of vector stars from (a) taking into account orthorhombic sample symmetry; (c)position of the fitted vector stars (without specimen symmetry); (d)--representation of vector stars from (c) taking into account orthorhombic specimen symmetry.
was used for the fit the comparison of experimental and Gaussian {111} and {110} pole figures may give an idea concerning the efficiency of the proposed method.In Figure 4 the pole figures are shown which are recalculated via spherical harmonics analysis.In Figure 5 constant-x-sections are given at ;t 30 for the experimental {100} pole figure, the recalculated one and the Gaussian fitted pole figure as well.
The ODF calculated by means of Eq. ( 1) is presented in Figure 6.For -1 0 3O 60 90 Figare $ Constant-z-section of {100} pole figures from Figs. 1, 3 and 4 as well.The continuous line corresponds to the experimental pole figure, the medium curve to the spherical harmonics recalculated one and the lower dotted line to the Gaussian fitted pole figure respectively.
comparison Figure 7 shows the ODF determined by spherical harmonics analysis.
The ODF obtained by the proposed method can be seen to be close to the spherical harmonics one from ODF sections for several constant cF2, angles in Figure 8.The even part of the Gaussian ODF deviates from the complete one in the same angular ranges and in the same sense as the spherical harmonics ODF refering to ghost effects.The errors of the proposed method can be estimated by the following characteristics S exp R 2 /;S exp 2 02 (P,, (y)   dy P,(y)) P., (y)) dy=O.O15 exp R IP, (y)-P,(y)I RP 100%/K 16% ,=1 Ie ,7 (y)l Comparing the ODFs from Figures 6 and 7 the maximum of the odd part of the spherical harmonics ODF is 1.01 and the maximum negative value is -0.99.The quality of results obtained by the proposed method is essentially determined by the accuracy of the Gaussian parameter fit.It can be estimated comparing the pole figures which were not used for analysis and the correspond- ing ones reproduced from the fitted parameter set.

CONCLUSION
In the present paper a method is proposed to reproduce the ODF from pole figures by means of a superposition of Gaussian distributions satisfying the 8.00 6.00 4.00 2.00 1.00 Central limit theorem.The application of the new formalism is illustrated for an example of the rolling texture of copper.
The representation of the ODF as a superposition of Gaussians fitted in the pole figure space provides the following characteristics and consequences: -the ODF performs the positivity condition and is free of ghost effects -1 pole figure is required to determine the ODF by the given method for cubic lattice symmetry, 1 or 2 pole figures are necessary for hexagonal lattices, respectively ligare 7 Representation of the ODF in 2-const.-sectionsreproduced by spherical harmonics analysis.
-in the neighbourhood of maxima the ODF is Gaussian shaped if the Central limit theorem is satisfied, i.e. locally the ODF can be considered as a superposition of a large number of random rotations of individual grains.
Up to now, the lack of a simple fully determined approach to find the number and position of Gaussian maxima in the case of overlappings seems to be a drawback of the method.
The circles label the exit points of the vector stars.(a)--somc possible vector stars (triclinic sample symmetry); (b)--rcprcscntation of vector stars from (a) taking into account orthorhombic sample symmetry; (c)position of the fitted vector stars (without specimen symmetry); (d)representation of vector stars from (c) taking into account orthorhombic specimen symmetry.
4 Pole figures recalculated by means of the spherical harmonics method (]max 22).

Figure 6
Figure 6 Representation of the Gaussian fitted ODF in 02 const.-sections.The Euler angles in the text and on the Figs. 6and 7 are related by 01 tr + r/2, Figare 8 (Continued) Constant-go 2, -sections for different angles I02, (in Figure6athe sections are labelled by dotted lines).