• Aucun résultat trouvé

Cubic approximants in quasicrystal structures

N/A
N/A
Protected

Academic year: 2021

Partager "Cubic approximants in quasicrystal structures"

Copied!
17
0
0

Texte intégral

(1)

HAL Id: jpa-00212565

https://hal.archives-ouvertes.fr/jpa-00212565

Submitted on 1 Jan 1990

HAL is a multi-disciplinary open access

archive for the deposit and dissemination of

sci-entific research documents, whether they are

pub-lished or not. The documents may come from

teaching and research institutions in France or

abroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, est

destinée au dépôt et à la diffusion de documents

scientifiques de niveau recherche, publiés ou non,

émanant des établissements d’enseignement et de

recherche français ou étrangers, des laboratoires

publics ou privés.

Cubic approximants in quasicrystal structures

V.E. Dmitrienko

To cite this version:

(2)

Cubic

approximants

in

quasicrystal

structures

V. E. Dmitrienko

Institute of

Crystallography,

117333, Moscow, U.S.S.R.

(Received

7 mai 1990,

accepted

in

final form

7 August

1990)

Abstract. 2014 The

regular

deviations from the exact icosahedral symmetry,

usually

observed at the diffraction patterns of

quasicrystal alloys,

are

analyzed.

It is shown that

shifting, splitting

and

asymmetric broadening

of reflections can be attributed to

crystalline phases

with the cubic

symmetry very close to the icosahedral one

(such pseudo-icosahedral

cubic

approximants

may be called the Fibonacci

crystals).

The Fibonacci

crystal

is labelled

as Fn+1/Fn>,

if in this

crystal

the most intense vertex reflections have the Miller indices

{0,

Fn,

Fn + 1}

where

Fi

are the Fibonacci numbers

(Fi

= 1, 1, 2, 3, 5, 8, 13, 21,

34...).

The deviations of x-ray and electron reflections from their icosahedral

positions

are calculated. The

comparison

with available

experimental

data shows that at least four different Fibonacci

crystals

have been observed in Al-Mn and Al-Mn-Si

alloys :

2/1>

(MnSi

structure), 5/3> (03B1-Al-Mn-Si), 13/8>, and 34/21>

with the lattice constants

4.6

Å,

12.6

Å,

33.1

Å,

86.6 Å respectively.

It is

interesting

to note that there are no

experimental

evidences for the intermediate

approximants 3/2>, 8/5> and 21/13>.

The

possible

space groups of the Fibonacci

crystals

and their

relationships

with

quasicrystallographic

space groups are discussed.

Classification

Physics

Abstracts

61.16 - 61.50e - 61.55H

1. Introduction.

Among

many

problems

concerning quasicrystals

the most

important

ones are

1)

the

characterization of

quasicrystalline

atomic structure and

2)

the nature of the disorder in

quasicrystalline alloys

[1,

2]. Imperfectness

of

quasicrystal

structures is observed as

peak

splitting, shifting

and

asymmetric broadening

in x-ray and electron diffraction

pattern

[1-9J.

In

the icosahedral

glass

model,

the disorder lies in the basis of the

quasicrystalline

structure. In

other theoretical

models,

the

perfect quasicrystalline

structures are

possible

and the

imperfectness

of real

samples

is

supposed

to be a result of frozen

phason

strains

[1,

2, 5,

6].

In the

present

paper another

approach

is

developed.

It is shown that in many cases the

asymmetry

and deviations of the

Bragg

reflections from the icosahedral

positions

can be connected with the presence of

crystalline ,phases

which

approximate

the

quasicrystalline

phase

very

closely.

Below, only

cubic

approximants

are considered in detail

(they

are called

the Fibonacci

crystals).

The sequence of Fibonacci

crystals

with ever

larger

unit cells forms a

chain between conventional

crystals (with

small unit

cells)

and

quasicrystals.

Thus,

above

all,

the Fibonacci

crystals

give

us a convenient

starting point

for the

understanding

of the

quasicrystalline

phase.

(3)

Although

the

possibility of crystal approximants

has been discussed in many papers

[1,10-17],

the solid

arguments

for their real existence are not numerous

(excepting a-Al-Mn-Si).

It

can be shown from the

comparison

between theoretical

predictions

and available

experimental

data that at least four different Fibonacci

crystals

can exist in Al-Mn and Al-Mn-Si

alloys ;

their lattice constants are about 4.6

À (MnSi

type

structures),

12.6

Á

(a-Al-Mn-Si),

33.1 À

[15],

and 86.6

Á (see below). Although

the latter

approximant

looks rather fantastic

(the

unit cell contains more than 4 x

104 atoms !),

its existence seems to be confirmed

by

x-ray and electron diffraction. In other

alloys (Al-Li-Cu, etc.)

the deviations from the icosahedral

symmetry

are also observed

[6-9]

and their

analysis

will be

given

elsewhere. For

two-dimensional

quasicrystals,

the

periodic approximants

have been considered in details earlier in reference

[18],

where the

importance

of such

approximants

was shown.

The remainder of this paper will

proceed

as follows : in section 2 we discuss the

symmetry

of the Fibonacci

crystals

and relations between the

Bragg

reflections in cubic and icosahedral

structures. In sections 3-4 the deviations of cubic reflections from their icosahedral

positions

are calculated.

Then,

in section

5,

the

possible

structure of the Fibonacci

crystals

is illustrated

by

the

simple

low-order

approximant

with 8 atoms per unit cell

(MnSi

structure,

P213

space

group). Finally,

in section 6 the

comparison

of our results with the

experimental

data are

given.

2.

Crystallography

of Fibonacci

crystals.

Both the icosahedral

quasicrystals

and their Fibonacci

approximants

can be obtained from the suitable

projection

of six-dimensional cubic lattice on our three-dimensional space

[10-12,

16,

17,

19].

In this paper we use more

simple approach, considering quasicrystal (or crystal)

as a

superposition

of

density

waves with some set of the

reciprocal

lattice vectors

Q :

where p

(r )

is a

density.

This

approach

is convenient if one focuses on the diffraction

patterns

rather than on the

real-space

structure.

Any reciprocal

lattice vector

Q

of icosahedral

quasilattice

may be written as a linear combination of six basic vectors

Qm :

where the

Nm are integers

and may be

thought

of as the six-dimension Miller indices of the

Bragg

reflections. At first we consider the

primitive

icosahedral lattices. As the basic vectors

we will use six vertex vectors

Qvm

directed toward the icosahedron vertices

(the

fivefold

axes).

We have chosen as

(100 000)

the most intense reflection of this

series,

which for Al-Mn occurs

at

2.9 Â -

and corresponds roughly

to interatomic distances :

d

= 2 w

/

1 Qv.

1

= 2.17

Á.

The

advantage

of this choice is that the most of the intense reflections can be indexed with a small

number of nonzero

Nm [19]. Using

three-dimensional Cartesian

coordinates,

those six basic

vectors

Qvm

may be written as

(4)

basic vectors

Qvm by

six vectors

Qnm

of a cubic lattice with the Miller indices

equal

to the

Fibonacci numbers :

an is the lattice constant. The n-th order

approximant

is convenient to label as

(F n + 1/ F n> ;

its lattice

constant an

increases with n :

The

relationship

between an

and aR was first obtained in reference

[10].

For

example,

in our

notation the well-known a-Al-Mn-Si

crystal

should be labelled

(5/3)

(because

of another choice of basic vectors, in reference

[10]

this

approximant

was

labelled (1/1),

but this

produces

some difficulties in

labelling

of low-order

approximants

with n =

1, 2).

To avoid confusion it should be noted that in the

crystallographic

literature some definite

choice of cubic axes is used

[20] ;

if the space group is Pa3,

then,

as a result of this

choice;

the

first index of hk0 reflections must be even ; to meet the case,

Fn

and

F n +

1 in

equation (4)

must be

permuted

if

Fn +

1 is even

(see

an

example

in Sect.

5).

Using equations (2)

and

(4),

the

h,

k

and f

Miller indices can be

expressed

via

Nm:

For another

indexing

scheme,

that of reference

[21],

the

quasicrystal

reflection

(h /h’,

k/k’,

f /Î’)

corresponds

to the cubic reflection with the Miller indices

hFn-3

+ h’

Fn-2’

kF,,-3 + k’F,,-2

and iF,, -3 + iF,,

-2-The cubic vectors

Qnm

can

approximate

the icosahedral vectors

Qvm

very

closely,

even if the

order n is not very

large.

For

example,

the

angular

deviation from the fivefold axis is about 2° for

(230),

0.75° for

(350),

0.29° for

(580),

and 0.11 ° for

(8,

13,

0).

Using Qnm

as the basic vectors in

equation (2),

we obtain a set of

reciprocal lattice

vectors

with cubic

symmetry.

In

fact,

for cubic lattice any three basic vectors may be

used ; however,

it is convenient to use those six-vector

basis,

because in this case the vectors of cubic lattice

are close to the vectors of

quasicrystal

with the same

Nm

(examples

are

given below).

The

point

symmetry

of cubic lattice is 23 or

m3 (that is,

the

symmetry

is lower than icosahedral

point

symmetry

532 or

532/m).

The

Landau-theory

calculations of the

spatial

structure of Fibonacci

crystals [15]

show that the most

possible

space groups are

Pm3

or

Pa3

(if

both

Fn

and

Fn +

1 are

odd,

the

possible

space groups are

Im3

or

I213).

It is

interesting

to note that the Landau

theory

predicts

the same absolute values of the

strongest

harmonics in

Pm3

and in

Pa3,

whereas the

signs

of p Q

are

predicted

to be different.

Although

the

Landau-theory

approach

is

hardly

suitable for a detail

description

of

crystal

structure, there is a

hope

that the

qualitative

results,

such as

spatial

symmetry

of a cubic

phase

and the

signs

of the

Fourier harmonics

p Q,

are correct.

If n = 1,

the basic vectors

become 111 0}

and

symmetry

of the Fibonacci

crystal

is

Im3m

[17].

(5)

When the

point

symmetry

reduces from icosahedral to

cubic,

the icosahedral

symmetry

axes

(planes)

become

unequal.

As a

result,

the icosahedral reflections may

split

up into several

types

of cubic reflections.

Inequality

of the

symmetry

axes

(planes)

is

especially

evident if we

put

icosahedron into cube

(see

Fig. 1).

Fig.

1. - The

relationships

between symmetry axes in cubic

crystals

and icosahedral

quasicrystals.

The

symbols

of axes and

pseudo-axes (with asterisk)

are evident from the

figure.

It is clear that instead of

thirty

twofold icosahedral axes,

only

six

twofold (100)

axes are

held in cubic

crystal.

The other

twenty-four (,r 2 r 1 >

directions may be

regarded

as twofold

pseudo-axes,

because for these directions the deviations from twofold

symmetry

are not very

large (in

the Fibonacci

crystals

but not in an

arbitrary

cubic

crystal !).

Then,

instead

of twenty

threefold axes, in a Fibonacci

crystal

we have

eight

threefold (111)

axes and twelve threefold

(7» 2 Io>

pseudo-axes.

The twelve

fivefold (lrO>

axes are turned into twelve

pseudo-axes (all

of them are

equivalent). Finally,

instead of fifteen mirror

planes,

we have three mirror

planes

and twelve

pseudo-planes

which are normal to twofold axes and

pseudo-axes

of cubic

crystal

respectively.

The

inequality

between axes and

pseudo-axes

leads to

inequivalence

and

splitting

of reflections. For

example,

one can

easily

obtain from

(2)-(4)

that

thirty

twofold

(110000)

reflections

(directed along

icosahedron

edges) split

up into two

types

of cubic reflections : six

(2

Fn,

0,

0 ?

reflections and

twenty-four

{F n +

1,

Fn,

F n - 1}

reflections which are directed

along

twofold axes and

pseudo-axes (approximately).

The vectors of these reflections form

deformed « icosahedron » in the

reciprocal

space of the Fibonacci

crystal.

The threefold icosahedral reflections

split

up also into the cubic reflections of two

types ;

for

example,

the

( 110 001 )

reflections are turned into

eight

{Fn - , F n -1’ F n - 1}

reflections and twelve

{F n’

Fn - 2,

0}

reflections. Those icosahedral

reflections,

which are

parallel

to mirror

planes,

split

up into cubic reflections of three

types :

twelve of them are

parallel

to mirror

planes

of cubic lattice and the rest 48 reflections are close to

pseudo-planes.

As a

result,

the

(111 000)

reflections

split

up into

{F n + 1 + F n - , F n + , O} ,

{F n + 2, F n - 1, F n - 1 } ,

, and

{2

Fn, Fn,

F n +

t}.

Finally,

an

arbitrary

icosahedral reflection

splits

up into five

types

of cubic reflections

(5

x 24 = 120 reflections in

all).

The

quantitative description

of reflection

splitting

and

shifting

is

given

in the next sections.

At the same

time,

the Fibonacci

crystals

have many common features with

quasicrystals.

The

general

feature of

quasicrystalline

lattices is the inflation of the reflection

wavelengths

with factors T or

r 3[

1, 2].

In the Fibonacci

crystals

the

approximate

inflation takes

place [ 151.

For

example,

in addition to the basic « icosahedron », formed

by thirty

vectors

{F n +

1,

F n,

F n - 1}

and ( 2

Fn,

0,

0 ) ,

we see

(for

rather small

Nm

in

(2))

the sequence of

larger

(6)

(2

Fn +

k,

0,

0 },

where k = ±

1,

±

2,

... ; each « icosahedron » differs from the basic one

by

factors about

r k (the

T inflation).

The Fibonacci

crystal,

of course, contains the smallest

« icosahedron », formed

by

the

vectors (21 1 ) and {200},

(or

very

degenerate

«

icosahed-ron », formed

by

vectors ( 1 1 0) and ( 200 )

of the same

series)

whereas the

reciprocal

space of

quasicrystals

contains

(in

the

general case)

the infinite series of ever smaller icosahedra. On

the other

hand,

the

approximate

T3-inflation

takes

place

for the

pseudo-fivefold

{F n’ F n + l’ O}

reflections: if all

Nm are

not very

large,

one obtain from the basic

{F n’ F n + l’ O}

vectors the

{Fn - 3, F n - 2, O}

reflections,

but neither

{Fn - l’ F n’ O}

nor

{F n -

2,

F n -1,

O}.

However,

in the Fibonacci

crystals

the inflation rules are not

rigorous

and,

for

example,

the

{F n - 3’ F n - 2’ O}

reflections can be obtained for

sufficiently

large

Nm

(but intensity

of reflections is small if

Nm

are

large).

It is

interesting

to note that the

approximate

T

3-inflation

becomes more

pronounced

in the

Fibonacci

crystals

with

Pa3, Im3,

and

I213

space groups because of the extinctions rules and because of the

parity

of the Fibonacci numbers whose

cycle

is three

(odd,

odd,

even,

...).

For

example,

let us consider

Pa3

group ; if n = 2 and the basic reflections are

(210) ,

then the

T3-inflated {850}

reflections are

possible,

whereas the intermediate

reflections (320)

and

{530}’

are forbidden

by

the extinction rules of

Pa3

group

(in

{hkO}

reflections h must be

even).

In the end of this section we discuss the

symmetry

of those Fibonacci

crystals

which

approximate body-centered

and face-centered icosahedral

quasilattices.

In the

body-centered

quasicrystals E

Nm

is even ; in this case, h + k

+ f

is even for all reflections and the Fibonacci

m

crystals

are bcc too. More

intriguing

situation is in the case of the face-centered

quasicrystals

(like

A165Cu2oFe15),

where all

Nm

are of the same

parity :

it follows from

equations (6)

that all Miller indices are even for all reflections of the

corresponding

Fibonacci

crystals.

Thus,

choosing

another cubic unit cell

(twice

smaller in each

direction),

we return to the same

indexing

of the Fibonacci

crystal

as in the case of

primitive

icosahedral lattice

(with

one

important exception :

the inflation factor is now about T for all

types

of

reflections).

3. Déviations of cubic reflections from icosahedral

positions.

In the Fibonacci

crystal,

the deviation

AQ

of an

arbitrary

cubic reflection from its icosahedral

position

can be written as a sum of the deviations

âQnm

of the basic reflections :

where

àQ_ = Qnm - Qvm.

To obtain the

explicit

form of

AQ,,.,

we use the

relationship

between an

and aR

(see

Eq.

(5))

and the

following expression

for

Fn :

After

simple

calculations we obtain for

âQnm :

where m =

1,

(7)

For any Fibonacci

crystal,

dQnm

are normal to the

corresponding

basic vectors

Qvm;

thus,

the deviations of the absolute values of

Qnm

from the absolute value of

are

very small

(of

the order of

A n 2

Qv,,,

This is also valid for any fivefold reflection.

For other

strong

reflections the deviations of

Q

are of the order

of

1 An 1

for weak reflections

(large Nm)

the deviations may be not very small. The

simple

calculations show

(see figures

below)

that both the vertex vectors and the

edge

vectors of the Fibonacci

crystals

fit the

corresponding

icosahedral vectors very

closely.

Therefore,

careful

analysis

of x-ray and electron diffraction

patterns

is needed to

distinguish quasicrystals

and their Fibonacci

approximants.

,

4. Diffraction

patterns

of the Fibonacci

crystals.

To show the main features of the cubic distortions of icosahedral

quasicrystals,

we illustrate

the above

equations

with

computer-simulated pictures.

At

first,

let us consider the x-ray

diffraction on

powder

samples.

In this case,

only

the absolute values of the

reciprocal-lattice

vectors are relevant

(but

not their

directions). Figure

2 shows a

computer

simulation of x-ray

powder

diffraction

patterns

for three Fibonacci

crystals.

The half-width at half-maximum of individual reflection is chosen 0.009

Â-

1 and

quasicrystal

parameter

aR = 4.60 À

(about

those ones observed in real

samples) ; positions

of icosahedral and cubic reflections are

calculated from

equations (2)-(4).

For

simplicity,

the total intensities of all reflections

(before

splitting)

are chosen of the same value and the intensities of the

split

components

of each

reflection are

supposed

to be

proportional

to their

general multiplicity

factors

(24

for

hkf,

12 for

hk0,

8 for hhh and 6 for

h00). Figure

2 shows that in

the (5/3)

approximant

the

splitting

of

strong

reflections is

complete.

In

thé (13/8)

Fibonacci

crystal,

the values of

splittings

are

comparable

with the reflection

widths ;

as a

result,

the reflections are broadened

asymmetrically

and their

apparent

widths are different. For

the (34/21)

Fibonacci

crystal,

the

Fig.

2. -

Computer

simulation of x-ray

powder-diffraction picture

for (5/3),

(13/8),

and

(34/21 )

Fibonacci

crystals.

The

peak positions (when

not

evident)

are marked

by

vertical lines ; for

(8)

splittings

are so small that the reflections are almost

symmetrical

and their shifts from the

icosahedral

positions

are

hardly

observable

(see

below

Fig. 7).

For non-shown intermediate

approximants (8/5) and (21/13),

the

signs

of the

splittings

and, hence,

the

signs

of

asymmetry

are

opposite.

The electron-diffraction

pattern

is shown at

figure

3 for the case of

pseudo-fivefold

symmetry

(here

and below the incident beam is assumed to fall

along -

z direction so that the

x, y and z axes form the

right-hand triad).

This

picture

is universal for the Fibonacci

crystals

of

any

order ;

the

only

difference is on a scale. The scale can be determined from the

positions

of the

strongest

edge

reflections

(2

Fn,

0,

0 }, {Fn + l’ F n’ F n - 1}

which are at about 3

Â-’.

The

positions

of all

spots

in

figure

3 are determined from the

projections

of the

reciprocal-lattice

vectors

hkf

on the

properly

chosen x

and y

axes : xhk p = h and yhk p =

(Tk + f ) / It

is

important

that the most of relevant vectors deviate from their icosahedral

positions

not

only

in the

plane

of the

pattern

but in

off-plane

direction too : Zhkf =

(rf2013A;)/B/1+T;

such

off-plane

deviations can cause a

disappearance

of some weak

spots.

Fig.

3. -

Universal

pseudo-fivefold

electron-diffraction pattern of the Fibonacci

crystals (incident

beam falls

along -

z

direction).

All x-coordinates are

integer :

xhkl = h.

Only

about one fourth of the pattern is

shown ; the

eomplete picture

can be obtained

by

the mirror reflections : x --+ - x and

(or)

y - - y. The scale of the pattern are determined

by

the

positions

of the most intense

edge

reflections

(Q =

3 A - 1 ) ;

for the different Fibonacci

crystals

those

positions

are marked

by

arrows. The Miller indices are shown for outer reflections

only ;

the indices of :inner reflections can be obtained from the

outer ones after

multiplication by

factor T -

i (i

= 1, 2,

3, ...)

and

round-up

to the closest

integer.

The

innermost ten reflections are of the

type (200)

and ( 21 1 ) .

The

approximate

T inflation is

clearly

seen

in the

figure (pentagons

with strongest distortions are marked with solid

lines).

Pseudo-twofold axes are

shown

by

dashed lines.

Several features of the

pattern

should be noted. The reflections are deviated from their

icosahedral

positions

in x-direction

mainly ;

therefore,

the twofold h00 reflections are

deviated at radial

direction,

whereas the

pseudo-twofold

reflections

(shown

with dashed

lines)

(9)

pseudo-twofold axis directed at 72° to the

x-axis).

For

(pseudo-)twofold

reflections the deviations are

usually

smaller than for other reflections with

comparable

1 Q 1.

. Note also the

« zigzag »

character of the deviations

(this

is

clearly

seen at

glancing angle).

The deviations are

larger

for reflections with

small

[ Q

1

and(or)

with

large Nm ; they

are

especially

visible when at least one

of the Miller indices becomes small. The inflation process

(multiplication

of

hkf

by

the factor

T and

round-up)

decreases the distortions of the

spots

from their icosahedral

positions (see,

for

comparison, proposition

4 from Ref.

[16]).

The

(pseudo-)threefold

patterns

are obtained as a

projection

of the

reciprocal-lattice

vectors on any axes which are normal to a

(pseudo-)threefold

axis. These

patterns

can be of three different

types ;

two of them are shown at

figure

4 and the third one is the mirror

reflection of

figure

4a

(the

choice of axes is

displayed

at the

figure).

At the threefold

patterns

the spots

are deviated in

tangential

directions in

zigzag

manner, whereas at the

pseudo-threefold

pattern

the deviations are directed

along

x-axis and all x-coordinates are

integer.

Fig.

4. - Threefold

(a)

and

pseudo-threefold (b)

electron-diffraction patterns. Other outer reflections

can be indexed for

(a) using

60°-rotation

(cyclic permutation

of the indices and

change

of their

signs)

and for

(b) using

the mirror symmetry

(x - -

x and

(or) y - - y).

The indices of inner reflections can be obtained as for the

pseudo-fivefold

pattern. The twofold and

pseudo-twofold

axes are shown with the dashed lines. The third type of the pattern

(the

threefold one with z

Il 111 ])

is the mirror reflection of

(a).

There are also three different

types

of

(pseudo-)twofold

patterns

(see

Fig. 5).

To avoid

confusion,

it should be

emphasized

that the

pseudo-twofold

patterns

(which

correspond

the

incident beam direction

along

the

pseudo-twofold axes)

have the exact twofold

symmetry

because of the inversion

symmetry

of the electron-diffraction

patterns.

The different

types

of

spot

displacements

are observed for the reflections

lying along

symmetry

axes

(radial

displacement)

and

along pseudo-axes (tangential displacement mainly).

The

off-plane

deviations are also

present

both at the

pseudo-twofold

and at the

pseudo-threefold

patterns.

5. A

simple example :

MnSi cubic

crystal.

It was shown earlier that two

approximants

were observed in Al-Mn and Al-Mn-Si

alloys :

(10)

Fig.

5. -

Twofold

(a)

and

pseudo-twofold (b)

electron-diffraction patterns ; the third type of the pattern

(pseudo-twofold)

is the mirror reflection of

(b).

The spots show the

positions

of reflections but not their intensities.

approximant

of this series

should

be

(2/1)

with the space group either

Pm3

or

Pa3

[15] (or

close to

them)

and with the

following

strong

reflections:

{ 210 } , {211}

and

{200}, {231}

and

{400} ;

the lattice constant should be about

T 2

smaller than that of a-Al-Mn-Si.

Turning

to the

crystallographic

handbooks,

we find a

possible

candidate

[22] :

MnSi

cubic

crystal

(a

= 4.6

Â)

with

P213

space group

(close

to

Pa3).

This structure contains

only

eight

atoms per unit cell : Mn in 4a

positions

x, x, x with xMn = 0.138 and Si in 4a

positions

with xs; =

0.846 ;

each Mn atom is surrounded

by

seven Si atoms and vice versa. There are a

lot of

compounds

with this structure :

FeSi, CoGeSi,

AlNi2Si, NaBr03,

AIPt,

Al4GePd5

and

even

AI,Mn,,,Si2; they

may be used for the search of new

quasicrystals.

To demonstrate the hidden icosahedral

symmetry

of MnSi

crystal (both

in the real space and in the diffraction

patterns),

let us consider the idealized structure AB with xA =

1)/4=t

and xB = 1 - xA = 0.845. This structure looks like a

periodic

system

of

interpenetrating

dodecahedra

(see Fig. 6) :

each A atom is at the center of an ideal

dodecahedron and seven vertices of the dodecahedron are

occupied by

B atoms

(and

vice

versa) ;

the closest A-B distance is

/3/(2 )

and the closest A-A

(or

B-B)

distance is

T - 1.

If we do not

distinguish

A and B atoms

(A = B ),

the space group becomes

Pa3

instead

ofP2t3

for

A =1= B)

as

expected

from the Landau

theory [15].

The structure factor

F(hkf)

for such

Pa3

crystal

is

given by

the

following equation

where

f = f (h 2+ k 2+ J2)

is the atomic factor. From this

equation

the

approximate

icosahedral

symmetry

of diffraction

patterns

can be seen

(at

least,

in

high-Q region).

For

example,

the structure factors of the twofold and

pseudo-twofold

reflections are almost

equal

(11)

Fig.

6. -

Idealized MnSi structure. Cube with the inscribed dodecahedron shows the unit-cell dimensions

(conventional

unit cell has another

origin [20]).

Full circles : A-atoms and empty circles :

B-atoms. Atom 1 is at the center of the cube and atom l’ is at the cube

diagonal ;

the rest atoms are at the cube faces.

whereas 1 F (Fi ,

1

Fi Fi - 1 ) 1

8

If 1.

. This is also valid for the threefold and

pseudo-threefold

reflections : if i = 3 k then

and if i =A 3 k then

All these

approximate equalities

become more and more accurate with

growing i

(to

prove

this,

one should remember that

Fi +

1

7-F,

and that the

cycle

of the

parity

of the Fibonacci numbers is three :

odd, odd,

even,

...).

It is

interesting

to note that there is a connection between MnSi structure and more

simple

1 / 1 >

approximant.

The latter

approximant

should have

Im3m

symmetry

if A = B

[17]

or

Pm3m

symmetry

(like FeAl)

if

A =F

B. We can see at

figure

6 that there is a cluster of

eight

atoms which are

arranged

as in

Im3m

cubic structure

(A-atoms

marked with

1, 5, 6,

7 and B-atoms marked with

l’, 5’, 6’, 7’).

The threefold axis of the cluster is directed

along

the

threefold axis of

MnSi,

whereas the fourfold axes of

Im3m

structure are directed

along

pseudo-twofold

axes of MnSi structure. It seems that MnSi structure looks like

microtwinning

of

FeAl-type

clusters ;

it would be very

interesting

to

generalize

this connection between

neighbouring approximants

for the

high-order

case.

Note also that the hidden icosahedral

symmetry

of MnSi

crystal

results in some

physical

properties.

For

example,

the elastic

properties

of MnSi are almost

isotropic (the

ratio of the

shear moduli is very close to unit : 2

c441 (CI I -

C12) = 1.05) ;

then,

the tensor form of the

tensor structure

amplitudes [23]

of MnSi is close to that of

quasicrystals [24].

The direct

proof

shows that there is no

tendency

to the

weakening

of those MnSi reflections

which

correspond

to the forbidden reflections of

nonsymmorphic

icosahedral space groups

[25].

That’s

why

we should conclude that the

nonsymmorphic

MnSi

crystal

is the

approximant

(12)

6. Discussion.

Now we turn to the

high-order

Fibonacci

crystals

((13/8)

and

(34/21 )

and

begin

with discussion of

experimental

evidences in

support

of their existence. The direct evidences of this

sort can be seen at diffraction

patterns.

Most of the x-ray diffraction

patterns

were obtained

from

powder

samples ;

in this case

(see

Fig. 2),

one can

try

to

identify (13/8)

approximant

using

careful

analysis

of the

shape

of reflections. The

only

available

example

of such

analysis

provides

reference

[4],

where reflection with

Q -

3.04

 -

was

fitted to two Gaussian

peaks.

It was shown that the widths of both

peaks

are

equal

to the width of fivefold

peak

(Q

2.894

 - 1) ;

their fitted

positions

are 3.031

Â-

1 and 3.046

A-

I

(those

values are

marked

by

squares at

Fig. 7),

and the ratio

of peak’s

intensities is about

1/3.

Those values are

very close to the theoretical

predictions

for

{16,0,0}

and

{13,8,5}

reflections in

(13/8)

approximant :

3.033

 -

and

3.045

 -

l respectively

(the

ratio

of peak’s

intensities is

expected

to be about

1/4) ;

icosahedral

peak

should be

positioned

at 3.043

A-

1.

In the

pioneering

paper

[3]

the

splitting

of this reflection was also

observed,

but the

intensity

ratio was estimated as

1/1.

In any case, the proper

fitting

of

asymmetrically

broadened reflections

(like

in Ref.

[4])

may

give

decisive

arguments

for the choice’ of

approximant (or

another

phason-strain model).

Fig.

7. - The deviations of cubic reflections from their icosahedral

positions

in different Fibonacci

crystals : (A)

for the

twofold { 2

Fi,

0,

0 }

reflections and

(B)

for the

pseudo-twofold {Fi + h

F i, F i -

i}

reflections. Theoretical values are at the vertices of broken lines :

(34/21) -

dash lines and

13 /8 > -

solid lines.

Experimental

data : circles

[5] -

Mn

implantation

into

single-crystal

Al ; squares

[4]

-

AlnMn22Si6;

rhombi - from

figure

5 of reference

[7] - A186Mn14’

Indices of reflections see at

figure

2. For

(5/3)

approximant (a -Al-Mn-Si)

all deviations

(not-shown)

are about

T 4 times

greater than for

(13/8)

approximant.

For

(34/21)

approximant,

the

powder

diffraction

pattern

is

practically indistinguishable

from

quasicrystal

one

(see

Fig. 2). Fortunately,

in reference

[5]

the very accurate

measure-ments of reflection

positions

were made for an orientated

sample

obtained

by

Mn

implantation

into Al

monocrystal.

The measured deviations of reflections from their

icosahedral

positions (shown

by

circles at

Fig. 7)

are in well accordance with the theoretical

(13)

More obvious

(but

may be less

accurate)

evidences for the Fibonacci

crystals

can be found

at electron diffraction

patterns.

Although

some deviations from icosahedral

symmetry

were

observed in many

works,

for the decisive conclusion about their nature

convergent

beam electron diffraction

[7,

8]

is

preferable.

The size of the area should be less than

(or,

at

least,

of the order

of)

the coherent

length

f coh

of the

positional

order ;

in AIMnSi

quasicrystal alloys

that

length

is about 300-500

 (it

can be estimated from the intrinsic width of x-ray

peaks).

The shifts of reflections at the fivefold

pattern

from reference

[7]

are close to those

expected

for the

(13/8)

approximant (see

discussion in Ref.

[15]);

as an

example,

the shifts of two

reflections are shown

by

rhombi at

figure

7. The deviations observed in electron diffraction

patterns

from

Al78Cr17Rus

alloy [19]

can be also attributed

to (13/8)

approximant : figure

3

Fig.

8. - The

superposition

of the

pseudo-fivefold

patterns from different domains with coherent

orientations ; (a), (b)

and

(c) -

for the same

approximant

with different orientations of twofold axes :

(a)

- 360 and

(b) -

72° between twofold axes in different

domains ; (c) -

five different orientations, i.e. the

polydomain

diffraction area ;

(d) -

for the

F n + 1/ F n >

approximant

in one domain and the

F n + 3/ F n + 2>

approximant

in another with the same direction of the twofold axes. Coordinate axes are

labelled for the

(Fn + 1/ F n >

approximant

with the twofold axis

along

x-direction. Note that the spots are

(14)

from reference

[19]

is very close to the innermost

part

of our

figure

3

for (13/8)

approximant.

Those

pentagons

with the

right angles,

marked at

figure

3,

can be seen at

figure

11.2 of reference

[26].

On the other

hand,

the

convergent

beam electron diffraction

pattern

of an

Al74Mn2oSi6

alloy (Fig.

2 of Ref.

[8])

seems to be close to the

pattern

expected

for the

(34 /21 )

approximant.

Note

also,

that the absence of some weak

spots,

violating

the

symmetry

of

patterns

[8],

may be caused

by

the

off-plane

deviations of

reciprocal

lattice

vectors in cubic

approximants (those

deviations are stronger for weak

spots).

As a

result,

some weak reflections can appear or

disappear

when the direction of incident electron beam is

slightly changed.

The

qualitative

analysis

of electron diffraction

patterns

is rather difficult

because,

even on

the best

patterns,

the

shape

of diffraction

spots

is distorted too : the

spots

are

elongated

in one or in different directions and the faintest

spots

show the

greatest

distortion. Such

picture

can

be

explained

if we

adopt

different orientation of the

symmetry

axes of the Fibonacci

crystals

in different domains

(the

size of the domains is about

icoh)-

It is

interesting

that the small value of

ecoh

in

quasicrystalline alloys

can be also

explained

within this model.

Indeed,

let us

suppose that there is the

phase

transition from

quasicrystal

to its

approximant

(the

transition of this sort seems to be observed in Ref.

[27]).

If the transition is

rapid enough,

then the domains with lower

(cubic)

symmetry

can grow

independently

and their

symmetry

axes may

be orientated

along

different axes of

quasicrystal

matrix. For

example,

some twofold axis of

quasicrystal

can transform either into twofold or into

pseudo-twofold

axis of the Fibonacci

crystal ;

the same is true for any threefold axis. Such coherent orientation of axes and

pseudo-axes is well known for Al-Mn

alloys.

There is no need of visible

sharp boundary

between different domains of

high-order approximants,

because their local structures are similar and

the differences between them are accumulated at the distance of the order of

fcoh-

In this

model,

the natural estimation for

ecoh

is

fcoh -

2

’TT /

1 dQnm 1,

where

àQnm

is the difference between the basic vectors in cubic and in icosahedral lattices

(see

Eq.

(7)) ;

at

that distance the mismatch between

quasicrystal

matrix and its cubic

approximant

riches about one

interatomic

spacing.

The

superposition

of diffraction

patterns

of

differently

orientated domains are shown at

figure

8a, b ;

the close

pairs

of reflections look like

elongated

spots.

If the electron beam size is much more than

fcoh,

then all

possible

orientations of cubic axes with

respect

to those of

Fig.

9. - Threefold

(15)

quasicrystal

are

presented

in the diffraction area and the calculated diffraction patterns for such

polydomain

structure are shown at

figures

8c and 9. Note strange «

triangular »

form of the

spots

which is

frequently

observed in Al-Mn and Al-Li-Cu

alloys

[1,

19,

26] ;

the centers

of the

spots

are almost

exactly

at

correspondent

icosahedral

positions.

Another

type

of the

pattern

(with

spots

elongated

in the same

direction)

may be a result of the same orientation of

axes in domains of the Fibonacci

crystals

of different order

(Fig.

8d).

7. Conclusion.

We have

presented

above the

arguments

for existence of

high

order Fibonacci

approximants

in Al-Mn-Si

alloys (for

other

alloys

and for

low-symmetry approximants

see Ref.

[28]).

This

problem

has

given

rise to many controversies and in conclusion we shall

try

to relieve them. The most

popular

argument

against crystal approximants

with a =

30 Â

is that these

approximants

cannot

explain

the

low-Q

reflections with the

d-spacings

of

8.85(7)

and

5.42(3) Â

[29].

However,

this

argument

seems to be

unacceptable

because in the Fibonaci

crystals

the ratio of these

d-spacings

should be close to

[(Ff+l +Ff+Ff-I)/(Ff+F’f-1 + Fl-2)]o.s

rather than to

Fi /Fi _

1

(see

Fig. 2).

Thus,

within the

experimental

errors, the

d-spacings

of

8.85(7)

and

5.42(3) Â

can be attributed to

pseudo-twofold almost-parallel

reflections (321 ) and {532}

in

the (13/8)

Fibonacci

crystal

with a =

33 À ;

therefore,

this

approximant

cannot be ruled out.

Moreover,

as it was

emphasized

above,

the average

position

of the

partially splitted

cubic reflections is very close

to

corresponding

icosahedral

position

even for

low-Q

reflections

(see Figs.

2,

8c and

9).

In

any case, more careful

fitting

of reflections is desirable.

Now we must

emphasize

that there is no contradiction between the

phason-strain

description

of the distortions of diffraction

patterns

[19, 27]

and our

description.

Indeed,

the

Fibonacci

crystals correspond

to some fixed values of

phason

deformation

[14, 16-18,

30] ;

those values are fixed

owing

to the commensurate character of

crystal

structures.

Speaking

in the

phason-strain language,

we should say

accurately

that the observed shifts of reflections can be caused

by

those

anisotropic phason

strains,

which

correspond (at

least,

approximately)

to (13/8)

Fibonacci

crystal (in

many cases

[4, 7, 19])

or

to ( 34 /21 )

Fibonacci

crystal (in

the

best case

[5]).

The next

controversy

is connected with the

Pauling

model

[31]

(twinned

cubic

crystal

with

a - 23.36

Â).

It should be noted that all

reflections,

which were indexed

by Pauling

at

low-Q

region

(where

the differences between

crystal

and

quasicrystal

reflections are more

pronounced),

can be

easily

indexed

in (13/8>

Fibonacci

crystal

because the ratio of lattice

constants

(33.1 /23.36)

is very close to

J2

(hence,

the ratios of

h2 +

k2

+ £2

are very close to

2).

Moreover,

those

reflections,

which can not be indexed in the

Pauling

model

[31 ],

may be

indexed in

(13/8)

approximant;

for

example,

reflection in

A174Mn2lSi5

with

d-spacing

of

8.98 Â

(or

8.85(7) Â

[29])

can be indexed as

{321}.

Some theoretical

arguments

against crystal approximants

are

given

in reference

[32] ;

it is

shown that the

gain owing

to lock-in terms in free energy is smaller than the loss

arising

from the

gradient

term

(at

least,

for

high-order approximants).

However,

within the model used in

reference

[32],

the Fibonacci

crystals give

no loss in

gradient

term, because the

lengths

of

wave-vectors are

equal

for all basic harmonics

(compare Eq. (3)

and

Eq.

(4)).

Moreover,

even more

complex

model is unstable with

respect

to variations in the

lengths

of icosahedral

wave vectors

[33]

and the

phason

strains may be favorable.

Indeed,

most of harmonics may be

shifted toward lower

gradient

energy when

symmetry

is broken to cubic. The mode

locking

can fix those

phason

strains on the values

corresponding

to the Fibonacci

crystal

of some

order

(more

detail discussion see in Ref.

[34]).

The fixed

sign

of the

phason

strains can

(16)

intermediate

approximants

seem to be absent. It is

unclear, however,

whether

All2Mn

structure

(Im 3

space group, a = 7.47

Á)

can be

regarded

as n = 3 member of this series of

approximants, becausein (3/2)

approximant

the

reflections {320} or (230)

are expected to

be

strong

but those reflections are forbidden in

Im3

structure.

In the

end,

it should be noted that the Fibonacci

crystals

are not

unique

and other

types

of cubic

approximants

can exist. As a curious

(but

not

impossible)

example

we can consider the

(Ln + 1/ Ln)

Lucas

approximants,

where the Lucas numbers

Li

are connected with the

Fibonacci numbers :

Li = Fi +

1 +

Fi -

1

(jL,

=

1, 3, 4, 7, 11, ...,

i =

1, 2, 3,

4, ...).

The dif-fraction

patterns

of the Lucas

approximants

are similar to the

patterns

of the Fibonacci

approximants (zigzag

shifts of

reflections,

asymmetric broadening, etc.),

whereas the values of all those effects are

different ;

the consideration of the Lucas

approximants

will be

given

elsewhere.

Note added

during

revision : .’

In the recent work

[35],

Prof. L.

Pauling

also considers the Fibonacci

approximants

instead of his twinned model. He

interprets

the

experimental

data of reference

[5]

as (13/8)

approximant,

whereas it seems evident from

figure

7

that (34/21)

approximant gives

better

fitting

of the data. In reference

[36],

the series of cubic and non-cubic

approximants

are

constructed

using

the canonical-cell

tiling (the Fn +

1 IF,

approximant

of Ref.

[36]

corresponds

to

the (Fn + 4/ Fn + 3)

approximant

of the

present

paper).

Note also reference

[37],

where the

computer

simulations of x-ray diffraction

patterns

have been

presented

for the Fibonacci cubic

approximants

of different order.

Many approximants

other

than just

cubic

approximants

may occur both in Al-Mn-Si

[1, 28, 38]

and in other

alloys (see

Refs.

[28,

39,

40]

for Al-Li-Cu and Ref.

[41]

for

Al-Cu-Fe).

Such

crystal-quasicrystal polymorphism

seems to be a common

feature of different

alloys

and further studies are needed to

clarify

its

origins.

Acknowledgments.

The author is

grateful

to Drs. V. M.

Kaganer,

E. B.

Loginov,

E. M.

Terentjev

and S. B.

Tochilin for the aid in

computer

calculations and

graphics.

References

[1]

HENLEY C. L., Comments Cond. Matter

Phys.

13

(1987)

59-117.

[2]

JANOT Ch. and DUBOIS J. H., J.

Phys.

F. Met.

Phys.

18

(1988)

2303-2343.

[3]

SHECHTMAN D. and BLECH I. A., Metall. Trans. a 16

(1985)

1005.

[4]

ROBERTSON J. L., MISENHEIMER M. E., MOSS S. C. and BENDERSKY L. A., Acta Metall. 34

(1986)

2177.

[5]

BUDAI J. D., TISCHLER J. Z., HABENSCHUSS A., ICE G. E. and ELSER V.,

Phys.

Rev. Lett. 58

(1987)

2304-2307.

[6]

CHEN H. S., KORTAN A. R. and PARSEY J. M.,

Phys.

Rev. B 38

(1988)

1654.

[7]

TANAKA M., TERAUCHI M., HIRAGA K. and HIROBAYASHI M., Ultramicrosc. 17

(1985)

279-286.

[8]

TANAKA M., TERAUCHI M., SUZUKI S., HIRAGA K. and HIRABAYASHI M., Acta

Cryst.

B 43

(1987)

494-501.

[9]

VECCHIO K. S. and WILLIAMS D. B., Metall. Trans. 19A

(1988)

2875.

[10]

ELSER V. and HENLEY C. L.,

Phys.

Rev. Lett. 55

(1985)

2883.

[11]

MACKAY A. L., Scr. Metall. 20

(1986)

1205.

(17)

[15]

DMITRIENKO V. E., Pis’ma Zh.

Eksp.

Teor. Fiz. 45

(1987)

31 ;

Engl.

translation in : JETP Lett. 45

(1987)

38.

[16]

VERGER-GAUGRY J.-L., J.

Phys.

France 49

(1988)

1867.

[17]

MAI Z., TAO S., ZANG B. and ZENG L., J.

Phys. :

Condens. Matter 1

(1989)

2465.

[18]

ENTIN-WOHLMAN O., KLEMAN M. and PAVLOVITCH A., J.

Phys.

France 49

(1988)

587.

[19]

BANCEL P. A. and HEINEY P. A., J.

Phys. Colloq.

France 47

(1986)

C3-341.

[20]

International Tables for

Crystallography,

Vol. A., ed.

by

T. Hahn

(D.

Reidel

Publishing

Com-pany, Dordrecht,

Holland)

1983.

[21]

CAHN J. W., SHECHTMAN D. and GRATIAS D., J. Mater. Res. 1

(1986)

13.

[22]

VILLARS P. and CALVERT L. D., Pearson’s Handbook of

Crystallographic

Data for Intermetallic

Phases,

Vol. 3

(American Society

of Metals, Metal Park,

Ohio)

1985.

[23]

DMITRIENKO V. E., Acta

Cryst.

A 40

(1984)

89.

[24]

DMITRIENKO V. E., Pis’ma Zh.

Exper.

Teor. Fiz. 50

(1989)

153 ;

Engl.

translation in : JETP Lett.

50

(1989)

171.

[25]

LEVITOV L. S. and RHYNER J., J.

Phys.

France 49

(1988)

1853.

[26]

LUBENSKY T. C., in Introduction to

Quasicrystals,

Ed. M. V. Jaric

(Academic

Press, New

York)

1988, p. 199.

[27]

BANCEL P. A.,

Phys.

Rev. Lett. 63

(1989)

2741.

[28]

SHOEMAKER D. P. and SHOEMAKER C. B., in reference

[26]

p. 1.

[29]

BANCEL P. A., HEINEY P. A., STEPHENS P. W. and GOLDMAN A. I., Nature 319

(1986)

104.

[30]

KRAMER P., Acta Cryst. A 43

(1987)

486.

[31]

PAULING L.,

Phys.

Rev. Lett. 58

(1987)

365 ; Proc. Nat. Acad. Sci. USA 85

(1988)

8376.

[32]

BIHAM O., MUKAMEL D. and SHTRIKMAN S.,

Phys.

Rev. Lett. 56

(1986)

2191 ; in reference

[26],

p. 171.

[33]

GRONLUND L. and MERMIN N. D.,

Phys.

Rev. B 38

(1988)

3699.

[34]

ISHII Y.,

Phys.

Rev. B 39

(1989)

11862.

[35]

PAULING L., Proc. Nat. Acad. Sci. USA 86

(1989)

8595.

[36]

HENLEY C. L., submitted to

Phys.

Rev. B

(1990).

[37]

ISHIHARA K. N., Mater. Sci. Forum 22-24

(1987)

223.

[38]

FOLLSTAEDT D. M. and KNAPP J. A., J. Mater. Res. 4

(1989)

1398.

[39]

AUDIER M. and GUYOT P., Acta Metall. 36 (1988) 1321.

[40]

KNOWLES K. M.,

Quasicrystalline

Materials, Eds. Ch. Janot and J. M. Dubois

(World Scientific,

Singapore)

1988, p. 158.

Références

Documents relatifs

In the Janner-Janssen- Bak description, it consists of three concentric spherical shells of respectively Mn, Al and Al aligned in perpendicular space around the

Approximate icosahedral periodic tilings with pseudo-icosahedral symmetry in reciprocal

2.2 2-DIMENSIONAL NONCRYSTALLOGRAPHIC SYM- METRIES.- In 2 dimensions the noncrystallographic point groups are just the symmetry groups GN of the regular N-gons with N =

The a-phase (long dash) and i-phase (solid) of MnAlSi and the i-phase (short dash) of MnAl are shown... i-phase, we favor it over the

The actual three-dimensional crystal is represented by a three-dimensional cut in a regular six-dimensional periodic crystal with symmetry described by a six-dimensional space

L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des

both models give the same kind of images but at different defocus values. Unfortunately, the experimental defocus is determined a posteriori by comparing through

Thus, if we assume a centrosymmetric structure for Alcufe, any possible superspace group of the modulated structure can be described by a pair, (G,, G~ ), of icosahedral 6D