• Aucun résultat trouvé

Dislocations and disvections in aperiodic crystals

N/A
N/A
Protected

Academic year: 2021

Partager "Dislocations and disvections in aperiodic crystals"

Copied!
20
0
0

Texte intégral

(1)

HAL Id: jpa-00246464

https://hal.archives-ouvertes.fr/jpa-00246464

Submitted on 1 Jan 1992

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.

Dislocations and disvections in aperiodic crystals

Maurice Kléman

To cite this version:

Maurice Kléman. Dislocations and disvections in aperiodic crystals. Journal de Physique I, EDP

Sciences, 1992, 2 (1), pp.69-87. �10.1051/jp1:1992124�. �jpa-00246464�

(2)

Classification

Physics

Abstracts

61.42 61.70

Dislocations and disvections in aperiodic crystals

Maudce Kl£man

Laboratoke de

Physique

des Solides (Associ£ au C-N-R-S-), Universit£ de Paris-Sud, Bit. 510, 91405

Orsay

Cedex, France

(Received 25

July

J99J,

accepted

4 October J99J)

Abstract. The search for

topological

invariants of dislocations in a

periodic

crystal

employs mappings

of the

Burgers

circuit

surrounding

the defect on 2 types of manifolds : the

perfect

lattice (this

mapping yields

the

Burgers

vector) and the order parameter space

(which

classifies the defects as elements of the

homotopy

groups of this op

space).

Both classifications are

equivalent

because the

perfect

lattice is a

covering

(in a

topological

sense) of the op space. We deime an

aperiodic crystal

as the dj -dimensional boundary of a d-dimensional crystal and get two similar manifolds : an op space U which is the acceptance domain in

Pi

with suitable identifications of the faces, and a

perfect

lattice H~ which is a curved

periodic

lattice in a space of negative curvature and a covering of the former. H~ is invariant under the action of

a non-abelian group of translations H~ which classifies all the

(di 2)-dimensional singularities

of the

aperiodic crystal,

viz. al complete dislocations and b) novel

topological

defects which we call disvections. The latter

are classified as the elements of some normal

subgroup

3~ of H~ which includes the commutator

subgroup.

The former appear as the elements of the quotient group of H~ mod.

3~

or else as the elements of a

subgroup

of the group of dislocations Z~ of the

hypercubic

lattice the other elements of Z~ which do not enter in H~

correspond

to partial dislocations. We argue that disvections are the usual

phason

defects (mismatches, etc.). The above

theory

of defects

applies,

mutatis mutandis, to

approximants,

for which we

provide

a

topological

classification.

1. Introduction.

Defects in

aperiodic crystals (commonly

called

quasicrystals

and abbreviated hereunder as

qc)

show up a number of features which make them

essentially

different from defects in

periodic crystals [1-3].

These differences and their

peculiarities

appear best when one defines the qc as

some

projection

from a

crystal

of

higher

dimension.

Let us recall that the structure of a

quasicrystal

obtains

by

the

cut-and-project

method

[4]

performed

in a d-dimensional Euclidean

hypercrystal E~:

a

djj-dimensional planar

cut

Pjj is introduced at some irrational

angle

with

respect

to the d-dimensional

lattice,

and a restricted set of vertices and

edges

of E~,

belonging

to a

cylindrical strip parallel

to Pjj and

spanned by

a unit cell of

E~,

is

projected

on Pjj ;

they

constitute the

« Bravais » lattice of the

quasicrystal

in

question.

Pjj can be

thought

of as the

physical

space. The

hypercrystal

is

usually

a

hypercubic lattice,

with

d=6, dj

=3 for the icosahedral

crystal, d=5,

(3)

djj = 2 for the

pentagonal crystal (the

Penrose

pattem).

Pjj is a

subspace

of E~ which is

globally

invariant under the action of the icosahedral group

Y,

which is a

subgroup

of the

hyperoctahedral

group in d

= 6

(resp.

the

pentagonal

group

D~~,

which is a

subgroup

of the

hyperoctahedral

group in

d=5).

The Euclidean space which is the substratum of E~ is the

product

Pjj x

Pi,

where

Pi

is the

orthogonal complement

of

Pi

in E~. In order to go

farther than the Bravais

lattice,

a more

general

method consists in

attaching

a

di

(=

d djj

)-dimensional

motif to each cell of the

hypercrystal

and

constructing

the

quasicrystal

as the set of intersections of the ensemble of motifs

(the

so-called atomic

su~fiace [5] E)

with

Pi.

Each intersection is

generically

a

point

and

represents

an atom. The qc appears then as a

dj -boundary Pi

of a

d-hypercrystal.

This

picture

of the qc has been

already

used with some

success

[6]

for the purpose of

analyzing

random

tilings.

The results which are

presented

below are in fact

independent

of the nature of the atomic surface

(they depend only

on the

symmetries

of the construction

process)

hence we shall for the sake of

simplicity

invoke the

cut-and~project

method

(and

the Bravais

lattice)

each time a

reference to the construction of the qc appears necessary or convenient.

This paper deals with the classification of defects of a

quasicrystal.

For reasons of

simplicity,

we shall discuss

only

the defects of the qc which break the

symmetries

of translation of

E~,

I,e. the dislocations and a new class of line defects

(in 3D)

which we call disvections and which are

singularities

of the

phase. Disclinations,

which break rotational

symmetries,

will therefore not be considered. Defects of other dimensionalities will not be considered either :

they

are far less

interesting

in the

proposed

frame of

study.

The paper is

organized

as follows.

In the next

section,

we comment at a

general

level on some not very well-known aspects of the

general theory

of defects which are of

particular

use for

qc's

; one of them refers to the

utility

of the

algebraic theory

of

coverings,

which

provides

a

deep insight

on the links between the

Volterra-Burgers approach

to dislocations and the

topological

methods ; a second one

concems the

specific approaches

which have been

developed

for the

study

of

crystal

surface

defects. The

impatient

reader will first

glance

over those sections, and come back to them afterwards if necessary. Sections 3 and 4

give

a short account of the nature of the

phase singularities

in

qc's

and of the type of order parameter

(abbreviated

as

op)

space we expect.

For reasons to be

discussed,

the acceptance domain with faces identified

U, despite

its

usefulness,

is not a

perfect

op space, and must be

supplemented by

a second

manifold, H~,

the universal

covering

of

U,

which in a sense is the

equivalent

of the

perfect crystal

of the

Volterra-Burgers mapping.

The

topologies

of these two

manifolds,

and how

they

relate to the classification of dislocations and

disvections,

are discussed in details in section 4. In section

5,

we introduce a third manifold

E~,

whose

topological properties

separate very

nicely

the

dislocations and the

disvections,

and allow for a

study

of their

topological

interaction ; it is indeed well-known that each dislocation is escorted

by

a

phason

strain

[I] analyzable

as a

cloud of disvections which cannot

mutually

anneal. A somewhat

unexpected

result of our

analysis

is that there are two

types

of dislocations in a qc,

partial

dislocations and

complete dislocations, according

to the value of their

Burger's

vector. The word «

partial

» refers to the

same concept of

partial

as in usual

crystals.

The «

complete

» dislocations are not the

analogs

of the usual «

perfect

» dislocations. Both

types

of dislocations are indeed escorted

by phasons strains,

but with

qualitative

differences which are

currently

under

study.

2. The

theory

of defects revisited.

2. I VOLTERRA PROCESS AND TOPOLOGICAL THEORY. All the results which follow in this

work concem line defects in a 3D medium

(or topological point

defects in a 2D

medium).

We

shall have to invoke the two well-tried methods of

approach

of the classification of defects in

(4)

an ordered

medium,

I-e- the usual method known as the Volterra process

[7, 8],

and the

topological

method

[9].

Both will appear necessary, and

they

will

play

a

complementary

role.

This is in a sense

quite surprising,

since it is

precisely

for the line defects that the two methods

give equivalent results,

while it is well-known that the Volterra process is ineffective to

explain

defects of other dimensionalities. Let us make some comments on the relation between the two methods in a usual

periodic crystal.

The traditional method

(derived

from the Volterra

process)

to

recognize

the presence of a

singularity

of the

displacement

phonon »)

field a dislocation in a

periodic crystal

makes use of a

mapping

of a

loop

70

(the Burgers circuit), surrounding

the

dislocation,

onto the «

perfect crystal

» E~

[7, 8]

; this

mapping yields directly

the

Burgers'

vector b as the

closure failure of the

image

r of To in the considered

mapping.

The

topological

method makes use of a

mapping

of yo on the order parameter space V

=

T~ (the d-torus),

the

image

is a

loop

y and the class of the defect is then an element of the

homotopy

group

arj

(T~)

which is the group of oriented

loops

with base

point

on V. Of course the two answers are

isomorphic,

a result which

originates

in the fact that the «

perfect crystal

» considered here is

nothing

else than a

tiling

of a d-dimensional space with

equivalent hypercubes (unit cells),

I.e. is the universal

covering

of the order parameter space V

=

T~.

In

brief,

at the end of this

analysis,

we have tmJo manifolds at our

disposal,

V and E~. In one of

them, V,

the defects are

represented by loops

in the other one, E~,

they

are

represented by

open

paths joining equivalent points.

V is the

analog

of a reduced Brillouin Zone

(rBZ); similarly,

E~ is an extended BZ

(eBZ).

A natural

question

to ask is whether there are

intermediary

manifolds which enable a finer classification of

defects,

some

being represented by loops,

other

by

open

paths

on the same manifold which would be then a iBZ. It appears necessary at this stage to

deepen

the

concept

of

covering

space

[10]

we have

just

alluded to.

As a illustration of the

concept (Fig. I),

take the 2-torus

T~

which is the

op space V for the square lattice E~. V

=

T~

is the square Q with

opposite edges

a and b identified ; its universal

cover is E~

itself,

which is obtained

by glueing together copies

of Q in all

possible

manners

along

the identified

edges.

This construction maps any

loopy belonging

to

T~

onto a

path

r in E~

and, conservely,

«

projects

» onto

T~ along

y any

path

r

joining equivalent points

in

~2

p:E~-T~~p(r)

=y.

(1)

-1 a

b_~ ~

A

a

a) b) c)

Fig.

I. Illustration of the concept of

covering

space :a) C, a

cylindrical covering

of Q b) its universal

cover E2; c) il

(Q

)

=

T~ represented

as a square Q with

opposite edges

a and b identified. The inverse

projections

of a

loop belonging

to il is shown stretched in both covers.

(5)

As a consequence, we see that the group of translations

Z~

of E~ is

isomorphic

to the

homotopy

group of

T~ Z~

is indeed the

group with two generators a and

b,

with the relation

r w aba~ b~

= l. We have :

"i(T~)

m

Z~

m

(a,

b ; aba~ b~

=

l) (2j

This relation r

=

I expresses the way a rotation about a vertex of the square lattice is

performed

with the

help

of successive translations which respect the identifications.

But,

while the square lattice is the universal cover of

Q,

there are other covers of

Q,

in-between lli and the square lattice : for

example

any

cylindrical

lattice C tiled with squares whose

edges

are

along

the

generatrices

of the

cylinder.

Observe that the first group of

homotopy

of the

cylinder

ar

j

(C

is a

subgroup

of arj

(T~),

since

some

loops belonging

to V are still

loops

on

C,

while other ones open up in the inverse

projection

p~ : C

-

E~.

More

generally [10],

l'is a

covering

space of X if there is a continuous

mapping p:v-x

which maps

topologically

each arcwise-connected open

neighborhood

of a

point

y e l' onto a

neighborhood

of xe X. The

mapping

p is often called a

projection.

The fundamental theorem of

covering theory

is that the first

homotopy

group

arj(l')

is a

subgroup

of the first

homotopy

group

arj(X ).

For the universal cover, we have

arj(l')

=

l. This theorem states in fact in a

precise

form the

relationship

between

loops

y ~ in X and their

inverse-projections p~ ~(f~ )

in the

covering

space : the

loops

whose class of

homotopy belongs

to some

subgroup

of

arj(X)

lift to

loops p~~(y~)=y~

in

l',

and the other ones lift to open

paths

P

(i~

=

r ~

We shall find

illuminating

in the qc case to use three different

manifolds,

viz.

I)

a

particular

order parameter space

U, it)

its universal

covering H~,

which

plays

the same role as a «

perfect crystal

», but lives in a space of

negative

curvature, and

it) in-between,

its Euclidean

covering E~

in E~

(E~

is a

special

type of atomic

surface). Unfortunately,

none of those manifolds

yields

a

mapping

as easy to use as the universal

covering

of

T~

for dislocations

(there

is no

simple equivalent

of the Volterra process for

disvections)

; and none

provides

either a

mapping

of the same type as

T~

itself for both dislocations and disvections. But in a sense these difficulties are not

artificial,

and

they

shed a new

light

on the nature of the

singularities,

in

particular

the

singularities

of the

phase

in

qc's.

2.2 SINGULARITIES OF A djj -BOUNDARY IN A d-DIMENSIONAL CRYSTAL. We have all

advantage

to

keep

the

picture

of the qc as a djj

-boundary

Pjj of a

d-crystal

: it tells us that the defects divide into two classes :

a)

those which

belong

to the

hypercrystal

E~ and cut the

physical plane Pjj,

and

b)

those which are

specific

of the

boundary

of the

hypercrystal.

Such a

partition

is not new in the

theory

of defects in condensed matter ; it has been used first

by

Volovik

[I

I for the purpose of

classifying

surface

singularities

with constrained boundaries in usual ordered

media,

with the

help

of the

topological

method of classification of defects

[9].

Consider,

as a

simple (and helpful

for what we have in mind later

on)

illustration of Volovik's

method,

the case of a nematic

phase

with

parallel anchoring

conditions of the director at the

boundary

:

I)

in the bulk the order parameter space V is the

projective plane P~

and the defects of class

a)

are classified

by

the classes of

equivalence

of the various groups of

homotopy ar~(P~), I.e.,

for n

=

I say,

arj(P~)

=

Z~,

the Abelian group with two elements which tells us that the line defects

belong

to two classes,

experimentally recognized

as the

thins

(which

are

topologically singular

lines,

represented by

the non trivial element in

Z~)

and the thicks

(these

lines are

represented by

the trivial element in

Z~,

I-e-

they

are

topologically

unstable lines of

defects) [12]

;

it)

on the

boundary,

the order

parameter

space is

(6)

restricted to U

=

§~

; each

point

of the standard circle

§~

represents

a

possible

direction of the director at the

boundary. arj(§~ )

=

Z,

the group of

integers,

classifies both types of line defects

a)

and

b) belonging

to the

boundary

;

a)

defects are

represented by

the odd

integers

S

= 2 p + I in

Z,

and

b)

defects

by

the even

integers

S

= 2 p. The way U is included in V tells

us

something

about the

relationship

between the surface and the bulk

defects;

here the solution is rather easy to grasp

(Fig. 2)

: the odd

integer,

class

a),

defects map on the non- trivial element of

Z~

and are intersections of thins with the

boundary

;

conversely,

the class

b)

surface defects of even

integer strength

in

Z,

which are

singular

on the

boundary,

are the terminations of the thicks and map on the trivial element of

Z~. they

are isolated

point

defects on the

boundary,

since the thicks are

usually non-singular (Friedel's

nuclei are most

probably

surface defects of this type in small molecules

nematics).

These considerations

point

towards an

homomorphism

between the two groups

ii arj(§~

-

art(P~) (3)

whose kemel ker

ii

is the invariant

subgroup

which consists of the even

integer

elements S

= 2 p in art

(§~ ).

All the results of some interest

concerning

the surface defects in a nematic with

parallel anchoring

conditions are included in

equation (3).

This is a rather

simple

case. In

more

complex instances,

the

analysis

of the surface defects would

require

some more

algebraic-topology

theoretical

results, involving higher homotopy

groups. Notice that in the

example

under

study

the

singular points

of the

boundary

can also be defined 0

directly

as the elements of the relative

homotopy

group

ar~(V, U),

since this group classifies indeed the classes of

equivalence

of the defects surrounded

by loops

To

belonging

to the

boundary

and

capped

in the bulk

by

a 2D surface

ahomotopic

to a

half~sphere.

We find indeed

ar~(V, U)

= ker I

trim j~ (see below).

Results of a similar nature extend to any situation in which two order parameter spaces V and U are related

by

an

operation

of inclusion ; more

precisely,

it exists in each such case an exact sequence

of homomorphisms

between

homotopy

groups, which reads

a~ ,~ j~ a~ ,,

-

ar~(u)

-

ar~(il)

-

ar~(il, u)

-

arj(u)

-

arj(il) (4)

-

~~~/

a

/~~

Fig. 2. Defects of classes a) and b) at the boundary of a nematic with

parallel anchoring

conditions (see text).

(7)

The property of exactness means that the

image

of any

homomorphism belonging

to the sequence is the kemel of the

homomorphism

which follows in the sequence ; for

example

:

im a~ = ker

ii-

Note that

j~

is another

operation

of inclusion which consists in

considering

any

oriented 2D

cycle

in U as

belonging

to

V,

and a is the

operation

which consists in

restricting

« to its

boundary

am = y. The last two

homomorphisms

of the sequence suffice to discuss the

relationships

between defects of different classes in the case of

aperiodic crystals.

But it is clear that the

knowledge

of arj

(U

is

enough

to

classify

all the defects in

Pjj.

Observe that in the nematic case we have

ar~(V)

= Z ;

hence,

because of the property of exactness of the

sequence,

ar~(V, U)

is as stated above.

3. General remarks on the

topology

of defects in

aperiodic crystals.

Since the defects

belonging

to the

hypercrystal

E~ can be

readily

classified

by

the standard methods relevant to ordered media

by simply generalizing

them to

high dimensionalities,

it is rather easy to

classify

the defects of the qc which

belong

to class

a).

The dislocations and the

disdinations of the qc are indeed

nothing

else than the intersections with

Pi

of the defects of the same

topological

nature in E~, I-e- classified

by

arj

(V ).

If one

forgets

the

disclinations,

the order parameter space V for the defects in E~ is the unit cell with

opposite

faces

identified,

I-e- the d-dimensional torus

T~.

Defects of class

a)

of other dimensionalities than dislocations

originate

in

non-generic

intersections and are

probably

of small

physical

interest. We refer the reader to the relevant papers for more details

[13~15].

Defects of class

b)

are much more subtle and more novel.

Let us define the nature of the constraint on the

boundary

Pjj of the

hypercrystal

: all the

parallel positions

of Pjj which intersect

Pi

inside the

projection /ti

of the unit cell on

P~ provide equivalent ground

states of the qc,

differing

one from the other either

by

a

global

translation and/or a

phase

shift

(Fig. 3).

We call

/ti

the acceptance domain. It is a

tdacontahedron TR in the icosahedral case and a set of 4

pentagonal

sections of a rhombic icosahedron al in the

(slightly

more

involved)

Penrose case.

By conveniently identifying opposite

faces and

edges

of

/ti,

we obtain a closed manifold U which is the op space of the

hypercrystal boundary

Pjj constrained as defined

above, although

some difficulties are

immediately

apparent

I)

a

given phase

state of the qc is

represented by

an

infinity

of

points

in U

(and

not

by

a

single point) differing by

translations e

belonging

to the group of symmetry of E~

(Fig. 3)

note that the set of vectors e do not form a group ;

it)

the set of such translations e which connect

equal phase

states in

/ti (hence

in

U)

is not the set of all the translations of the symmetry group, but

only

those which connect two

points

of the

strip,

I-e- it excludes a number of translations

(forming

a

group) belonging

to the set

£

m, y,

(see Fig. 3).

In other

words we have in U too many values of the same

phase

and not

enough

of the different

translations. These

peculiar

characters can be rationalized so that U

obeys

the usual definition of an op space, as follows.

The translations e and y

together

form a group

Z~ again, they

are

represented

in il

by

different

loops

which are not

homotopic

to zero

(except

the zero element of the

group),

or

equivalently, by

segments which connect

equivalent points

on the

boundary

of the unfolded version

T~

of

V,

I-e- of the unit cell in E~

(Fig. 4).

But observe that a segment of type e

projects orthogonally

in

/ti

as a segment which does connect two

points

which are both in

/t~,

i.e. which does not

yield

a

loop

in

U,

whereas a

segment

of

type

y generates a

loop

in U.

Therefore y connects

equivalent points

in

/t~,

while e connects

points

which

qualify

as

«

inequivalent

»,

although they

generate

qc's differing by

a «

phonon

» shift

only,

I-e- a pure

displacement.

Henceforth the

only

dislocations of E~ which

are classified as defects in

arj(U)

are those of type y

(they

form a group

isomorphic

to

Z~).

We call them

complete

(8)

3 P

I

I

i

Fig.

3. The acceptance domain

Jti

of

equivalent perfect crystalline

states for the d= 2 case.

PI

and

Pi

have the same

phase

state, but differ

by

a translation of the type e =

£

n, e,, where e is a free

vector

belonging

to the

strip

whose cross~section is

Jti

;

PI

and

PI

differ

by

a

phase

shift (the translation

which

brings PI

upon

PI

is not a lattice translation)

PI

and

PI

do not

belong

both to

Jti,

have the same

phase

state, but differ

by

a translation t

= e + y, where y

=

£

m, y, is not a free vector of the strip.

e

Ai

~

Fig.

4. Relation between

Jti

and the d-dimensional torus T~. Dislocations of type e are

partials,

dislocations of type y are

complete.

d

= 2. See text for more details.

dislocations. On the

contrary,

dislocations of

type

e, which are of course true defects in

Pi,

since

they

can be constructed as intersections of the relevant defects in

E~,

should be considered as

partial

dislocations of the qc, since their

representation

in

/ti

connects two

points

which are

inequivalent. They

are true

partials

in the exact sense of usual

crystals,

where the set of

partial Burgers'

vectors

emerging

from a common

origin

form a star in the unit cell

(9)

(this

star is

easily

related to the

Thompson representation

of

partial Burgers'

vectors, which is

a tetrahedron in fcc lattices

[7])

and a set of open

paths

in the op space

T~.

In the qc, the

relevant star is the

projection

of the vertices and

edges

of the Bravais lattice of the qc in

/t~. They

are also

partials

in the sense of

approximants

the y vectors transform to the true lattice parameters of the

approximant

when the

boundary

Pjj is

rational,

and the e vectors tum to the

partials (in

the usual sense of this

term)

of the

approximant crystal.

We can summarize this discussion

by stating

that the

homomorphism

ii, gr~(u)

-

grj(v) (5)

is not onto im

ij

does not contain all the elements of

arj(V).

The first

homotopy

group

arj(U) yields

all the types of

complete

defects

a)

and

b)

which break the « translational »

symmetries

of

Pjj.

Some of those defects

(we

have

just

characterized them

by

their

Burgers' vectors)

are

dislocations,

I-e-

singularities of

the «

phonon

»

field,

we call the other ones

disvections

;

they

are

clearly singularities of

the

«

phason

»

field

: and we shall argue that

they correspond precisely

to « mismatches » of the Penrose

lattice,

in the d

= 5 case

(Fig. 5),

and to their

generalizations.

Both fields are indeed modified when Pjj is

displaced parallel

to itself

and spans U. The

partial

defects appear as the elements of

arj(V)

which are not in

arj(U). They

are all of the dislocation type. The rest of this paper deals

essentially

with disvections.

Again

U is not the

complete

order

parameter

space

(although

it is a valid op

space)

because it does not include rotational

symmetries.

,,

,' 1 "

""#*'

'

'

1 , Al

i , 15

1 ,,

1 , ,

1 , ,

' 35

' '

54

1 , 3

1 , ,

, , , 42

, , ,

',

1

~~

23

' '

',

,

$~ 43 25

34 ~~

, ,

, , ,

, 51

,

' ' ' "

' ~,

' '

' ' '

24 j 4

' '

' $

, ' ,

' '

*

, 2

'

13

a)

~~

b)

Fig.

5. Triacontahedron TM a) bird's eye view b) its Schlegel diagram with faces identified by the symmetries of translations of the parent

hypercubic

cell.

4. Defects classified as a whole.

In this

section,

we define U and

H~

and indicate how these manifolds are used. We

delay

a

similar discussion for

E~

to the next section. As

already indicated,

U and

H~

are

respectively

(10)

the

analogs

of the order

parameter

space and of the «

perfect crystal

»

(the

universal

covering

of the op

space).

But

E~

is a manifold much more novel in the

theory

of defects

by

many

aspects.

4.I THE ORDER PARAMETER SPACE U AND ITS COVERING SPACE

H~(d= 6).

In the

icosahedral case the op space U is a TR with

equivalent

faces identified. We do not enter into the details of the geometry of this manifold and of the relevant identifications this is

published

elsewhere

[16],

as well as the more delicate case of the Penrose

tiling.

The identifications in

question

are the identifications induced on

/ti by

the

orthogonal projection

on

Pi

of the identifications of the

opposite

faces and

edges

of the unit cell in

E~.

Figure

5

gives

as an

example

the identifications of the faces of

/ti

=

TR. These

identifications

provide

the group of translations of the universal cover

H~

of U.

P:H~-U. (6)

For

TR,

the

corresponding

group of translations has 15

generators

y, =

A~~ (I

=

1,

,

15 p, q = 1,

,

6)

and there are 12 relations between generators, of the form

[17]

:

~l"~12~13~16~14~15~ ~12"~46~26~56~36~16

"1

(7)

where

A~~

is the translation which

brings

face pq over face qp

(Fig. 5).

The reader will

recognize

that each of them involves a series of faces

belonging

to the same zone on TR. In these

expressions,

the

operators

r; act to the

right, I-e-,

in rj, say,

Aj~

acts

first,

then

Ai~,

etc. The group

H~

=

lY,

; r; =

11 (8)

which is defined

by

generators and relations is the group of translations of the universal cover

H~

of U

(TR ) and, by

definition of an universal

covering,

it is also the first

homotopy

group of U.

gri

(u )

-

H6 (9)

I.e. it is the group which classifies

complete

line defects in a qc

(partials

are excluded from

H~,

as it has

already

been

emphasized

above ; but these

points

will be taken over in more details in the

sequel). H~

is a

non-abelian,

infinite group, since the number of

independent

generators is

larger

than the number of relations. Therefore

H~

is a «

crystal

» tiled with TR'S in a space of

negative

curvature

[16, 17]

of dimension

di

= 3.

4.2 MAPPINGS OF LOOPS OF THE QUASICRYSTAL ON U AND

H~.

The

precise

definition of

the order parameter of a deformed

quasicrystal

presents some fundamental

difficulties,

whose nature can be

already

realized when

trying

to

give

a measure of this order

parameter

in the

simple

case of a

perfect qc.

Of course, if this qc is defined

by

some djj

-plane Pjj,

it

might

seem

sensible to

give

to the order parameter a

unique

value

w(Pjj

in

Pi,

as follows : lift any vertex r = Mjj

belonging

to the Bravais lattice of the qc to the

point

JL e

E~

from which it has been

projected orthogonally,

then

project orthogonally

JL on

Mi

in

Pi,

take the

envelop

of all these

projections,

which is a window

congruent

to the acceptance domain

/ti (we

denote it also

by

the same

symbol /ti,

since there is no ask of confusion, and since this

property

of congruence allows us to use the window as an

acceptance domain,

I-e- as a copy of the rBZ of the

op)

and define

w(Pjj

as the value of

w at the center of

/ti,

for all r's. However this

unicity

of

w(Pj ) depends strongly

on the fact that the infinite qc is known. If this is not the

case, there is some

indeterminacy

in the choice of the window in which to locate the

(11)

projection

of the finite qc. A solution to this

difficulty

is to

assign

a variable order parameter to a

perfect

qc, viz. the value

w(Mi

to each vertex Mjj of the qc. Let us show that this does

not lead to

topological

inconsistencies. Note indeed

that, although

all the

physical points

Afj do not

project

at once on the same

point

in

Pi,

all the

loops

To in the qc lift to a

image loop 7~

in E~ which

belongs entirely

to the

strip

§ of the

cut-and-project

method, since the qc is

perfect.

Hence its

orthogonal projection

yi in

Pi

is

entirely

inside the window

/ti

and can be

smoothly

reduced to a

point

inside

/ti

without ever

meeting

the boundaries of

/ti Consequently,

this

property

of smooth

reductibility

to a

point

in

/ti

is not modified

by

an identification of the

edges

and faces of

/ti

which transforms

/ti

to U and

yi to 7. The

image yof

To in U is

homotopic

to zero, And the

image

r of To in

H~

is

obviously

a

loop,

I.e. is also

homotopic

to zero in

H~,

since any

loop

is reducible to a

point

in the universal cover. In that sense, the infinite

perfect

qc is

truly homotopic

to zero.

This is true also of any finite

perfect crystal,

for which the above method of the

envelop

does not

yield

a

unique w(Pi ).

The above discussion extends to any

loop

yo in the qc

(not only loops joining vertices)

it suffices for the sake of the argument to lift any

point Mj

of the

loop

yo to the

corresponding point

JL e E~

belonging

to the djj -face in E~ which is the lift of the tile in

Pi

to which M~

belongs.

Has this

mapping

a sense for a deformed qc ? The answer is

straightforwardly positive

for a

«

phonon

» strained qc, and for dislocations. For in this case the

image y~

in E~ of a

loop

To

surrounding

a dislocation is an open

path,

and its closure failure measures

precisely

the

Burgers'

vector b in d dimensions. The

projection y~

of

y~

in

/t~

is also an open

path,

but its

image yin

U is either an open

path

if To surrounds a

partial dislocation,

or a

loop

if it surrounds

a

complete

dislocation. In this latter case, this

loop

is

represented by

some class of

arj(U)

which we shall soon

display.

The use of the same

mapping

to the case of a

phason

strained

crystal

does not

display

at once the

topological

nat~Jre of the

phasons, although

it can be shown

[18]

that the union

~J M~

of the

images

in

P~

of a

phason

strained qc, finite or

infinite,

cannot be included in any window

/t~.

We then first

modify

our

perception

of this

mapping by noticing

that any

point

JL e E~ and

belonging

to the

loop y~

can be attached

arbitrarily

to a

particular

unit cell of the

hypercubic lattice,

for

example

any of the

C(d, di )

=

2~~

hypercells containing

the djj -dimensional face to which JL

belongs.

We prove below that, this first choice

being

made for some

starting point

on 7~ the other

hypercells

which follow when

traversing

yo are no

longer arbitrary.

In the case of a

dislocation,

the

starting

unit cell and the final unit cell are

equivalent by

a translation

precisely equal

to the

Burgers'

vector measured in E~. Observe that this

perception

of the

mapping

is reminiscent of the de

Bruijn's pentagrid [19],

because in the

pentagrid

each mesh

(which

is the dual of a vertex of the Penrose

pattem)

is the intersection of some

hypercell

in E~ with

Pi.

Therefore the

pentagrid method,

which can

be extended to the icosahedral case,

provides

without any arbitrariness an

hypercube

specifically

attached to each vertex of the qc, while the direct method suffers from the arbitrariness in the choice of the first cube at the

starting point

of the circuit. The

unicity

of the result

provided by

the dual method proves in a way the result we were

stating

above. But

it can be proven in another way we schematize the argument.

Assume that the

starting point JL°

on

7~

is attached to some

hypercell H°

which contains the

djj-dimensional

face

to which it

belongs.

This

dj"face belongs

itself to a set of

(d

I

)-dimensional

faces of the

hypercell,

which are

C(d I, dj )

=

d~

in number. When

JL°

moves to another

position JL~

in

some other

djj-face h~,

two cases arise ; either h~

belongs

to the same

hypercell H°,

and we then make H~

=

H°,

or it

belongs

to some other

(12)

hypercells.

We look for the condition that there is no choice on

H~.

Observe that JL

belongs

to a set of

(d

I

)-dimensional

faces which are also C

(d 1, ~ )

=

d~

in number

but are different from the former. These two sets

belong

to the same

hypercell

H~ if and

only

if 2 x C

(d I, dj )

is smaller or

equal

to the maximum number of faces of a d-

hypercell,

viz. 2xd. This condition reads

d~

w2xd and is satisfied for

d~ =3,

d

=

2 or 3.

5. Dislocations and disvections discriminated.

We used above a

mapping

of To in E~

(yo

-

7~)

which

played

an

intermediary

role.

Although

it presents in itself some interest

(it

allowed us, if some

precautions

are

taken,

to define

unambiguously

the

mapping

on

U),

it has a

disadvantage

which limits its use : E~ does not

have the same local

topology

as U.

Hence,

there is no relation of

covering

between

yo and

y~.

It is the

identity

of the local

topologies

of U and

H~

which allows for an

unambiguous mapping

from one to the

other,

since

H~

covers U

(infinitely

many

times).

We

show now that it is

possible

to embed

7~

in a manifold which is a

covering

of

U,

and which has the same

topology

than

P~,

in a sense.

5.I

i~,

AN ABELIAN COVERING OF THE op SPACE OF AN APERIODIC CRYSTAL.

H~

is the universal cover of

U,

but there are

midway

other covers of U which are of

interest. We define now a cover

E~

of U where the

Burgers'

circuits of dislocations

(partial

and

complete ones)

in the qc are

represented by

open

paths

and in which other circuits in the qc map on non-trivial

loops

:

they

surround

topological

defects which we call disvections.

Ea

consists of the set of all the

orthogonal projections /ti

of all the

hypercells

of the

hyperlattice

in

Pi,

with rules of incidence inherited from the

projection

in

Pi

of the rules of

incidence of the

hypercells

in

E~.

In the d=6 case,

/ti

= TM. A better view of

Ea

is obtained

by lifting

each

di

-dimensional

/ti

inside the d-dimensional

hypercell

from

which it is

projected,

in such a way that each face of the lift /t is

glued along

the

«

silhouetting

» face of the

hypercell

from which it is

projected

onto

/ti (Fig. 6).

These two

tx

I

I

Y ,

fl

Fig.

E

~ in theunit of E~.

The

edges a,

a', p, p', y, y'

are

in « silhouetting » ositions

PI [I I Ii- Tlie

projection Jti of the

unit

cell (or of Jt) on Pi is

an

hexagon.

Pi

is a ii I Ii plane.

Références

Documents relatifs

contrast of individual dislocations in silicon; by analysing their results they concluded that the con- trast model proposed by Donolato [3, 4] was unable.. to

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

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

I) d'affecter la mobilit6 des dislocations vis dont le mouvement semble contr6- 16 par la propagation de crans simples a. 2) de bloquer le fonctionnement de sorlrces

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

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

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

Its elastic interaction energy with the limiting surfaces is calculated by taking into account ngorously the surface tension y This formula is used to discuss the dislocation