• Aucun résultat trouvé

Inelastic mean free path and phase-shift determinations in NiO, using EXELFS

N/A
N/A
Protected

Academic year: 2021

Partager "Inelastic mean free path and phase-shift determinations in NiO, using EXELFS"

Copied!
12
0
0

Texte intégral

(1)

HAL Id: jpa-00246823

https://hal.archives-ouvertes.fr/jpa-00246823

Submitted on 1 Jan 1993

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.

Inelastic mean free path and phase-shift determinations in NiO, using EXELFS

Mohammad Tafreshi, Stefan Csillag, Zou Yuan, Christian Bohm, Elisabeth Lefèvre, Christian Colliex

To cite this version:

Mohammad Tafreshi, Stefan Csillag, Zou Yuan, Christian Bohm, Elisabeth Lefèvre, et al.. Inelastic

mean free path and phase-shift determinations in NiO, using EXELFS. Journal de Physique I, EDP

Sciences, 1993, 3 (7), pp.1649-1659. �10.1051/jp1:1993207�. �jpa-00246823�

(2)

Classification

Physic-s

Abstracts

71.20 78.70D 61.14

Inelastic

mean

free path and phase-shift determinations in NiO, using EXELFS

Mohammad A. Tafreshi

(I),

Stefan

Csillag ('),

Zou Wei Yuan

('),

Christian Bohm

('),

Elisabeth Lefbvre

(2)

and Christian Colliex

j2)

(')

Institute of

Physics,

University of Stockholm,

Vanadisvhgen

9, S- ii 3 46 Stockholm, Sweden (2) Laboratoire de Physique des Solides, Bit. 510, Universitd Paris Sud, 91405 Orsay, France

(Receit'ed 22 September1992, accepted in final form 2 March 1993)

Abstract. Structural and chemical information about the local atomic environment in

a

specimen

can be obtained

by analysing

the EXELFS-modulations

occurring beyond

energy loss ionization edges.

Using

spectra recorded from a thin NiO sample, this paper reviews the problems and

possibilities

associated with distance and

intensity analysis

in EXELFS.

Knowing

a priori the

structure of the sample, one can determine unknown parameters involved in the

analysis

of the data

such as an average value of the inelastic mean free

path

for the

ejected

electrons, a linear approximation for the k-dependence of the oxygen d~~, oxygen-oxygen pair d~~_~, and nickel

backscattering

d~

~~, phase shifts.

1. Introduction.

The energy distribution of electrons

passing through

a thin foil of material is a measure of the electronic excitations

produced

in the

sample by

the beam.

During

inelastic

scattering

events, the incident electrons excite inner-shell electrons. Interference between the

outgoing

excited

inner-shell electron waves and electron waves backscattered from the

surrounding

atoms

affects the

energy-loss

spectrum. The

probability

for a

given

energy loss increases or decreases

depending

on whether constructive or destructive interference occurs

[Il.

Approximating

the

ejected

electron wave function at the

backscattering

atom

by

a

plane-

wave while

assuming single scattering

formalism with a small

scattering angle,

the

interference

amplitude

can be described

[2] by

:

where

N~

is the number of atoms in shell

j,

i~ is the

radius,

W~

(k)

is the total

phase change

of the electron wave after

t?avelling through

the field of the

emitting

and

backscattering

atoms,

f~(k)

is the

backscattering amplitude

which

depends

on the type of

backscattering

atoms,

(3)

A is the mean free

path

for inelastic

scattering

of the

ejected

electron, and «~ is a parameter

including

thermal vibrations and static disorders of atoms.

The total

phase shift, W~(k)

for K-shell ionisation

edges

is described

[3] by

:

mj(k>

~

ma(k>

+

m~(k>

ar

(2>

where

W~(k)

and

W~(k)

are

phase

shifts due

respectively

to the central and

backscattering

atoms.

Changing f~(k)

to

f~(k,

r) and

W~(k)

to

4l~(k, r),

in

expression (21,

recasts this

expression

from the

plane-wave approximation

to full curved-wave formalism

[4].

Depending

on the a

priori

known

parameters, equation (I)

can

provide

means for

deriving

information about the local atomic environment,

[I.e.,

q,

N~, A,, 4l~(k), W~(k), f~(k)

and

«jl 15].

2.

Experimental

data

acquisition.

EELS spectra have been recorded under a

primary

beam

voltage

of 100 kV with a Vacuum Generators dedicated STEM

equipped

with a Gatan

parallel

EELS spectrometer

(see

Bouchet et al.

[6]

for a

complete description

of the instrument

operated

in

Orsay). During

spectrum

acquisition,

the incident

probe

is

permanently

scanned over

specimen

areas of

typically

10 to 30 x 10 to 30

nm~.

The

angle

of acceptance of the inelastic electrons is determined

by

the size of the inner hole in the annular dark field

detector,

which

corresponds

to about 25 mrad at the exit surface of the

specimen.

NiO

specimens

were

specially prepared by

C-E-

Rojas

as part of a

study

on model

catalysts.

After

impregnation

and calcination of a mixture of

amorphous

silica and nickel nitrate, the reduction to metallic nickel is carried out in a

thermoprogrammed

reduction

(TPR)

system. A

large peak,

observed

during

this

procedure

at 392

°C,

is

normally

associated with the presence

of bulk NiO. In order to be

observed,

the

samples

were crushed in an agata mortar and

suspended

in ethanol before

deposition

and

drying

on a thin

holey

carbon foil. The

particles

observed showed an average size of 90 nm but looked

generally

like clusters of smaller size.

Selected area diffraction pattems

correspond

to

crystalline

NiO. EELS spectra are recorded

from areas close to the

edge

of such thin

microcrystals protruding

over the holes of the

supporting

films. The average thickness was estimated to be less than 0.5

A,~,

where A,~ is the inelastic mean free

path

of the incident electrons. This was done to avoid the

superposition

of

plasmon

type satellites, due to

multiple

inelastic events, over the first 50 ev

following _the edge.

Typical

spectra are shown in

figure

I. When

focusing

on

edge

fine structures, the energy

dispersion

is set to cover a

typical

100 eV window.

Energy

resolution better than I eV can then be achieved with a sufficient

signal-to-noise

ratio. The

analysis

of the

origin

of the fine structures observed within the first 50 eV above threshold is discussed in another paper

by

Kurata et al.

[7].

For

EXAFS-type studies,

the whole energy range from the O-K

edge

to the next

edge (I.e., Ni-L~~

at 855

eV)

must be recorded

(see Fig. lb)

with a sufficient

signal-to-

noise ratio, in order to detect the

periodic

oscillations over this extended energy loss domain.

Since the

Ni-L~

and

Ni-L~ edges

are

relatively

close in energy and thus

overlap,

structure

obtained

beyond

these two

edges

cannot be used for EXELFS

analysis,

otherwise such an

analysis

would have

hopefully strengthened

and confirmed the results obtained from the O-K

edge.

30 General comments about the

analysiso

The structure

parameters

of NiO

(I,e.,

of Nacl

type

with a = 4,1767

1) previously

determined

by

other

techniques [8] together

with the

energy-loss

spectra recorded from the oxygen

K-edge

(4)

in a thin NiO

sample (Fig. lb),

have been used to :

Perform and

investigate

the

EXELFS-intensity analysis.

Determine the average value of the inelastic mean free

path

of electrons

ejected

from oxygen atoms in NiO.

Obtain

experimental

values for

4lo_o (I,e., phase

shifts due to the O-O

pair),

and a mixture of the

experimental

and theoretical values for

4lo_

and 4l_N,

(I.e., phase

shifts due to oxygen central and nickel

backscattering

atom

respectively),

as well as the linear fit to these

phases

over the

experimental

data range.

Provided

optimum sample

thickness is

achieved,

the errors introduced in the determination of the non

phase-shift

corrected interatomic distances and

corresponding intensities, originate mainly

from the

following

sources :

Low resolution effects and truncation artefacts introduced as a consequence of the short data interval and limited number of data

points.

These effects can however be estimated and corrected for

by analysing

simulated spectra

corresponding

to the same conditions as the

experimental

one.

Inaccuracy

in determination of the K~~~

K~,~ value,

caused

by

an inaccurate choice of the

Eo position (I.e.,

energy threshold

position)

and inaccurate calibration of the energy axis.

The main effect of this

inaccuracy

is a shift of all measured distances with

equal

percentage in the same direction

[9].

§~ 27

O 7

U fi

]O 5

1 f3

520 540 560 580 600

Energy Lou (cV) Ch&""tl ~~.

~~ b>

Fig.

I. al Energy loss spectrum

displaying

the near edge structure of the O-K edge from a thin NiO sample b) energy loss spectrum from the same NiO area exhibiting the O-K and Ni-L~~ edges. Energy loss interval is from 460 to 948 eV.

The near

edge

structure of the O-K

edge

in NiO has been

theoretically

calculated

[10].

Comparing

the

experimental

spectrum with the

theoretically

calculated spectrum allows an

accurate determination of the

Eo position.

The value obtained for

Eo using

this

procedure

is

found to be 530 eV, I-e- 10 eV below the

position

of the most intense

peak (observe

that the

same

value,

I-e. 530 eV, has also been used

by Leapman

et al.

[I II).

Since calibration of the energy axis has been

accurately

determined

by using

the energy difference between oxygen K-

edge

and nickel L~

~

edges,

the

inaccuracy

introduced

by

this step should be

negligible.

Inaccurate

background

subtraction can

give

rise to false structures in the RDF

(I.e.,

Radial Distribution Function or

magnitude

of the

FFT)

at small I-values.

Background

subtraction has been carried out

by fitting

two third order

polynomial

functions to different

(5)

parts of the data interval of interest. In order to minimize the

inaccuracy,

the

fitting regions

have been chosen as

long

as

possible

(I,e., one from channel 157 to 691 and the other from channel 435 to

803,

see

Fig, lb).

The first

peak,

obtained in the RDF

(Fig.

2b,

peak Po),

is more

likely

due to this

background

subtraction effect and to the truncation artefacts introduced

by

the second

peak. Analysis

of

simulated data shows that this

peak

has a few percent influence on the

intensity

of

Pi (I.e., Ip,,

see Tab. I and

Fig. 2b),

and none on its

position.

Since intensities in our

analysis

are determined

by fitting

the RDF of simulated data to the

experimental

data, the effect of this

peak

can be

compensated

for

by including

the

corresponding frequency

in the simulation

model.

Q ~

W fi

§

I

E ]

« s s a 1 4 o la u u 13

A'~ AngsUom

a> b>

Fig. 2. al Spectrum from

figure

16 after background subtraction, conversion from energy to k-space, multiplication by k~ and filter function ; b) Magnitude of the FFT of (al, P4 is not observable due to its low intensity (less than 0.5 times Ip~ and 0.4 times Ip~), however its position should be at about 3.88 A.

4. Data

analysis.

The data

analysis procedure

used in this

work,

which has been

developed by

the Stockholm group and discussed

extensively by

Tafreshi et al.

[9, 12],

includes the

following

steps

selection of

Eo Position

at 530 eV,

subtracting

the main

background,

conversion from energy space to

k-space, multiplying

the data

by

k~ in order to compensate for the

damping

effect of the

k-dependent

factors

[I,e., Ilk, f~

(k) and «-term, see

Eq. II II

and

multiplying by

a filter

function which varies as cosine square at the ends of the data range but is flat over 50 fib of the central

region (Fig. 2a).

As a consequence of the short data interval

corresponding

to the

wavelengths

of the

principal

modulations, it is useful to zero-extend the data outside the

original

data interval

prior

to the Fourier transform in order to

improve

the apparent resolution.

However this

improvement

is achieved at the cost of increased transformation time and the appearance of smaller satellite

peaks

due to the truncation effect.

Special

care must be taken when

interpreting

these

peaks.

In order to minimize

inaccuracy,

fifteen different data intervals have been chosen and Fourier transformed. The mean and standard deviations of the FFT results obtained from these

intervals

(Fig.

2a and

Fig. 2b), (Tab. I),

is

being

used as the

experimental

values.

(6)

Table I. The lst column is the central and backscatterer atoms in the

first

seven shells

expected

to

give

most contribution to the EXELFS

modulations,

the 2nd column is the

expected

values

of

the

corresponding

interatomic distances in

~ngstrom

;

3rd, expe;imental

non

phase-shift

cot.i"ected distances in

~ngstrom 4th, dijfiet.ences

bemleen the known interatomic distances

(I,e.,

column

2)

and the

experimental

non

phase-shift

corrected

(I,e.,

column 3

)

; sth,

expe;imental

intensities

(I.e., height ofthe peaks

) ; 6th,

expected

coo;dination

numbers

;

7th, expe;imental

q~o (w>ill be discussed in the

following)

in ;adian.

2 3 4 5 6 7

O-Ni 2.09 1.81 ± 0.02 0.28 ± 0.02

lpi

= 100 ± 0 6 0.30 ± 0,14

O-O 2.95 2.65 ± 0.02 0.30 ± 0.02

Ip2

= 80 ± 2 12 4. 84 ± 0,19

O-Ni 3.62 3.36 ± 0.04 0.26 ± 0.04

Ip3

= 41 ± 2 8 3, I1 ± 0.23

O-O 4,18 6

O-Ni 4.67 4.32 ± 0.04 0.35 ± 0.04

Ips

=

36 ± 3 24 0.92 ± 0.31

O-O 5.12 4.79 ± 0.06 0.33 ± 0.06

Ip~

=

34 ± 2 24 4.39 ± 0.40

O-O 5.91 5.53 ± 0.02 0.38 ± 0.02

Ip~

=

32 ± 2 12 3.09 ± 0.19

4.I ANALYSIS OF THE PHASE SHIFTS. For many elements the

phase

shift

[W~(k)],

is

approximately

linear in k over the

analysing

interval and can be written

[13]

as

~°j(k>

+ " ~

~°a(k>

+

ibb(k>

# §~o + §~I k.

(3>

While q~o is a constant

phase

term which cannot affect the distances obtained from the

RDF,

q~j introduces a shift of these distances towards lower values. For those elements for which

W~

(k)

cannot

accurately

be described

by equation (3),

this

expression

can still be considered as

a linear

approximation.

A

comparison

between

expressions

I and 3

implies

that the

experimental

distance obtained from the RDF is

equal

to r +

q~j/2.

Thus

subtracting

the

expected

value of ; from the

experimentally

measured value

(I,e.,

; +

~j/2), yields q~j/2.

In

general

the

imaginary

and real parts,

corresponding

to a

particular peak

in the

magnitude

spectrum of a Fourier

transform,

can be used to calculate the

phase

value «

= arctan

(Im/Re ).

This is the

phase

of the wave

corresponding

to that

particular peak,

at the

beginning

of the transformed data interval. Since the

frequency corresponding

to this wave is

already

known

(I,e.,

the

position

of the

peak

in the

magnitude spectrum),

it is

possible

to determine the

phase

value of this wave in

regions

outside the transformed data interval.

In our case the

imaginary

and real parts

corresponding

to a

particular peak

in the

magnitude

spectrum

(see Fig. 2b), yield

the

phase

value at k~~~

(in

the

k-space). Using

this

phase

and the

corresponding frequency

allows us to determine the value of q~o

w

(I,e., phase

shift at

ko or

Eo).

Column 7 in table I shows the obtained

phase

shifts with correction for

w

(I.e., ~o)

for different atomic shells.

Analysis

of the simulated data indicates that as an effect of

overlapping neighbouring

shells

(or

low

resolution),

with the

exception

of the

peaks Pi

and

P~

the

position

of other

peaks

in the RDF

(particularily P~),

cannot

accurately

describe the

corresponding

interatomic distances.

Since the obtained value of q~o

depends

on the

position (I.e.,

distance or

frequency)

of the

peaks,

the

relatively

poor agreement between the q~o values obtained

using

these

peaks (particularily

for P~, see

Fig. 2b)

can

mainly

be

explained by

this effect. Curved-wave effects

[4]

also introduce a

slight r-dependence

in the

phase

shifts.

The

experimental

value for the

non-phase

shift corrected first O-O distance is 2.65 ± 0.02

1

(see

Tab.

I),

while the

expected

distance for this shell is

2.951 (see

Tab. Il.

Subtracting

2.95

(7)

from 2.65 ± 0.02 and

multiplying by

two

gives

q~

j as 0.60 ±

0.041.

Since the

experimental

q~o for this shell is 4.84 ± 0.19

radians,

the

phase

shift for the first O-O shell in the

analysing

interval

(I,e.,

2.3 to 8.5

l~ ')

can be written as :

4lo_o

=

(0.60

± 0.04

)

k

(4.84

± 0.19

(4)

where k is in

l~

' and 4l in radians. The

phase

shift calculated

by

Teo & Lee

[14], using

the

plane-wave approximation

in the k-interval 3.8 to 8.5

l~ ',

leads to

4lo_o

=

0.67 k 3.49.

Using 4lo_ (I,e.,

oxygen central atom

phase shift)

calculated

by

Teo & Lee and

4l_o

calculated

by

McKale

[4], considering

curved-wave formalism, over a shorter k-interval

(I.e.,

from 3.8

l~

to 8.5

l~') yields 4lo_o

= 0.59 k 4.90 (see

Fig. 3a).

-6 00 5 50

, ,

, ,.:'~--.

, :

, ". ',

,

,

".

, 5 00 :~ "..., ,'

', i "., "

-8.00

.. , I [ I, ".. '

". ',

~, ".., ',

'" ',

450 I "., ',

, c

' ,

~,

,

~

£ ".

,

".. ',

i

". "

... , ~ ".

".., '

400 ". ,

".., ',

",

"".. ',

".._ '

".._

....

~

"

3 50

3 oo 16

A"~ A"1

a> i~~

Fig. 3. k-dependent phase shifts

: a) the dashed curve is d~~_~ using d~o and d~

o calculated

by

Teo and Lee [14], where d~

o is calculated

using

the plane-wave

approximation.

The dotted curve is the 4~o-o, where d~o- is the same as in the previous case but d~_~ is calculated

using

the curved-wave

formalism [4]. The solid line is the

experimental phase

shift extracted from this work, I.e. a linear fit to the 4~o-o values over the

corresponding

k-interval (from 2.3 to 8.5 A~ ' b) 4~

~~

values : the dashed curve is calculated using the plane-wave

approximation

[14], while the dotted curve uses the curved-wave

formalism [4]. The solid line is the linear fit to d~

~, obtained in the present study.

The

good

agreement between

theoretically

calculated and

experimentally

determined values for

4lo_o

and the accuracy

[15]

of the calculated

4lo_,

support the claim that the calculated

values for 4l

o,

using

curved-wave formalism, are accurate. The first order

polynomial

fitted to this

phase

shift over the

experimental

range

(I,e.,

2.3 to 8.5

l~ yields 4l_o

=

-0.25k+

0.54,

subtraction of this from the

experimentally

obtained

4lo_o (I.e., expression (4))

allows us to determine the

phase

shift of the central atom over the

experimental

data range as :

4~o_ =

(0.35

± 0.04 k

(5.38

± 0,19

). (5)

This can be

compared

to

4lo_

=

0.40 k 5.10, the value obtained

using

the

phase

shift calculated

by

Teo & Lee over a shorter data range

(I.e.,

3.7 to 8.5

l~

~).

(8)

The

experimentally

obtained value for the first

non-phase

shift corrected O-Ni distance is 1.81 ± 0.02

1 (see

Tab.

Ii.

This value

together

with the

expected

value for this distance

(I.e.,

2.091)

leads to

4lo_~~ =

(0.56

± 0.04 ) k

(0.30

± 0.14

(6)

Subtracting

the

previously

determined

4lo_ (I.e., expression (5))

from this

yields

the nickel

backscattering phase

shift

(see Fig. 3b)

4l

~, =

(0.21

± 0.06

)

k +

(5.08

± 0.24

(7)

While the calculated

[14] phase

shift

using

the

plane-wave approximation

over a shorter k- interval

(I.e.,

3.8 to 8.5

l~ ')

resulted in 4l

~, = + 0.10 k +

4.60,

the calculated

[4]

values

using

curved-wave formalism over a k-interval

equal

to the

experimental

interval leads to 4l ~, = + 0.00 k + 5.00

(see Fig. 3b).

The

disagreement

between

experimental

and theoretical values of

4l_~,

can be

explained partly by

inaccuracies in the calculated

backscattering phase

shift values

(a

common

problem

for

relatively heavy

atoms

[15]),

and

partly by

the fact that the data interval used for EXELFS

analysis

includes the lower part of the k-interval

(see Fig. 3),

where the

reliability

of the

calculated

phase

shift values is limited and the

validity

of the linear

approximation

questionable.

The

phase

shifts values for

4l_o

and

4l_~,

have been calculated

by

McKale

[4]

for two distances

(I.e.,

2,10 and

3.751

in the

case of 4~_o and 2.75 and

41

in the

case of

4l_~,),

therefore linear

interpolations

have been used to obtain the calculated

phase

shifts in

figure

3

corresponding

to 2.95 and

2.091distances respectively.

Since the 4l

N, has not been calculated

(by

Teo & Lee

[14]),

the 4l_N, in

figure

3b has been obtained

by

a linear

interpolation

between the calculated 4l

~~ and 4l

c~.

4.2 INTENSITY ANALYSIS. To the extent that the inelastic mean free

path

of the

ejected

electron

(I.e., A,)

over the actual k-interval can be considered to be

independent

of

k,

the

intensity

of the

peaks

in the RDF can be described

by

the

following relationship

:

>~

fit 2

Ip~ cc S~

fl

e ~'

(8)

r~

where Ip~ is the

experimentally

measured

intensity (I,e., height

of the

peak)

due to the shell

j

(see Tab.

I),

and

S~ is a parameter

describing

the combined effect of the

product

of

backscattering amplitude

and disorder term

[I.e., f~(k) exp(-

2

ml k~)]

on the

intensity.

Since both the first and fifth

peaks

in the RDF

(see Fig. 2b)

contain the same backscatterer

atoms

(I,e., nickel), Sj

should

approximately

be

equal

to S~. On this account

using

equation (8),

the ratio of

lpi /Ip~ yields

2(1"~

rj)

A, =

(9)

In

(lpi N~ ;()

In

(Ip~

N

j

r))

Since all parameters in

equation (9)

are

known,

it is therefore

possible

to determine the value of

A,

as 6.5

1.

It should be mentioned however that the

same calculation could also be carried out

using intensity

of the third

peak (I,e., Ip~)

which also

corresponds

to a nickel shell.

However the low

intensity

and poor resolution of this

peak

suggests the use of the fifth

peak

in the calculations.

(9)

Using equation (8),

the ratio of

Ipj/Ip~

leads to the

following expression.

S~~

lpi N~ r(

i-j r~

= exp 2

(10)

So

I

p~ N

i~

A

This

expression

describes

S~,/So

as a function of, among

others,

the measured intensities

corresponding

to shell I and 2.

Using

this

expression together

with the

already

calculated A, value (I.e.,

6.51), lpi

and

Ip~,

the value of

S~,/So

ratio can be calculated as 0.96.

In order to prove the accuracy of the results, a NiO

EXELFS-spectrum including

the first ten shells around the oxygen as central atom has been simulated over the same k-interval as the

experimental

data (see

Fig. 4a).

The known values of the interatomic

distances,

coordination numbers, and the

previously

determined values of 4~o_,

4l_o

and 4l_~~ have been used. In order

to better

approximate

the real

physical

conditions, a

background equal

to the

remaining

background

in the

experimental

RDF

(Fig. 2b)

has also been included in the simulated data.

Since the resolution is not sufficient to

distinguish

the difference between different disorder parameters due to the difl'erent

shells,

and since before

attempting

to Fourier transform, a

compensation

for the

damping

effect of the

k-dependent

factors has been

applied,

the disorder

parameters has been taken as zero, while the obtained values for A,

(I.e.,

A, =

6.51)

and

S~,/So (I.e., SN,/So

=

0.96)

have been used.

Analysis

of the simulated spectrum indicates that, due to the effects of low resolution and trucation artefacts, the ratios between intensities in the

RDF,

are not

equal

to the

expected

ratios

originating

from the EXELFS modulations. This in turn means that

using

the

experimental

intensities in table I can cause a

relatively large

error in the determination of A, values and

S~~/So

ratios. In other words we cannot use the measured intensities

(I,e., lpi, Ip~

and

Ip~)

without

considering

the effect of low resolution and truncation artefacts.

To solve this

problem

we simulate data with different A~ and

S~,/So

values to find those values which

give

the best fit between the intensities of

Pi,

P2 and P5 (see

Fig. 2b)

in the RDFS of the simulated and

experimental

spectrum. The best fit

(Figs. 4a, b)

is achieved for A, =

6.51

and

SN,/So

= 1,10.

The standard deviation of the measured intensities

(Tab.

I, column

5),

causes error

margins

in determined values of the

A,

and SN~/So ratio.

Considering

these deviations the determined values become

A,

= 6.5 ± 0.7

1

and

SN,/So

=

I.10 ± 0.04

respectively.

Assuming approximately

the same disorder parameters «~, for all atomic shells around the oxygen centre atom, the

S~,/So

becomes a parameter

describing

the average ratio between the

backscattering amplitudes

of the nickel and oxygen atoms

(I,e., f~,/fo). Using

calculated

backscattering amplitudes

based on curved-wave formalism

[4],

the average value of the

fNi/fo

over the

analysed

interval

(I.e.,

2.3 to 8.5

l~

~)

yields

I.

II,

I-e- a very

good

agreement with our

experimental

results.

Using

calculated values based on the

plane-wave approximation [13],

a shorter interval

(I.e.,

3.8 to 8.5

l~'), yields

1.57.

4.3 THE EFFECTS OF MULTIPLE SCATTERING.

Equation

(I takes into account

only

a

single

backscattering

from the

neighbouring

atoms, while the

ejected

electron in

principle

can be backscattered after

multiple scattering [16]

from different atomic shells. Due to the

longer

effective

path (I,e., longer distance), multiple scattering

of the

ejected

electron does not affect either the

position

or the

intensity

of the

peaks corresponding

to the first interatomic distance.

Theoretical and

experimental

work about the

multiple scattering

effect has been

performed by

different authors

[10, 16-20]. Study

of the MS effect on near

edge

structure of the NiO shows

[10]

that

beyond approximately

20 eV above the O-K

edge

the so-called type I -MS is indeed dominant, so that an

expansion

in order to take other types of MS into account is not

necessary. The type I-MS is when

scattering

occurs from atoms which are

arranged

(10)

z 3 « s 7

h-1

a)

3 5 6 la 14 15 is

Angs~om

b)

Fig.

4. a) Simulated EXELFS data including the first ten atomic shells around an oxygen atom, over

the same k-interval as in figure 2a, with A, =

6.5 A and SNI/S~ = I, lo b) magnitude of the FFT of (al.

approximately colinearly

with the central atom so that the

intervening

atom focuses the

ejected

electron wave onto the

backscattering

atom and enhances

scattering

from it. This can, in turn, enhance the

intensity

of some

pealcs

in the

experimental

RDF

(see Fig. 2b).

In NiO, the atoms in the first shell have coordinates

(1/2,

0,

0)

and lie

colinearly

with atoms in the fourth shell with coordinates

(1,

0,

0)

and hence the

multiple scattering

can affect the

intensity

of the P4 in the

experimental

RDF. The second shell has coordinates

(1/2,

1/2,

0)

and

can enhance the

intensity

of the seventh

shell,

which has coordinates

(I,

1,

0).

The

intensity

of the ninth shell, with coordination

(I,

I,

I)

can also be

amplified by

atoms in the third shell,

which have the coordinates

(1/2, 1/2, 1/2).

Since the

positions

of the nearest

peaks,

Pi and P2, have been used in the

phase

shift

analysis, multiple scattering

cannot have any

significant

effect on the results of this

analysis.

In the

intensity analysis lpi, Ip~

and

Ip~

have been used which indicates that even here

multiple scattering

effects can

only

have a very limited effect on the results.

The

relatively high intensity

of the 7th and 9th

peak

in the

experimental

RDF

compared

with the RDF of the simulated data, can be

partly explained by

the noise effects and

partly by

the

multiple scattering

contribution.

(11)

5. Discussion.

EXELFS

analysis

over several short data intervals indicates that the

phase

shift values obtained in this work are valid even for shorter

intervals,

which in turn

implies

that

they

can be used for

typical

EXELFS

analysis.

The curved-wave formalism suggests that the obtained

4lo_o

is valid for r =

2.951

and

4l_~,

for ;

=

2.091.

However calculated

phase

shifts

using

curved-wave formalism show that the

phase

shift

corresponding

to a

particular

distance is also valid within a certain interval

around it without any

significant changes [4].

Comparing

the values for

4l~,

obtained in this work

[I.e., (-0.21±0.06)k

+

(5.08

± 0.24

)]

with

corresponding

values obtained

using

curved-wave formalism

[4] (I.e.,

+ 0.00 k +

5.00),

indicates a

discrepancy

between the

slopes

which cannot be

explained by

the

error

margins.

It should be mentioned here that the

slope

of the calculated

phase

shift in the

higher

energy part is about

0.181,

which is close to the

experimentally

determined value.

The value obtained for

4l_~,

in this work is a linear fit to this

phase

and cannot therefore be

regarded

as the real

phase.

The

discrepancy

between calculated and

experimentally

obtained values

emphasizes

the

necessity

of

performing

EXELFS

analysis

on

specimens

with a known structure in order to determine

experimental

values for the

backscattering phase

shift and

amplitude, particularly

for relative

heavy backscattering

atoms.

As discussed

here,

the

intensity

of a

peak

in the RDF can be influenced

by

several factors

by

noise as well as

by

low resolution and truncation artefacts due to the

neighbouring peaks.

For an accurate

intensity analysis,

these effects should therefore be considered and accounted

for. In other words when resolution in the RDF is

relatively

low

(which

is the

typical

case in the

EXELFS),

the effect of

overlapping peaks

makes the use of back Fourier

filtering procedure

rather inaccurate and therefore we have

adapted

the method described in this work for data

analysis.

6.

Summary.

For an accurate EXELFS

intensity analysis,

the effect of low resolution and truncation

artefacts must also be considered. The method used in this

work, analysing

simulated spectra

corresponding

to the same conditions as the

experimental

ones, can be used to test the

reliability

of the obtained results.

In

spite

of the limitations above, an accurate

intensity analysis

is still

possible by fitting

the RDF from a simulated

spectra

to the RDF of the

experimental

data.

Since

relying entirely

on calculated values of the

backscattering phase

shift of

relatively heavy

backscatterer atoms can lead to

large

errors, EXELFS

analysis

of known

specimens

seems to be necessary to determine

experimental

values of these effects.

The average value of the mean free

path

of

inelastically

scattered electrons in the

typical

EXELFS

region (I,e.,

2.3-8.5

l~')

in NiO has been estimated to be A, = 6.5 ± 0.7

ji.

The linear

approximation

for the nickel

backscattering phase

shift in the

analysed

k-interval

(I.e.,

2.3-8.5

l~ ')

at a distance of about

2.091,

can be described

by

4l ~, =

(0.21

± 0.06 k +

(5.08

± 0.24

).

The

phase

shift

corresponding

to the O-O shell in the same k-interval, where the backscatter oxygen atom is at about

2.951,

as well as a linear fit to it can be described

by

4lo_o

=

(0.60

± 0.04 k (4.84 ± 0.19 )

(12)

The linear fit to the oxygen central atom

phase

shift in the

analysed

k-interval can be described

by

4lo_

=

(0.35

± 0.04 k

(5.38

± 0.19 ).

References

[ii KINCAID B. M., MEIXNER A. E., PLATzMAN P. M.,

Phys.

Rev. Lett. 40 (1978) 1296.

[2] STERN E. A.,

Phys.

Rev. B10 (1974) 3027.

[3] TED B. K., LEE P. A., SIMMONS A. L., EISENBERGER P., KINCAID B. M., J. Am. Chem. Soc.. 99 (1977) 3854.

[4] MCKALE A. G., KNAPP G. S., CHAN S.-K., Phys. Rev. B 33 (1986) 841.

[5] EGERTON R. F., Electron Energy-Loss

Spectroscopy

in the Electron

Microscope,

p. 225.

[6] BOUCHET D., COLLIEX C., FLORA P., KRIVANEK O., MORY C., TENCL M., Microsc. Microanal.

Microstruct 1 (1990) 443.

[7] KURATA H., LEFtVRE E., COLLIEX C.. BRYDSON R., Phys. Ret'. B (1993) to be published.

[8] LEV losifovicH MIRKIN, Handbook of X-Ray

Analysis

of Polycrystalline Materials, Translation from the Russian by J. E. S. Bradley. B. Sc., Ph. D.

[9] TAFRESHI M. A., BOHM Ch., CSILLAG S., Microsc. Microanal. Microstract. 1(1990) 199.

[10] VVEDENSKY D. D., PENDRY J. B., Phys. Rev. Lett. 54 (1985) 2725.

[I ii LEAPMAN R. D., GRUNES L. A., FEJES P. L., SiLcox J., EXAFS

Spectroscopy techniques

and applications, B.-K. Teo, D. C. Joy Eds, ISBN 0-306-40654-3 (1980) p. 217-239.

[12] TAFRESHI M. A., CSILLAG S., YUAN Z. W., BOHM C., Microsc. Microanal. Microstract. 2 (199 ii 515.

[13] STERN E. A., SAYERS D. E., LYTLE F. W., Phys. Ret'. B II (1975) 4836.

[14] TED B. K., LEE P. A., Ab initio calculations of amplitude and phase functions for extended x-ray

absorption

fine structure spectroscopy, J. Am. Chem. Soc. 101 (1979) 2815.

[15] BUNKER B. A., An X-Ray

Absorption Study

of Tetrahedral Semiconductors, University of

Washington

(1980) p. 174.

[16] TED B. K., EXAFS

Spectroscopy

Techniques and

Applications.

B. K. Teo, D. C. Joy Eds., ISBN 0-306-40654-3 (1980) p. 13.

[17] LEE P. A., PENDRY J. B., Phys. Rev. B II (1975) 2795.

[18] STERN E. A., HEALD S. M., in Handbook on Sunchrotron Radiation, E. E. Koch Ed, I (North- Holland, New York, 1983) p. 986.

[19] BUNKER G., STERN E. A., Phys. Rev. Lett. 52 (1984) 1990.

j20] NATOLI C. R., BENFATTO M., J. Phys. Colloq. Franc-e 47 (1986) C8 II-

Références

Documents relatifs

The energy stores in the z modes cannot come from charges as it is stored in the wide space between the plates. This energy can only come from the time derivation of the

Given a learning task defined by a space of potential examples together with their attached output values (in case of supervised learning) and a cost function over the

On the other hand, in a dilute scattering medium containing high- contrast inclusions, each scattering event introduces a possibly large phase shift which depends on the

- Taking into account the strong interrelation between thermal disorder and surface potentials in ionic crystals, such potentials have been determined for AgCl

Using this hypothesis the Matthiessen rule for the reciprocal mean free path leads to a factorization of K\ with respect to its temperature and concentration dependence.. This is

other examples in that they represent a situation as bounded. This obser vation has important consequences for the expression of temporal relations in past time sentences.

In section 3, we will discuss the coupling of ddCOSMO with an existing code and review what are the quantities that the existing code needs to assemble in order to compute

Vertrauen reduziert in zwischen- menschlichen Beziehungen die Komplexität der Um- welt und stellt damit eine Strategie dar, welche die Handlungsfähigkeit des Einzelnen auch