2. Data analysis - umich.edujphgroup/XAS_Course/Harbin/Lecture2.pdf · Bruce Ravel's notes on using...
Transcript of 2. Data analysis - umich.edujphgroup/XAS_Course/Harbin/Lecture2.pdf · Bruce Ravel's notes on using...
1
Lecture 2Principles of EXAFS and
XANES data analysisHow to avoid common errors in EXAFS and XANES analysis
Tutorials and other Training MaterialBruce Ravel's notes on using FEFFIT for data analysisDaresbury Laboratory lectures on data analysis (EXCURV98)Grant Bunker's XAFS tutorialsFrenkel et al on comparing PCA with other methodsChantler (Uni. Melbourne) on the absolute determination of x-ray
absorption
ProgramsESRF Software catalogXFITEXAFSPAKAthena and Artemis
FEFFIT and IFEFFITDL EXCURVWinXAS
2
where
and
-10
-5
0
5
10
0 2 4 6 8 10 12
EX
AF
S•k
3
k (Å -1 )
Amplitude
coordinationnumber
Frequency
bondlength
Phase
scatterer
Shape
scatterer
k N sAs k S02
kRas2 exp 2k 2as
2 exp 2Ras sin 2kRas as k
k 2me E Eo 2
k E o E o E
E s E
b E
Z ± 10N ± 1R ± 0.02 Å
k N sAs k S02
kRas2 exp 2k 2as
2 exp 2Ras sin 2kRas as k +E0, b ,
Radial structure only
Fourier transformation can be used to visualize frequencies contributing to EXAFS
k N sAs k S02
kRas2 exp 2k 2as
2 exp 2Ras sin 2kRas as k
R2sinA kkk
22
10 RR
Å R
3
Common errors in EXAFS analysis
• Least-squares minimization
• Fourier Filtering
• Resolution
• The job of a least-squares fitting program is to give you the best (smallest deviation) solution, not to give you the right solution.
• If you see something, it tells you something; if you see nothing, it tells you nothing.
4
R
Carbonscattering
Iterative refinements are especially susceptible to multiple minima
Pd
P P
CCH2C
Me
ca b Me
N
Phosphorusscattering
Clot, et al., J. Am. Chem. Soc 2004, 126, 9079-9084
Minima give very different structures but nearly identical EXAFS
M-P
M-C
sum
Nfree vs. Nobs
Nobs ~ 300
(k = 0.05 Å; kmax= 15 Å)
Nfree ~ (2 k R ) /
5
Anatoly Frenkel
EXAFS Data Collection and Analysis Workshop, NSLS
Further explorations of the difficulty of multiple minima
N and 2
R and E0
# Pt foil, T=200 K
guess S02 = 0.9 guess ss1 = 0guess dr1 = 0guess th1 = 0guess e0 = 0
data = ptfoil-200avk.chiout = ptfoil-200avk
rmin = 2.1 rmax = 3.3kmin = 2 kmax= 20 w = 2 dk=2
!% 1st path: e0shift 1 e0 amp 1 S02path 1 p1.datid 1 SS Pt-Pt(1), r=2.7719delr 1 dr1sigma2 1 abs(ss1)third 1 th1
0 2 4 6 8 10 12 14 16 18 20 22-2.5
-2.0
-1.5
-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
2.5 200 K
k2 (k
), Å
-2
k, Å-1
0 2 4 6 8 10 12 14 16 18 20 22-1.0
-0.5
0.0
0.5
1.0
673 K
k2 (k
), Å
-2
k, Å-1
How to model metal (Pt) foil data:
)(2sin
e)()(
333
4
2eff2
20
22
kkCkr
kfkr
NSk k
6
0 2 4 6 8 100.0
0.2
0.4
0.6
0.8
1.0
FT
Mag
nitu
de,
Å-3
r, Å0 2 4 6 8 10
0.0
0.2
0.4
0.6
0.8
1.0
FT
Mag
nitu
de,
Å-3
r, Å
S02 = 0.747364 0.044827ss1 = 0.009313 0.000385 dr1 = 0.005673 0.007571 th1 = 0.000259 0.000076e0 = 8.069036 0.594447
0 2 4 6 8 100
1
2
3
4
5
FT
Mag
nitu
de,
Å-3
r, Å0 2 4 6 8 10
0
1
2
3
4
5
FT
Mag
nitu
de,
Å-3
r, Å
S02 = 0.874046 0.033009ss1 = 0.003609 0.000106dr1 = -0.009912 0.003630th1 = -0.000030 0.000019e0 = 8.286025 0.590203
This is not physically reasonable….
What caused S02 to be different at 200 K and 673 K?
- correlation with other fit variables:
)(2sin
e)()(
333
4
2eff2
20
22
kkCkr
kfkr
NSk k
Fit Results
)(2sine)()( 333
42eff2
20
22
kkCkrkfkr
NSk k
0 2 4 6 8 10 12 14 16 18 20 22-2.5
-2.0
-1.5
-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
2.5 200 K 300 K 473 K 673 K
k2(k) of Pt foil: temperature dependence
k2 (k
), Å
-2
k, Å-1
One possible solution:a multiple-data-set (mds) fit.
What variables are not expected to change at different temperatures?
How to break the correlation?
E0, N
)exp(1
)exp(1
2 E
E2
222
T
Td
ds
2s E,
7
0 2 4 6 8 100
1
2
3
4
5
200 KF
T M
agn
itude
, Å
-3
r, Å
0 2 4 6 8 100.0
0.5
1.0
1.5
2.0
2.5
3.0
FT
Ma
gnitu
de,
Å-3
r, Å
300 K
0 2 4 6 8 100.0
0.2
0.4
0.6
0.8
1.0
FT
Ma
gnitu
de,
Å-3
r, Å
673 K
MDS fit results
ss011 = 0.000533 0.000093 theins1 = 189.743073 2.311668
s02 = 0.836704 0.017830
dr11 = -0.011222 0.002248 dr12 = -0.009361 0.003034 dr13 = -0.000354 0.003642 dr14 = 0.006588 0.004801
th11 = -0.000035 0.000013 th12 = -0.000017 0.000022 th13 = 0.000113 0.000033 th14 = 0.000267 0.000060
e0 = 8.064717 0.271896
Physical (chemical, engineering, mat.science, life science etc.)
reality checks:
1) Debye temperature: 240 K for PtAs obtained (through ): 253(3) K
2) Static disorder : ~03) Corrections to model distances: ~0
4) Thermal expansion: evident
5) S02: reasonable (between 0.7 and 1.0)
2s
E
0 2 4 6 8 100.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
FT
Mag
nitu
de,
Å-3
r, Å
473 K
How to model XAFS data in nanoparticles?
A priori knowledge or a working hypothesis must exist(the “zero” approximation)
otherwise: the transferability of amplitude/phase will not work!)
1) Hemispherical2) Crystal order3) Size: about 20 Å
What information can be obtained from 1st shell EXAFS analysis?
1) Size of the particle (via N)2) Distances, thermal vibration,
expansion3) Static disorder (icosahedral?
surface tension?)
Rel
ativ
e A
bund
ance
Nanoparticle Diameter (A)
0 5 10 15 20 25 30 35 40 45 50
Ave
rage
Fir
st-S
hell
Coo
rdin
atio
n
4
5
6
7
8
9
10
11
12
(Å)
Ave
rage
CN
8
0 2 4 6 8 100.0
0.5
1.0
1.5
2.0
2.5
200 K
FT
Ma
gnitu
de
, Å
-3
r, Å0 2 4 6 8 10
0.0
0.2
0.4
0.6
0.8
1.0
FT
Ma
gn
itud
e,
Å-3
r, Å
473 K
0 2 4 6 8 100.0
0.5
1.0
1.5
FT
Mag
nitu
de,
Å-3
r, Å
300 K
0 2 4 6 8 100.0
0.2
0.4
0.6
FT
Ma
gni
tud
e, Å
-3
r, Å
673 K
MDS fit (1shell) to the nanoparticles EXAFS
- Coordination number is now guessed (a variable)
- is fixed to be equal to that in Pt foil EXAFS
- E0 is fixed to be equal to that in Pt foil EXAFS
20S
ss011 = 0.001676 0.000177 theins1 = 191.842209 3.893480
n1 = 7.879327 0.197850
dr11 = -0.015809 0.003938 dr12 = -0.011870 0.002064 dr13 = -0.008558 0.003883 dr14 = -0.000845 0.004875 th11 = -0.000017 0.000030 th12 = 0.000055 0.000019th13 = 0.000159 0.000047th14 = 0.000421 0.000079
Summary
• The more variables you control, the more likely you are to obtain a unique solution
• Multiple data sets (elements, temperature, concentration, time, etc.) almost always help
• Conclusions are only as good as your model
9
Fourier filtered data can be distorted
M-O at 2.0 and 2.2 Å give two apparently well resolved peaks in FT
0
5
10
15
20
25
30
35
40
0 1 2 3 4
Fou
rier
tran
sfor
m m
agni
tude
R + (Å)
10
Fitting each filtered peak gives the appearance of M-O and M-S EXAFS
-4
-2
0
2
4
6
8
10
4 6 8 10 12
EX
AF
S•k
3
k (Å-1)
"M-O" peak "M-S" peak
Effect of limited resolution on EXAFS
11
EXAFS resolution is ~ /2k =
0.13 Å
-2
0
2
4
6
8
10
12
0 2 4 6 8 10 12
EX
AF
S•k
3
k (Å-1)
R1=1.75 Å, R
2=2.00 Å
R1=1.85 Å, R
2=2.00 Å
R1=1.85 Å, R
2=2.00 Å
R=1.875 Å
R=1.925 Å
R=1.925 Å,damped
Clark-Baldwin, et al. "The limitations of X-ray absorption spectroscopy for determining the structure of zinc sites in proteins. When is a tetrathiolate not a tetrathiolate?" J. Am. Chem. Soc. 1998, 120, 8401-8409.
A case study in data under-determination
12
Zn EXAFS is remarkably insensitive to changes in ligation.
-10
-5
0
5
10
15
20
25
30
2 4 6 8 10 12
EX
AF
S•k
3
k (Å-1)
ZnS2N
2
ZnS3N
ZnS4
0
10
20
30
40
50
0 1.5 3 4.5 6 7.5
FT
Mag
nitu
de
R + (Å)
ZnS2N
2
ZnS3N
ZnS4
ZnSR
SR
L1
L2
Z ± 10 ???
XANES spectra are sensitive to ligation
0
100
200
300
400
9650 9660 9670 9680 9690 9700
Nor
mal
ized
Abs
orpt
ion
(cm
2 /g)
Energy (eV)
Peptide
ZnS2N
2
ZnS3N
ZnS4
0
100
200
300
400
9650 9660 9670 9680 9690 9700
Nor
mal
ized
Abs
orpt
ion
(cm
2 /g)
Energy (eV)
Inorganic
ZnS2N
2
ZnS3N
ZnS4
but show greater variation between different compounds than with changes in ligation.
13
Zn-S and Zn-N EXAFS signals are approximately out of phase
-3
-2
-1
0
1
2
3
2 4 6 8 10 12 14
k3 E
XA
FS
k (A-1)
Zn-S Zn-N
The observed EXAFS for mixed S/N sites is dominated by Zn-S scattering
-15
-10
-5
0
5
2 4 6 8 10 12 14
k3 E
XA
FS
k (A-1)
Zn-S Zn-N ZnS2N
2
14
One solution is to measure data over wide k range(ZnS2N2 inorganic)
0
10
20
30
40
50
60
70
0 1.5 3 4.5 6 7.5
FT
Mag
nitu
de
R + (Å)
k=2-11 Å-1
k=2-12 Å-1
k=2-13 Å-1
k=2-14 Å-1
k=2-15 Å-1
k=2-16 Å-1
Note – R ~ 0.25 /2k = 0.25 Å kmin ~ 6.3
High resolution EXAFS is required to reliably distinguish Zn-S from Zn-N …
-15
-10
-5
0
5
2 4 6 8 10 12 14
k3 E
XA
FS
k (A-1)
Zn-S Zn-N ZnS2N
2
15
…and even with high resolution data, extremely high signal/noise ratios are required
to detect Zn-N in the presence of Zn-S
-15
-10
-5
0
5
2 4 6 8 10 12 14
k3 E
XA
FS
k (A-1)
Zn-S Zn-N ZnS2N
2
-100
-50
0
50
20 30 40 50 60 70 80 90 100% sulfur
P i(%)
Peptide
ZnS2N
2
ZnS3N
ZnS4
-100
-50
0
50
20 30 40 50 60 70 80 90 100
Pi(%
)
% sulfur
Inorganic
ZnS2N
2ZnS
3N
ZnS4
It is possible to reliably distinguish between ZnS4, ZnS3N, and ZnS2N2
if variable para-meters are carefully controlled.
Note that fit quality always improves for mixed ligation fits.
Clark-Baldwin, K et al, J. Am. Chem. Soc. 1998, 120, 8401-8409.
16
In addition to Pi, 2 depends on ligation
-2
0
2
4
6
8
10
20 30 40 50 60 70 80 90 100
2 (Å
2 x10-3
)
% sulfur
Threshold energy changes apparent ligation
50
60
70
80
90
100
4 6 8 10 12 14 16
Sm
ax
ZnS2N
2
ZnS3N
ZnS4
E0,S
(eV)
Calibrated E0
5 eV is enough to change a sulfur into a nitrogen!
17
Distinction between S and N rests largely on phase, which depends on E0
-8
-6
-4
-2
0
2
4
4 6 8 10 12
k3 E
XA
FS
k (A-1)
Zn-S Zn-N
Geometry determination
Multiple scattering in EXAFS
18
0
2
4
6
8
10
12
14
0 1 2 3 4 5 6 7
Fou
rier
Tra
nsfo
rm M
agni
tude
Radius + (Å)
N
NH
Zn
Outer shell scattering can provide ligand identification and geometric information
Multiple scattering makes EXAFS sensitive to angular arrangement of ligands
N
NH
Zn
19
M CC
C
C
C
N
N
N
N
N
C
N
M N
O
O
N
O
OC
O R1
R2
R3
R4
1
2
Pa
ram
ete
rsR
Info
rma
tion
However, ability to reliably determine geometry is limited
XANES spectra contain useful information regarding structure
• Quantitative comparisons (e.g., titration) requires accurate normalization.
• Correction for various artifacts (self-absorption) requires accurate normalization.
• Common normalization procedures were developed for extracting EXAFS and do not necessarily work well for XANES.
20
Normalization Schemes
Conventional
MBACK
Conventional normalization is sensitive to background shape
MBACK shows much weaker sensitivity
21
Conventional normalization is sensitive to range of data
0
0.5
1
1.5
9300 9400 9500 9600 9700 9800 9900 10000 10100
980098209840986098809900992099409960998010000100201004010060
Energy (eV)
MBACK shows only slight sensitivity for Emax150 eV above edge
-50
0
50
100
150
200
250
300
350
9300 9400 9500 9600 9700 9800 9900 10000 10100
980098209840986098809900992099409960998010000100201004010060
Ab
sorp
tion
(cm
2 /g)
Energy (eV)
22
Conventional normalization misses changes in XANES
MBACK reveals subtle changes when thiolate is added
4 possible difference specta –should all be the same
-0.04
-0.03
-0.02
-0.01
0
0.01
0.02
0.03
0.04
9700 9750 9800 9850 9900 9950 10000 10050 10100
Conventional
Energy (eV)
23
With new normalization, difference signal is detectable
-10
-5
0
5
10
9700 9750 9800 9850 9900 9950 10000 10050 10100
Energy (eV)
Dependence of XANES on Oxidation State
0
400
800
6535 6555 6575 6595Energy (eV)
Nor
mal
ized
Abs
orpt
ion
II/II
II/III
III/III
III/IV
24
Edge “energy” is poorly defined
-80
-60
-40
-20
0
20
40
0
200
400
600
800
1000
1200
d(A
bs)/
dEA
bsorption
Simon Bare, NSLS workshop
25
Factor Analysis in Chemistry, 2nd Ed. John Wiley & Sons, NY, 1991
Coulston et al., Science 275 (1997) 191-193
T
1
T
k
l
kkk
l sX
VSUX
vu
is the closest l-rank matrix to X(see http://public.lanl.gov/mewall/kluwer2002.html)
EnergyTime