XAFS Data Analysis using LCF / PCA NSLS XAFS Short Course Nov 13-15, 2014
XAFS Data Analysis using LCF / PCA
NSLS XAFS Short Course Nov 13-15, 2014
1. A XAFS of a sample of multi-component 2. Good knowledge on the sample & each component 3. Need to quantify fractions
− Mn2+
− Mn3+
− Mn4+
6520 6540 6560 6580 66000.0
0.4
0.8
1.2
1.6
2.0
Norm
alize
d Ab
sorp
tion
Coef
ficie
nt
Energy (eV)
V-XAFS series from an in-situ reaction 1. XAFS series show systematic change 2. Little information on the process or distinct
reaction phases. 3. Need to understand kinetics and composition
Mn-XAFS spectrum from a sample
LCF vs. PCA
LCF: Linear Composition Fitting PCA : Principal Component Analysis
5460 5480 5500 55200.0
0.4
0.8
1.2
Norm
alize
d Ab
sorp
tion
Coef
ficie
nt
Energy (eV)
• A Factor Analysis Method • Applied to a series of XAFS • Obtain number of distinct phases in
this group of spectra • Advantages:
– No prior knowledge is required – Detect the intermediate, if applied to
in-situ reaction; assuming complete reaction.
• Requirements: – Large # of spectra – Data point # > spectrum # – Good data quality, minimum noise as
possible – Data are collected and processed in
the same manner.
• A Specification Method – Applied to an spectrum for a sample – Quantify the composition for the
sample of multi-component. • Advantages:
– Straightforward implementation – Applicable to both single spectrum
& XAFS series • Requirements:
– Good knowledge on the sample(s) and the references
– Good reference spectra: • Experiment • data processing
Same condition to the sample
NSLS XAFS Short Course, Nov 13-15, 2014
Software and Packages
Freeware: • DAthena – B. Ravel http://bruceravel.github.com/demeter/
• SixPack – S. Webb: http://home.comcast.net/~sam_webb/sixpack.html
Distribution Permission / Purchase Option: • XAMath – S. R. Wasserman (Mathematica 4. & above) • WinXAS – T. Ressler: http://www.winxas.de/
NSLS XAFS Short Course, Nov 13-15, 2014
Result: Fe2+/Fe3+ fractions
NSLS XAFS Short Course, Nov 13-15, 2014
Data One Spectrum (XANES, EXAFS)
Least Square Fitting
Linear Combination Fitting
Reference Spectra
Result Fraction
of Each Component
LCF analysis: fits data to LC of spectra of Fe2+ and Fe3+
Determine the Fraction of Each Component that can be combined to reproduce the experimental spectrum within statistical error
7100 7120 7140 7160 7180-0.2
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
Isosbestic points
Norm
alize
d Ab
sorp
tion
Coef
ficie
nt
Energy (eV)
001 002 003 004 005 006 007
Fe K-edge
0.5C Charge
Result: Fe2+/Fe3+ fractions
2.5
3.0
3.5
4.0
4.5
5.0
0.0 0.2 0.4 0.6 0.8 1.00.0
0.2
0.4
0.6
0.8
1.0 LiFePO4
FePO4
X in Li1-xFePO4
Frac
tion
of L
iFeP
O 4 or F
ePO 4
0.5 C Charge
Fe2+
Fe3+
1 2 3 4 5 6 7
Mix
ing
frac
tion
Fe2+
Fe3+
Single data group
Sequence: Batch mode
data
data
NSLS XAFS Short Course, Nov 13-15, 2014
A
B
LCF in DAthena
E
A) Load Data B) Display C) LCF setup D) Actions
E) LCF Results
D C
Principal Component Analysis
NSLS XAFS Short Course, Nov 13-15, 2014
Original Data XAFS series
Linear Algebra
• Principal components • Mathematical constructs
derived original data • Each component is unique
Abstract Standards
PCA Mathematical Matrix
Vector = Spectrum
Target Transformation Real Standards
• Examine each standard • Reproduce spectra
Determine the number of components that can reproduce a series of experimental spectra within statistical error.
PCA: Algorithm
Singular value decomposition (SVD) algorithm – Matrix: [D] = m x n ( m: row; n: column; m ≥ n) – [D] = [E] x [V] x [W]t
– Eigenvectors: Columns of [E]
– Eigenvalues: Diagonal elements of [V]
NSLS XAFS Short Course, Nov 13-15, 2014
mnmm
n
n
n
XXX
XXXXXXXXX
............
..
..
..
21
33231
22221
11211
mnmm
n
n
n
eee
eeeeeeeee
............
..
..
..
21
33231
22221
11211
nnv
vv
v
...0......0..0..00..0
33
22
11
nnn
n
n
ww
wwwwww
.............
..
..
1
22221
11211
= x x
D E V W = x x t
Ressler, Environ. Sci. Technol. 2000, 34, 950
m x n n x n n x n
PCA: Data Requirement & Selection
[D] = m x n XAFS data Data points in each spectrum : m Number of spectra n
NSLS XAFS Short Course, Nov 13-15, 2014
• Number of data points are larger than numbers of XAFS spectra • All XAFS spectra are interpolated to the same energy grid • Number of spectra are larger than number of principal components
XAFS data requirement
data
A
D B
F
C
A) Load Data B) Display C) PCA setup D) PCA results E) Actions F) Target Transformation
NSLS XAFS Short Course, Nov 13-15, 2014
DAthena PCA Demo
Demeter © Bruce Ravel
E
PCA 1. PCA + LCF 2. PCA + FEFF Fit
Applications
• Mixed phases: – Environmental Science – Biomedical Science
• Sequential processes: – Chemical Transformation – Catalysis – Battery
NSLS XAFS Short Course, Nov 13-15, 2014
LCF
5460 5480 5500 5520 55400.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
Norm
alize
d Ab
sorp
tion
Coef
ficie
nt
Energy (eV)
NSLS XAFS Short Course, Nov 13-15, 2014
Combined PCA and LCF
in-situ battery lithiation
Experiment: Real-time lithiation of V-based battery material
XANES data: 1. Pre-edge change 2. Edge shift 3. Above-edge & EXAFS
NSLS XAFS Short Course, Nov 13-15, 2014
• Maximum number of components equals the number of original spectra – Data reconstructed using all components reproduce exactly the
original spectra. • Principal components contain real spectra features • Other components: data noises & errors
PCA: Components
0 5 10 15 20 25 30 35 40
0
1
2
3
4
5
6
A
Row Numbers5460 5480 5500 5520 5540 5560-0.7
-0.6
-0.5
-0.4
-0.3
-0.2
-0.1
0.0
0.1
AU
Energy (eV)
1 2 3 4
0 5 10 15 20 25 30 35 40-5
0
5
10
15
20
25
30
35
40
Eige
n Va
lue
Component #NSLS XAFS Short Course, Nov 13-15, 2014
Scree Plots
PCA: Principal Components • Contains the spectra features • Can reproduce the data • Determined by:
- Eigen-values - Error Analysis - Reproduction of data
Eigen Value # 1: 36.679264 # 2: 0.294768 # 3: 0.024383 # 4: 0.001016 # 5: 0.000139
Component 1 ~ 4
5440 5460 5480 5500 5520 5540 5560-0.2
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
data
energy
data fit residual
2-compent fit
5440 5460 5480 5500 5520 5540 5560
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
data
energy
data fit residual
3 Component fit
0.0000
0.0002
0.0004
0.0006
0.0008
0.0010
0.00120.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6
Resid
ual
Residual using 3 components
0 50 100 150 200 2500.0000
0.0002
0.0004
0.0006
0.0008
0.0010
0.0012
Resid
ual
Capacity
Residual using 2 components
Data Reproduction
• Examine reproduction by using varied number of components – Original data vs. Reproduced spectrum – Reproduction residuals
• Very useful for intermediate detection in a multiple-phase reaction
NSLS XAFS Short Course, Nov 13-15, 2014
2-compoent fits and residuals 3-component fits and residuals
A reaction involving 3 phases
Target Transformation
• Examine the standards • A correct standard can be fitted by principal components
– Determine whether a selected standard is legitimate – Eliminate impossible standards
NSLS XAFS Short Course, Nov 13-15, 2014
Wrong Standard Reasonable Standard
NSLS XAFS Short Course, Nov 13-15, 2014
0.0 0.4 0.8 1.2 1.6
0.0
0.2
0.4
0.6
0.8
1.0
Starting Phase Intermediate Phas Ending Phase
Mixin
g Fra
ction
# Li
Obtain Mixing Fractions - Composition of a sample. - Indicate reaction dynamics of a chemical change
LCF: Mixing Fractions
Conclusion for Analysis: 1. Relates the mixing fraction with inserted Li# 2. Only need to analyze data for 3 reaction phases to understand the
chemical compositions and structures
8960 8980 9000 9020 9040 9060 9080
0.0
0.2
0.4
0.6
0.8
1.0
1.2
Norm
alize
d ab
sorp
tion
coef
ficie
nt
Energy (eV)
Combined PCA & FEFF Fit
Experiment: quick XAFS @ Cu Ke-edge on reduction/re-oxidation of 20% Cu/CeO2
Reduction at 200oC • XANES (8959 ~ 9059 eV) showing structural evolution • No isosbetic points
17 Q. Wang, et al. J. Chem. Phys., 129, 234502 (2008).
NSLS XAFS Short Course, Nov 13-15, 2014
in-situ catalysis
Reproduction: determine Intermediate
18 NSLS XAFS Short Course, Nov 13-15, 2014
3-component fit: • Presence of Intermediate
0 400 800 1200 1600 20000.000
0.004
0.008
0.012
0.016
Resid
uals
(a.u.
)
Reaction time t(s)
1)()()( =++ tztytx
)()()()()()(),( EtzEtyEtxtE PIR χχχχ ++= t = t* 𝒚 𝒕 = 𝟏
Intermediate Phase
0 400 800 1200 1600 20000.000
0.004
0.008
0.012
0.016
Resid
uals
(a.u.
)
Reaction time t (s)
t=t*
2-component fit: • Uneven residual 660< t <900s • Intermediate phase
FEFF fit to EXAFS of Intermediate
19 NSLS XAFS Short Course, Nov 13-15, 2014
8960 8980 9000 9020 9040 90600.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
0 2 4 6 8 10
-1
0
1
k2χ(
k), Å
-2
k, Å-1
Norm
alize
d ab
sorp
tion
coef
ficie
nt
Energy (eV)
A Intermediate phase:extracted by PCA
FEFF
)(223)3(eff2
20 22
)(342sin)()( kRk
iiiii
ii
iii eekkkRkfkR
nSk λσδσχ −−
+−=
0 1 2 3 4 5 60.0
0.2
0.4
0.6
0.8
1.0
FT M
agni
tude
إ ,-3
Data Fit
B
r,Å
Intermediate Analyzed by FEFF
NCu-O = 2.2 ± 0.5
NCu-Cu = 5.8±3.5
RCu-O = 1.86 ±0.02 Å
RCu-O = 3.07 ±0.02 Å
Conclusion for Analysis: 1. intermediate is detected. 2. Reaction dynamics is elucidated by applying PCA followed by LCF. 3. Intermediate structure was determined by combined use PCA and FEFF fit.
Cu2O
2
12
1.85 Å
3.02 Å
Summary
NSLS XAFES Short Course, Nov 13-15, 2014
• LCF and PCA are important speciation and quantification methods applied to XAFS spectra
• LCF: – simple and straight forwards – require knowledge of number and types of references.
• PCA: – mathematically accurate – no priori assumption for standards – requires additional information to relate the mathematical vectors
to physically meaningful spectra. • Combined use of PCA and LCF would provide insights on reactions • Broadly applied in a variety of research fields
Some References
• B. Ravel, “http://bruceravel.github.io/demeter/ • S.R. Wasserman, et al . “EXAFS and principal component analysis: a new shell game” , J . Synchrotron Rad. (1999) 6, 284 • A. I. Frenkel, et al. “Phase speciation by extended x-ray absorption fine structure spectroscopy “ , J. Chem. Phys. (2002) 116, 1473193 • S. Beauchemin et al., “Principal component analysis approach for modeling sulfur K-XANES spectra in humic acids”, Soil. Sci. Soc.Am.J. (2002) 66,83. • T. Ressler et al., “Quantitative speciation of Mn-bearing particulates emitted from autos burning mcp-Mn gasolines using XANES spectroscopy”, Environ. Sci. Technol., (2000) 34, 950.
NSLS XAFS Short Course, Nov 13-15, 2014