Measurements Lecture

Complex numbers
Matrix solutions
Singular Value Decomposition (SVD)

NC State University
Absorption spectroscopy

Beer-Lambert law
Allowed and forbidden transitions
Transition metal vs. lanthanide ions
 Beer-Lambert Law

A is the absorbance.
() is the extinction coefficient.
The unit of () is M-1cm-1.
c is the concentration (M). xx dx
d is the pathlength (cm).
I0 I’ I’+dI’ I
The exponential attenuation
of the intensity is shown
d
in the Figure.
Fundamentals of the extinction coefficient
The absorption cross section, A
The absorption cross section has units of area (cm2) gives
the probability for absorption. It depends on the
transition dipole moment M12 but also on a shape factor.
For allowed transitions it is the Franck-Condon factor, FC.

The absorption cross section is proportional to the well

known extinction coefficient.

The extinction coefficient has units of M-1cm-1.

Electromagnetic radiation interacts
with ground and excited state
Pyridine
molecule
E E = electric field
B = magnetic field

H H

H H
N

B E = Eocos(2t)
 The dipole moment
The dipole moment operator is

 = e z ir i
i

where zie is the electronic charge at

a nucleus and ri is a vector from an
arbitrary origin. Along the x-direction


=e  x dx
*

0
 Examples of ground state
dipole moments

The ground state dipole moment of

hydrogen halides can be calculated from
the fractional charges −
+ 

1 Debye = 3.33 x 10-30 Cm

+ - R(pm) (x 10-30 Cm)

HF 0.42 0.42 91.7 6.37  1.9 D
HCl 0.16 0.16 127.5 3.44  1.0 D
HBr 0.11 0.11 141.4 2.64  0.8 D
HI 0.05 0.05 160.9 1.40  0.4 D
 Transition dipole moment
The transition dipole moment results from the interaction
of electromagnetic radiation with the molecule, where
E = E0cos(2t). The hamiltonian for interaction is:

𝐻 = −𝑀12 𝐸0 cos(2𝜋𝜈𝑡)

where 𝑀𝑔𝑒 is the transition dipole moment

𝑀12 = 𝑒 න Ψ1∗ 𝑥Ψ2 𝑑𝑥

 Selection Rules
A transition will be allowed only if the transition
dipole moment integral is non-zero. The general
rules are:

Electronic: l =  1, m = 0
Rotational: J =  1, M = 0
Vibrational: v =  1

This can be seen by looking at atomic or

molecular orbitals of the ground and excited state.
 The interaction of electromagnetic
radiation with a transition moment
The electromagnetic wave has an angular momentum of 1.
Therefore, an atom or molecule must have a change of 1
in its orbital angular momentum to conserve this quantity,
as shown for the hydrogen atom:

Look at the nodal plane of the electromagnetic radiation.

Electric vector
of radiation
ℓ=1
ℓ=0

ℓ=1
 Copper sulfate spectroscopy
Forbidden d-d transitions
The ground state electron configuration of Cu is [Ar]3d104s1.
However, Cu(II) has a configuration [Ar]3d9. In the hexahydrate
there is an octahedral ligand field. It is an unusual case because the
water molecules all have the same Cu-O bond length.
The electronic transition is
LaPorte forbidden because
Both ground and exited state
d-orbitals. d-d transitions are
broad because of a
vibronic coupling mechanism.
 Lanthanide spectroscopy
Forbidden f-f transitions
The ground state electron configuration of Nd is [Xe]4f46s2.
However, Nd(III) has a configuration [Xe]4f3. Nd transitions observed
in the absorption spectrum start from the 4I9/2 ground state.
Transitions to the following
2S+1L levels can be observed:
J
4F , 2H , 4F , 4F , 4S ,
3/2 9/2 5/2 7/2 3/2
4F , 2H 4 2 4
9/2 11/2, G5/2, G7/2, G7/2,
4K 4 2 4 2
13/2, G9/24, K15/2, G11/2, D3/2
and 2G9/2. The spectrum shown
is a high-resolution spectrum of
Nd-doped LaTiO3. f-f transitions
are symmetry forbidden.
 Why are forbidden transitions ever observed?
An effect in the environment or higher lying states must break the
symmetry.
1. Vibronic coupling permits observation of d-d transitions, but
also, symmetry forbidden −* states. These formally forbidden
transitions are weak, i.e. not very intense.
2. Judd-Ofelt perturbation theory predicts magnetic dipole and
quadrupole terms that break the f-f symmetry and permit the
weak observed bands.
 Complex numbers

Introduction to complex numbers

The Euler relation
Matrix rotation
 Complex numbers
General complex number
Complex conjugate

Complex product

The complex product is just

another statement of the
Pyrthagorean theorem.

Complex plane
 The complex plane
On the complex plane we can represent a point as

relationship to Cartesian coordinates

This is known as the Euler relation

Complex conjugate
on the plane
 Matrix rotation

A counter-clockwise rotation is given by

A clockwise rotation is given by

Spectral libraries
Commercial spectral library example
Wide range of applications
Wide range of applications
Government and public spectra libraries
 Application of UV-vis spectral library
We can use a spectral library to determine the identity of
compounds in mixtures in solution. We will obtain standard
spectra of pure components. Using a concentration-dependent
calibration, the molar absorptivity can be obtained. This can
be used to analyze the spectrum of an unknown concentration
in a mixture.
Creating a spectral library from UV-vis data
Raw data for Cu(NO3)2
Expansion of the baseline of Cu(NO3)2
Baseline corrected data for Cu(NO3)2
Linear regressions for four wavelengths
The slope is 𝜖, the molar absorptivity (extinction coefficient).

832 nm

662 nm

575 nm
424 nm
Raw data for Nd(NO3)3
Expansion of the baseline of Nd(NO3)3
Baseline corrected data for Nd(NO3)3
Linear regression calibration for Nd(NO3)3
The slope is 𝜖, the molar absorptivity (extinction coefficient).

575 nm

832 nm 662 nm 424 nm

Raw data for Ni(NO3)2
Expansion of the baseline for Ni(NO3)2
Baseline corrected data for Ni(NO3)2
Linear regression calibration for Ni(NO3)2
The slope is 𝜖, the molar absorptivity (extinction coefficient).
Raw data for K3Fe(CN)6
Expansion of the baseline for K3Fe(CN)6
Baseline corrected data for K3Fe(CN)6
Linear regression calibration for K3Fe(CN)6
The slope is 𝜖, the molar absorptivity (extinction coefficient).
Concentration of unknown using absorption spectroscopy
The wavelengths selected for analysis were 424 nm, 575 nm,
662 nm, and 832 nm. Find wavelengths the optimize the
ability to detect a specific molecule without interference.

0 0 1.829 879.5
𝜀= 0 6.85 0.307 0
3.59 0 1.82 0
11.5 0 0.535 0

0 0 −0.028 0.0956
𝜀 −1 = 0 0.145 −0.027 0.0084
0 0 0.603 −0.1885
0.00113 0 −0.0012 0.00039

𝒄 = 𝜺−𝟏 𝑨

But wait! Aren’t we throwing away most of the data?

Multiple algebraic equations

Application to UV-vis measurement

The matrix inverse (square matrix only)
This is the simple-minded solution
 Problem in linear analysis using analytical chemistry
A student determines absorbance of a solution known to contain 4
strongly absorbing dyes (labeled dye A, B, C and D) at four different
wavelengths λ = 495,420, 470 and 590 nm.
The extinction coefficients ελ(dye) for the 4 compounds at these 4
wavelengths are known (units lit/mol):

A B C D

395 13500 2000 28000 7200

420 180000 4500 14000 8000
470 700 700 3000 32000
590 8000 70000 400 1350
 Problem in linear analysis using analytical chemistry

The absorbance values in a 1 cm pathlength are:

A395 0.5
A420 0.6
A470 0.3
A590 0.4

What are the four concentrations?

(Hint: Absorbance A= ελLc is an additive quantity, so you can write out the
problem as a set of linear equations. Then write this as a matrix formula
and see if you can solve it by matrix algebra.
 Method using matrix analysis
The equations have the form

𝐴1 = 𝜀11 𝑐1 + 𝜀12 𝑐2 + 𝜀13 𝑐3 + 𝜀14 𝑐4

𝐴2 = 𝜀21 𝑐1 + 𝜀22 𝑐2 + 𝜀23 𝑐3 + 𝜀24 𝑐4
𝐴3 = 𝜀31 𝑐1 + 𝜀32 𝑐2 + 𝜀33 𝑐3 + 𝜀34 𝑐4
𝐴4 = 𝜀41 𝑐1 + 𝜀42 𝑐2 + 𝜀43 𝑐3 + 𝜀44 𝑐4

We can write these compactly in matrix form as:

𝑨 = 𝜺𝒄
Where the knowns are the vector A of absorbances and the matrix of the
extinction coefficients. We can solve for the concentrations using the matrix
inverse:
𝜺−𝟏 𝑨 = 𝜺−𝟏 𝜺𝒄
Which tells us that
𝒄 = 𝜺−𝟏 𝑨
 Multiple linear regression

Uses the entire spectrum

LINEST applied to spectra
The basis of spectral library applications
 Generalization of linear least squares

The data consist of multiple sets of n observations 𝑥𝑖 , 𝑦𝑖 𝑛 of

p points each ( i = 1...p). If there is a single type of observation,
then we can use ordinary least squares (linear regression)

𝑦𝑖 = 𝛽1 𝑥𝑖 + 𝛽0 + 𝜀𝑖

where 𝛽1 is the slope and 𝛽0 is the intercept. 𝜀𝑖 represents the

error. This method can be expanded to simultaneous solution of
multiple linear relationships.

𝑦𝑖 = 𝛽0 + 𝜀𝑖 + 𝛽1 𝑥𝑖1 + 𝛽2 𝑥𝑖2 + 𝛽3 𝑥𝑖3 … + 𝛽𝑗 𝑥𝑖𝑗

 Application to a UV-vis spectral library
In this application the meaning of 𝑥𝑖𝑗 could be a collection of j of
known molar absorptivities. Then 𝑦𝑖 represents a mixture of
species that have in the collection j. The 𝛽𝑗 are the concentrations
of these species. This method can also be used to identify a species
of unknown origin.

𝐴𝑖 = 𝛽0 + 𝜀𝑖 + 𝑐1 𝜖𝑖1 + 𝑐2 𝜖𝑖2 + 𝑐3 𝜖𝑖3 … + 𝑐𝑗 𝜖𝑖𝑗

where 𝑐𝑗 are the concentrations and 𝜖𝑖𝑗 are the molar absorptivities
(extinction coefficients). Note that 𝜀𝑖 and 𝜖𝑖𝑗 represent the error
and molar absorptivity, respectively.
Generalization of inverse matrix method
If the data conform to a linear model, then we can write

𝑇
𝑦𝑖 = 𝑥𝑖𝑗 𝛽𝑗 + 𝜀𝑖

where  is p x 1 vector of unknown parameters and the i are

random variables (errors). For the total of n observations this
set of equations can be written in matrix notation as

𝒚 = 𝑿𝑻 𝜷 + 𝜀𝑖

This is known as multiple linear regression (MLR) or also as

multilinear regression
 Least square vs. SVD
We can write this equation in terms of the application to a spectral
library for UV-vis (or FTIR) spectroscopy.

𝑨 = 𝝐𝑻 𝒄 + 𝜀𝑖

There are two ways to solve this problem.

1. Generalize ordinary least squares from one slope to many
simultaneous slopes.
2. Find the pseudo-inverse of the molar absorptivity matrix.

If the matrix were square, i.e. if we had the same number of

wavelengths as species, we use matrix inversion. Finding the
pseudo-inverse involves using singular value decomposition
(SVD).
Using LINEST for multiple
regression
Application to determination of
concentration of unknowns
 Setting up the multiple regression
for the UV-vis experiment
If we treat the LINEST as a multiple regression, then we can use the data for
absorbance
𝐴 = 𝜀ℓ𝑐
Where ℓ = 1 𝑐𝑚 and thus we can simply write it as 1. Thus, a plot of A vs. 𝜀
should be linear with a slope equal to c, the concentration. If we have multiple
concentrations then we can set up the regression as
𝜀11 𝜀21 𝐴1
𝜀12 𝜀22 𝐴2
𝜀13 𝜀23 𝐴3
𝜀14 𝜀24 𝐴4
And the slopes of the multiple regression will be c1 and c2. The intercept should
be zero. The standard errors can be used as the error estimate in this solution
of the problem.
When defining the LINEST for this
problem you need a 3 x 5 array
Implement the LINEST using
<crtl><shift><enter>
 Definitions of the solution array

m1 = slope of regression 1
m2 = slope of regression 2
b = intercept
Next line is standard error in each
R = coefficient of determination
sey = standard error in y
F = F statistic
df = degrees of freedom
Singular value decomposition

Non-symmetric eigenvalue problem

Eigenvectors in U are basis spectra
Eigenvector in V are species
 Singular value decomposition
Singular value decomposition takes a rectangular matrix of
gene expression data (defined as A, where A is a n x p matrix)
in which the n rows represents the genes, and the p columns
represents the experimental conditions. The SVD theorem
states:
Anxp= Unxn Snxp VTpxp

Where UTU = Inxn VTV = Ipxp (U and V are orthogonal)

The columns of U are the left singular vectors; S has singular

values and is diagonal; and VT has rows that are the right
singular vectors. The SVD represents an expansion of the
original data in a coordinate system where the covariance
matrix is diagonal.
 A

A
Example of a data matrix
Raw data

U1

VT1
Raw data

U2

VT2
Raw data

U1

VT1
Raw data
U2

VT2
Using the pseudoinverse of the data
matrix to solve for coefficients
Simulated data: Four Gaussian functions
Coefficients used to construct trial data

0.4 0.2 0.3 0.1

4 species of unknown concentration
Matrix inversion limited to four wavelengths

If we limit ourselves to only the four wavelengths at the

peaks we can determine the concentrations in a mixture
containing 4 unknowns using matrix inversion.
−1
𝑐𝑠 𝜀1 (𝑠) 𝜀1 (𝑡) 𝜀1 (𝑢) 𝜀1 (𝑣) 𝐴1
𝑐𝑡 𝜀2 (𝑠) 𝜀2 (𝑡) 𝜀2 (𝑢) 𝜀2 (𝑣) 𝐴2
𝑐𝑢 = .
𝜀3 (𝑠) 𝜀3 (𝑡) 𝜀3 (𝑢) 𝜀3 (𝑣) 𝐴3
𝑐𝑣 𝜀4 (𝑠) 𝜀4 (𝑡) 𝜀4 (𝑢) 𝜀4 (𝑣) 𝐴4
Matrix inversion limited to four wavelengths
We could use more wavelengths (of which we have a
spectrum full!) but then the matrix ε with extinction
coefficients will not be a square, so we should use a
generalized inverse like (εTε)-1 εTA rather than a simple ε-1A.
Methods
1. Multiple linear regression
2. Singular Value Decomposition (SVD)
 SVD can calculate the pseudo-inverse
Calculate the SVD of the original matrix A:
𝐴 = 𝑈 𝑆 𝑉𝑇
To calculate the pseudoinverse of A, you ccan use the
components of the SVD directly,
𝐴+ = 𝑉 𝑆 + 𝑈 𝑇
where 𝑆 + is the pseudoinverse of the diagonal matrix 𝑆.
Since the eigenvalue matrix 𝑆 is a diagonal matrix, its
pseudoinverse is it inverse.
 Important detail: use asymmetric UT
The 𝑈 matrix needs to be truncated to include only the
number of components in the eigenvalue or 𝑆 matrix. To
make this explicit, if 𝐴 consists of 4 spectra with 600
wavelengths each then 𝐴(𝑟𝑜𝑤𝑠, 𝑐𝑜𝑙𝑢𝑚𝑛𝑠) has dimensions
𝐴(600,4). The SVD decomposition gives
𝐴 600,4 = 𝑈 600,600 𝑆 4,4 𝑉 𝑇 (4,4)
The transpose of 𝑉 𝑇 is trivial and is called 𝑉. However, the
transpose of 𝑈 should only include the first four columns
(basis spectra). Thus, the transpose matrix has dimensions
𝑈 𝑇 4,600 .
 Using the pseudo-inverse
to obtain concentrations
Thus, the A pseudoinverse has dimensions
𝐴+ (4,600) = 𝑉(4,4) 𝑆 + (4,4) 𝑈 𝑇 (4,600)
A spectrum consisting of unknowns that has 600
absorbance values is a column vector X(600). To solve for
coefficients in the 𝐴 600,4 matrix that give rise to the
X(600) matrix we solve the matrix equation
𝑐(4) = 𝐴+ (4,600)𝑋(600)
Data matrices
SVD result U S VT
UT truncated to number of
experimental spectra
S+ inverse
V = VT transpose
 A+ Pseudoinverse using SVD matrices

𝐴+ (4,600) = 𝑉(4,4) 𝑆 + (4,4) 𝑈 𝑇 (4,600)

In this simulated example the coefficients
are reproduced exactly.
𝑐(4)
= 𝐴+ (4,600)𝑋(600)