WingZ version (2 or 3 components) CLS1: two-component model with 26 wavelengths |
Excel and OpenOffice Calc versions (5 components) Five components with 100 wavelengths |
[Operating instructions] [Instructor's Notes] [Cell definitions and equations] [Student assignment handout]
This is a simulation of UV multicomponent absorption
spectroscopy of a mixture consisting of two or three
absorbing components whose absorption spectra are known (but whose
concentrations in the mixture are unknown and are to be
determined). The simulation allows you to create simulated
mixtures of the components and then to analyze the mixture using
the observed total absorption spectrum and the absorptivities of
the components, using the "classical
least-squares" (CLS) method. Realistic amounts of random
noise can be added to the "observed" mixture spectrum in order to
demonstrate the precision obtainable with this method.
Operating instructions (Excel and
OpenOffice Calc versions)
In these versions, the simulated mixture is defined by the peak
amplitudes, positions, and widths of the 5 components that you
type into the table in columns A-F, rows 6-8. The noise level is
determined by typing into Noise, in column B, row 10. The
spreadsheet automatically plots the spectra of the five components
(each component with a different color) and the spectrum of the
mixture with noise added, then automatically calculates the
amplitude (concentrations) of each component in the simulated
mixture and the percent error (relative difference between the
true and measured amplitudes). You can test the limits of this
method by placing the peak positions of the components closer
together, increasing their widths, or increasing the noise,
thereby creating a more challenging problem to solve. If you are
curious how this calculation is performed, the details are laid out step-by-step
in rows 123 to 158.
Operating instructions
(WingZ version)
The relative concentrations of the components are controlled by
the colored sliders on the right. The green slider controls the
concentration of component 1, the yellow slider controls the
concentration of component 2, and, for the three-component
version, the blue slider controls the concentration of component 3
(or of the background absorption, in CLS1b and CLS4). The spectra
of the components separately are shown on the graph as the solid
lines matching the color of the slider that controls it. These
concentrations are the "known unknowns" for the simulation. Every
time you change these sliders, you are simulating a different
sample mixture.
The small "noise" slider below the compute button determines the amount of total absorbance noise. Its range is from zero (left-most position) to 0.001 (right-most setting). This amount of random noise is added to the mixture spectrum to approximate the effect of random instrumental noise (flicker, photon, and detector dark noise). Without this noise, the results of the simulation would be unrealistically "perfect". Every time you change this slider, you are simulating the effect of a different instrument (with more or less random instrumental noise).
The Spectral Difference Index slider controls the similarity of the shapes of the component spectra. When this is zero, there is no difference between the shapes of the components. This control basically controls the overlap between the spectra. The purpose of this control is to allow you to simulate different mixture systems of varying degrees of similarity so you can determine how much difference between spectra is necessary for good measurement. Every time you change these sliders, you are simulating a different chemical system (different compounds).
When you click on the Compute button, a classical least
squares computation is performed, based on the simulated measured
mixture spectrum with noise added (red dots) and using the known
shapes of the component spectra. That is, it attempts to calculate
the component concentrations from the mixture spectrum. The
results of this calculation are shown in the table below the graph
as the "Measured concentrations". The "Actual" concentrations
(that you set using the sliders) is also shown, as well as the
relative percent difference between them. Every time you click on
this button, you are simulating another reading of the mixture
spectrum and a repeat of the CLS calculation. The results will be
slightly different each time because of the noise.
Note: The classical least squares method works properly only if the absorbance is directly proportional to concentration according to the Beer-Lambert Law. If non-linearity occurs because of polychromaticity or stray light, a more advanced computational method, the transmission fitting method, may be employed.
Download links for the WingZ version:
CLS1 has two absorbing components measured at 26 wavelengths
(readings taken every 0.4 nm). Download link: CLS1.wkz
CLS1b is like CLS1, with the addition of (uncorrected) background
absorption to the mixture spectrum. Download link: CLS1b.wkz
CLS2 has two absorbing components measured at 51 wavelengths
(readings taken every 0.2 nm). Download link: CLS2.wkz.
CLS3 has three absorbing components measured at 26 wavelengths
(readings taken every 0.4 nm). Download link: CLS3.wkz.
CLS4 is like CLS3, except that the third component is a flat
background absorption. Download link: CLS4.wkz.
Wingz player application and basic set of simulation modules, for
windows PCs or Macintosh
RegressionDemo.xls and RegressionDemo.ods (for Excel and OpenOffice Calc, respectively)
Other related simulations:
Curve Fitting B:
Multicomponent Spectroscopy
Signal-to-noise ratio of absorption
spectrophotometry
Transmission fitting method
(a multicomponent method that does not require Beer's Law
linearity)
Instructor's note on the different versions.
Some mathematical background on the classical least squares method applied to multicomponent spectroscopy can be found in Curve fitting B: Multicomponent Spectroscopy.
CLS1 can be compared to CLS2 to demonstrate the slightly increased precision resulting from the use of a greater number of wavelengths (better averaging out of the random noise). CLS1 can be compared to CLS3 to demonstrate the additional computations required to handle one more component (and the increasing computation time, noticeable only on slow computers). CLS1 can be compared to CLS1b to demonstrate what happens if there is absorption occurring in the mixture solution that is not included in the model (namely, gross errors). CLS1b can be compared to CLS4 to demonstrate how the error resulting from a flat background absorption can be corrected by including the background absorption as a third component and performing a 3-component computation.
These models can be operated using only the mouse-activated on-screen sliders, for use in a lecture-demonstration environment with a computer video projection system, where it is difficult to use the keyboard data entry in a darkened room. However, the concentrations of the components can also be typed into the table below the graph (in the row labeled "Actual concentrations") in order to specify concentrations beyond the limits of the sliders.
Pull down the Sheet menu and the Matrix sub-menu
and notice the Transpose, Invert, and Multiply
operations which are needed for the CLS calculation. "Compute" button script that performs the
Classical Least Squares matrix calculations):
repaint off
select range A107..B132
Copy
select range C107
paste values
Transpose
select range C107..AB108
select more range A107..B132
select more range F110
Multiply
select range F110..G111
select more range F113
Invert
select range F113..G114
select more range C107..AB108
select more range F116
Multiply
select range F116..AE117
select more range B30..B55
select more range F119
Multiply
recalc
repaint on
select range A1
1. Set the Spectral Difference Index slider to -20 and both component concentrations to 1.0. Would it have been possible to measure either of these components by a simple single-wavelength calibration? That is, are there wavelengths where only one of the components absorb?
2. How does the noise level effect the average accuracy of measurement? What happens if you to the error if you make the noise zero?
3. Does varying the concentration of component 1 while holding component 2 constant cause a systematic change in the measured concentration of component 2?
4. What is the smallest concentration of component 1 you can measure with a relative accuracy of 1% when the concentration of component 2 is held at 2? How does the noise effect this?
5. Set both concentrations to 1 and the noise to 0.001 (maximum). Vary the Spectral Difference Index slider to make the two spectra more similar. How similar can the be and still be measured to a relative accuracy of 1% . How does the noise effect this?
6. Set the noise to zero (left-most position) and force the Spectral Difference Index to exactly zero by typing a 0 into the gray cell to the right of the = sign in "Spectral Difference Index =". This simulates two components with absolutely identical spectra, but with zero noise. What happens when you click on the Compute button. Why?
7. To see how the computations are performed, scroll the window to the right (click on the right-pointing arrow on the scroll bar at the bottom of the window) until you see the "Step-by-step calculation procedure", which shows the actual matrices as gray blocks, each labeled, for each step of the calculation. Which matrices change when you recalculate after changing the concentration sliders? The noise slider? The Spectral Difference Index slider? Why?
8. Open up the file called "CLS2". This is similar to CLS1 except that readings are taken every 2 nm, for a total of 51 wavelengths (w =51). Can you expect any differences in calculation speed or accuracy? Do you observe any difference? Explain. (Hint: There is a difference, theoretically, but it's small and you may miss it).
[RETURN] to the Spreadsheet
Simulations Index.
[RETURN] to Chem 623 WingZ page.