Information • Textbooks • Media • Resources

Analysis of the Infrared Spectra of Diatomic Molecules W

Richard W. Schwenz
Department of Chemistry and Biochemistry, University of Northern Colorado, Greeley, CO 80639

William F. Polik*
Department of Chemistry, Hope College, Holland, MI 49423; *[email protected]

One of the most common physical chemistry laboratory displacement of the bond as vibrational energy increases.
exercises involves the infrared spectrum of a heteronuclear While some of the spectroscopic literature includes more
diatomic molecule, usually HCl, from which the internuclear molecular constants, these five are sufficient to describe the
separation, the vibrational force constant, and higher-order infrared spectra recorded for most diatomic molecules in the
molecular constants can be determined (1–5). The reasons undergraduate physical chemistry laboratory. Thus, a
for frequent use of this exercise include the relevance to the complete expression for the rovibrational energy levels of a
harmonic oscillator and rigid rotor models taught in quantum diatomic molecule is given by
mechanics, the high accuracy of the spectroscopic measurements ∼
(achieving essentially literature values for the line positions), E (v, J ) = ωe(v + 1⁄ 2 ) +Be J( J + 1) – ωexe(v + 1⁄ 2 )2 –
(3)
the existence of several isotopic variants of HCl allowing tests De J 2( J + 1)2 – αe(v + 1⁄ 2 ) J( J + 1)
of the Born–Oppenheimer approximation, and the oppor-
tunity for students to perform a rigorous data reduction and where all the constants have wavenumber units.
error analysis to determine molecular parameters from spec- The lines in an absorption or emission spectrum arise
troscopic data. It is this last feature with which this paper is from transitions between two energy levels. The wavenum-
concerned. In particular, several methods of data reduction ber ν∼ of a transition can be obtained by subtracting the wave-
are examined and an improved method using the multiple number energy of the lower state from that of the upper state
linear regression feature available on most microcomputer ∼ ∼
ν∼(v′, J′,v′′, J′′) = E (v′, J ′) – E (v′′, J ′′) (4a)
spreadsheet programs is presented that eliminates some of the
problems inherent in other methods. ∼ ∼
E (v′, J ′) – E (v′′, J ′′) =
Theory ωe[(v′ + 1⁄ 2) – (v′′ + 1⁄ 2 )] + Be[ J′( J′ + 1) – J ′′( J ′′ + 1)] –
All textbooks, laboratory manuals, and original spectros- ωexe[(v′ + 1⁄ 2)2 – (v′′ + 1⁄ 2)2] – De[ J ′2(J′ + 1)2 – J ′′2(J′′ + 1)2] –
copy papers treat the infrared spectra of diatomic molecules
similarly. The molecules are first treated in the harmonic αe[(v′ + 1⁄ 2 )J′( J′ + 1) – (v′′ + 1⁄ 2)J ′′( J ′′ + 1)] (4b)
oscillator–rigid rotor approximation with energy levels given by
where the single prime indicates the upper state and the
E(v,J ) = [(h/2π)(k/µ)1/2](v + 1⁄ 2 ) + [h 2/8π2Ie ] J( J + 1) (1) double prime indicates the lower state. In order to apply
where v is the vibrational quantum number, J is the rotational eq 4b, appropriate quantum numbers (v ≥ 0, J ≥ 0) and
quantum number, h is Planck’s constant, k is the vibrational selection rules (∆J = ±1) must be used.
force constant, µ is the reduced mass, and Ie = µre2 is the For the HCl molecule, the infrared absorption spectrum
moment of inertia with re being the internuclear separation. consists of about 20 pairs of closely spaced lines for both the
Dividing through by hc yields the molecular energy and fundamental (v′′ = 0 → v′ = 1) and first harmonic (v′′ = 0 →
parameters in wavenumber units (cm{1), which is indicated by v′ = 2) bands for each of the H and D isotopic variants. The
the use of a tilde,
∼
E(v,J )/hc = E (v, J ) = ωe(v + 1⁄ 2 ) + Be J( J + 1) (2a)
where
ωe = (k/µ)1/2/2 πc (2b)
and
Be = h/8 π 2cI e (2c)
are the harmonic vibrational frequency and rotational con-
stant, respectively.
Three correction terms are then usually introduced to
this energy level expression. The first of these, the anharmo-
nicity constant ωexe, arises from the departure of the vibra-
tional energy levels from the harmonic approximation, in part
to allow for bond dissociation. The second, the centrifugal
distortion constant De, arises from the stretching of the
molecule as it rotates faster. The third, the vibration–rotation Figure 1. Infrared absorption spectrum of the fundamental band
interaction constant α e, arises from the increase in mean for HCl.

1302 Journal of Chemical Education • Vol. 76 No. 9 September 1999 • JChemEd.chem.wisc.edu

 Information • Textbooks • Media • Resources

spacing within a pair arises from the slight difference of the and the difference between adjacent transitions by
molecular constants for the 35Cl and 37Cl isotopic variants.
∆ν∼(m) = ν∼(m + 1) – ν∼(m) = (2Be – 5αe) – 4αem (8)
Figure 1 presents the fundamental band of HCl. The overall
appearance of the spectrum is due to the ∆J = ±1 rigid rotor The fundamental band origin ω10 may be combined with knowl-
selection rules, which result in a P branch (∆J = {1) and an edge or measurement of the first overtone ω20 = 2ωe – 6ωexe
R branch (∆J = +1) of almost equally spaced pairs of lines to determine the values of ωe and ωexe.
surrounding the band origin. The spacing between each pair of The successive differences method readily introduces the
lines is approximately 2Be. The relative intensity of each pair concept that vibration–rotation interaction is responsible for
of lines arises from the 2J + 1 degeneracy of the rigid rotor the unequal spacing of spectroscopic transitions in the HCl
energy levels, the Boltzmann distribution of molecules in the spectrum. However, this method results in an inaccurate value
ground state rotational levels, and the rotational absorption of Be due to neglect of the centrifugal distortion term De,
intensity factors (squared dipole transition moment) (6 ). The often yields inconsistent values of B e and αe from the inde-
relative intensity of the lines within each pair is due to the pendent analyses of the fundamental and overtone bands,
relative isotopic abundance of 35Cl and 37Cl. and does not correctly calculate the uncertainties in Be and
α e due to statistical correlation between the coefficients of m
Analysis of the Spectrum and m2. It should be noted that the successive difference
method can be extended to determine De by retaining it in
Average Spacing the energy level expression to yield an equation for ν∼(m) that
The simplest method for analyzing the infrared spectrum is cubic in m (6, 7 ). This equation can be analyzed by least
of a diatomic molecule relies on the harmonic oscillator–rigid squares fitting to cubic polynomial or by taking the second
rotor energy level expression of eq 2. In this approximation, successive difference ∆2ν∼(m); however, this method is not used
the spacing between all the lines is averaged, this average is in any physical chemistry laboratory manual owing to its
set equal to 2Be, and the moment of inertia Ie and internuclear complexity.
separation re are calculated. The force constant k for the bond is
Combination Differences
simply calculated from the wavenumber of the band center.
This method has the virtue of simplicity. It also has the dis- An analysis technique used by spectroscopists prior to
advantage of being vastly oversimplified, as it neglects the modern computational abilities was the combination differ-
additional terms present in eq 3 that do not occur in eq 2. ences method (6, 8–10). In this method, it is recognized that
This method is suitable for homework problems in textbooks, certain pairs of spectral lines share a common upper or lower
but not for the quantitative analysis of an actual spectrum. state, allowing an energy difference between states to be
calculated. For example, the R( J ′′ = 0) and P( J ′′ = 2) transi-
Successive Differences tions both have J ′ = 1 in the upper state; thus, the difference
It is readily apparent from the spectrum of HCl that the in transition energies for the R( J ′′ = 0) and P( J ′′ = 2) lines is a
lines are not equally spaced, but that the spacing between direct measurement of the energy separation between J ′′ = 0
lines in the P branch expand and lines in the R branch contract and J ′′ = 2 in the lower state. Formulas involving the rota-
as J increases. The method of successive differences found in tional constants of the upper and lower vibrational states and
laboratory manuals (2, 5, 6 ) accounts for this observation by the centrifugal distortion constant can be derived by taking
recognizing that the transition wavenumbers in the spectrum the expression for spectroscopic transitions, substituting upper
can be approximated as a parabolic function of an index vari- state quantum numbers using the rigid rotor selection rules,
able m where m = J ′′ + 1 for the R branch and m = {J ′′ for identifying appropriate pairs of combination difference states,
the P branch. Neglecting the effect of De (since it contrib- and eliminating the upper or lower state energy (1, 3, 4, 6 ).
utes significantly only at high J ) and using this substitution, The two simplest resulting formulas are
the transition wavenumbers in the fundamental band are
ν∼R( J ′′) – ν∼P( J ′′) = 2(2J ′′ + 1)Bv′ – 4[( J ′′ + 1)3 + J ′′3]De (9)
found to be given by
and
ν∼(m) = ωe – 2ωexe + (2Be – 2αe)m – αem2 (5)
ν∼R( J ′′) – ν∼P( J ′′ + 2) =
Typically, a successive difference plot of the wavenumber dif-
2(2J ′′ + 3)B0 – 4[( J ′′ + 1)3 + ( J ′′ + 2)3]De (10)
ference between adjacent transitions versus m is constructed
allowing the determination of Be and αe from a linear fit to
where Bv ′ = Be – αe(v′ + 1⁄ 2 ) is the rotational constant of the
∆ν∼(m) = ν∼(m + 1) – ν∼(m) = (2Be – 3αe) – 2αem (6) upper vibrational state and B0 is the rotational constant of the
lower vibrational state. A formula involving the band origin
The fundamental band origin ω10 = ωe – 2ωexe may be calcu- can be similarly derived:
lated by rearranging eq 5 to isolate ωe – 2ωexe and substituting
the values of Be and αe along with the observed transition ν∼R( J ′′) + ν∼P( J ′′ + 1) =
wavenumbers for several values of m near the band center. 2[v ′ωe – v ′(v ′ + 1)ωexe] + 2( J ′′ + 1)2(Bv ′ – B0) (11)
The intercept (m = 0) of a plot of ν∼(m) – (2Be – 2αe)m + αem2
versus m yields the best estimate of the fundamental band To determine the values of B1 and De from the fundamental
origin ω10. Analysis of the first overtone band is similar, with band, for example, eq 9 is linearized by dividing through by
the transition wavenumbers given by 2(2J ′′ + 1) and a plot of [ν∼R( J ′′) – ν∼P( J ′′)]/[2(2J ′′ + 1)] versus
{4[(J ′′ + 1)3+ J ′′3]}/[2(2J ′′ + 1)] is constructed. The intercept
ν∼(m) = 2ωe – 6ωexe + (2Be – 3αe)m – 2αem2 (7) equals B1 and the slope is {De. Similar linearized plots of eqs

JChemEd.chem.wisc.edu • Vol. 76 No. 9 September 1999 • Journal of Chemical Education 1303

 Information • Textbooks • Media • Resources

10 and 11 yield B0, another determination of De, ω10, and J ′′(J ′′ + 1), (v′ + 1⁄ 2 )2 – (v′′ + 1⁄ 2 )2, J ′2(J ′ + 1)2 – J ′′2(J ′′ + 1)2,
the difference between upper and lower state rotational constants (v′ + 1⁄ 2 )J ′(J ′ + 1) – (v′′ + 1⁄ 2 )J ′′(J ′′ + 1), in which each formula
B1 – B0. Analysis of the first overtone band yields B2, two is calculated from the quantum numbers in the first five
additional values of De, another value of B0, ω20, and B2 – B0. columns of the spreadsheet. The most straightforward pro-
The molecular constants Be and αe are then determined from cedure is to create these formulas in the first row and then
the intercept and slope of a plot of Bv ′ versus v ′ + 1⁄ 2 , and De copy them into the rest of the rows.
is taken to be its average determined value. The constants ωe The remaining step is to initialize and perform the
and ωexe are determined from the values of ω10 and ω20. multiple linear regression. The regression variables must be
The difficulties encountered by students using this defined in the spreadsheet. The y (or dependent) variable is
method are legendary. Problems include understanding the the wavenumber of the transitions. The x (or independent)
derivation of the combination difference formulas, the immense variables consist of the five computed columns defined in the
amount of data that must be handled to produce the three previous paragraph, which contain the quantum number
graphs per vibrational band per isotope required to determine expression associated with each molecular constant. Since the
the molecular constants, and resolution of inconsistent values form of the energy expression does not contain a constant
of molecular constants determined from independent fits. In term, the regression should have a defined zero value for the
addition, spectroscopists recognize that the combination intercept. Once the regression is performed, the output will
differences method introduces errors when missing or blended include five x coefficients with their associated errors and the
lines occur, alters the weights of lines of differing J values standard error of the regression analysis. The five regression
because of the form of the linearized equation, and completely coefficients are ωe, Be, { ωexe, {De, and { αe, which correspond
neglects the statistical correlation between the molecular directly to the energy expression coefficients of eq 3. The
constants in the upper and lower states (11). coefficient errors are the errors in each molecular constant.
The standard error is a measure of how well the observed
A New Method of Analysis transitions fit eq 4b. The equilibrium internuclear separation,
vibrational force constant, and their associated uncertainties
We suggest here that a different analysis method be used are easily calculated from the regression output using eqs 2b
by physical chemistry students, which takes advantage of the and 2c.
relative simplicity of the energy expression in eq 3 and the
ability of a microcomputer spreadsheet program to perform Fundamental Only
multiple simultaneous regression on a large amount of data. The preceding treatment assumes that measurements
The preceding data analysis methods can be frustrating to have been made on both the fundamental and first overtone
students because of the repetitive calculations, the graph bands of HCl. Should measurements be made only on the
production, and the indirect method of obtaining molecular fundamental band, then information is not present to deter-
constants. In addition, the methods themselves suffer from mine both the harmonic frequency ωe and the anharmonicity
inherent problems in determining molecular constants when ωexe, and the matrix inversion in the regression routine will
the energy expression coefficients are highly correlated. The new fail. In this case, the known vibrational quantum numbers,
method dramatically simplifies the effort of data reduction, v′ = 1 and v′′ = 0, should be incorporated into the expres-
provides a more direct connection between experimental data sion for transition wavenumbers, eq 4b, resulting in
and molecular constants, and yields results that are more
accurate and precise than the traditional methods. ν∼(v′=1, J ′,v′′=0, J ′′) =

Fundamental and Overtone ωe – 2ωexe + Be[ J ′( J ′ + 1) – J ′′( J ′′ + 1)] –

In our method, the wavenumber and the associated De[ J ′2( J ′+ 1)2 – J ′′ 2( J ′′+ 1)2] –
quantum numbers of each transition are entered into a spread- αe[ 3⁄2 J ′( J ′+ 1) – 1⁄2 J ′′( J ′′ + 1)] (12)
sheet row. The data occupy five columns of the spreadsheet,
one for the transition wavenumber, one each for the upper state Three data columns are needed for the values of ν∼,
vibrational and rotational quantum numbers, and one each for J ′, and J ′′. Three formula columns are used to compute
the lower state vibrational and rotational quantum numbers. J ′( J ′ + 1) – J ′′( J ′′ + 1), J ′ 2( J ′ + 1)2 – J ′′ 2( J ′′ + 1)2, and
3⁄ J ′( J ′ + 1) – 1⁄ J′′( J ′′ + 1). Since eq 12 contains a constant
Thus, five pieces of data are entered for each observed tran- 2 2
sition: ν∼, v′, J ′, v′′, J ′′. The same information is entered into term, that is, a term not multiplied by a quantum number
a new row for each transition observed in the spectrum. expression, the regression is performed with three indepen-
Five columns are subsequently computed that allow the dent variables that correspond to the three formula columns
determination of the molecular constants directly via mul- and the intercept must be allowed to vary. The three regres-
tiple linear regression. These five columns are computed from sion coefficients will correspond to Be, {De, and { αe, and
the five terms in the wavenumber transition expression, eq the
∼
intercept will ∼
correspond to the vibrational origin, ω10 =
4b. Each term in eq 4b consists of a molecular constant mul- E(v′=1, J ′=0) – E (v′′=0, J ′′=0) = ωe – 2ωexe. The coefficient
tiplied by a combination of quantum numbers for the upper errors are the errors in each molecular constant, and the
state minus the same combination of quantum numbers for intercept error is the error in the vibrational origin. An
the lower state. For example, Be is multiplied by J ′( J ′ + 1) – alternate method of fitting the constant term in eq 12 is to
J ′′( J ′′ + 1). The five computed columns consist of the quan- include a column of ones for it in the spreadsheet and perform
tum number expressions that multiply the five molecular con- the regression with four independent variables while defining
stants that are being determined. Thus, these five columns the intercept to be zero. The vibrational origin will then
will contain formulas for (v′ + 1⁄ 2) – (v′′ + 1⁄ 2 ), J ′(J ′ + 1) – correspond to the coefficient for the column of ones. This

1304 Journal of Chemical Education • Vol. 76 No. 9 September 1999 • JChemEd.chem.wisc.edu

 Information • Textbooks • Media • Resources

alternate method is particularly useful if the spreadsheet Table 1. Transition Wavenumbers

program does not output the error for the intercept.1 and Assignments for Fundamental
and First Overtone Bands of H 35 Cl
Calculation of Residuals ∼
v ′′ v′ J ′′ J′ ν/cm{1
Regardless of whether data from the overtone band are 0 1 0 1 2905.995
used, it is critical to use two more columns in the spread- 0 1 1 2 2925.581
sheet to calculate the predicted transition wavenumber
0 1 2 3 2944.577
from the determined constants and to calculate the residual
0 1 3 4 2962.955
difference between observed and predicted wavenumber for
0 1 4 5 2980.689
each transition. This not only illustrates how to use molecular
constants to calculate spectroscopic transitions, but also serves 0 1 5 6 2997.788
as an essential check for data entry or calculational errors. 0 1 6 7 3014.202
The regression method is numerical rather than graphical, 0 1 7 8 3029.941
and data entry errors are not easily caught before performing 0 1 8 9 3044.965
the regression analysis. Since line positions are determined 0 1 9 10 3059.234
with high precision (typically 0.2 cm{1 or better) with modern 0 1 10 11 3072.771
FTIR instrumentation and eq 3 fits the observed energy levels 0 1 11 12 3085.600
extremely well over the observed quantum number range, the 0 1 12 13 3097.550
calculated residuals should be on the order of the precision 0 1 13 14 3108.914
of the spectrometer. It is very easy to pick out a transition 0 1 14 15 3119.418
with a data entry error, as its residual will be much larger 0 1 15 16 3129.099
than the standard error of the regression. Also, errors in the 0 1 1 0 2864.834
first-order spectroscopic constants (ωe and Be) may easily be as 0 1 2 1 2843.315
small as 1 part in 105 with data taken on a FTIR instrument, 0 1 3 2 2821.249
so the existence of a data entry or calculational error is flagged
0 1 4 3 2798.641
by an unreasonably large error for a regression constant.
0 1 5 4 2775.499
0 1 6 5 2751.817
Comparison of Methods
0 1 7 6 2727.624
The regression method of data analysis is computation- 0 1 8 7 2702.907
ally superior to both the successive-difference and combina- 0 1 9 8 2677.697
tion-difference methods. To demonstrate this, an H35Cl data 0 1 10 9 2651.932
set recorded with a Nicolet 730 FTIR spectrometer is pre- 0 1 11 10 2625.689
sented in Table 1. This data set has been analyzed using each 0 1 12 11 2598.979
of the methods described above. A spreadsheet with these 0 1 13 12 2571.861
analyses is available on the World Wide Web (12, 13). Imple- 0 1 14 13 2544.220
mentation of the regression method using Mathcad is also
0 1 15 14 2516.141
possible (14, 15). The results of these analyses are compared
0 2 0 1 5687.494
to accepted literature values (16, 17) in Table 2. The regression
0 2 1 2 5705.926
analysis results are both more precise and more accurate than
0 2 2 3 5723.158
the successive-difference and combination-difference methods.
The successive-difference method neglects the centrifugal 0 2 3 4 5739.109
distortion constant De, and its effect is therefore incorporated 0 2 4 5 5753.793
into the determination of the other constants. This is apparent 0 2 5 6 5767.262
in the successive difference plot of Figure 2, in which a 0 2 6 7 5779.441
systematic curvature is observed resulting in a significant 0 2 7 8 5790.312
underestimate of Be when the slightly curved data set is fit to 0 2 8 9 5799.833
a straight line. The combination-difference method results 0 2 9 10 5808.141
from a linear least squares analysis with equal weights are also 0 2 1 0 5646.969
given in Table 2. The numerical difficulties of this method 0 2 2 1 5624.896
are apparent from the deviation from linearity near the 0 2 3 2 5601.612
intercept in the plot of Figure 3 (the result of fitting eq 10) 0 2 4 3 5577.185
and from the inconsistent De values among plots despite the 0 2 5 4 5551.571
fact that all values should be the same at this level of theory. 0 2 6 5 5524.865
Owing to division by 2(2J ′′ + 3) when linearizing eq 10, for 0 2 7 6 5496.971
example, the y uncertainties for low J ′′ data points are greater
0 2 8 7 5467.968
than high J ′′ data points, resulting in larger deviation of the
0 2 9 8 5437.895
data from the best fit near the intercept in Figure 3. The
determination of De by averaging its value from four separate
analyses causes an unrealistically large error for De in Table 2.
The regression method simultaneously determines all the
molecular constants without neglecting any terms. In addition,

JChemEd.chem.wisc.edu • Vol. 76 No. 9 September 1999 • Journal of Chemical Education 1305

 Information • Textbooks • Media • Resources

Table 2. H35Cl Molecular Constants Determined by Various Methods

Value (SD) /cm{1
Quantity Average Successive Combination Multiple
Ref 16 Ref 17
Spacing Difference Difference Regression
ωe 2860(37) 2989.215(23) 2989.206(55) 2989.281(34) 2989.74 2990.97424
Be 9.8(31) 10.41(12) 10.58982(85) 10.58919(89) 10.5909 10.593404
ωexe — 51.763(8) 51.765(18) 51.796(12) 52.05 52.84579
De — — 0.0005041(181) 0.0005206(29) 0.00053 0.000532019
αe — 0.30421(118) 0.30318(57) 0.30167(10) 0.3019 0.307139
ω10 — 2885.689(8) 2885.676(18) 2885.690(41) — 2885.9775
ω20 — 5667.852(5) 5667.823(7) 5667.789(97) — 5667.9841
Devia- 49 1.431 0.172 0.046 — —
tion a
aRoot-mean-square residual.

no skewing of the weights is introduced from linearizing fitting of the energy level expression, eq 3, and the concept that
equations, correlation between the molecular constants in the spectroscopic transitions arise from transitions between energy
upper and lower state constants is accounted for, and each levels, eq 4, over the details of a data analysis method that is
observed transition is weighted equally. It should be noted specific to the analysis of diatomic rovibrational spectra.
that equal weighting is appropriate for data acquired with Third, it gives students experience with the technique of
an FTIR instrument, as the wavenumber error in line posi- multiple linear regression. Physical chemistry students get
tions is constant over the entire spectral range rather than much experience fitting data to a straight line, y = mx + b
varying with wavelength, as occurs with grating or prism in- (linear regression), but almost no experience fitting data to
struments. In all cases, care must be taken when comparing an equation of the form y = m1x1 + m2x2 + m3x3 + … + b
fit values and literature values to make sure they use a con- (multiple linear regression). This latter form is critical when
sistent set of molecular constants and the same range of ro- dealing with data that depend on more than one variable, as
tational and vibrational levels in the analyses. Inclusion of ad- is the case for most real data sets. Finally, it introduces students
ditional constants or more molecular states will affect the to a method that roughly corresponds to the data analysis
values obtained in any analysis. method used by practicing spectroscopists, namely, multiple
There are also pedagogical advantages to using the re- nonlinear regression (nonlinear least squares fitting) to the
gression method to analyze the infrared spectrum of diatomic energy level expression (Hamiltonian matrix) of a molecule.
molecules. First, it replaces the tedium of preparing and fitting
multiple plots with a much easier and far less time-consuming Conclusion
use of computer technology. Students learn the concepts
underlying the experiment and gain experience analyzing data Analysis of the infrared spectra of diatomic molecules
using spreadsheets, rather than worrying about the details of by multiple linear regression in a spreadsheet dramatically
graph preparation. Second, the method emphasizes the form simplifies the reduction of spectroscopic data compared to
/ cm{1
νR J ′′ – νP J ′′ + 2
2 2 J ′′ + 3

3 3
{4 J ′′ + 1 + J ′′ + 2
/ cm{1
2 2 J ′′ + 3

Figure 3. Combination difference plot, eq 10, for HCl fundamental

Figure 2. Successive difference plot, eq 6, for HCl fundamental band with least squares linear fit. Inclusion of the centrifugal distor-
band with least squares linear fit. The systematic curvature of the data tion constant De results in a more linear plot than the successive differ-
points is caused by neglect of the centrifugal distortion constant ence plot in Figure 2. However, low J ′′ data points near the y-axis
De. The lowered intercept of the least squares fit results in an under- have larger uncertainties and should be weighted less than high
estimate for the value of the rotational constant Be. J ′′ data points.

1306 Journal of Chemical Education • Vol. 76 No. 9 September 1999 • JChemEd.chem.wisc.edu

 Information • Textbooks • Media • Resources

the traditional successive difference and combination difference Physical Chemistry, 6th ed.; McGraw-Hill: New York, 1996; pp
methods. It also provides a more direct connection between 397–404.
experimental data and the molecular constants. The resulting 3. Crockford, H. D.; Nowell, J. W.; Baird, H. W.; Wetzen, F. W.
molecular constants are shown to be both more accurate and Laboratory Manual of Physical Chemistry, 2nd ed.; Wiley: New York,
1975; pp 255–269.
more precise than the results of traditional analysis methods. 4. Sime, R. J. Physical Chemistry: Methods, Techniques, and Experiments;
Saunders: Philadelphia, 1990; pp 676–687.
Acknowledgments 5. Halpern, A. M. Experimental Physical Chemistry: A Laboratory Text-
book; Prentice-Hall: Upper Saddle River, NJ, 1997; pp 567–587.
This work was supported by National Science Foundation
6. Herzberg, G. Molecular Spectra and Molecular Structure I. Spectra
grants CHE-8851651 (RWS) and CHE-9157713 (WFP). of Diatomic Molecules, 2nd ed.; Krieger: Malabar, FL, 1989; pp
110–114, 121–128, 175–183.
Note 7. Richards, L. W. J. Chem. Educ. 1966, 43, 552–554.
W 8. Stafford, F. E.; Holt, C. W.; Paulson, G. L. J. Chem. Educ. 1963,
Supplementary materials for this article are available on JCE
Online at http://jchemed.chem.wisc.edu/Journal/issues/1999/Sep/ 40, 245–249.
abs1302.html. 9. Roberts, B. J. Chem. Educ. 1966, 43, 357.
1. After submission of this paper, an alternative spreadsheet-based 10. Prais, M. G. J. Chem. Educ. 1986, 63, 747–752.
analysis method was published which uses the Solver tool in Excel to 11. Albritton, D. L.; Harrop, W. J.; Schmeltekopf, A. L.; Zare, R. N.;
perform a nonlinear least squares optimization of a transition wave- Crow, E. L. J. Mol. Spec. 1973, 46, 67–88.
number expression to a data set (18). This method could be adapted to 12. JCE Online Journal Supplements; http://jchemed.chem.wisc.edu/
optimize eq 4 or eq 12 and should give the same results as the multiple Plus/JCEExtras/index.html (accessed July 1999).
linear regression method. The Solver method has the advantage of work- 13. hcl.xls (Excel 5.0/95 spreadsheet; requires Analysis ToolPak Add-In).
ing for more complex, nonlinear functional forms. However, it is possible http://www.chem.hope.edu/~polik/doc/hcl.xls (accessed July 1999).
for the method to converge at a local minimum giving an incorrect 14. Mathcad Instructional Documents for Physical Chemistry; http://
result, and the determination of errors in fitting parameters is not www.monmouth.edu/~tzielins/mathcad/mathindx.htm (accessed
performed as easily as with the multiple linear regression method. May 1999).
15. hcl.mcd (Mathcad 6.0 document). http://www.chem.hope.edu/
Literature Cited ~polik/doc/hcl.mcd (accessed May 1999).
16. Herzberg, G. Molecular Spectra and Molecular Structure I. Spectra
1. Daniels, F.; Williams, J. W.; Bender, P.; Alberty, R. A.; Cornwell, of Diatomic Molecules, 2nd ed.; Krieger: Malabar, FL, 1989; p 534.
C. D.; Harriman, J. E. Experimental Physical Chemistry, 7th ed.; 17. Rank, D. H.; Eastman, D. P.; Rao, B. S.; Wiggins, T. A. J. Opt.
McGraw-Hill: New York, 1970; pp 247–256. Soc. Am. 1962, 52, 1–7.
2. Shoemaker, D. P.; Garland, C. W.; Nibler, J. W. Experiments in 18. Iannone, M. J. Chem. Educ. 1998, 75, 1188–1189.

JChemEd.chem.wisc.edu • Vol. 76 No. 9 September 1999 • Journal of Chemical Education 1307