Computational Approach to

Materials Science and Engineering

Prita Pant and M. P. Gururajan
October, 2012

c 2012, Prita Pant and M P Gururajan. Permission

Copyright
is granted to copy, distribute and/or modify this document under
the terms of the GNU Free Documentation License, Version 1.3
or any later version published by the Free Software Foundation;
with no Invariant Sections, no Front-Cover Texts, and no BackCover Texts. A copy of the license is included in the section
entitled GNU Free Documentation License.

Module: Fitting
1

Pre-requisites

The modules in Part II of this course material.

Learning goals
To be able to fit a given data to a given polynomial and plot it.

Fitting

In this module, we use the term fiting to describe the process of finding
out the best function that describes the given data. There are more than one
way of defining what the best function is, and different definitions can lead
to different results. However, we do not discuss these details here. Further,
we also do not discuss the mathematical ideas behind the notion of design
matrix. We refer the interested reader to books on linear regression.

Lattice parameter variations with composition: linear fit

The lattice parameter of a material changes with composition as well as with

temperature. Generally, for small changes in composition, the change in
lattice parameter is assumed to be linear. This is known as Vegards law.
In Table 1 we show the data on the variation of lattice parameter in the
Si-Ge system. We want to know if we can fit the data to a linear function; if
so, we also would like to know what the slope and intercept of the linear fit
is.
In the GNU Octave script given below, we show how to carry out the linear
fitting. In addition, after fitting, the fit is also plotted along with the experimental data and the plot is shown in Fig. 1 below. The fitting exercise
2

Mol. % of silicon

Lattice constant

Density

100
87.4
85.8
75.7
57.5
44.3
22.9
15.0
13.5
12.6
0

5.434
5.454
5.461
5.473
5.518
5.549
5.593
5.620
5.613
5.626
5.657

2.328
2.72
2.80
3.03
3.62
3.95
4.70
4.86
4.89

5.323

Table 1: The variation of density and lattice parameter with composition in

SiGe alloys. The data is taken from Some properties of Germanium-Silicon
alloys, E. R. Johnson and S. M. Christian, Phys. Rev., 95, pp. 560-561,
1954.

indicates that the lattice parameter decreases by about 0.0022345 Angstroms

for every mol % of Silicon; further, the fit gives the lattice parameter for 0
mol % of silicon as 5.6498279 while the experimentally observed value is 5.657
(indicating that the fit is quite good).
# codes/LinearFitVegardsLaw.oct
#
# Copyright (C) 2011 Prita Pant and M P Gururajan
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of GNU General Public License as published by
# the Free Software Foundation; either version 3 of the License, or (at
# your option) any later version.
#
# This program is distributed in the hope that it will be useful, but
# WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
# General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program; if not, write to the Free Software Foundation,
# Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA
#
#
% Load the given data from the data file
X = load("data.dat");
% Read the first column of the data as composition
comp = X(:,1);
% Read the second column of the data as lattice parameter
a = X(:,2);
% Use polyfit to fit the data in the least square sense; note that the
% third parameter, which is unity, is what tells Octave that we are
% looking for a linear fit to the data
z = polyfit(comp,a,1);
% Use the fit to generate data at equal intervals
fitx = 0:10:100;
fity = polyval(z,fitx);
% Plot the experimental data and the fit
4

Composition versus lattice parameter plot

5.7

5.65

Lattice paramter (in Angstroms)

5.6

5.55

5.5

5.45

5.4
0

20

40

60

80

100

Si (in mol%)

Figure 1: The linear least squares fit for the experimental data on lattice
parameter variation with composition in Si-Ge alloy.
plot(comp,a,o,fitx,fity)
% Set the axes values
axis([0 100 5.4 5.7]);
% Label the axes and title the plot; also make the plot square
xlabel "Si (in mol%)"
ylabel "Lattice paramter (in Angstroms)"
title "Composition versus lattice parameter plot"
axis(square)
% Print the plot as an eps file
print -depsc VegardsLawGOCT.eps

Note that the GNU Octave command used for fitting is polyfit. There are
three arguments for polyfit. The first is the independent variable (say, x);
5

the second is the dependent variable (say, y); and the third is the order of
the fit. By choosing the third variable to be unity, we do a linear fit in the
least squares sense. However, by choosing it to be n, we can fit the data
to an n-th order polynomial, namely, y = an xn + an1 xn1 + ... + a1 x + a0
and the coefficients an -a0 are returned to the user. In the script above, the
coefficient vector is stored as z.

Low temperature specific heat of copper:

Polynomial fit

In the previous section, we discussed polynomial fit for the given data. However, sometimes, the exact form of the polynomial is known, and we wish to fit
the data to the given polynomial and not just to any polynomial. For example, consider the variation of specific heat with temperature. Depending on
the mechanisms of absorption of heat in a material, the temperature dependence of specific heat will be different. In general, at very low temperatures,
the specific heat has the following dependence on temperature: Cp = aT +bT 3
where a and b are constants. At high temperatures, the specific heat has the
following dependence on temperature: Cp = a + bT + CT 2 + dT 2 where
a, b, c and d are constants. More information on the variation of specific
heat with temperature is available in Swalin listed in References and further
reading section.
In Table 3 we show the data on the variation of specific heat with temperature
for copper. Note that the data here is for the temperature range 0 to 20 K
which is different from the one that was discussed in the plotting module.
Since the temperature range in question is 0 to 20 K, it can be considered
as the low temperature regime for copper. Unlike in the earleir case, since
we know that at low temperatures, the variation of the specific heat with
temperature is polynomial (given by the polynomial Cp = aT + bT 3 ), we
want to fit the data to this given polynomial and identify the constants a
and b. In this section, we will show how to carry out such a polynomial
fitting to the given data.
The following GNU Octave script fits the given specific heat data to a polynomial of the type aT + bT 3 .

Temperature

Specific Heat

1
2
3
4
5
6
7
8
9
10
12
14
16
18
20

7.43x104
1.77x103
3.37x103
5.82x103
9.43x103
1.45x102
2.13x102
3.01x102
4.14x102
5.55x102
9.36x102
1.49x101
2.25x101
3.28x101
4.62x101

Table 2: The variation of specific heat with temperature in copper in the

low temperature regime (0 20K). The data is taken from Heat capacity of
reference materials: Cu and W, G. K. White and S. J. Collocott, J. Phys.
Chem. Ref. Data, 13, pp. 1251-1257, 1984.

% Load the specific heat data

S = load("CuSpHeatData.dat");
% Get the temperature and specific heat from the data file
T = S(:,1);
Cp = S(:,2);
% Plot the experimental data with open circles
plot(T,Cp,o)
hold on
% Generate the design matrix
X = [T T.*T.*T];
% Calculate the coefficients a and b which are consistent with the given data
a = X\Cp
% Using the obtained a and b values, generate data points
t = (0:1.:25);
Y = [t t.*t.*t]*a;
% Plot the fit
plot(t,Y,.-)
% Label the axes
xlabel ("Temperature (in K)");
ylabel ("Specific heat (in J/mol/K)");
% Make the plot square and save the output file

0.8

0.6

0.4

0.2

-0.2
0

10

15

20

25

Figure 2: The polynomial fit for the experimental data on specific heat variation with temperature in copper at low temperatures (in the range 0-20).
axis(square)
print -depsc CuSpHeatFit.eps

From this script, it is clear that the given data fits the polynomial Cp =
2.2460 1004 T + 5.7411 1005 T 3 . The plot generated using this script
is shown in Fig. 2.
Note the use of the so-called design matrix; the design matrix is a matrix
in which for each data point, the independent variable is put in the required
polynomial form, and knowing the value of the dependent variable, the unknown coefficients for the polynomial are evaluated. The solution of this
matrix equation gives us the coefficients that is consistent with the given
data.

Self-assessment questions
1. What is the GNU Octave command for fitting a given data to a polynomial of the type y = a3 x3 + a2 x2 + a1 x + a0 ?
2. What is the GNU Octave command to calculate values using a given fit
(that is obtained using polyfit)?
3. To fit a given data to the polynomial y = a + bx5 one can use polyfit.
True or false?

Answers to self-assessment questions

1. polyfit(x,y,3)
2. polyval
3. False. One has to use design matrix and solve the set of resultant
equations to obtain such a fit.

Exercises
Problem 1. Linear fit
Consider the consumption of copper as a function of temperature in a
reaction. The data is as shown in Table 3. Assume that the consumption is related to temperature through an Arrhenius type of reaction,
Q
namely, h = A exp RT
where R is the universal gas cosntant, T is
the absolute temperature, Q is the activation energy for the reaction,
h is the copper consumption (in microns), A is the pre-exponential
constant. From the given data, estimate Q and A.
Problem 2. Polynomial fit
The data for linear thermal expansion of cubic BN is given in Table. 4.
Fit the data to a polynomial and plot the data and the fit.

10

Temperature (in K) h (in microns)

473
493
513

0.26
0.35
0.42

Table 3: The copper consumption in a reaction as a function of time. The

data is taken from Kinetic analysis of the soldering reaction between eutectic
SnPb alloy and Cu accompanied by ripening , H K Kim and K N Tu, Phys.
Rev. B, 53, pp. 16027-16034, 1996.

Temperature (in K) Linear thermal expansion

77
200
297
535
668
785
889
1008
1061
1137
1205
1289

a
a0

103

-0.380.05
- 0.280.05
0.00
+0.440.14
+0.860.14
+1.4580.14
+2.020.14
+2.660.14
+2.96 0.14
+3.600.14
+3.900.14
+4.400.14

Table 4: The variation of linear thermal expanion with temperature in BN

in the temperature regime (77 1289 K). The data is taken from Thermal
expansion of some diamondlike crystals, G. A. Slack and S. F. Bartram, J.
Appl. Phys., 46, pp. 89-98, 1975.
11

Temperature

Specific Heat

300
350
400
450
500
550
600
650
700
800
900
1000
1100
1200
1250
1300

24.45
24.88
25.25
25.61
25.91
26.21
26.48
26.73
26.99
27.48
28.04
28.66
29.48
30.53
31.12
32.16

Table 5: The variation of specific heat with temperature in copper in the low
temperature regime (300 1300 K). The data is taken from Heat capacity of
reference materials: Cu and W, G. K. White and S. J. Collocott, J. Phys.
Chem. Ref. Data, 13, pp. 1251-1257, 1984.
Problem 3. Non-linear, polynomial fit Given below is the data for
specific heat of copper at constant pressure in the temperature range
300 1300. It is known that in this range, the specific heat of copper is
described by the polynomial Cp(T ) = 22.61 + 6.27 103 T Verify that
this indeed is so.

Solution to exercises
Solution to Problem 1.
12

Q
Consider the expression: h = A exp RT
Taking logarithm on both
Q
sides, one obtains, log (h) = log (A) RT Thus, one can see that
the plot of (1/T) versus log(h) is a straight line. The slope of the
curve, multiplied by R, the unviersal gas constant gives the activation
energy (in J/mol) and the intercept value, when exponentiated gives
the pre-exponential constant A. The GNU Octave script below does this
Arrhenius plot as well as fitting. By executing the script, we calculate
the activation energy to be 24.252 kJ/mol and the pre-exponential constant to be 125.87 microns. The Arrhenius plot that we have fit the
data to, along with the data is shown in Fig. 3.

# CuConsumption.oct
#
# Copyright (C) 2011 Prita Pant and M P Gururajan
#
# This program is free software; you can redistribute it and/or modify
# it under the terms of GNU General Public License as published by
# the Free Software Foundation; either version 3 of the License, or (at
# your option) any later version.
#
# This program is distributed in the hope that it will be useful, but
# WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
# General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program; if not, write to the Free Software Foundation,
# Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA
#
#
% Load the given data from the data file
X = load("CuConsumptionData.dat");
% Read the first column of the data as temperature
T = X(:,1);
% Read the second column of the data as dh
dh = X(:,2);
% Make the relevant x and y parameters; we would like to plot the logarith
% of dh as a function of (1/T).
13

x = 1./T;
y = log(dh);
% Use polyfit to fit the data in the least square sense; note that the
% third parameter, which is unity, is what tells Octave that we are
% looking for a linear fit to the data
z = polyfit(x,y,1);
% Use the fit to generate data at equal intervals
fitx = 0.0018:0.0001:0.0022;
fity = polyval(z,fitx);
% Plot the experimental data and the fit
plot(x,y,o,fitx,fity)
% Set the axes values
axis([0.0018 0.0022 -1.5 0]);
% Label the axes and title the plot; also make the plot square
xlabel "1/T"
ylabel "ln(dh)"
title "Arrhenius plot"
axis(square)
% Print the plot as an eps file
print -depsc CuArrheniusPlot.eps
% From the linear fit data, one can now calculate the activation energy
% in J/mol and the pre-exponential constant
z(1)*8.314
exp(z(2))

Solution to Problem 2.
#
#
#
#
#
#
#
#

ThermalExpansionBN.oct
Copyright (C) 2011 Prita Pant and M P Gururajan
This program is free software; you can redistribute it and/or modify
it under the terms of GNU General Public License as published by
the Free Software Foundation; either version 3 of the License, or (at
your option) any later version.
14

Arrhenius plot
0

-0.2

-0.4

ln(dh)

-0.6

-0.8

-1

-1.2

-1.4

0.0018

0.00185

0.0019

0.00195

0.002

0.00205

0.0021

0.00215

0.0022

1/T

Figure 3: Arrhenius plot for the consumption of copper in a reaction. From

this plot, one can obtain the activation in energy (in units of R, the universal
gas cosntant) and the pre-exponential constant.

15

#
# This program is distributed in the hope that it will be useful, but
# WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
# General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program; if not, write to the Free Software Foundation,
# Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA
#
#
% Load the given data from the data file
X = load("BNalpha.dat");
% Read the first column of the data as temperature
T = X(:,1);
% Read the second column of the data as linear thermal expansion (epsilon)
epsilon = X(:,2);
epsilon = epsilon.*1000;
% Use polyfit to fit the data in the least square sense; note that the
% third parameter, which is unity, is what tells Octave that we are
% looking for a linear fit to the data
z = polyfit(T,epsilon,2);
% Use the fit to generate data at equal intervals
fitx = 0:10:1300;
fity = polyval(z,fitx);
% Plot the experimental data and the fit
plot(T,epsilon,o,fitx,fity)
% Set the axes values
axis([0 1300 -1000 5000]);
% Label the axes and title the plot; also make the plot square
xlabel "Temperature (in K)"
ylabel " Linear thermal expansion"
title "Temperature verus linear thermal expansion plot"
axis(square)
% Print the plot as an eps file
print -depsc ThermalExpansionBN.eps

16

Temperature verus linear thermal expansion plot

5000

4000

Linear thermal expansion

3000

2000

1000

-1000
0

200

400

600

800

1000

1200

Temperature (in K)

Figure 4: The polynomial fit for the experimental data on linear thermal
expansion variation with temperature in Boron Nitride in the temperature
range 771289 K.

17

 Solution to Problem 3.
Using the script below, we see that the data can be fit to the polynomial
Cp (T ) = 22.342 + 6.8475 103 T as opposed to the given expression,
namely, Cp(T ) = 22.61 + 6.27 103 T Thus, we see that the given data
does fit to the known Cp expression quite well.
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
%

Solution3.oct
Copyright (C) 2011 Prita Pant and M P Gururajan
This program is free software; you can redistribute it and/or modify
it under the terms of GNU General Public License as published by
the Free Software Foundation; either version 3 of the License, or (at
your option) any later version.
This program is distributed in the hope that it will be useful, but
WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software Foundation,
Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA

Load the specific heat data

S = load("CuCpHighTVariation.dat");
% Get the temperature and specific heat from the data file
T = S(:,1);
Cp = S(:,2);
% Plot the experimental data with open circles
18

plot(T,Cp,o)
hold on
% Generate the design matrix
X = [ones(size(T)) T];

% Calculate the coefficients a and b which are consistent with the given d
a = X\Cp
% Using the obtained a and b values, generate data points
t = (300:1.:1300.);
Y = [ones(size(t)) t]*a;
% Plot the fit
plot(t,Y,.-)
% Label the axes
xlabel ("Temperature (in K)");
ylabel ("Specific heat (in J/mol/K)");
% Make the plot square and save the output file
axis(square)
print -depsc CuHighTSpHeatFit.eps

10

References and further reading

1. Thermodynamics of solids, R. A. Swalin, 2nd edition, Wiley-Interscience

19

34

Specific heat (in J/mol/K)

32

30

28

26

24
200

400

600

800

1000

1200

1400

Temperature (in K)

Figure 5: The polynomial fit for the experimental data on specific heat variation with temperature in copper at temeperatures in the range 300-1300
K).

20

publishers, 1972.
2. Advanced Engineering Mathematics, E. Kreyszig, 8th edition, John Wiley and Sons, 1999.
3. Elementary Numerical Analysis, K. E. Atkinson, 3rd edition, Wiley
India, 2003.

21