## How to do a nonlinear fit using least squares

on 14 Sep 2019
Latest activity Commented on by Star Strider

on 15 Sep 2019

### Star Strider (view profile)

I have a set of data points giving me the values for the second virial coefficient, for various values of T, of the virial expansion which is an equation that corrects the ideal gas law for empirical deviations: I'm trying to do a least squares fit to determine how well the van der Waals equation predicts using MATLAB.
The equation you derive for B using van der Waals' equation ends up being: where a and b are unknown constants I need to determine and R is just the ideal gas constant.
Now my data points for true values are here:
B = [-160 -35 -4.2 9 16.9 21.3]*10^-6 ;
T = [100 200 300 400 500 600]; (kelvin)
How can I write some code which will use the least squares method to generate estimates for a and b using the given data points?
My problem is that this is essentially a equation which I'm not sure how to represent in MATLAB.
I can only find options for quadratic, cubic, polynomial fit etc.

### Tags ### Star Strider (view profile)

on 15 Sep 2019
Edited by Star Strider

### Star Strider (view profile)

on 15 Sep 2019

This is actually a linear problem, so a linear approximation will estimate the parameters correctly:
R = 8.314462; % J K^−1 mol
B = [-160 -35 -4.2 9 16.9 21.3];
T = [100 200 300 400 500 600];
ab = [ones(size(B(:))) - 1./(R*T(:))] \ B(:); % Linear Approximation Parameter Estimation
Bfit = [ones(size(B(:))) - 1./(R*T(:))] * ab; % Linear Fit
figure
plot(T, B, 'pg')
hold on
plot(T, Bfit,'-r')
hold off
grid
xlabel('T')
ylabel('B')
legend('Data','Linear Least-Squares Fit', 'Location','E')
The parameters are:
a = 64.2001320479734
b = 182307.574287957
EDIT — Star Strider

### Star Strider (view profile)

on 15 Sep 2019
My pleasure!
Note that your equation is linear because it is linear in the parameters. That is because the partial derivatives of ‘B’ with respect to either parameter are not functions of that parameter or other parameters. So a linear approximation is appropriate, and provides the correct result.

on 15 Sep 2019
got it. Thanks for the clarification!
Star Strider

on 15 Sep 2019
My pleasure!

### Jackson Burns (view profile)

on 15 Sep 2019

Attached is code generated by cftool using a custom equation. Hope this helps!

on 15 Sep 2019
Pardon my ignorance but I'm having trouble implimenting the command:
[result,~] = createFit(T,B)
its telling me that "result" is unused. Is there something I need to link to that spot? Thanks again
Jackson Burns

### Jackson Burns (view profile)

on 15 Sep 2019
If you're making this call in a function, it will tell you that result is unused if you don't later on reference result or return it from the function.