Goo day. I'm just having trouble creating model with less residuals in Vstar using DCDFT with period range as an option. As you may see in the image, I performed period analysis and the results were given and I have tried to get the number of harmonics (which suggests to be equal to 1) but when I tried to create the model, it seems that the model is off to the data. I tried to increase the number of harmonics and found out that increasing it upto 12 also increases the accuracy of the fit though, RS Gru has thought to only have 1 period. Does that mean something?
Here's the description of my procedures.
Target object: RS Gru (downloaded all the data directly from Vstar itself)
Performed DCDFT with period range (Low period: 0.13 ,high period: 0.15, resolution of 0.0000001)
Created model with harmonic=1
Hello Joe
It looks like your model uses Harmonic = 1. This will fit a sine cuve nicely, but you have higher harmonics in the data. Try increasing to 3, 5, 7, and 9. Watch what you get. Post your results. We can go from there.
Gary
Thanks for a quick response Gary. I've tried to increase the harmonics upto 12 (which gives me the best fit). I've calculated the residuals using SSR (Sum of Squared Residuals) which gives me the following results:
Harmonics and SSR Values
3=14.3283601291422
5=13.3148714338014
7=13.2028411760553
9=13.1786120408334
12=13.1496988447827
The values suggests that as the SSR value decreases, the fit is getting great, which again tells that increasing the harmonics decreases the value of residuals.
Am I doing it right?
One other thing to try, is that your residuals may be dominated by the Vis observations. Are you fitting all or just the V? Select just the V and the residuals may reflect what you see when looking.
Gary
Yes, I've tried to filter the observation which in turn, makes the residuals smaller. I selected the points with lesser error and the observations conducted in Johnson V band and still, the residuals are getting smaller when the harmonics is approaching to 12.
One more question, is this always applicable to all observations? I mean, if increasing the harmonics will increase the precision of the model, then why is it not by default? I'm sorry if I might bother you with my question. I can't read more of the articles regarding harmonics applied to this kind of field.
Mathematically, the model will get better and better with more terms. However, if you take data again, will the optimum model hold, or will it change slightly? In the engineering world, hi order models are always suspect. You have to do a lot more testing to make sure they are real.
Gary
Thanks. It has just cleared my worries and doubts. I'm not good at data analysis but still, I'm trying to understand. I'll try your suggestions and might experiment on the data in the future.