Comparing Shock Models
I regenerated the images for comparing my shock models (dashed) with Anna's (solld). I am comparing density (blue), temperature (red), and ionization fraction (black). I also made plots showing the percent relative error. The error does not look that good even though the plots of the primitive variables look okay. It just goes to show you how deceiving plots can be, and this is why quantifying differences and errors is so important.
Primitive Variables | Percent Relative Error |
---|---|
UPDATE 1 (10/23 1:00 PM)
I discovered something that might be screwing up the relative error plots. Anna's data does not have a fixed step size while mine does. I set it up so that we have the same number of data points, but Anna's data points are more concentrated where gradients are higher whereas my simulations are a fixed grid. Therefore, visit could be doing the relative error based on the point number rather than the actual distance behind the shock. Perhaps if I use gnuplot it will do some type of interpolation to get a more accurate error plot.
UPDATE 2 (10/23 9:00 PM)
As I previously mentioned, visit could not correctly interpolate the data to produce accurate relative error plots. gnuplot can interpolate data but only as a plotting option. I couldn't figure out how to do operations on the interpolated curves. So I did something different…I fitted polynomials to the data points and then found the relative errors between the polynomial curves. To achieve better fits, I used 10th order polynomials. I have only done this for vs = 30 km/s, but I will do the rest tomorrow. I will also post images of the fits vs. the data so that you're convinced that the fits are okay.
UPDATE 3 (10/25 11:00 AM)
The above method worked pretty well for vs = 30 km/s. The fits get a little oscillatory for temperature and ionization fraction, but they are not too bad. However, the fits for the higher shock velocities did get pretty bad (especially for temperature). So I will have to come up with something else. I think my only option at this point is to write my own little fortran routine to interpolate the data and calculate the errors for me.
UPDATE 4 (10/25 3:30 PM)
Finally got accurate relative error plots. I wrote a little fortran program that resamples/interpolates the data, then calculates relative errors and outputs to a curve file to easily create the plots in visit. I also redid the primitive variable plots using gnuplot instead of visit, so that I could have better control of the tick marks.
Attachments (7)
- v50_errors.png (32.2 KB) - added by 12 years ago.
- v30_errors.png (32.1 KB) - added by 12 years ago.
- v80_errors.png (33.3 KB) - added by 12 years ago.
- v30_shock.png (8.0 KB) - added by 12 years ago.
- v50_shock.png (8.1 KB) - added by 12 years ago.
- v80_shock.png (8.7 KB) - added by 12 years ago.
- errorplot.p (3.7 KB) - added by 12 years ago.
Download all attachments as: .zip
Comments
No comments.