sph.image loses particles in some cases? about pynbody HOT 8 CLOSED

From [email protected] on March 16, 2011 21:22:20
It seems like the problem is introduced when plotting the 'dsp' array. There are dispersion values for s['dsp'][i](when i is correct), so I'm not sure why it seems to die here...

from pynbody.

From [email protected] on March 16, 2011 21:55:32
It doesn't look to me like there are no particles in those white regions... more like they have been contaminated by a NaN or inf? This is suggested by the fact the regions are exactly square.

Could you generate the image through pynbody.sph.render_image and inspect the values in the 'white' regions?

from pynbody.

From [email protected] on March 16, 2011 22:17:43
If where returns an empty array, doesn't that mean it didn't find anything that matched the condition?

also, if I do a max() and min() on the gas 'mach' array, I get real numbers.

Anyway:
In [48]: a = pynbody.sph.render_image(s.g,qty='dsp')
In [63]: a[250][250]
Out[63]: nan

You're right that there are NaNs there... how do they get there? The 'dsp' array has all real numbers, as far as I can tell:

In [67]: i = where(s['dsp'] == float(nan))
In [68]: i
Out[68]: (SimArray([], dtype=int64),)

Same result for "inf."

from pynbody.

From [email protected] on March 17, 2011 08:59:49
Your nan test (in[67] above) won't work because nan!=nan by definition. You want something like where(s['dsp']!=s['dsp']).

Could you also check what the smoothing length for the affected particles is?

from pynbody.

From [email protected] on March 17, 2011 15:15:43
Ah, yep you're right as usual!

In [5]: i = where(s['dsp'] != s['dsp'])
In [6]: i
Out[6]: (SimArray([ 18618, 19851, 24200, ..., 7316807, 7316809, 7316856]),)
#there they are. This is before trying any plotting... though I'm not sure I understand why the particles are found and then not found in my original report.
In [9]: s['smooth'] = numpy.power(s['mass']/s['den'],1./3.)
In [11]: s['smooth'][i]
Out[11]:
SimArray([ 2.71418938e-08, 2.67177526e-08, 3.81532428e-08, ...,
1.38790972e-10, 1.12029037e-10, 1.58774446e-10])
In [12]: sm = s['smooth'][i]
In [13]: max(sm)
Out[13]: 0.012417326985919935
In [14]: min(sm)
Out[14]: 3.2986661680289594e-11

#And for the whole array:
In [15]: max(s['smooth'])
Out[15]: 0.019023386110121238
In [16]: min(s['smooth'])
Out[16]: 1.7852063890462095e-11

Is that ok for finding smoothing lengths?

from pynbody.

From [email protected] on March 17, 2011 18:23:15
The minimum smoothing length looks a bit tiny... is that physical units?

Incidentally you shouldn't need to explicitly calculated s['smooth'] as it is a @derived_array already.

However, if you do, then to maintain unit information you want:

s['smooth'] = (s['mass']/s['den'])**(1,3)

then s['smooth'] will have unit information

from pynbody.

From [email protected] on March 17, 2011 18:54:27
Sorry, those weren't in physical units. Here it is again:

In [78]: i = where(s.g['dsp'] !=s.g['dsp'])
In [80]: s.g['smooth'][i]
Out[80]:
SimArray([ 11.63911516, 10.02338895, 11.82276064, ..., 13.55228082,
10.60649354, 25.1879408 ], 'kpc')
In [81]: sm = s.g['smooth'][i]
In [82]: min(sm)
Out[82]: 0.0039075758281437454
In [83]: max(sm)
Out[83]: 25.187940795333986

Minimum is 4 pc??

from pynbody.

From [email protected] on March 23, 2011 18:35:02
We discussed this offline and agreed the particle losses were due to incorrect syntax.

The white rectangles have been confirmed due to NaNs.

I'm now closing the issue.

from pynbody.