Southworth Det Eclips Binary Catalog

On Thursday, December 5, 2013 1:51:20 AM UTC-5, wlandsman wrote:

winds observed in the spectra of red giants. The discrete self-similar

paradigm can't address these questions because it has thrown out the

physics and is built on only metaphor. (I would however be interested

in learning how the paradigm is supposed to account for the high mass

loss rates seen in red giant stars.)

It is false and pejorative to say that Discrete Scale Relativity is
"built on only metaphor". The discrete self-similar paradigm is based
on a very large amount of empirical data from the subatomic to the
galactic scales, AND it is built on discrete conformal
invariance/symmetry (a topic that fascinated Weyl and Dirac, I might
add). Your comment shows that you have not taken a serious idea
seriously.

When a highly excited atom makes a transition to a lower energy state,
or more appropriately here, undergoes an ionization event leaving an
ionic core in a low energy state, it loses mass/energy. Right? See any
possible connection with a red giant blowing off its outer layers and
a white dwarf being left behind?

[Mod. note: that's *precisely* the sort of metaphor that the OP was
talking about -- mjh]

http://www3.amherst.edu/~rloldershaw
Discrete Scale Relativity/Fractal Cosmology
Robert L. Oldershaw

In article , "Robert L.
Oldershaw" writes:

(1) Using induction and pattern recognition I search empirical
knowledge for unforeseen patterns and correlations.

(3) Then the predictions and subsequently observed results can be
compared at face value. You would say it does not matter until the
statistical significance reaches a certain level. I do not agree.

Note that it is well documented that humans see many patterns where none
really exist. This has an obvious evolutionary explanation: it is
better to have a false positive (think something is significant when it
is not) than a false negative (think something doesn't matter when it
does). There is a huge literature on this topic.

This is how Democritus could learn so much about nature without any
statistics or computers or advanced observational equipment. He was
right about so many very fundamental things that saying he just made
lucky guesses is not an acceptable answer.

Note that there is a selection effect here. If many people make
guesses, most will be wrong (and much Ancient Greek science was wrong),
but some will be right. Some of those who were right were right just
due to chance.

On Friday, December 6, 2013 3:10:14 AM UTC-5, Robert L. Oldershaw wrote:

[Mod. note: Jonathan's suggestion would take about 5 minutes to

implement with the current large dataset -- why not try it? -- mjh]
----------------------------------

Looked up CDFs and K-S test.
Spent 5 minutes.
Generated more questions than I started with.
Will pay teacher/coach when time is propitious (we have plenty here).
Time more important than money to me in this case.
And when I release a paper on this subject into the public domain I would like it to be relatively "bullet-proof" because it will need to be, as has been so clearly shown at SAR.

On Friday, December 6, 2013 1:30:23 PM UTC-5, Robert L. Oldershaw wrote:

[Mod. note: that's *precisely* the sort of metaphor that the OP was

talking about -- mjh]

There is a very large and scientifically important difference between "metaphor" and self-similarity.

"*precisely*"? Where has our statistical precision and accuracy gone now?


[Mod. note: quoted text trimmed -- mjh]

In article ,
Robert L. Oldershaw wrote:
On Friday, December 6, 2013 6:14:45 AM UTC-5, Martin Hardcastle wrote:

Sorry, does that mean that you don't have a prediction for the

intrinsic scatter about zero -- which makes your model untestable,

since the scatter could in principle be much larger than 0.145 solar

-- or that you aren't prepared to tell us what it is?
---------------------------

The former is the current situation. I think we would need more
empirical guidance before inducting a theoretical explanation. In the
case of the Solar System (Tech Note #1 at my website) the delta M is
about 0.003 solar mass. In the case of binary star systems the delta M
can be much larger, say 0.01 to 0.04 solar mass.

I made the assumption that the delta M is the maximum you quote, 0.04
Then to give the model the maximum possible credit I assume that any
data point with |Q|0.04 contributes zero to the chi^2 sum (this is
equivalent to assuming that the distribution is uniform in the range
-0.04--0.04 -- in reality it would be peaked, which would give worse
results, but since you can't tell us the shape of the distribution, I
have to take the one that most favours the model). Then I take all the
values of |Q|0.04, subtract 0.04, and find chi^2, for the full
dataset we discussed earlier.

The result? The model is still ruled out at the 99.99[x18] per cent
confidence level. You can only save it if you make the intrinsic
scatter so broad that it will be impossible to see any signal.

Actually this is the way I personally test the merit of scientific ideas.

(1) Using induction and pattern recognition I search empirical
knowledge for unforeseen patterns and correlations.

(2) The perceived pattern can be extended into the unknown (one's
personal unknown, or much better, everybody's unknown) so as to make
predictions.

(3) Then the predictions and subsequently observed results can be
compared at face value. You would say it does not matter until the
statistical significance reaches a certain level. I do not agree. I
use simple yes/no tests. No single test is sufficient in this method
to prove the theory. BUT, if a theory can consistently pass these
yes/no tests, especially if previous theories stumble on these
results, then my confidence in that theory grows. If this method
violates your statistical approach or your logic, then I would say
that your statistics and logic are wrong.

Straw man. Of course it's possible to have a yes/no test. For example,
the answer to the question 'does your model describe the mass
distributions of real stellar systems' is unambiguously 'no'.

And, of course, knowing when this is the case and when it isn't
often requires some knowledge of statistics. Not always, but often.

I think that's all I have to say on the matter.

Martin
--
Martin Hardcastle
School of Physics, Astronomy and Mathematics, University of Hertfordshire, UK
Please replace the xxx.xxx.xxx in the header with herts.ac.uk to mail me

On Saturday, December 7, 2013 3:40:46 AM UTC-5, Phillip Helbig---undress to reply wrote:

Note that there is a selection effect here. If many people make

guesses, most will be wrong (and much Ancient Greek science was wrong),

but some will be right. Some of those who were right were right just

due to chance.
-------------------------------------------------------

I got interested in Democritus in the 1990s and read the definitive
source by Cyril Bailey, The Atomists... Oxford UP, 1928. I wrote a
paper that was published in 1998 on Democritus.

Democritus discovered/taught the following things about nature.

- Atomic basis of matter (actually more like molecular, but...)
- idea of vacuum between atoms
- constant motion of atoms
- basic concept of mass/energy conservation
- basic theory of colors (primary and secondary)
- constant formation and destruction of systems on macroscopic scales
- Milky Way composed of stars
- Sun is a star
- heliocentric Solar System
- light is composed of corporeal emanations from surfaces
- basic ideas of perception
- inherent limitations of observations

Not too shabby for 2500 years ago.

Too bad the Aristotelians and Platonists stopped this progress and led
us into the Dark Ages for many centuries, until the Renaissance
scientists rediscovered the wisdom of the Democritus and other ancient
Greek scientists and got the ball rolling again.

"Chance" is a ludicrous explanation in the case of Democritus.

[Mod. note: we seem to be wandering away from research in astronomy
and astrophysics -- mjh]

On Saturday, December 7, 2013 3:47:10 AM UTC-5, Martin Hardcastle wrote:

The result? The model is still ruled out at the 99.99[x18] per cent

confidence level. You can only save it if you make the intrinsic

scatter so broad that it will be impossible to see any signal.


I need a clarification on one point.

First, forget about any intrinsic physical variability of stellar
masses, for this question.

If we had a sample of 100 binary systems and their total masses were
all known to =/ +/- 0.03 solar mass.

As a thought experiment, when we compare them to multiples of 0.145
solar mass, say *hypothetically* we find that 80 have deviations of
=/ 0.02 solar mass, while 14 have deviations in the 0.21 to 0.04
solar mass, and 6 have deviations of 0.41 to 0.07 solar mass.

Is there any chance that your preferred chi-squared test would
indicate that this was a non-uniform distribution? Likewise would it
possibly indicate that the distribution favors the preferred DSR
masses over a random result.

Or is the model still ruled out with high (low) probability.

I don't need an answer based on a full analysis, just an informed
guess would do. I want to get an intuitive feeling for what properties
a sample would have to have in order to support my model.

On 12/8/2013 9:24 AM, Robert L. Oldershaw wrote:
On Saturday, December 7, 2013 3:47:10 AM UTC-5, Martin Hardcastle wrote:

The result? The model is still ruled out at the 99.99[x18] per cent

confidence level. You can only save it if you make the intrinsic

scatter so broad that it will be impossible to see any signal.


I need a clarification on one point.

First, forget about any intrinsic physical variability of stellar
masses, for this question.

If we had a sample of 100 binary systems and their total masses were
all known to =/ +/- 0.03 solar mass.

As a thought experiment, when we compare them to multiples of 0.145
solar mass, say *hypothetically* we find that 80 have deviations of
=/ 0.02 solar mass, while 14 have deviations in the 0.21 to 0.04
solar mass,

0.21 can never happen. They can only have deviations up
to 0.0725 solar mass if you compare them to the nearest
multiple of 0.145 (and if you don't take the nearest
then it's not clear which one you *are* taking).

... and 6 have deviations of 0.41 to 0.07 solar mass.

Likewise, this interval is almost completely outside
the range of attainable outcomes! In addition it is
overlapping the previous range you gave (0.04 to 0.21)
so you are using overlapping bins, which would make
your question ill-defined. Maybe you miscalculated
the example values, but in this way it is completely
unclear what the question actually is.

--
Jos

[Mod. note: I would assume he means 0.021 to 0.040, 0.041 to 0.07 -- mjh]

In article ,
Robert L. Oldershaw wrote:
First, forget about any intrinsic physical variability of stellar
masses, for this question.

If we had a sample of 100 binary systems and their total masses were
all known to =/ +/- 0.03 solar mass.

As a thought experiment, when we compare them to multiples of 0.145
solar mass, say *hypothetically* we find that 80 have deviations of
=/ 0.02 solar mass, while 14 have deviations in the 0.21 to 0.04
solar mass, and 6 have deviations of 0.41 to 0.07 solar mass.

Is there any chance that your preferred chi-squared test would
indicate that this was a non-uniform distribution?

The chi^2 test is checking for consistency with the model, not a
non-uniform distribution.

Calculating the chi^2 statistic is easy: add up the squares of the
ratios of the deviation from the model prediction to the error bar. So
in this case, I'll assume that the deviations are all in the middle of
your range and that the error bars are all 0.03, and then we have

chi^2 = 80*(0.01/0.03)^2 + 14*(0.03/0.03)^2 + 6*(0.055/0.03)^2

This would give a chi^2 of 43 for 100 degrees of freedom, which (you
can check this on an online calculator) would very definitely be
consistent with your model.

(However, it would also be literally too good to be true -- the
typical deviation from the model prediction for most objects is
significantly less than the error bars, which in real data would be a
sign either that the error bars were wrong or that someone was cooking
the books.)

If instead the errors were all 0.01, say, then instead we would have a
chi^2 of 388 and the model would be ruled out at a very high
confidence level.

This is why the errors matter.

Martin
--
Martin Hardcastle
School of Physics, Astronomy and Mathematics, University of Hertfordshire, UK
Please replace the xxx.xxx.xxx in the header with herts.ac.uk to mail me

On Monday, December 9, 2013 3:51:13 AM UTC-5, Martin Hardcastle wrote:
[stuff]

Many thanks for this post.

Long ago and far away (Seattle) as an oceanography/chemistry student I
took a course in statistics and actually learned how to do chi-squared
tests, among other things.

Thanks to your post I have the motivation to pull out my old
statistics text and relearn how to do this basic stuff. Statistics is
like a foreign language that I once had a working knowledge of but
that I forgot due to lack of use. My only excuse is that I needed time
and brain-space for other pursuits that involved a very different set
of discovery and evaluation tools.

I heartily agree that errors matter, especially systematic errors
that, when combined with high precision, can give the false impression
of virtually "final" answers. One thinks here of the ongoing proton
radius issue.

Precision appears to be fairly straightforward to determine. Accuracy
is a much more difficult thing to determine.

[Mod. note: reformatted -- mjh]