[Aavso-photometry] Peranso vs Minima V2.3 question

[Michael Newberry]

See below...

----- Original Message ----- 
From: "Michael Koppelman" <michael at slackerastronomy.org>
To: "Michael Newberry" <mnewberry at mirametrics.com>
Cc: "Yenal Ogmen" <yenalogmen at yahoo.com>; <aavso-photometry at mira.aavso.org>
Sent: Tuesday, February 17, 2009 2:59 PM
Subject: Re: [Aavso-photometry] Peranso vs Minima V2.3 question


> We're physicists, Michael, so 3.7 ~ 3.
>
But only in the spherical chicken approximation.

> :)
>
> Seriously, though, your software and Minima's might have used similar 
> techniques or libraries such that more of your calculations are  similar.

I know that there are no common libraries between mine and the others, at 
least as far as any numerical methods go.

> It is possible that you are both doing it wrong and Peranso is doing it 
> right!

Agreed. But I don;t think there's a right or wrong issue here, just a tiny 
difference. Look at the graph: 
http://www.mirametrics.com/images/KvW%20Solution%201.png  First notice how 
clean Yenal's data are. The data points are spaced 82 seconds, on average 
and each marker is about 0.01 mag tall. The vertical line marks my result 
and the result from Minima. The "outlying" Peranso value is 14 seconds to 
the right, or about 1/6 the horizontal spacing of the data points. We need 
to keep it in perspective about how small this difference really is.

> But, Dear Reader, when in doubt Dr. Newberry is right and I am wrong. :)
    Ha!
>
> Also, one must, in general, resist the temptation to just do a 
> statistical average of these results since they are not truly 
> independent.

Michael Newberry

>
> M.
>
>
>
> On Feb 17, 2009, at 3:52 PM, Michael Newberry wrote:
>
>> Hi Michael,
>>
>> You have a good point and I appreciate what you are saying. But the 
>> situation is a little more complex. The difference between his  Peranso 
>> and Minima values is 0.00016 day. The uncertainty reported  by Minima is 
>> sigma=0.00007 and that by Peranso is sigma=0.000045. I  got 
>> sigma=0.000065. Using the Minima sigma makes them 2.3 sigmas  apart but 
>> using the Peranso sigma makes them 3.7 sigmas apart. My  value and its 
>> uncertainty agree with Minima to beyond 5 places,  despite our 
>> implementations of the KvW algorithm being independent.
>>
>> In this case we are talking comparing small differences of 6 seconds  or 
>> 14 seconds. But it would be useful to know two things: 1) whether  a 
>> difference of this order is typical or it is usually larger or  smaller, 
>> and 2) whether the difference is random or systematic. In  other words, 
>> if, on multiple tests, Minima is sometimes higher and  other times lower 
>> than Peranso, and by similar amounts, then you can  say they are 
>> basically giving similar results within random  differences (which is 
>> what you said). On the other hand, if one had  a tendency to give values 
>> that were typically higher or lower than  the other, then there would be 
>> a systematic error that needs to be  addressed---as in, who's error is 
>> it?
>>
>> Michael Newberry
>>
>> ----- Original Message ----- From: "Michael Koppelman" 
>> <michael at slackerastronomy.org
>> >
>> To: "Michael Newberry" <mnewberry at mirametrics.com>
>> Cc: "Yenal Ogmen" <yenalogmen at yahoo.com>; 
>> <aavso-photometry at mira.aavso.org
>> >
>> Sent: Tuesday, February 17, 2009 2:21 PM
>> Subject: Re: [Aavso-photometry] Peranso vs Minima V2.3 question
>>
>>
>>> My first reaction to Yenal's original email was: welcome to the  world 
>>> of computational physics! There are many factors which can  make 
>>> different implementations of the same algorithms vary, many  of which 
>>> Michael mentioned.
>>>
>>> The error reported for Minima (as an example) was 0.00007 or about  6 
>>> seconds. The difference between Minima and Peranso is about 14 
>>> seconds, which is ~2 sigma of the Minima result. So they agree!  You 
>>> could use any of the results and be fine since they all fall  within 3x 
>>> of the uncertainty. Your ToM is never known better than  your 
>>> uncertainty (of course).
>>>
>>> A long time ago I tried to write my own ToM finding algorithm and  it 
>>> failed miserably. Another method that has served me well is  curve- 
>>> fitting with polynomials or some other function.
>>>
>>> Cheers,
>>> M.
>>>
>>>
>>> On Feb 17, 2009, at 1:14 PM, Michael Newberry wrote:
>>>
>>>> In general, using the same algorithm in different software does  not 
>>>> assure that the result will be the same.
>>>
>>
>
>