Title: accuracy of glitch detection? Post by: claude on May 08, 2017, 06:33:43 PM It seems m_db_z (used for Pauldelbrot glitch detection threshold) is an array of double even for long double and floatexp iterations, which makes me wonder if glitches could be missed due to the double underflowing to 0.
Changing it to use the appropriate type shouldn't be too difficult, but a bit boring with all the copy/paste iterations code... Title: Re: accuracy of glitch detection? Post by: Kalles Fraktaler on May 09, 2017, 04:10:15 PM No, it is only measuring some 0.0000001 or something, so it is safe for double.
I haven't seen any missed glitches because of this, ever! Only case when glitches may be missed are if SA skips beyond the iterations where the glitch is detectable. Title: Re: accuracy of glitch detection? Post by: claude on May 09, 2017, 05:00:10 PM No, it is only measuring some 0.0000001 or something, so it is safe for double. Ah true, I forgot it wasn't the small delta but the large orbit. Makes sense. Thanks! With a near-nucleus of a minibrot it could underflow to 0 (and should in fact be exactly 0 at the nucleus) after each NxPeriod iterations, but that doesn't cause any problems as we know. Then it remains the case of "the second-nearest to 0 iterate of the reference", chances that it is so small might even be imposible at achievable zoom levels, something to investigate. |