Dear Linjun He,
PyFAI's geometry is composed of 7 parameters and as Jon mentioned several are correlated:
* distance and wavelength.
* rot1 and poni2
* rot2 and poni1
* rot3 is undefined, in recent versions of pyFAI it is enforced to 0 and unrefined
For a SAXS experiment, there is a "SAXS-constraints" that enforces all
rotations to 0. With this, the calibration should be much more stable.
typically in the order of the 1e-4m, thus much better than the pixel
size (1.72e-5m)
Concerning the ring extraction, as Emily mentionned, there are several
algorithm used for peak extraction and most of them are sampling pixel
using some randomness, this is why there is non-reproducibility of the
ring extraction. That said, you can save precisely the control-points
you have in the "peak-picking" tab. The geometry fitting should be more
deterministic (but since we start without initial geometry, there can
be divergence of the algorithm at the very beginning).
Concerning the number of rings: one should use a much information as possible from the image:
* Use all your ring at least in SAXS setup. In WAXS, be aware of parallax effect beyond ~30° incidence angle.
* Do not extract too many peaks per degree since points not actually on the ring will degrade the fit.
* Do not refine simultaneously distance and wavelength/energy unless you have very good data up to 30°
As Emily mentioned, detector distortion can be important (and
compensated for). Usually it is visible in WAXS calibration where
calibrant produce very thin rings. In SAXS, it is less of a problem
since peaks are bread.
Cheers,
Jerome
On Tue, 17 Mar 2026 12:14:20 +0800
"\"Time\"" (via pyfai Mailing List) <pyfai@???> wrote:
> Dear pyFAI Developers,
>
> I hope this email finds you well.
>
> My name is Linjun He, and I am a graduate student working on SAXS-related scattering data calibration and detector geometry analysis. During our use of pyFAI for calibrating a large number of experimental datasets, we observed a certain degree of instability in the fitted geometric parameters. I am therefore writing to seek your advice and would sincerely appreciate any guidance you may be able to provide.
>
> Our experimental conditions are as follows, and here I use a single SAXS image as an example for illustration:
>
> Data type: SAXS image
>
> X-ray energy: 10 keV
>
> Standard sample: AgBH
>
> Detector: Pilatus 2M
>
> Exposure time: 5 s
>
> During the calibration process in pyFAI, we consistently used the same raw image （Attachment1） and applied the same mask file（Attachment2） throughout the entire workflow. In the peak-picking interface, we manually selected the peaks and then performed calibration under different ring number settings (1, 2, 3, 4, and 5).
>
> We observed that, although the raw image and the mask file were exactly the same, the final fitted geometric parameters still showed noticeable differences when different numbers of rings were used for calibration (the settings interface is shown in Attachment 3, and the corresponding results are summarized in Attachment 4). The parameters that changed include:
>
> poni1
>
> poni2
>
> rotation1
>
> rotation2
>
> rotation3
>
> distance
>
> In addition, we noticed that even for the same diffraction ring, the number of valid points identified in the peak-picking interface could sometimes differ (see the comparison in Attachment 5; the corresponding data are highlighted in red in Attachment 4).
>
> Therefore, we would like to ask the following questions:
>
> 1.Under the condition of using the same image, the same mask, and the same manually selected peaks, why can the number of valid points on the same ring still change?
>
> 2.When the ring number is set to 1, 2, 3, 4, and 5, the fitted values of poni1, poni2, rotation1, rotation2, rotation3, and distance show certain differences. Is this behavior considered normal in pyFAI?
>
> 3.Are these variations mainly caused by differences in the number of points selected during the peak-picking stage, or are they primarily due to the different constraints introduced into the geometric fitting when different ring number settings are used?
>
> 4.For SAXS image calibration, would you recommend a more stable peak-picking or ring-selection strategy to reduce this type of parameter fluctuation?
>
> We have organized the fitting results obtained from the same image under different ring number settings into a table, and it can be clearly seen that these parameters do vary to a certain extent. I would like to understand whether this behavior is expected in the pyFAI fitting process, or whether there are additional settings or precautions that we should pay particular attention to.
>
> Thank you very much for taking the time to read this email. I would greatly appreciate your guidance and reply.
>
> Sincerely,
> Linjun He
>
>
> 从QQ邮箱发来的超大附件
>
> Attachment 1-AGB-5s_00001.tif (9.45M, 无限期)进入下载页面：https://mail.qq.com/cgi-bin/ftnExs_download?k=6f3361632ef15b9eeac9c4254439014e414700025701500414570507031450590a514c57500f074c0f02005a535a05560d035301621833204d4700000a54560f4d13504e237e714c0c403e5352090350191b524a4c4d5a07390e&t=exs_ftn_download&code=93acb93a
>
>
>
>
> Attachment 4.xlsx (12.2K, 无限期)进入下载页面：https://mail.qq.com/cgi-bin/ftnExs_download?k=7a66646612fecb9ee39cc1201131064e48120600010302511d000757511c5659510049545405054c05545d5356520651055e5c5437243420441205055f5c510f444650461f051d4f480a171e370c&t=exs_ftn_download&code=0fdf714a
>
>
>
>
> Attachment 2-mask.edf (2.36M, 无限期)进入下载页面：https://mail.qq.com/cgi-bin/ftnExs_download?k=2a3965662501b99bb7c3c0201035514b1c4d50560104025d4909005e50185a5d570a48520e005049015853560f045701535a5704362c6325104d04055e58060a1019574b5b54100f4411514f185007026404&t=exs_ftn_download&code=d9ef65cd
>
>
>
>
> Attachment 3.docx (1.96M, 无限期)进入下载页面：https://mail.qq.com/cgi-bin/ftnExs_download?k=733763625ed6d0c6eacdc62417610216414306565057080014560652544c050a0d514e5009555214090054070502050a58055451317430784d430201590c55574d175042195519175d58001a315c&t=exs_ftn_download&code=97cb1a09
>
>
>
>
> Attachment 5.docx (4.63M, 无限期)进入下载页面：https://mail.qq.com/cgi-bin/ftnExs_download?k=7362363971878dcfea98937f16660a1f4116570d07040c0014010e5a074b0c020b541b5c09055a1d5f00520c05050b095c01570e307338714d16575a580b5d5e4d4203191852111e5d0d5541305b&t=exs_ftn_download&code=9b690f80
--
Jérôme Kieffer
tel +33 476 882 445
