pith. sign in

arxiv: 1906.10479 · v1 · pith:44SHGHNRnew · submitted 2019-06-25 · 🌌 astro-ph.IM

Conversion of Tycho-2 to Johnson-Cousins Magnitudes in the Gaia Era

Stephen C. Schiff This is my paper

Pith reviewed 2026-05-25 16:11 UTC · model grok-4.3

classification 🌌 astro-ph.IM
keywords Tycho-2Johnson-Cousins magnitudesmagnitude transformationphotometryGaia DR2machine learningcomparison stars
0
0 comments X

The pith

Machine learning on Gaia data converts Tycho-2 magnitudes to the Johnson-Cousins system for use as photometry comparison stars.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper derives a set of equations that map Tycho-2 BT and VT magnitudes into Johnson B and V magnitudes. It trains the mapping on 558 bright standard stars using one-step supervised learning with weight decay and 10-fold cross validation, then applies the same process to the associated standard deviations. The resulting transformations show average errors much smaller than 1 mmag with standard deviations that match the input precision, allowing a large share of the 2.5 million Tycho-2 stars to serve as reliable comparison sources in two-color bright-star ensemble photometry.

Core claim

The learned mapping from Tycho-2 to Johnson-Cousins magnitudes is essentially unbiased, with average errors much less than 1 mmag in both B and V, and the predicted standard deviations accurately reflect the actual errors, so that they can be used directly to rank comparison-star candidates from the full Tycho-2 catalog.

What carries the argument

One-step supervised learning with weight decay regularization and 10-fold cross validation applied to the 558 standard stars to produce both magnitude transformations and uncertainty transformations.

If this is right

  • A substantial portion of the 2.5 million Tycho-2 stars can now be treated as usable comparison stars for two-color ensemble photometry.
  • The derived standard-deviation transformations allow reliable suitability ranking of candidate comparison stars.
  • The in-sample and cross-validation error statistics remain comparable to the input data precision.
  • The approach supplies both transformed magnitudes and their uncertainties in a single consistent framework.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same supervised-learning procedure could be retrained on additional Gaia-calibrated standards to extend coverage to fainter Tycho-2 stars.
  • Comparison of the new transformations against existing literature relations on overlapping stars would quantify any remaining color-dependent residuals.
  • The availability of uncertainty estimates opens the possibility of weighting individual comparison stars by their predicted precision in ensemble solutions.

Load-bearing premise

The 558 standard stars brighter than magnitude 11 form a representative training set whose learned mapping generalizes without significant bias across the full magnitude and color range of the Tycho-2 catalog.

What would settle it

Independent B and V measurements of Tycho-2 stars outside the training set that reveal systematic offsets larger than the reported millimeter-level errors would falsify the claim of unbiased generalization.

read the original abstract

We take advantage of the availability of precision parallax data from Gaia Data Release 2 together with machine learning to develop a set of equations for transforming Tycho-2 (VT, BT) magnitudes into the Johnson-Cousins (J-C) system. Starting with data for 558 standard stars with apparent magnitudes brighter than 11.0, we employed one step supervised learning with weight decay regularization and 10-fold cross validation to produce a set of transformation equations from Tycho-2 into J-C, which in turn were used to derive transformations of the Tycho-2 standard deviations into the J-C system. Both the aggregated cross validation data sets and the in-sample results from the final training were essentially unbiased (average errors << 1 mmag in both B and V) and had error standard deviations comparable to those of the input data. Comparison of errors in- and out-of-sample indicate modest generalization error growth. Moreover, testing of the distributions of the normalized errors indicated that the predicted standard deviations are accurate, enabling them to be reliably employed in the suitability ranking of comparison star candidates. These results thus enable utilization of a substantial portion of the 2.5 million star Tycho-2 data set as comparison stars for two-color bright star ensemble photometry.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript develops a set of transformation equations from Tycho-2 BT and VT magnitudes to Johnson-Cousins B and V using one-step supervised machine learning with weight decay regularization, trained and 10-fold cross-validated on 558 standard stars brighter than 11.0 mag. It reports near-zero average bias, error standard deviations comparable to input data, and accurate normalized residual distributions, concluding that these enable use of a substantial portion of the 2.5 million Tycho-2 stars as comparison stars for two-color bright-star ensemble photometry.

Significance. If the learned mapping generalizes reliably, the work would expand the available pool of comparison stars for ensemble photometry by providing usable transformations and error estimates for the large Tycho-2 catalog, a practical contribution in the Gaia era. The reported cross-validation, in-sample/out-of-sample error comparison, and normalized-error distribution tests provide concrete empirical support for the transformations within the training magnitude range.

major comments (2)
  1. [Abstract] Abstract and results: the central claim that the transformations enable utilization of a substantial portion of the full 2.5 million star Tycho-2 catalog rests on the untested assumption that the mapping learned from the 558 stars <11.0 mag generalizes without bias to fainter magnitudes; all validation (10-fold CV, residual distributions) is confined to the bright training subset, leaving potential magnitude-dependent systematics unexamined.
  2. [Results] Methods and results: the supervised learning produces data-driven fits whose generalization error is described as modest, yet no independent test set at V>11 (or color/magnitude slices outside the training distribution) is presented to quantify any systematic offset that would affect the suitability ranking of comparison-star candidates across the catalog.
minor comments (1)
  1. [Methods] The role of Gaia DR2 parallax data in sample selection or feature construction is mentioned in the abstract but not detailed in the transformation procedure; clarifying this would improve reproducibility.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for the careful review and constructive comments highlighting the limits of our validation. We address each major comment below. Where the points identify genuine gaps, we will revise the manuscript to add explicit caveats rather than overstate the results.

read point-by-point responses

  1. Referee: [Abstract] Abstract and results: the central claim that the transformations enable utilization of a substantial portion of the full 2.5 million star Tycho-2 catalog rests on the untested assumption that the mapping learned from the 558 stars <11.0 mag generalizes without bias to fainter magnitudes; all validation (10-fold CV, residual distributions) is confined to the bright training subset, leaving potential magnitude-dependent systematics unexamined.

    Authors: We agree that the 10-fold CV and residual tests are confined to the <11.0 mag training stars and that no direct test of magnitude-dependent bias exists for fainter Tycho-2 stars. The transformations are derived from color terms that are in principle magnitude-independent, and the training sample spans the color range of the catalog, but this does not substitute for an explicit check at fainter magnitudes. We will revise the abstract and add a limitations paragraph stating that the transformations are validated only for V < 11 and that application to fainter stars involves an untested extrapolation. revision: yes

  2. Referee: [Results] Methods and results: the supervised learning produces data-driven fits whose generalization error is described as modest, yet no independent test set at V>11 (or color/magnitude slices outside the training distribution) is presented to quantify any systematic offset that would affect the suitability ranking of comparison-star candidates across the catalog.

    Authors: No independent test set at V > 11 was available because the 558 standards with published Johnson-Cousins photometry and Gaia DR2 parallaxes are all brighter than this limit; fainter Tycho-2 stars lack equivalent ground-based J-C measurements for direct comparison. The modest in-sample vs. out-of-sample error growth we report is therefore only within the bright regime. We will expand the results section to discuss this constraint and its effect on suitability ranking for any fainter candidates. revision: partial

standing simulated objections not resolved
  • Absence of an independent test set at V > 11; no additional standard stars with both Tycho-2 and Johnson-Cousins photometry exist in the literature that could be used for such a test.

Circularity Check

0 steps flagged

Empirical ML calibration on standard stars with CV yields independent transformations for Tycho-2 catalog

full rationale

The derivation fits transformation equations via supervised learning (weight decay, 10-fold CV) on 558 standard stars with known J-C magnitudes and Tycho-2 photometry, then applies the resulting equations to the remaining Tycho-2 stars. This is a conventional empirical mapping; the outputs for the 2.5 M catalog stars are not equivalent by construction to the training inputs, and CV supplies explicit out-of-sample checks within the sampled distribution. No self-citations, uniqueness theorems, or definitional loops appear in the chain. Generalization beyond the bright training subset is a separate empirical-validity question, not a circularity reduction.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the representativeness of the 558 standard stars and the assumption that Gaia DR2 parallaxes add useful information for the magnitude transformation; the ML weights constitute fitted parameters whose values are not reported.

free parameters (1)
  • ML model weights and regularization strength
    Determined by supervised learning with weight decay on the 558-star training set; exact values not stated in abstract.
axioms (1)
  • domain assumption Gaia DR2 parallaxes provide accurate distances for the selected standard stars that improve the magnitude transformation.
    Abstract states the use of precision parallax data from Gaia DR2 as a key input to the learning process.

pith-pipeline@v0.9.0 · 5757 in / 1202 out tokens · 47707 ms · 2026-05-25T16:11:26.090790+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

11 extracted references · 11 canonical work pages

  1. [1]

    Spectrophotometric

    pp. 57; 133; 150 Bessell, Michael S., PASP 112 : 961o965, 2000 July, DOI 10.1086/664083 Bowker, A.H. & G.J. Lieberman: Engineering Statistics, 2nd Ed. (PrenticeoHall,

    work page doi:10.1086/664083 2000

  2. [2]

    1, 1997 pp 57o61 Fabricius, C., et

    p 97 ESA, The Hipparcos and Tycho Catalogs, ESA SPo1200, Vol. 1, 1997 pp 57o61 Fabricius, C., et. al., A&A 384 :180o189, 2002, DOI 10.1051/0004o6361/20011822 Feller, W, An introduction to probability theory and its applications, 3 rd Ed. (Wiley,

    work page doi:10.1051/0004o6361/20011822 1997

  3. [3]

    230 Fowler, C.A., Proc

    p. 230 Fowler, C.A., Proc. 1975 IEEE Int. Radar Conf

    work page 1975

  4. [4]

    2016, DOI 10.1051/0004o6361/201629272 Gaia Collaboration, A&A:616 , id A1, 22pp, 2018 August, DOI 10.1051/0004o6361/201833051 Høg, E., et

    Gaia Collaboration, A&A 595 , idA1, 36pp. 2016, DOI 10.1051/0004o6361/201629272 Gaia Collaboration, A&A:616 , id A1, 22pp, 2018 August, DOI 10.1051/0004o6361/201833051 Høg, E., et. al., A&A:355 , L27oL30, 2000 March Kilkenny, D. et al. Mon. Not. R. Astron. Soc. 294 :93o104 1998 Landolt, Arlo U., AJ 88 (3):439o460, March 1983, DOI 10.1086/113329 Lilliefors...

    work page doi:10.1051/0004o6361/201629272 2016

  5. [5]

    Astronomy Reports , year = 2017, month = jan, volume =

    pp. 136o137 Samus', N.N., et al. ARep 61 (1),80o88, January 2017, DOI 10.1134/s1063772917010085 Sachs, Lothar Applied Statistics, 2nd Ed. (Springer,

    work page doi:10.1134/s1063772917010085 2017

  6. [6]

    pp. 255; 258; 259; 330o331 8 Schiff Appendix: The Database For training, the data were sorted in order of increasing right ascension to randomize the magnitude and χ values to form training and validation sets. Here, the sort is by color index. In the table, the first column indicates the source; the next list the apparent magnitudes from the respective s...

    work page 1998

  7. [7]

    027 2.051 0.031 0.043 2.4663 M 9459 1323 1 19 52 53.1 o76 10 46 7.977 8.015 1.455 0.01 5 1.406 0.011 0.049 4.8442 M 9239 473 1 12 01 50.8 o73 17 55 6.856 6.905 1.669 0.015 1 .614 0.010 0.055 8.9045 M 7832 2493 1 14 56 24.8 o44 42 15 6.754 6.804 1.194 0.01 5 1.137 0.010 0.057 7.5158 M 9239 1582 1 12 17 06.6 o73 32 02 8.645 8.724 o0.053 0.016 o0.113 0.013 0...

    work page 2016

  8. [8]

    031 0.890 0.034 0.188 1.1217 M 7639 818 1 06 43 33.6 o44 03 18 8.269 8.433 1.686 0.015 1.497 0.012 0.189 4.3867 M 7882 17 1 17 17 05.5 o44 46 43 6.458 6.714 o3.724 0.014 o3.913 0.010 0.189 0.8231 M 7639 2310 1 06 48 18.2 o43 48 03 7.416 7.576 1.804 0.015 1.607 0.010 0.197 6.8478 M 8167 362 1 09 26 25.4 o46 07 32 7.995 8.175 1.417 0.016 1 .218 0.011 0.199 ...

    work page 1925

  9. [9]

    497 0.010 0.258 3.8171 12 Schiff K, L, M TYC RA-2000 Dec-2000 V B bt σBT vt σVT χ Plx M 9314 780 1 19 49 25.3 o72 30 12 5.389 5.625 2.494 0.014 2.234 0.009 0.260 23.0320 M 8233 3211 1 12 03 06.7 o48 11 42 6.615 6.849 0.763 0.014 0.502 0.010 0.261 5.9471 M 9412 431 1 12 27 37.1 o75 28 21 8.024 8.255 2.842 0.016 2.579 0.011 0.263 8.0891 M 7548 698 1 01 36 0...

    work page 2000

  10. [10]

    490 0.015 1.034 2.0846 M 9142 2815 1 01 08 06.8 o73 00 48 9.096 10.008 2.984 0.0 29 1.948 0.018 1.036 3.5936 M 8167 1048 1 09 29 05.3 o45 27 25 9.078 10.005 1.324 0.02 7 0.280 0.016 1.044 1.6655 M 8889 524 1 04 58 16.6 o66 15 36 9.274 10.197 2.021 0.03 4 0.975 0.021 1.046 2.1195 M 7832 701 1 14 52 17.7 o43 49 05 8.707 9.643 7.233 0.025 6.179 0.015 1.054 3...

    work page 2016

  11. [11]

    730 0.013 1.174 8.2083 M 8229 2807 1 12 10 39.9 o46 31 45 8.347 9.374 2.105 0.019 0.930 0.012 1.175 3.1218 M 8170 165 1 09 19 00.2 o47 58 04 7.026 8.045 4.141 0.016 2.964 0.010 1.177 14.6343 M 8010 425 1 22 46 36.0 o43 53 09 9.230 10.240 2.006 0.034 0.825 0.019 1.181 1.9941 M 8104 1227 1 06 39 42.4 o45 33 38 8.584 9.601 1.996 0.024 0.811 0.014 1.185 2.673...

    work page 2083