A fixed annual term reduces omitted-year squared error by at least 10% in 9 of 77 selected HARPS velocity differences
We tested whether a fixed annual term can predict a difference between two reductions of the same stellar spectra. We used a public HARPS-RVBank GitHub snapshot. Fixed quality, mask, count, and span rules selected 59978 exposures in 14149 nights for 77 archive target names before the 2015 instrument change. We compared a linear trend with the same trend plus a 365.256-day sine and cosine. We omitted each year bin with at least five nights and fitted all coefficients to the other nights. The annual term lowered squared prediction error for 32 of 77 targets. It lowered error by at least 10% for nine targets. The median error increased by 0.78%. Undoing the two source night corrections raised the 10% count to 36 targets. A median-based outlier control removed 43 nights and raised that count to 16. Eleven removed nights had paired deviations above one million m/s despite passing the source flag and error cuts. Eight of the original nine targets kept at least 10% lower error under all four stated controls. The package preserves the source columns, provenance, all predictions, figures, and code. These are retrospective tests conditional on published velocity products. They do not identify a planet, choose the correct reduction, or establish a unique calibration cause.
HARPS · radial velocity · annual calibration · SERVAL · DRS · omitted year prediction · reproducible benchmark
References
5- doi:10.1088/0004-637X/808/2/171
This work established an annual HARPS artifact from lines crossing detector block boundaries. Our paired test does not assign that cause.
- usesdoi:10.1051/0004-6361/201936686
We use the HARPS-RVBank velocity products and night corrections through a fixed public GitHub snapshot.
- usesdoi:10.1051/0004-6361/202348263
This paper describes the expanded bank. We use the version 02 GitHub table at a stated commit and hash.
- A free BERV term removes over 99.9% of annual circular signal information at some phases in 151 HARPS designs under white noise sci:sha256:4f2be1c5e46742f40b4598bf5427841e077428f6d351fc726285900bc213d5ad
This verified SCI study measures annual signal information loss in HARPS designs with a free Earth-velocity term. We add measured paired-velocity prediction results.
- Fits to 57 fixed HARPS candidate curves give an 8.85 m/s DRS–SERVAL difference for GJ 579.2 sci:sha256:19d09eb0566dc992a9b1d674dec9352a8a308e851342e29b1c3fe514d0256f47
This verified SCI study compares paired reductions for fixed candidate curves and reports extreme DRS errors. We test a fixed annual term across a separate sample rule.
Cited by
3 publicationsPublic catalog entries on Sciscape.
- A free scale on each released correction curve keeps at least 67% of annual circular signal information in 105 CARMENES series under the stated diagonal weights
It supplies a related test of annual differences between HARPS reductions and their sensitivity to nightly corrections. Our test measures in-sample changes, not omitted-year prediction.
- Earlier linear Earth velocity fits reduce later squared loss by at least 10% in 11 of 74 selected HARPS difference series before the adopted 2015 boundary, with free year means
This close prior study tests annual paired differences in omitted years and includes a separate forward split. We add fixed BERV functions, earlier-only year-contrast fits, and an explicit transfer test on the corrected CDS release. We read its public main text but did not verify its complete package or reuse its files.
- A 1417-day term reverses the prediction gain at a fixed 71.45-day period in the GJ 9527 HARPS velocity difference
Prior work tests annual velocity-difference predictions in another HARPS sample.