California Legacy Survey: Exoplanets & Stellar Activity

• California Legacy Survey is a long-term program that systematically catalogs exoplanets and stellar magnetic cycles from high-precision spectroscopic data.
• It combines data from Keck-HIRES, APF-Levy, and Lick-Hamilton to achieve 1–3 m/s precision, enabling robust detection from 0.03 to 30 AU.
• The survey also tracks chromospheric activity to link planetary occurrence, multiplicity, and host-star properties for comprehensive demographic studies.

California Legacy Survey (CLS) is a long-term, high-precision radial-velocity and spectroscopic program developed by the California Planet Search team to deliver a blind, uniform census of exoplanets and long-period companions around nearby FGKM stars, while also using the same spectra to characterize stellar chromospheric activity and magnetic cycles (Rosenthal et al., 2021, Isaacson et al., 2024). In its core radial-velocity release, the survey assembled 103,991 measurements for 719 stars and produced a uniform catalog of 178 companions, including 164 previously known exoplanets and 14 newly discovered or revised exoplanets and substellar companions (Rosenthal et al., 2021). The broader CLS framework also includes the California-Kepler Survey (CKS), described as the spectroscopic cornerstone of the broader California Legacy Survey, which extends the legacy of uniform stellar characterization to Kepler planet hosts and connects host-star properties, chromospheric activity, stellar rotation, and exoplanet demographics (Isaacson et al., 2024).

1. Survey conception, sample definition, and observational scope

CLS was constructed from several “blind” California Planet Search occurrence-rate surveys, most notably the Keck Planet Search with 585 stars, the Eta-Earth sample with 166 stars, and a 25 pc volume-limited subset, supplemented by 31 additional CPS-observed stars meeting uniform quality and statistical-rigor criteria (Rosenthal et al., 2021). The target-selection logic explicitly excluded biases toward metal-rich stars or previously known planet hosts, and also excluded N2K/M2K, Friends-of-Hot-Jupiters, transit-survey targets, young or IR-excess stars, and subgiant surveys (Rosenthal et al., 2021). The resulting stellar sample spans spectral types F–M on the main sequence, with selection cuts requiring at least 20 total radial-velocity measurements, at least 10 post-2004 HIRES observations, and a baseline of at least 8 years (Rosenthal et al., 2021).

The sample occupies a broad region of nearby stellar-parameter space: $V \approx 3$–11, distances $\lesssim 60$ pc with median $\sim 20$ pc, $T_{\rm eff} \approx 3200$–6500 K, stellar masses 0.2–1.5 $M_\odot$ with median 0.93 $M_\odot$, and $[{\rm Fe/H}]$ from $-0.8$ to $+0.5$ (Rosenthal et al., 2021). Stellar activity in the parent sample is low to moderate, with median $\log R'_{\rm HK} \lesssim -4.7$ (Rosenthal et al., 2021). This construction was designed to support population-level occurrence measurements rather than case-by-case follow-up of already unusual systems.

A parallel CLS activity program targeted slowly rotating FGKM dwarfs within $\lesssim 60$0 pc, with median distance $\lesssim 60$1 pc, chosen to be amenable to $\lesssim 60$2-level radial velocities (Isaacson et al., 2024). That activity sample comprises 710 stars and extends from mid-2005 through 2023 October, or $\lesssim 60$3 years, with the authors citing “over twenty years” when archival data are included (Isaacson et al., 2024). This dual structure—planet census plus activity monitoring—is central to the CLS design.

2. Instrumentation, radial-velocity pipeline, and completeness formalism

CLS combines three major spectroscopic platforms. Keck-HIRES, operating from 1996 to the present, provided $\lesssim 60$4–86,000 spectra with iodine-cell calibration, with pre-2004 radial-velocity precision of $\lesssim 60$5 and post-2004 precision of $\lesssim 60$6, contributing $\lesssim 60$7 radial velocities (Rosenthal et al., 2021). APF-Levy, in use from 2013 onward, uses the same iodine Doppler code as HIRES and achieves $\lesssim 60$8, contributing $\lesssim 60$9 radial velocities (Rosenthal et al., 2021). Lick-Hamilton, operating from 1987 to 2011, delivered 3–10 $\sim 20$0 precision and added $\sim 20$1 radial velocities (Rosenthal et al., 2021). Overall, the survey reaches per-star medians of $\sim 20$2 and baseline $\sim 20$3 yr, while stars with detected planets were observed more intensively, with median $\sim 20$4 (Rosenthal et al., 2021).

All radial velocities were derived through forward modeling of iodine-superimposed spectra against a deconvolved stellar template, fitting each $\sim 20$5 Å segment for Doppler shifts, line-spread function, and continuum (Rosenthal et al., 2021). The automated search pipeline, RVSearch, performs iterative planet detection and characterization using a likelihood model of the form

$\sim 20$6

with

$\sim 20$7

including per-instrument zero points, jitter added in quadrature, Keplerian terms, and optional linear and quadratic trends (Rosenthal et al., 2021). The search proceeds through an iterative grid in $\sim 20$8 from 2 days to $\sim 20$9 the observational baseline, using $T_{\rm eff} \approx 3200$0 as periodogram power and adopting a detection threshold set at an empirical false-alarm probability of 0.1% (Rosenthal et al., 2021). Posterior sampling uses RadVel plus emcee in the basis $T_{\rm eff} \approx 3200$1, while TheJoker is used for sparsely constrained orbits (Rosenthal et al., 2021).

Survey completeness is a defining methodological component. For each star, completeness $T_{\rm eff} \approx 3200$2 was computed by injecting Keplerian signals that are log-uniform in period and mass and $T_{\rm eff} \approx 3200$3-distributed in eccentricity into the real radial-velocity time series, then measuring the fraction recovered above the detection threshold (Rosenthal et al., 2021). Typical 50% completeness curves show sensitivity above 50% to $T_{\rm eff} \approx 3200$4 out to 5 AU, above 50% to $T_{\rm eff} \approx 3200$5 out to 10 AU, and falling to $T_{\rm eff} \approx 3200$6 beyond 20 AU (Rosenthal et al., 2021). Population occurrence in a region of period–mass space is then written as

$T_{\rm eff} \approx 3200$7

providing the statistical basis for the occurrence studies that followed (Rosenthal et al., 2021).

3. Giant-planet occurrence across orbital separation

CLS II used the uniform radial-velocity sample to measure giant-planet occurrence as a function of orbital separation from 0.03 to 30 AU and found that giant planets are more prevalent at 1–10 AU than at smaller or larger separations (Fulton et al., 2021). In non-parametric form, the survey reports $T_{\rm eff} \approx 3200$8 planets per 100 stars, $T_{\rm eff} \approx 3200$9 per 100 stars, and $M_\odot$0 per 100 stars (Fulton et al., 2021). The $M_\odot$1 enhancement between $M_\odot$2 AU and 1–10 AU is highly statistically significant, and the decline beyond 10 AU is favored over models with flat or increasing occurrence at the $M_\odot$3 to $M_\odot$4 level (Fulton et al., 2021).

The same framework applied to sub-Jovian planets with masses 0.1–1 $M_\odot$5 yields $M_\odot$6 per 100 stars and $M_\odot$7 per 100 stars, with the 10–30 AU rate not well constrained because of few detections (Fulton et al., 2021). The survey therefore identifies an orbital region near and just beyond the water-ice line where both Jovian and sub-Jovian planets are enhanced.

To describe the semi-major-axis dependence in a compact analytic form, CLS II fit a broken power law,

$M_\odot$8

where $M_\odot$9, obtaining $M_\odot$0, $M_\odot$1 au, $M_\odot$2, and $M_\odot$3 (Fulton et al., 2021). Because 99.4% of the posterior samples have $M_\odot$4, the data favor a turnover or decline beyond the ice line (Fulton et al., 2021).

CLS II also corroborates that cold-gas-giant occurrence increases with host-star mass and metallicity. Restricting to planets $M_\odot$5 between 1 and 5 AU, occurrence rises from $M_\odot$6 per 100 stars for $M_\odot$7 to $M_\odot$8 per 100 for $M_\odot$9, and from $[{\rm Fe/H}]$0 per 100 for $[{\rm Fe/H}]$1 to $[{\rm Fe/H}]$2 per 100 for $[{\rm Fe/H}]$3 dex (Fulton et al., 2021). The paper further notes statistical consistency with direct-imaging and microlensing surveys when those results are cast into common mass–semi-major-axis bins (Fulton et al., 2021).

4. Planetary architectures, multiplicity, and host-star metallicity

The uniform CLS catalog enabled direct comparison between systems containing a single giant planet and systems containing multiple giant planets. Of the 178 cataloged companions, 134 have minimum masses $[{\rm Fe/H}]$4; these include 65 “lonely” giants and 69 giants in 31 multi-giant systems (Rosenthal et al., 2023). Rosenthal et al. modeled the eccentricity distributions of singles and multis with hierarchical Bayesian Beta distributions,

$[{\rm Fe/H}]$5

and found that the two populations are distinct at $[{\rm Fe/H}]$6 (Rosenthal et al., 2023). The multiple-giant population has $[{\rm Fe/H}]$7, whereas the single-giant population extends to $[{\rm Fe/H}]$8, while both have $[{\rm Fe/H}]$9 of orbits at $-0.8$0 (Rosenthal et al., 2023). The reported morphology is a pile-up of nearly circular orbits plus a long high-eccentricity tail for singles, and a distribution concentrated at modest eccentricity for multis (Rosenthal et al., 2023).

Host-star metallicities also differ. High-resolution spectra analyzed by SpecMatch yielded $-0.8$1 dex for singles and $-0.8$2 dex for multis (Rosenthal et al., 2023). Using $-0.8$3 realizations of metallicity cumulative distributions and Anderson–Darling testing, the study found $-0.8$4 in 82.2% of trials; a Kolmogorov–Smirnov test gave similar results (Rosenthal et al., 2023). The occurrence density as a function of semi-major axis further shows a $-0.8$5 significant hot-Jupiter pile-up at $-0.8$6 AU among singles that is not seen among multi-giant systems (Rosenthal et al., 2023).

The same comparison identified a mass difference and an intra-system mass correlation. The median $-0.8$7 of giants in singles is 0.92 $-0.8$8, versus 1.71 $-0.8$9 in multis, with an Anderson–Darling test giving $+0.5$0 (Rosenthal et al., 2023). For adjacent giants ordered by semi-major axis within multi-giant systems, the Pearson correlation coefficient is $+0.5$1; bootstrapping the observed mass catalog yields a 99.71% exclusion of this correlation under reshuffling, and synthetic populations with a uniform $+0.5$2 distribution produce $+0.5$3 only 1% of the time (Rosenthal et al., 2023). The resulting “giant peas in a pod” effect has $+0.5$4, implying paired giants tend to lie within a factor of $+0.5$5 in mass (Rosenthal et al., 2023).

A separate CLS analysis addressed whether distant giant companions are enhanced around systems hosting close-in small planets, especially in metal-rich stars. Using the Rosenthal et al. 2021 sample filtered to 594 Sun-like stars with $+0.5$6, measured $+0.5$7, and at least 20 radial-velocity observations, and adopting the same planet definitions as Bryan and Lee (2024), the study found $+0.5$8 and $+0.5$9 (Zandt et al., 2024). The conditional rate is therefore statistically indistinguishable from, or slightly lower than, the field rate, and a factor-of-two enhancement is in the 97.5% upper tail of the posterior, constituting a $\log R'_{\rm HK} \lesssim -4.7$0 tension with Bryan and Lee’s result (Zandt et al., 2024). The paper attributes the discrepancy to small-number statistics, sample inhomogeneity, and completeness-correction methodology, and states that larger, homogeneous samples will be required to settle the question at $\log R'_{\rm HK} \lesssim -4.7$1 confidence (Zandt et al., 2024).

5. Chromospheric activity monitoring and stellar magnetic cycles

CLS V repurposed the long time-series HIRES archive to construct a chromospheric activity survey of 710 main-sequence stars, based on 52,372 Ca II H and K spectra after quality cuts (Isaacson et al., 2024). The spectra were obtained mainly with the B5 and C2 deckers, giving $\log R'_{\rm HK} \lesssim -4.7$2, with occasional B1 and B3 templates at $\log R'_{\rm HK} \lesssim -4.7$3 (Isaacson et al., 2024). S-values on the Mount Wilson scale were measured from the H and K cores and two 20-Å continuum windows according to

$\log R'_{\rm HK} \lesssim -4.7$4

with $\log R'_{\rm HK} \lesssim -4.7$5 and $\log R'_{\rm HK} \lesssim -4.7$6 determined from standard stars and individual-observation uncertainties on $\log R'_{\rm HK} \lesssim -4.7$7 of $\log R'_{\rm HK} \lesssim -4.7$8 (Isaacson et al., 2024). The corresponding dimensionless chromospheric metric was then computed as

$\log R'_{\rm HK} \lesssim -4.7$9

with $\lesssim 60$00 (Isaacson et al., 2024).

For cycle searches, the survey required at least 45 post-2005 observations, producing a cycle-search subsample of 285 stars (Isaacson et al., 2024). Periods from 100 to 10,000 days were scanned with a generalized Lomb–Scargle periodogram, after which the strongest peak was fit with a zero-eccentricity sinusoid,

$\lesssim 60$01

and subjected to robustness filters: peak power $\lesssim 60$02, second-tallest peak $\lesssim 60$03 of the main peak, and $\lesssim 60$04 (Isaacson et al., 2024). Under those criteria, 138 stars, or 48% of the sufficiently sampled subsample, show robust cycles (Isaacson et al., 2024).

The detected cycle periods span 2.3–23 yr, with a broad peak near 7–12 yr, and amplitudes $\lesssim 60$05–1.6, corresponding to $\lesssim 60$06–0.4 dex (Isaacson et al., 2024). A particularly tight empirical relation appears for main-sequence stars with $\lesssim 60$07 between $\lesssim 60$08 and $\lesssim 60$09: cycle period increases systematically as $\lesssim 60$10 decreases, with 300 K bins yielding $\lesssim 60$11 yr for 5800–6100 K, $\lesssim 60$12 yr for 5500–5800 K, $\lesssim 60$13 yr for 5200–5500 K, and $\lesssim 60$14 yr for 4900–5200 K (Isaacson et al., 2024). For stars less active than $\lesssim 60$15, the correlation breaks down and periods scatter to 10–20 yr (Isaacson et al., 2024). The Sun, with $\lesssim 60$16 K, $\lesssim 60$17, and $\lesssim 60$18 yr, sits squarely on the G-star trend line (Isaacson et al., 2024). Within exoplanet surveys, these S-value and $\lesssim 60$19 time series are used as chromospheric diagnostics that correlate with radial-velocity jitter and can be incorporated through regression or quasi-periodic Gaussian-process modeling (Isaacson et al., 2024).

6. The California-Kepler Survey within CLS and the survey’s long-term legacy

The California-Kepler Survey represents the spectroscopic cornerstone of the broader California Legacy Survey, and its eleventh installment extended CLS methodology to Kepler planet hosts by extracting chromospheric activity measurements from CKS Gaia survey spectra (Isaacson et al., 2024). The parent sample comprises 1189 Kepler Objects of Interest with confirmed or candidate planets, selected to be among the brighter hosts in the Kepler field with $\lesssim 60$20, of which 879 stars have sufficient signal-to-noise in the blue chips to measure the cores of the Ca II H line at 3968 Å and the K line at 3934 Å (Isaacson et al., 2024). All spectra were obtained with HIRES on Keck I at resolving power $\lesssim 60$21 over $\lesssim 60$22–8000 Å (Isaacson et al., 2024).

Using Spectroscopy Made Easy and an updated line list, the survey homogeneously derived $\lesssim 60$23 with typical uncertainties $\lesssim 60$24 K, $\lesssim 60$25 to $\lesssim 60$26 dex, $\lesssim 60$27 to $\lesssim 60$28 dex, and isochronal ages from Dartmouth tracks with median uncertainties of order 25–30%, or roughly $\lesssim 60$29 Gyr for a 5 Gyr star (Isaacson et al., 2024). The Mount Wilson S-index was computed from triangular bandpasses of full-width at half-maximum 1.09 Å centered on the Ca II H and K cores and two 20 Å pseudocontinuum windows, achieving an external precision of $\lesssim 60$30 after calibration to the Mount Wilson scale (Isaacson et al., 2024). In practical form, the pure chromospheric flux ratio is written as

$\lesssim 60$31

where $\lesssim 60$32 is the dimensionless photospheric contribution (Isaacson et al., 2024).

Rotation periods were inferred by inverting the empirical $\lesssim 60$33–Rossby relation of Mamajek and Hillenbrand (2008), with $\lesssim 60$34, and convective turnover times obtained from the Noyes et al. polynomial

$\lesssim 60$35

valid for $\lesssim 60$36 (Isaacson et al., 2024). Comparing activity-derived rotation periods to photometric rotation periods from Kepler light curves revealed a systematic offset along the “Rossby Ridge,” $\lesssim 60$37, with stars older than $\lesssim 60$38 Gyr spinning down more slowly than predicted by standard Skumanich-type braking laws, in direct support of weakened magnetic braking (Isaacson et al., 2024). The same analysis found no statistically significant metallicity trend in the $\lesssim 60$39–$\lesssim 60$40 relation for $\lesssim 60$41, and identified approximately 25 stars with $\lesssim 60$42 that lie slightly below the main sequence in the $\lesssim 60$43 versus $\lesssim 60$44 plane as potential Maunder Minimum analogs, while noting that multi-year time series are not yet available to verify those states (Isaacson et al., 2024).

Taken together, the CLS data products comprise radial velocities from Keck-HIRES, APF-Levy, and Lick-Hamilton, Ca H&K S-values, APT photometry, machine-readable stellar and planetary tables, completeness contours, and the public RVSearch codebase (Rosenthal et al., 2021). The survey has therefore become a benchmark dataset for giant-planet occurrence measurements, differential studies of single and multiple giant-planet systems, chromospheric cycle statistics, age–activity–rotation calibrations, and host-star characterization for exoplanet demographic analyses (Rosenthal et al., 2021, Isaacson et al., 2024). A plausible implication is that CLS functions not only as a planet catalog, but as an integrated empirical framework linking planetary architectures, stellar magnetic evolution, and survey completeness over decadal baselines.