PACO: Co-eval Radio-to-mm Extragalactic Survey

• PACO is a coordinated radio-to-millimetre survey that integrates ATCA and Planck data to study compact extragalactic sources and overcome variability challenges.
• It employs a co-eval observational strategy with near-simultaneous flux measurements from 4.5 to 217 GHz, enabling precise broadband spectral modeling.
• The project refines AGN spectral classifications, validates Planck photometry, and improves radio source counts for accurate CMB foreground estimation.

PACO, the Planck–ATCA Co-eval Observations project, is a coordinated radio-to-millimetre survey of compact extragalactic sources designed to combine quasi-simultaneous Australia Telescope Compact Array measurements between 4.5 and 40 GHz with ESA Planck compact-source photometry up to 217 GHz. Its defining methodological feature is co-eval scheduling with Planck’s first two all-sky surveys, implemented to minimize the impact of strong source variability on broadband spectral energy distributions. The project produced a homogeneous catalogue for 464 southern sources and established a detailed empirical description of mm-selected radio-source spectra, variability, and photometric cross-calibration in the Planck era (Massardi et al., 2015).

1. Scientific rationale and co-eval strategy

PACO was conceived in response to two closely connected observational circumstances. First, Planck opened a broad compact-source window from 33 to 857 GHz, including mm-wave regimes that had been only sparsely characterized for extragalactic radio sources. Second, the relevant source population is strongly variable, so non-simultaneous flux measurements from different instruments can distort spectral indices, introduce artificial curvature, and complicate physical interpretation. PACO therefore targeted frequencies below and overlapping the Planck Low Frequency Instrument bands and scheduled ATCA observations within 10 days of Planck scans at 30, 44, and 70 GHz, so that the combined spectral energy distributions would trace essentially the same source state (Massardi et al., 2015).

The project used the On-the-Flight Forecaster to predict Planck scan times and maximize co-eval scheduling. This observational design was especially important because Planck compact-source flux densities are time-averaged over survey windows, whereas PACO measured individual source states with higher angular resolution. The project rationale was thus simultaneously astrophysical and methodological: to recover reliable broadband spectra, to characterize variability on timescales relevant to Planck averaging, and to provide an external validation set for Planck compact-source photometry.

Earlier PACO studies had already shown why this was necessary. The bright sample established that most 20 GHz-selected sources are neither simple power laws nor temporally stable at the level required for naive catalogue matching, while the faint sample showed that variability on multi-year baselines is substantial and that source counts below the Planck completeness limit materially affect foreground estimates in CMB maps (Massardi et al., 2010, Bonavera et al., 2011).

2. Observational setup and sample architecture

PACO used ATCA with the Compact Array Broadband Backend, observing in three pairs of 2 GHz-wide bands: 5.5/9 GHz, 18/24 GHz, and 33/39 GHz. Each 2 GHz band was split into $4\times 512$ MHz sub-bands, which allowed the project to trace spectral curvature more accurately than single-band averages. The PACO frequency coverage overlaps Planck LFI at 30, 44, and 70 GHz and, in the combined catalogue analysis, extends through the High Frequency Instrument bands at 100, 143, and 217 GHz; above 217 GHz, potential Galactic dust contamination becomes important for some sources (Massardi et al., 2015).

The parent sample comprised 482 AT20G-selected extragalactic sources at Galactic latitude $|b|>5^\circ$, excluding the Large Magellanic Cloud region. Valid flux densities were obtained for 464 sources in at least one epoch, and these constitute the PACO catalogue. Nineteen extended sources were flagged and excluded from spectral analysis, yielding a statistical sample of 445 sources. Observations were carried out over 65 epochs, amounting to approximately 450 hours between July 2009 and August 2010 (Massardi et al., 2015).

Three principal sub-samples structure the project history. The bright sample contains 189 sources with $S_{20\,\mathrm{GHz}}>500$ mJy and $\delta<-30^\circ$; only 14% are consistent with single power laws in 5–40 GHz, most are down-turning, and 14% are peaked, with a mean observed peak at 16.3 GHz (Massardi et al., 2010). The faint sample contains 159 sources with $S_{20\,\mathrm{GHz}}>200$ mJy in the Southern Ecliptic Pole region; compared with the bright sample, it shows a higher steep-spectrum fraction at lower flux density, while otherwise preserving broadly similar high-frequency spectral behavior (Bonavera et al., 2011). The spectrally selected sample contains 69 sources brighter than 200 mJy at 20 GHz that were inverted or upturning between 5 and 20 GHz; it was used to separate candidate High Frequency Peakers from blazar candidates and to show that genuine HFPs with peaks above 10 GHz are rare, amounting to $<0.5\%$ of the $S_{20\,\mathrm{GHz}}\ge 200$ mJy population (Bonaldi et al., 2012).

3. Calibration, variability, and Planck photometry

PACO flux calibration relied on PKS1934–638 at all bands and Uranus at 7 mm, with bandpass and phase self-calibration and triple-product visibilities for compact-source flux estimation. A known discrepancy between the PKS1934–638 and Uranus models was corrected by adopting the revised PKS1934–638 model, which implied 7 mm corrections of 5.2–10.5% and corrections of at most 0.5% at lower frequencies. Gain-term uncertainties were characterized per epoch, with $\sigma_g\simeq 0.012$ for point sources and 0.05 for extended sources (Massardi et al., 2015).

Variability was explicitly incorporated into the analysis. PACO adopted the variability index

$$V_{\rm rms}=\frac{100}{\langle S \rangle}\sqrt{\frac{\sum (S_i-\langle S \rangle)^2-\sum\sigma_i^2}{n}}.$$

For the catalogue analysis, median variability at 20 GHz is approximately 6% over 6 months and approximately 9% over 9 months; at about 35 GHz the corresponding medians are approximately 7% and approximately 11%. Over 2–4 year baselines, variability rises to about 18–27%, increasing with frequency and marginally with time lag (Massardi et al., 2015). In the faint sample, comparison with earlier AT20G measurements showed an rms amplitude of approximately 40% at 20 GHz over a few years, underscoring the practical importance of the co-eval strategy (Bonavera et al., 2011).

PACO also provided a direct comparison with Planck compact-source photometry. Cross-matching used search radii of half the Planck FWHM, and among the four PCCS photometric measures—DETFLUX, APERFLUX, PSFFLUX, and GAUFLUX—DETFLUX showed the best consistency with PACO for bright point sources.

PCCS measure	Mean relative difference vs PACO
DETFLUX	$0.8\pm1.6\%$ at $|b|>5^\circ$0 GHz; $|b|>5^\circ$1 at $|b|>5^\circ$2 GHz
PSFFLUX	$|b|>5^\circ$3 at $|b|>5^\circ$4 GHz; $|b|>5^\circ$5 at $|b|>5^\circ$6 GHz
APERFLUX	$|b|>5^\circ$7 at $|b|>5^\circ$8 GHz; $|b|>5^\circ$9 at $S_{20\,\mathrm{GHz}}>500$0 GHz
GAUFLUX	$S_{20\,\mathrm{GHz}}>500$1 at $S_{20\,\mathrm{GHz}}>500$2 GHz; $S_{20\,\mathrm{GHz}}>500$3 at $S_{20\,\mathrm{GHz}}>500$4 GHz

DETFLUX and, to a comparable extent, PSFFLUX therefore provide the most appropriate PCCS flux densities for the PACO source population, whereas APERFLUX shows larger scatter and fewer $S_{20\,\mathrm{GHz}}>500$5 detections. GAUFLUX is optimized for extended sources, whereas PACO predominantly targets point-like objects at Planck resolution. The project also identified Eddington bias below the 90% completeness limit and found that in 11 of 16 extreme outliers with PCCS/PACO $S_{20\,\mathrm{GHz}}>500$6, local extended Galactic emission likely contaminated the Planck photometry (Massardi et al., 2015).

4. Spectral modelling and empirical source classes

PACO parameterized spectra using the conventional index definition $S_{20\,\mathrm{GHz}}>500$7 and, for broadband fitting, adopted a continuous double power law,

$S_{20\,\mathrm{GHz}}>500$8

This model was fitted in logarithmic space by $S_{20\,\mathrm{GHz}}>500$9 minimization using nonlinear Generalized Reduced Gradient optimization. The fit contains no discontinuity at a nominal break; instead, curvature emerges from the two asymptotic power-law slopes (Massardi et al., 2015).

The principal result is the predominance of smooth spectra. Among the 290 sources with PACO averages and detections in at least two Planck bands up to 217 GHz, 91% are well fitted by the double power law, with mean reduced $\delta<-30^\circ$0. Only 26 sources, about 9%, fail the adopted goodness-of-fit threshold. This whole-catalogue result generalizes what earlier PACO analyses had already shown at lower frequency coverage: in the bright sample only 14% were consistent with single power laws over 5–40 GHz, and in the spectrally selected sample 57 of 67 sources were adequately described by the same smooth double-power-law form (Massardi et al., 2010, Bonaldi et al., 2012).

Across the bright sample, the median spectral indices trace a continuous steepening with frequency: $\delta<-30^\circ$1 at 5–9 GHz, $\delta<-30^\circ$2 at 9–18 GHz, $\delta<-30^\circ$3 at 18–33 GHz, $\delta<-30^\circ$4 at 44.1–70.4 GHz, $\delta<-30^\circ$5 at 70.4–100 GHz, and $\delta<-30^\circ$6 at 100–143 GHz. The same trend is visible in the faint sample, where the median index changes from $\delta<-30^\circ$7 at 5–10 GHz to $\delta<-30^\circ$8 at 30–40 GHz (Massardi et al., 2015, Bonavera et al., 2011).

PACO also quantified distinct empirical spectral classes. In the bright sample, 66.0% of classified point sources are down-turning, 14.5% are peaked, 10.3% are flat, 4.8% are self-absorbed, 3.6% are steep, and 0.6% are inverted (Massardi et al., 2010). In the faint sample, the principal flux-density dependence is an increase in the steep-spectrum fraction to 13.3% below 500 mJy, while the down-turning fraction remains 65.3% (Bonavera et al., 2011). For the combined PACO+Planck analysis, 47 sources satisfy the “peaked” criterion $\delta<-30^\circ$9 and $S_{20\,\mathrm{GHz}}>200$0; their observed median peak frequency is 18.3 GHz, and for the 29 with redshifts the median rest-frame peak is 39.6 GHz (Massardi et al., 2015).

For peaked spectra fitted by the double-power-law form, the peak frequency is

$S_{20\,\mathrm{GHz}}>200$1

In the bright sample the observed mean $S_{20\,\mathrm{GHz}}>200$2 is approximately 16.4 GHz, whereas the spectrally selected sample found that sources with peaked PACO spectra show a decrease of peak frequency with time at a mean rate of $S_{20\,\mathrm{GHz}}>200$3 GHz yr$S_{20\,\mathrm{GHz}}>200$4 on an average timescale of $S_{20\,\mathrm{GHz}}>200$5 yr (Massardi et al., 2010, Bonaldi et al., 2012).

5. Physical interpretation of the PACO spectra

The PACO results altered the standard phenomenology of compact radio sources at millimetre wavelengths. The classical cm-wave dichotomy between flat-spectrum compact sources and steep-spectrum extended sources does not persist into the PACO–Planck frequency range. Most sources show nearly flat or mildly inverted spectra at centimetre wavelengths and then steepen above about 30 GHz, consistent with synchrotron emission becoming optically thin (Massardi et al., 2015).

The smoothness of the spectra is central to the physical interpretation. Because 91% of the broadband spectra are well fitted by a single continuous double power law across roughly two decades in frequency, PACO concluded that the emission is typically dominated by a single compact emitting region. This stands against the traditional explanation that “flat” cm-wave spectra are produced by the superposition of many compact regions self-absorbed up to different frequencies. The inference is reinforced by the spectrally selected sample, in which about 89% of blazar candidates have smooth spectra of the same form (Massardi et al., 2015, Bonaldi et al., 2012).

PACO also examined whether the spectra show the high-frequency steepening expected from synchrotron ageing. In the continuous-injection picture with electron energy distribution $S_{20\,\mathrm{GHz}}>200$6, the injected synchrotron index is

$S_{20\,\mathrm{GHz}}>200$7

and above the break one expects

$S_{20\,\mathrm{GHz}}>200$8

The standard spectral-ageing relation quoted by PACO is

$S_{20\,\mathrm{GHz}}>200$9

with $<0.5\%$0 in $<0.5\%$1G and

$<0.5\%$2

No clear spectral break of this kind is seen up to 217 GHz. PACO therefore argued that the absence of a detectable $<0.5\%$3 may imply young source ages and/or Doppler up-shifting of the break frequency in relativistic jets; non-standard acceleration can also mitigate ageing (Massardi et al., 2015).

Broadband radio–infrared spectral energy distributions in the spectrally selected sample further support a predominantly low-synchrotron-peak blazar population. Using PACO plus WISE data, most blazars showed $<0.5\%$4 peaks below $<0.5\%$5 GHz, with a median peak wavelength near $<0.5\%$6m; only six objects had $<0.5\%$7 GHz (Bonaldi et al., 2012).

6. Catalogue, source counts, and lasting significance

The PACO catalogue, released as an online supplement, lists positions, flux densities per 512 MHz sub-band and per 2 GHz band across multiple epochs, morphology flags for extended sources, and quality assessments. Cross-identifications with the Planck PCCS are included, and 290 PACO sources have detections in at least two Planck bands up to 217 GHz. For DETFLUX at $<0.5\%$8, the combined catalogue contains 259 detections at 30 GHz, 149 at 44 GHz, 175 at 70 GHz, 282 at 100 GHz, 311 at 143 GHz, and 282 at 217 GHz (Massardi et al., 2015).

One immediate legacy of the project is photometric validation. PACO showed that DETFLUX is the most consistent PCCS flux-density measure for compact point sources at 30–44 GHz, thereby informing best practice for Planck source analysis. Earlier comparison with the ERCSC in the faint-sample region had already indicated that the catalogue reliability there exceeded 95%, with missing matches largely attributable to the Large Magellanic Cloud, low Galactic latitude, or source extension (Bonavera et al., 2011).

A second legacy is statistical control of unresolved-source foregrounds. The faint sample extended the source counts at approximately 33 and 40 GHz downward by a factor of about five, to approximately 200 mJy, relative to Planck ERCSC counts. This improved estimates of the Poisson contribution of unresolved radio sources to small-scale CMB anisotropy, especially below about 100 GHz where radio sources dominate the contaminating signal (Bonavera et al., 2011). The whole-catalogue analysis sharpened this further by quantifying the systematic steepening of source spectra toward mm wavelengths, which directly affects extrapolations below the Planck detection threshold (Massardi et al., 2015).

A third legacy is the physical reframing of compact AGN spectra in the Planck era. PACO established that smooth, single-component spectra are the rule rather than the exception across 5–217 GHz, that the flat/steep cm-wave taxonomy fails in the mm regime, and that no electron-ageing break is evident up to 217 GHz. These results constrain jet-emission models, support interpretations in which one dominant compact region shapes the broadband radio-mm spectrum, and provide a variability-controlled empirical basis for subsequent work on blazar cores, source counts, and Planck foreground validation (Massardi et al., 2015).