Filename
stringlengths 22
64
| Paragraph
stringlengths 8
5.57k
|
|---|---|
Processed_Using_Program_Induction_to_Interpret_Transition_Sy.txt
|
Harrison, Brent, Ehsan, Upol, and Riedl, Mark O. Ra- tionalization: A neural machine translation approach to generating natural language explanations. arXiv preprint arXiv:1702.07826, 2017.
|
Processed_Using_Program_Induction_to_Interpret_Transition_Sy.txt
|
Hendricks, Lisa Anne, Akata, Zeynep, Rohrbach, Marcus, Donahue, Jeff, Schiele, Bernt, and Darrell, Trevor. Gen- erating visual explanations. In European Conference on Computer Vision, pp. 3–19. Springer, 2016.
|
Processed_Using_Program_Induction_to_Interpret_Transition_Sy.txt
|
Brendel, William and Todorovic, Sinisa. Learning spa- tiotemporal graphs of human activities. In Computer vi- sion (ICCV), 2011 IEEE international conference on, pp. 778–785. IEEE, 2011.
|
Processed_Using_Program_Induction_to_Interpret_Transition_Sy.txt
|
Cai, Jonathon, Shin, Richard, and Song, Dawn. Making neural programming architectures generalize via recur- In International Conference on Learning Repre- sion. sentations (ICLR), April 2017.
|
Processed_Using_Program_Induction_to_Interpret_Transition_Sy.txt
|
Devlin, Jacob, Uesato, Jonathan, Bhupatiraju, Surya, Singh, Rishabh, Mohamed, Abdel-rahman, and Kohli, Pushmeet. Robustfill: Neural program learning under noisy i/o. arXiv preprint arXiv:1703.07469, 2017.
|
Processed_Using_Program_Induction_to_Interpret_Transition_Sy.txt
|
Duchi, John, Hazan, Elad, and Singer, Yoram. Adaptive subgradient methods for online learning and stochastic optimization. Journal of Machine Learning Research, 12(Jul):2121–2159, 2011.
|
Processed_Using_Program_Induction_to_Interpret_Transition_Sy.txt
|
Garnelo, Marta, Arulkumaran, Kai, and Shanahan, Murray. Towards deep symbolic reinforcement learning. arXiv preprint arXiv:1609.05518, 2016.
|
Processed_Using_Program_Induction_to_Interpret_Transition_Sy.txt
|
Lake, Brenden M., Salakhutdinov, Ruslan, and Tenen- baum, Joshua B. Human-level concept learning through probabilistic program induction. Science, 350(6266): 1332–1338, 2015.
|
Processed_Using_Program_Induction_to_Interpret_Transition_Sy.txt
|
Letham, Benjamin, Rudin, Cynthia, McCormick, Tyler H, Interpretable classifiers using Madigan, David, et al. rules and bayesian analysis: Building a better stroke pre- diction model. The Annals of Applied Statistics, 9(3): 1350–1371, 2015.
|
Processed_Using_Program_Induction_to_Interpret_Transition_Sy.txt
|
Ling, Wang, Yogatama, Dani, Dyer, Chris, and Blun- som, Phil. Program Induction by Rationale Gener- ation:Learning to Solve and Explain Algebraic Word Problems. arXiv preprint arXiv:1705.04146, 2017.
|
Processed_Using_Program_Induction_to_Interpret_Transition_Sy.txt
|
Maclaurin, Dougal, Duvenaud, David, and Adams, Ryan P. Autograd: Effortless gradients in numpy. In ICML 2015 AutoML Workshop, 2015.
|
Processed_Using_Program_Induction_to_Interpret_Transition_Sy.txt
|
Mnih, Volodymyr, Kavukcuoglu, Koray, Silver, David, Rusu, Andrei A, Veness, Joel, Bellemare, Marc G, Graves, Alex, Riedmiller, Martin, Fidjeland, Andreas K, Ostrovski, Georg, et al. Human-level control through deep reinforcement learning. Nature, 518(7540):529– 533, 2015.
|
Processed_Using_Program_Induction_to_Interpret_Transition_Sy.txt
|
Tokui, Seiya, Oono, Kenta, Hido, Shohei, and Clayton, Justin. Chainer: a next-generation open source frame- work for deep learning. In Proceedings of workshop on machine learning systems (LearningSys) in the twenty- ninth annual conference on neural information process- ing systems (NIPS), 2015.
|
Processed_Using_Program_Induction_to_Interpret_Transition_Sy.txt
|
Negrinho, Renato and Gordon, Geoff. Deeparchitect: Au- tomatically designing and training deep architectures. arXiv preprint arXiv:1704.08792, 2017.
|
Processed_Using_Program_Induction_to_Interpret_Transition_Sy.txt
|
Ribeiro, Marco, Singh, Sameer, and Guestrin, Carlos. Why should I trust you?: Explaining the predictions of any classifier. In ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD), 2016.
|
Processed_Using_Program_Induction_to_Interpret_Transition_Sy.txt
|
Yin, Pengcheng and Neubig, Graham. A syntactic neu- ral model for general-purpose code generation. arXiv preprint arXiv:1704.01696, 2017.
|
Processed_Using_Program_Induction_to_Interpret_Transition_Sy.txt
|
Zahavy, Tom, Ben-Zrihem, Nir, and Mannor, Shie. Visual- izing Dynamics: from t-SNE to SEMI-MDPs. In Proc. 2016 ICML Workshop on Human Interpretability in Ma- chine Learning, pp. 111–115, New York, NY, June 2016.
|
Processed_Using_Program_Induction_to_Interpret_Transition_Sy.txt
|
Solar-Lezama, Armando, Tancau, Liviu, Bodik, Rastislav, Seshia, Sanjit, and Saraswat, Vijay. Combinatorial sketching for finite programs. ACM SIGOPS Operating Systems Review, 40(5):404–415, 2006.
|
Processed_Using_Program_Induction_to_Interpret_Transition_Sy.txt
|
standing convolutional networks. ence on computer vision, pp. 818–833. Springer, 2014. Zoph, Barret and Le, Quoc. Neural architecture search with reinforcement learning. In International Conference on Learning Representations (ICLR), April 2017.
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
We exploit data from the Pan-Andromeda Archaeological Survey (PAndAS) to study the extended structures of M31’s dwarf elliptical companions, NGC147 and NGC185. Our wide- field, homogeneous photometry allows to construct deep colour-magnitude diagrams (CMDs) which reach down to ∼ 3 mag below the red giant branch (RGB) tip. We trace the stellar components of the galaxies to surface brightness of µg ∼ 32 mag arcsec−2 and show they have much larger extents (∼ 5 kpc radii) than previously recognised. While NGC185 retains a regular shape in its peripheral regions, NGC147 exhibits pronounced isophotal twisting due to the emergence of symmetric tidal tails. We fit single Sersic models to composite surface brightness profiles constructed from diffuse light and star counts and find that NGC147 has an effective radius almost 3 times that of NGC185. In both cases, the effective radii that we calculate are larger by a factor of ∼ 2 compared to most literature values. We also calculate revised total magnitudes of Mg = −15.36 ± 0.04 for NGC185 and Mg = −16.36 ± 0.04 for NGC147. Using photometric metallicities computed for RGB stars, we find NGC185 to exhibit a metallicity gradient of [Fe/H]∼ −0.15 dex/kpc over the radial range 0.125 to 0.5 deg. On the other hand, NGC147 exhibits almost no metallicity gradient, ∼ −0.02 dex/kpc from 0.2 to 0.6 deg. The differences in the structure and stellar populations in the outskirts of these systems suggest that tidal influences have played an important role in governing the evolution of NGC147.
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
Table 1. Assumed properties of NGC147 and NGC185. Columns are (1): name of the galaxy; (2-3): equatorial coordinates from the NED database (http://ned.ipac.caltech.edu/ ); (4-5) Galactic longitude and latitude (NED); (6) absolute V magnitude, obtained using the apparent, de-reddened magnitude listed in NED (originally from de Vaucouleurs et al. 1991) and the distance modulus derived by Conn et al. (2012); (7-9): apparent TRGB i-band magnitude and deprojected distances from the Milky Way and from M31 (Conn et al. 2012, uncertainties stem from the resulting Posterior Probability Distributions); (10-11): mean extinction values (SDSS bands) from NED; (12-13): mean position angle (measured from north through east) and ellipticity as derived in this study (see Sect. 4).
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
larly sensitive to processes that affect gas content, and thus star for- mation, and both internal feedback and environmental effects are likely to play major roles in governing their evolution. For exam- ple, the origin of dwarf elliptical (dE) galaxies is a longstanding question: are they simply scaled-down versions of more massive early-type galaxies, which experience the main episode of star for- mation at early epochs, or do they result from the morphological transformation of gas-rich dwarf irregular (dIrr) galaxies which are depleted of their gas content through environmental processes (e.g., ram-pressure stripping, tidal interactions)? The role of internal ver- sus environmental factors has been discussed extensively in the lit- erature (see, e.g., Mateo 1998; Grebel et al. 2003) but there is as of yet no consensus. Investigating the structural profiles and stel- lar populations gradients in these systems has the potential to help distinguish between possible origins.
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
NGC147 probed stellar kinematics to large radii and discov- ered significant rotation in both galaxies (Geha et al. 2010), con- trary to what had been previously found by smaller FoV studies (Bender et al. 1991; Simien & Prugniel 2002). This provides im- portant insight into the formation mechanism of this class of galax- ies, suggesting a rotationally-supported origin possibly followed by a morphological transformation.
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
In this work, we present a photometric analysis of the stel- lar content of NGC147 and NGC185 derived from very wide- field survey data. Many galaxies have recently been shown to have large low surface brightness extensions (e.g. Barker et al. 2012; Bernard et al. 2012; Crnojevi´c et al. 2013; Ibata et al. 2014), the properties of the which hold important clues to their past histories, and it is natural to ask whether these two systems do too. We exploit data from the Pan-Andromeda Archaeological Survey (PAndAS) that imaged the M31 halo out to projected distances of ∼ 150 kpc (McConnachie et al. 2009) and which provides deep uniform pho- tometry out to unprecedented radii for both dwarfs. This dataset has already led to the discovery of a prominent stellar stream em- anating from NGC147, which is reported in detail elsewhere (e.g., Ibata et al. 2014; Martin et al. 2013, Irwin et al., in prep.), as well as several new globular clusters around both systems (Veljanoski et al. 2013). In this paper, we analyse the resolved stellar populations across large swaths of the two dEs and re-examine their extents, structural properties and chemical contents.
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
The paper is organized as follows: in §2 we present the data and the photometry, and show the resulting colour-magnitude dia- grams (CMDs) in §3. We analyse the spatial distribution of popu- lations in §4, their radial density and surface brightness profiles in §5, while in §6 we derive photometric metallicity distribution func- tions (MDFs) and investigate the metallicity gradients. We discuss our results in §7, and draw our conclusions in §8.
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
The the was MegaPrime/MegaCam camera on the 3.6 m Canada-France- Hawaii Telescope (CFHT) under excellent average seeing conditions (< 0.8′′). MegaCam consists of a mosaic of 36 2048 × 4612 CCDs with a total FoV of ∼ 1 × 1 deg2 and pixel scale 0.187′′ per pixel. The effective area covered by each pointing is about 0.96 × 0.94 deg2 due to small gaps between the chips. In total, roughly 400 deg2 was mapped around M31 and M33 via 413 distinct MegaCam pointings. The observing strategy, data processing and calibration have been fully described else- where (McConnachie et al. 2009, 2010; Richardson et al. 2011; Ibata et al. 2014).
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
Figure 1. De-reddened CMDs of the dEs, showing only stars falling within an ellipse of semi-major axis 0.3 deg. Overlaid are Dartmouth isochrones (green lines) with a fixed age of 12 Gyr and varying metallicity ([Fe/H]= −2.5, −1.3, −0.7 and −0.3), shifted to the appropriate distance. The adopted RGB boxes are outlined in red, while blue boxes indicate areas dominated by foreground sequences (Galactic halo: 0.1 < (g − i)0 < 0.6, 19 < i0 < 21; Galactic disc: 1.2 < (g − i)0 < 2.7, 18 < i0 < 19). The dash-dotted red line isolates candidate young stars. We do not consider luminous AGB stars in this work as their study is deferred to a future paper incorporating a NIR dataset.
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
As a brief summary, the raw data are pre-processed through the Elixir CFHT pipeline, and the source detection and aperture photometry are obtained with the CFHT/MegaPrime adapted ver- sion of the Cambridge Astronomical Survey Unit (CASU) pipeline (Irwin & Lewis 2001). PSF-fitting photometry is subsequently per- formed using DAOPHOT/ALLSTAR (Stetson 1987). Although PSF-fitting photometry leads to smaller photometric uncertainties (see Ibata et al. 2014), the quality of fit parameters vary greatly be- tween CCDs and thus cannot be used to clean our catalogue of non-stellar sources. Instead, the PSF photometry and the CASU aperture photometry are cross-matched and the source classifica- tions from the latter are adopted. For the bulk of our analysis, only objects classified as stellar or probably stellar (i. e. within 2σ of the stellar locus; Irwin & Lewis 2001) in both bands are used. The resulting g and i magnitudes are on the AB system, and the final band-merged catalogue provides both morphological classification and accurate astrometry for each source. Sources in our catalogue have a signal-to-noise ratio of ∼ 5 (corresponding to a photometric error of 0.2 mag) at i0 ∼ 24 − 24.5 and g0 ∼ 25 − 25.5.
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
Throughout this study we use de-reddened magnitudes, adopt- ing the extinction values obtained from the Schlegel et al. (1998) E(B − V) maps interpolated on a star-by-star basis. Additionally, we apply the Bonifacio et al. (2000) correction for E(B − V) > 0.1 (their formula 1), which returns slightly smaller values than the original ones. The mean extinction for a ∼ 3 × 3 deg2 field centered on NGC185 and NGC147 is < E(B − V) >∼ 0.13 and presents spa- tial variations of up to ∼ 40%. We apply the following conversion to obtain the final de-reddened magnitudes: g0 = g−3.793 E(B−V) and i0 = i − 2.086 E(B − V) (Schlegel et al. 1998).
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
projected elliptical radii (r = px2 + y2/(1 − ǫ)2) computed using our mean values of position angle (PA) and ǫ.
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
In Fig. 1 we show de-reddened colour-magnitude diagrams (CMDs) for NGC185 and NGC147. We plot all stars in each system lying within an ellipse of semi-major axis 0.3 deg to highlight the main CMD sequences. The most prominent feature in the CMDs is a broad red giant branch (RGB), indicative of old (∼ 2 to 13 Gyr) populations that have evolved off the main sequence (MS). Dart- mouth isochrones in the CFHT photometric system (Dotter et al. 2008; McConnachie et al. 2010) have been overlaid on the RGB. The isochrones have been shifted to the distance of each galaxy and have an age of 12 Gyr, [α/Fe] = +0.0 and varying metallicity to en- compass the full width of the RGB. For our subsequent analysis, we define a selection box on the main RGB locus with the bright limit set by the TRGB luminosity and a faint limit of i0 = 23.5. The faint limit has been chosen to limit incompleteness in our data and minimise contamination from unresolved galaxies which become significant at magnitudes fainter than this (see Ibata et al. (2014) and Fig. 3).
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
Figure 2. Spatial distribution of sources falling within the NGC185 RGB box (in standard coordinates centered on NGC185). North is up and east is left. The red boxes indicate the position of the two adopted field regions (each covering a ∼ 1.9 deg2 area) used to monitor contaminants. Large red ellipses are drawn with radii of 0.5 deg around NGC185 (left) and 0.8 deg around NGC147 (right). Also shown are the dwarf spheroidal galaxies Cass II (ξ = −0.39◦, η = 1.31◦, And XXV (ξ = −1.51◦, η = −1.46◦) and And XXVI (ξ = −2.55◦, η = −0.36◦), where the ellipse in each case indicates 4reff . The arrow points in the direction of M31.
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
ture paper which will also incorporate near-infrared (NIR) data, en- abling a cleaner and more robust separation from foreground stars. Finally, a few objects are visible at colours bluer than (g−i)0 . 0 (vertical dashed line in Fig. 1), suggesting the presence of young, massive MS stars. While NGC185 has been previously shown to have signs of recent star formation in its main body (Lee et al. 1993; Mart´ınez-Delgado et al. 1999; Butler & Mart´ınez-Delgado 2005; Marleau et al. 2010), NGC147 is not known to contain young stars. Some foreground stars could contaminate this CMD region, however. We define a selection box for blue stars at −1.0 < (g−i)0 < 0.0, 17 < i0 < 23.5 and consider in detail their spatial distribution in the following analysis, though we note that crowding in the central regions – where these sources typically lie – limits straightforward interpretation.
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
While our photometric catalogue will be dominated by stars in NGC147 and NGC185, there will also be contaminant sources. Ac- counting for these is crucially important when analysing the low surface density peripheral regions of the two dwarfs. The are three types of contaminating sources to consider – Galactic foreground stars, M31 halo stars, and unresolved background galaxies. While M31 halo stars and background galaxies are expected to be rela- tively smoothly distributed over our FoV (we are considering large galactocentric distances from M31 where the halo density falls off slowly, Ibata et al. 2014), the contribution from Galactic disc stars significantly increases in the north-west direction due to the low Galactic latitude (b ∼ −14◦) of the systems.
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
with gradient in foreground density due to the Galactic disc. Our field regions are chosen to avoid the north-south direction because of the presence of NGC147’s extended tidal tails and the dwarf spheroidal galaxies Cass II (Irwin et al., in prep.) and And XXV (Richardson et al. 2011). A third low-mass galaxy, And XXVI (the least luminous of the three), lies within our west field region but we have excised stars falling within a 4 reff ellipse centered on this dwarf (there are virtually no stars belonging to And XXVI outside this radius).
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
The CMD of the combined field regions is shown in the left panel of Fig. 3, scaled to an arbitrary area of 0.25 deg2 to ease the identification of the main features. This is comparable to the sur- veyed area within r = 0.3 deg (after gap subtraction) for NGC185 (∼ 0.20 deg2) and NGC147 (∼ 0.27 deg2) and hence should rep- resent the level of contamination present in the CMDs of Fig. 1. Besides the prominent foreground sequences, sources found at magnitudes greater than i0 ∼ 23.5 (i.e. below our RGB box faint limit) are largely unresolved background galaxies (see, e.g., Fig 9 of Barker et al. 2012). The RGB of M31’s stellar halo is barely vis- ible due to its low density at these galactocentric radii. Finally, a few sources bluer than (g − i)0 ∼ 0 are also present, representing an important source of contamination for the identification of possible young stars belonging to the dEs.
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
Figure 3. The CMDs of contaminant sources. Left panel: the CMD of point sources lying in the field selection boxes, scaled to an arbitrary area of 0.25 deg2. The inset shows the colour histogram of sources per unit area with (i0,T RGB NGC185 + 2.0) < i0 < (i0,T RGB NGC185 + 2.5) (this magnitude range is denoted with red lines on the y-axis). At fainter magnitudes, unresolved galaxies dominate the CMD. Right panel: simulated Galactic foreground from the Besanc¸on model, scaled to an area of 0.25 deg2. The simulation has been convolved with photometric errors (as a function of magnitude only) stemming from our photometry, but is not corrected for the data incompleteness.
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
our contaminant population in subsequent analyses. Finally, for the observed field CMD, we compute the colour histogram of sources that lie within the RGB selection box and have magnitudes in the range (i0,T RGB NGC185 + 2.0) < i0 < (i0,T RGB NGC185 + 2.5). The over- density of sources at colours (g − i)0 ∼ 1 results from M31 halo stars, while at redder colours the Galactic foreground dominates.
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
falling within the RGB selection boxes drawn in Fig. 1. The gaps between the CCDs in the MegaCam FoV and the saturated stars are clearly visible. Due to the higher central density of NGC185, the innermost ∼ 0.03 deg is mostly unresolved. Even in these point source density maps, one can see the significant larger extent of NGC147 compared to NGC185 and evidence of the tidal tails em- anating in the north-west/south-east directions at large radii. These low surface brightness tails extend well beyond the area analysed here and are the focus of a companion paper (Irwin et al., in prep).
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
To explore how the stellar populations vary across the face of each galaxy, we construct CMDs in four elliptical bins spanning a range of radii and assuming the average PAs and ellipticities derived in this study (see Sect. 4). For NGC185, we restrict our analysis here and throughout the rest of the paper to 6 0.5 deg to avoid contam- ination from NGC147’s tidal tails. The CMDs are shown in Fig. 4 and 5, with and without isochrones overlaid.
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
In both systems, a clear signature of an RGB population is seen at all radii demonstrating the significant extents of these dwarf galaxies. In addition, the width of the RGB clearly decreases as a function of increasing galactocentric radius, suggestive of a smaller dispersion in metallicity in the outermost regions. To demonstrate this, colour histograms for RGB stars (per unit area) in a narrow range of magnitudes ((i0,T RGB + 2.0) < i0 < (i0,T RGB + 2.5)), assum- ing the TRGB of each galaxy) are constructed and shown as insets in Fig. 4 and 5. For NGC185, the width of the histogram distribu- tion decreases from the first to the fourth radial bin, and the peak of the distribution also shifts bluewards. For NGC147, the width decreases from the first two to the second two radial bins, but the peak of the distribution does not significantly change. We note that the innermost radial bin for each galaxy is more incomplete at faint magnitudes, due to stellar crowding in these regions.
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
To create smoothed density maps, we artificially fill these small gaps and holes through populating them to match source counts in adjacent areas. From the gap-filled distribution, we com- pute the RGB stellar density per arcmin2 and smooth the result with a 2 × 2 arcmin2 grid to produce the contour plots shown in the mid- dle panels of Fig. 6. The density contours of NGC185 are regular over its full extent, while NGC147 presents clear twisting of the isophotes at large radius due to the tidal material.
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
Figure 4. CMDs in four radial bins (in 0.125 deg steps) for NGC185, assuming the average PA and ellipticity derived in this study. Upper panels: The radial limits and enclosed areas are reported in each subpanel. The insets show colour histograms of sources per unit area with (i0,T RGB NGC185 + 2.0) < i0 < (i0,T RGB NGC185 + 2.5) (this magnitude range is denoted with red lines on the y-axis). Lower panels: As above, but with Dartmouth isochrones of fixed age (12 Gyr) and varying metallicity ([Fe/H]= −2.5, −1.3 and −0.7) overlaid. Also shown is the adopted RGB selection box and the number of stars per bin.
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
Figure 5. Same as Fig. 4, for NGC147 and with radial bins of 0.2 deg width. The insets show colour histograms of sources per unit area with (i0,T RGB NGC147 + 2.0) < i0 < (i0,T RGB NGC147 + 2.5) (this magnitude range is denoted with red lines on the y-axis).
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
Figure 6. Top panels: The spatial distribution of RGB stars in standard coordinates for NGC185 (left) and NGC147 (right). White stripes indicate gaps between CCDs and holes stem from saturated stars. We draw ellipses at 0.125, 0.25, 0.375 and 0.5 deg for NGC185 and at 0.2, 0.4, 0.6 and 0.8 deg for NGC147, with average position angle and ellipticity derived in this study, corresponding to the radial bins shown in Fig. 4 and 5. Middle panels: RGB density maps. The survey gaps/holes have been filled with artificial stars, and the isodensity contours are logarithmically spaced and indicate densities of ∼ 4, 16 and 64 stars per 1 arcmin2. Position angle and ellipticity are recomputed for 0.1 deg radial annuli (blue dashed ellipses). Bottom panels: Position angle and ellipticity as a function of distance along the semi-major axis, calculated for annuli centred on each radial point. The errorbars are smaller than the symbols.
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
tion angle twists strongly, both a direct result of the emergence of the tidal tails.
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
Figure 7. Spatial distribution of blue sources, consistent with being candidate MS stars, in each of the target galaxies. Ellipses are as in the top panels of Fig. 6.
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
We also investigate the spatial distribution of sources with (g − i)0 < 0 which are candidate young MS populations. Fig. 7 shows a tentative overdensity of such sources in the central re- gion of NGC185, while no obvious enhancement is observed in NGC147. This is consistent with previous studies, as further dis- cussed in Sect. 5.2. Stars falling within the two foreground selec- tion boxes drawn in Fig. 1 are homogeneously distributed, as ex- pected for Milky Way populations. Sources classified as “extended” (and therefore rejected by our quality cuts) are also mostly uni- formly distributed but show enhancements in the central regions of both dEs. This suggests that some fraction of genuine RGB stars in these parts are misclassified as extended sources due to blending. We will return to this issue in the next section.
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
In this section, we derive radial surface brightness profiles for the dEs by combining diffuse light and resolved star counts. While it is challenging to trace diffuse light in faint galactic outskirts, this is the only technique applicable to the central regions, where high stellar crowding hampers the resolution of individual sources (see, e.g., Fig. 6). On the other hand, resolved star counts allow us to trace the light profiles out to the largest galactocentric distances probed by our dataset.
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
Geha et al. (2010) used the r-band photometry of Kent (1987) and the assumption of a constant B − V colour to derive V-band surface brightness profiles out to radii ∼ 7′ and 5′ in NGC185 and NGC147, respectively. They find slightly lower values of nNGC185 = 1.76 and nNGC147 = 1.04. Their Sersic fit is performed with fixed values of effective radius, adopted from De Rijcke et al. (2006). These values are obtained from NIR (2MASS) photometric data in the innermost ∼ 4′ of each dE, and are reff = 1.50′ for NGC185 and reff = 2.04′ for NGC147, thus at least a factor of two smaller than our best-fit values. The factor of two difference holds true when we extend the comparison to previous scale-length estimates resulting from fitting exponential functions or King profiles (Hodge 1963, 1976; Kent 1987; Caldwell et al. 1992; Lee et al. 1993; Kim & Lee 1998).
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
We exploit the PAndAS images to derive a g-band surface bright- ness profile from the diffuse light in the innermost 10′ × 10′ of each galaxy. After masking out bright stars, we compute the median of the pixel values in concentric ellipses (with our adopted values of average PA and ǫ). In order to derive the sky value, we average the mode of the pixel values in two ∼ 2.5′ × 2.5′ regions located at 15′ on opposite sides of each galaxy, and take the sky uncertainty to be the standard deviation of these values. We convert the resulting profile into mag per arcsec2, and de-redden by adopting the median E(B − V) value within the considered area.
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
To construct accurate star count profiles, we need to account for the contaminant level and its uncertainty. Fig. 9 shows the radial surface density profiles as a function of elliptical radius for RGB stars and blue sources in each galaxy, with the field contamination level and uncertainty indicated. These are calculated as the average and standard deviation of the number densities of objects falling within the RGB and blue selection boxes in our two adopted field regions (Fig. 2).
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
Figure 8. Surface brightness profiles (g-band) for NGC185 (left) and NGC147 (right), derived from diffuse light analyses alone. The best-fit Sersic profiles are overlaid, and the respective parameters reported.
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
Figure 9. Radial star count profiles for NGC185 (left) and NGC147 (right). Red circles are RGB stars, while blue diamonds are blue sources. The error bars are Poissonian. The red dashed line indicates the surface density of sources falling within the RGB box in the adopted field regions, i.e. the contaminant level (see Fig. 2), while red dotted lines indicate the dispersion around this value. The blue dot-dashed line is the contaminant level for sources falling within the blue star selection box.
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
sonable to expect that unresolved galaxies and M31 halo stars are smoothly distributed across our FoV, the Galactic foreground may vary substantially. To check this, we have inspected the maps of the Milky Way foreground populations within the PAndAS area pre- sented by Martin et al. (2013). Their Fig. 2 reveals that there are no obvious foreground overdensities in the vicinity of NGC185 and NGC147 at magnitudes that would affect our RGB selection box. As a further check, we also use the colour-magnitude contamina- tion model presented in Martin et al. (2013, their formula 12) in order to estimate the foreground number density in our RGB se- lection box, at the position of the target dEs. The values we obtain from the model are consistent with, but lower by ∼ 5 − 10% with respect to, our direct estimates. This may be due to the additional sparse M31 halo component, and/or to low-level foreground sub- structure.
|
Processed_A_PAndAS_view_of_M31_dwarf_elliptical_satellites:_.txt
|
NGC147’s tidal tails which start to contribute in these parts. One the other hand, the RGB profile of NGC147 declines smoothly and remains well above the field level out to 0.7 deg. Fig. 10 exam- ines how the contaminant-subtracted RGB star count profiles vary as a function of azimuthal angle. Each galaxy is split into eight 45◦-wide wedges, with the central angle of each wedge measured clockwise from the north-eastern semi-major axis. For NGC185, this analysis reveals no significant change in the radial profiles as a function of azimuthal angle. For NGC147, there is a clear excess of stars at large radius in the directions of 45◦ − 90◦ and 225◦ − 270◦, consistent with the presence of the tidal tails (see Fig. 6). It can be seen that the excess stars in these directions is what causes the azimuthally-averaged radial density profile to remain above the field level out to 0.7 deg. In directions away from the tidal tails, the RGB star count profile falls off more steeply at large radius, and reaches the background level at ∼ 0.5 deg. Note that the RGB den- sity profiles in all cases show a drop in number counts at small radii (. 0.05 deg), due to incompleteness in these parts.
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
Recurrent waves which are often observed during long pandemics typically form as a result of several interrelated dynamics including public health interventions, population mobility and behaviour, varying disease transmissibility due to pathogen mutations, and changes in host immunity due to recency of vaccination or previous infections. Complex nonlinear dependencies among these dynamics, including feedback between disease incidence and the opinion-driven adoption of social distancing behaviour, remain poorly understood, particularly in scenarios involving heterogeneous population, partial and waning immunity, and rapidly changing public opinions. This study addressed this challenge by proposing an opinion dynamics model that accounts for changes in social distancing behaviour (i.e., whether to adopt social distancing) by modelling both individual risk perception and peer pressure. The opinion dynamics model was integrated and validated within a large-scale agent-based COVID-19 pandemic simulation that modelled the spread of the Omicron variant of SARS-CoV-2 between December 2021 and June 2022 in Australia. Our study revealed that the fluctuating adoption of social distancing, shaped by individual risk aversion and social peer pressure from both household and workplace environments, may explain the observed pattern of recurrent waves of infections.
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
The presence of recurrent waves of infections [1, 2] has been reported in pandemics of several pathogens [3, 4, 5]. For influenza, multiple waves have been noted as a distinguishing feature of pandemics, and an opportunity for informing public health responses [6]. Accordingly, there is much interest in understanding contributing factors that can explain the observed pattern of waves [7, 8, 9, 10]. Seasonal effects, evolutionary dynamics of the pathogen, and interactions with other co-circulating pathogens are each important, however many of these explanatory models focus on the impact of human behaviour. The contact patterns of infected hosts with immunologically naive populations often determine early pandemic dynamics. As a pandemic progresses and populations observe the impacts of the disease and of control measures, it can be expected that host behaviours as a result of opinions will play an increasing role.
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
A pattern of recurrent waves of infections was observed during the COVID-19 pandemic, in Australia between December 2021 to June 2022 the Omicron variant was dominant. During this period, the population had a relatively high vaccine coverage (i.e., over 90% within adult population) [11, 12, 13]. This Omicron wave of infections resulted in a a prominent peak reaching 105,000 new cases per day around early January 2022 (over 4,100 cases per million), followed by two subsequent smaller peaks reaching 66,000 new cases per day around early April 2022 (over 2,500 cases per million) and 59,000 new cases per day around late May 2022 (approximately 2,300 cases per million) [14]. The persistence of the Omicron variant has been partially attributed to fluctuating adoption of non-pharmaceutical interventions (NPIs), particularly social distancing (SD), during this period.
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
Using our census-calibrated agent-based model (ABM), we previously identified a plausible sequence of step changes in the fraction of the population adopting social distancing (i.e., SD profile) which reproduced the observed pattern of recurrent waves over a period of approximately 29 weeks when Omicron variant was dominant [15]. The SD profile was determined retrospectively by tuning the step changes to minimise the difference between simulated and actual disease incidence. The resultant profile included an initial period of low SD adoption fraction (30% of the population), followed by a period of relatively high SD adoption (70% of the population) when the incidence approached the first peak, a rapid decline to a moderate level of SD adoption (by 50 − 60% of the population) during the second wave, and a drop to low SD-adoption (at 30%) around the third mid-year wave. We hypothesised that declining SD adoption could be attributed to pandemic fatigue [16]. Although this study revealed an important relationship between social distancing adoption and pandemic spread, the underlying opinion formation mechanism and contributing factors shaping social behaviour were not explored.
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
In this study, we aimed to develop a model of opinion formation and dynamics which can shape complex SD adoption behaviours without relying on post hoc fitting to observation. Previous studies have modelled opinion dynamics using two perspectives: (i) opinion formation based on personal beliefs (e.g., individual risk evaluation due to fear of infection [17, 18, 19], or perception fatigue [20]); and (ii) social influence (e.g., imitation [21, 22], averaging opinions [23], or majority-following opinions [24]). Our opinion formation model incorporates both of these aspects into our pandemic ABM. At each simulated cycle, the current opinion of each agent depends on their perceived level of risk (risk aversion) and on the opinions of the agents they interact with (peer pressure). An agent’s opinion determines its behaviours on SD adoption. We quantified the impact balancing risk aversion and peer pressure on opinion dynamics, and explored how a combination of the individual and social perspectives can reproduce the recurrent pandemic waves. In doing so, we identified the effect on opinion formation of varying memory horizon, perception fatigue, and social contexts.
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
The remainder of the paper is organised as follows. We describe the ABM, including the pandemic transmission and control, i.e., NPIs and vaccination rollout (Section 2.1), and the opinion dynamics model (Section 2.2). We then report the key factors contributing to opinion dynamics that generate recurrent pandemic waves (Section 3). Finally, we discuss the limitations and implications of our study (Section 4).
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
We simulate the viral transmission of the Omicron variant of COVID-19 in Australia, as well as pandemic control measures (NPIs and vaccination roll-out), using a high-resolution agent-based model which was previously calibrated and validated [25, 26, 15, 27, 28]. Our model comprises 25.4 million anonymous agents, each assigned several demographic attributes derived from the latest 2021 Australian Census [29], so that the artificial population represents the key demographic characteristics of Australia. These demographic attributes, including age, gender, residential area, student enrollment, and workforce/educational groups, also determine the social contexts in which agent-to-agent interactions and thus, disease transmission, take place [28, 30].
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
In our model, disease transmission follows a discrete-time simulation, updating each agent’s health state over time: Susceptible, Latent, Infectious (asymptomatic or symptomatic), and Recovered. The simulation begins with infections seeded around international airports (mostly in metropolitan areas) and continues to spread within the artificial population through interactions within multiple social contexts. These interactions are simulated in two phases per workday with different contact and transmission rates: “daytime” cycle where interactions take place in workplace or education contexts (e.g., class, grade, school), and “nighttime” cycle where interactions take place in residential context (e.g., household, household cluster, neighbourhood and community). On weekends, interactions are assumed to occur solely within residential contexts, consisting of two “nighttime” cycles instead [28].
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
The probability of an exposed agent becoming infected is adjusted by checking if the agent is (i) vaccinated prior to the start of the pandemic wave, and/or (ii) adopting or complying with non-pharmaceutical interventions (NPIs), such as social distancing, case isolation, and so on. Out of all infections, only a fraction is assumed to be detected, to match the voluntary self-reporting system adopted in Australia during the Omicron stage. Here, we approximate the case detection rate based on the prevalence of anti-nucleocapsid antibodies in Australia [31] (see Appendix for more details).
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
Following previous studies [27, 26, 15, 28], we assumed a high level of preemptive vaccination coverage in the population (90%) with two types of vaccines (priority vaccine, with a higher efficacy; and general vaccine, with a lower efficacy) distributed prior to the start of the simulation (i.e., the start of the Omicron stage in December 2021), in line with the reported vaccination coverage and vaccine distribution in Australia [11]. In addition, we introduced a waning immunity (from both vaccination and previous infections) into the model by tracking the vaccination and infection record of each agent, with the immunity declining over time (see Appendix). Consequently, in this extension of our model, agents can be re-infected after recovering.
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
NPI adoption reduces disease transmission by reducing the strength of interactions between agents across various social contexts. These NPIs include: case isolation (CI, affecting symptomatic infectious agents and detected asymptomatic agents), home quarantine (HQ, affecting household members of infected agents), social distancing (SD, affecting susceptible agents), and school closures (SC, affecting school-aged agents, their households and teachers). Given our focus on the interplay between opinion dynamics and recurrent pandemic waves, we model SD compliance as a voluntary, opinion-driven decision which may change as the pandemic progresses (while assuming static compliance with other NPIs). Here, SD is modelled as a generic intervention which reduces the overall intensity of agent interactions, intended to capture the combined effect of reduced travel, mask-wearing, and physical distancing (see Section 2.2).
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
We divide the population into three groups of agents: (i) committed, compliant agents which always comply with SD (25% of the population, randomly assigned); (ii) non-compliant agents which never comply with SD, due to being either a contrarian or an essential worker (25% of the population, randomly assigned); and (iii) rational agents which change their SD adoption based on the current pandemic severity (the remaining 50% of the population). A rational susceptible agent has a choice between two behaviours: (i) “live as usual" behaviour, and (ii) “socially distancing" behaviour. A rational agent may choose to switch between these two behaviours at any time as a result of the risk evaluation process following three steps: (i) self-evaluation where the opinion on SD adoption is formed individually by assessing the risk of infection (see Section 2.2.1); (ii) partial peer pressure where each agent may be partially influenced by opinions of the agents in their social contexts (see Section 2.2.2); and (iii) integration of the self-evaluation and peer pressure steps using a weighted average (see Fig. 1).
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
I(d, T ) averages the daily incidence, I(d), between day (d − T ) and day (d − 1). T acts as a memory horizon, limiting the effect of earlier incidence data on an agent’s behaviour. N c/N n is the relative size of a typical community, defined by the ratio between the typical community size N c = 103 and the national population size N n = 25.4 × 106, according to 2021 census [29]. Informally, this normalisation converts the global incidence to the local level at which agents perceive the disease severity. Parameter β quantifies the agent’s attitude towards risk aversion. A higher β leads to higher Es, so that agents become more likely to adopt social distancing.
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
u 1 + e−v(d−L) where T0 is the baseline memory horizon length at the beginning of the simulation, L, determines the simulation day when the memory horizon reaches its midpoint value, and u and v determine the shape of the transition region in the sigmoid function. We fixed these parameters to T (0) = 7, L = 60, u = 28, and v = 0.25, resulting in T (d) ranging from 7 to 35 days as illustrated in Fig. 2.
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
Figure 2: Sigmoid function for varying memory horizon, T (d), in Eq. 3. Shaded areas show the initial period with memory horizon of 7 days (yellow colour), the intermediate period of increasing value (grey), and the period with the final maximum value of 35 days (blue). Dashed lines mark the midpoint value of 21 days reached on simulation day d = 60.
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
where Es j(d) as before is the perceived risk of infection of agent j who belongs to the same context g as agent i, Ag\{i} is the set of all agents in context g excluding agent i, and ψg ∈ [0, 1] is the relative weight of the context g in all social contexts Gi that agent i belongs to, satisfying (cid:80) ψg = 1. In our study, we considered two representative types of opinion-affecting social contexts: households (HH) and working groups (WG).
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
where λ ∈ [0, 1] is the weight of peer pressure and 1 − λ is the weight of self-evaluation in forming the current opinion.
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
We present results in two parts, following the risk evaluation steps: (i) self-evaluation only (Section 3.1), and (ii) both self-evaluation and partial peer pressure (Section 3.2).
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
While simulating the opinion dynamics formed only by using self-evaluation, we examined three scenarios: SE1, the baseline model; SE2, where only the memory horizon varies with time, i.e. T = T (d) and γ = 0; and SE3, where both the memory horizon and the perceived probability of infection vary globally with time, i.e. T = T (d) and γ = 0.002.
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
For SE1, we tested different values of the constant β to explore its effects on opinion dynamics and pandemic spread (Fig. 3 and Appendix Fig. 5). We found that a moderate value of β = 0.5 produced some recurrent waves, however varying β alone could not generate a satisfactory match the observed magnitude of the first peak and the interval between waves.
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
Introducing the time-dependent memory horizon (SE2: Fig. 3), we observed improved accuracy in reproducing the second wave, but the difference with the first incidence peak could not be reconciled. We found that the combined effect of the time-dependent memory horizon and pandemic fatigue (SE3: Fig. 3) could reproduce the first peak and the timing of the second wave, but not the amplitude of the second incidence peak. These results suggested that the complex pandemic dynamics (e.g., recurrent waves with a prominent first wave and slower subsequent waves) observed during the Omicron stage in Australia could not be reproduced by our opinion dynamics model based solely on self-evaluation of risks.
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
Figure 3: A comparison of simulated incidence (a) and SD-adoption fraction (b), with the opinion dynamics produced by self-evaluation only (SE1, baseline in purple; SE2, with changing memory horizon in blue; and SE3, with changing horizon and fatigue in orange). Refer to Table 1 for parameterisation. Shaded areas around the solid line show standard deviation. Each simulated profile is averaged over 25 runs. The actual daily incidence between December 2021 to June 2022 is shown in scattered diamonds. Note that SE1 is equivalent to the purple profile in Fig. 5.
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
To examine the peer pressure component coupled with scenario SE3, we varied the peer pressure weight λ, while selecting different social contexts. Specifically, we modelled three scenarios where peer pressure originated from household (PP1, ψHH = 1), working group (PP2, ψWG = 1), and both household and working group with equal influence (PP3, ψHH = ψWG = 0.5). Parameters setting in these scenarios are summarised in Table 1.
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
Figure 4: A comparison of (a) simulated incidence and (b) SD-adoption fraction, with opinions produced by combining self-evaluation and peer pressure in different social contexts (PP1, household only, in blue; PP2, working groups only, in purple; and PP3, household and working groups with equal weight, in green). Parameters relating to self-evaluation are set to the same values as in SE3 (see Table 1). Shaded areas around the solid line show standard deviation. Each simulated profile is averaged over 25 runs. The actual daily incidence between December 2021 to June 2022 is shown in scattered diamonds.
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
incidence peaks around day 130 and day 170 (see Table 2 for the comparison between simulated and actual incidence dynamics).
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
These findings were in concordance with our previous study [15]: the fluctuating adoption of social distancing strongly contributed to the recurrent waves observed during the Omicron pandemic stage in Australia. In this study, we further demonstrated that the changes in the SD adoption can themselves be attributed to several opinion-forming factors: risk aversion, memory horizon, perception fatigue, peer pressure, as well as selection of the peer-pressure forming social context(s).
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
Recurrent pandemic waves are a systemic phenomenon observed during spread of many infectious diseases, including during the COVID-19 pandemic [36]. One of the factors contributing to this phenomenon is the fluctuating compliance with or adoption of non-pharmaceutical interventions, such as social distancing (SD) [15]. However, the quantitative understanding of the nonlinear dependencies and feedback between the pandemic spread and the opinion-driven compliance/adoption of social distancing behaviour remains elusive.
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
Table 2: Mean incidence peaks and the corresponding simulation day in the considered simulated scenarios.
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
To address these questions we modelled the dynamics of opinions which shape the decisions on SD adoption, using a large-scale pandemic ABM. Our integrated model captured complex feedback between opinion-driven SD adoption and pandemic spread, producing a fluctuating SD-adoption profile that generated recurrent pandemic waves. The identified SD-adoption profile is in a qualitative agreement with the actual mobility reduction during this period, observed across the workplace, retail, and transport [37]. Our prior study [15] retrospectively selected six SD-adoption fractions in a sequence of five-step changes, while minimising the difference between simulated and actual incidence curves. In contrast, the opinion dynamic model proposed and validated in this study produced a more intuitive and constructive explanation, using five underlying factors, attributing the varying SD-adoption to a combined effect of individual risk aversion and social peer pressure. We identified an opinion formation scenario that best matches the recurrent waves observed during the Omicron stage in Australia by fine-tuning these five factors: risk aversion β, fatigue γ, memory horizon T , peer pressure weight λ, and selected opinion-affecting social contexts ψg.
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
opinion leaders and community champions). One possible future direction is to model several agents assigned a higher weight of opinion affecting a large social context (e.g., community) and explore the impact of these agents on the coupled pandemic and opinion dynamics. We also acknowledge that there are other methods to model peer pressure, for example, conformity [24].
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
We quantitatively modelled a nonlinear interplay between opinion dynamics and pandemic spread in a heterogeneous population, revealing a balanced impact of individual risk perception and social peer pressure on the SD-adoption decisions and resultant recurrent pandemic waves. We believe that a better understanding of intricate underlying factors shaping opinion dynamics during pandemics may inform policy-makers in designing more effective crisis response approaches.
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
This work was supported by the Australian Research Council grant DP220101688 (MP, QDN, SLC) and the University of Sydney’s Digital Science Initiative (SLC, QDN, CS, RR, VS, TS, AM, MP). In addition, SLC is supported by the University of Sydney Infectious Diseases Institute Seed Grant 220182. AM is supported by NHMRC Investigator Grant. The simulations involved in this work used the high-performance computing cluster (Artemis) provided by the Sydney Informatics Hub at the University of Sydney.
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
We generated an artificial population representing the main demographic characteristics and commuting network of the Australian population, by using several high-resolution datasets, including the Australian Census of 2021 [38], international air traffic reports with incoming passenger flows in Australian airports [39], and educational registration records including the number of schools and students [40]. The population generation process assigned each agent with multiple demographic characteristics (e.g., age, gender, residency location) and social contexts in different settings: residential (e.g., household, household cluster, etc.); educational (e.g., classroom/school for agents aged 18 years or younger); and workplace (for agents aged over 18 years). While the residential contexts are determined by residential demographics, the workplace and educational contexts follow the commuting network (travel to work and travel to school respectively, depending on agents’ age groups) [41]. The detailed up-to-date description of the population generation process can be found in Supplementary Materials of a recent study [28] and the user guide of our open-source software [42].
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
Using the generated surrogate population, we stochastically simulate infection transmission based on interactions within the agents’ social contexts. If an agent in Susceptible state is exposed to the infection within one or more of its social contexts, it may get infected (dependent on infection probabilities), and then progress through several states: Latent, Infectious (asymptomatic or symptomatic), and Recovered. We assumed no population dynamics during the simulated period (i.e., no births and deaths).
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
We use the daily probability of interaction qj→i(g), potentially transmitting infection from agent j to agent i, which depends on the ages of agents j and i, and on the context g in which the interaction occurs. These probabilities qj→i(g) have been defined and calibrated in previous studies [25, 15].
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
where κ is a global transmission scalar used to calibrate the reproductive number R0, nj denotes the infection onset time for agent j, and the agent-specific function fj(n − nj) is the natural history of the disease, reflecting the infectivity of agent j as its infection progresses. At the time cycle n, if agent j is not infected, fj(n − nj) = 0. If agent j is infected, fj(n − nj) ≥ 0.
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
Table 3: Model parameterisation for Omicron variant adopted by AMTraC-19. The last two rows show the corresponding basic reproductive number (R0) and generation/serial interval (Tgen), calibrated previously [15, 28].
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
where Gi(n) denotes the set of all social contexts g that agent i interacts with during the time cycle n, Ag\{i} denotes the set of agents in g (excluding agent i).
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
where for adults (age > 18) σi = σa; and for children (age ≤ 18) σi = σc. We calibrated the associated parameters (summarised in Table 3) to match the characteristics of the Omicron variant following [28, 15]. In this study, we also introduced re-infections, so that agents may become susceptible again upon recovery. The transition between Recovered and Susceptible states takes place after a fixed post-recovery period (e.g., 60 days), which aligns with clinical observation of re-infection periods for Omicron infections [43, 44].
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
Non-pharmaceutical interventions (NPIs) may reduce the infection probability for susceptible agents. We characterise each NPI by (i) macro-distancing level, defined as the NPI-adopting fraction of all population, and (ii) micro-distancing level, defined as the adjusted interaction strengths between the NPI-adopting agents and other agents in the same social group (see Table 4 for NPI parameterisation).
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
At each time point, relevant NPIs incorporated in our model (CI, HQ, SD, and SC) affect the agents according to their state, over a certain duration Di. While CI and HQ are activated at the start of the simulation and last throughout the simulated period, SD and SC are triggered only if the national cumulative incidence exceeds a certain threshold. Assignment of NPI-adopting agents follows a Bernoulli process and an agent may adopt multiple NPIs. However, the interaction strengths are adjusted to the value of Fj(g) following a descending priority assignment order: CI, HQ, SD, and SC. In this study, we assumed a constant level of adoption or compliance with for CI, HQ, and SC, while the SD adoption level is formed by the opinion dynamics detailed in Section 2.1. Therefore, there are no macro-compliance or micro-duration parameters for SD.
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
where Fj(g) is the strength of the interactions between agent j and other agents in the social context g. For NPI-adopting agents j, the interaction strength is Fj(g) ̸= 1. For non-adopting agents j, the interaction strength remains unchanged, Fj(g) = 1. Note that Eq.10 reduces to Eq. 8 if Fj(g) = 1.
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
Table 4: The macro-distancing parameters (population fractions) and micro-distancing (interaction strengths) for the considered NPIs. The micro-duration of CI is limited by the disease progression in the affected agent i, Di.
|
Processed_Impact_of_opinion_dynamics_on_recurrent_pandemic_w.txt
|
Vaccination reduces the infection risk for susceptible agents. In this study, we modelled a high pre-emptive vaccination coverage (90% of the population, or 22.89 million agents), attained before the Omicron stage. The vaccine distribution follows an age-dependent proportion [27]: 8.33% of vaccinated agents are aged 18 years or younger (age ≤ 18), 83.34% are aged between 18 and 65 years (18 < age < 65), and 8.33% are aged 65 years or older (age ≥ 65). Two types of vaccines are rolled out in our model, each covering 11.443M agents (or 45% of the simulated population): the “priority” vaccine with a higher clinical efficacy (ηpri), and the “general” vaccine with a medium efficacy (ηgen). Table 5 shows the efficacy values aligned with clinical observations following the booster vaccination against the Omicron variant [45, 46].
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.