Split (1)

train · 132k rows

Filename stringlengths 22 64	Paragraph stringlengths 8 5.57k
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	where the sequence Mn = max{X1, . . . , Xn} and parameter θ ≤ 1 is the extremal index.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	where G(y) is the limiting extremal distribution of the suitably normalized maxima and takes the form of one of three types of distributions.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	A Type 1 (Gumbel) distribution has an upper exponentially decaying tail; Type 2 (Fréchet) corresponds to an upper polynomial decaying tail; and Type 3 (Weibull) corresponds to a bounded tail.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	where µ is the location parameter, σ is the scale parameter, and ξ is the shape parameter which determines the tail-behavior or type of distribution [11].
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	One key point that should be emphasized here is the relationship between the GEV for i.i.d. sequences and the GEV for weakly dependent, stationary sequences with extremal index θ < 1.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	Provided the data meet the assumptions of stationarity and weak dependence, the limiting distribution of (Mn) is known, and creating a numerical model for (Mn) is then reduced to applying an appropriate statistical fitting method that results in the most accurate numerical approximation of the parameters ξ, µ, and σ. Since traditional statistical fitting methods require a frequency plot of the data to estimate parameters, the block maxima method is typically employed to generate such a frequency plot in which the data is blocked into pieces of length m (long enough to guarantee convergence to the GEV), the maxima are calculated over each block, and this sequence of maxima is used to approximate the parameters of the corresponding GEV.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	Recent advancements in understanding and modeling consecutive climate extremes have centered around redefining the sequence (Xn) as an average over some window and inter- preting returns of extreme highs or lows of this average as indicative of several high or low values, respectively, occurring within the window. However, it was only recently that the extreme value theory and dynamics communities began investigating the numerical accuracy of these models [30] with the first expectations on the limiting distribution type established this year [7]. For ease of interpretation, we focus on results for extremes of the minimum over a window; however, we recognize the importance of the windowed aver- age in climate applications and remark that expectations on the limiting distribution for certain windowed averages have also been established [7]. We now restate for our context the dynamical lemma which will be central to our investigation on multiday precipitation extremes across eastern Australia.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	for some unknown function g(k, χ) depending on the window size k and the system χ.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	Introducing functionals, such as the windowed minimum or average, reduces the amount of data available for fitting by a factor of the window size. Hence, for increasing window sizes, even for window sizes as low as k = 2 depending on the available pool of data, the numerical error of the extremal parameter estimates significantly increases. To our knowl- edge, this manuscript is the first to use techniques that reduce numerical error when using functionals to estimate consecutive extreme weather events. The fundamental importance of the previous lemma is that it proves max-stability (e.g. stability of the shape parameter) for the maximal process (Bn). As a result, this information provides us with greater control over the numerical error of our model by forcing us to only choose models of (Bn) with shape parameters close to the true shape estimate from (Mn). In the sections to follow, we illustrate where this information can be used in practice to obtain more robust modeling results on the basis of model stability.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	Description of the data. Daily precipitation data for eastern Australia from 1984 to 2021 was obtained from the Australian Bureau of Meteorology. The study area includes New South Wales (NSW), Victoria (VIC), Queensland (QLD), and Tasmania (TAS). The dataset consists of 454 stations that have long-term records of precipitation observations with less than 1% missing values each year. The full list of stations is available in the Supplementary Material.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	To assess the impact of ENSO on the distribution of extreme events, we used the monthly Southern Oscillation Index (SOI) to provide information about the development and inten- sity of El-Niño or La-Niña events in the Pacific Ocean obtained from the NOAA Climate Prediction Center.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	in a short window. There is extensive literature on the interpretation and approximation of the extremal index. We refer the reader to [27] for a nice summary of the current methods used for numerical approximation of the extremal index.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	where Tj is the wait time associated to the exceedance Zj > u for j = 1, . . . N over some high (fixed) threshold, often taken as the 95% or 99% quantile of the data (Xi) for i = 1, . . . n.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	Immediately obvious is the spatial structure of the ex- tremal index from stations across eastern Australia (Fig- ure 1). We note that the extremal index is smaller (e.g. cluster-size is larger) for stations in northern Queensland, and along the coast. This structure alone tells the story of large storms with multi-day rainfall moving in from northern Queensland and the coast, breaking up as they traverse south and inland. In 2022, the flooding across southern Queensland and northern New South Wales was attributed to the "2-3 consecutive days of heavy rainfall coming from a large, slow-moving storm system" [1]. To illustrate the importance of this index, note that the ex- tremal index for Brisbane is estimated at θ ≈ 0.45 so that 1/θ ≈ 2.25 giving an expectation of about 2.25 days of extreme rainfall which corresponds well to what was observed during the 2022 flooding event [1].
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	Figure 1. Ferro-Segers ex- tremal index estimate for daily rainfall from stations across eastern Australia.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	occurrences of extremes. One way of addressing this issue is to introduce functionals that preserve the consecutive property of extremes.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	We focus our study on applications of the Generalized Extreme Value distribution; how- ever, there is one other common approach for modeling extremes of a sequence. Before introducing our functional, we briefly remark on the appropriateness of this method and where interpretations for consecutive extremes fail.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	A note on the appropriateness of the Generalized Pareto. Another common approach in extremal literature is to apply a peaks over threshold method which involves setting a high threshold, creating a frequency distribution from over-threshold values in sequence, and fitting the Generalize Pareto distribution (GPD). However, the extremal index must be equal to 1 (e.g. θ = 1) to guarantee a GPD models these over-threshold values. A common solution is to remove all clustered over-threshold values in the sequence, keeping only a single value to model the GPD, then factoring the extremal index back into the resulting return-level estimate by multiplying the corresponding return-time by 1/θ. The problem with this approach is that it removes all interpretation of consecutive extreme events, in favor of some average return-time.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	on our daily time-series of precipitation (Xn) for j = 1, . . . n. Provided our sequence (Xn) is stationary, and blocks are taken long enough so that observations are roughly independent and converge to the GEV, we may apply the block maxima method to estimate the maximal process Bi,m from blocks of length m given by, Bi,m = max{Yim, . . . Yi(m+1)}. For this investigation, we find block lengths m of one-year long enough to guarantee convergence to the GEV.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	We refer the reader to a nice, detailed investigation on modeling precipitation extremes across eastern Australian in the single-daily case, using the Generalized Extreme Value distribution [31]. Scatter plot for covariates against block maxima used to analyze these relationships can be found in Figure 6.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	Remark 2.1. Coefficients for the location and scale estimate decrease for increasing window size k. This is in line with the more robust theoretical result from [7, Theorem 5.1] by noting that, rainfall is maximized at the center of a storm system (invariant set), hence g(k, χ)θ2/θ1, decreases for increasing k. Estimating this relationship is beyond the scope of this investigation which is centered around applications of the proven max-stability property. However, a preliminary investigation using rainfall from Germany (in the ap- proximately stationary setting) has been performed with encouraging results that support the idea of future numerical investigations.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	Yearly non-stationarity was also examined as a potential covariate; however, we find that the inclusion of this effect does not result in a statistically significantly improved model over ENSO alone. We refer to Table 1 for model estimates, noting that we observe a Fréchet distribution (ξ > 0) for precipitation extremes across eastern Australia.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	Table 1. Chosen nonstationary model coefficient estimates. There are three models, each corresponding to consecutive k daily extreme rainfall. Standard errors can be found in Table 3.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	fit the distribution. To our knowledge, this hidden numerical error, coming from reduced convergence rates of (Bi,m) to the GEV which is not reflected in the confidence intervals of likelihood methods, has not been considered in previous literature when forming distri- butions for functionals of a sequence. We illustrate this phenomenon by fitting the shape of the GEV for increasing window size over increasing block sizes in Figure 2.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	Remark 2.2. Although we use a fixed block length for ease of interpretation and com- parisons across different window sizes, a straight-forward solution to the issue of reduced convergence rates of (Bi,m) is to increase block length m. However, given the novelty of these results, there is currently no literature addressing optimal increases of block length.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	Applying the max-stability property pro- vided by Lemma 1.2 implies that the shape parameter for the maximal process (Bi,m) should not change with increasing window size. We use this property as a benchmark for our extremal distribution fits. We find that we are able to control the shape pa- rameter (to within 52.26% relative error) up to window size k = 3, before lack of tail sam- pling results in a significant jump in numer- ical error of the shape parameter (greater than 100% relative error).
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	Figure 2. Shape parameter estimate of the GEV for Bi,m for increasing window size. Longer block lengths are able to capture the true shape parameter for greater window sizes.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	If ˆF is a reasonable estimate of F or the model is a good fit, the plot should be roughly a straight line of unit slope through the origin.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	Provided the chosen covariates appropriately explain the non-stationarity in the parameters of the GEV corresponding to Zt, the standardized variable ˜Zt is expected to follow a standard Gumbel distribution. A conventional quantile plot for block maxima can be generated using the standardized variable ˜Zt compared to a standard Gumbel distribution. Anderson-Darling (A-D) and Mann-Kendall (M-K) tests are used to assess how well the data fit the distribution, and how well our covariates explain the non-stationarity within the data. Model diagnostics, including M-K and A-D test results can be found in Table 4 and quantile plots to assess the fit against a stationary model can be found in Figure 7.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	Assessing model complexities and improvements for multiday extremes. We test each non-stationary model against its stationary counterpart to determine whether increased model complexity statistically significantly improves model performance. We refer the reader to [23] which provides a nice summary on appropriate model selection In line with the results of this investigation, criteria under nonstationary assumptions. we use the AIC to select the appropriate model out of our series of nested models; how- ever, we remark that likelihood ratio and BIC were also in agreement with this choice. Nested nonstationary models were used to assess the statistical significance against model complexities. These models are described in the appendix A. In agreement with previous literature on single-day precipitation extremes across eastern Australia [31], we find a non- stationary model with latitude, longitude, distance from the coast, and SOI as the most representative of the non-stationarity observed in multiday precipitation extremes.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	wi = 1/m. In this setting, ¯xp is interpreted as the return level one expects to exceed in m years with probability equal to 1.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	Multiday precipitation extreme model predictions for historical and future flooding events. We first examine the spatial and temporal accuracy of our model to represent multiday extreme rainfall for years of historical relevance. Then, we report on the return level expectations under varying ENSO conditions.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	Benchmarked return levels to empirical rainfall data from historical flooding events. During the summer of 2010-2011, extensive parts of Queensland, New South Wales and Victoria experienced one of the worst flooding events on record, according to economic and loss of life measures. Following in 2022, Queensland and New South Wales experienced 2-3 consecutive days of extreme rainfall, causing extensive property damage, delaying harvests and reducing grain quality in the growing regions of New South Wales and Victoria. A natural question is whether we can use our non-stationary models of consecutive extremes to inform predictions on expected returns of consecutive days of extreme precipitation.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	We adopt the approach in section 2 which allows us to answer the question: What level of extreme rainfall do we expect to observe for k = 1, 2, or 3 consecutive days 10 years out (in 2022) from the last flooding event (in 2011)? Return level estimates and 95% confidence intervals are shown in Figure 3 against empirical quantiles. In noting that these empirical quantile estimates are more accurate with increasing data, we also provide the 20 year return level for 2022 (e.g. predictions out from 2001). Our models agree well with the empirical quantiles of multiday extremes in 2022.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	Model behaviour for historical El Niño and La Niña conditions. We use these methods to illustrate differences in the probability of observing multiday extreme precipitation under historical El Niño and La Niña conditions. We note that our model results suggest the contribution of SOI on the variance for k = 3 consecutive days is double that of k = 1 or 2 consecutive days of extreme precipitation, against an overall decreasing trend for variance against k, which agrees well with the conclusions of [35], suggesting the SOI is most influential for longer duration storms. Our model also suggests that changes to SOI have greater effects on inland compared to coastal regions. Table 2 shows the k = 1, 2 and 3 estimates of consecutive extreme rainfall using El Niño and La Niña conditions from 2015 and 2010, respectively. As an example, Karange Model Farm reports a 4.3% increase in probability of observing 3 consecutive days of extreme rainfall above the regionally high threshold compared to approximately 1-1.5% increase for coastal regions. The percent increase in observing multiday extreme rainfall from El Niño to La Niña years increases across duration (the number of consecutive days).
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	Figure 3. 2022 return levels, 10-years out from 2011 and 20-years out from 2001, for mutliday extreme rainfall across stations in eastern Australia for k = 1, k = 2, and k = 3 consecutive days (from top).
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	Predicted return levels under varying ENSO conditions. For the readers’ interest, we sim- ulate varying ENSO conditions for 20, 40, and 100 years by random sampling from the distribution of averaged yearly historical SOI taken from the NOAA (Figure 4). We em- phasize that these simulations are meant to provide ideas for tools and techniques for cli- mate scientists and should be interpreted in that context. Return levels for multiday ex- treme rainfall from these simulations for all locations along the eastern coast of Australia can be found in Figure 5.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	Figure 4. Example simulated SOI from randomly sampling from the distribution of averaged yearly historical SOI.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	A note on the prediction horizon. The prediction accuracy of our non-stationary model for returns and magnitudes of multiday precipitation extremes across eastern Australia is dependent on the accuracy of predictions of the proceeding years’ ENSO (SOI or similar) index.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	Table 2. Probabilities of multiday extreme rainfall exceedances for select stations and 2010, 2015 ENSO conditions.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	Conclusions and further applications. We have summarized current, popular methods used to model extreme rainfall over varying durations and the relationship of these methods to the theoretical advancements made in the dynamical framework. We have shown where classical extreme value modeling fails to appropriately interpret returns of consecutive extreme events (e.g. through averaging over the extremal index). We have introduced the minimum functional, taken from dynamical systems, as a way to preserve the consecutive property of multiday extremes in rainfall. We illustrate how advancements in dynamics can be translated to improve hidden numerical error (e.g. not detectable by current statistical methods) on multiday extreme rainfall models.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	We have shown that in the case of extreme rainfall following a Fréchet distribution (ξ > 0), the max-stability results from [7, Lemma 4.1] can be used to improve model estimates for returns of multiday extreme rainfall by limiting investigations to window sizes which guar- antee a stabilized shape parameter. This is the first time max-stability properties have been used to appropriately model consecutive extreme rainfall in a non-stationary setting; however, we note that these results are complementary to those observed empirically in previous literature on extremes of rainfall taken over higher resolutions, but shorter du- rations where only small fluctuations in shape were observed [35]. We have shown that our models agree with the empirical historical estimates of rainfall and the conclusions of these models agree with current literature, in particular on the empirically observed effects of SOI on extreme rainfall duration. Although we have used eastern Australia as a case study, we note the general use of these methods for modeling multiday rainfall in many locations where the maxima of daily rainfall follow a Fréchet distribution.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	in recent literature. For example, the authors in [35] reported a positive shape parameter with some variation depending on storm duration (6 minutes to 72 hours) across Australia. The authors in [16] report that their empirically estimated shape parameters range from slightly negative to positive values for 1-3 days durations at sites in New South Wales. On the other hand, the authors in [36] obtained negative shape parameters for 1- and 2-day durations at a specific site in Victoria and [21] reports a negative shape parameter for 1-day durations is common across meteorological stations in Australia. It is not clear whether the negative shape parameters observed for these regions is due to a true upper bounds on the quantity of rainfall for the chosen durations or whether it is the result of lack of available data. We illustrate that the data for modelling extreme multiday rainfall follow the general assumptions of [7, Lemma 4.1], so that there is no theoretical expectation for a varying shape parameter over storm duration and that the observation of such a change may be the result of increased numerical error in limiting the available pool of data used to fit the GEV.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	The authors in [7] have also investigated max-stability for averaging functionals (e.g. ex- tremes of the average over some window of length k). In future investigations, we would like to develop techniques for modeling heat-waves using such results. Due to the novelty of these results, convergence rates for appropriate block-size choice have not yet been inves- tigated. We intend on performing numerical investigations to estimate these rates which can then be used to choose appropriate block-sizes against window size k (duration of consecutive extreme events) and perhaps incorporated into more robust statistically fitting methods for consecutive extreme events.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	Figure 5. From the left to right, 20-, 40-, and 100-year return levels for (top to bottom) k = 1, 2, and 3 consecutive days of extreme rainfall.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	Figure 6. Scatter plots of the block maxima against covariates assess for the nonstationary models.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	Table 4. Model diagnostic and model selection: log-likelihood from max- imum likelihood estimation, AIC, BIC, Likelihood ratio test, and goodness of fit tests for nested nonstationary models.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	Funding. The authors would like to acknowledge the University of Queensland for facili- ties and funding under the research start-up scheme.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	Authors’ contributions. M.C. wrote the main manuscript text, formatted all tables and figures, prepared figures 3, 4, and 5, and completed the formal analysis. R.K. wrote the main manuscript text, prepared figures 1, 2, 6 and 7, prepared tables 1, 3, and 4, and completed the formal analysis. Both authors worked together on sections of the code (R and MATLAB) for the numerical results and reviewed the manuscript.
Processed_Modeling_Multiday_Extreme_Precipitation_Across_Eas.txt	Acknowledgements. M Carney would like to thank the Max Planck Institute for the Physics of Complex Systems, Dresden, Germany, where part of this work was completed.
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	ABSTRACT Ultra-diﬀuse galaxies (UDGs) are attractive candidates to probe cosmological models and test theories of galaxy formation at low masses; however, they are diﬃcult to detect because of their low surface brightness. In the Local Group a handful of UDGs have been found to date, most of which are satellites of the Milky Way and M31, and only two are isolated galaxies. It is unclear whether so few UDGs are expected. We address this by studying the population of UDGs formed in hydrodynamic constrained simulations of the Local Group from the HESTIA suite. For a Local Group with a total enclosed mass 𝑀LG(< 2.5 Mpc) = 8 × 1012 M(cid:12), we predict that there are 12±3 isolated UDGs (68% conﬁdence) with stellar masses 106 ≤ 𝑀∗ / M(cid:12) < 109, and eﬀective radii 𝑅e ≥ 1.5 kpc, within 2.5 Mpc of the Local Group, of which 2+2 −1 (68% conﬁdence) are detectable in the footprint of the Sloan Digital Sky Survey (SDSS). Accounting for survey incompleteness, we ﬁnd that almost the entire population of UDGs in the Local Group ﬁeld would be observable in a future all-sky survey with a depth similar to the SDSS, the Dark Energy Survey, or the Legacy Survey of Space and Time. Our results suggest that there is a population of UDGs in the Local Group awaiting discovery.
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	et al. 2015; Martínez-Delgado et al. 2016; Román & Trujillo 2017); in galaxy groups (Trujillo et al. 2017); and in the ﬁeld in between (e.g. Leisman et al. 2017). These extended objects have stellar masses and magnitudes typical of bright dwarf galaxies (𝑀∗=106−9 M(cid:12) and 𝑀V < −8, respectively); how- ever, they are signiﬁcantly larger, with sizes approaching those of massive galaxies such as the Milky Way. As a result they have very low surface brightness, usually between 𝜇e=24 and 28 mag arcsec−2, earning them the sobriquet “ultra-diﬀuse galaxies” (UDGs).
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	The circumstances leading to the emergence of such diﬀuse galaxies are not understood fully and several scenarios have been proposed to explain their formation. These are divided broadly into two main categories: i) internal processes that drive stars toward the outer regions of the galaxy, as could happen in haloes with high spin (Amorisco & Loeb 2016), and during episodes of powerful stellar feedback (Di Cintio et al. 2017; Chan et al. 2018; Cardona-Barrero et al. 2023); and ii) the disturbance caused by external mechanisms such as stripping and tidal heating (Carleton et al. 2019; Jiang et al. 2019; Tremmel et al. 2020; Benavides et al. 2021), and galaxy mergers (Wright et al. 2021). A compelling test of these proposals requires a large sample of UDGs, the catalog of which has grown rapidly in recent years because of advances in instrumentation and observational techniques. However, UDGs remain challenging to detect so their census in the nearby universe is likely far from complete.
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	Similarly, the census of dwarf galaxies within the Local Group is also incomplete (Garrison-Kimmel et al. 2014; New- ton et al. 2018; Nadler et al. 2019; Drlica-Wagner et al. 2020; Fattahi et al. 2020b). Using the Di Cintio et al. (2017) deﬁ- nition of UDGs, only eight Local Group galaxies satisfy the criteria: And II, And XIX, And XXXII, Antlia II, Crater II, Sagittarius dSph, WLM, and IC1613 (Collins et al. 2013; Kirby et al. 2014; Torrealba et al. 2016; Caldwell et al. 2017, see also McConnachie 2012 for observational data). Of these, six are satellites of the Milky Way and M31 and only IC1613 and WLM are found “in the ﬁeld” of the Local Group, i.e. they are outside the virial radii of the Milky Way and M31, which we take to be 230 and 275 kpc, respectively. It is un- clear whether the dearth of UDGs in the ﬁeld of the Local Group arises primarily from environmental inﬂuences that prevent most galaxies from becoming UDGs, or if observa- tional limitations are the main obstacle impeding their detec- tion. Indeed, if such a UDG population exists it would be partly obscured by the foreground of Milky Way stars and the background of other galaxies, making it diﬃcult to detect with current instruments. Therefore, in this Letter we use high-resolution simulations to quantify the number of UDGs that we expect to ﬁnd in the ﬁeld within 2.5 Mpc of the Local Group, and study their potential detectability in current and forthcoming surveys.
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	galaxies in this environment. The HESTIA suite does this (Libeskind et al. 2020), and consists of 13 zoom-in simula- tions of Local Group analogs that were run with the arepo moving mesh code (Springel 2010) and the auriga galaxy formation model (Grand et al. 2017). Using estimates of the peculiar velocity ﬁeld derived from observations (Tully et al. 2013), the initial conditions are constrained to repro- duce the major gravitational sources in the neighborhood of the Local Group. Consequently, at 𝑧=0 the Local Group analogs are embedded in large-scale structure that is consis- tent with the observations when assuming the Λ+cold dark matter (ΛCDM) cosmological model (see e.g. Hoﬀman & Ribak 1991; Doumler et al. 2013; Sorce et al. 2016).
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	The Local Group analogs are simulated at “low” and “in- termediate” resolution in a Planck Collaboration et al. (2014) cosmology. Three were resimulated at higher resolution us- ing ∼200M DM particles in a high-resolution region con- sisting of two overlapping spherical volumes with radii of 2.5 ℎ−1 Mpc, each centered on the Milky Way and M31 analogs at 𝑧=0. The spatial resolution achieved is 177 pc, and the eﬀective masses of the DM and gas particles are 𝑀DM = 2 × 105 M(cid:12) and 𝑀gas = 2.2 × 104 M(cid:12), respec- tively. The simulations are labeled 09_18, 17_11, and 37_11, after the initial seed used, and their physical properties can be found in Libeskind et al. (2020, table 1). We use the Amiga Halo Finder (ahf) algorithm (Gill et al. 2004; Knollmann & Knebe 2009) to identify and characterize gravitationally bound structures in the simulations.
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	In the auriga model the star particles represent simple stel- lar populations of a given age, mass, and metallicity. Upon creation they are initialized using a Chabrier (2003) initial mass function, and the mass that is subsequently lost due to stellar evolution is calculated using the yield curves from Portinari et al. (1998) and Karakas (2010). The photometric properties of each star particle are computed using the Bruzual & Charlot (2003) stellar population synthesis model while neglecting the eﬀects of dust attenuation. Consequently, the stellar populations may be as much as 0.75 mag brighter in the V– and r–bands that are of interest in this Letter than they would be if dust attenuation was accounted for. However, we note that the isolated, low-mass galaxies in HESTIA experi- ence minimal star formation activity at 𝑧=0, so we expect that including dust attenuation would produce only a small eﬀect.
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	Figure 1. The two-dimensional eﬀective radius, 𝑅e, as a function of the r–band eﬀective surface brightness, 𝜇e, of the ﬁeld UDGs in the three high-resolution HESTIA simulations. The size of each marker is inversely proportional to the distance to the nearest host galaxy. The dashed lines show two of our selection criteria applied to the ﬁeld haloes in the simulations. The galaxies that satisfy all of the selection criteria described in Section 2.1 are plotted with ﬁlled symbols, while unﬁlled symbols show the rest of the ﬁeld galaxies. The faint ﬁlled symbols show galaxies that satisfy less stringent selection criteria (dotted lines) that are often used in the literature.
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	Figure 2. The projected mass-weighted densities of DM (red), gas (green), and stars (white) within 2.5 Mpc of the Local Group in the 17_11 simulation. The two bright galaxies at the center are the analogs of the Milky Way and M31 and we mark the projected positions of the UDGs with light-blue circles.
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	In Fig. 2, we show the distribution of the UDGs in one representative high-resolution simulation (17_11; chosen ar- bitrarily). This shows the projected DM, gas, and stellar density in a spherical region with a radius of 2.5 Mpc cen- tered on the midpoint of the Milky Way and M31 analogs. The distribution of UDGs throughout the volume is not uni- form: at small distances from the center of the Local Group the UDGs cluster close to 𝑅200 of the Milky Way and M31 analogs, and in the other simulations they congregate nearer to the splashback radii of the hosts (as deﬁned in Diemer 2021). However, at larger distances from the center of the Local Group the UDGs are aﬃliated preferentially with the large structures that compose the Local Group analog and the ﬁlaments and sheets that deliver matter to it.
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	Table 1. The 𝑧=0 properties of the three simulations. We provide the total enclosed mass, 𝑀LG, the number of ﬁeld galaxies, 𝑁ﬁeld, tot, and the number of UDGs, 𝑁UDG, tot, with 106 ≤ 𝑀∗ / M(cid:12) ≤ 109 within 2.5 Mpc of the midpoint of the primary haloes. Note the diﬀerent observer position to that in Fig. 3.
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	ter simulation that shows that a large fraction of low-mass galaxies at 𝑧=0 are UDGs (Tremmel et al. 2020).
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	In Fig. 3, we show the cumulative radial distributions of ﬁeld galaxies and ﬁeld UDGs in each high-resolution volume with respect to the nearest host galaxy analog at 𝑧=0. We also overlay the incomplete census of observed ﬁeld galaxies in the stellar mass range described above. Their distances with respect to the Milky Way and M31 are calculated using the equatorial coordinates and distance moduli compiled in the most recently updated catalog of McConnachie (2012). As we described in Section 1, we exclude galaxies that are within the virial radius of the Milky Way or M31. The observational data are limited by incomplete sky coverage and insuﬃcient sensitivity to low-surface-brightness objects, which partly ex- plains the discrepancy between these data and the number of ﬁeld galaxies we identify in the simulations (see the bottom panel of Fig. 3). As we will discuss in Section 3.2, we think that the 17_11 and 09_18 simulations could be 1.25 − 1.5 times more massive than the Local Group. We expect that this would increase the number of ﬁeld galaxies we identify in the simulations by a similar factor. Between 62% and 80% of the ﬁeld UDGs in the simulations are found within 1.5 Mpc of the center of the Local Group, and approximately half are between 𝑅200 and 3 × 𝑅200 of the host galaxies (see the upper panel of Fig. 3). The latter distance is consistent with the splashback radius deﬁned in Diemer (2021). At larger radii, the UDGs are aﬃliated preferentially with the ﬁlaments and sheets that feed the growth of the Local Group (see also Fig. 2). This is in agreement with the results of Fattahi et al. (2020a), who used the APOSTLE simulations to show that most undiscovered dwarf galaxies should lie near the virial boundaries of the primary haloes.
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	Figure 3. Lower panel: the whole-sky radial cumulative distribu- tions of UDGs (solid lines) and all ﬁeld galaxies (dotted lines) with stellar masses, 106 ≤ 𝑀∗ / M(cid:12) < 109, as a function of the distance to the nearest host galaxy, 𝑟nearest, normalized to the host galaxy’s 𝑅200. We overlay the incomplete census of observed ﬁeld galaxies as points with error bars showing the 68% distance uncertainties (using data compiled by McConnachie 2012). Upper panel: the cumulative distribution function of the UDGs in each HESTIA sim- ulation. To aid the eye, we mark the 50% threshold with a horizontal dotted line.
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	Survey (DES; Abbott et al. 2018), and the Legacy Survey of Space and Time (LSST; Ivezić et al. 2019), we apply lim- iting surface brightness cuts in the r–band of 29, 30, and 31 mag arcsec−2, respectively. We choose these with ref- erence to the response function of the SDSS determined by Koposov et al. (2008), and the design sensitivity of the LSST (Laine et al. 2018), both of which bracket the sensitivity of the DES. These limits apply to studies of resolved galax- ies, and are several magnitudes deeper than the limits that are achievable in unresolved searches (V. Belokurov 2022, private communication). Under these assumptions few of the HESTIA UDGs are too faint to be detected (Fig. 4, left panel). This simpliﬁed scenario suggests that almost the entire ﬁeld UDG population is detectable in all-sky surveys with SDSS-, DES-, and LSST-like surface-brightness limits. In practice, the surface-brightness limits depend on distance such that nearby very-low-surface-brightness galaxies are not observ- able (see, e.g. Koposov et al. 2008). We ﬁnd that most of the ﬁeld UDGs in HESTIA are close to the Milky Way and M31 analogs, so they could be diﬃcult to detect; however, their surface brightnesses are high enough that they are detectable in the surveys.
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	Figure 4. The total luminosity functions of ﬁeld UDGs within 2.5 Mpc of each high-resolution Local Group simulation in the HESTIA suite. Left panel: we estimate the absolute V–band magnitude luminosity functions as a fraction of the total number of UDGs, 𝑁UDG, tot, likely to be observed in whole-sky SDSS-, DES-, and LSST-like surveys by applying a limiting surface-brightness cut in the r–band when assuming that the stars in each galaxy are individually resolved (SDSS: 29 mag arcsec−2, DES: 30 mag arcsec−2, LSST: 31 mag arcsec−2). With the exception of the SDSS and DES curves in the 09_18 simulation, the luminosity functions of all surveys overlap with the “Total” luminosity function. Right panel: the cumulative number of ﬁeld UDGs in the simulations as a function of apparent V–band magnitude, 𝑚V. The solid and dashed lines show the luminosity functions measured by observers in the Milky Way and M31 analogs, respectively. The vertical arrows indicate the faintest dwarf galaxies that could be detected in several past and future surveys: SDSS (𝑚V=16), DES (𝑚V=17.5), HSC (𝑚V=20), and LSST (𝑚V=21.5).
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	In the right panel of Fig. 4, we plot the apparent V–band magnitude luminosity functions of the UDG populations. We generate luminosity functions for an observer located in the Milky Way and M31 analogs; however, there is little dif- ference between them because the distributions of relative distances to the UDGs are similar. The magnitude limits of the surveys are marked with arrows and suggest that, on the basis of apparent magnitude alone, almost the entire UDG population is detectable in SDSS-, DES-, and LSST-like sur- veys. Taken together, the panels in Fig. 4 illustrate that UDGs in the Local Group should be detectable in all extant wide- area surveys. The most signiﬁcant factors that could limit their detectability are likely to be their inclination with re- spect to the observer, which we discuss in Section 3.2; and the obscuration of the sky by the Galactic disk, which we do not model here.
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	to ﬁnd in the footprints of current surveys such as the SDSS and whether they are, in principle, detectable using exist- ing data sets.
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	To generate a mock observation, we place an observer at the center of one of the primary haloes. We model the mock survey as a conical volume with an opening angle of 14,555 deg2, corresponding to the sky coverage of the SDSS, and orient it so that its apex coincides with the observer. To account for the eﬀects of the viewing angle, we assign each UDG a random orientation with respect to the observer and recalculate 𝑅e and 𝜇e. UDGs that fail the selection criteria described in Section 2.1 are discarded before the analysis pro- ceeds. Using the relative distances of the UDGs with respect to the observer and the recomputed values of 𝜇e, we calculate 𝜖 using eq. (1). This represents the probability of detecting each UDG, and we use it to randomly select a set of UDGs that are detectable in the mock survey. As most galaxies have 𝜖∼1 the eﬀect of the random sampling is small.
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	Figure 5. Mock SDSS observations of the population of ﬁeld UDGs. The solid and dashed lines show the median predictions obtained by observers in the Milky Way and M31 analogs, respectively. The shaded regions represent the 68% scatter in the Milky Way analog luminosity functions over 15,000 mock observations.
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	simulated UDGs that are detectable in each mock SDSS obser- vation are misclassiﬁed as non-UDGs because of projection eﬀects arising from their orientation with respect to the ob- server. Only ﬁve per cent of the mock observations contain at least one simulated UDG that is not detected at all. The pro- jection eﬀects impose the most signiﬁcant limitation on the discoverability of UDGs in extant surveys. As we showed in Section 3.1, survey incompleteness has only a minimal eﬀect on the number of UDGs that can be found.
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	and 2 mag, for each respective survey), we expect that 3+3 −2 UDGs are detectable in the combined SDSS+DES footprint, and 4 ± 2 will be detectable in the LSST (see Table 2). Us- ing our selection criteria, no UDGs have been observed in the SDSS footprint to date. Disregarding the eﬀects of dust attenuation, we estimate that the chance that there are no ﬁeld UDGs detectable in the SDSS footprint is less than 13.1%. In Table 2 we provide the predicted number of ﬁeld galax- ies and UDGs for diﬀerent choices of 𝑀LG and other UDG selection criteria.
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	of projection eﬀects. The total number of UDGs that are detectable is also subject to the variation in the masses of the three Local Group volumes we use. When renomaliz- ing these to 𝑀LG (< 2.5 Mpc) = 8 × 1012 M(cid:12), we ﬁnd 12±3 ﬁeld UDGs within 2.5 Mpc of the Milky Way–M31 midpoint, of which 2+2 −1 are detectable in the footprint of the SDSS (see Section 3.2 and Table 2). A full-sky survey with a response function similar to that of the SDSS, DES, or LSST will detect the entire population of ﬁeld UDGs.
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	In this Letter, we provide quantitative predictions of the size and luminosity function of the population of isolated UDGs within 2.5 Mpc of the Local Group (which we call “Local Group ﬁeld UDGs”), and estimate how many could be detectable in dedicated searches of current data sets, and in future surveys. We produce these predictions using the popu- lations of UDGs in the highest-resolution hydrodynamic sim- ulations from the HESTIA suite that are constrained to repro- duce the local large-scale structure at 𝑧=0. This is the ﬁrst time that such spatially extended galaxies have been simulated self- consistently in such environments (see Figs 1 and 2). To ob- tain our results, we rescale the simulations to a common mass, 𝑀LG (< 2.5 Mpc) = 8 × 1012 M(cid:12), which is consistent with current estimates of 𝑀LG (< 1 Mpc) from the Local Group timing argument (see Section 3.2). We predict that there are 12±3 (68% CL) low-surface-brightness UDGs in the ﬁeld of the Local Group with stellar masses, 106 ≤ 𝑀∗ / M(cid:12) < 109, and eﬀective radii, 𝑅e ≥ 1.5 kpc; and as many as 27 ± 5 when selecting UDGs with 𝑅e ≥ 1 kpc. The UDGs account for approximately one-quarter and one-half, respectively, of the total population of 52±7 (68% CL) ﬁeld galaxies with similar stellar masses. As many as 80% of these systems are within 1.5 Mpc of the Milky Way–M31 midpoint and cluster close to these two primary haloes (see Fig. 3), in agreement with the results of Fattahi et al. (2020a).
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	To generate mock SDSS observations, we used a simple model of the Koposov et al. (2008) SDSS response function. This depends on several physical properties of the galaxies such as their sizes and luminosities, their orientation with re- spect to the observer, and their physical locations, i.e. their heliocentric distances and projected positions on the sky. The latter are important because galaxies that are partially or to- tally obscured by the Milky Way can be more diﬃcult to detect against the high-density Galactic stellar foreground, i.e. the ZoA. In HESTIA, we ﬁnd that 13+20 −9 % (68% CL) of the total UDG population is in the ZoA on average. We do not account for this when estimating the detection eﬃciencies of the UDGs, and we further assume that the UDGs do not suﬀer from dust extinction. Correcting for both of these ef- fects would likely reduce the predicted number of UDGs that are detectable in the surveys.
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	As we have shown, UDGs are challenging to observe be- cause they are extremely diﬀuse. However, those that contain large reservoirs of neutral hydrogen, such as most isolated ob- served UDGs as well as the simulated ﬁeld UDGs in HESTIA (S. Cardona–Barrero et al. 2023, in preparation), could be de- tected more easily. In H i surveys the neutral hydrogen could appear as ultracompact high-velocity clouds (UCHVCs; Gio- vanelli et al. 2009; Adams et al. 2013). Recent searches for UCHVCs and other H i-bright systems using ALFALFA (e.g. Janesh et al. 2019), DES (Tanoglidis et al. 2021), and HIPASS (Zhou et al. 2022) have produced promising results that could expand the catalog of targets for dedicated follow-up studies. Our results suggest that there is a population of low-surface- brightness, spatially extended galaxies in the Local Group awaiting discovery.
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	Table 2. The total number of ﬁeld galaxies, 𝑁ﬁeld, tot, UDGs, 𝑁UDG, tot, and the number of UDGs detectable in the SDSS footprint, 𝑁UDG, SDSS, for combinations of 𝑀LG (< 2.5 Mpc) and UDG selection criteria. Our ﬁducial choice is in bold.
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	Foundation grant ISF 1358/18. JS acknowledges support from the ANR LOCALIZATION project, grant ANR-21- CE31-0019 of the French Agence Nationale de la Recherche. AK is supported by the Ministerio de Ciencia e Innovación (MICINN) under research grant PID2021-122603NB-C21. ET acknowledges support by ETAg grant PRG1006 and by the EU through the ERDF CoE grant TK133. The au- thors acknowledge the Gauss Centre for Supercomputing e.V. (www.gauss-centre.eu) for providing computing time on the GCS Supercomputer SuperMUC-NG in support of the HES- TIA project.
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	Software: astropy (The Astropy Collaboration et al. 2018), matplotlib (Hunter 2007), numpy (Harris et al. 2020), pynbody (Pontzen et al. 2013), python (Van Rossum & Drake 2009), scipy (Virtanen et al. 2020), and NASA’s Astrophysics Data System.
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	A repository of reduced data and scripts to produce the ﬁgures in this Letter is available on GitHub1 and archived in Zenodo (Newton 2023). Requests for access to the raw HESTIA simulation data should be directed to a CLUES Col- laboration PI.
Processed_The_undiscovered_ultra-diffuse_galaxies_of_the_Loc.txt	Planck Collaboration et al., 2014, A&A, 571, A16 Pontzen A., Roškar R., Stinson G., Woods R., 2013, ASCL, p.
Processed_The_Radio_Remnant_of_Supernova_1987A_-_A_Broader_V.txt	Abstract. Supernova remnants (SNRs) are powerful particle accelerators. As a supernova (SN) blast wave propagates through the circumstellar medium (CSM), electrons and protons scatter across the shock and gain energy by entrapment in the magnetic ﬁeld. The accelerated parti- cles generate further magnetic ﬁeld ﬂuctuations and local ampliﬁcation, leading to cosmic ray production. The wealth of data from Supernova 1987A is providing a template of the SN-CSM interaction, and an important guide to the radio detection and identiﬁcation of core-collapse SNe based on their spectral properties. Thirty years after the explosion, radio observations of SNR 1987A span from 70 MHz to 700 GHz. We review extensive observing campaigns with the Australia Telescope Compact Array (ATCA) and the Atacama Large Millimeter/submillimeter Array (ALMA), and follow-ups with other radio telescopes. Observations across the radio spec- trum indicate rapid changes in the remnant morphology, while current ATCA and ALMA ob- servations show that the SNR has entered a new evolutionary phase.
Processed_The_Radio_Remnant_of_Supernova_1987A_-_A_Broader_V.txt	forward SN shock, i.e. the outer blast wave, and the reverse shock, which causes decelera- tion and heating of the inner ejecta, accelerate the shocked material, with some fractions of electrons and protons reaching cosmic ray energies. The acceleration mechanism usu- ally associated with the shock propagation is diﬀusive shock acceleration (DSA), also known as the ﬁrst-order Fermi mechanism (Drury 1983). According to DSA theory, ener- getic particles at shock fronts undergo spatial diﬀusion in a uniform magnetic ﬁeld. If the magnetic ﬁeld is not uniform, as in the case of young SNRs, the charged particles can get partially trapped in the magnetic ﬁeld structure, while still gaining energy by scattering between the upstream and downstream regions of the forward shock (Kirk et al. 1996). This leads to a sub-diﬀusive particle transfer (sub-DSA), since the particle density at the shock front is lower than it is far downstream. In either case, as energetic electrons gyrate in the local magnetic ﬁeld, they emit synchrotron emission. The synchrotron emission να, with Sν the ﬂux den- observed in SNRs can be well described by a power law Sν sity measured at the observing frequency ν. Simpliﬁed synchrotron physics tells us that (cid:112)ν/B GeV, emission at the frequency ν is primarily by electrons with energy E with ν expressed in GHz and the magnetic ﬁeld strength, B, expressed in µG (Reynolds 2011).
Processed_The_Radio_Remnant_of_Supernova_1987A_-_A_Broader_V.txt	SN 1987A in the Large Magellanic Cloud, identiﬁed as Type IIP (Leibundgut & 20 M(cid:12) (Smartt 2009), is the only nearby CCSN Suntzeﬀ 2003) with progenitor of mass observed with a telescope since its early stages. Given its proximity, SN 1987A has al- lowed unique studies of the radio evolution of CCSNe and of the SN CSM interaction. The complex CSM distribution in the remnant is believed to have originated from a red supergiant (RSG) which has evolved into a blue supergiant (BSG) about 20,000 years before the explosion (Crotts & Heathcote 2000). In its transition from supernova to SNR, SN 1987A has continued to brighten for the past three decades, thus making possible to monitor the rapid changes of the radio emission with a number of telescopes over an increasingly broad frequency range.
Processed_The_Radio_Remnant_of_Supernova_1987A_-_A_Broader_V.txt	The interaction of the SN shock wave with the CSM in the equatorial plane (equatorial ring, ER) is the most notable feature of the radio remnant of SN 1987A. The ER deﬁnes the waist of the ‘hourglass’ environment created by the progenitor star prior to explosion. As the SN propagates outwards and impacts the denser CSM in the ER, the shocks expand above and below the ER plane (see Figure 1) and interact with high-latitude material (Potter et al. 2014), conﬁned within the inner hourglass structure.
Processed_The_Radio_Remnant_of_Supernova_1987A_-_A_Broader_V.txt	Figure 1: Stages of the shock expansion from HD simulations by Potter et al. (2014) (see their Figures 1, 6, and 7), as modelled around 1000 days (top right), at day ∼2000 (bottom right), and at day ∼8000 (bottom left) since the SN. By day ∼8000, the forward shock has overtaken the equatorial ring (ER). The colour scheme varies from white to dark brown, indicating the increasing particle number density in log10 scale (∼ 5 − 11 × 107 m−3, see Potter et al. 2014). A 3D view of both the forward and reverse shocks expanding above and below the ER, is shown on the top left image, which is a rendering of the volumetric models developed by Potter et al. (2014) (see their Figure 9) at day ∼8000.
Processed_The_Radio_Remnant_of_Supernova_1987A_-_A_Broader_V.txt	km s−1 from day 1200 (Gaensler et al. 1997), when the radio emission started to increase 5000 the radio emission has been rising at an exponential (Ball et al. 2001). Since day ∼ CSM interaction is observed rate (Zanardo et al. 2010, Figure 2). A new phase of the SN − at all frequencies after day 8000 (Figure 2; Zanardo 2014), when the shock has likely become engulfed in the densest CSM in the ER, while still expanding above and below the ring.
Processed_The_Radio_Remnant_of_Supernova_1987A_-_A_Broader_V.txt	Figure 2: Left − Flux densities for SNR 1987A. Plots include data from observations at 1.4, 2.4, 4.8, and 8.6 GHz from August 1990 to June 2012 and are stacked with an oﬀset of 0.1 mJy (Zanardo et al. 2010; Zanardo 2014). The red dashed lines are the exponential ﬁts of data from November 2000 to February 2009, as derived by Zanardo et al. (2010). The blue dashed lines are 1500-day ﬁts of data from February 2009. The colour bands are included to highlight diﬀerent rates of the ﬂux density increase. Top Right − Remnant radius resultant from images at 9 GHz, from August 1992 to March 2013, as derived by Ng et al. (2013) via ﬁtting the SNR with a torus (black) and an elliptical ring (red: semi-major axis; green: semi-minor axis). The radius of the radio remnant is compared with that derived from X-ray images (blue, from Helder et al. 2013). The solid lines indicate the best-ﬁt expansion. Bottom Right − Evolution of the remnant asymmetry at 9 GHz as derived from the torus (black) and ring (red) models (Ng et al. 2013). Credit: Zanardo (2014), Ng et al. (2013).
Processed_The_Radio_Remnant_of_Supernova_1987A_-_A_Broader_V.txt	Figure 3: 9 GHz images of SNR 1987A from 1992 to 2013, adapted from Ng et al. (2013) (see Figure 1 therein). Images include data from Gaensler et al. (1997), Ng et al. (2008), and Ng et al. (2013). Colour-bar units are Jy beam−1. Credit: Ng et al. (2013), Zanardo (2014).
Processed_The_Radio_Remnant_of_Supernova_1987A_-_A_Broader_V.txt	Figure 4: Contours of the 2010 VLBI image of SN 1987A at 1.4 GHz are overlaid with the 44 GHz image derived from 2011 observations (Zanardo et al. 2013). The 44 GHz image has a resolution of 0.(cid:48)(cid:48)2, while the VLBI image has a restoring beam of 61 × 95 mas. The VLBI observations have been instrumental to align images prior to the derivation of spectral index maps (see Zanardo et al. 2013, 2014). Credit: Zanardo (2014).
Processed_The_Radio_Remnant_of_Supernova_1987A_-_A_Broader_V.txt	have shown small-scale structures more prominent on the eastern lobe, which are likely due to the shock wave interacting with discrete clumps near the inner surface of the SN shell (see Ng et al. 2011).
Processed_The_Radio_Remnant_of_Supernova_1987A_-_A_Broader_V.txt	Figure 5: Mosaic of radio images of the remnant of SN 1987A at day 9280. Each image frequency is indicated on the top right corner, in GHz. Images below 100 GHz are derived from ATCA observations (Laki´cevi´c et al. 2012a; Ng et al. 2013; Zanardo et al. 2013), images above 100 GHz are from ALMA data (Zanardo et al. 2014). All images have been super-resolved with a circular beam (lower left corner). The bottom right image is the spectral index map obtained by combining images at ﬁve frequencies, i.e. 44, 94, 102, 213, and 345 GHz (see Zanardo et al. 2014). The spectral image is overlaid with contours (in blue) of the 15% and 60% emission levels at 44 GHz (Zanardo et al. 2013). Credit: Zanardo (2014).
Processed_The_Radio_Remnant_of_Supernova_1987A_-_A_Broader_V.txt	Figure 6: SED of SNR 1987A from radio to FIR, with data from: ATCA at 1.4 GHz (Zanardo 2014), 9 GHz (Ng et al. 2013), 18 and 44 GHz (Zanardo et al. 2013), and 94 GHz (Laki´cevi´c et al. 2012a); ALMA (Zanardo et al. 2014); the Atacama Pathﬁnder EXperiment (APEX) at 345 and 857 GHz (Laki´cevi´c et al. 2012b); the Herschel Space Observatory at 600−3000 GHz (Matsuura et al. 2011). The brown dash-dot-dotted curve is the amorphous carbon dust ﬁt for ALMA data and Herschel observations carried out in 2012 (Matsuura et al. 2015). To match the average epoch of ALMA observations, ATCA data are scaled to day 9280, via exponential ﬁtting parameters derived for ATCA ﬂux densities from day 8000, as measured at 1.4, 8.6 and 9 GHz (Zanardo 2014). The hollow red diamond indicates the central emission measured at 44 GHz as reported by Zanardo et al. (2013) scaled to day 9280. The diﬀerence between possible spectral ﬁts is highlighted in light orange. Credit: Zanardo et al. (2014).
Processed_The_Radio_Remnant_of_Supernova_1987A_-_A_Broader_V.txt	Figure 7: An outline of the equatorial ring and inner debris, as seen with HST (green/blue contours), on top of ALMA observations of the remnant at 345 GHz (Zanardo et al. 2014) (red/orange). The dust in the central region is rendered. (Hubble data: NASA/STScI/CfA/P. Challis and R. Kirshner.) Credit: Zanardo (2014).
Processed_The_Radio_Remnant_of_Supernova_1987A_-_A_Broader_V.txt	The search for a compact object in the inner regions of SNR 1987A has been under- taken on multiple fronts. It has been shown that ﬂat spectral synchrotron indices can be associated either with accretion onto a black hole or with a pulsar wind nebula (PWN; Gaensler & Slane 2006). Spectral index maps of the remnant derived at diﬀerent fre- quencies and epochs have consistently featured a ﬂat central component (Laki´cevi´c et al. 2012a; Potter et al. 2009; Zanardo et al. 2013, 2014).
Processed_The_Radio_Remnant_of_Supernova_1987A_-_A_Broader_V.txt	et al. 2014). An asymmetric explosion of the progenitor could have led to a high-velocity neutron star (Janka 2012) and, thus, to the formation of a PWN oﬀset from the SN site. After unsuccessful searches for the pulsar signal with the Parkes telescope (Manchester 1988, 2007), the preliminary analysis of data from an extended pulsar search campaign in 2013 has tentatively identiﬁed several candidates with period 30 (cid:46) P (cid:46) 300 ms (Zanardo 2014). However, there is yet no convincing detection and the search continues.
Processed_The_Radio_Remnant_of_Supernova_1987A_-_A_Broader_V.txt	References Ball, L., Campbell-Wilson, D., Crawford, D. F., & Turtle, A. J. 1995, ApJ, 453, 864 Ball, L., Crawford, D. F., Hunstead, R. W., et al. 2001, ApJ, 549, 599 Bell, A. R., & Lucek, S. G. 2001, MNRAS, 321, 433 Bell, A. R. 2004, MNRAS, 353, 558 Boggs, S. E., Harrison, F. A., Miyasaka, H., et al. 2015, Science, 348, 670 Callingham, J. R., Gaensler, B. M., Zanardo, G., et al. 2016, MNRAS, 462, 290 Chevalier, R. A., & Fransson, C. 1987, Nature, 328, 44 Chevalier, R. A., & Soker, N. 1989, ApJ. 341, 867 Crotts, A. P. S. & Heathcote, R. S. 2000, ApJ, 528, 426 Drury, L. Oc. 1983, Rep. Prog. Phys., 46, 973 Gaensler, B. M., Manchester, R. N., Staveley-Smith, L., et al. 1997, ApJ, 479, 845 Gaensler, B. M. 1998, ApJ, 493, 781 Gaensler, B. M., & Slane, P. O. 2006, ARAA, 44, 17 Helder, E. A., Broos, P. S., Dewey, D., et al. 2013, ApJ, 764,11 Indebetouw, R., Matsuura, M., Dwek, E., et al. 2014, ApJ, 782, L2 Janka, H.-T. 2012, Annual Review of Nuclear and Particle Science, 62, 407 Kamenetzky, J., McCray, R., Indebetouw, R. et al. 2013, ApJ, 773, L34 Kjær K., Leibundgut, B., Fransson, C., et al. 2010, A&A, 517, A51 Kirk, J. G., Duﬀy P., & Gallant, Y. A. 1996, A&A, 314, 1010 Laki´cevi´c, M., Zanardo, G., van Loon, J. Th., et al. 2012, A&A, 541, L2 Laki´cevi´c, M., van Loon, J. Th., Stanke, T., et al. 2012, A&A, 541, L1 Larsson, J., Fransson, C., Kjær, K., et al. 2013, ApJ, 768, 89 Leibundgut, B., & Suntzeﬀ, N. B. 2003, Supernovae and Gamma-Ray Bursters, edited by K.
Processed_Unconfused_Ultraconservative_Multiclass_Algorithms.txt	Besides the theoretical questions raised by the learning setting considered, we may depict the following example of an actual learning scenario where learning from noisy data is relevant. This learning scenario will be further investigated from an empirical standpoint in the section devoted to numerical simulations (Section 4).
Processed_Unconfused_Ultraconservative_Multiclass_Algorithms.txt	Example 1 One situation where coping with mislabelled data is required arises in scenarios where labelling data is very expensive. Imagine a task of text categorization from a training set S = S(cid:96) ∪ Su, where S(cid:96) = {(xi, yi)}n i=1 is a set of n labelled training examples and Su = {xn+i}m i=1 is a set of m unlabelled vectors; in order to fall back to a realistic training scenario, we may assume that n << m. A possible three-stage strategy to learn a predictor is as follows: ﬁrst learn a predictor f(cid:96) on S(cid:96) and estimate its confusion C error via a cross- validation procedure, second, use the learned predictor to label all the data in Su to produce the labelled traning set (cid:98)S = {(xn+i, tn+i := f (xn+i))}m i=1 and ﬁnally, learn a classiﬁer f from (cid:98)S and the confusion information C. Contributions Our main contribution is to show that it is both practically and theoreti- cally possible to learn a multiclass classiﬁer on noisy data as long as some information on the noise process is available. We propose a way to compute update vectors for any ultracon- servative algorithm which allows us to handle massive amount of mislabeled data without consequent loss of accuracy. Moreover, we provide a thorough analysis of our method and show that the strong theoretical guarantees that caracterize the family of ultraconservative algorithm carry over to the noisy scenario.
Processed_Unconfused_Ultraconservative_Multiclass_Algorithms.txt	Organization of the paper. Section 2 formally states the setting we consider throughout this paper. Section 3 provides the details of our main contribution: the UMA algorithm. Section 4 presents numerical simulation that support the soundness of our approach.

Subsets and Splits

No community queries yet

The top public SQL queries from the community will appear here once available.