Datasets:

SouthernCrossAI
/

Arxiv

Dataset card Data Studio Files Files and versions Community

Split (1)

train · 132k rows

Filename stringlengths 22 64	Paragraph stringlengths 8 5.57k
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	Concentration bound. We ﬁrst state a concentration in- equality that controls the underestimation probability based on an alternative Chernoff bound for a sum of independent binary random variables (Lemma 13 proved in Appendix A).
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	This lemma only holds for a sequence of pulls ﬁxed before- hand, independently of realizations, i.e., the values of At do not depend on the sequence of Xs. Although with a restrictive scope, it provides intuition on the construction of the algorithm.
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	where dPois(p, q) = p log p/q + q − p denotes the Poisson Kullback-Leibler divergence.
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	Even though the Kullback-Leibler divergence does not have the same expression for Bernoulli and Poisson random vari- ables, the following lemma (proved in Appendix A.1) shows that for a certain range or parameters they are actually very close.
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	the parameter estimator. In both cases, one may also consider either of the two UCB or KL-UCB index deﬁned in the previ- ous section, resulting in the DelayedUCB and DelayedKLUCB algorithms. We provide a ﬁnite-time analysis of the regret of these algorithms, when using an exploration function of the form β(cid:15)(t) = (1 + (cid:15)) log(t), for some positive (cid:15).
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	The last term is actually O((cid:112)log(T )), giving the desired result. Details, as well as explicit constants and dependencies can be found in Appendix C.1.
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	Proof The analysis of DelayedUCB given in Appendix C.1 (in the censored setting) shows that the performances of DelayedUCB in the uncensored setting can be upper-bounded by its performances in the censored setting, where the threshold m can be arbitrarily ﬁxed to some value. Choosing m will only have an impact on the analysis of the algorithm. The speciﬁc choice of m satisfying τm ≥ 1 − (cid:15) gives the claimed result. As indicated in Appendix C.1, the dependency of o(cid:15),m(log(T )) is actually only linear in m. As a consequence, along with the assumption on the decay of 1 − τm, this yields that the overall dependency in the parameter m is reduced to 1/(cid:15).
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	We emphasize that the assumption that 1 − τm ≤ 1/m, is actually rather natural. Indeed, if 1 − τm ≤ c/mγ, for some constants c, γ > 0, then the ﬁniteness requirement on the ex- pected delay is satisﬁed if and only if γ > 1.
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	Proof As for the proof of Corollary 10, the performance of DelayedKLUCB in the uncensored case can be bounded as in the censored case by a speciﬁc choice of m((cid:15)) such that τm ≥ 1 − (cid:15), namely m((cid:15)) ≥ c/(cid:15). As shown in the proof of Theorem 11 in the censored case, the dependency in m of the term of rest is linear, reducing to 1/(cid:15).
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	These indices can only be computed after at least m rounds. The proof technique used for the analysis of our algorithms in the censored case actually shows that the DISCARDINGUCB and DISCARDINGKLUCB policies are asymptotically opti- mal. Nonetheless, in practice it is very undesirable to have an arbitrarily long linear regret phase at the beginning of the learning until the threshold m is reached. This is especially true if the threshold m is large as compared to the horizon T . In that case, we empirically show in Section 7 that our algorithms achieve drastically improved short-horizon performance.
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	In this section we perform simulations in our two delayed feed- back frameworks. The algorithms described in the previous section will be denoted D-UCB and D-KLUCB in the censored setting, and UD-UCB and UD-KLUCB in the uncensored setting. As a matter of fact, the bottleneck of such policies is to com- pute ˜N (t) which is theoretically a weighted sum over all past actions and, without any assumption on the weights (τs)s≥0, it requires to store all previous rewards and recompute ˜N (t) at each iteration.
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	In practice this can be done by maintaining a buffer of size m containing the last m pulls that are multiplied by the probability of observing a reward with the delay corresponding to their current position in the buffer. In addition to this buffer, we add old pulls in a separate count Nk(t − m) for which the weight will stay τm.
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	Comparing DelayedUCB and DelayedKLUCB. We com- pare the regret of both delayed bandits policies in the cen- sored and uncensored setting for T = 10000, µ = 500 and m = 1000.
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	two problems, θH = (0.5, 0.4, 0.3) on the left, and θL = (0.1, 0.05, 0.03) on the right, display the classical pattern that while UCB-based algo- rithms perform satisfactorily for central values (close to 0.5) of the conversion rate, they are clearly sub-optimal with more re- alistic values for the conversion rate. The two right plots also conﬁrm that, for the KLUCB-based algorithms, the loss with respect to the optimal regret growth rate due to the use of the Poisson divergence is – as expected from Theorem 11 – not signiﬁcant for low values (here θ∗ = 0.1) of the conversion rates.
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	DelayedUCB and DelayedKLUCB vs. DISCARDING. In this section, we illustrate the good empirical initial perfor- mance of DelayedUCB and DelayedKLUCB, when compared to the heuristic DISCARDING approach presented in Section 6.
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	Figure 1: Expected regret of D-UCB and D-KLUCB (censored setting), and UD-UCB and UD-KLUCB (uncensored setting) for two bandit problems: θH = (0.5, 0.4, 0.3), θL = (0.1, 0.05, 0.03). For all experiments, T = 10000, µ = 500, m = 1000 (if censored) and the results are averaged over 100 runs.
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	Figure 2: Expected regret of D-UCB and D-KLUCB in the censored setting vs the equivalent discarding policies when µ = 500 and m = 1000. Results are averaged over 200 in- dependent runs.
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	Figure 2 compares results for both DelayedUCB and DelayedKLUCB with θ = (0.1, 0.05, 0.03), T = 10000, µ = 500 and m = 1000 in the censored setting. We ob- serve that discarding policies incur a linear regret phase at the beginning of the learning and happen to catch up with the ex- pected regret growth rate only after a large number of rounds. These ﬁgures reveal a non-negligible gap in performance be- tween the naive DISCARDING approach and our delay-adapted quasi-optimal algorithms.
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	The stochastic delayed bandit setting introduced in this work addresses an important problem in many applications where the feedback associated to each action is delayed and censored, due to the ambiguity between conversions that will never hap- pen and conversions that will occur at some later – perhaps unobservable– time. Under the hypothesis that the distribution of the delay is known, we provided a complete analysis of this model as well as simple and efﬁcient algorithms. An interest- ing generalization of the present work would be to relax the model hypothesis and estimate the delay distribution on-the- go, possibly using context-dependent delay distributions.
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	and Olivier Capp´e. Multiple-play bandits in the position-based model. In Ad- vances in Neural Information Processing Systems, 2016.
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	We require variants of Lemma 9 and Theorem 10 in [5] adapted to our setting.
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	Lemma 13. For θ ∈ [0, 1], (τi)1≤i≤L ∈ [0, 1], let (Xi,j)1≤i≤L,j≥1 be a collection of independent Bernoulli random variables such that E(Xi,j) = τiθ and ((cid:15)i,j) ∈ {0, 1} associated deterministic indicators. For 1 ≤ i ≤ L, denote by ni = (cid:80)∞ j=1 (cid:15)i,j and whe shall assume that all ni are ﬁnite an that at least one of them is non-zero. Let X = (cid:80)L j=1 (cid:15)i,jXi,j and denote by φ(λ) = log E [exp(λX)] its log-Laplace transform and by φ∗(x) = supλ xλ − φ(λ) the associated convex conjugate (Fenchel– Legendre transform).
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	where dPois(p, q) = p log p/q + q − p denotes the Poisson Kullback-Leibler divergence.
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	Lemma 13 bounds the log-Laplace transform of the Bernoulli distribution with that of the Poisson distribution with the same mean and uses the stability of the Poisson distribution. Using dPois(p, q) instead of d(p, q) –where d(p, q) = p log p/q + (1 − p) log[1 − p)/(1 − q) denotes the Bernoulli Kullback-Leibler divergence– will of course induce a performance gap, which is however not signiﬁcant for low values of the probabilities as shown by the Lemma 8, which we recall below.
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	We can now prove the concentration result stated in Lemma 7 that we recall below for readability purpose.
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	where dPois(p, q) = p log p/q + q − p denotes the Poisson Kullback-Leibler divergence.
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	Theorem 14. Consider (τi)1≤i≤L ∈ [0, 1], θ ∈ (0, 1), and independent sequences (Xi(s))s≥1 of independent Bernoulli ran- dom variables such that EXi(s) = τiθ. Let Ft denote an increasing sequence of sigma-ﬁelds, such that for each t and all i, σ(Xi(1), . . . , Xi(t)) ⊂ Ft. Also consider a predictable sequence of indicator variables (cid:15)i(s) ∈ {0, 1}, that is, such that σ((cid:15)1(t + 1), . . . , (cid:15)L(t + 1)) ⊂ Ft.
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	Proof The proof is analogous to that of Theorem 10 of [5] and we only detail the step that differs, namely, the identiﬁcation of the supermartingale W λ t .
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	The rest of the proof is as in [5] replacing N (t) by ˜N (t) and φµ(λ) by θ (cid:0)eλ − 1(cid:1).
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	Now, considering speciﬁc changes of measures θ(cid:48) that only modify the distribution of one single suboptimal arm, we are going to obtain lower bounds on each expected number of pulls E[Nk(T )] for k (cid:54)= 1 as in Appendix B of [13].
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	Now, in order to obtain an expression that involves E[Nk(T )], we need to bound from above this sum using Lemma 5 of the Appendix B of [11], which we recall here for completeness.
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	Lemma 15. Let p, q be any ﬁxed real numbers in (0, 1). The function f : α (cid:55)→ d(αp, αq) is convex and increasing on (0, 1). As a consequence, for any α < 1, d(αp, αq) < d(p, q).
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	Then, we use this upper bound on the indices in order to bound the relevant sum of probabilities.
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	where the last inequality comes from the fact that for all s ∈ {1, . . . , T }, (cid:80)T 1 {At+1 = k, Nk(t − m) = s} ≤ m and from the proof of Fact 2 for exponential family bandits in [2] that proves the existence of the constants C2(η) and f (η) that achieve the bound.
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	As a last additional contribution to this work, we suggest a distribution-agnostic heuristic that estimates the CDF parameters (τd)d≥0 in an online fashion. Indeed, as the delay distribution is assumed to be shared between actions, each observed reward provides an information on the delays that can be exploited to estimate the CDF without having to deal with the exploration-exploitation dilemma.
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	Then we use this estimator as a plug-in quantity to compute Ok(t) deﬁned in (6) for all k.
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	In the Censored setting however, no observation comes after the threshold m and this does not allow us to Censored setting. directly estimate the expected delay µ as the longest observations are censored. To circumvent this problem, we choose to estimate biased parameters for τ1, . . . , τm. Concretely, we initialize counts for the observed delay values δ0 = (0, . . . , 0) ∈ Nm+1 (delay can be null). Then, after each observation Dt, we increment all the counts δs for s ≥ Dt. Then, the biased empirical CDF is obtained by normalizing those counts by the total number of observations received up to time t, nd(t). We emphasize that the obtained estimators are biased: For each s ∈ {0, . . . , m}, E[δs(t)/nd(t)] = τs/τm as all observed delays are smaller or equal to m. Thus, plugging those estimates in ˜Nk(t) actually allows to have an estimate of ˜Nk(t)/τm instead of ˜Nk(t) and consequently an estimate of τmθk instead of θk.
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	Figure 3: Expected regret of DelayedKLUCB with and without online estimation of the CDF in both the censored and uncensored setting. Results are averaged over 100 independent runs.
Processed_Stochastic_Bandit_Models_for_Delayed_Conversions.txt	Figure 3 compares both our policies to its equivalent, delay-agnostic heuristic using the same conﬁdence intervals with plug- in estimates of the (τd)d≥0. It is clear from these experiments that using delay parameters estimated on-the-go does not hurt the cumulated regret overall.
Processed_The_DES_Bright_Arcs_Survey:_Candidate_Strongly_Len.txt	We report the combined results of eight searches for strong gravitational lens systems in the full 5,000 sq. deg. of Dark Energy Survey (DES) observations. The observations accumulated by the end of the third observing season fully covered the DES footprint in 5 ﬁlters (grizY), with an i−band limiting magnitude (at10σ) of 23.44. In four searches, a list of potential candidates was identiﬁed using a color and magnitude selection from the object catalogs created from the ﬁrst three observing seasons. Three other searches were conducted at the locations of previously identiﬁed galaxy clusters. Cutout images of potential candidates were then visually scanned using an object viewer. An additional set of candidates came from a data-quality check of a subset of the color-coadd “tiles” created from the full DES six-season data set. A short list of the most promising strong lens candidates was then numerically ranked according to whether or not we judged them to be bona ﬁde strong gravitational lens systems. These searches discovered a diverse set of 247 strong lens candidate systems, of which 81 are identiﬁed for the ﬁrst time. We provide the coordinates, magnitudes, and photometric properties of the lens and source objects, and an estimate of the Einstein radius for 81 new systems and 166 previously reported. This catalog will be of use for selecting interesting systems for detailed follow-up, studies of galaxy cluster and group mass proﬁles, as well as a training/validation set for automated strong lens searches.
Processed_The_DES_Bright_Arcs_Survey:_Candidate_Strongly_Len.txt	Gravitational lensing occurs when the gravitational ﬁeld of a massive foreground object, the “lens”, If the lens is massive enough, and the deﬂects the light of a more distant object, the “source”. source, lens, and observer suﬃciently collinear and distant from one another, the observed image of the source object can be distorted into rings, arcs, or multiple images. This eﬀect is known as strong gravitational lensing (SL), and has become a powerful and popular tool for studying a wide variety of extragalactic physics and cosmology.
Processed_The_DES_Bright_Arcs_Survey:_Candidate_Strongly_Len.txt	lensing of very distant objects by foreground galaxies was pre- dicted (Zwicky 1937) four decades prior, the ﬁrst discovery of a strongly lensed object was that of QSO 0957+561 (Walsh et al. 1979), a double image of a single quasar at redshift z = 1.4. Not long after, it was realized that the long, narrow, curved features bent around the centers of galaxy clusters were actually SL systems as well (Soucail et al. 1987; Lynds & Petrosian 1989; Grossman & Narayan 1989). In the subsequent 40-year interval, the count of candidate strongly lensed galaxies has expanded to more than 2000. At the same time, so has their scientiﬁc interest.
Processed_The_DES_Bright_Arcs_Survey:_Candidate_Strongly_Len.txt	matter (e.g. Koopmans et al. 2009; Wiesner et al. 2012; Treu & Ellis 2015; Newman et al. 2015), and explore the connection between these characteristics and the occurrence of strong lensing (Robertson et al. 2020; Fox et al. 2021; Sonnenfeld & Cautun 2021). Strong lens systems are now being used to study galaxy and cluster substructures and to, in turn, constrain the interaction physics of dark matter (Meneghetti et al. 2020; Gilman et al. 2021).
Processed_The_DES_Bright_Arcs_Survey:_Candidate_Strongly_Len.txt	Strong gravitational lensing enables studies of cosmology. Studies of characteristics of ensembles of strong lensing systems can be used to constrain ΩM in wCDM and ΛCDM (Leaf & Melia 2018), albeit loosely, with samples of O(100) conﬁrmed lensing systems. Studies of gravitationally lensed time-varying systems (Refsdal 1964; Blandford & Narayan 1992; Birrer et al. 2016; Oguri 2019), such as lensed quasars (Schechter et al. 1997; Suyu et al. 2013) or supernovae (Grillo et al. 2018), are used to constrain the expansion history between source, lens, and observer.1 Individual measurement of strongly-lens quad quasars systems provide (Bonvin et al. 2017; Shajib et al. 2020) constraints on H0 with about 4% precision, and combinations (Suyu et al. 2017; Wong et al. 2020) are now in mild tension, or not (Birrer et al. 2020), with early Universe probes, depending (Birrer & Treu 2021) on assumptions about the lens’ radial mass proﬁles. Lens systems with multiple sources at diﬀering redshifts can provide (Link & Pierce 1998; Gavazzi et al. 2008; Collett et al. 2012) complementary information about the expansion history, independent of the Hubble constant. Constraints on ΩM and the dark energy equation of state, w, are at about the 20% level from individual lensing systems (Jullo et al. 2010; Collett & Auger 2014). In these systems it can be critical to accurately model massive lens substructures to obtain unbiased cosmological constraints (D’Aloisio & Natarajan 2011).
Processed_The_DES_Bright_Arcs_Survey:_Candidate_Strongly_Len.txt	The count of candidate SL systems has increased rapidly with the advent of deep wide-ﬁeld surveys with dedicated SL searches. In the CFHTLS Strong Lensing Legacy Survey (Cabanac et al. 2007), 54 systems with promising lenses were identiﬁed by the ARCFINDER algorithm (More et al. 2012) in 150 sq. deg., 49 conﬁrmed strong lens systems identiﬁed with RingFinder (Gavazzi et al. 2014), and 29 promising and 59 total (More et al. 2016) from a crowdsourcing (Marshall et al. 2016) eﬀort. CFHTLS imaging yielded 16 more probable to deﬁnite SL systems using neural network- based searches (Jacobs et al. 2017). The Survey of Graviationally-lensed Objects in HSC Imaging (SuGoHI, Sonnenfeld et al. (2018)) used three diﬀerent methods, including YattaLens, to search the Hyper Suprime-Cam Subaru Strategic Program (HSC SSP) images. The program yielded 333 galaxy- galaxy SL candidates from an area of 442 deg2. The Kilo Degree Survey images were searched using machine-learning (ML) classiﬁers that discovered 227 high grade SL candidates, plus an additional 200 candidates discovered serendipitously (Petrillo et al. 2017, 2019; Li et al. 2020). Ca˜nameras et al. (2020) discovered 330 new SL systems in the 30,000 sq. deg. Pan-Starrs 3π Survey (Chambers et al. 2016). The 1.64 deg2 Hubble Space Telescope (HST) COSMOS survey ﬁeld yielded 67 galaxy- galaxy lens candidates (Faure et al. 2008). HST was also used to conﬁrm a total of 110 SL systems identiﬁed (Bolton et al. 2008; Shu et al. 2017) in the Sloan Digital Sky Survey (SDSS) spectroscopic data. SDSS also yielded 19 conﬁrmed systems to the Sloan Bright Arcs Survey (Allam et al. 2007; Kubo et al. 2009; Lin et al. 2009; Diehl et al. 2009; Kubo et al. 2010), and more than 30 conﬁrmed and 50 additional candidate lenses to the CASSOWARY survey (Belokurov et al. 2009; Pettini et al. 2010; Stark et al. 2013). Searches of galaxy clusters identiﬁed in SDSS yielded 16 strong lens systems with > 10(cid:48)(cid:48) radius and 21 additional SL candidates (Hennawi et al. 2008), 68 giant arcs in Wen et al.
Processed_The_DES_Bright_Arcs_Survey:_Candidate_Strongly_Len.txt	Galaxy clusters, among the most massive objects in the universe, are ideal candidates for strong lenses, and optical observations of galaxy cluster catalogs have yielded many SL systems. Galaxy cluster catalogs selected via optical observations, the Sunyaev-Zel’dovich eﬀect (Sunyaev & Zeldovich 1972), or as extended high-ﬂux x-ray sources have been used. Follow-up of South Pole Telescope (SPT) (Reichardt et al. 2013) detections yielded 34 new and 10 previously known SL systems in Bleem et al. (2015), and 44 new SL systems in Bleem et al. (2020). Follow-up of Atacama Cosmology Telescope galaxy cluster identiﬁcations (Hilton et al. 2021) yielded 67 new candidates including many in the DES footprint that are characterized here for the ﬁrst time.
Processed_The_DES_Bright_Arcs_Survey:_Candidate_Strongly_Len.txt	The DES data have proven to be a productive source of SL candidate systems. Searches for lensed quasars have yielded four spectroscopically conﬁrmed systems (Agnello et al. 2015; Ostrovski et al. 2017; Lin et al. 2017), two new quad SL systems and 96 more quasar lens candidates ripe for spectroscopic follow-up (Agnello & Spiniello 2019). Spectroscopic observations by the STRIDES collaboration conﬁrmed 10 new lensed quasars and 10 quasar pairs (Lemon et al. 2020). Visual scanning of 400, 000 color cutout images at coordinates selected, using techniques similar to those of this paper, from the ∼250 sq. deg. Science Veriﬁcation (SV) and ∼1,800 sq. deg. Y1 catalogs yielded 348 new SL candidates in Diehl et al. (2017), as well as one serendipitous SL discovery (Bettinelli et al. 2016). The convolutional neural network-based search technique described in Jacobs et al. (2017) contributed (Jacobs et al. 2019a; Jacobs et al. 2019b) 485 ranked candidates in the full ∼5,000 sq. deg. DES footprint, principally of the galaxy-galaxy lens conﬁguration. The latter paper also included 26 additional SL candidates identiﬁed by visual scanning of ∼54,000 potential lensing targets identiﬁed using procedures similar to those of Diehl et al. (2017). Spectroscopic conﬁrmation of 17 DES systems is described in Nord et al. (2016); Lin et al. (2017); Collett et al. (2017); Nord et al. (2020).
Processed_The_DES_Bright_Arcs_Survey:_Candidate_Strongly_Len.txt	The recent competitive success of automated and machine learning methods for identifying strong lens candidates is evident from a simple comparison of the numbers of candidates in these references. Neural network and ML algorithm searches for SL systems have advanced because of increased attention to the technique’s power and capabilities. Advances in computer vision have given rise to automated searches for strong lens candidates, and for classiﬁcation, deblending, and modelling of strong lens candidates (Alard 2006; Seidel & Bartelmann 2007; Kubo & Dell’Antonio 2008; Furlanetto et al. 2013a; Joseph et al. 2014; Paraﬁcz et al. 2016; Li et al. 2016; Xu et al. 2016; de Bom et al. 2017; Lanusse et al. 2018; Metcalf et al. 2019; Plazas 2020). Nonetheless, these ML searches continue to beneﬁt from visual scanning and ranking of candidate systems. Furthermore, continued eﬀorts to attain large samples of real lenses with a variety of morphologies is important for training and validation of new automated lens-ﬁnding algorithms, particularly for identiﬁcation of group and cluster-scale SL systems, where ML has been less successful.
Processed_The_DES_Bright_Arcs_Survey:_Candidate_Strongly_Len.txt	In this paper we report 247 strong gravitational lens candidates from searches of the full 5,000 sq. deg. DES footprint, including 81 previously unreported candidates. These lenses constitute a diverse set of primarily galaxy group- and cluster-scale systems, with both red and blue sources. We provide details of the locations (RA & DEC), and g,r,i,z,Y band magnitudes of source and lens objects, as well as Einstein radii and a rank, quantifying the conﬁdence level of the selection, for each SL candidate system. This paper is organized as follows. In §2 we describe the Dark Energy Survey observations and catalogs. In §3 we describe our strong gravitational lens search procedures. In §4 we describe the results from the searches and provide the properties of the candidate lens systems. We highlight some of the systems that have notable properties. Finally, in §5 we summarize and discuss our results.
Processed_The_DES_Bright_Arcs_Survey:_Candidate_Strongly_Len.txt	The Dark Energy Survey has completed a six season program that imaged 5,100 sq. deg. of the southern galactic cap using the Dark Energy Camera (DECam) (Flaugher et al. 2015), which is operated on the 4m Victor M. Blanco Telescope at Cerro Tololo Interamerican Observatory (CTIO) near La Serena, Chile. Following a “Science Veriﬁcation” period (Dark Energy Survey Collaboration et al. 2016), DES data collection typically occurred from mid-to-late August to mid-February, starting on August 31, 2013 and ending on January 9, 2019. Over the six year survey, we accumulated ten high-quality, 90-second duration observations of the full survey ﬁeld in each of the four ﬁlters g, r, i, and z-bands, and six to ten high-quality observations of 45 or 90 seconds length for the Y - band (totalling 450 seconds). In addition to the wide-ﬁeld survey, DES performed a time-domain supernova survey during the same time period, visiting ten ﬁelds in the g, r, i, and z-band ﬁlters with an approximately weekly cadence and at much greater depth (Kessler et al. 2015) than the wide-ﬁeld survey. Details of the observation strategy, survey operations, and the survey progress for each of the seasons are available (Neilsen et al. 2019; Diehl et al. 2016, 2018, 2019). Most of the searches described in this paper used as a starting point the object catalogs and images from the ﬁrst three seasons of observations. Much of the information that we provide about the candidate strong lens systems comes from the full six-season data set. When we refer to “Y1” (Drlica-Wagner et al. 2018), “Y3” (Abbott et al. 2018), and “Y6” (Dark Energy Survey Collaboration et al. 2021) data sets, we mean to include all the preceding seasons of observations. For instance, Y3 refers to the data from the ﬁrst three seasons.
Processed_The_Life_and_Times_of_the_Parkes-Tidbinbilla_Inter.txt	Abstract: The Parkes-Tidbinbilla took advantage of a real-time radio-link connecting the Parkes and Tidbinbilla antennas to form the world’s longest real-time interferometer. Built on a minus- cule budget, it was an extraordinarily successful instrument, generating some 24 journal papers including 3 Nature papers, as well as facilitating the early development of the Australia Telescope Compact Array. Here we describe its origins, construction, successes, and life cycle, and discuss the future use of single-baseline interferometers in the era of SKA and its pathﬁnders.
Processed_CMB_observations_using_the_SKA.txt	We examine the prospects for observations of CMB anisotropy with the SKA; we discuss the advantages of in- terferometric SKA imaging, observing strategies, calibration issues and the achievable sensitivity. Although the SKA will probably operate at cm wavelengths, where discrete source confusion dominates the CMB anisotropy, its extreme sensitivity to point sources will make it possible to subtract the source contamination at these wavelengths and thereby image the low surface brightness CMB anisotropies on small angular scales. The SKA, operating at 10-20 GHz, may usefully make high-l observations of the CMB anisotropy spectrum and survey the sky for Sunyaev-Zeldovich decrements.
Processed_CMB_observations_using_the_SKA.txt	Observations of the angular anisotropy power in the Cosmic Microwave Background (CMB) temperature ﬂuctu- ations on diﬀerent angular scales, leading to a measurement of the power spectrum of the anisotropy, has proved to be very useful in determining the parameters of the space-time structure of the Universe, its constituents, and understanding the ingredients and mechanism for large-scale structure growth and formation. Current goals for CMB research include extending the measurements of the anisotropy spectrum to small multipole orders, measuring the CMB polarization anisotropy, and surveying the sky for CMB decrements that have been created by cosmological clusters via the Sunyaev-Zeldovich eﬀect (SZE). The aims of these observations are to provide additional constraints on cosmological parameters, break parameter degeneracies, probe the cluster evolution and deriving the equation of state for the dark energy.
Processed_CMB_observations_using_the_SKA.txt	The CMB anisotropy observations at high multipole orders, of the primary anisotropy as well as of the SZE decrements towards cosmological clusters, are currently being pursued with radio interferometers and bolometric arrays at mm wavelengths. Today, bolometers have the advantage of wider bandwidths and better sensitiv- ity particularly at the mm wavelengths where most CMB telescopes choose to operate in order to avoid the contamination and confusion arising from the non-thermal discrete foreground sources in the sky. However, in- terferometers have much lower systematic errors and many ground based small angle anisotropy measurements are being done with purpose built interferometers operating at wavelengths around 10 mm. Because recent de- velopments in low noise ampliﬁers at mm wavelengths have made short wavelength interferometer sensitivities competitive, CMB interferometers operating at 3 mm are being planned.
Processed_CMB_observations_using_the_SKA.txt	A consortium of major radio astronomy institutions across the world is currently planning the world’s next generation large radio telescope, the Square Kilometre Array (SKA). The telescope will have a total collecting area of 106 m2 and will be capable of interferometric continuum imaging at metre and centimetre wavelengths down to at least 3 cm and perhaps lower. Although the SKA will operate at cm wavelengths, where foreground discrete source confusion limits the achieveable sensitivity for the imaging of the brightness temperature ﬂuctua- tions in the background CMB sky, its extreme sensitivity due to its large collecting area may make it possible to measure and subtract the point source contamination at these relatively low operating frequencies, and thereby image the low surface brightness CMB anisotropies on small angular scales. We examine the prospects for CMB observations with the SKA in the following sections.
Processed_CMB_observations_using_the_SKA.txt	Discrete synchrotron sources in the sky are a foreground ‘contaminant’ in images of CMB anisotropy. They may be assumed to be Poisson random distributed on the sky with possibly some small degree of clustering. The clustering is indeed tiny: to our knowledge, the NVSS was the ﬁrst radio survey to detect 2-point angular correlation between radio sources on the sky [1].
Processed_CMB_observations_using_the_SKA.txt	The mean spectral index α (Sν ∼ να) of extragalactic sources is about −0.7 at metre wavelengths and ﬂattens somewhat to about −0.5 at cm wavelengths: most discrete sources in the sky have lower ﬂux density at smaller wavelengths. Therefore, in images of the sky made with a constant telescope beam, the ﬂux density variations (in Jy beam−1) owing to discrete extragalactic sources are expected to be smaller at shorter (mm) wavelengths as compared to longer (cm) wavelengths. In contrast, at wavelengths longward of about 1 mm, in the Rayleigh- Jeans part of the CMB spectrum, the expected variance in ﬂux density owing to CMB temperature anisotropies in sky images made with a ﬁxed telescope beam increases towards shorter wavelengths as ∆Sν ∼ ν2. Searches for small angle anisotropy that have been done at cm wavelengths, like the 3.4 cm ATCA search for arcmin scale CMB anisotropy [2], were done in ﬁelds pre-selected to be relatively devoid of the stronger point sources; however, the sensitivity was limited by confusion due to weaker sources which could not be subtracted because of inadequate sensitivity to these point sources. For this reason, most planned CMB telescopes are being designed for operation at mm wavelengths to avoid the known extragalactic discrete source confusion that dominates the CMB anisotropy at cm wavelengths.
Processed_CMB_observations_using_the_SKA.txt	All surveys of the sky for discrete sources, which have covered signiﬁcant parts of the celestial sphere, have been made at metre and cm wavelengths. Our catalogs of sources detected at mm wavelengths have been made from observations of the mm ﬂux densities of sources detected in the long wavelength surveys. Our knowledge of the source counts in the mm sky is based on extrapolations of source populations identiﬁed at cm wavelengths and may miss sources that are bright at mm wavelengths and undetected at cm wavelengths. Recent discoveries of high frequency peakers [3] and sources which have spectral index α ∼ 2 at cm wavelengths [4] are indications that there may be surprises for us when mm source counts and populations are directly determined via mm surveys. To sumarize, the movement of CMB anisotropy search telescopes to short wavelengths certainly avoids the contamination from most known source populations; however, the mm observations may now have to confront new source populations and ill understood dust contamination. An alternate approach is to make small angle CMB surveys at cm wavelengths and overcome confusion from the well studied cm source populations via high sensitivity detections of these sources.
Processed_CMB_observations_using_the_SKA.txt	Any sky survey is made with a certain thermal noise sensitivity and a beam size. There is a threshold ﬂux density Sc at which the integral source count N (> S), at S = Sc and within the beam area, is unity. We expect one source on the average with ﬂux density exceeding Sc in any beam area. If a survey has a thermal noise that is less than or close to Sc, we would expect that sources in the sky which have a ﬂux density well exceeding Sc would stand out and could be reliably identiﬁed by the survey as ‘foreground sources’. These sources would occupy a small fraction of the image pixels (which I assume are roughly the size of the beam). If these identiﬁed sources are subtracted from the image, or if these few pixels that obviously contain sources are omitted from consideration, the weak sources with ﬂux density less than Sc will contribute an image rms of value approximately equal to the ﬂux density threshold Sc.
Processed_CMB_observations_using_the_SKA.txt	Sc and ∆T are, respectively, the confusion noise limit and corresponding brightness sensitivity limit of the sky survey.
Processed_CMB_observations_using_the_SKA.txt	The relatively short baselines of the SKA would be useful for imaging the CMB sky with high brightness sensitivity; we assume herein that CMB images are made with 1 arcmin resolution. At 10 GHz, these images would have a confusion noise of 8 µJy. The corresponding brightness sensitivity limit is 30 µK.
Processed_CMB_observations_using_the_SKA.txt	The CMB angular power spectrum is expected to have a band power of pl(l + 1)Cl/(2π) ∼ 6 µK on arcmin scales owing to the SZE in cosmological clusters. A SZE survey for distant clusters with the SKA requires lowering the confusion noise via the subtraction of weak sources in every beam area.
Processed_CMB_observations_using_the_SKA.txt	The proposed SKA is to have a sensitivity: Aef f /Tsys = 2 × 104 m2 K−1; we assume a 10 percent observing bandwidth. The entire collecting area (all baselines) would potentially be useful for detecting the discrete source confusion in the ﬁelds. We may, therefore, expect the high resolution continuum images made using the full array to have a thermal noise of about 50 nJy with 1 hr integration time at 10 GHz. Placing a 5-σ threshold for the reliable detection of point soucres in the ﬁeld, we may expect sources with ﬂux density exceeding about 250 nJy to be subtracted. There would be, on the average, about 102 such sources in every arcmin area at 10 GHz.
Processed_CMB_observations_using_the_SKA.txt	Assuming that foreground sources above Sm ∼ 250 nJy are successfully subtracted, the residual confusion in arcmin resolution images may be expected to be as low as 6 µK at 10 GHz.
Processed_CMB_observations_using_the_SKA.txt	It appears that the SKA could usefully image the background CMB at a frequency >∼10 GHz, with arcmin resolution, and with a confusion noise limit less than 6 µK. If the thermal noise is to be less than the residual confusion noise, say a tenth of the confusion noise, a quarter of the SKA baselines are to be useful in making the arcmin scale CMB survey images.
Processed_CMB_observations_using_the_SKA.txt	Interferometers are the instruments of choice for precision measurements of extremely weak sky signals like the CMB anisotropies because they have much lower systematic errors. The SKA, being an interferometric imaging telescope, would have substantial advantages of lower systematics. The lower element sensitivity, owing to the smaller bandwidths at the longer operating wavelengths, would be oﬀset by the sensitivity gain obtained by the extremely large number of antenna and receiver elements. Additionally, at the longer cm wavelengths, system temperatures are lower owing to lower receiver noise temperatures, lower losses in signal transmission, lower atmospheric emission temperature.
Processed_CMB_observations_using_the_SKA.txt	The inherent stability of interferometers implies that highly precise calibration, including polarization calibra- tion, is possible. In some of the SKA conﬁguration solutions, the stations have a large number (n) of elements in compact arrays. These arrays have of order n2 longer baselines that will resolve the CMB but will measure the discrete sources with high accuracy. Since the same elements measure the long and short baselines simul- taneously, the subtraction of self calibrated intensity and polarization should be accurate to better than 1 part in 1,000.
Processed_CMB_observations_using_the_SKA.txt	Finally each SKA patch can make an independent observation, and these could either be used to decrease noise in one patch of sky, or to observe patches over a larger area to increase the survey area, in searches for cosmological clusters, or to reduce cosmic variance, in statistical measurements of CMB anisotropy.
Processed_The_Static_Baryon_Potential.txt	Using state of the art lattice techniques we investigate the static baryon potential. We employ the multi-hit procedure for the time links and a variational approach to determine the ground state with suﬃcient accuracy that, for distances up to ∼ 1.2 fm, we can distinguish the Y - and ∆- Ans¨atze for the baryonic Wilson area law. Our analysis shows that the ∆-Ansatz is favoured. This result is also supported by the gauge-invariant nucleon wave function which we measure for the ﬁrst time.
Processed_The_Static_Baryon_Potential.txt	Due to the important role which the q ¯q poten- tial plays in our understanding of the structure of mesons, there exist numerous studies of this quantity on the lattice. For a review see ref. [1]. The three quark potential plays an equally im- portant role in the understanding of baryon struc- ture. Mass relations between baryons and mesons can be made more exact if the form of the bary- onic potential is known [2]. However, very few lattice studies have been made of the baryonic potential. Moreover, two such recent studies [1,3] have reached diﬀerent conclusions for the area law behaviour of the baryonic Wilson loop: they give support to two diﬀerent Ans¨atze, called Y - and ∆-law. Since the maximal diﬀerence between these is a mere 15% for SU (3), a reliable extrac- tion of the ground state as well as noise reduc- tion techniques are essential in order to resolve the dominant area law behaviour [4].
Processed_The_Static_Baryon_Potential.txt	The issue of the dominant area law behaviour arises for any gauge group SU (N ). Therefore it is also interesting to study SU (4) and test whether the SU(4) results corroborate the conclu- sions reached in SU(3). As for SU (3), we choose lattice geometries which maximize the diﬀerence between the two Ans¨atze which for SU (4) is at the 20% level.
Processed_The_Static_Baryon_Potential.txt	As yet a further check, we evaluate the gauge- invariant nucleon wave function, using a den- sity insertion for each quark line, and examine whether it is better described according to the Y - or ∆- law [5].
Processed_The_Static_Baryon_Potential.txt	P denotes path ordering and Γ(j) is the path from x to y for quark line j.
Processed_The_Static_Baryon_Potential.txt	Figure 1. The baryonic Wilson loop in SU (3). The quarks are located at positions r1, r2 and r3.
Processed_The_Static_Baryon_Potential.txt	The Y -Ansatz: In the strong coupling limit, minimization of the static energy amounts to giving the shortest length, LY , to the ﬂux tubes join- ing the quarks. For SU (3), this is realized in general if the three ﬂux tubes meet at an interior point [6], known as the Steiner point, where their mutual angles are 1200. [If one of the angles of the triangle formed by the three quarks exceeds 1200, the ﬂux tube coming from that summit has length zero, and the other two ﬂux tubes meet there.] Time evolution of this state in the general case produces a three-bladed area similar to Fig.1, known as the Y - area law.
Processed_The_Static_Baryon_Potential.txt	Figure 2. The ﬂux tubes joining four quarks. The quarks are located at positions r1, r2, r3 and r4. The upper graph shows the local minimum of the energy with one Steiner point A, and the lower is the minimum with two Steiner points A and B.
Processed_The_Static_Baryon_Potential.txt	single Steiner point (X-Ansatz), and one with two Steiner points A and B (Y - Ansatz). The double string between the two Steiner points has tension 1.357(29) times greater[7] than the other four, single strings. If we neglect this diﬀerence for simplicity, then the Y -Ansatz always has lower energy than the X-Ansatz. Here, we make no at- tempt to distinguish between the Y - and X- Ans¨atze, and we assume that the dou- ble string has the same tension as the sin- gle strings. Since this assumption has the eﬀect of reducing the potential of the Y - Ansatz, which is itself lower than in the X-Ansatz, it turns out to have no bearing on our conclusions. In contrast to SU (3) where for any given location of the three quarks, the Steiner point and therefore the Y -Ansatz energy can be computed analyt- ically, in SU (4), the two Steiner points in the Y -Ansatz are obtained by an iterative numerical procedure.
Processed_The_Static_Baryon_Potential.txt	quark lines i and j. Because of its shape in SU (3), this Ansatz is known as the ∆- area law. Therefore, we denote by L∆ the total length of all interquark distances.
Processed_The_Static_Baryon_Potential.txt	For SU (3) the maximal diﬀerence of 15% be- tween the two proposed area laws is obtained when the 3 quarks form an equilateral triangle.
Processed_The_Static_Baryon_Potential.txt	Figure 3. The geometries considered in SU (4). For geometry 1 the quarks are placed on a plane at equal distances from the origin; for geometry 2 the positions of the four quarks are along the three axes at (l,0,0), (0,l,0) (0,0,l), (0,0,-l); and for geometry 3 at (0,0,0), (l,0,0), (0,l,0), (0,0,l).
Processed_The_Static_Baryon_Potential.txt	with CF = (N 2 1)/2N and σ the string ten- sion of the q ¯q potential. Note that in contrast to ref. [3] we do not allow σ to vary. The factor of 1/(N 1) < LY always. In addition, we directly compare the three- or four-quark potential with the sum of two-body potentials measured on the same gauge conﬁgurations, with no adjustable parameters.
Processed_The_Static_Baryon_Potential.txt	I.Multi-hit procedure: We carried out a com- parison between the multi-hit procedure and hypercubic-blocking, as proposed in ref. [9], for In brief, hypercubic blocking the time links. smears the link using staples which all belong to the 24 hypercube surrounding it. Compared with the multi-hit procedure, we ﬁnd that hypercubic blocking tends to give larger errors, especially for the large Wilson loops. Since accuracy at large distances is our objective, we have adopted the multi-hit procedure. Moreover, for SU (3) the group integral needed to obtain the mean value of the link can be computed exactly [10]. We note that the reduction factor in the statistical error from applying the multi-hit procedure to Wilson loops of time extent t grows exponentially with t, and that the exponent is further multiplied by 3/2 in the case of baryonic loops.
Processed_The_Static_Baryon_Potential.txt	used as a check in the extraction of the ground state, ensuring that the contamination is less than e−2 in the values considered for the plateau.
Processed_The_Static_Baryon_Potential.txt	For the baryonic loop in SU (3) we used 220 conﬁgurations at β = 5.8 and 200 at β = 6.0 for a lattice of size 163 32 from the NERSC archive. For SU (4) we generated 100 conﬁgurations at β = 10.9 which gives a similar string tension σa2 as for SU (3) at β = 6.0.
Processed_The_Static_Baryon_Potential.txt	Figure 4. The static SU (3) baryonic potential at β = 5.8 (ﬁlled circles). The crosses show the sum of the static q ¯q potentials. The curves for the ∆ and Y Ans¨atze are also displayed. The quarks are located at (l, 0, 0), (0, l, 0), (0, 0, l) and r = r12 = r13 = r23 = l√2 .
Processed_The_Static_Baryon_Potential.txt	Figure 6. The static SU (3) baryonic potential at β = 5.8 for this work (ﬁlled circles) and for ref. [12] (crosses) versus the perimeter L∆ of the triangle formed by the quarks.
Processed_The_Static_Baryon_Potential.txt	In Fig. 6 we compare our results with those obtained in ref. [12]. As it can be seen the two sets of data are in agreement for small loops which correspond to the bulk of the data of ref. [12]. For larger loops, the few results of ref. [12] tend to lie above ours. The statistical errors on these data are not quoted but we expect them to be larger than ours, especially since the multi-hit procedure was not used. The analysis of ref. [12] diﬀers from ours in that we do not allow the string tension to vary but take it from the ﬁt to the q ¯q potential. So in our approach having ﬁxed the q ¯q potential there are no adjustable parameters that enter in the two Ans¨atze.
Processed_The_Static_Baryon_Potential.txt	Figure 7. Upper graph: The nucleon wave func- tion versus LY . Lower graph: The nucleon wave function versus L∆/2, ﬁtted by a linear potential ansatz. All quantities are in lattice units.
Processed_The_Static_Baryon_Potential.txt	can be seen, this simple asymptotic form pro- vides a remarkably good description of the nu- cleon wave function.
Processed_The_Static_Baryon_Potential.txt	Finally we display the results obtained in SU (4) for geometry 3. They give yet more sup- port to the ∆-area law for the baryon Wilson loop. The other two geometries show the same behaviour.
Processed_The_Static_Baryon_Potential.txt	of a many-body component. Nevertheless, for the distances up to 1.2 fm that we were able to probe in this work, the ∆-area law gives the clos- est description of our data. More reﬁned noise- reduction techniques for the large loops will be needed in order to clarify or rule out a many- body component at larger distances. Preliminary results on the nucleon wavefunction also support a potential in accord with the ∆-Ansatz.
Processed_Optimising_MWA_EoR_data_processing_for_improved_21.txt	Abstract The redshifted cosmological 21 cm signal emitted by neutral hydrogen during the ﬁrst billion years of the universe is much fainter relative to other galactic and extragalactic radio emissions, posing a great challenge towards detection of the signal. Therefore, precise instrumental calibration is a vital prerequisite for the success of radio interferometers such as the Murchison Wideﬁeld Array (MWA), which aim for a 21 cm detection. Over the previous years, novel calibration techniques targeting the power spectrum paradigm of EoR science have been actively researched and where possible implemented. Some of these improvements, for the MWA, include the accuracy of sky models used in calibration and the treatment of ionospheric eﬀects, both of which introduce unwanted contamination to the EoR window. Despite sophisticated non-traditional calibration algorithms being continuously developed over the years to incorporate these methods, the large datasets needed for EoR measurements requires high computational costs, leading to trade-oﬀs that impede making use of these new tools to maximum beneﬁt. Using recently acquired computation resources for the MWA, we test the full capabilities of the state-of-the-art calibration techniques available for the MWA EoR project, with a focus on both direction dependent and direction independent calibration. Speciﬁcally, we investigate improvements that can be made in the vital calibration stages of sky modelling, ionospheric correction, and compact source foreground subtraction as applied in the hybrid foreground mitigation approach (one that combines both foreground subtraction and avoidance). Additionally, we investigate a method of ionospheric correction using interpolated ionospheric phase screens and assess its performance in the power spectrum space. Overall, we identify a reﬁned RTS calibration conﬁguration that leads to an at least 2 factor reduction of the EoR window power contamination at the 0.1 hMpc−1 scale. The improvement marks a step further towards detecting the 21 cm signal using the MWA and the forthcoming SKA low telescope.
Processed_Optimising_MWA_EoR_data_processing_for_improved_21.txt	galaxies, improving our knowledge on the evolution of the universe as a whole (see reviews in, Fan et al. 2006; Furlanetto et al. 2006; Pritchard & Loeb 2012; Zaroubi 2013).
Processed_Optimising_MWA_EoR_data_processing_for_improved_21.txt	Current interferometers lack the sensitivity required to directly detect the cosmological signal, estimated to be a few tens of milliKelvin in brightness temperature. A more feasible pursuit has been targeting a statistical detection of the signal, with most eﬀorts going towards measuring its power spectrum (PS). Further, radiation from emitters (foregrounds) that are 2 to 3 orders of magnitude brighter, make detection of the signal an up- hill task. Deriving ways of mitigating these foregrounds has remained a key area in observational EoR research. Most current foreground mitigation methods hinge on the contrast between the spectral signature of the foregrounds and the 21 cm signal. In contrast to the foregrounds, which are mostly spectrally smooth, the 21 cm signal varies rapidly over frequency. Therefore, in Fourier (k) space, the 21 cm signal and the fore- grounds occupy separate distinguishable k modes. The 21 cm signal dominates the foreground-free modes and in principle, with enough sensitivity, is detectable. The MWA foregrounds mitigation strategy is a hybrid be- tween the foreground avoidance strategy and foreground subtraction, the two main methods employed in the ﬁeld. In foregrounds avoidance, foreground-dominated modes (the wedge feature in 2D PS, Datta et al. 2010; Par- sons et al. 2012; Vedantham et al. 2012; Morales et al. 2012) are discarded. On the other hand, foreground sub- traction attempts to directly remove a model of the unwanted sky from the data by use of either parametric or non-parametric methods, leaving the cosmological signal behind. So far, speciﬁcally for the MWA, fore- ground subtraction alone has not been successful in helping recover the foreground dominated modes. How- ever, Kerrigan et al. 2018 showed that such a hybrid method can result in improved results for the modes in the EoR window. In this hybrid mitigation method, the main beneﬁt of foregrounds subtraction is realised from the fact that by subtracting power from the wedge, we are by extension reducing the power that can leak to the window and therefore keeping the window as con- tamination free as possible for a 21 cm detection (Liu & Shaw, 2020). Foregrounds necessitate precise instru- mental calibration, with the success of any foreground mitigation strategy relying on a calibration that achieves a challenging dynamic range of ∼ 105.
Processed_Optimising_MWA_EoR_data_processing_for_improved_21.txt	solutions that contaminate the EoR window and could inhibit a 21 cm detection. Errors in positions of the sources themselves will also introduce unwanted calibra- tion errors and as a result, custom source catalogues for enhanced accuracy during MWA sky modelling are used (Line et al., 2017). There are also continuous ef- forts to include a diﬀuse emission component in EoR sky models (Byrne et al., 2021), improved radio frequency interference (RFI) mitigation (Wilensky et al., 2019), as well as improving the accuracy of the instrumental beam response (Sokolowski et al., 2017).
Processed_Optimising_MWA_EoR_data_processing_for_improved_21.txt	We focus on the ionosphere as a cause of calibration er- rors ﬁrstly because for the MWA, ionospheric refraction, which results in positional oﬀsets of compact sources, is a dominant eﬀect. The ionosphere manifests as a direction dependent eﬀect based on the MWA array properties; see Lonsdale (2005). MWA PS measurements usually exclude data observed in durations with high ionospheric activity Trott et al. (2018, 2020); Rahimi et al. (2021); Yoshiura et al. (2021). Secondly, the Real-Time System (RTS, Mitchell et al. 2008), the calibration algorithm used in this work, is designed to perform a thorough direction dependent calibration, focusing almost entirely on the ionosphere. Since the primary ionospheric eﬀects are expected to manifest in the foreground dominated k modes, using simulations, this paper shows how iono- spheric refraction leads to contamination in k modes that otherwise should be clean and also test whether increased sky model sources combined with aggressive foregrounds subtraction helps in reducing the contami- nation. The primary eﬀects due to the ionosphere are expected to manifest in the foreground dominated k modes. Using simulations, this paper shows how iono- spheric refraction leads to ionospheric contamination leaking into k modes that otherwise should be clean. In addition, we test whether increased sky model sources applied together with aggressive foregrounds subtraction helps in reducing the contamination.
Processed_Optimising_MWA_EoR_data_processing_for_improved_21.txt	Despite ongoing development of EoR-speciﬁc calibra- tion algorithms that implement the aforementioned new non-traditional calibration routines that would poten- tially improve the MWA PS measurements, a full imple- mentation and testing of these methods has not been feasible over the previous years. The task involves high computational costs, exacerbated by the need for EoR studies to average over potentially thousands of hours of data to reach the required sensitivity.
Processed_Optimising_MWA_EoR_data_processing_for_improved_21.txt	In 2020, the MWA acquired g a r r awa r l a, a super- computer dedicated to MWA data processing hosted by the Pawsey Supercomputing Centre. With this recent signiﬁcant upgrade in our processing capabilities, we are able to perform these tests and enhance our calibration knowledge, a big motivation for this work. This work will feed into other ongoing software development eﬀorts; more eﬃcient and faster tools targeting the MWA and the forthcoming Square Kilometre Array (SKA).
Processed_Optimising_MWA_EoR_data_processing_for_improved_21.txt	The main objective of this work is to identify the most accurate calibration routine for MWA EoR that fully utilises the capabilities of the currently available calibration tools, speciﬁcally the RTS, with a focus on ionospheric correction, foregrounds subtraction and sky model completeness.
Processed_Optimising_MWA_EoR_data_processing_for_improved_21.txt	This paper is organised as follows: In Section 2, we describe the methods applied and the data used. This involves our simulated models for the ionosphere and the real data used, as well as the calibration routine and PS estimation applied in this work. We then introduce the MWA real observations dataset and the validation procedure adopted. Section 4 presents the results, and Section 7 investigates more methods of ionospheric mod- elling. A discussion and conclusion of the work is given in Section 8.

Subsets and Splits

No community queries yet

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