The TRAPUM Small Magellanic Cloud pulsar survey with MeerKAT: I. Discovery of seven new pulsars and two Pulsar Wind Nebula associations

The sensitivity of the MeerKAT radio interferometer is an opportunity to probe deeper into the population of rare and faint extragalactic pulsars. The TRAPUM (TRAnsients and PUlsars with MeerKAT) collaboration has conducted a radio-domain search for accelerated pulsars and transients in the Small Magellanic Cloud (SMC). This partially targeted survey, performed at L-band (856-1712 MHz) with the core array of the MeerKAT telescope in 2-h integrations, is twice as sensitive as the latest SMC radio pulsar survey. We report the discovery of seven new SMC pulsars, doubling this galaxy's radio pulsar population and increasing the total extragalactic population by nearly a quarter. We also carried out a search for accelerated millisecond pulsars in the SMC Globular Cluster NGC 121 using the full array of MeerKAT. This improved the previous upper limit on pulsed radio emission from this cluster by a factor of six. Our discoveries reveal the first radio pulsar-PWN systems in the SMC, with only one such system previously known outside our galaxy (the"Crab pulsar twin"in the Large Magellanic Cloud, PSR J0540$-$6919). We associate the 59 ms pulsar discovery PSR J0040$-$7337, now the fastest spinning radio pulsar in the SMC, with the bow-shock Pulsar Wind Nebula (PWN) of Supernova Remnant DEM S5. We also present a new young pulsar with a 79 ms period, PSR J0048$-$7317, in a PWN recently discovered in a MeerKAT radio continuum image. Using the multi-beam capability of MeerKAT, we localised our pulsar discoveries, and two previous Murriyang discoveries, to a positional uncertainty of a few arcseconds.


INTRODUCTION
The Small and Large Magellanic Clouds (SMC and LMC) are the only galaxies outside our own in which radio pulsars have been discovered to date.They are nearby galaxies that are unobstructed by the Milky Way's (MW) galactic plane and therefore they are good targets for extragalactic pulsar searches.Indeed, the Small Magellanic Cloud is just 60 kpc away (Karachentsev et al. 2004), and the expected Milky Way Dispersion Measure (DM) contribution in its direction ★ E-mail: emma.carli@outlook.com is low: about 30 pc cm −3 according to the YMW16 electron density model (Yao et al. 2017), and 42 pc cm −3 according to the NE2001 model (Cordes & Lazio 2002).
The SMC has a lower metallicity than our galaxy ( SMC ≃ 0.2  MW , Luck et al. 1998), which can lead to an increased Neutron Star (NS) birth rate (Heger et al. 2003).Indeed, the latest surveys confirm that the SMC has a much higher density of High-Mass X-ray Binaries (HMXBs) than the Milky Way (Haberl & Sturm 2016).The SMC has also undergone recent episodes of star formation (Harris & Zaritsky 2004), which should allow for the detection of a higher proportion of short-lived and young objects than in the Milky Way's older population.Together, these galactic properties make the SMC an excellent target to search for pulsars, particularly in rare young systems.
Of the nearly 3400 radio pulsars discovered, only 31 are extragalactic (see the ATNF pulsar catalogue for a list, Manchester et al. 2005).Their radio emissions were discovered using Murriyang, the Parkes 64 m radio telescope.Of these discoveries, seven are in the Small Magellanic Cloud, discovered by McConnell et al. 1991 (1 pulsar), Crawford et al. 2001 (1), Manchester et al. 2006 (3), and Titus et al. 2019 (2).Their spin periods range from 191 to 937 ms, their characteristic ages span 1-4 Myr and their median DM is approximately 115 pc cm −3 .
In addition, two young (not accretion-powered) pulsars have been discovered, via their X-ray pulsations only, in the SMC: the magnetar CXOU J010043.1−721134 by Lamb et al. 2002 (recently associated with a new supernova remnant by Cotton et al. 2024) and the rotationpowered pulsar J0058−7218 by Maitra et al. 2021 (see also Carli et al. 2022).The latter is embedded in a Pulsar Wind Nebula (PWN) (Maitra et al. 2015) in the supernova remnant (SNR) IKT 16 (Inoue et al. 1983;Mathewson et al. 1984;Owen et al. 2011).A second PWN in the SMC was discovered by Alsaberi et al. (2019) in the SNR DEM S 5 (Haberl et al. 2000;Payne et al. 2007) but no pulsar was detected.There are no other SMC pulsars known to be associated with supernova remnants so far.None of the SMC's known radio pulsars is young: their characteristic ages are all larger than 1 Myr.There are no known extragalactic millisecond pulsars (old pulsars which have been spun up to millisecond rotation periods by accretion from a companion).The SMC hosts the only discovered extragalactic binary pulsar: PSR J0045−7319, an ordinary pulsar with a massive main sequence B star companion (Kaspi et al. 1994).
The larger population of the LMC shows more extensively the wealth of exotic neutron star systems that can be found outside our galaxy.It includes the 'Crab pulsar twin' PSR J0540−6919 in the PWN of SNR 0540−69.3(Manchester et al. 1993b;Gotthelf & Wang 2000;Brantseg et al. 2013), the only known extragalactic multiwavelength pulsar, discovered in the X-ray band by Seward et al. (1984), then via optical pulses by Middleditch & Pennypacker (1985), then via radio pulsations by Manchester et al. (1993a), later seen to emit giant radio pulses (Johnston & Romani 2003), and finally in gamma-rays by Marshall et al. (2016).Along with PSR J0534−6703 (Manchester et al. 2006), a long period (≃1.8 s) radio pulsar with no SNR or PWN association and no high-energy emission, these are the only two known young (<100 kyr characteristic age) radio pulsars in the LMC, and thus outside our galaxy.There is one additional young protation-powered ulsar in the LMC, PSR J0537−6910, in a PWN in the SNR N157B (Wang et al. 2001;Wang & Gotthelf 1998).It is only detected via its X-ray pulsations and is the fastest spinning (16.1 ms) young pulsar known (Marshall et al. 1998).The LMC also hosts one magnetar, SGR 0526−66 (Helfand & Long 1979;Kulkarni et al. 2003), possibly associated with the SNR N 49 (e.g.Gaensler et al. 2001).Finally, PSR J0529−6652 is also known to emit giant pulses (Crawford et al. 2013).There are a further three known PWNe in the LMC with no pulsar associations (Gaensler et al. 2003;Williams et al. 2005;Bamba et al. 2006;Haberl et al. 2012).
Thus, despite only a very small sample of pulsars outside our galaxy being known, many systems are showing characteristics that are relatively rare in the Milky Way population, and are prized for a variety of astrophysical studies.TRAPUM 1 (TRAnsients and PUlsars with MeerKAT) is a Large Survey Project of the MeerKAT telescope (Stappers & Kramer 2016), and one of its core science goals 1 trapum.org is to increase the known population of the rare extragalactic neutron stars.The collaboration has already discovered over 200 pulsars2 in the Milky Way's galactic plane (Padmanabh et al. 2023;Colom i Bernadich et al. 2023), gamma-ray sources (Clark et al. 2023;Dodge et al. 2024), and Globular Clusters (Ridolfi et al. 2021;Abbate 2023;Vleeschower et al. 2022;Zhang et al. 2022;Ridolfi et al. 2022;Douglas et al. 2022;Barr et al. 2024).We have conducted the first survey of the SMC with the MeerKAT telescope.This TRAPUM survey searches for ordinary and young pulsars, single pulses, Fast Radio Bursts (FRBs), binary systems and millisecond pulsars in a variety of environments.
We begin this paper by presenting the novel observing (subsection 2.1) and processing (subsection 2.2) techniques of this survey.The resulting survey sensitivity is compared to other surveys in subsection 2.3.The different observing and processing methods used for the SMC Globular Cluster NGC 121 are reported in section 3. Our discoveries are described in section 4. A summary of this work, section 5, concludes this paper.The timing solutions and peculiarities of our discoveries will be shown in Paper II of this series.The discoveries will enable an analysis of the impact of the low-metallicity, recent star formation environment of the SMC on its neutron star population and comparisons with predictions (e.g.Titus et al. 2020).We will present this population study in Paper III of this series.

Introduction
Historically, pulsar searching has been dominated by large single-dish surveys 3 , and in the Southern Hemisphere by Murriyang, the Parkes 64-m radio-telescope in Australia (hereafter Murriyang).Multi-beam techniques were introduced for Murriyang by Staveley-Smith et al. (1996) and Crawford et al. (2001) conducted the first survey of the Magellanic Clouds that benefited from this increased field of view.This allowed for greater integration times per pointing, improving the sensitivity of surveys.To further increase the observing sensitivity and find more of the faint extragalactic pulsars, one can either turn to very large unsteerable aperture dishes (e.g. the FAST telescope, Nan et al. 2011), or to a combined array of smaller dishes: an interferometer.
The MeerKAT telescope (Jonas 2016;Camilo 2018), inaugurated in 2018, is of the latter type.It is a precursor to the mid-frequency component of the Square Kilometer Array, situated in the Karoo desert in South Africa.It consists of 64 13.5-m Gregorian antennas, 44 of which are arranged in a core of 1 km diameter.The maximum baseline is 8 km.The full coherent array of MeerKAT has a gain about four times that of Murriyang.This is a major sensitivity boost for finding faint or distant pulsars in the Southern Hemisphere, notably extragalactic pulsars.MeerKAT operates in three observing bands.Our observations were performed at L-band (see Table 1 for the specifications).The UHF band of MeerKAT is also available but was not used due to data rate and processing time considerations.
This survey uses the Filterbanking Beamformer User Supplied Equipment (FBFUSE), a MeerKAT backend specific to pulsar and transient search users designed by the Max Planck Institute for Radio Astronomy (MPIfR).The beamformer can generate a chosen number  (Lehmensiek & Theron 2012;Bailes et al. 2020).

Parameter Value
Bandwidth (MHz) Δ 856 Centre frequency  (MHz) 1284 System temperature  sys (K) 18 Gain  (64 coherent antennas or 'full array', K Jy −1 ) 2.8 Gain  (44 coherent antennas or 'core array', K Jy −1 ) 1.925 Gain  (62 incoherent antennas, K Jy −1 ) 0.344 Number of polarisations recorded  pol 2 of coherent 'tied-array' beams on the sky by varying delays calculated by katpoint4 between antennas before coherently combining their signals.The Point Spread Function (PSF), orientation and optimal tiling of the coherent beams (so that they overlap at a chosen fractional sensitivity level of the PSF) is simulated by the mosaic software suite (Chen et al. 2021).To increase the coherent beam size and therefore the field of view covered by the coherent beam tiling in each pointing of this survey, we use the core array configuration of MeerKAT (40 or 445 antennas out of 64) with a maximum baseline length of about 1 km.This reduction from the full array's 8 km maximum baseline strikes a good balance between beam area and sensitivity (Chen et al. 2021).The gain is reduced to just under three times that of Murriyang.We also observe the circumpolar SMC around the time it is furthest from the observer's Meridian, in the intermediate elevation range, to maximise beam size (Chen et al. 2021).With this setup, the area of a single coherent beam inside its 50 per cent sensitivity contour at the central frequency of L-band is around 1 square arcminute.This is about 150 times smaller than Murriyang's single beam area.This means that a much larger number of beams must be recorded than in the Murriyang surveys to cover the galaxy's area.However, when a pulsar is detected in a MeerKAT coherent beam, its precise position is immediately pinpointed to a beam size error of the order of an arcminute.It is even possible to constrain the pulsar location to a few arcseconds by combining detections in multiple MeerKAT beams (Bezuidenhout et al. 2023), which is described in subsubsection 2.2.4.In comparison, the Murriyang telescope has a position error on discovered pulsars of 14 ′ diameter, the Full Width at Half Maximum (FWHM) of its beam.The Parkes Multi Beam (PMB, Staveley-Smith et al. 1996) receiver of Murriyang can observe 13 simultaneous beams, covering 154 square arcminutes each with a sensitivity above 50 per cent of the maximum, or around 0.56 square degrees per pointing.The area covered with above 50 per cent of the maximum sensitivity by a 768 MeerKAT core array coherent beam tiling is only about 768 square arcminutes or 2.6 times less than a PMB pointing.
There is also an 'incoherent' or 'primary' beam obtained from summing all the antenna signals with no delay corrections.It has a FWHM of 1.1 degrees at the centre frequency of L-band (Asad et al. 2021).At the lowest frequency of the band, the incoherent beam has a FWHM of 1.7 degrees, and at the highest, a FHWM of 0.9 degrees.The incoherent beam's gain is half that of Murriyang and covers 0.95 square degrees with over 50 per cent sensitivity at the central frequency of L-band, nearly double the area of the PMB.Coherent beams can be placed anywhere within the MeerKAT incoherent beam, but their sensitivity is governed by the incoherent beam response.The gains of the different configurations of MeerKAT are provided in Table 1.Beam tilings for each of our observations are shown in Appendix B.
The filterbank data generated by FBFUSE are processed locally (subsection 2.2), in the MPIfR APSUSE computing cluster (Accelerated Pulsar Searching User-Supplied Equipment) located in the Karoo Array Processor Building.The cluster contains 60 nodes, each with 24 CPUs, two GPUs and 97 GB of RAM.APSUSE pipeline processing is performed using MPIfR pipelines6 running on 56 nodes, while the remaining four are used for manual processing tasks and testing.Observations, processing and data products are tracked using a MPIfR database7 .The pulsar searching backends of MeerKAT are presented in further detail in Barr (2017), Padmanabh (2021) and Padmanabh et al. (2023).

Survey pointings and targets
To maximise the number of beams available and thus the field of view observed in each pointing, we must reduce the data rate per beam.This decreases the amount of data that needs to be stored and processed as well as the APSUSE filterbank recording rate.We have thus opted to use 2048 frequency channels (a higher resolution 4096-channel mode is available), and a sampling time of 153 μs, rather than the minimum available 76 μs.This enabled us to record 768 simultaneous coherent beams8 and one incoherent beam, with a total data rate of 10 gigabytes per second.We observed each pointing for 2 hours with 8-bit sampling.One pointing in the survey needs 74 terabytes of APSUSE disk space.The characteristics of the observations of each pointing of the survey are given in Table 2.We observed each pointing twice to allow for candidate pulsar confirmations, except for pointings 5 and 7, which second passes we replaced by an additional pointing to the survey (SMCPOINTING9, observed once) and an observation of the SMC Globular Cluster NGC 121 (see section 3).
Pointing centres were picked to loosely follow the High Mass Xray Binary distribution of the SMC (Haberl & Sturm 2016) to favour areas with high neutron star density.Furthermore, the pointing positions were optimised to include as many targets that could host pulsars as possible within the incoherent beam (IB) area with over 50 per cent of the maximum sensitivity.For each pointing, coherent beams were tiled on relevant targets, overlapping at a chosen sensitivity level.Any remaining beams were tiled at boresight, where sensitivity is highest.
Table 2.The SMC survey observation parameters.A high resolution coherent beam PSF was simulated with the latest version of mosaic at the centre coordinates of the pointing (Right Ascension and Declination), at the central time of the observation, with the antennas used during the observation and at the central frequency of L-band.The coherent beam ellipse semi-axes are given in arcseconds in the Right Ascension and Declination directions, with the ellipse position angle.The observation parameters of the observation of the SMC Globular Cluster NGC 121 are given in Table 6.We opted to tile most of our targeted beams on and around 9 the supernova remnants of the Small Magellanic Cloud to find associated young pulsars.We observed all the SNRs and SNR candidates in the multiwavelength census by Maggi et al. (2019).A few additional SNR candidates were also obtained from Titus et al. (2019) and their observation proposal.A complete list of the observed SNRs is given in subsection 4.3.The new SNRs identified in the SMC by Cotton et al. (2024) were not published at the time of observing.For the SNRs IKT 16 (Maitra et al. 2015;Carli et al. 2022) and DEM S5 (Alsaberi et al. 2019), a single beam was placed on the associated PWN.Our target list also comprises five out of the seven known radio pulsars in the SMC, for localisation (see subsection 4.4) or processing validation purposes, and the two known SMC X-ray pulsars that are not accretion-powered.There are no confirmed Central Compact Objects 10 (CCOs) in the Magellanic Clouds (e.g.Long et al. 2020), so none was targeted by our tilings.In later pointings of the survey, we placed a beam on HMXBs with periods under 10 s (Haberl & Sturm (2016), Bartlett et al. (2017), Vasilopoulos et al. (2020), see Table 3), as it is our longest period searched (see subsubsection 2.2.1).We also targeted most steep spectrum continuum point sources (from the Joseph et al. (2019) ASKAP EMU SMC image, see Table A1) that fell within the IB FWHM and were not within planned tilings in those later pointings.Earlier in the survey, some beams were overlapping with the position of more of these fast-spinning HMXBs and steep spectrum point sources (see Appendix B).Finally, we tiled star clusters with ages over 5 Gyr near our pointings (Bica et al. 2020): Lindsay 110 in both passes of pointing 8 (though it was outside of the IB FWHM), and NGC 416 in the second pass of pointing 5. NGC 361 was omitted due to being in a lower sensitivity area of a pointing.An observation of the SMC's only Globular Cluster, NGC 121, was conducted and processed with a different setup detailed in section 3. The beams recorded for each observation are shown in Appendix B. Sensitivity limits for various targets are given in subsection 4.3.

Periodicity searches
We first processed the survey data with a periodicity search pipeline.Radio Frequency Interference (RFI) cleaning was performed in the early stages of the survey with iqrm (Morello et al. 2021), presto's rfifind (Ransom 2011), and a multi-beam rfifind wrapper aiming to remove signals detected in several beams with sufficient spatial separation 11 .These methods were eventually replaced by a new, performant RFI removal tool, pulsarX's filtool (Men et al. 2023).We also masked frequency channels that are known to be significantly RFI-corrupted.
We calculated the dispersion smearing for our observations and the corresponding DM trial step sizes with presto's DDplan.pyscript.The SMC survey de-dispersion plan spans 50 to 350 pc cm −3 .We selected this maximum as the median DM of published SMC pulsars is 115 pc cm −3 and the maximum is 205 pc cm −3 (Manchester et al. 2006).Conversely, the minimum published DM for a SMC pulsar is 9 Pulsars can be kicked out of their birth site due to a supernova kick (e.g.Lyne & Lorimer 1994). 10CCOs are soft thermal X-ray sources near the centre of a supernova remnant, with no detected hard X-ray emission, radio emission, or PWN environment (De Luca 2017). 11https://github.com/mcbernadich/multiTRAPUMby Miquel Colom i Bernadich Table 3.The accretion-powered X-ray pulsars with a period under 10 s observed in this survey.These were taken from the Haberl & Sturm (2016) census (unless specified) with Gaia DR3 positions (Vallenari et al. 2023) to place a single beam on these sources from SMCPOINTING6 onwards.Four fast-spinning HMXB sources were observed with beams overlapping their position while targeted at different sources (see Appendix B). 71 pc cm −3 (Manchester et al. 2006) and the Milky Way DM contribution in the direction of the SMC is 30 or 42 pc cm −3 according to the YMW2016 or NE2001 models respectively (Yao et al. 2017;Cordes & Lazio 2002).Halving the frequency channels to 2048 after the first observation of the survey (to maximise the number of coherent beams, see subsubsection 2.1.2) doubled dispersion smearing within channels, which prohibits detection of the fastest spinning MSPs at the highest DMs.Indeed, at the central frequency of Lband, a maximum total smearing (including de-dispersion step size, sample time, and intra-channel dispersion) of 0.75 ms is reached at 350 pc cm −3 .At the lowest frequency of L-band, the total smearing is greater than 1 ms above a DM of 160 pc cm −3 .We used the dedisp library (Levin 2012;Barsdell et al. 2012) for de-dispersion12 .We performed periodicity searches using peasoup (Barr 2020), a GPU-based time-domain linear acceleration pulsar searching software suite (described in Morello et al. 2019 andSengar et al. 2023).The 2-hour data were not corrected for acceleration before performing a Fast Fourier Transform (FFT), while 20-minute segments were resampled for constant accelerations from a binary companion up to |50| m s −2 .This segmenting reduces the portion of the orbit for which a potential binary is observed, and thus the range over which the Doppler-shifted pulsar spin frequency is smeared.While our potential to detect highly accelerated systems is thus increased, the shorter integration time reduces our sensitivity by a factor of √ 6 compared to the full observation search.We chose an acceleration tolerance parameter of 10 per cent, resulting in 153 acceleration trials13 at 0 pc cm −3 .This means the acceleration broadening from one acceleration step to the next cannot exceed 10 per cent of the combined pulse smearing due to sampling time, intra-channel frequency dispersion, and de-dispersion step size (Levin 2012, used in Morello et al. 2019).The peasoup searches were limited to consider candidates with a minimum FFT spectrum Signal to Noise Ratio (S/N) of 8 and to a maximum period of 10 s.The 2-hour and 20-min timeseries were zero-padded to the next nearest power of 2 (2 23 and 2 26 respectively), resulting in a better resolved and more quickly computed Fourier spectrum.Candidates were harmonically summed by peasoup up to the eighth harmonic (Taylor & Huguenin 1969).A list of common local RFI fluctuation frequencies was removed from the searches.

Name of accretion
The FFT candidates were sifted by peasoup, i.e. clustered in DM, period, acceleration and harmonics.The full observation periodicity search processing time was de-dispersion dominated.These searches resulted in about 9×10 5 candidates per pointing, i.e. around 1200 candidates per full periodicity search of one beam.Due to an oversight, pointings 1 and 2 (first passes only) were limited to 1000 candidates per beam, resulting in a spectral S/N cut of about 8.4.They were also expected to retrieve around 1200 candidates per beam.
The segmented acceleration search processing time was FFTsearch dominated and resulted in around 12 million candidates per pointing, i.e. about 2600 candidates per segment of a single beam.Again, survey pointings 1, 2 and 3 (first passes only) were erroneously limited to 1000 candidates per beam segment in the acceleration search, increasing the spectral S/N cut to about 8.5.When acceleration trials were reduced by more than half from SMCPOINTING7 due to oversampling, the acceleration search resulted in approximately 7 million candidates per pointing or 1500 per beam segment.
The candidates were then input into a multi-beam spatial sifting programme14 which uses the expected spatial S/N distribution of real sources to identify RFI and performs clustering of candidates across beams, after which about 100 candidates per beam remained15 for the full periodicity search and 600 (200 after the reduction in acceleration trials) per segment for the acceleration search.This reduced the total number of candidates per pointing, both searches combined, from around 13 million to 3 million (8 million to 1 million from SMCPOINTING7).We note that a recently discovered bug in the multi-beam filtering code over-removed candidates from an inbuilt list of RFI fluctuation frequencies.It is possible to retroactively check if a specific candidate has been removed in a list kept in the data products of this pipeline.

Candidate folding
Due to the high total number of spectral candidates to fold, we applied the following cuts: we folded candidates with periods longer than 8 time samples (1.216 ms, 1 ms for the segment search of the first pass of pointings 1 and 2), and a spectral S/N greater than 9.This is on the high end of S/N cuts compared to other surveys, while not being the highest (Lyon et al. 2016).This resulted in about 30 candidates to fold for each full integration periodicity search of a beam and 65 (32 after the reduction of acceleration trials) for each beam segment acceleration search.We folded the raw data of all 3.5×10 5 (2×10 5 from SMCPOINTING7) candidates per pointing of this survey with the parameters of the filtered FFT candidates using pulsarX's psrfold_fil, a new software suite designed to reduce the computation costs of high candidate numbers from interferometer-based multi-beam pulsar surveys (Men et al. 2023).psrfold_fil cleans the full resolution data from RFI with filtool before folding and applies the clfd (Morello 2023) RFI removal software on the folded data.In this search, we downsampled the raw data into folded archives as follows: 64 time samples (bins) per pulse profile (128 for signals with  > 100 ms from SMC pointings 1 and 2 second pass onwards), each sub-integration in time of the observation is 20 seconds long, and the number of frequency channels is reduced to 64.The folds were optimised to the highest S/N pulse profile over the default psrfold_fil period and period derivative range around the peasoup-detected topocentric spin frequency so that the maximum smearing is 1 pulse period over the whole observation time, and over a DM range so that the maximum dispersion delay is 1 pulse period over the whole bandwidth.Due to electricity failures during processing, 18 and 12 segments could not be folded in the second pass of pointings 3 and 4 respectively.
Due to an oversight, the segment candidates from the first pass of pointing 1 and 2 were folded with opposite acceleration from the peasoup value.Candidates with an acceleration greater than twice the optimisation range of psrfold_fil were thus not folded correctly.These data were deleted before the bug was noticed and the first pass accelerated segment candidates could not be refolded.The pointings were observed and searched anew in their second pass.The acceleration convention was fixed after the candidates of the second pass were already folded.We refolded candidates with a S/N greater than 10 from the segmented acceleration searches of these second passes.

Candidate selection
As survey sensitivities and processing capabilities advance, the number of pulsar candidates is ever increasing at all stages of a pulsar search.Thanks to the high performance software employed in this survey, we were able to retrieve and fold hundreds of thousands of candidates from each observation.However, this is too many to inspect by eye -the so-called 'candidate selection problem' (Lyon et al. 2016).Part of this issue originates from being sensitive to weaker sources of RFI.However, the main source of spurious folds are so-called 'noise candidates'.These arise from the background noise levels and are simply random noise fluctuations (Groth 1975).With a large number of samples (long high-resolution observations) and search trials, along with strict RFI removal, our candidates are noise-dominated.
We partially classified the folded candidates with pics (Zhu et al. 2014, a Pulsar Image-based Classification System based on Machine Learning).We used a conservative minimum score of 10 per cent pulsar-like for both the original and TRAPUM training sets (Padmanabh et al. 2023), as well as a minimum16 folded S/N of 7.This returned a handful of candidates for each beam, or of the order of a thousand candidates per pointing, with few RFI candidates.
We used a Graphical User Interface candidate classifier tool, CandyJar 17 , to classify the folded candidate plots by eye and check for harmonically related candidates or known pulsars.Two or three volunteers viewed and classified the candidates as noise, RFI, known pulsar, new pulsar, or candidate pulsar.In the entire survey, seven candidates were classified as new pulsars.They are described in subsection 4.1.The folded discovery data were refolded at multiples of the period (2, 3, 1/2 and 1/3) to check that the pipeline did not retrieve a harmonic as the best detection, but this did not happen.Usually, around 50 low S/N folded candidate plots were clasified as pulsar candidates.These signals were cross-matched between segments and passes, but none was detected and classified as pulsar candidates more than once.Furthermore, from the second pass of pointings 1 and 2, weak pulsar candidates found in segments were folded over double their detection integration time, with the default psrfold_fil range search around their candidate period, DM and acceleration (the DM search was removed in later pointings).This did not strenghthen any segment candidates.Due to this low success rate, pointings 5 and 7 were not re-observed and the time was used to observe SMCPOINTING9 and the Globular Cluster NGC 121 (section 3).Pointings 3 and 4 were only partially re-observed due to a failed observation setup (Table 2).All pulsar candidates that were not observed twice are noted for potential re-observation.

Multi-beam localisation method
When a pulsar was discovered in a coherent beam, the beams in neighbouring positions were folded with the discovery parameters to look for more detections.As seen in the previous section, only the strongest detection would have been folded by the TRAPUM pipeline, as the duplicates are clustered spatially by the multi-beam filter.We thus obtained the pulsar discovery's S/N in each detected beam with psrfold_fil.These detection S/Ns, beam positions and point spread functions (generated by mosaic) were input into the seeKAT software (Bezuidenhout et al. 2023).seeKAT can then constrain the localisation of the pulsar to better than the detected beams' combined area (each beam having an area of about 1 square arcminute at 50 per cent sensitivity).
However, while beam positions are tiled by the FBFUSE beamformer to attain a chosen fractional sensitivity overlap when reaching half of the observation's integration time18 , the beams have a different overlap at the start and end of the observation as the observed source moves through the sky.To account for this in seeKAT, we split in time the strong detections of pulsars and their S/Ns into four segments with psrchive (Hotan et al. 2004) and generated a PSF for each segment.We could then generate a localisation for each segment and combine them with seeKAT.Weaker observations could not be split into segments, as the decrease in S/N would result in unreliable detections.In these cases, we assumed that using a PSF simulated for the middle of the observation time is sufficient.We also generally removed weak segment and/or beam detections that would not strengthen the localisation.Usually, discovery multi-beam localisations resulted in position errors of a few square arcseconds.

Single pulse search
We also performed a single pulse search on every beam19 from pointing 3 onwards, aiming to find giant pulses from e.g.young pulsars that are too faint to be detected in our survey, or FRBs.As described in section 1, two LMC pulsars are known to emit giant radio pulses, but no single pulse emission has been detected from radio pulsars in the SMC.With the two magnetars located in the Magellanic Clouds, it is worth searching for transient radio emission like that of the Milky Way FRB-emitting magnetar SGR 1935+2154 (Bochenek et al. 2020).Moreover, dwarf galaxies are known to host FRBs (see e.g. the repeating FRB 121102, Tendulkar et al. 2017), even if it is unlikely that one will be active presently.
We used the software transientX (Men & Barr 2024) to perform the searches on the raw data downsampled to a sampling time of 306 μs, RFI-cleaned by a rfifind or standard channel mask and pulsarX's filtool.We used a range of widths from 306 μs to 100 ms over DMs 0 to 5000 pc cm −3 (using DM trial step sizes from presto's DDplan.pyscript) and a minimum single pulse S/N of 8. Depending on the RFI during the observation, between tens of candidates to several tens of thousands per observation could be returned.Inspection of the candidates indicated they were consistent with RFI or noise, except for two single pulses from the known SMC pulsar J0045−7319 with a S/N of about 8.The flux density of these pulses is 17 times stronger than the average single pulse flux over the 2h SMCPOINTING3 observation.No single pulses were detected from any of our discoveries.
As well as this offline processing using transientX, a commensal search for transients was carried out on the incoherent beam and coherent beams using the MeerTRAP backend TUSE (Transient User Supplied Equipment).This means the raw data of both passes of Pointings 1 and 2 was searched by MeerTRAP, despite being deleted before transientX was run.The capabilities and pipelines for MeerTRAP are described in Rajwade et al. (2022).All identified candidates were consistent with RFI or noise or were from the known pulsar.

Future reprocessing
Once the processing of each pointing was finished, a downsampled version was stored for future reprocessing.The data were first RFI cleaned by iqrm.A typical pointing was then reduced by a factor of four in frequency (512 channels each de-dispersed at 115 pc cm −3 ), and by a factor of eight in time (1.225 ms samples).This reduces the storage space occupied by a typical pointing from 74 terabytes to 2.3 terabytes.We plan to search these data with a Fast Folding Algorithm (using e.g.Morello et al. 2020) in the future, with the possibility of extending the search to higher DMs.Due to electricity failures during processing, 72 and 48 beams could not be kept in reduced resolution for the second pass of pointings 3 and 4 respectively.The raw data of all pointings were deleted immediately after processing as they were too large to store.All secondary processing data products were retained for possible future inspection.

Sensitivity
In this section, we calculate radio flux density sensitivity limits using the radiometer equation (Equation 1) applied to pulsar observations (Lorimer & Kramer 2005, p. 265): In Table 4, we list the parameters used in the sensitivity calculations for this TRAPUM survey.Additionally, the MeerKAT L-band specifications are given in Table 1.We use a FFT efficiency factor  = 0.7 to perform a spectral to folded S/N conversion (Morello et al. 2020).Thus, a minimum S/N cut of 9 in the FFT corresponds to a folded S/N of approximately 13.The best flux density sensitivity limit of this survey's full integration periodicity searches is  1284 MHz,full = 7.6 μJy.We assume an integration time t obs of 7200 s with the core array (44 antennas), a pulsar with a rotation period of 100 ms (so that pulse broadening effects are negligible) placed at the centre of the pointing (incoherent beam location), and at the centre of a coherent beam (where it has maximum sensitivity).Assuming a power law radio spectral index of −1.60±0.54(Jankowski et al. 2018), our radio flux density sensitivity upper limit rescaled to 1400 MHz is  1400 MHz,full = 6.6±0.3 μJy.This translates to a radio pseudo-luminosity upper limit of  pseudo,1400 MHz =  1400 MHz ×  2 = 24.1±1.0mJy kpc 2 , assuming an approximate distance to the SMC of  = 60 kpc (Karachentsev et al. 2004).As seen later in Figure 1, this survey is thus twice as sensitive as the latest SMC pulsar survey by Titus et al. (2019), conducted with Murriyang.For the acceleration search, the segmenting of our observations reduces our sensitivity by a factor of √ 6, yielding  1284 MHz,segments = 18.7 μJy and  1400 MHz,segments = 16.3±0.8μJy with the same assumptions as before.Further from the pointing boresight, the sensitivity of coherent beams is reduced by the response of the incoherent beam.This is taken into account for the upper limits on the pulsed flux density of various targets presented later in subsection 4.3.Note that the incoherent beam constitutes a parallel survey with a large area and reduced gain.The full integration incoherent observations have a flux density sensitivity limit of  1284 MHz,incoherent = 42.7 μJy and  1400 MHz,incoherent = 37.2±1.8μJy with the same assumptions as before, which is a similar sensitivity to the Manchester et al. (2006) and Crawford et al. (2001) surveys.
In Figure 1, we show the flux density sensitivity of all the SMC pulsar surveys.The parameters for these surveys, including their S/N cut, are taken from their respective papers, which we list in Table 5.We assume the gain of Murriyang to be 0.735 K Jy −1 in all cases, taking the boresight value (Manchester et al. 2001).Two polarisations are recorded for all the Murriyang SMC surveys.The digitisation correction factors are taken from Kouwenhoven & Voûte (2001): 1.0 for 8-bit digitisation, 1.25 for 1-bit, and 1.06 for 2-bit.We derived the sky temperature in the direction of the SMC at each survey's central frequency using the Zheng et al. (2017) model using pyGSM (Price 2016).We rescale all survey flux density sensitivities to 1400 MHz assuming a −1.60 pulsar spectral index (Jankowski et al. 2018).Note that the McConnell et al. (1991) survey had a central frequency outside of L-band, at 610 MHz.This means that the spectral index conversion is speculative as there is a range of pulsar spectral indices that may vary the resulting flux density upper limit.The other surveys compared are all conducted at L-band, close to a central frequency of 1400 MHz.We use the median DM of the SMC, 115 pc cm −3 .We assume an intrinsic pulsar duty cycle of 2.5 per cent.We take into account pulse broadening due to sampling time, DM smearing (Lorimer & Kramer 2005, p. 109) in a frequency channel (2048 channels for this survey), and due to DM scattering -i.e.multipath propagation through the interstellar medium (using a formula from Bhat et al. 2004).Due to these effects, at shorter periods, the duty cycle  departs from the assumed value.Temperature contributions from the ground and the atmosphere, scintillation effects, effective bandwidth (due to RFI masking and band characteristics), de-dispersion step size, and harmonic summing contributions are not taken into account, as they are not all provided for previous surveys.
Similarly, the limiting fluence for this survey's single pulse search down to S/N=8 is  pulse,1284 MHz = 80.4 mJy ms, for a 1 ms pulse width (and a 1 ms integration time).We use the same assumptions as before, for a pulse originating from the centre of the pointing and at the centre of a coherent beam observed with the core array (44 antennas).
We also show in Figure 1 the discovery flux densities of the SMC pulsars, scaled to 1400 MHz assuming a −1.60 pulsar spectral index.This is calculated using Equation 1 as before, with the values in tables 1, 4 and 5.We use the folded S/N (thus not needing the spectral to folded S/N conversion factor ) and duty cycle (pulse width divided by period) of the discovery observation (see subsection 4.1 and Table 8).The MeerKAT core gain is adjusted depending on the number of antennas used in the discovery observation.The observation length is approximated to 7200 s despite the variation between pointings.Observed pulsar flux densities can be highly variable, therefore the values plotted are only valid for the discovery observation.Furthermore, the pulsar can be discovered away from the beam centre, which results in a decreased gain.To reflect discovery values, this is not taken into account in this plot.This results in an artificially decreased pulsar flux density.Finally, the S/N can be greatly impacted by the folded data resolution (as seen later in sections 4.2.1 and 4.2.2).We do not correct for this to reflect the search process.For the McConnell et al. (1991); Crawford et al. (2001) and Manchester et al. (2006) surveys, the pulse width and S/N in the discovery observations are not provided, thus we use directly their flux density value as an approximation.The values these authors used in Equation 1 to calculate the pulsar flux densities are unknown and different from Table 5, and thus are not directly comparable with the sensitivity curves.Most SMC pulsar discoveries were bright enough to be discovered in the previous survey, but were not due to survey coverage, beam losses and flux density variability.We discuss the detectability of our discoveries in the Titus et al. (2019) survey in subsection 4.1, and the detectability of their discoveries in this survey in subsection 4.4.

GLOBULAR CLUSTER NGC 121
Millisecond pulsars are the fastest spinning and oldest pulsars known.They have been 'recycled', i.e. spun up by accretion from a binary companion.They have science applications such as gravitational wave detection and tests of relativistic gravity (see Lorimer 2008).Globular Clusters are known to harbour exotic compact objects and binaries, thanks to their old age and the high density of high-mass stars in their core that favours interactions.About 60 per cent of millisecond pulsars (here conventionally assumed to be pulsars with a period of under 10 ms) and over 40 per cent of binary pulsars are located in Globular Clusters of the Milky Way (Manchester et al. 2005).As there are no known extragalactic millisecond pulsars (MSPs), the detection of the first rotationally stable extragalactic MSP could  1983) is given with the sky temperature in the direction of the LMC added, that we removed using a pyGSM value.The spectral S/N cut in Titus et al. (2019) is 4, but we assumed a minimum folded S/N value of 7 to be able to distinguish a signal by eye.improve Pulsar Timing Arrays by greatly increasing their longest baseline distance (Lee et al. 2008).The first extragalactic pulsar in a double neutron star would inform the extragalactic neutron star merger rate.NGC 121 is the only Globular Cluster of the SMC, and its oldest stellar cluster, with an age of 11 billion years.Situated 65 kpc away on the outskirts of the dwarf galaxy, it is a mass-segregated, but not core-collapsed cluster.Its core, half-light and tidal radii are 4.2, 8.5 and 51.9 pc respectively (Glatt et al. 2008, 2009, andreferences therein).The expected Milky Way Dispersion Measure contribution in its direction is as expected for the SMC (30-40 pc cm −3 ) according to both the Yao et al. (2017) and Cordes & Lazio (2002) electron density models.Since it is situated away from the main body of the dwarf galaxy, we can expect that the SMC DM contribution will be low.We conclude it is an ideal target to search for extragalatic MSPs or relativistic binaries.
We conducted two observations of NGC 121 with the same back-ends introduced in subsubsection 2.1.1.However, the observation parameters, shown in Table 6, are different to those of the rest of the survey (described in subsubsection 2.1.2).Thanks to the small angular size of the cluster, we were able to use the full array of MeerKAT, despite its much reduced beam size.We used 63 coherent beams to cover an area that extended beyond the half-light radius of the cluster in the first pass observation, and seven to cover its core in the second pass.Thanks to the low number of beams recorded, we were able to double the frequency and time resolution to 4096 channels and a sampling time of 76 μs.This also reduces the pulse smearing in a frequency channel due to dispersion, and the sampling time pulse smearing, that strongly affect millisecond pulsars due to their extremely short pulse length.
The central beam of the first pass observation (see Table 6) was RFI-cleaned with filtool then processed with pulsar miner20 v1.1.5,on the four manual processing nodes of the APSUSE cluster (subsubsection 2.1.1).We followed the same pulsar miner processing steps as detailed in Ridolfi et al. (2021).We produced just over 6000 de-dispersed timeseries with prepsubband (using 64 subbands) from 30 to 335 pc cm −3 .We chose this maximum DM as the dispersion smearing within a frequency channel becomes larger than twice our sampling time at higher DMs.We ran accelsearch with  max = 0 and 200 to target isolated and binary pulsars, on segments of 15, 30, 60 minutes and the full observation, from periods spanning 0.91 ms to 10 s.This results in a maximum acceleration searched of 1.2 or 74 m s −2 for a 1 ms pulsar and 116 or 7.4×10 4 m s −2 for a 100 ms pulsar, in the full observation and 15-minute segments respectively.Candidates were harmonically summed up to the eighth harmonic.A list of common local RFI fluctuation frequencies was removed from the candidates.After sifting, candidates with a DM above 10 pc cm −3 and a minimum Fourier significance of 6 σ were kept.The raw data were folded with these candidates' parameters, using 64 time subintegrations, 128 frequency channels, and 128 pulse profile bins.All candidates (about 2000) were visually classified by three viewers and about 40 were identified as faint pulsar candidates in segments or full observations.None of the segment candidates was confirmed by doubling their integration time in a psrfold_fil fold optimised over the default range, or by appearing in several segments.10 full observation candidates remained.To confirm or dismiss these candidates, a second pass observation of the core of the cluster was conducted (Table 6).All seven beams of the second pass were folded using psrfold_fil at the parameters of the 10 candidates, and optimised over twice the default range.This did not yield any pulsations and the candidates were dismissed as noise.
The full tiling of 63 coherent beams of the first pass observation was then searched following the steps described in subsection 2.2.Some adjustments were made to the peasoup parameters: segments of 20 and 40 minutes, as well as the full observation, were searched with no acceleration, using the same de-dispersion plan described in the previous paragraph.The parameters of psrfold_fil were also altered: we folded candidates with S/Ns greater than 8 and subintegrations were reduced to 10 s for the segment searches.These searches cumulatively returned around 800 candidates after multibeam filtering and AI scoring.These were classified by two or three viewers.Around 15 low confidence segment candidates were found.Half of these were refolded with double their integration time, which did not strenghthen their S/N.The other half were refolded at the wrong parameters due to an oversight and could not be confirmed or discarded before the raw data were deleted.The reduced resolution stored data (subbanded at a DM of 100 pc cm −3 ) did not have a sufficiently high time resolution to detect them or was not available (due to an electricity failure, three beams of the first pass could not be stored).The second pass observation did not cover their position in the edge of the cluster and was also stored in reduced resolution.Consequently, no candidates were retained from the peasoup segmented search of the 63 coherent beams of the first pass observation.Single pulse searches were performed on data downsampled to 153 μs (first pass); or 306 μs and 2048 frequency channels (second pass, incoherent beam not searched).They returned candidates consistent with RFI or noise.
Using the values and assumptions detailed in subsection 2.3, we can calculate a flux density sensitivity limit with Equation 1 for our peasoup periodicity (no acceleration) search of the full integration, full tiling of the first pass observation of NGC 121:  1284 MHz, NGC 121 = 10.2 μJy.We used the appropriate gain, S/N cut and sky temperature to reflect our observation.We set the DM to 100 pc cm −3 and the duty cycle to 10 per cent (approximately the median duty cycle of pulsars in Globular Clusters, Manchester et al. 2005).Due to the very small size of the tiling (see Figure B9a), the incoherent beam loss is negligible (less than 1 per cent) at the edge of the tiling which extends beyond the half-light radius of the cluster.This limit is valid for a pulsar with a rotation period of 50 ms (so that pulse broadening effects are negligible).For a pulsar with a rotation period of 1 ms, the limit is  1284 MHz, NGC 121, MSP = 12.9 μJy, taking into account pulse broadening as detailed in subsection 2.3.These upper limits rescaled at 1400 MHz are  1400 MHz, NGC 121 = 8.9 ± 0.4 μJy and  1400 MHz, NGC 121, MSP = 11.2 ± 0.5 μJy.This is six times more sensitive than the latest pulsar search of this cluster, Titus et al. (2019).Their beam loss for this target was 50 per cent, which is taken into account here, but this was still the most sensitive search of NGC 121 previously.We show the sensitivity curves for both surveys in Figure 2. Our flux density upper limit translates to a radio pseudo-luminosity upper limit of  pseudo, 1400 MHz, MSP =  1400 MHz, NGC 121, MSP × 65 kpc 2 = 47±3 mJy kpc 2 .Only a handful of Milky Way pulsars with a period under 10 ms have a pseudo-luminosity higher than this value (Manchester et al. 2005), thus potential NGC 121 MSPs could still be too faint to be detected in this search.We reserve a more in-depth population limit of this cluster for Paper III.The limiting fluence for the first pass single pulse search is  pulse,1284 MHz = 58 mJy ms, for a 1 ms pulse with a S/N of 8 (and a 1 ms integration time).Figure 3.The discovery pulse profiles from the folded data of the pulsarX pipeline candidate.The pulse smearing is due to the number of phase bins: the raw data smearing due to sampling time or dispersion within a channel is negligible compared to the bin size.The data were folded with 64 phase bins for the first pass of pointings 1 and 2, then this increased to 128 phase bins for periods over 100 ms.For PSR J0044−7314 the original candidate data were corrupted by an electricity failure, so we display the RFI-cleaned raw discovery data folded to the resolution of the pipeline.

Pulsar discoveries
We present the discovery parameters of the seven new pulsars discovered in this survey in Table 8.The period and dispersion measure are approximate values from the pulsarX discovery plots.A short description of the discovery of each pulsar is provided in the following sections, and their discovery pulse profiles are shown in Figure 3.These parameters and others will be robustly estimated in Paper II of this series.All Parkes (Murriyang) Pulsar Data Archive observations 21 near the position of the discoveries were folded with dspsr (Van Straten & Bailes 2011) at the pulsars' period and dispersion measure.RFI was removed from the folded data with clfd and optimised over a large period search range with psrchive's pdmp tool (Hotan et al. 2004).None of the pulsars could be detected.Notably, none of our pulsar discoveries was located within any of the survey beams of Titus et al. (2019) (considering the beam area with over 50 per cent of the full sensitivity), except for PSR J0040−7337 which we discuss in subsubsection 4.2.2.This explains why our discoveries that appear detectable in this latest Murriyang SMC survey in Figure 1 were not found then.We provide the details of the folded archival observations in Table 7.

PSR J0048−7317
PSR J0048−7317 was discovered in the full observation search of the central tiling of the first pass of SMCPOINTING1 (see figure B1a) with a pulsarX folded S/N of 32.1.It was also detected by the search pipeline in a segment with a folded S/N of 11.2.Following the procedure described in subsubsection 2.2.4, the pulsar was localised using segments of six coherent beam detections.In the second pass of SMCPOINTING1, this SeeKAT localisation position was targeted with a single beam (see figure B1b) and resulted in a pipeline detection in the full observation with a S/N of 55.2 and in all six segments.It was also detected by the pipeline in SMCPOINTING9 as a high S/N harmonic detection in the full observation and in 5 segments.We note that this pulsar has the highest Dispersion Measure known in the SMC.The DM range of the SMC previously spanned 71.3 to 205.2 pc cm −3 and now extends to 292.3 pc cm −3 .We discuss a new PWN association for this pulsar in subsubsection 4.2.1.This pulsar has a short spin period of 79.3 ms.No single pulses were detected from this pulsar.

PSR J0054−7228
PSR J0054−7228 was detected in in the full observation search of the central tiling of the first pass of SMCPOINTING2 (see figure B2a) with a pulsarX folded S/N of 26.9.The pulsar was localised using segments of five coherent beam detections (one was removed as its S/N was too weak).In the second pass of SMCPOINTING2, this position was targeted with a single beam (see figure B2b) and resulted in a pipeline detection in the full observation with a S/N of 45.5 and in all six segments.It was also detected by the pipeline in the full integration of SMCPOINTING9 with a low S/N due to being in a low sensitivity area of the incoherent beam. 21Accessed through CSIRO's Data Access Portal https://data.csiro.au/domain/atnf4.1.3PSR J0044−7314 PSR J0044−7314 was detected in the full observation search of the second pass of SMCPOINTING1, in a coherent beam tiling covering the position error of known pulsar PSR J0043−73 (see figure B1b) with a pulsarX folded S/N of 23.5.The pulsar was localised using full integrations of three coherent beams.In the first pass of SMCPOINTING3, this position was targeted with a single beam (see figure B3a) but did not result in any significant detections, due to a combination of flux variability and being further away from the pointing centre (i.e. the gain at the pulsar location was reduced).Similarly, no significant detection was made in SMCPOINTING9.The pulsar was later confirmed in follow-up observations.

PSR J0040−7337
PSR J0040−7337 was detected in the full observation search of the first pass of SMCPOINTING3, in a beam targeted at the PWN of DEM S5 (see figure B3a) with a pulsarX folded S/N of 22.8.We discuss the new PWN association for this pulsar in subsubsection 4.2.2.The pulsar was localised using full integrations of four coherent beam detections (the additional beams came from the untargeted tiling the targeted beam was overlapping with) and was confirmed in follow-up observations.We show the output seeKAT plot as described in Bezuidenhout et al. (2023) in Figure 4.This pulsar is now the fastest spinning known radio pulsar in the SMC with a spin period of 59.9 ms.No single pulses were detected from this pulsar.

PSR J0040−7335
PSR J0040−7335 was detected in the full observation search of the first pass of SMCPOINTING3 with a pulsarX folded S/N of 29.3.It was detected in a beam neighbouring PSR J0040−7337 as seen in B3a.Owing to their very different DMs and different (not harmonically related) periods, we do not believe these two pulsars are associated, as detailed in subsubsection 4.2.2.The pulsar was localised using full integrations of five coherent beams and was confirmed in follow-up observations.

PSR J0040−7326
PSR J0040−7326 was detected in the full observation search of the first pass of SMCPOINTING3 (in the central tiling, see figure B3a) with a pulsarX folded S/N of 11.9, making it the weakest pipeline detection.The pulsar was localised using the full integrations of three coherent beams and was confirmed in follow-up observations.

PSR J0105−7208
PSR J0105−7208 was detected in the full observation search of the first pass of SMCPOINTING5 (in the central tiling, see figure B7a) with a pulsarX folded S/N of 22.0.The pulsar was localised using the full integrations of three coherent beams and was confirmed in follow-up observations.Its vicinity to a SNR is discussed in subsubsection 4.2.3.

Associations
The latest radio continuum images of the SMC (Joseph et al. 2019;Cotton et al. 2024) were checked for extended emission at the locations of our new pulsars.This revealed the associations detailed in the next sections.dio pulsars in the LMC (see section 1).We reserve a detailed analysis of this pulsar's age and energetics to Paper II.
Due to the suggested young age of the pulsar and its localisation, we classify this nebula as a PWN.This is the third such object in the SMC after IKT 16 (Maitra et al. 2015;Carli et al. 2022) and DEM S5 (Alsaberi et al. 2019, discussed in the next section).With PSR J0040−7337 described in the next section, these discoveries are the first radio pulsar-PWN systems in the SMC and only one other such system was previously known outside our galaxy: the 'Crab pulsar twin' PSR J0540−6919 (see section 1).Cotton et al. (2024) measure the size of the nebula to be about 100 arcseconds or 29 parsec at a distance of 60 kpc.They also state that there is no visible SNR on their radio continuum image near the pulsar.Thus, the PWN and pulsar system has no known parent SNR, like several others, including the 'Potoroo' PWN (Lazarević et al. 2024).The nearest known SNRs are IKT 4 and IKT 5.However, to travel from their centres to its current position in 38 ky, the pulsar would have needed to have a projected velocity of about 1500 km s −1 , which is at the limit of the known distribution of pulsar velocities (Lyne & Lorimer 1994;Verbunt et al. 2017).The tail of emission extending to the south likely marks the direction the pulsar is moving in.It points to neither of the nearby remnants.Morever, IKT 4 is a Core-Collapse supernova, IKT 5 is of possible Type Ia supernova origin, but both are estimated to be younger than the pulsar (Maggi  Leahy & Filipović 2022).This further favours the suggestion that the parent SNR is not visible, but a better estimate of this pulsar's age will inform this further.This is the only known extragalactic PWN with no SNR association (see section 1).
The SMCPOINTING1 second pass detection of PSR J0048−7317 was obtained from a coherent beam targeted at our seeKAT localisation from the first pass (see subsubsection 4.1.1).We created high resolution folded data with dspsr and optimised the folding parameters with psrchive's pdmp tool to obtain a S/N of 69.1 and a pulse width of 2.9 ms.This second pass S/N is different to the pulsarX S/N in subsubsection 4.1.1due to the higher resolution data improving the signal.Similarly, the pulse width is slightly different to the discovery value in Table 8.We can estimate a preliminary flux density for PSR J0048−7317 of 56 μJy at 1284 MHz, noting however that it can vary significantly between observations23 .We used Equation 1and corrected the MeerKAT gain for the incoherent beam sensitivity loss (2 per cent) due to the pulsar being slightly away from the centre of the beam (see figure B1b).No FFT efficiency factor  is needed as we do not use a spectral S/N value.Cotton et al. (2024) state that no emission from the nebula could be found in images of the region in any other wavelength, suggesting the nebula is non-thermal.No high energy gamma-ray source was detected in a recent survey of the SMC by HESS (Haupt 2020).A nearby X-ray source, 4XMM J004855.2−731753, is identified in the XMM-Newton DR13 catalogue (Webb et al. 2020), but it has a low significance of about 4 σ.PSR J0048−7317 is located off-axis of all nearby XMM observations: stronger emission may be revealed with targeted observations.(Haberl et al. 2000;Filipović et al. 2008) using radio and XMM-Newton X-ray images.In both the X-ray and radio data, they were able to resolve the PWN's soft thermal emission from the SNR's.Within the 14 arcsecond soft X-ray PWN, they detected a confused, unresolved hard X-ray point source; coincident with a steep spectrum, non-thermal, resolved radio point source: a putative pulsar's magnetospheric emission.Indeed, steep spectrum radio emission from pulsars' magnetospheres is powered by their rapid rotation, which can also power hard X-ray emission when the pulsar is young and most rapidly rotating, with a characteristic hard powerlaw spectrum (Becker & Trümper 1997, and references therein).They also resolved a 10 pc radio bow shock nebula and tail due to the putative pulsar's fast displacement through the SNR medium from its centre, and estimated the age of the PWN-SNR system to be between 10 and 28 kyr.The latest radio continuum image (Cotton et al. 2024) of this PWN-SNR system is shown in Figure 6.Alsaberi et al. (2019) searched for the pulsar using Murriyang observations but could not find any pulsations.No gamma-ray source was detected in a recent survey of the SMC by HESS (Haupt 2020).
In the first pass of SMCPOINTING3, as detailed in subsubsection 4.1.4,we discovered PSR J0040−7337 in a beam targeted at the position of the steep spectrum radio point source resolved by Alsaberi et al. (2019) in a 5500 MHz ATCA image: RA(J2000) = 00 h 40 m 46.s 49 Dec (J2000) = −73°37 ′ 05.′′ 8.This targeted beam was overlapping with an untargeted tiling which allowed detections in multiple coherent beams and a seeKAT localisation.The localisation is shown in figures 6 and 4, and overlaps the position of the ATCA radio point source within 3 σ (with the position error region approximated as Gaussian, see Bezuidenhout et al. 2023).As the second pass of SMCPOINTING3 failed (see Figure B3b), this survey only has one observation of the new PSR J0040−7337 and we cannot estimate its period derivative and age from survey observations.However, the period of the pulsar is indicative of a young pulsar, with a characteristic age of order 100 ky or less (Manchester et al. 2005).We therefore believe this pulsar is associated with DEM S5 and powers its PWN.We reserve a detailed analysis of this pulsar's age and energetics that could strengthen this association to Paper II.This is the first radio pulsar-SNR system in the SMC and only one other such system was previously known outside our galaxy (the 'Crab pulsar twin' PSR J0540−6919).There are only two known young (<100 kyr) radio pulsars outside our galaxy, in the LMC (see section 1).If the age of this pulsar was to match that of the PWN, this would be, with PSR J0048−7317, the first two young radio pulsars in the SMC.
We created high resolution folded data of the discovery observation with dspsr and optimised the folding parameters with psrchive's pdmp tool to a S/N24 of 37.7.We can estimate a preliminary flux density for PSR J0040−7337 of 16 μJy at 1284 MHz, noting however that it can vary significantly between similar observations.Using Equation 1, we corrected the MeerKAT gain for the incoherent beam sensitivity loss (2 per cent) due to the pulsar being slightly away from the centre of the beam (see figure B3a).No FFT efficiency factor  is needed as we do not use a spectral S/N value.This is much lower than the steep spectrum radio point source flux25 in Alsaberi et al. (2019), therefore it is possible that foreground emission, likely from the shock between the pulsar wind and the surrounding medium, contributed to the point source flux they measured.
We downloaded the targeted Murriyang observations conducted by Alsaberi et al. (2019) from the Parkes Pulsar Data Archive26 .The bandwidth (256 MHz), length (40 minutes) and gain were reduced compared to our discovery observation.We folded the data with high resolution using our discovery period and dispersion measure with dspsr.We then optimised the S/N of the folded profile over a large period search range with psrchive's pdmp, but the pulsar could not be detected.As seen in Figure 1, PSR J0040−7337 could also be below the detection threshold of the latest SMC survey, Titus et al. (2019).We followed the same procedure as above on these observations, and the pulsar could not be detected.We note also that it was placed off-centre of the Murriyang beam, resulting in further gain loss.
The localisation of our new pulsar J0040−7335 (see Table 8 and subsubsection 4.1.5)is coincident with the SNR DEM S5, just north of the PWN of PSR J0040−7337 as seen in Figure 6.However, the DM of PSR J0040−7335 is nearly double that of PSR J0040−7337, so we believe it is a chance alignment and not an association, with PSR J0040−7335 in the background of the SNR.We checked the images in Alsaberi et al. (2019) for multiwavelength emission from this pulsar within the DEM S 5 emission but none could be found.

A pulsar in DEM S128 but not associated?
We have noted, using recent SMC radio continuum images (Joseph et al. 2019;Cotton et al. 2024), that our pulsar discovery J0105−7208 is situated just at the edge of the radio extent of the supernova remnant DEM S128 (Filipović et al. 1998).It is not known whether this remnant originated from a Type Ia or core-collapse supernova (Roper et al. 2015) but evidence leans towards the former (Maggi et al. 2019;Leahy & Filipović 2022).The age of the SNR is 12 ky according to Leahy & Filipović (2022).To travel from the centre of the SNR to its position in that time, PSR J0105−7208 would have needed to have a projected velocity of over 2000 km s −1 , which is well beyond the known distribution of supernova birth kicks (Lyne & Lorimer 1994;Verbunt et al. 2017).Therefore, we suggest that the pulsar is not associated with the SNR.There is a central radio and X-ray source in the SNR (XMMU J010532.0−721017) that is tentatively classified as a background Active Galactic Nucleus but not excluded as a PWN (Sturm et al. 2013).There are no known X-ray sources at the pulsar position.The determination of the age of this pulsar in Paper II will inform further this potential association.

Upper limits
No other pulsars were found in SNRs in this survey.In Table 9, we show the full observation flux density upper limits on radio pulsations for all the SNRs observed and also for the magnetar CXOU J010043.1−721134.We use the same method as in subsection 2.3, taking into account the integration time and number of antennas of each observation (see Table 2 ) as well as the decrease in sensitivity away from boresight.We note that these values are valid at the centre of each coherent beam.The coherent beams were tiled to overlap at a decreased sensitivity level, usually 75 per cent.Our upper limit on the rotation-powered X-ray pulsar J0058−7218 in a PWN of the supernova remnant SNR IKT 16 (Maitra et al. 2021) is detailed in Carli et al. (2022).

Localisation of two Murriyang discoveries
The two pulsars discovered in Titus et al. (2019) were not localised: their position was not constrained within the 14 arcminute diameter Murriyang beam in which they were detected.In the second pass of Pointings 1 (PSR J0043−73, figure B1b) and 2 (PSR J0052−72, figure B2b), we tiled these error regions with about 170 MeerKAT coherent beams.PSR J0052−72 was detected by the full integration search pipeline with a pulsarX folded S/N of 15.5.Using three full integration detections in neighbouring beams, we localised the pulsar with seeKAT to RA(J2000) = 00 h 51 m 36.s 5 ± 1. s 5 Dec(J2000) = −72°04 m 26.′′ 1 ±3.′′ 5 (2σ error).PSR J0043−73 was not detected by the pipeline.We used Titus et al. 2019's discovery period and DM to fold every beam in the tiling, optimised over the default psrfold_fil range.This yielded a single detection in an edge beam of the tiling with a pulsarX folded S/N of 9.5 (which is below our search pipeline limit of folded S/N 13, see subsection 2.3).A second observation was necessary to obtain simultaneous detections in several coherent beams.In the first pass of SMCPOINTING3, we observed a tiling of 45 coherent beams overlapping at the 70 per cent sensitivity level centred on the position of the SMCPOINTING1 detected beam.The pipeline again did not yield any detections.We Table 9. Flux density upper limits for various targets observed in this survey.The targets' Right Ascension and Declination coordinates are given.The most sensitive observation is used.The parameters of the observations are given in Table 2.The method used to calculate the limits is detailed in subsection 2.3.The gain is decreased away from the centre of the pointing by the incoherent beam loss, using a beam model from Asad et al. (2021).Reference 1 is Lamb et al. (2002), 2 is Maggi et al. (2019), 3 is Titus et al. (2019) and their proposal, and 4 is Cotton et al. (2024).used our SMCPOINTING1 detection period and DM to fold every beam in the tiling, optimised over the default psrfold_fil range.This recovered four coherent beam detections, the strongest of which had a pulsarX folded S/N of 12.9.Using full integrations, we localised the pulsar with seeKAT to RA(J2000) = 00 h 43 m 17. s 6 ± 3. s 9 Dec(J2000) = −73°19 m 47. ′′ 5 ± 15. ′′ 0, just outside the Murriyang survey beam area with over 50 per cent sensitivity.We note that PSR J0043−73's flux density has decreased by about a factor of 10 since its discovery in Titus et al. (2019).Indeed, PSR J0043−73 could not be detected by our pipeline despite its discovery flux (shown in Figure 1) being several times brighter than our sensitivity limit.Furthermore, its intrinsic flux upon discovery was at least twice as strong as reported in Titus et al. (2019) as we have localised it to outside the Murriyang beam 50 per cent sensitivity contour.These pulsars are now named PSR J0043−7319 and J0051−7204.

SUMMARY
This paper described a radio-domain search for accelerated pulsars and transients in the Small Magellanic Cloud, a low-metallicity galaxy with recent star formation history expected to host a population of young compact objects.The survey, conducted at L-band with the core array of MeerKAT in 2-h integrations, took advantage of TRAPUM's beamforming capabilities to simultaneously cover the Supernova Remnants of the Maggi et al. ( 2019) census, as well as point sources and untargeted areas.About 800 coherent beams were recorded for each observation.Each observation's large dataset of high-resolution time and frequency domain records was processed locally on the TRAPUM computing cluster and yielded considerable numbers of candidates.This number was reduced using new software adapted to the sizeable outputs of interferometric surveys.The survey is generally about twice as sensitive as the latest SMC pulsar search, conducted at Murriyang.Individual upper limits on pulsed emission from Supernova Remnants are provided where no discovery was made, reaching down to 8 μJy at 1400 MHz.
We reported the discovery of seven new SMC pulsars, doubling this galaxy's known radio pulsar population and increasing the total extragalactic population by nearly a quarter.Localisations were computed upon discovery using multi-beam techniques to a position error of a few arcseconds.We surveyed the position error region of two previous Murriyang discoveries to localise them in the same manner.Further, we conducted an expansive search for accelerated millisecond pulsars in the SMC Globular Cluster NGC 121 using the full array of MeerKAT.This improved by a factor of six the previous upper limit on pulsed radio emission from this cluster, and started to probe into the high end of known millisecond pulsar luminosities.
Finally, our discoveries revealed the first radio pulsar-PWN systems in the SMC, with only one such system previously known outside our galaxy (the Crab pulsar twin in the Large Magellanic Cloud, PSR J0540−6919).In particular, we presented a new young pulsar with a 79 ms period, PSR J0048−7317, powering a Pulsar Wind Nebula recently discovered in a MeerKAT radio continuum image.We also associated the 59 ms pulsar discovery PSR J0040−7337, now the fastest spinning pulsar in the SMC, to the bow-shock PWN of Supernova Remnant DEM S5.
The timing solutions and peculiarities of our discoveries will be reported in Paper II of this series.The discoveries will enable an analysis of the impact of the low-metallicity, recent star formation environment of the SMC on its neutron star population and comparisons with predictions.We will present this population study in Paper III of this series.

APPENDIX B: BEAM LAYOUT MAPS
In the following figures B1a to B9b, we show the layout of the MeerKAT coherent beams for each observation, aimed at targets detailed in subsubsection 2.1.2.We will show the full survey coverage of the SMC in Paper III.The regions are overlaid on the Joseph et al. ( 2019) ASKAP EMU radio continuum image of the SMC shown in greyscale.The beam positions were retrieved from the FBFUSE record of the observation.A high resolution coherent beam PSF was simulated with the latest version of mosaic at the centre coordinates of the pointing, at the central time of the observation, with the antennas used during the observation and at the central frequency of L-band.An ellipse was then fitted by mosaic to the PSF fractional sensitivity level chosen for the tiling overlap.This is the beam size displayed on the figures.
The beam placement and packing is not optimal for all observations.This can be due to several reasons.First, for both passes of pointings 1 and 2, the beams were immediately tiled with the requested overlap.The tilings' overlap departs from the requested value as the observed sources move through the sky during the observation.From pointing 3 onwards, the beams were tiled so that they would reach the desired overlap in the middle of the observation, giving minimal deviation from this optimal tiling.Secondly, the beam shape is simulated with mosaic at very low resolution during the observation setup by FBFUSE, which results in an approximation of the true beam shape.Finally, for both passes of pointings 1 and 2, mosaic used a Gaussian fitting technique which was later replaced by a more performant contour ellipse fitting.

Figure 1 .
Figure 1.The minimum flux density sensitivity curves and the pulsar discovery flux densities (modified by beam losses) for the pulsar surveys of the SMC.All 14 known SMC radio pulsars are plotted.The calculations used to produce this figure and interpretations are detailed in subsection 2.3.

Figure 2 .
Figure 2. The minimum flux density sensitivity curves for the two most recent pulsar surveys of the SMC Globular Cluster NGC 121.The calculations used to produce this figure and interpretations are detailed in section 3.

Figure 4 .
Figure 4.The seeKAT localisation output plot of PSR J0040−7337.The localisation's 1 to 4σ error regions are depicted in red.The coherent beams in which PSR J0040−7337 was detected are shown (50 per cent sensitivity contour).The position of the radio point source resolved by Alsaberi et al. (2019) in a 5500 MHz high-resolution ATCA image (see subsubsection 4.2.2) is shown as a dark blue cross.

Figure 5 .
Figure 5.The new Pulsar Wind Nebula associated with PSR J0048−7317 from the Cotton et al. (2024) MeerKAT image of the SMC and the seeKAT localisation region with its 3σ error approximated to a green ellipse.This localisation is consistent with the brightest emission in the head of the nebula at the top of the figure.The southern tail of the nebula overlaps with the brightest source in this figure, a steep spectrum background source.

Figure 6 .
Figure 6.We show the Cotton et al. (2024) MeerKAT radio continuum image of the SMC around the SNR DEM S5.On the left, we use a color scale that shows the bow-shock PWN with a shock front, and a pulsar displacement tail as discovered in Alsaberi et al. (2019).The soft X-ray extent of the PWN from Alsaberi et al. (2019) is shown in green.On the right, we use a color scale that reveals the fainter emission from the SNR, with its extent as stated in Maggi et al. (2019) shown as a large red ellipse.In SMCPOINTING3, a coherent beam was centred on the position of the radio point source resolved by Alsaberi et al. (2019) in a 5500 MHz high-resolution ATCA image (see sections 4.1.4and 4.2.2).This position is shown as a blue cross -the ATCA beam size is too small to display here.Our PSR J0040−7337 seeKAT discovery localisation 3σ error contour (approximated as a symmetrical ellipse) is shown in black.Our background pulsar discovery PSR J0040−7335 is shown in yellow (seeKAT discovery localisation 2σ error contour, approximated as a symmetrical ellipse).

( a )
The first pass of SMCPOINTING1.(b) The second pass of SMCPOINTING1.

Figure B1 .
Figure B1.(a): The MeerKAT core array coherent beams observed in the first pass of SMCPOINTING1 are shown as small black ellipses.The large black circles show the extent of the full array MeerKAT incoherent beam with a FWHM of 1.1 degrees (solid line) at the centre frequency of L-band and 0.9 degrees (dashed line) at the highest frequency of the band (Asad et al. 2021).Regions are overlaid on the Joseph et al. (2019) ASKAP EMU radio continuum image of the SMC at 1320 MHz.Supernova remnants (bona fide or candidate, see Table 9) are displayed as red ellipses.Five supernova remnants are observed with a 75 per cent coherent beam overlap requested and a slightly lower overlap obtained (see Appendix B) and one with a single beam displayed at 50 per cent contour.Two SNRs from Cotton et al. (2024) are depicted as brown ellipses.Most of these new SNRs discovered with MeerKAT were not observed as they were not known at the time of observation.The remaining untargeted beams were tiled at boresight where sensitivity is highest, and overlapped with one SNR.One pulsar was discovered in these beams: PSR J0048−7317 (see subsection 4.1 and subsubsection 4.2.1).The strongest S/N discovery beam is highlighted in purple.Single beams (shown at 50 per cent sensitivity contour) were placed on the known pulsar J0045−7319 (small pink box with position error box inside) and the published position of PSR J0043−73.However, the error on the position of the latter (large pink box covering the Murriyang beam area) was not taken into account, and the pulsar was not detected in this pass of SMCPOINTING1.Steep spectrum point sources from Joseph et al. (2019) (orange boxes) were not yet targeted and none was covered by our coherent beam tilings.HXMBs from the Haberl & Sturm (2016) census are displayed as dark blue circles with their position error box inside.Those with a position error larger than 100 ′′ are omitted.HMXBs with a period of under 10 s (dark blue boxes, see Table 3) were not yet targeted but one was observed in the central tiling: SXP 9.13.In the top left corner, one SNR (red ellipse) observed in SMCPOINTING2 is visible.This figure was generated with the aplpy Python package.(b): The second pass of SMCPOINTING1.The regions displayed are as detailed in Figure B1a.Due to the larger beam sizes, a candidate SNR (red circle) and part of a SNR newly discovered in Cotton et al. (2024) (brown circle) were observed in the central tiling.The position error area of known pulsar J0043−73 was tiled with beams overlapping at 50 per cent sensitivity (covering two untargeted steep spectrum point sources).This resulted in the detection and localisation of the latter (see subsection 4.4) as well as the discovery of PSR J0044−7314 (purple beam, see subsection 4.1).A single beam, pictured at 50 per cent sensitivity, was placed on the seeKAT localisation position of PSR J0048−7317.
The second pass of SMCPOINTING2.

Figure B2 .
Figure B2.(a):The first pass of SMCPOINTING2.The regions displayed are as detailed in FigureB1a.Beam tilings overlapping at 75 per cent sensitivity are positioned on five SNRs (red ellipses).The remaining beams were tiled with the same overlap at boresight, which resulted in the discovery of PSR J0054−7228 (purple coherent beam, see subsection 4.1).One SNR observed in SMCPOINTING6 is visible in the bottom left corner as a small red ellipse.One steep spectrum point source and one HMXB with period under 10 s (SXP 6.88) were observed without being targeted (see Table3).The two SMC X-ray pulsars that are not accretion-powered are visible on this plot: the magnetar CXOU J010043.1−721134(cyan box, observed in later pointings) and PSR J0058−7218(Maitra et al. 2021) in PWN IKT 16 (green box with PWN size as a green ellipse inside,Maitra et al. 2015).The latter was targeted with a single beam, for which the 50 per cent sensitivity contour is shown (seeCarli et al. 2022).Similarly to the first pass of SMCPOINTING1, a beam was placed on the published position of PSR J0052−72 but its position error due to the Murriyang beam size (large pink box) was not taken into account, resulting in its non-detection in this pass of the pointing.(b): The second pass of SMCPOINTING2.The regions displayed are as detailed in FigureB2a.Due to the larger beam sizes, part of one SNR fromCotton et al. (2024) (brown circle) was observed in the central tiling.The position error area of known pulsar J0052−72 was tiled with beams overlapping at 50 per cent sensitivity (covering an untargeted steep spectrum point source).This resulted in the detection and localisation of this pulsar (see subsection 4.4).A single beam, pictured at 50 per cent sensitivity, was placed on the seeKAT localisation position of PSR J0054−7228.

Figure B3 .
Figure B3.(a):The first pass of SMCPOINTING3.The regions displayed are as detailed in FigureB1a.The central tiling covers the SNR DEM S5 (central red ellipse).A single beam was nevertheless placed on its associated PWN (green box,Alsaberi et al. 2019).The PWN, PSR J0045−7319 (small pink box), PSR J0044−7314 (discovered in the second pass of SMCPOINTING1, see FigureB1b) and seven steep spectrum point sources (orange boxes, see TableA1) were targeted with single beams shown with their 50 per cent sensitivity contour, even those that fell in the central tiling for this pointing.Three pulsars were discovered in this pointing (purple beams): PSR J0040−7326, PSR J0040−7337 (with the single beam on the PWN the strongest detection), and nearby pulsar J0040−7335 in the main tiling.A smaller coherent beam tiling overlapping at 70 per cent sensitivity was placed around the detected position of PSR J0043−73 (from the second pass of SMCPOINTING2, see FigureB2b) to localise it further than its large Murriyang position error (large pink box, see subsection 4.4).SXP 7.59(Hong et al. 2017) was observed without being targeted.SNRs observed in both passes of SMCPOINTING1 are seen in the top left corner.(b): The second pass of SMCPOINTING3.The regions displayed are as detailed in FigureB3a.The setup of this observation failed, and used the full array instead of the core, resulting in a single small coherent beam tiling overlapping at 80 per cent.MNRAS 000,1-32 (2024)

Figure B4 .
Figure B4.(a): The first pass of SMCPOINTING4.The regions displayed are as detailed in Figure B1a and Figure B2a.Three SNRs (red) are tiled with beams overlapping at 70 per cent sensitivity.The PWN within IKT 16 (green box with PWN size inside), the magnetar CXOU J010043.1−721134(cyan box), and nine steep spectrum point sources (orange boxes, see TableA1) were targeted with single beams shown at their 50 per cent sensitivity contour, even those that fell in the central tiling for this pointing.NGC 361 (thick black circle) was omitted due to being in a lower sensitivity area of the incoherent beam.At the bottom centre and right of the figure, SNRs and the error box of PSR J0052−72, both observed in SMCPOINTING2 are shown.At the bottom left, SNRs observed in SMCPOINTING5 (see FigureB7a) appear.HMXBs with a period of under 10 s (dark blue boxes) were not yet targeted and none was observed in this pointing.(b): The second pass of SMCPOINTING4.The regions displayed are as detailed in FigureB3a.The setup of this observation failed, resulting in a small core array coherent beam tiling overlapping at 80 per cent.

Figure B5 .
Figure B5.(a): The first pass of SMCPOINTING6.The regions displayed are as detailed in Figure B1a and Figure B2a.Three SNRs (red) are targeted with coherent beam tilings overlapping at 70 per cent sensitivity.One HMXB with a period of under 10 s (dark blue box) and 10 steep spectrum point sources (orange boxes) are targeted with single beams that are outside tilings (50 per cent sensitivity contours displayed).The central coherent beam tiling, overlapping at 50 per cent sensitivity, covers a further five untargeted steep spectrum point sources.It is interesting to note that steep spectrum point source EMU-ESP-SMC-5407 is observed inside SNR J0112−7304, newly discovered by Cotton et al. 2024 (see Table 9).Other new SNRs from Cotton et al. (2024) are visible in the top right corner but were not observed in this survey due to not being known at the time.A candidate SNR from Titus et al. (2019) observed in SMCPOINTING7 is visible at the bottom of the figure.In the top left of the figure, a small pink box shows the known pulsar J0113−7220, with its position error inside.Other targets observed in SMPOINTING5 (see Figure B7a) are visible in the top middle of the figure.(b):The second pass of SMCPOINTING6.The regions displayed are as detailed in FigureB5a.MNRAS 000,1-32 (2024) Figure B5.(a): The first pass of SMCPOINTING6.The regions displayed are as detailed in Figure B1a and Figure B2a.Three SNRs (red) are targeted with coherent beam tilings overlapping at 70 per cent sensitivity.One HMXB with a period of under 10 s (dark blue box) and 10 steep spectrum point sources (orange boxes) are targeted with single beams that are outside tilings (50 per cent sensitivity contours displayed).The central coherent beam tiling, overlapping at 50 per cent sensitivity, covers a further five untargeted steep spectrum point sources.It is interesting to note that steep spectrum point source EMU-ESP-SMC-5407 is observed inside SNR J0112−7304, newly discovered by Cotton et al. 2024 (see Table 9).Other new SNRs from Cotton et al. (2024) are visible in the top right corner but were not observed in this survey due to not being known at the time.A candidate SNR from Titus et al. (2019) observed in SMCPOINTING7 is visible at the bottom of the figure.In the top left of the figure, a small pink box shows the known pulsar J0113−7220, with its position error inside.Other targets observed in SMPOINTING5 (see Figure B7a) are visible in the top middle of the figure.(b):The second pass of SMCPOINTING6.The regions displayed are as detailed in FigureB5a.MNRAS 000,1-32 (2024)

Figure B6 .
Figure B6.(a): The first pass of SMCPOINTING8.The regions displayed are as detailed in Figure B1a.A SNR (red) is observed in the central tiling overlapping at 50 per cent sensitivity.Ten steep spectrum point sources (orange) outside this tiling were targeted with single beams (50 per cent sensitivity contour displayed).Two were covered by the central tiling.PSR J0131−7310 (pink box with position error inside) was observed with a single coherent beam (50 per cent sensitivity contour shown).The cluster Lindsay 110 (thick black circle) is tiled with 70 per cent sensitivity overlap.However, sensitivity is limited in this area, as it is beyond the 50 per cent sensitivity contour of the incoherent beam at the central L-band frequency (large thin black circle).A new Cotton et al. (2024) SNR is displayed as a brown ellipse but was not known at the time.(b): The second pass of SMCPOINTING8.The regions displayed are as detailed in Figure B6a with two additional steep spectrum point sources observed in the central beam tiling due to the larger beam size.

Figure B7 .
Figure B7.(a): The single observation of SMCPOINTING5.The regions displayed are as detailed in Figure B1a and Figure B2a.Two SNRs (red) were observed with tilings of coherent beams overlapping at the 70 per cent sensitivity level, while four were observed with the central tiling overlapping at 50 per cent sensitivity.PSR J0105−7208 was discovered with the strongest S/N in the coherent beam highlighted in purple (see subsection 4.1).The SMC magnetar CXOU J010043.1−721134 and four steep spectrum point sources (orange) were targeted with a single beam (50 per cent sensitivity contour shown).No fast-spinning HMXBs (blue boxes) were thought close enough to boresight to target.Two steep spectrum point sources were observed in the main tiling without being targeted.The large position error area of a pulsar candidate from Titus et al. (2019) (pink dashed box) was not entirely observed, instead focusing on the cluster NGC 416 (thick black circle) with a tiling of coherent beams overlapping at the 70 per cent sensitivity level.Two SNRs (red) observed in SMCPOINTING6 are visible in the bottom centre of the figure, while the right side of the figure contains several targets from SMCPOINTING2.(b): The single observation of SMCPOINTING7.The regions displayed are as detailed in Figure B1a and Figure B2a.All beams are tiled at 50 per cent sensitivity overlap, covering a central zone, a candidate SNR (red), four steep spectrum point sources (orange), and one fast-spinning HMXB, SXP 0.72 also known as SMC X-1 (blue boxes).In the top right corner, targets observed in SMCPOINTING6 are visible.Single beams are placed on one fast-spinning HMXB and nine steep spectrum point sources.MNRAS 000,1-32 (2024)

Figure B8 .
Figure B8.The extra pointing SMCPOINTING9 consists of an ellipse tiling of coherent beams which 50 per cent sensitivity contour is shown, covering three steep spectrum point sources and part of a SNR newly discovered in Cotton et al. (2024) (brown ellipse).The regions displayed are as detailed in Figure B1a.Single beams were placed on 13 steep spectrum point sources (orange) and two fast-spinning HMXBs (dark blue boxes), as well as PSR J0044−7314, PSR J0048−7317 and PSR J0054−7228 discovered earlier in the survey.The SNR above the tiling (red) was observed in SMCPOINTING2 while the sources in the bottom right corner were observed in SMCPOINTING1.

Table 4 .
(Price 2016) (2017)tivity calculations parameters.The sky temperature in the direction of the SMC was retrieved fromZheng et al. (2017)using pyGSM(Price 2016).The given value of  is the median intrinsic duty cycle from the ATNF pulsar catalogue, for radio-emitting pulsars with a measured FHWM, excluding Globular Cluster pulsars, Millisecond Pulsars, Rapidly Rotating Radio Transients, magnetars and binary pulsars.

Table 5 .
The parameters of Murriyang SMC surveys used in this work's flux density sensitivity upper limit calculations.The system temperature inMcCulloch  et al. (

Table 6 .
The NGC 121 observations' parameters.The pointing position is given in Right Ascension and Declination.The coherent beam ellipse semi-axes are given in arcseconds in the Right Ascension and Declination directions, with the ellipse position angle.