A&A615, A98 (2018) 
https://doi.org/10.1051/0004-6361/201832837 
©ESO 2018 
Astronomy&AstrophysicsInvestigationofthe cosmicraypopulation and magnetic feld strengthinthe haloofNGC891? <BR>
D.D. Mulcahy1,2,A. Horneffer2,R. Beck2,M.Krause2,P.Schmidt2,A. Basu2,3, 
K.T.Chy˙zy4, 
R.-J.Dettmar5,M.Haverkorn6,G. Heald7,V. Heesen8,C. Horellou9,M. Iacobelli10, 

B. 
Nikiel-Wroczy´, 


nski4,R.Paladino11,A.M.M. Scaife1, SarrveshS.Sridhar10,12,R.G.Strom10,13 <BR>
F.S.Tabatabaei14,15,T. Cantwell1,S.H. Carey16,K.Grainge1,J. Hickish16,Y.Perrot16,N.Razavi-Ghods16, P. Scott16, andD.Titterington16 <BR>
1 Jodrell Bank Centre for Astrophysics, Alan Turing Building, School of Physics and Astronomy, The University of Manchester, OxfordRoad, ManchesterM139PL,UK 2 Max-Planck-InstitutfRadioastronomie,AufdemHel69,53121Bonn,Germany 
e-mail: rbeck@mpifr-bonn.mpg.de <BR>3Fakultät fPhysik,Universität Bielefeld,Postfach100131, 33501Bielefeld, Germany 4 Astronomical Observatory, JagiellonianUniversity, ul.Orla171, 30-244Krakow,Poland 5 Astronomisches Institut derRuhr-Universität Bochum,Universitätsstr.150,44780 Bochum, Germany 6 DepartmentofAstrophysics/IMAPP,RadboudUniversity,POBox9010, 6500Nijmegen, TheNetherlands 7 CSIROAstronomyand Space Science,POBox1130, Bentley,WA6102,Australia 8Universityof Hamburg, HamburgerSternwarte, Gojenbergsweg112,21029 Hamburg, Germany 9 Department of Space, Earth and Environment, Chalmers University of Technology, Onsala Space Observatory, 439 92 Onsala, 
Sweden 10 NetherlandsInstituteforRadioAstronomy(ASTRON),Postbus2, 7990 Dwingeloo,TheNetherlands 11 INAF/IstitutodiRadioastronomia,Via Gobetti101,40129 Bologna,Italy 12 KapteynAstronomical Institute,UniversityofGroningen,POBox 800,9700Groningen, TheNetherlands 13 Astronomical Institute ‘Anton Pannekoek’, Faculty of Science, University of Amsterdam, Science Park 904, 1098 Amsterdam, 
TheNetherlands 14 Instituto de Astrofísica de Canarias, Vía Láctea S/N, 38205 La Laguna, Spain 15 DepartamentodeAstrofísica,UniversidaddeLa Laguna, 38206La Laguna, Spain 16 AstrophysicsGroup,Cavendish Laboratory,19J.J. ThomsonAvenue, Cambridge CB3 0HE,UK 
Received15February2018/Accepted29 March2018 

ABSTRACT 
Context. Cosmicrays and magnetic feldsplayanimportantrolefortheformation and dynamicsofgaseous halosofgalaxies. Aims. Low-frequency radio continuum observations of edge-on galaxies are ideal to study cosmic-ray electrons (CREs) in halos via radio synchrotron emission and to measure magnetic feld strengths. Spectral information can be used to test models of CRE propagation.Free–freeabsorptionbyionizedgasatlowfrequenciesallowsustoinvestigatethepropertiesofthewarmionized medium inthe disk. Methods. Weobtainednew observationsoftheedge-onspiralgalaxyNGC891 at129–163MHzwith theLOwFrequencyARray (LOFAR)andat13–18GHzwiththeArcminuteMicrokelvinImager(AMI)and combinethemwithrecenthigh-resolutionVeryLarge Array(VLA)observationsat1–2GHz, enablingustostudytheradio continuum emissionovertwoordersof magnitudeinfrequency. Results. Thespectrumoftheintegratednonthermalfuxdensitycanbe fttedbyapowerlawwithaspectralsteepeningtowardshigher frequencies orby a curved polynomial. Spectral fattening at low frequencies due to free–free absorption is detected in star-forming regions of the disk. The mean magnetic feld strength in the halo is 7 ± 2 µG. The scale heights of the nonthermal halo emission at 146MHz are larger than those at 1.5 GHz everywhere, with a mean ratio of 1.7 ± 0.3, indicating that spectral ageing of CREs isimportantandthatdiffusivepropagation dominates.The halo scale heightsat146MHz decreasewith increasing magnetic feld strengths whichisa signatureof dominating synchrotron lossesof CREs.Ontheother hand,the spectral indexbetween146MHz and 1.5GHz linearlysteepens from the disk to the halo, indicating that advection rather than diffusion is the dominating CRE transport process. This issue callsforrefned modellingof CREpropagation. Conclusions. Free–free absorptionisprobablyimportantatandbelow about150MHzinthedisksofedge-ongalaxies.Toreliably separatethethermaland nonthermal emission components,toinvestigate spectralsteepeningduetoCRE energy losses,andtomeasure magnetic feldstrengthsinthe disk and halo, widefrequency coverage and high spatialresolution are indispensable. 
Keywords. galaxies: halos–galaxies: individual:NGC891–galaxies: ISM –galaxies: magnetic felds– cosmicrays – radio continuum:galaxies 
? LOFAR and AMI images (FITS fles) are onlyavailable atthe CDS via anonymousftpto cdsarc.u-strasbg.fr <BR>(130.79.128.5)or via http://cdsarc.u-strasbg.fr/viz-bin/qcat?J/A+A/615/A98 <BR>
Article publishedbyEDP Sciences A98, page1of 21 A&A615, A98 (2018) 
1. Introduction 
Magnetic feldsand cosmicraysare dynamically relevantinthe disksof spiralgalaxies becausethe magnetic energy densityis similartothekinetic energy densityof turbulencebutlargerthan thethermal energy density (e.g. Beck2015).Partofthe energy inputfrom supernovaremnantsgoesintothe accelerationof cosmicraysandintotheamplifcationof turbulent magnetic felds. This scenariocanexplainthetightcorrelationbetweenradioand far-infrared emissionthat holdsfortheintegrated luminositiesof galaxies aswell asforthe local intensities withingalaxies (e.g. Tabatabaei etal. 2013, 2017;Heesen etal. 2014). 
The physical relationships between the various components ofthe interstellar medium (ISM) are less understood ingaseous galaxy halos. Continuum emission from thick disks or halos is observedfromtheradiototheX-rayspectralranges,aswellas HIradio line emissionof neutralhydrogen andoptical emission lines of ionized gas. As sources in the disk probably provide mostofthe energy,warm andhotgas, cosmicrays, and mag-netic felds have to be transported from the disk into the halo. The required pressure could be thermal or nonthermal (cosmic rays and magnetic felds).Possible transport mechanisms area “galactic wind”withavelocitysuffcientfor escape(Uhligetal. 2012),a“galacticfountain” withcoldgasreturningtothegalaxy (Shapiro&Field1976;Fraternali2017),or“chimney” outbreaks drivenbyhot superbubbles(Norman& Ikeuchi1989). Cosmic rays canpropagaterelativetothegas with avelocity limitedto the Alfvén speedbythestreaming instability(Kulsrud&Pearce 1969)or can diffuse along or across the magnetic feld lines (Buffe et al. 2013). Especially unclear is the role of magnetic felds in outfows.Aturbulent feld maybe transportedtogether withthegas in an outfow while an ordered feld can support or suppress the outfow, depending on its strengthand orientation. This is complicated further duetothe interplayofthe magnetic feld with an inhomogeneous outfow in which isotropic turbu-lent felds are converted into anisotropic turbulent felds due to shearandcompression,thuscreatinganordered feldinthehalo (Elstneretal.1995;Moss&Sokoloff2017). 
Edge-ongalaxies areideal laboratoriestostudythe disk–halo connection andtoinvestigatethedrivingforcesof outfows. The discovery of a huge radio halo around NGC 4631 by Ekers & Sancisi(1977)revealedtheimportanceof nonthermalprocesses. Thetheoryof cosmic-raypropagationintothe haloswasdeveloped ingreat detail(Lerche& Schlickeiser1981,1982;Pohl &Schlickeiser1990), but comparisons with observationswere inconclusiveduetothe limitedqualityoftheradiodataatthat time. 
Recently,thistopic has beenrevived withhigh-qualityradio spectral index maps becoming readily available. This is driven both by the advent of broadband correlators at existing inter-ferometric telescopes such astheAustraliaTelescope Compact Array (ATCA) and the Very Large Array (VLA), operating at GHzfrequencies,aswellasthearrivalof entirely newfacilities such astheLOwFrequency ARray(LOFAR)thatopensa hithertounexplored windowonradio halosatMHzfrequencies.Asa promisingstep, Heesenetal. (2016),usinga1D cosmicraytrans-port model SPINNAKER (SPectralINdexNumerical Analysis of K(c)osmic-ray Electron Radio-emission), were able to extract propertiesofthe cosmic-raypropagationinthe halooftwoedgeongalaxies1.Asakey result,theycould showthatthevertical proflesofthe spectralindexcanbeusedtodistinguishbetween advection-and diffusion-dominated halos,the latterrepresenting the case of no signifcant outfows. Extending the sample 
1 www.github.com/vheesen/Spinnaker <BR>
Table 1. Observational dataofNGC891. 
Morphologya Sb Positionofthe nucleusb RA(J2000) =02h22m33.s4 
Dec(J2000) = +42◦ 200 5700 Position angle of major axisc 23◦ (0◦ is north) Inclination of diskd 89.8◦ (0◦ isface on) Distancea 9.5 Mpc(100 ≈ 46 pc) Star-formation ratee 3.3 M&#12; yr−1 Total massc 1.4× 1011 M&#12; 
References. (a) van der Kruit&Searle (1981). (b)Vigottiet al.(1989). 
(c) Oosterlooet al. (2007). (d) Xilouriset al. (1999). (e) Arshakian et al. (2011). 
to12edge-on spiral andMagellanic-typegalaxies, Krauseetal. (2018)and Heesenet al. (2018a)showedthatmany halos are advection dominated with outfow speeds similar to the escape velocity, raising the possibility of cosmic ray-driven winds inthem. 
Aprimetargetfor such studiesisNGC891 thatisafairly nearbyedge-on spiralgalaxy.NGC891is similar to our own MilkyWayin termsofoptical luminosity(deVaucouleursetal. 1991), Hubble type (Sb; van der Kruit & Searle 1981), and rotationalvelocity (225kms−1;Rupen1991), but hasa highstar<BR>formation rate (3.3 M&#12; yr−1, Arshakian et al. 2011)compared to the Milky Way (the Galactic value is 1.66 ± 0.20 M&#12; yr−1; Licquia&Newman2014);thisisin accordance with thepresenceof approximatelytwicethe amountof moleculargasofthe MilkyWay, withtheradial distributionofCO remarkablysimilarinbothgalaxies(Scovilleet al.1993). Dueto itsproximity andvery high inclination,NGC891is an observational testing groundforthestudyof disk and halo interactions andthegalactic halo. Thephysical parametersofNGC891 are summarized inTable 1. 
NGC891possessesabright,well-studied haloandfor which a plethora of ancillary data from various gas components is available. Randet al.(1990)andDettmar (1990)independently detected diffuse Hα emission from ionized gas up to 4kpc distance from the galaxy’s plane with an exponential scale height of about 1kpc. A huge halo of neutral atomic H I gas withupto22kpcextent was observedby Oosterloo et al. (2007). Howk& Savage (1997)detected prominent dust lanes emergingverticallyintothe haloofNGC891whichcouldpartly be associated with energetic processes connected to massive starformation inthe disk. Sofue(1987)interpreted such dust lanes as tracers of vertical magnetic felds. Diffuse line emission fromCOmoleculesis observedupto about1kpc distancefrom the plane(Garcia-Burilloet al.1992). Infrared emissionfrom polycyclic aromatic hydrocarbons (PAH) particles and warm dust is detected to about 2.5 kpc from the plane with similar scale heights of about 300pc(Whaley et al. 2009), excited by photons escaping from the disk. Emission from cold dust inthe sub-millimetrerangeextendsupto about2kpc(Alton et al. 1998)and has a larger scale height than the warm dust. DiffuseX-rayemissionfromhotgaswasfoundinthe haloupto 4kpcfromthe plane(Bregman&Pildis1994;Hodges-Kluck& Bregman 2013). 
NGC891 has also been extensively observed in radio continuum emissionthroughoutthe pastfew decades. The frst extensiveinterferometricinvestigationinradio continuumat610, 1412and 4995MHz withtheWesterborkSynthesisRadioTele-scope(WSRT)by Allenetal. (1978)revealedastrongsteepening 
A98, page2of 21 

D.D. Mulcahyetal.: Cosmicrays and magnetic feldsinthe haloofNGC891 
of the spectral index in the halo, but this result was affected by missing large-scale emission in the 4995 MHz image. Hummelet al. (1991) observedNGC891 at 327MHz and 610MHz withtheWSRTand at1490MHz withthe VLA. The spectral indexbetween610MHz and1.49GHz witharesolution of 4000 showed that the inner and outer radio disks (at small and large distances from the centre) have signifcantlydifferent spectra,partlyduetothelargerthermalfractioninthe innerdisk. The spectral steepening towards the halo is mild on the eastern side butsteep onthewestern side. Theradio disk at1490MHz was describedby anexponential scale heightof1.2kpc (scaled tothe distance adoptedinthis paper, seeTable 1). Beck et al. (1979)andKleinetal. (1984)observedNGC891at8.7GHzand 
10.7GHzwiththe100-mEffelsbergtelescope.Whencomparing theirdatatothoseat610MHz,theyfoundonlya mildsteepening ofthe spectral index2 from α ≈−0.75 inthe diskto about −0.9 inthe halo and no furthersteepening untilupto6kpc abovethe galaxy’splane (scaledtothedistancetoNGC891adoptedinthis paper).The spectralindexinthehaloofNGC891allowedforthe frsttimeacomparisonwithCREpropagation models.Themodelsof diffusion and advectionby Strong(1978)predicted almost lineargradientsof spectral index whichwerein confict withthe observationsby Allenet al. (1978), but consistent withthoseby Beck et al. (1979)andKleinet al. (1984). Dumkeet al. (1995) observed a sample of edge-on galaxies, including NGC891, withtheEffelsbergtelescopeat10.55GHz.Theverticalprofle was describedby twoexponential scale heightsof270pc and 
1.8kpcforthe disk and halo,respectively(Krause2012). 
Thestructureofthe magnetic feldinthe haloofNGC891 has beeninvestigatedthrough measurementof linearlypolarized radio synchrotron emission. The aforementioned observations of Dumkeet al. (1995)withtheEffelsbergtelescopeat10.55GHz revealed diffuse polarized emissionfromthe disk with an orientationpredominatelyparalleltothe plane. Such a feldstructure istobeexpectedfrom magnetic feld amplifcationbythe action of a mean-feld αΩ-dynamointhe disk (e.g. Beck et al.1996) or from a small-scale dynamo in a differentially rotating disk (Pakmoretal.2014)orfrom shearingofan initiallyvertical feld (Nixonetal.2018).FromEffelsberg observationsofNGC891 at 8.35GHz, Krause (2009)showedthatthe large-scale orientation of the halo magnetic feld in the sky plane appears to be “X-shaped”. Sucha feld structure could arise from a mean-feld dynamo including agalactic wind(Mosset al.2010)orfrom a helical feld generated by a velocity lag of the rotating halo gas(Henriksen&Irwin2016).Withthe aidofFaraday rotation measures, Krause (2009)alsofound indicationofa large-scale regular magnetic feld withinthe diskofNGC891, likelypartof a spiral magnetic feld. 
Israel&Mahoney (1990)foundthatthe integrated fux den-sities of 68 galaxies at 57.5 MHz are systematically below the extrapolation from measurements at 1.4 GHz if one assumes a power-law spectrum with a constant slope. They also reported thatthe57.5MHz fux densityisa functionof inclinationangle of the disk with respect to the sky plane, with lower fux den-sities observedforgalaxies withlarger inclination angles which they interpreted as increasingfree–free absorption causedby a clumpy medium with an electron temperature of Te ≈ 1000K and an electron density of order 1cm−3. However, no direct observational evidence of such a medium exists, not even in our own Galaxy. Hummel (1991), re-analyzing the data of Israel& Mahoney (1990), and Marviletal. (2015)observed a fattening in the integrated spectra of nearby galaxies, but 
Iν ∝ να where Iν isthe intensityatfrequency ν. 
found no dependence on inclination; this makes it less likely that free–free absorption is the cause of this fattening. Therefore,low-frequency observationsof nearbygalaxies, specifcally ofedge-ongalaxies, needtobeperformedto clearup someof these contradictions. 
The LOw Frequency ARray (LOFAR; van Haarlem et al. 2013)opened a new era of studying the diffuse, extended radio continuum emissionin halosof nearbygalaxies andtheir mag-netic felds –thestudyofwhichhassofarbeenhamperedatGHz frequenciesbyspectralageingof CREs.TherangeofshortbaselinesoftheHighBand Antenna(HBA)Arrayallowthedetection ofextended emissionfrom nearbygalaxies.To date,results on twofewstar-forminglow-inclinationgalaxies withLOFARwere published(Mulcahy et al.2014; Heesenet al.2018b), butthe natureof halosaroundedge-onspiralgalaxiesstill needstobe investigated at low radio frequencies. 
Inthis paper, wepresent and analysethe frstLOFAR observationsofNGC891with theHBA array ata centralfrequency of 146MHz. These data are complemented by observations from the Arcminute MicroKelvin Imager (AMI) at 15.5 GHz, thus providing us with two decades of radio frequency cover-age. This paper is organized as follows: In Sects. 2 and 3, we describethe observational setup oftheLOFAR and AMI observations along with the data reduction and imaging process. In Sect.4,wepresentthe mapsofNGC891atboth146MHzand 
15.5GHz;thisisfollowedbya discussionin Sect. 5onthe mor-phology of the galaxy in comparison with other wavelengths; here,wepayparticular attentiontobroadband observationswith the VLA at central frequencies of 1.5 and 6GHz from the CHANG-ES survey(Continuum Halos inNearbyGalaxies: An EVLA Survey; Irwin et al. 2012). In Sect. 6, we present the separation of thermal and nonthermal emission components, discuss the spectrum of the integrated nonthermal emission, and present maps of the total and nonthermal radio spectral indices. Estimates of the magnetic feld strength are given in Sect. 7. In Sect. 8, we measure the scale heights of the non-thermal emission in the halo at different distances along the projectedmajor axisofthegalaxy.We discusstheimplications of our fndingsin Sect. 9:free–free absorption, energy losspro<BR>cesses of CREs, and CRE propagation.A summaryis given in Sect.10. 
2.LOFAR observations and dataprocessing 
2.1. Observational setup and data selection 
The observationsofNGC891 were donein interleaved mode, switching between scans on the calibrator 3C48 (at RA(J2000) = 01h34m41s3, Dec(J2000) = +33◦0903500) and the target NGC891.Atotalof44stationswere usedforthis observation, of which32were corestations and12wereremotestations.Full detailsofthe observational setup are showninTable 2. 
Two of the scans on the target were not recorded properly and one had excessive radio-frequency interference (RFI), so that these scans were discarded after pre-processing. After sub-tracting overheads for the calibrator observations and for the switching of the beam positions, the remaining on-source inte-gration timewas5.6h. Duringfacetcalibration (see Sect. 2.2.2), the self-calibration solutions of the calibrator for the target facet, containingNGC891, did not convergefor frequencies >163.6 MHz, so those data were excluded as well. Hence, the resulting bandwidth that was used for the imaging was 
34.4 MHz. Finally, the automatic fagging in the pipelines faggedafurther15%ofthe visibilities, mostlydataaffectedby 
A98, page3of 21 


Fig. <BR>1. <BR>Imageofthe compact source 3C66Aandtheextended source 3C66Bafterthe initial direction-independent calibration(left panel)and afterthe direction-dependentfacetcalibration(right panel).Bothimagesare smoothedtothe sameresolutionof 1400 and are displayed onthe same colour scale.Thestrongartefactsaround3C66Aafterthedirection-independent calibrationintheleftmapare causedbythe calibrationnotbeing adequateforthis direction.After calibrating specifcallyforthis directionthese artifacts are mostlygone. 
Table 2. ParametersoftheNGC891LOFARHBAobservations. 
Startdate(UTC) 31March2013/09:00 Enddate(UTC) 31March2013/16:59 Interleaved calibrator 3C48 Scan length on calibrator 2min Scan length ontarget 15.8min Duration of observations 8h Final time ontarget 5.56h(21scans) 
Frequency range 129.2–176.8 MHz Final frequency range 129.2–163.6 MHz Total bandwidthontarget 47.6MHz Final bandwidthontarget 34.4MHz Reference frequency 146.38 MHz 
RFI that were identifed as spikes when plotting amplitudes as function of time or frequency. 
Most of the short baselines between the “ears” of the core stations were fagged to avoid cross-talk. The remaining shortest baselinesbetweenthe corestationsof about50m ensure the detection of large diffuse emission on scales of up to2◦ . 

2.2. Calibration 
Low-frequency, wide-feld imaging data are subject to several effects that are usually negligible at higher frequencies, most notablydistortions of the Earth’s ionosphere and artifacts caused by the large number of bright background sources. In order to overcome these diffculties, we applied the novel tech-nique of facet calibration. This method uses three main steps. First, a direction-independent calibration of the data with the prefactor pipeline that applies the fux density scale, corrects the Faraday rotation of the Earth’s ionosphere, corrects for instrumental effects such as clock offsets between different LOFAR stations, and performs an initial round of phase calibra-tion; second, the subtraction of all sources that are present in the supplied sky model with the Initial Subtract pipeline; last,the applicationofthe Factor pipeline. This pipeline divides the feldofviewintoa mosaicof smallerfacetsusing Voronoi tesselation (e.g. Okabe etal. 2000). 

2.2.1. Initial calibration 
To calibrate our LOFAR datasetofNGC891,we usedaprototype version of the prefactor3 pipeline for the initial, direction-independent calibration and Version 1.0pre of Factor4forthe direction-dependentfacetcalibration. 
The dataofthe calibrator andtarget were frstpre-processed with the New Default Pre-Processing Pipeline (NDPPP) that includes RFI excision with aoflagger (Offringa et al. 2010, 2012),removalofedge-channels,andaveragingto4s time-and 49 kHz (4 channels per subband) frequency-resolution. 
Thedataofthe calibrator observationswerethen calibrated against a known model characterizing the detailed frequency dependence ofthe fux density of 3C48(Scaife&Heald 2012). This modelthus setsthe fux density scaleof our data. Thegain solutions from this calibration were then used to extract instrumental calibration parameters: gain amplitudes, station clock delays, and phase offsets between the X-and Y-dipoles within a station. 
These solutions were copied to the target data in order to correctforinstrumentaleffects.To dealwiththeeffectsofstrong off-axis sources,wepredictedthe visibilitiesofthefourstrongest sources (CasA,CygA,TauA, andVirA) and fagged all vis-ibilities to which they contributed more than an apparent (not correctedfor primary-beam attenuation) fux density of5Jy5. Thenthe datawereaveragedtothe fnalresolutionof8s and 98kHz(2channels per subband) and concatenated intogroups of 11 subbands to form 16 bands with 2.15MHz total band-width each. This resolution is suffcient to avoid decorrelation due to rapid changes in the ionospheric phase, while keeping thedatavolume manageableat14.5GByteper band.An additional round of automatic RFI excision with aoflagger was then performed onthe concatenated data, becausethe algorithm is more effective in detecting RFI in data sets withwide band-widths as compared withsingle subbands.Asa fnalstepforthe direction-independent calibration,the datawere phase calibrated 
3  https://github.com/lofar-astron/prefactor <BR> 
4  https://github.com/lofar-astron/factor <BR> 
5  This threshold was found to be the best compromise between fag- 

gingtoo muchdata andthus increasingtherms noise and faggingtoo fewdataandthusnotremovingtheresidual sidelobesofthese sources (vanWeeren,priv. comm.). 
A98, page4of 21 

D.D. Mulcahyetal.: Cosmicrays and magnetic feldsinthe haloofNGC891 

Fig. <BR>2. <BR>Imageofthe fullFoVoftheLOFAR observationsafter initial calibration, withNGC891inthe centre, 3C66A/B abovethe centre, and 3C65attheloweredgeoftheimage.Overlaidisthefacetingschemethatwasusedforthefacetcalibration:inorangethesquaresofthe calibrator regionsthatwere usedto calibratethefacets andingreentheresultingfacetsthat are defnedby Voronoi tessellation aroundthe calibrators. 
on a model generated from the LOFAR Global Sky Model (GSM; van Haarlem etal. 2013). 
Finally, the 16 bands were imaged separately, frst at a medium resolution (outer uv cut at7kλ, about 2000 resolution) then – after subtracting the sources found in the medium res-olution images – at a lower resolution (outer uv cut at 2kλ, about 1.50 resolution). The feld of view (FoV) of the medium resolution images was 2.5 times the full width to half maximum (FWHM) of the station beam (between 9.4◦ and 12.8◦ depending on the frequency of the band) and the FoV of the low-resolution images 6.5 timesthe FWHMofthestation beam (24.5◦–33.3◦).Thereasontocreatedifferent kindsofimagesisto pickuplow surface-brightness emission andtobe ableto image a largerFoV withoutprohibitive computingrequirements. The fnal result is the combined list of sources from both imaging steps,theresidual visibilitiesin whichalldetected sourceshave been subtracted,andthephase solutionsfromthelastcalibration step,for eachofthe16bands. 
2.2.2. Facet calibration 
The frst stepofthefacet calibrationisthe selectionofthe cali-bration directionsthat containthefacetcalibrators.Forthisstep andfor whatfollowsweusedthe Factor pipelinethat automates most of the necessary steps. In the frst step, Factor uses the skymodelforthehighestfrequencybandand searchesforstrong and compact sources. Thiswas done withthefollowing selection parameters: 
– 
maximum size of a single source: 20; 

– 
minimum apparent fux density of a single source: 100 mJy; 

– 
maximum distanceof single sourcetobegrouped into one calibrator region: 60; 

– 
minimum total apparent fux density of the sources of a 


calibrator region: 250 mJy. The resulting list of calibration directions was then manually modifedinordertotailoritto our specifcrequirements.Apart from removing some of the weaker calibrators that were too close to each other, we also experimented how to defne best the calibratorregion around 3C66(Fig. 1)–byfarthebright-est source in our FoV. In the end we decided to fully include both 3C66A and 3C66B into the region: while the extended emissionof 3C66Bis notsuitedfor calibration,thereis enough compact fux density in 3C66A to allow for good calibration solutionseven onthe long baselines, whereasexcluding 3C66B made calibration worse, probably because the short baselines couldnotbe calibratedwellasthesky modeldidnotincludethe extended emission. 
The 29 calibration directions after our manual adjustments are shown in Fig. 2. The facets are set up by Voronoi tesselation aroundthe calibrationdirectionuptoa maximum 
A98, page5of 21 A&A615, A98 (2018) 

radius of 2.5◦ in RA and 2.7◦ in Dec6 aroundthe pointing centre. Thefacet boundaries are slightlydefectedtoavoid intersecting withdetected sourcesinthesky model.We defnedaregionwith a radius of 10.20 aroundNGC891 asthetargetfacet, sothat sources,aswellas ourtarget,arenot splitbetweentwofacets. Theresultingfacet boundariesareshownas linesinFig. 2.The onedirection outsidethefacetingradius(seetheorangebox out-sideofthegreen ellipseinFig. 2)has onlya calibrationregion butnofacetassociatedwithitwhichmeansthatonlyasmallarea aroundthe calibratoris imaged. 
The core of Factor is the direction-dependent calibration. For each direction the residual visibilities are phase shifted to the direction of the calibration region and the sources within that region are added back to the visibilities. Then these data are self-calibrated, starting with the direction-independent cal-ibration. The self-calibration has two loops: in the frst loop, Factor only solves for a fast phase term to track the ionospheric delay;in the second loop, Factor solves againforthis fast phase term and simultaneously for a slowly varying gain (phase+amplitude)termtoalso correctforresidualeffectsfrom discrepancies in the beam model and from any other causes. After the calibration region has been calibrated, the full facet is imagedagain:the visibilities arepreparedina similarfashion asthey werefor self-calibration,exceptthat all sources within thefacet are added backin andthatthe data are corrected with the self-calibration solutions. From this image an updated sky modelforthefacetis created which,together with the calibra-tion solutions for this direction, is used to update the residual visibilitiesbysubtractingthedifferencebetweenthenewandthe original model.Thefollowingdirectionsarethenprocessedwith the improved residual visibilities. This way the strong sources can be subtracted frst and relatively weak sources can be used as calibrators. 
Weonlyprocessed onedirectionatatime,startingwith3C66 whichhasthe largest apparent fux density andthusthe highest signal-to-noise ratio, proceeding down in apparent fux density. Orderedinthisway,theregionthat containsNGC891isinthe 12thfacet. Beforeprocessingthetargetfacet,the directlyadjacentfacetswereprocessed,too. Theremainingfacets contribute only little noisetotheregion aroundNGC891, sowestopped processingthere. The visibilities we usedforthe fnal imaging werethe onesthat Factor generatedforthe imagingofthe full targetfacet;this meansthat all detected sources outsidethetargetfacetwere subtracted,sothatitwassuffcienttoimageonly therelativelysmall areaofthetargetfacet. 
2.3. Final imaging 
We used theCommon Astronomy Software Applications7 (CASA; McMullin et al. 2007), Version 4.7, to image the facet containingNGC891. WhilstCASA does not implement the LOFARprimarybeam,NGC891is muchsmallerthanthe size oftheLOFARprimarybeam (3.8◦ FWHMat146MHz) andis located atthe phase centre of our observation. Thus, systematic fux density errors due to the missing primarybeam correction are minimal. 
We created four different images. The frst image, with no uv taper applied and a robustweighting of −1.0, achieves a high resolution of 8.300 × 6.500,suffcienttoresolvevariousfeaturesin the disk and the inner halo. Another version of this image was 
6 Theregionis ellipticalto accountforthe elongationoftheLOFAR primarybeam at lower elevations. 
7 http://casa.nrao.edu <BR>
Table 3. ParametersoftheNGC891AMI observations. 
1stSmall Arraytime (UTC) 7–8 Dec. 2016/20:07–01:50 2nd SmallArraytime(UTC) 9–10Dec.2016/17:23–01:42 LargeArraytime(UTC) 7Dec.2016/19:01–22:51 Flux calibrators 3C48 (LA); 3C286(SA) 
Frequency range 13–18GHz Reference frequency 14.5 GHz 
createdbyconvolutionto 1200 × 1200. Thethirdimage hasa mod-erateresolution, withan outer uv taper of 7kλ,arobustweighting of0,and convolvedto 2000 × 2000.Thefourth, ourlow-resolution image, with an outer uv taper of 2kλ and a robust weighting of +1.0, is best suited to detect the low surface brightness of the extended halo. 
For all images, we performed multi-scale and multifrequency synthesis CLEAN (Hbom 1974; Cornwell 2008; Rau& Cornwell 2011) onthefacet containingNGC891,for which we used CLEAN manually drawn masks. The imaging parameters aregiveninTable 4. 
3. AMI observations and data processing 
3.1. Observational setup 
The Arcminute Microkelvin Imager (AMI) telescope (Zwart et al. 2008)consists of two radio arrays, the Small Array (SA) andtheLargeArray(LA), locatedatthe MullardRadioAstronomyObservatory(Cambridge,UK).TheSAisa compactarray of ten 3.7-m paraboloid dishes and is sensitive to structures on angular scales between 20 and 100. The LA is an arrayof eight 12.8-m dishesandis sensitivetoscalesbetween 0.50 and 30. Each array observes at a frequency range of 13–18 GHz with 4096 channelssplitintotwo bands(Hickishetal.2018). 
TheSAdataweretaken asa single pointing withinterleaved observations of the phase calibrator J0222+4302 and 3C286 usedasa fux densityand bandpasscalibrator.TheLAdatawere taken as a mosaic, consisting of seven pointings arranged on a hexagonalraster centred onNGC891. Individual pointingswere cycledbetweenthe pointing centersevery60s, switchingtothe phase calibrator J0222+4302every10minfor2min.Detailsof these observationsaregiveninTable3. 
3.2. Calibration and imaging 
Calibrationand imagingofthe visibilitiesfrombotharrayswere carried out using CASA. The frst step, in calibrating AMI data, is the generation of the rain gauge correction which is a correction for the system temperature dependent on the weather during the observation. An initial round of fagging was car-ried out on the full resolution data, using the rflag option in CASA’s flagdata task,inordertoremovestrong narrow-band RFI. The datawerethenaveragedfrom2048to64 channelsfor each band. We used the fux density scale of Perley & Butler (2013)for calibrators 3C48and 3C286.Atailored, in-housever<BR>sion of the CASA task setjy was used to correct for the fact that AMI measures single polarization I+Q. An initial round of phase only calibration was performed on the fux calibra-tor which was then used for delay and bandpass calibration. Thiswasfollowedby another phase and amplitude calibration, applying the delay and bandpass solutions on the fy. The cal-ibrated data were used for a second round of fagging, before performing a phase and amplitude calibration on the phase 
A98, page6of 21 

D.D. Mulcahyetal.: Cosmicrays and magnetic feldsinthe haloofNGC891 

Fig. <BR>3. <BR>LOFAR mapsofNGC891 observedata centralfrequencyof146.4MHz with a bandwidthof34.4MHzattwo differentresolutions. Left panel:Resolution of 8.300 × 6.500 (shownbythe flled ellipseinthebottomleft corner). Contoursareat3,5,8,12,18,32,44,64 ×σ where σ = 0.29mJy beam−1istherms noiselevel. The locationof SN1986Jis shownbythe white arrow. Right panel:Resolution of2000 × 2000 (shown bythe flledellipseinthebottomleftcorner).Contoursareat3,5,8,12,18,32,44,64,76 ×σ where σ = 0.8mJy beam−1istherms noiselevel. Thefeatures denotedby“(a)” and “(b)” are discussedin Sect. 4. The colour scaleisin unitsof Jy/beam. 
Table 4. Imaging parametersfortheLOFAR images. 

High-resolution image Medium-resolution image Low-resolution image 
uv taper – 7kλ 2kλ Weighting −10 +1 Angular resolution 8.300 × 6.500 (381 × 299pc2) 2000 × 2000 (920 × 920pc2) 40.000 × 35.700 (1.84 × 1.64 kpc2) 
1.000 3.000 4.000 
Cell size 
calibrator, applying again the delay and bandpass solutions on the fy. After the fux density of the phase calibrator was boot-strapped (e.g. Lepage&Billard1992)the calibrationtableswere appliedtothetargetdata whichwerethenaveragedto8channels for imaging. 
The data from both arrays were imaged with multi-scale CLEAN (Cornwell 2008)and each image was cleaned interactively. TheSAimage withits single pointingwasprimarybeam corrected using the task PBCOR in AIPS with the defned SA primary beam8. Each of the seven LA pointings were imaged separately and converted into FITS fles. These images were combined into a mosaic using the AIPS task FLATN which was then correctedfor attenuationbytheLAprimarybeam. 
4.Results:NGC891at146MHz and15.5GHz 
4.1. The LOFAR images 
TheimagesofNGC891 ata centralfrequencyof146MHzare shown atthe highest resolution(8.300 × 6.500)inFig.3(left), at medium resolution(2000 × 2000)inFig.3(right), andatlowres-olution(40.000 × 35.700)inFig.4. The imagecharacteristics are giveninTable 4. The imageat 1200 × 1200 resolution is shown in the compositeofFig.18. 
The meanroot-mean-square(rms) noiseathighresolutionis approximately σ ' 0.29 mJy beam−1 nexttoNGC891and about 
8 AIPS, the Astronomical Image Processing System, is free software available from NRAO. 
0.26 mJy beam−1 in a quiet region. The noise at medium res-olutionis approximately1.0mJy beam−1 nexttoNGC891 and 
0.8 mJy beam−1 inaquiet region. 
The morphology ofNGC891 at146MHz at lowresolutionisquite similartothatat higherfrequencies(327MHz and 610MHz)as seeninHummeletal. (1991).Theradiohalobulges outinthe northernsectorofthegalaxy. Thisis especiallyevident inthe high-resolution image(Fig. 3left) where diffuse emission extends to the north-east and north-west. It is known that the northernpartofthe diskhasalargerstar-formationratethanthe southernpart(Stricklandetal. 2004). 
Interestingextensions(marked “(a)”and “(b)”inFig. 3)are seen in the western halo which were also observed at higher frequencies (Schmidt 2016, Schmidt et al. in prep.), but are more prominent at lower frequencies. Signs of feature “(a)” are also seen as low brightness emission in Fig. 4. These fea<BR>tures are revisited later with information on their spectrum in Sect.6. 
The easternradio haloofNGC891 displaysa “dumbbell” shape in Fig. 4, similar to what is observed on both sides of the haloofNGC253(Heesenet al. 2009). Such a shape can be a signature of dominating synchrotron losses (see Sect. 9.2). The two extensions on the eastern side of the galaxy out to approximately 5.5 kpc from the major axis are similar to the onesfoundinNGC5775(Soidaetal.2011)andmaybe signs of outfows which ejectgasfromgalaxies and enrich the local intergalactic medium.Betweenthesetwogalactic spurstheradio emissionextends outto about 4.5kpcfromthe plane. The maxi-mumextentofthewestern haloofNGC891(outermostcontour 
A98, page7of 21 


Fig. <BR>4. <BR>LOFAR map ofNGC891 ataresolution of 40.000 × 35.700 shownby the flled ellipseinthebottom left,overlaid onto anoptical imagefromthe DigitizedSkySurvey.Contoursareat 3, 5, 8, 12, 18, 32, 44, 88, 164, 200 × σ where σ = 1.1mJy beam−1istherms noiselevel. 
in Fig. 4)is about 9kpc from the major axis. The extents are measured usingthe 3σ level inFig. 4. 
Theradio diskextendsto about16kpcfromthe centre along themajoraxisinthe northand south.Withthefullextentsofdisk andthehaloof about32kpcalongtheplaneand14.5kpcperpendiculartothe plane,respectively,the halo-to-diskextentratioat 146 MHz is ≈0.45. This is similar to the ratio for NGC 253 at 200 MHz(Kapi´
nska et al. 2017). However, we caution that the halo-to-disk extent ratio is stronglydependent on sensitivity and angularresolution;abetterwaytoquantifythe halo emissionis viathe scale height (see Sect. 8). 
Severalfeaturesinthe diskareresolvedinthe high-resolution image(Fig. 3,leftpanel).The mostintenseradio emissioninthe galaxyis seeninthe centralregionofNGC891andthe north ofthe diskduetothelargerstar-formationrateinthisregionof the disk.Tothe southofthe disk, less intenseradio emissionis observed comparedtothe north. 
The positionoftheradio supernova SN1986Jis indicatedby anarrowinFig. 3(leftpanel).SN1986J(vanGorkometal.1986) is one of the most luminous radio supernovas ever discovered (Bietenholz et al. 2010)and has been studied extensivelysince its discovery.The dateofitsexplosionis uncertain;thebestesti-mate is 1983.2 ±1.1(Bietenholzetal. 2002).WedetectSN1986J in our LOFAR high-resolution image (Fig. 3 left) as an unresolved point source locatedinthe south-westofthe disk.AGaussianftgivesa fux densityat146MHzof 5.5 ± 0.2mJy beam−19 above the background disk emission at the position of RA(J2000) = 02h 22m 30s .8, Dec(J2000) = +42◦ 190 5700 . 
4.2. The AMI images 
The images ofNGC891 ata central frequency of15.5GHz observed withboththe AMI arrays are shown inFig. 5.Forthe SA image the fnal resolution is 14200 × 12100 (6.5 × 5.5 kpc2) and 3600 × 2400 (1.6 × 1.1 kpc2) for the LA image. For the SA and LA images the rms noise is 1.5 mJy beam−1 and 0.11mJy beam−1, respectively. 
9 The modelby Bietenholz&Bartel (2017)predictsa fux densityof 
4.8 mJyfor an age of 30.0y and frequency of146.4 MHz, consistent with our observations. 
The AMISAimage shows no distinctfeaturesforNGC891 duetoitslowresolution.TheAMILAimageis abletoresolve thestar-forming diskofNGC891andis similartotheimageof theHα line emissionfrom ionizedhydrogengas(Fig. 6)andto the infrared image of thermal emission of warm dust at 24 µm (see Fig. 6 in Whaley et al. 2009). The northern star-forming region stands also out in the emission of cold dust at 850 µm (Altonetal.1998), sub-millimetre line emissionfromrotational transitionsoftheCOmolecule,atracerfor molecularhydrogen (Garcia-Burilloetal.1992),andinopticalHα line emission (e.g. Dahlemetal.1994). 
We measure an integrated fux densitythatis onlyabout2% lowerfortheLAimagethanfortheSAimage(avoidingthe negative sidelobes), indicatingthat no signifcant fux densityis lost in the LA image due to missing spacings. This seems surprising because the largest visible structure the LA is sensitive to is onlyabout30. However, due to the highlyelliptical shape of NGC891the visibilities at short baselines that are aligned at the position angle of the major axis can detect the entire fux density. 
Veryfewprevious observationsofNGC891 with a suffciently small beam size to resolve the disk exist at similar frequencies.The closestinfrequencyarethoseofGioiaetal. (1982) and Dumkeet al. (1995)who usedtheEffelsberg 100-m telescope at10.7GHz and10.55GHz,respectively, andfound inte-grated fux densities consistent withthe AMIvalue (seeTable5). 
Theextentofthe disk abovethe major axisis approximately 2kpc which is similar to the beam size at the distance of the galaxy, so that the disk thickness cannot be properlymeasured. No extended halo emission is observed at the given sensitivity. One feature, observed emanating from the region to the north of the galaxy (marked by “I”), extends to ≈3kpc from the plane and coincides withX-rayemissionofhotgas observed with XMM-Newton at 0.4–0.75keV(Hodges-Kluck&Bregman 2013).At13–18GHz,a substantialfractionofthe emissionwe are observing is thermal, so this feature could be the result of outfow of warm gas (in addition to hot gas) from star-formingregionsinthe northernregionofthegalaxy wherethe star-formationrateis larger(Dahlemetal.1994). 
5. Comparison with other wavelengths 
5.1. Radio continuum 
The most recent radio continuum data for comparison are the 
1.5GHzand6GHzVLA observationsfromtheCHANG-ESsur-vey(Wiegertetal. 2015;Schmidt 2016;Schmidt etal., in prep.). Whenthe146MHzimageis smoothedtoalargerbeam(Fig. 4) in order to detect the most extended emission, the halo extends about asfar out asinthe1.5GHz D-arrayimage (with a simi-lar beam size), but not further out. This is due to the relatively limited sensitivityof ourLOFAR image. Theexponential scale heightswillgiveusabetter indicationofthehaloextent,aswill be shownin Sect. 8. 
5.2. Hα 
The diffuseHα emissionfromthe haloofNGC891hasanexponential scale heightof about1kpc(Dettmar1990).Randet al. (1990)andRossaet al. (2004)observed manyverticalHα fl-aments or“worms” extendingupto2kpcoff the planeofthe galaxy. They interpreted these “worms” as providing evidence foragalactic“chimney” mode(Norman&Ikeuchi1989).Rossa etal. (2004)speculatedthattheverynarrowHα flaments could be magneticallyconfned. 
A98, page8of 21 

D.D. Mulcahyetal.: Cosmicrays and magnetic feldsinthe haloofNGC891 

Fig. <BR>5. <BR>AMImapsofNGC891observedata centralfrequencyof15.5GHzwith the SmallArray(SA)at 14200 × 12100 resolution(left)and the Large Array(LA) at 3600 × 2400 resolution(right).FortheSAimage, contoursareat3,5,8,16,32,64 ×σ where σ = 1.5mJy beam−1 istherms noiselevel.FortheLAimage, contoursareat3(white),5,8,12,18,36,72,108 ×σ where σ = 0.11mJy beam−1istherms noiselevel. The colour scales arein unitsof Jy/beam. The sizesofthe synthesized beams are shownbythe flled ellipsesinthebottomleft corners. 

Fig. <BR>6. <BR>Hα imageofNGC891from Randetal.(1990)(in colours)with the high resolution LOFAR image (Fig. 3 left) overlaid as contours. Contours are at12,18, 25, 32,44,64 ×σ where σ = 0.29mJy beam−1 istherms noiselevel. 
We observearelationbetweenthe high-resolution146MHz radio continuum andHα emission onlyintheHα complex inthe northern disk(Fig. 6).Atthislowfrequency,wedo not expect signifcantthermal emission (Sect. 6.1)whilethe15.5GHz emis-sion(Fig. 5right)hasa muchlargerthermalfraction (Sect. 6.2) and henceis more similartotheHα image.However,the observations at 15.5 GHz are not sensitive enough to detect diffuse thermal emissionfromthe halo. 
Figure 18 shows a composite of radio, Hα, and optical emission. 
5.3. Neutral gas 
Thegalacticfountain modelisinvokedtoexplainthehuge halo of neutralatomic HIgasofNGC891(Oosterlooetal.2007). The extent is up to 22 kpc from the plane in the north-western quadrant. Theexponential scale height increasesfrom1.25kpc inthe centralregionsto about2.5kpcinthe outerpartsbeyond about15kpcradius (“faring”).The bulkofthe coldCO-emitting molecular gas and the cold dust, on the other hand, are much more concentratedtothe plane(Scovilleetal.1993;Altonetal. 1998), but some CO emission could be traced up to 1.4 kpc height abovethe plane(Garcia-Burilloet al.1992)and infrared emissionupto 2.5kpc height(Whaley etal. 2009). 
We do not observe such a large extension in the northwesternquadrant as seenin HIbyOosterlooet al. (2007). The sizeoftheradio halois limitedby the synchrotron lifetimeof the cosmic-rayelectrons of about 2 × 108 yr (Sect. 9.2), longer thanthedutycycleofa typicalgalacticfountainof about108yr (Fraternali2017).However,the north-westernextensionis much largerthanatypicalfountain becauseeitherits timescaleislarger orthe originis different. 
5.4. X-rays 
X-rayobservationsperformedbyBregman&Pildis (1994)were able to detect a considerable amount of diffuse X-ray emissionfromthe haloofNGC891.Theverticalprofleis Gaussian with a vertical scale height of 3.5 kpc, corresponding to a full width at half maximum of 5.8kpc(Bregman& Houck1997). Templeet al. (2005)observedX-ray emissionprotrudingfrom the disk in the north-western direction up to approximately 6kpc which showed a sharpcut-off, suggesting that this is the maximum extent that the outfowing hot gas has reached. The authors also concludedthatNGC891hasa largerstar-formation rate than a normal spiral, but not as extreme as the starburst galaxy NGC253. Cosmic rays and magnetic felds will also be transported by the outfow, but radio emission cannot be detectedat suchlarge heightsdueto energy lossesofthe cosmicray electrons and the limited sensitivity of present-day radio observations. 
6.NGC891’s spectralproperties 
6.1. Thermal and nonthermal emission 
To measure the nonthermal intensity In and the magnetic feld strength, subtracting the free–free (thermal) emission is 
A98, page9of 21 


Fig. <BR>7. <BR>Thermal map(left panel)and nonthermal map(right panel)ofNGC891at146MHz ataresolution of1200,shownbythe circleinthebottom left corner. The colour scale is in mJy beam−1. 

Fig. <BR>8. <BR>Thermalfraction mapofNGC891at146MHz ataresolutionof 1200. The colour scale is in percent. The beam is shown in the bottom left corner. 
essential.Furthermore,in ordertoinvestigatethe energy losses of the CREs, the nonthermal spectral index αn(In ∝ ναn)needs to be known. 
Our AMI LA image cannot be used as a tracer of thermal emission because the angular resolution is too coarse and the emissionisstill dominatedby nonthermal emission (Sect. 6.2). A1.5GHzthermal mapataresolutionof1200 was derived from theCHANG-ES VLA data(Schmidt2016,Schmidt et al.in prep.), based on anHα image correctedfor internalextinction with help of 24 µm dust emission. In the inner disk of edge-on galaxies, the 24µm dust emission may become optically thick, so that the extinction correction may be insuffcient and hence may leadto an underestimateofthethermal emission(Vargas etal.2018). Thiseffectwastakeninto account whenestimating thethermal emissionthe1.5GHz, but leadstoa largerelative uncertainty of about 40%. 
The integrated thermal fux density of 53 ± 23 mJy correspondstoathermalfractionof 7 ± 3%at1.5 GHz. Thisthermal mapwas scaledto146MHz, usingthe spectral indexofoptically thin thermal emission of −0.1. If the assumption of optically thin emission is not valid at 146 MHz in dense regions of the disk,thethermal emission willbeoverestimated. Our highresolutionLOFAR imagewas convolvedtothe sameresolution and re-griddedtothe samegrid size asthethermal map. The scaled thermal mapwas subtractedfromtheLOFAR imagetoproduce the nonthermalimageofNGC891at146MHz. 
Thethermaland nonthermalmapsat146MHzandataresolution of1200 are showninFig. 7. Thethermal image displays a thin disk plus a weak diffuse halo, similar to the maps of thermal emission of cold dust observed at 850µm(Alton et al. 1998; Israel et al. 1999)and of warm dust observed at 24 µm (Whaley et al. 2009). In contrast to the thermal emission, the nonthermal diskisnot“thin”andrevealsasmoothtransitionto the halo. 
Thethermalfractionimageat146MHzwas computedfrom thethermalimage(Fig. 7left)andatotal intensityimageatthe same resolution of1200 andis showninFig. 8. 
Thethermalfractionsinthe diskarebetween5%and10%at 146MHz comparedto10–20%at1.5GHz.Thelargestthermal fractionof approximately16%at146MHzand30%at1.5GHz occursinthe northern diskofthegalaxy.The halorevealsvery small thermal fractions at 146 MHz of 0.1–0.2% compared to 0.7–0.9%at1.5GHz, meaningthattheradio emission observed inthe halois almostentirelynonthermal. 
Free–free absorptionby ionizedgas inthe disk lowersthe radio synchrotron intensity at 146 MHz (Sect. 6.4)and causes a further overestimation of the thermal fraction. Therefore, the thermal fractions at 146MHz estimated for the disk should be regarded as upper limits. Further observations at frequencies lowerthan146MHz are neededto measurethermal absorption and correctthe synchrotron intensity (see Sect. 9.1). 
6.2. Integrated spectrum of NGC 891 
We obtain an integrated fux density ofNGC891withLOFAR at146MHz of 2.85 ± 0.28Jy. The fux density of SN1986J was subtracted. The largest cause of uncertainty (about 10%) is the limited accuracy of the beam model of LOFAR affectingthetransferofgains.Theintegratedfux densitywithAMI at15.5GHzis 120 ± 12mJy, assuminga10% uncertainty. 
A whole range of fux density measurements from the literature was found with many of the fux density mea-surements taken from Hummel et al. (1991). Several fux density measurements from the literature were found to have 
A98, page10of 21 

D.D. Mulcahyetal.: Cosmicrays and magnetic feldsinthe haloofNGC891 
either no uncertainty quoted or seriously underestimated their uncertainties by only including the rms noise (of typically a few %) and not including any calibration uncertainty. In these cases we have inserted a 10% uncertainty to these fux density values. A full list of the fux density measurements is given inTable 5. 
We subtracted the thermal emission at all frequencies, based on the value of 53 ± 23mJy at1.5GHz(Schmidt 2016; Schmidtetal.,inprep.) and usingthe spectral indexofoptically thinthermal emission of −0.1. Thethermal fraction ofthe inte-grated fux density is≤ 2%at146MHz, 7 ± 3%at1.5GHz, and 35 ± 15%at15.5 GHz. The spectrum of integrated nonthermal emissionis plottedinFig. 9. 
Thethermal fraction ofNGC891 at1.5GHz fts wellto theaveragethermalfractionsat1.4GHzobtainedforthe sam-plesof spiralgalaxiesby Niklaset al. (1997; 8 ± 1%), Marvil et al. (2015; 9 ± 3%), and Tabatabaeiet al. (2017; 10 ± 9%)10 . Thisresult indicatesthatthe24 µm intensitiesofedge-ongalaxies indeed needstobe correctedforextinction, asproposedby Vargasetal.(2018). 
The shapeofthe spectrumoftotalradio continuum emission isdeterminedby therelative contributionofthethermalfree– free emission and the shape of the nonthermal (synchrotron) spectrum. As the disk dominates the radio emission from NGC891, spectraleffectsinthe diskare moreimportantforthe integrated emissionthanthoseinthe halo.Forexample,athigh frequencies, typicallyabove5GHz,the increasingthermalfractioninthe disk could leadtoa spectral fattening.Atlowerfrequencies, depending ontheleveloffree–free absorptionbyionizedgas inthe disk andthe nature ofthe nonthermal spectrum, thetotalradio continuum spectrum coulddevelopa spectral fattening, typically at .300 MHz. The shape of the nonthermal spectrum depends onthe dominant energy loss/gain mechanisms which the synchrotron-emitting CREs undergo (see Sect. 9.2). A transition from dominating bremsstrahlung and adiabatic losses to dominating synchrotron and/or inverse-Compton (IC) losses leads to a spectral steepening by −0.5 beyond a certain frequency. Dominating ionization losses of low-energy CREs (.1.5 GeV) could also lead to a fattening of the nonthermal spectrum at low radio frequencies by +0.5 in regions of high gas density (e.g.Basuet al.2015).We notethatthroughoutthe disk and halo spatially varying magnetic felds andgas densities (both neutral and ionized) lead to locally varying breaks in the CRE energy spectrum, suchthatthe correspondingbreaksinthe galaxy’s integrated radio continuum spectrum are smoothed out (Basu etal. 2015). 
Tostudythe nonthermal spectrumofNGC891wefrstmodel it using a simple power law of the form S n(ν) = a0 ναn. Here, S n is the nonthermal fux density, αn is the nonthermal spec-tral index and a0 isthe normalization at1GHz.We fndthe best ft αn = −0.78 ± 0.02, with a reduced χ2 = 1.87. The best-ft power-law spectrum is shown as the black dashed line in Fig. 9 (bottom panel). The expected total intensity spectrum after addingthethermal spectrumtothe nonthermalpower-law spectrum is shown as the black dashed line in the top panel of Fig.9. Thistotal fux density spectrum hasareduced χ2 = 2.62. Clearly, a simple power law does not represent the integrated spectrum well. 
10 The average thermal fraction derived for a sample of star-forming galaxiesby Kleinet al. (2018)is morethan twicelarger, butthese galaxies show indicationsforabreak or anexponential declineintheir nonthermal radio spectra which makes the estimate of the thermal fraction diffcult. 
Table 5. Integrated fux densitiesofNGC891. 
ν (GHz)  Flux density (Jy)  Ref.  
15.5  0.120 ± 0.012  This work  
10.7  0.152 ± 0.026  Gioia etal. (1982)  
10.7  0.155 ± 0.010  Israel & van der Hulst (1983)  
10.55  0.183 ± 0.010  Dumke etal. (1995)  
8.7  0.171 ± 0.023  Beck etal. (1979)  
6.0  0.25 ± 0.03  Schmidt etal. (in prep.)  
4.995  0.29 ± 0.03  Allen etal. (1978)  
4.85  0.25 ± 0.03  Gregory&Condon (1991)  
4.8  0.29 ± 0.03  Stil etal.(2009)  
4.75  0.30 ± 0.03  Gioia etal. (1982)  
2.695  0.43 ± 0.06  Kazès etal.(1970)  
2.695  0.38 ± 0.03  de Jong (1967)  
1.5  0.74 ± 0.04  Schmidt etal. (in prep.)  
1.49  0.74 ± 0.02  Hummel etal. (1991)  
1.49  0.66 ± 0.06  Gioia &Fabbiano (1987)  
1.49  0.70 ± 0.07  Condon (1987)  
1.412  0.77 ± 0.08  Allen etal. (1978)  
0.75  1.4 ± 0.14  Heeschen &Wade (1964)  
0.61  1.53 ± 0.08  Hummel etal. (1991)  
0.61  1.6 ± 0.16  Allen etal. (1978)  
0.408  1.8 ± 0.1  Gioia &Gregorini (1980)  
0.408  1.7 ± 0.2  Baldwin &Pooley (1973)  
0.327  2.1 ± 0.2  Hummel etal. (1991)  
0.330  2.1 ± 0.1  Rengelink etal.(1997)  
0.146  2.85 ± 0.28  This work  
0.0575  6.6 ± 1.8  Israel &Mahoney (1990)  

Notes: Uncertainties markedin boldwere increasedfromtheir original values, asexplainedinthetext. 
In orderto assess any curvature inthe nonthermal spectrum ofNGC891,weempiricallymodelitwith a second-orderpolynomial of the form log S n(ν) = log a0 + α log ν + β (log ν)2. Here, α is the spectral index and β is the curvature parame-ter. The best ft values are found to be a0 = 0.94 ± 0.02, α = −0.76 ± 0.02, and β = −0.14 ± 0.03. The best-ft model is shown as the green solid line in Fig. 9. The reduced χ2 for the ft is 
0.86 
andthatforthetotal emissionis1.32, suggestingthatthe nonthermal spectrum signifcantlydeviatesfromthatofasimple power law. 

To understand the physical origin of the curvature, we also performed modelling ofthe nonthermal spectrum ofNGC891 with a spectral break given by S n(ν) = a0 ναinj /[(ν/νbr)0.5 + 1]. Here, αinj is the injection spectral index of the CREs and νbr is the break frequency beyond which the spectrum steepens by −0.5. Unfortunately, with the current data the parameters for this model cannot be well constrained. We therefore fxed αinj = −0.6 (see Fig. 3 in Caprioli 2011). This yields a break at νbr = 3.7 ± 1.3 GHz which corresponds to a synchrotron age (Eq.(3)) of . 2.5 × 107 yr for the CREs in the disk emitting in a magnetic feld of & 10 µG (Sect. 7). The best ft is shown as the blue dashed–dotted line in Fig. 9. It gives a reduced χ2 = 1.19 for the nonthermal emission and a reduced χ2 = 

1.57 
for the total emission, better than the values for a simple power law. 


As discussed above, the curvature in the nonthermal spectrum couldalsoariseastheresultofsynchrotron–free absorption and/or ionization losses at frequencies below about 150MHz (seeFig.3in Basuetal.2015).Toinvestigatethe frst scenario, a rigorous modelling of the synchrotron radiative transfer in an 
A98, page11of 21 Fig. <BR>9. <BR>Top: integratedtotal fux densities ofNGC891 as listed in Table 5. The two newtotal fux densities fromthis paper are marked in red. Bottom: integrated nonthermal fux densities ofNGC891 after subtracting an estimate of the optically thin thermal free–free emission (shown asthegreen dotted line) fromthetotal fux densities. The blackdashed, solidgreen, and blue dashed–dotted linesinthebottom panel showthe best-ftpowerlaw, second-orderpolynomial model, and spectralbreak modeltothe nonthermal emission,respectively. The corresponding lines in the top panel show the expected total fux density spectrum after adding the thermal free–free emission to the different modelsofthe nonthermal spectrum. 


inclined disk is necessary, while for the later case additional information onthe γ–rayspectrumis needed. Giventhe scarcity of low-frequency observations and the large uncertainty in the estimated thermal emission, the current data are insuffcient to constrainthe originofthe curvatureinthe nonthermal spectrum ofNGC891. 
To better constrain the radio continuum spectrum, additional radio frequency observations at low frequencies would be helpful. The146MHz fux density derivedinthis paper seems to indicate a spectral fattening at low frequencies by free– free absorption. However, this is inconsistent with the value at 57.5MHz. The latter value is based on observations with the ClarkLakeRadio Observatorysynthesisradio telescope with a synthesized beam size of about 70 (Israel&Mahoney1990).As thefrequencyrange below100MHziscrucialtodetect spectral fatteningbyfree–free absorption, new observations withhigher resolution are needed, e.g. withtheLOFARLow Band Antenna (see Sect. 9.1). 
6.3. Spectral index map of NGC 891 
To derive spectral index maps of high accuracy, the frequency span should be as large as possible. However, our new AMI image at 15.5 GHz does not provide suffcient angular resolution to be combined with our new LOFAR image at 146MHz. Hence, spectral index maps were created from the CHANG-ES VLA imageofNGC891at1.5GHz(Schmidt2016; Schmidtetal.,inprep.)11 and ourLOFARimage.Inordertobe sensitivetothe same angular scales,bothimagesweregenerated with the same minimum and maximum uv distance and the same weighting scheme, smoothed to two different resolutions, 1000 and 2000, and placed ontothe samegrid viathe AIPS task OHGEO. Only pixels with fux densities above 8× the rms(σ) levelinthefourinput imageswere used. 
The minimumprojected baselineforthe VLAis about27m, correspondingin unitsofwavelengthtoa baseline about270m for LOFAR at 146 MHz. This uvmin value corresponds to an angular scale of about 250 whichis muchlargerthanthe angular scaleofNGC891. Hence,wedonot expectthe spectral index distributiontobeaffectedbysystematic errors dueto missing angular scales. 
Beforedeterminingthe spectral index,bothimageswereset onthe same fux density scale.The146MHzLOFARimagewas calibrated onthe Scaife& Heald fux density scale(Scaife& Heald 2012), while the 1.5 GHz image was calibrated on the Perley&Butler(2013)fux density scale.We frstconvertedthe 
1.5GHz imagetothe Baars scale(Baarset al.1977)byafactor of1.021takenfromTable13of Perley&Butler(2013).Wethen convertedthis fux densitytothe KPW scale(Kellermannet al. 1969)whichis identicaltothe Scaife&Heald fux density scale atfrequenciesabove325MHz, usingafactorof1.029takenfrom Table7ofBaarsetal. (1977). 
The spectral index between 146 MHz and 1.5 GHz was computed pixel by pixel and is shown in Fig. 10 along with the image of uncertainties (errors) due to rms noise for both resolutions. The spectral index map at 1000 resolution (Fig. 10 top) reveals great detail on the disk’s spectral features. We observe very fat spectra in the central region (α ≈−0.3), coincident with prominent H II regions. In other regions in the disk we observe spectra with α ≈−0.5. This is fatterthan what is observedathigherfrequencies.Schmidtetal.(inprep.)found the spectral indexbetween1.5GHz and6GHztobeα ≈−0.7 in the disk. Immediately away from the disk, we observe spectral indices of ≈−0.6to−0.7. 
The spectral index map at 2000 resolution (Fig. 10 bottom) shows a similar steepening of the spectral index from the disk to the halo. Due to the higher sensitivity with respect to weak extended emission,we cantracethe spectral indexfurtheraway fromthe disk intothe halo. 
6.4. Nonthermal spectral index of NGC 891 
The spectral indexis contaminatedby thermal emissionthatis unrelatedto CREs and magnetic felds. Therefore,we computed amapofthe nonthermal spectral index αn, usingthe nonthermal maps at146MHz at1200 resolution(Fig. 7right) andat1.5GHz atthe sameresolution. 
11 We did notusethe6GHzCHANG-ES image becausethe signal-tonoiseratiosare smallerthanthoseat1.5GHz. 
A98, page12of 21 

D.D. Mulcahyetal.: Cosmicrays and magnetic feldsinthe haloofNGC891 


The nonthermal spectral indicesinthe diskofthegalaxy regions(Tabatabaeietal.2007)andinthe halo. Therefore,we show signifcant signs of fattening. In the very centre of the expect to observe the injection spectral index in star-forming galaxy we measure a spectral index of−0.37.We measure spec-regions and a steeper spectral index in inter-arm regions and tral indices of −0.43 to −0.48 in the northern and southern in the halo. Indeed, a signifcant arm-interarm contrast was star-forming regions of the disk, respectively. Other regions in observed for M51 (Mulcahy et al. 2014), with αn = −0.8 the disk show spectral indices of−0.5to−0.6. between151MHz and1.4GHzinthe inter-armregions. 
InthedisktheobservedCRE populationisasuperpositionof CREsare acceleratedintheshockfrontsof supernovaremvariousspectralageswithyoungCREs locatedinthemajorstar-nants(SNRs). Observationsofthe γ-ray emission from bright formingregions, while older CREswouldexistinthe inter-arm SNRs yield anaverage energy spectral index of CRs inthe 
A98, page13of 21 


Fig. <BR>12. <BR>Equipartitionstrengthofthe magnetic feld B at1200 resolution, in units of µG, determinedby assuming energy equipartition between cosmic rays and magnetic felds. Pixels with αn > −0.6 are blanked. The beamis showninthebottom left corner. 
energyrange100 MeV–50 GeV of γinj −2.2 ± 0.2(Caprioli 2011) where the CRE number density spectrum is a power law as a function of energy, N ∝ Eγinj. Thisisequivalentto anaverage injection spectral indexintheradiorangeof αinj − 0.6 ± 0.1,consistentwiththeaverage spectralindexoftheradio emissionfrom SNRs of −0.5,though with a large dispersion(Green2014). 
Nonthermal spectral indices ofαn ≈−0.7 between1.4GHz and4.86GHzare observedinthediskofNGC891(Heesenetal. 2018a).CREsinthedisk emittingatfrequenciesofafewGHz arestillyoungandhave almost maintainedtheir injection spec-trum.Atlowerfrequenciesweare observing fatter spectrawith αn,146MHz−1.5GHz >αinj whichindicatesthatweareobserving signifcant thermal (free–free) absorption of synchrotron emission instar-formingregionsinthe disk (see Sect. 9.1). 
Inthe halowe fnd αn ≤−0.8,revealing CREageingthrough synchrotron and/or IC emission. These spectra are still signifcantlyfatter comparedto whatis observedat higherfrequencies (Heesenetal.2018a).The nonthermal spectrainthehalosteepen with increasing frequency, as expected from energy losses of CREsbysynchrotron orIC emission (see Sect. 9.2). 
7. Magnetic feldstrengthinthe haloofNGC891 
The strength of the magnetic feld (including turbulent and ordered components) can be determined from the nonthermal emission by assuming equipartition between the energy densities of total cosmic rays and magnetic felds, using the revised formula of Beck& Krause (2005). The equipartition magnetic feld strength B scales withthe synchrotron intensity Isyn as: 
!1/(3−αn) Isyn 
B ∝ (1) 
(K + 1) L 
where αn is the nonthermal spectral index. The adopted ratio of CR proton to electron number densities of K = 100 isa reasonable assumption in the star-forming regions in the disk (Bell 1978). Strong energy losses of CREs due to synchrotron and/or IC emission (Sect. 9.2)are signifed by αn ≤−1.1 in Fig.11and occur atthe outskirts ofNGC891. These leadto an increase of K byafactorofafew and henceto an underestimate of B intheseregionsbyabout10–30%. 

Fig. <BR>13. <BR>Image ofNGC891withindication ofthe boxes usedfor com-putingtheaverage intensities,atdifferentdistancesfromthemajoraxis ofthegalaxy’sprojected disk,andin9differentstripsalongthemajor axis. 
TheeffectivepathlengththroughNGC891, L,is assumedto be0.8timestheradio diameter,resultingin L = 19 kpc. Dependingontheshapeandextentofthe magneticfeldsinthehalo, L decreases with distance from the major axis and also with distancefromthegalaxy’s centrealongthemajor axis.We assumed a constant value of L which results in an underestimate of B by another10–30%.Animproved modelofthedistributionofmagnetic feld strengths will be presented in a forthcoming paper (Schmidt etal. in prep.). 
We created an image of the equipartition feld strength B inNGC891fromthe nonthermaltotal intensity(Fig. 7right) and the nonthermal spectral index (Fig. 11 left), presented in Fig. 12. We blanked pixels with αn > −0.6 as such regions are likely suffering from free–free absorption and would signifcantly underestimate the magnetic feld strength. Blanking regions where αn > −0.6 results in a large fraction of the inner disk without a magnetic feld strength estimate and highlights the diffculty posedbyfree–free absorptionindeterminingthe magnetic feld strength. 
Thefeldstrengthisfoundtobehighestneartotheplanewith a feldstrengthof10–11µG, similartotheaverage feldstrength inthe disksof mildlyinclinedgalaxies(Beck2015)andofedge<BR>ongalaxies(Krauseetal.2018).Theaverageequipartition feld strengthinthehalo,between1kpcand3kpc height,is7µGwith a standarddeviation of2µG. 
8. Radio emission profles and scale heights 
The vertical distributions of radio emission and their scale heightsinthe halosofstar-formingedge-ongalaxieshave been studiedin numerousworks(e.g. Dumkeetal.1995;Heesenetal. 2009; Krause et al. 2018)and is a vital tool for the analysis of cosmic raypropagation and CRE energy losses. Data below 300 MHz have never been used before, because sensitivity and angular resolution at low frequencies were insuffcient so far. TheMWA imagesofNGC253between88 and215MHzhave too low angular resolutions (between50 and 20)forthis purpose. 
The sensitivity of the observations limits the detectability of an extended halo. Therefore, the total observed extent is not suited asaphysical parameter describingthe halo emission, and a more objective parameter such as the scale height is required (Krause etal. 2018). 
ToallowfortestingCREpropagation models(seeSect.9.3), we need to measure the scale heights of the nonthermal emissionat146MHzand1.5GHz.However,free–free absorptionin the diskhampersareliableestimateofthermaland nonthermal emissionat146MHz,sothatwe useinsteadthetotal emissionat this frequency. The difference betweenthese twoquantities can be neglected becausethethermalfractions are small (Sect. 6.1). At1.5GHzwe usethe nonthermalmapfromSchmidtetal.(in prep.). 
A98, page14of 21 

D.D. Mulcahyetal.: Cosmicrays and magnetic feldsinthe haloofNGC891 
Table 6. Scale heights h ofthe ftstothetotal intensityat146MHzat 1200 (0.55kpc)resolutionforthe9 stripsat different distances, x, fromthe galaxy’s centrealongthemajoraxis,andtothenonthermal intensityat1.5GHz,takenfromSchmidtetal.(inprep.). 
x (00)  x (kpc)  hdisk (kpc) 146 MHz  hhalo (kpc) 146 MHz  χ2 red  hdisk (kpc) 1.5GHz  hhalo (kpc) 1.5 GHz  χ2 red  qdisk  qhalo  
+160  +7.37  0.47±0.11  3.6±2.0  2.5  0.25±0.03  1.66±0.07  0.31  1.9±0.5  2.1±1.2  
+120  +5.53  0.23±0.04  2.0±0.2  2.4  0.15±0.01  1.40±0.03  0.37  1.5±0.3  1.5±0.2  
+80  +3.68  0.21±0.07  2.0±0.2  4.4  0.14±0.03  1.38±0.06  4.8  1.5±0.6  1.4±0.2  
+40  +1.84  0.34±0.09  2.1±0.4  6.1  0.29±0.06  1.48±0.19  13  1.2±0.4  1.4±0.3  
0  0.00  0.26±0.06  1.8±0.3  12  0.13±0.02  1.13±0.09  30  2.0±0.6  1.6±0.3  
–40  –1.84  0.40±0.19  1.6±0.6  9.0  0.21±0.06  1.07±0.08  7.0  1.9±1.1  1.5±0.6  
–80  –3.68  0.30±0.06  1.7±0.2  2.8  0.21±0.04  1.14±0.08  4.6  1.4±0.4  1.5±0.2  
–120  –5.53  0.30±0.06  2.5±0.4  1.3  0.22±0.03  1.45±0.06  1.1  1.4±0.3  1.7±0.3  
–160  −7.37  0.39±0.08  3.5±1.2  3.9  0.21±0.02  1.51±0.05  0.26  1.9±0.4  2.3±0.8  

Notes: x is positive onthe north-eastern side and negative onthe south-western side. q istheratioofthe scale heightsat146MHzand1.5GHz. 

' 0.46 kpc),forthe9 stripsofNGC891at different distances, x,fromthegalaxy’s centre alongthe major axis, correspondingtoFig. 13. The frstpanelisatx=+7.37kpc(northern side),thelast oneat x = −7.37kpc (southern side). z isnegative (positive)onthe eastern(western) sideof the halo.Thered linesshowthetwo-componentexponential functions fttedtothedata. 
The disk scale heightsofthe nonthermal emissionat1.5GHz are onaverage12% largerthanthoseofthetotal emission, indicatingthatthe nonthermal diskisbroaderthanthethermal disk. The halo scale heights of the nonthermal and total emission at 
1.5GHz arethe same withinthe errors. 
Followinga similar approachasin Krauseet al. (2018),we ftted two intrinsically exponential distributionstothevertical profles of the disk and the halo. The vertical profle of the observed radio emission in an edge-on galaxy is a superpositionoftheverticalprofleofthe disk+halo emission,theradial profle of the disk emission projected onto the sky plane, and broadeningbythe telescope beam (see Mleret al. (2017)and Krause et al. (2018)for fulldetails).For an almost perfectly edge-ongalaxy such asNGC891,theradialprofleofthe disk emission does notcontributetotheverticalprofle. 
Toobtaintheproflesofradio emissionat146MHzwithdistance from the major axis for both the total intensity map and the nonthermal map (Fig. 7 right) at 1200 resolution, we per-formedaveraging withinstripsof4000 ('1.84 kpc) widthand 7500 ('3.45 kpc) maximum height above and below the major axis, 
A98, page15of 21 


Fig. <BR>15. <BR>Profles ofthe nonthermal spectral index, αn, as a function of height z abovethemajoraxis(inkpc)forthe9 stripsatdifferent dis-tances, x,fromthegalaxy’s centre alongthe major axis. Thetopfour panels(x > 0)correspondtothe northern side ofthegalaxy andthe bottomfour panels(x < 0)tothe southern side;the middle panelis cen-tred onthe minor axis. z is negative (positive) onthe eastern(western) sideofthe halo. 
applyingthe method describedin Mleretal. (2017). Eachdata point is separated by 500, slightly less than half the FWHM of the beam.We also calculated profles with different maximum heights above and below the major axis, without signifcantly differentresults.Theproflesofthetotal intensityarepresented in Fig. 14; those of the nonthermal intensity are very similar and are not shown.A visualisationofthe distributionofstrips isshownFig. 13.The sameprocesswasalsoperformedonthe VLA1.5GHztotal intensitymapfromSchmidtetal.(inprep.). The distributions were fttedby a least-squaresroutine and are shown asred linesinFig. 14. The corresponding scale heights, correctedfor beam smearing, are summarizedinTable 6. 
The exponential scale heights of the total radio emission at 146MHz for both the disk and halo are consistently larger thanthoseat1.5GHz.The meanscaleheightofthediskemission is 0.32 ± 0.08kpcat146MHz, comparedto 0.20 ± 0.05 kpc at1.5GHz.However,the scale heightsofthe diskat146MHz shouldberegarded withcaution. The146MHz emissioninthe inner diskis attenuatedbyfree–free absorption whichartifcially increasesthe scale heights. Hence,weregardthesystematic differenceinthe disk scale heightsbetween146MHzand1.5GHz nottobe signifcant. 
Themeanscaleheightofthehalo emissionis 2.30 ± 0.70 kpc at146MHz comparedto 1.36 ± 0.19kpc at1.5GHz. The mean ratioqhaloof scale heightsofthe halo emissionbetween146MHz and1.5GHz(Table 6)is1.7 ± 0.3. These results will be further discussed in Sect. 9.3. 
The profles of the nonthermal spectral index between 146MHz and 1.5 GHz for different strips along the major axis (seeFig. 13)usingtheaverage intensities are showninFig.15. Theprofles showa consistently fat spectrum(≈−0.4)on the major axisofthegalaxy andaroughly linearsteepening with increasing distance from the major axis on bothsides, reaching a spectral index of about −0.8at3kpc height. 
The halo scale heights and spectral index profles will be discussedin Sect. 9.3,usingsimplifed modelsfortheCREprop<BR>agation. A more thorough treatment, ftting the spectral index profles with numerical models of CRE propagation such as of Heesen etal. (2016, 2018a)andMulcahyetal. (2016)is deferred to future work. 
9. Discussion 
9.1. Thermal absorption in the disk of NGC 891 
Thetotal and nonthermal spectral indexbetween146MHz and 
1.5 GHz in the star-forming disk is observed to be signifcantly fatterthanthe injection spectrum(αinj '−0.6,see Sect.6.4)and far fatterthan whatis observedat higherfrequenciesbySchmidt etal.(inprep.).Synchrotron self-absorptionis unlikelyto occur for such a diffuse source andis seen onlyfor compact sources such as radio supernovae (Chevalier 1998). Bremsstrahlung losses of CREs dominate at low frequencies, but cannot fatten theradio spectrum (e.g.Basuetal.2015).Weproposethatfree– free absorptionof nonthermal emissionbyionizedgas occursin the disk. 
Observations with the VLA at 330 MHz and 74MHz suggested a fattening of the spectrum along the disk of the Milky Way(Kassim,private communication).Spectralturnoversdueto free–free absorptionhavebeen observedinthe integrated spectra of nearby starburstgalaxies suchM82 at ≈500 MHz(Adebahr et al. 2013),NGC253 at 230MHz(Kapi´
nska et al. 2017), and IC10at 322 MHz(Basu etal. 2017). Spatiallyresolved free–free absorptionhasbeen observedinM82withMerlinat408MHz (Willsetal.1997)and withLOFARat154MHz(Vareniusetal. 2015).Lacki&Thompson (2010)showedthatforM82free–free absorption from H II regions is the most important absorption process at frequencies down to 10MHz. Low-frequency observations of several starburst galaxies with the Murchison Widefeld Array(MWA) showed clear signs of free–free absorptioninthe spectrumof integrated fux densities(Galvinet al. 2018). Modelling of thermal absorption in nearby galaxies observed with the LOFAR Multifrequency Snapshot Sky Sur-vey (MSSS) shows that absorption effects in non-starburst galaxies are visible in the integrated spectra only much below 100MHz,but canbestronginthe spectraof individualregionsat higherfrequencies, especiallyforedge-ongalaxies(Chy˙zyetal., in prep.). 
While common instarburstgalaxies,itis under debateifa fatteningofthe spectrumof integrated emission canbe observed alsoin normalstar-forminggalaxies.From ourNGC891 data, we are able to study the spatially resolved distribution of free–free absorption in a highly inclined spiral galaxy for the frsttime.Thedistributionofnonthermal spectralindex(Fig. 11) 
A98, page16of 21 

D.D. Mulcahyetal.: Cosmicrays and magnetic feldsinthe haloofNGC891 
reveals regions in the disk with signifcant spectral fattening, whilethe spectrumof integrated emission(Fig. 9)showsa less obvious fatteningtowardslowerfrequencies,probablyduetothe contributionofsteep-spectrum synchrotron emissionfromthe galaxy’s halo. 
NGC891 hasalower surface densityofthestar-formation rate compared withM82 andthe nuclearstarburstofNGC253, sothatthe space densityof HIIregionsislowerinNGC891; consequently, the emission measure (the square of the elec-tron density integrated along the line of sight) and thus the spectral turnoverfrequency arelower aswell.Furthermore,the emission measure and hence the turnover frequency vary for different lines of sight through the galaxy disk. As a result, any fattening of the integrated spectrum is smeared out over a large frequency range. This may explain why the integrated spectrumofNGC891 around146MHzhardly deviatesfroma power law. 
To measurethe possible fatteningofthe integrated nonther-mal spectrum with higher accuracy, we frst should ascertain the nature ofthetotal emission spectrum ofNGC891. Since we are subtracting a simple optically thin free–free emission ofNGC891throughouttheradiofrequencyrangestudied here, any curvature in the total fux density spectrum itself will naturally give rise to a curvature in the nonthermal spectrum. A cursory look at the total fux densities in Fig. 9 (top panel) suggests that the measurements at the lowest frequencies, viz. 
57.5 and 146 MHz, play an important role in determining the shape ofthetotal intensity spectrum ofNGC891.New observations at similar frequencies, e.g. usingtheLOFAR Low Band Antenna (LBA) at around 55 MHz, could help us to ascertain whether the total intensity spectrum is indeed curved and therebyhelpto understand whetherthe curved nonthermal spectrum ofNGC891 is dueto fattening atthe lower frequencies or steepening at the higher frequencies. We performed fts to the total fux density spectrum by a power law, assuming varying values of the expected fux density at 55 MHz. If the LBA measures a fux density between 8 and 9Jy with less than about 10% uncertainty in fux density, one can confdently rule out a curvature in the total fux density spectrum ofNGC891. 
While spatially resolved observations of free–free absorptioninthe disksofexternalgalaxiesarenowfeasible,the same effect hampers the estimate of the magnetic feld strength. As mentionedin Sect. 7,thepresenceoffree–free absorption leads to an underestimate ofthe nonthermal intensity at low frequencies and henceofthe magnetic feldstrength.Ontheother hand, low-frequency observationsstillplayan essentialroleinestimatingthe magnetic feldstrengthinthe halo becausethey are less affectedbyspectralageingofthe CREs. 
High-resolution observations with the Giant Meterwave RadioTelescope (GMRT) andtheLOFARLBAare currentlyin progressand willhelpto constraintheopticaldepthoftheISM inthe disk ofgalaxies.WithNGC891being nearbyand having an inclination angle of ≈90◦ andthusa largepathlength,itis an idealtargetforfurtherstudiesoffree–free absorptionwithhigh spatial resolution. 
9.2. Loss processes of CREs in the halo of NGC 891 
The maximum contributiontothe synchrotron emission atthe frequency ν comesfrom CREsatthe energy: 
s &#18; &#19;!−1 
ν B⊥ 
Eν = GeV . (2) 
16 MHz µG 
CREs emittingat146MHzhavean energyof .1.0 GeV in the disk ofNGC891(B & 10 µG, Sect. 7)and'1.1 GeV in the halo(B'7 µG). 
The half-power lifetime of the observable synchrotron-emitting CREs is: 
&#18; Eν &#19;−1 B⊥ !−2 
tsyn = 8.35 × 109 yr , (3) 
GeV µG 
where B⊥ isthe magnetic feldstrengthperpendiculartothe line of sight (assuming an isotropic pitchangle distribution).For an 
√ 
isotropic turbulent feld, B⊥ = 2/3 B. InNGC891, tsyn varies between .1.2 × 108 yr in the disk and '1.9 × 108yrinthe halo at146 MHz. 
CREs losing energy via inverse Compton (IC) radiation losses have a half-power lifetime of: 
&#18;&#19;−1 !−1 tIC = 3.55 × 108 yr EUph , (4) 
GeV 10−12 erg cm−3 
where Uph is the total photon energy density. Heesen et al. (2014) found that the ratio between tsyn and tIC (identical to the ratio between Uph and the magnetic energy density) varies typicallybetween10% and 80%in normalstar-forminggalaxies, although it may be higher in galaxies undergoing a star-burst.InNGC891,this ratiois about 16%, estimated from the total infrared luminosity and assuming energy equipartition (Heesen et al. 2018a), so that synchrotron losses dominate over IC losses. 
The CRE half-power lifetime against bremsstrahlung losses is: 
&#18;&#19;−1 
n 
tbrems = 3.96 × 107 yr , (5) 
cm−3 
where n isthe densityofthe neutralgas.Accordingto Basuetal. (2015), bremsstrahlung losses of CREs emitting in the range 100–300 MHz dominate in regions with gas column densities beyond about5M&#12;/pc2, asisthe caseinthe diskofNGC891. 
The CRE half-power lifetime against ionization losses is (Longair 2011): 
&#18;&#19;−1 &#18;&#19;&#20; &#18;&#19; &#21;−1 
nE E 
tion = 4.1 × 109yr 3ln + 42.5 . 
cm−3 GeV GeV 
(6) 
Atfrequencies below 300 MHz, ionization losses are slightly lessimportantthanbremsstrahlung losses(Basuet al.2015), exceptperhapsin highlydensegas clouds. 
9.3. CRE propagation in NGC 891 
The scale heightsofthe nonthermal halo emissionat146MHz are consistentlylargerthanthoseat1.5GHz(Table 6).Theprob<BR>ablereasonisthatCRE energylossprocessesareweakeratlower frequencies, sothatthe CRE lifetime and hencethepropagation height of CREs are larger. 
We investigated two basic models of vertical CRE propagation, namelydiffusion or advection (outfow)(Tabatabaeiet al. 2013; Mulcahy et al. 2014, 2016;Krause et al. 2018). Mulcahy et al. (2016)showed via numerical modellingthat diffusion is the dominant CREpropagationprocessinthe diskofM51.As the observed synchrotron spectral indicesinthe disk aretoo fat tobeexplainedbydominating synchrotron losses,the CREsof M51must escape from the disk before losing their energy, e.g. bydiffusion perpendiculartothe disk. 
A98, page17of 21 


Fig. <BR>16. <BR>Scale heightsinkpc(darkblue)at146MHzand magnetic feldstrengthsin µG(red) at different distancesx fromthegalaxy’s centre along the major axis (in kpc). x > 0 correspondstothe north-eastern side ofthegalaxy and x < 0tothe south-western side. The disk scale heights are shown inthe left panel,the halo scale heightsinthe right panel. The magnetic feldstrengths are measuredfromthe1.5GHz data onthegalaxy’s major axis and are identicalin allfour panels. 
We assumethatthelifetimeof CREsinthe haloofNGC891 is dominatedby synchrotron loss (Eq.(3)), i.e.the haloisa calorimeterfor CREs.Inthe casethatradiation lossesareimportant,agalaxyisreferredto as an “electron calorimeter”, where the CREs arelosingtheir energythroughradiation and ionization losses before they escape. Further assuming that the diffusion coeffcient D is constant (no dependencies on height z, nor on CRE energy E,nor on feldstrength B),the diffusivepropagation height is: 
p
hdiff = DtCRE ∝ B −3/4 ν −1/4 (7) 
where tCREisthelifetimeoftheCREs(see Sect. 9.2)andB isthe feld strength. With the frequency ratio between 146 MHz and 
1.5 GHz of 1/10.3, the expected ratio of propagation heights is 
1.8.In caseof dominatingIC lossesofthe CREs,thefrequency dependenceofthe diffusivepropagation height hdiff remainsthe same, while there is only a weak positive dependence on feld strength(hdiff ∝ B1/4). 
Fora constant advection speedv and a lifetime of CRE dom-inatedbysynchrotron loss,the advectivepropagation height is: 
hadv = v tCRE ∝ B −3/2 ν −1/2 . (8) 
With the same frequency ratio as above, the expected ratio of propagation heights is 3.2. In the case of dominating IC losses of the CREs, the frequency dependence of the advective propagation height hadv remains the same, while the dependence on feldstrengthbecomes positive(hadv ∝ B1/2). 
SeveralspiralgalaxiesintheCHANG-ES sampleshow indications for advective CRE propagation and CRE timescales dominatedbyescapefromthe halo(Krauseet al.2018).Inthis case, the propagation height does neither depend on frequency nor on magnetic feld strength. 
The mean observed ratio qhalo of scale heights of the halo emission between 146 MHz and 1.5 GHz (Table 6) of 1.7 ± 
0.3 agrees well with the diffusion model and dominating synchrotron and/or IC losses. It seems that the CRE propagation from the disk to the halo of NGC891 is slower than in other spiral edge-on galaxies in the CHANG-ES sample, so that the CRE synchrotron lifetime is smaller than the escape time. 
The halo scale heightsfromTable 6maysufferfromvarious systematic biaseffects.Firstly,the halo scale heightsat146MHz canbeaffectedbyfree–free absorptioninthe disk.Asa test,we artifcially increased the nonthermal intensity in the inner disk byafactorof3. This leadsto about 2× smaller scale heights of the inner disk, whilethe halo scale heightsdo notchange within the limits of uncertainty. 
NGC891was observedinL-band withthe VLA D-arrayfor 3 × 10 min.Accordingtothe VLAwebsite,the maximum visible structureofasingle snapshotobservationat1.5GHzis about 80 whichis similartothe angularextentofNGC891.The maximum visible structure of a multiple snapshotobservation is somewhat larger but hard to estimate. We cannot exclude that some diffuse halo emissionis missingintheL-band imagebySchmidt et al. (in prep.). The effect on the total fux density must be small because Fig. 9 does not indicate an obvious defcit, but the emission in the outer halo may still be underestimated, so that the measured halo scale heights could be too small. As a test,we added an artifcial halo component whichis constructed bysmoothingthegalaxy’s image with a Gaussian functionof 50 full-width-at-half-maximum and decreasingtheresulting intensitiesby afactorof10. Hence,this component contributes10% tothetotal fux density whichisstill consistent withFig. 9.It leadsto an increaseofthe halo scale heightsbyonly5–10% and to an insignifcant decreaseofthe meanratioof scale heightsof the halo emissionbetween146MHz and1.5GHzto1.6. 
Both CRE propagation models discussed above are rather simplistic.A moredetailed analysis needsaproper description of CRE sources, magnetic feld distribution, CRE loss processes, and CRE propagation mechanisms. Such a study, based on the CHANG-ES data,isforthcoming(Schmidtetal.inprep.). 
Nevertheless,we canexplorethe above models further,taking advantageofthefactthat synchrotron emissionisthe only CRE loss process related to the magnetic feld strength. Hence, the local scale height, as measured in the strips, should depend 
A98, page18of 21 

D.D. Mulcahyetal.: Cosmicrays and magnetic feldsinthe haloofNGC891 
onthe local magnetic feldstrengthinthe same area.Weexpect an increase of scale height withdecreasing feld strengthwhich isdifferentfordiffusiveandadvectiveCREpropagation (Eqs.(7) and(8)). 
The scale heightsat146MHz(Table 6)areplottedinFig.16. A variation of the disk scale height (left panel) with distance fromthegalaxy centre is not obvious, whilethe scale height of the halo (right panel) increases with increasing distance (“faring”). This effect is similar to the one seen in the halo of the edge-on spiralgalaxyNGC253(Heesenetal. 2009). 
Toinvestigate whetherthe halo faringinNGC891is caused bya decrease of the synchrotron losses of CREs, we also show inFig. 16the magnetic feldstrengths onthe major axis which area measureofthelevelof synchrotron lossesinthe different strips (Eq.(3)).Aswewere unableto obtainreliable magnetic feldstrength estimatesinthe diskfrom our146MHz data due to free–free absorption effects (see above), we instead use the magnetic feld strength estimates from Schmidt et al. (in prep.) derivedfrom1.5GHzdata.Indeed,the scale heightsatbothfrequencies are smallestneartothe centre wherethe feldstrengths are largest. 
Therelationbetweenthe scale heightsat146MHzandthe magnetic feld strength is shown in Fig. 17. The exponent of −1.2 ± 0.6 is consistent withdiffusive or advective propagation (Eqs.(7)and(8))withinthe uncertainty.Thenegativesignofthe exponentsatbothfrequencies supportsouraboveresultthatsynchrotronradiationisthe dominating lossprocessof CREsinthe halo ofNGC891 andthat IC losses are less important, which is generally the case for late-type spiral and dwarf irregular galaxies(Heesenetal.2014). 
ThediskofNGC891seenwithLOFARisnotuniformeverywhere, but bulges outinthe northwherethehalo scale heightis large (Fig. 16, at x = 7.36 kpc), even larger than expected for theweak magnetic feldinthis area, as indicatedby the points above the ft line in the top left corners of Fig. 17. Possible reasons could be a larger diffusion coeffcient in the northern star-forming region or an increase of scale height by an enhanced,faster outfow. 
Accordingto Strong (1978),theverticalprofleofthe non-thermal spectral index is similar for the two types of CRE propagation,diffusionandadvection.Ontheother hand, Heesen et al. (2016, 2018a) showed that the vertical profle of the radio spectral index can help to distinguish between the two basic mechanisms of CRE transport. Advection of CREs with a constant outfow velocity should lead to linear profles of the nonthermal spectral index, where the spectral index steepens gradually with increasing distance from the disk plane. Con-versely, diffusion should leadto parabolic-shaped spectral index profles where the spectral index remains about constant until a few kpc height, from where on the synchrotron losses become strong and the nonthermal spectral index steepens rapidly. In NGC891,theverticalproflesoftheradio spectral indexbetween 146MHz and1.5GHz, aspresentedinFig. 15, arefalling into the frst category, indicating that advection in an outfow may be the dominating CRE transport process. In non-calorimetric halos where CRE escape is important, the nonthermal scale height and CRE lifetime are not related, and Eqs.(7)and(8)do notapply. 
In order to perform a more quantitative investigation, we need an estimate of the escape timescale for the cosmic rays inthe haloofNGC891 and hence a model ofthe outfow velocity as a function of distance from the disk. Based on previousradio data, Heesenet al. (2018a)could not fndawell<BR>constrainedadvection speedfora possible outfowinNGC891. 

Weplanto combineournewLOFAR146MHzdataofNGC891 and the CHANG-ES data at 1.5 and 6GHz in a forthcoming work. An analysis of the profles of synchrotron emission and spectral index will be performed, where the 1D diffusion-loss equation will be solved, such as has been done for the radial diffusion within the disk of the almost face-on galaxy M51(Mulcahy et al. 2016) andforthevertical transportby advectionanddiffusioninthe halosof manyedge-ongalaxies (Heesen et al. 2016, 2018a). This will allow us to measure dif-fusion coeffcients or advection speeds of CREs in the halo of NGC891. 
10. Conclusions 
Inthiswork,we performedradio continuum observationsofthe edge-on spiralgalaxyNGC891 withtheLOFAR High Band Antenna (HBA) Array with a central frequency of 146MHz andwiththeArcminute MicrokelvinImager(AMI)at15.5GHz. Usingthefacetcalibrationschemedetailedin vanWeerenetal. (2016),we achievedLOFAR images with a highquality nearto thermal noise which provide a sensitive view of the extended haloofNGC891 anditsphysical origin. Thisisthe frst time that low-frequency, high-resolution observations are presented ofthiswell-studied edge-ongalaxy. 
We used the AMI Small and Large Array to observe NGC891ata centralfrequencyof15.5GHz.Forthe frst time, a nearby galaxy has been studied using the upgraded correlatorof AMI.Nearbygalaxieshaverarelybeen observedatradio frequencieslargerthan10GHz. Thisworkpavesthewayforan AMI nearbygalaxy surveyat15.5 GHz. 
The main fndingsofthiswork are: 
– With our new measurements and assuming realistic uncertaintiesofprevious measurements,we derivedthe spectrum of the integrated nonthermal fux density. The spectrum is probably not asimplepowerlaw.It canbe fttedby apower law with a steepening in spectral index by −0.5 towards higherfrequencies, orby a curvedpolynomial, indicatinga fatteningtowardslowerfrequenciesbyfree–free absorption andasteepeningtowards higherfrequencies. Whichphysical mechanism produces the curvature in the spectrum cannotbe constrained with the current measurements.A clear detection of a spectral curvature in the integrated spectrum needs observations with LOFAR LBA at around 55MHz with < 10% uncertainty. 
A98, page19of 21 


Fig. <BR>18. <BR>Composite ofNGC891:LOFAR146MHz at 1200 × 1200 resolution (blue),opticalfrom DSS2(green),andHα from Randet al. (1990;red). The sizeofthe imageis 90× 120 . 
– 
We fnd no signifcant fattening of the spectrum by ther-mal emissiontowards higherfrequenciesofupto15.5GHz, suggesting that the synchrotron component is still dominantinthisfrequencyrange.Further observationsat higher frequencies are necessaryto confrmthisresult. 

– 
In the star-forming disk, our map of nonthermal spectral index between 146 MHz and the CHANG-ES data at 

1.5 GHz reveals areas with signifcant spectral fattening towards lower frequencies, with values signifcantly fatter than −0.5. This is likely causedby free–free absorption in the ionizedgas. 

– 
Inthe halo,we observe nonthermal spectral indicesbetween 146MHz and1.5GHz withinarangeof−0.6to−0.8,significantlyfatterthanthe spectral indicesat higherfrequencies. This supports the expectation that cosmic-ray electrons (CREs) emitting at low frequencies suffer less from energy losses. Consequently, low-frequency observations are better suited to estimate the magnetic feld in the halo if one assumes energy equipartition between cosmic rays and magnetic felds. 

– 
The mean magnetic feld strength in the halo is 7 ± 2 µG. Due to the signifcant free–free absorption, we cannot confdently estimate magnetic feld strengths in the disk ofNGC891. High-frequency observations are more suited for this task, while low-frequency observations are more suitableto estimate magnetic feldstrengthsinthe halo. 

– 
Radio emission from the halo at 146 MHz is detected out toa maximumof7.3kpcdistancefromthemajor axis.The similarextension comparedtothe newCHANG-ES image at 1.5 GHz is most likely due to the lower sensitivity at 146 MHz. 


– 
The scale heights of the nonthermal halo emission at 146 MHz are consistently larger at all radii than those at 

1.5 GHz, with a mean ratio of 1.7 ± 0.3, as predicted by diffusive CRE propagation. This ratio also suggests that spectral ageing of the CREs, caused by radiation losses, is important and hence that the halo is at least a partial calorimeter for CREs, i.e. they are losing a signifcant fractionoftheir energy beforethey can escape. 

– 
The scale height of the nonthermal halo emission at 146 MHz correlates with the magnetic feld strength with an exponent of −1.2 ± 0.6, consistent with diffusive or advective CRE propagation. The negative exponent is a signature of dominating synchrotron losses in the halo. IC losses appeartobe lessimportant. 

– 
Thelow-frequency radio haloinNGC891 seems tobe differentfromthe ones observedintheCHANG-ES sample at GHz frequencies that are escape-dominated (i.e. non-calorimetric). More low-frequency observations of edge-on spiralgalaxies are needed in orderto decide whetherthis is a specifcpropertyofthelow-frequency haloinNGC891 or holdstruefor spiralgalaxiesingeneral. 

– 
The linearly steepening spectral index profles of the nonthermal emission in the halo seem to favour advection of CREsinagalactic wind(Heesenet al.2016).Further studies are forthcoming, using refned modelling of CRE propagation and exploiting the combined LOFAR and CHANG-ES data at1.5 and6GHz. 

– 
To measure energy losses and propagation of CREs and magnetic feld strengths in the disk and halo with higher accuracy,further observationsofNGC891andotheredgeon spiralgalaxiesovera widefrequency coverage and with high spatial resolution are needed. 


Acknowledgements. This research was performed in the frameworkof the DFG ResearchUnit1254“Magnetisation of Interstellar and Intergalactic Media: The Prospects of Low-FrequencyRadio Observations”.LOFAR, designed and con-structedbyASTRON,hasfacilitiesinseveral countriesthatareownedbyvarious parties (eachwiththeirown funding sources) andthat are collectivelyoperated bythe InternationalLOFARTelescope(ILT)foundation underajointscientifc policy.We acknowledge supportbytheFZ Jichunder Jureca computinggrant HTB00. We thank the staff of the Mullard Radio Astronomy Observatory for their invaluable assistance in the commissioning and operation of AMI which is supportedby Cambridge and OxfordUniversities.We acknowledge support fromtheEuropeanResearchCouncil undergrantERC-2012-StG-307215LODESTONE. This research made use of the NASA/IPAC Extragalactic Database (NED) whichis operatedby theJetPropulsion Laboratory, California Institute ofTechnology, under contract withtheNationalAeronautics and SpaceAdministration.A.B. acknowledges fnancial supportbythe GermanFederal Ministry of Education andResearch(BMBF) undergrant 05A17PB1(VerbundprojektD-MeerKAT).F.S.T. acknowledges fnancial support fromthe Spanish Ministryof Economyand Competitiveness (MINECO) undergrantAYA2016-76219-P.We thankFrank Israel(Sterrewacht Leiden) and OlafWucknitz (MPIfR Bonn)for carefully readingofthe manuscriptand our anonymousrefereefor manyuseful comments. 
References 
Adebahr,B.,Krause,M., Klein,U.,etal.2013,A&A,555,A23 Allen,R.J., Sancisi,R.,&Baldwin,J.E.1978, A&A,62,397 Alton,P.B., Bianchi,S.,Rand,R.J.,etal.1998, ApJ,507,L125 Arshakian, T. G., Stepanov, R., Beck, R., Krause, M., & Sokoloff, D. 2011, 
Astron.Nachr., 332, 524 Baars,J.W.M., Genzel,R.,Pauliny-Toth,I.I.K.,&Witzel,A.1977, A&A,61, 
99 Baldwin,J.E.,&Pooley,G.G.1973, MNRAS,161,127 Basu,A.,Beck,R.,Schmidt,P.,&Roy,S.2015, MNRAS,449,3879 Basu,A.,Roychowdhury,S., Heesen,V.,etal.2017, MNRAS,471,337 Beck,R.2015, A&ARv,24,4 Beck,R.,Biermann,P., Emerson,D.T.,&Wielebinski,R.1979, A&A,77,25 
A98, page20of 21 

D.D. Mulcahyetal.: Cosmicrays and magnetic feldsinthe haloofNGC891 
Beck, R., Brandenburg, A., Moss, D., Shukurov, A., & Sokoloff, D. 1996, ARA&A,34,155 
Beck,R.,&Krause,M. 2005, Astron.Nachr.,326,414 
Bell,A.R.1978, MNRAS,182,443 
Bietenholz,M.F.,&Bartel,N.2017, ApJ,851,7 
Bietenholz,M.F., Bartel,N.,&Rupen,M.P. 2002, ApJ,581,1132 
Bietenholz,M.F., Bartel,N.,&Rupen,M.P.2010, ApJ,712,1057 
Bregman,J.N.,&Houck,J.C.1997, ApJ, 485,159 
Bregman,J.N.,&Pildis,R.A.1994, ApJ, 420,570 
Buffe,K., Heesen,V.,&Shalchi,A.2013, ApJ,764,37 
Caprioli, D. 2011, J. Cosmol. Astropart. Phys., 5, 026 
Chevalier,R.A.1998, ApJ, 499,810 
Condon,J.J.1987, ApJS,65, 485 
Cornwell,T.J. 2008, IEEEJ. Select.Top. SignalProcess.,2, 793 
Dahlem,M.,Dettmar, R.-J.,&Hummel,E.1994, A&A,290,384 
de JongM.L.1967, ApJ,150,1 
de Vaucouleurs, G., de Vaucouleurs, A., Corwin Jr., H. G., et al. 1991, Third ReferenceCatalogueofBright Galaxies(NewYork:Springer) 
Dettmar, R.-J.1990, A&A, 232,L15 
Dumke,M.,Krause,M.,Wielebinski,R.,&Klein,U.1995, A&A,302,691 
Ekers,R.D.,&Sancisi,R.1977, A&A,54,973 
Elstner,D., Golla,G.,Riger,G.,&Wielebinski,R.1995, A&A,297,77 
Fraternali,F. 2017, in Astrophysics and Space Science Library, GasAccretion onto Galaxies, eds.A.Fox,&R.Davé, 430, 323 
Galvin,T.J.,Seymour,N.,Marvil,J.,etal.2018, MNRAS,474,779 
Garcia-Burillo, S., Guelin, M., Cernicharo, J.,&Dahlem, M.1992, A&A, 266, 21 
Gioia,I.M.,&Fabbiano,G.1987, ApJS,63,771 
Gioia,I.M.,&Gregorini,L.1980, A&AS,41,329 
Gioia,I.M.,Gregorini,L.,&Klein,U.1982, A&A,116,164 
Green,D.A.2014, Bull.Astronom. Soc. India,42,47 
Gregory,P.C.,&Condon,J.J.1991, ApJS,75,1011 
Heeschen,D. S.,&Wade,C.M.1964, AJ, 69,277 
Heesen,V.,Beck,R.,Krause,M.,&Dettmar, R.-J.2009, A&A,494,563 
Heesen,V.,Brinks,E., Leroy,A.K.,etal.2014, AJ,147,103 
Heesen,V., Dettmar, R.-J., Krause, M., Beck, R.,&Stein,Y. 2016, MNRAS, 458, 332 
Heesen,V.,Krause,M.,Beck,R.,etal.2018a, MNRAS,476,158 
Heesen,V.,Rafferty,D.A., Horneffer,A.,etal.2018b, MNRAS,476,1756 
Henriksen,R.N.,&Irwin,J.A.2016, MNRAS, 458,4210 
Hickish,J.,Razavi-Ghods,N.,Perrott,Y.C.,etal.2018, MNRAS,475,5677 
Hodges-Kluck,E.J.,&Bregman,J.N.2013, ApJ,762,12 
Hbom,J.A.1974, A&AS,15,417 
Howk,J.C.,&Savage,B.D.1997, AJ,114, 2463 
Hummel,E.1991, A&A,251,442 
Hummel,E., Dahlem,M.,vanderHulst,J.M.,&Sukumar,S.1991, A&A,246, 10 
Irwin,J.,Beck,R., Benjamin,R.A.,etal.2012, AJ,144,43 
Israel,F.P.,& vander Hulst,J.M.1983, AJ,88,1736 
Israel,F.P.,&Mahoney,M.J.1990, ApJ,352,30 
Israel,F.P.,van derWerf,P.P.,&Tilanus,R.P.J.1999, A&A,344, L83 
Kapi´nska,A.D.,Staveley-Smith,L.,Crocker,R.,etal.2017, ApJ, 838,68 
Kazès,I.,LeSqueren,A.M.,&Nguyen-Quang-Rieu.1970,Astrophys.Lett.,6, 193 
Kellermann,K.I.,Pauliny-Toth,I.I.K.,&Williams,P.J.S.1969,ApJ,157,1 
Klein,U.,Wielebinski,R.,&Beck,R.1984, A&A,133,19 
Klein,U., Lisenfeld,U.,&Verley,S.2018, A&A,611, A55 
Krause,M. 2009, Rev.Mex.Astron.Astrofs.,36,25 
Krause, M. 2012, in MagneticFields intheUniverse III – From Laboratoryand StarstoPrimordialStructures, eds.M. Soida,K. Otmianowska-Mazur,E.M. de Gouveia Dal Pino,&A. Lazarian, (Krakow:JagiellonianUniversity) 155 
Krause,M.,Irwin,J.,Wiegert,T.,etal.2018, A&A,611,A72 
Kulsrud, R.,&Pearce,W.P.1969,ApJ,156,445 
Lacki,B.C.,&Thompson,T.A.2010, ApJ,717,196 
Lepage, R.,&Billard, L.1992, Exploringthe Limits of Bootstrap,WileySeries inProbability and MathematicalStatistics (NewYork:Wiley) 
Lerche,I.,&Schlickeiser,R.1981, ApJS,47,33 
Lerche,I.,&Schlickeiser,R.1982, A&A,107,148 Licquia,T.,&Newman,J.2014,in Amer.Astron.Soc.Meet.Abstr.,223, 336.04 
Longair, M. 2011, High Energy Astrophysics, 3rd ed. (Cambridge, UK: CambridgeUniversity Press) 
Marvil,J.,Owen,F.,&Eilek,J.2015, AJ,149,32 
McMullin,J.P.,Waters,B.,Schiebel,D.,Young,W.,&Golap,K.2007,in Astro-nomicalData AnalysisSoftwareandSystemsXVI, eds.R.A.Shaw,F. Hill, &D.J. Bell, ASP Conf. Ser.,376,127 
Moss,D.,&Sokoloff,D.2017, A&A, 598,A72 
Moss,D., Sokoloff,D., Beck,R.,&Krause,M.2010, A&A,512,A61 
Mulcahy, D. D., Fletcher, A., Beck, R., Mitra, D., & Scaife, A. M. M. 2016, A&A, 592, A123 
Mulcahy,D.D., Horneffer,A., Beck,R.,etal.2014, A&A, 568,A74 
Mler,P.,Krause,M., Beck,R.,&Schmidt,P.2017, A&A, 606,A41 
Niklas,S., Klein,U.,&Wielebinski,R.1997, A&A,322,19 
Nixon,C.J., Hands,T.O., King,A.R.,&Pringle,J.E.2018, MNRAS,477,3539 
Norman,C.A.,&Ikeuchi,S.1989,ApJ,345,372 
Offringa,A.R.,deBruyn,A.G., Biehl,M.,etal.2010, MNRAS,405,155 
Offringa,A.R.,vandeGronde,J.J.,&Roerdink,J.B.T.M.2012, A&A, 539, A95 
Okabe,A., Boots, B., Sugihara, K.,&Chiu, S. 2000, Spatial tessellations. Con-ceptsand ApplicationsofVoronoiDiagrams:WileySeriesinProbabilityand MathematicalStatistics, 2nd ed. 
Oosterloo,T.,Fraternali,F.,&Sancisi,R.2007, AJ,134,1019 
Pakmor,R., Marinacci,F.,&Springel,V.2014,ApJ,783, L20 
Perley,R.A.,&Butler,B.J.2013,ApJS,204,19 
Pohl,M.,&Schlickeiser,R.1990,A&A,234,147 
Rand,R.J.,Kulkarni,S.R.,&Hester,J.J.1990,ApJ, 352,L1 
Rau,U.,&Cornwell,T.J.2011,A&A, 532,A71 
Rengelink,R.B.,Tang,Y.,deBruyn,A.G.,etal.1997,A&AS,124,259 
Rossa, J., Dettmar, R.-J., Walterbos, R. A. M., & Norman, C. A. 2004, AJ,128,674 
Rupen,M.P.1991,AJ,102,48 
Scaife,A.M.M.,&Heald,G.H.2012, MNRAS,423,L30 
Schmidt,P.2016, PhD Thesis,Universityof Bonn, Germany 
Scoville,N. Z., Thakkar, D., Carlstrom, J. E.,&Sargent,A. I.1993, ApJ, 404, L59 
Shapiro,P.R.,&Field,G.B.1976, ApJ, 205,762 
Sofue,Y.1987, PASJ, 39,547 
Soida,M.,Krause,M.,Dettmar, R.-J.,&Urbanik,M.2011, A&A,531,A127 
Stil,J.M.,Krause,M.,Beck,R.,&Taylor,A.R. 2009,ApJ,693,1392 
Strickland,D.K.,Heckman,T.M., Colbert,E.J.M., Hoopes,C.G.,&Weaver, K.A. 2004, ApJS,151,193 
Strong,A.W.1978,A&A, 66, 205 
Tabatabaei,F.S., Beck,R., Krel,E.,etal. 2007,A&A,475,133 
Tabatabaei,F.S., Berkhuijsen,E.M.,Frick,P., Beck,R.,&Schinnerer,E.2013, A&A, 557, A129 
Tabatabaei,F.S.,Schinnerer,E.,Krause,M.,etal.2017,ApJ, 836,185 
Temple,R.F.,Raychaudhury,S.,&Stevens,I.R.2005,MNRAS,362,581 
Uhlig, M., Pfrommer, C., Sharma, M., etal. 2012, MNRAS, 423, 2374 
vanderKruit,P.C.,&Searle,L.1981, A&A,95,116 
vanGorkom,J.,Rupen,M., Knapp,G.,etal.1986,IAUCirc., 4248,1 
van Haarlem,M.P.,Wise,M.W., Gunst,A.W.,etal.2013, A&A, 556,A2 
vanWeeren,R.J.,Williams,W.L.,Hardcastle,M.J.,etal.2016, ApJS,223, 2 
Varenius,E., Conway,J.E., Martí-Vidal,I.,etal.2015,A&A,574,A114 
Vargas,C.J., Mora-Partiarroyo,S.C.,Schmidt,P.,etal.2018,ApJ, 853,128 
Vigotti, M., Grueff, G.,Perley, R., Clark, B. G.,&Bridle,A. H.1989,AJ, 98, 419 
Whaley, C. H., Irwin, J.A., Madden, S. C., Galliano,F.,&Bendo, G. J. 2009, MNRAS, 395,97 
Wiegert,T.,Irwin,J., Miskolczi,A.,etal.2015,AJ,150,81 
Williams,W.L.,vanWeeren,R.J.,Rtgering,H.J.A.,etal.2016,MNRAS, 460, 2385 
Wills, K.A.,Pedlar,A., Muxlow,T.W. B.,&Wilkinson,P.N.1997,MNRAS, 291,517 
Xilouris,E.M., Byun,Y.I.,Kylafs,N.D.,Paleologou,E.V.,&Papamastorakis, J.1999, A&A,344, 868 
Zwart,J.T.L.,Barker,R.W., Biddulph,P.,etal.2008, MNRAS,391,1545 
A98, page21of 21