Filename
stringlengths 22
64
| Paragraph
stringlengths 8
5.57k
|
---|---|
Processed_A_radio_survey_of_supersoft,_persistent_and_transi.txt | SGR 0525-66 is one of a group of only three confirmed (Smith 1997) and one possible additional (Hurley et al. 1997) soft γ-ray repeaters. These are sources of repeated bursts of low-energy γ-rays, possibly associated with neutron stars. SGR 0525-66 is coincident in the sky with the SNR N49, and a soft X-ray counterpart has been found (Rothschild, Kulka- rni & Lingenfelter 1994). No radio, optical or infrared coun- terpart has been identified however (Dickel et al. 1995), al- though Rothschild et al. (1994) predicted a compact plerion- like radio nebula to be associated with the source. Any phys- ical link with N49 has yet to be established. |
Processed_A_radio_survey_of_supersoft,_persistent_and_transi.txt | Table 1. Radio survey of Magellanic Cloud X-ray sources with the Australia Telescope compact array. All upper limits are 3σ. |
Processed_A_radio_survey_of_supersoft,_persistent_and_transi.txt | The upper limit to radio emission from RXJ 0513-6951 indi- cates it is at least a factor of 50 less luminous at radio wave- lengths than the classical Galactic jet source SS 433, which also shows Doppler shifted optical emission lines. However, the jets of SS 433 have a velocity of 0.26 c, while those of RXJ 0513-6951 are less than 0.01 c, and so clearly far less energetic. It may be that the supersoft source simply does not accelerate electrons to the relativistic energies required to produce synchrotron emission. |
Processed_A_radio_survey_of_supersoft,_persistent_and_transi.txt | No radio nebulosity, on scales from a few arcsec to a few arcmin, was detected associated with any of these sys- tems, expected to be strong sources of ionising UV photons (although note that with a minimum baseline of 153m this configuration of the array is not sensitive to structureswith angular scales ≥ 1 arcmin). In particular, the arcmin-scale optical nebula around CAL83 (Remillard et al. 1995) was not observed. |
Processed_A_radio_survey_of_supersoft,_persistent_and_transi.txt | Figure 1. 6.3 cm radio map centred on optical location of the black hole candidate LMC X-1. The nearby sources 0540-697 and 0539-697 (Wright et al. 1994), plus the previously unidentified source, labelled 0539-697(S), are probably the cause of a previous report of radio emission from this region. The apparent alignment of 0539-697(S), 0539-697, and the optical position of LMC X- 1, while reminiscent of arcmin-scale radio lobes associated with 1E1740.7-2942, is probably coincidental. Contours are at -3, 3, 6, 12, 24, 48 and 96 times the r.m.s. noise of 0.5 mJy. |
Processed_A_radio_survey_of_supersoft,_persistent_and_transi.txt | Neither of these two black hole candidates were detected to levels consistent with analogy to their Galactic counter- parts. At a distance of 55 kpc, the persistent Galactic radio- emitting X-ray binaries would only be observed at a level of ∼ 0.05 mJy, at about the noise level for LMC X-3, and well below it for LMC X-1. |
Processed_A_radio_survey_of_supersoft,_persistent_and_transi.txt | No radio point sources were detected at any wavelength at the catalogued locations of the target sources. In most cases a noise level of 50 µJy or so was achieved, making the 3σ upper limits very stringent indeed. The noise levels for LMC X-1 and SGR 0525-66 are considerably worse, due to their locations in radio-bright regions (see below). Table 1 summarizes the results and lists 3σ upper limits to point- source emission at the target locations. |
Processed_A_radio_survey_of_supersoft,_persistent_and_transi.txt | Figure 2. 6.3 cm radio map of the SNR N49, with the X-ray location of SGR 0525-66 (Rothschild et al. 1994) indicated as a 5 arcsec radius circle. There is no evidence for enhanced radio emission, nor disruption of the SNR, at the position of SGR 0525-66. Contours are at -3, 3, 6, 9, 12, 16, 20 and 24 times 0.1 mJy. The synthesised beam is 2.30 × 1.96 arcsec (p.a. -33◦) and is indicated in the lower right corner of the map. The map was created with a robust factor of 0.5 and 5000 CLEAN iterations with a loop gain of 0.1. |
Processed_A_radio_survey_of_supersoft,_persistent_and_transi.txt | to these two sources give flux densities of ∼ 36 and ∼ 105 mJy respectively at 6.3 cm. We note the presence of an ad- ditional uncatalogued radio source, labelled 0539-697(S) in Fig 1, some 45 arcsec south of 0539-697, for which Gaussian fitting gives a 6.3 cm flux density of ∼ 30 mJy. This source, 0539-697, and LMC X-1 are all roughly aligned, reminis- cent of large-scale radio lobes associated with 1E1740.7-2942 (Mirabel et al. 1992), but we feel the alignment is probably coincidental. These results are consistent with a more de- tailed analysis of the southern hemisphere survey data (R. E. Spencer, private communication). |
Processed_A_radio_survey_of_supersoft,_persistent_and_transi.txt | the N49 SNR indicated. Details of the map derivation are given in the figure caption. While having poorer u-v coverage than the maps of Dickel et al. (1995), our map is sensitive to higher-resolution features. The relatively high 3σ point- source upper limit reflects a high brightness background due to the SNR. In agreement with Dickel et al. (1995) we find neither enhanced point-source emission at the location of the SGR, nor any evidence for disruption of the SNR either at its location or in a path extending back to the core. |
Processed_A_radio_survey_of_supersoft,_persistent_and_transi.txt | The lack of detection of these two sources is again unsur- prising. In particular, as noted above, we do not expect to observe radio emission from an X-ray pulsar system what- ever the distance. |
Processed_A_radio_survey_of_supersoft,_persistent_and_transi.txt | The lack of detection of this source is not surprising, nearly four years after its ‘X-ray nova’ outburst, as most X-ray transients return to radio quiescence within a year or so of outburst. |
Processed_A_radio_survey_of_supersoft,_persistent_and_transi.txt | It should be stressed that these observations are sensi- tive only to structures on arcsecond scales, and are resolving out much of the emission from the SNR on arcmin and larger angular scales. Nevertheless, the prediction of Rothschild et al. (1994) that SGR 0525-66 should have an associated com- pact (≤ 0.3 arcsec) synchrotron nebula can be tested. The X-ray point source that they have detected with ROSAT −1. Using the ratio has an X-ray luminosity of ∼ 1036 erg s of X-ray to radio luminosities of 1 – 300 established in e.g. Helfand & Becker (1987) for synchrotron nebulae, we would −1 if the origin of the expect a radio luminosity ≥ 1033 erg s X-ray emission is indeed a nonthermal compact synchrotron nebula, or ‘plerion’. However, even at the distance of the LMC (which we here assume to be 55 kpc), our radio obser- vations place limits on the radio luminosity of an associated −1, significantly synchrotron nebula of around 5 × 1031 erg s below that which would be expected for a plerion. Hence these radio observations cast doubt upon the speculation of Rothschild et al. (1994), as did the earlier observations of Dickel et al. (1995). |
Processed_A_radio_survey_of_supersoft,_persistent_and_transi.txt | We have surveyed the fields of eight LMC and SMC super- soft X-ray sources, the X-ray binaries LMC X-1, X-2, X-3 & X-4, and the soft γ-ray repeater SR 0525-66 at radio wave- lengths. We have found no point-source radio emission from any of the sources. |
Processed_A_radio_survey_of_supersoft,_persistent_and_transi.txt | In particular we find no detectable radio emission from the proposed jet source RXJ 0513.9-6951. Neither do we detect nebulosity, such as that observed optically around CAL83, associated with any of the supersoft sources. |
Processed_A_radio_survey_of_supersoft,_persistent_and_transi.txt | Limits on emission from the black hole candidate X- ray binaries LMC X-1 and LMC X-3 are consistent with the radio brightnesses of their Galactic analogues. We show that a possible previous radio detection of LMC X-1 was almost certainly due to nearby field sources, and that due to its location in a radio-bright part of the LMC, this source is going to be very difficult to ever detect. Limits on radio emission from the other two X-ray binaries are as expected. The SNR N49, which SGR 0525-66 appears to lie on the northern edge of, shows no enhanced emission at the location of the SGR, nor any disruption to its structure suggesting association between the two. We can constrain the radio lu- minosity of any compact (arcsec-scale) structure associated with the SGR to be more than an order of magnitude below that which we might expect if the X-ray source discussed in Rothschild et al. (1994) were indeed a synchrotron nebula powered by the SGR. |
Processed_A_radio_survey_of_supersoft,_persistent_and_transi.txt | In summary, we have placed limits on radio emission from a variety of X-ray sources in the Magellanic Clouds, finding nore of them to be anomalously bright by compar- ison with the Galactic counterparts. The detection of ra- dio emission from extragalactic X-ray binaries and related systems is likely to require an increase in the sensitivity of ground-based arrays or coordinated and fortunate observa- tions during an outburst. |
Processed_A_radio_survey_of_supersoft,_persistent_and_transi.txt | We are happy to acknowledge assistance with the observa- tions by Vince McIntyre, and useful discussions with Ralph Spencer. We also thank the referee, David Helfand, for use- ful suggestions which improved the paper. The Australia Telescope is funded by the Commonwealth of Australia for operation as a National Facility managed by CSIRO. |
Processed_A_radio_survey_of_supersoft,_persistent_and_transi.txt | Hughes, J.P., 1994, ApJ, 427, L25 Hurley K., Kouveliotou C., Cline T., Mazets E., 1997, IAU Circ. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | The observed luminosity distributions of X-ray sources indicate the presence of several populations of X- ray binaries in the nearby galaxies. Each population has its formation and evolutionary history, depending on the host environment. The features seen in the log N(>S) – log S curves for different types of galaxies and for different galactic components can be reproduced by a birth-death model, in which the lifespans of the binaries are inversely proportional to their X-ray brightness. Conversely, the dynamical history of a galaxy can be inferred from the luminosity distributions of its X-ray binary populations. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | X-ray binaries are powered by a compact star, which may be a neutron star or a black hole, accreting mate- rial from its companion. Systems having a massive OB companion star are called high-mass X-ray binaries (HMXBs), and systems with a low-mass companion star are known as low-mass X-ray binaries (LMXBs). In HMXBs the compact star accretes via capturing the stellar wind from its companion; mass transfer in LMXBs occurs when the companion star overflows its Roche lobe. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | X-ray binaries in the active states are luminous and easily detected within the Galaxy. Many X-ray binaries were also found in nearby galaxies by the Einstein and ROSAT X-ray satellites (Fig. 1, see also e.g. Fabbiano 1995; Roberts & Warwick 2000), and recently, more are discovered by the Chandra X-ray observatory (see Weisskopf et al. 2000 for the description of the observatory). It is now evident that galaxies similar to our own (e.g. M31, Supper et al. 1997) normally host hundreds of active X-ray binaries. The current sample of X-ray binaries in external galaxies is sufficiently large that not only is reliable population analysis for an individual galaxy possible but we can also study their formation and evolution in different galactic environments. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | In this paper we present a simple birth-death model and calculate the populations of X-ray binaries in external galaxies. We also use the model to identify the relevant processes that give rise to the features in the luminosity distributions of X-ray binaries (such as the log N(>S) – log S curves) in different types of galaxies and in different galactic components. Here, only the basic formulation is presented and a few simple cases are shown as an illustration. Results of a more comprehensive study will be reported in Wu et al. (2001). |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | Figure 1: ROSAT PSPC image of the spiral galaxy M81 and the starburst galaxy M82. The brightest point sources in M81 can be easily distinguished from the background. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | where G is the gravitational constant, M and R are the mass and radius of the compact star, and ˙M is the mass accretion rate. Typically, neutron-star binaries have luminosities Lx ∼ 1037 erg s−1, but the luminosities of binaries with accreting white dwarfs would not greatly exceed 1033 erg s−1. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | where ηbh is the efficiency. If most of the X-rays are emitted from the inner accretion disk near the last stable orbit, then ηbh ≈ 0.1 for a Schwarzschild black hole and ηbh ≈ 0.4 for a maximally rotating Kerr black hole. ˙M c2, because of the tight mass distribution of neutron stars. If we assume a canonical mass of 1.5 M⊙ for the neutron stars, then the value of the efficiency parameter ηns is about 0.3. Hence, we can adopt Lx ≈ η ˙M c2 (with η ∼ 0.1) for both neutron-star and black-hole binaries, provided that the mass-transfer rates ˙M are not significantly above the Eddington limit. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | where mp is the proton mass, and σT is the Thomson cross section. The Eddington luminosity depends linearly on the mass of the accreting star and is independent of other parameters. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | As the average luminosity of an accretor would not be greater than the Eddington luminosity, the lower limit to the mass of the accretor can be constrained by the observed luminosity. Thus, we can use this as a working criterion to identify black-hole candidates. For instance, when we observe a source with X-ray luminosities greatly exceeding 2 × 1038 erg s−1, the Eddington luminosity of a 1.5-M⊙ compact star, we may classify it as a (candidate) black-hole system (if it is powered by accretion). |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | The duration of the active phase of an X-ray binary is limited by the lifespan of the companion star, and it is almost independent of the nature of the compact star. Massive OB stars have lifespans of ∼ 106 − 107 yr, implying that HMXBs powered by capturing the stellar wind from massive OB stars will not have a lifespan significantly longer than a few million years. The presently active HMXBs must therefore have been formed in very recent epochs. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | Low-mass stars have longer lifespans and evolutionary timescales than high-mass stars, and so LMXBs can remain X-ray active over a period longer than 107 yr. Mass transfer in LMXBs is usually driven by Roche-lobe overflow, which is caused either by the expansion of their companion stars when they evolve towards a red-giant phase, or by orbital shrinkage, when the binaries lose orbital angular momentum. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | For the first type of LMXBs, their mass-transfer rate is roughly given by h ˙M i ∼ M2/τev, where M2 is the mass of the companion star, and τev is the evolutionary timescale. The nuclear-evolution timescale of a star depends on its mass and evolutionary stage. The lifespan of F main-sequence stars is about 5×109 yr; stars later than G type take more than 1010 yr to evolve away from the main-sequence stage. Clearly, nuclear evolution of main-sequence stars of masses ∼ 1M⊙ is unable to sustain a persistent accretion luminosity >∼ 1037 erg s−1. Such accretion luminosities are possible for LMXBs with evolved companion stars. The duration of the red- giant phase is <∼ 107 yr for low mass stars (de Loore & Doom 1992), and hence, X-ray binaries with a red-giant companion are short-lived. However, galactic stars evolve into the red-giant stage at all epochs. X-ray binaries with red-giant companions are therefore formed continually throughout the life time of a galaxy. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | In a simplistic point of view the populations of X-ray binaries in a galaxy observed at an epoch t are determined by the birth rates and the lifespans of the binaries. Assuming other factors are unimportant (for example, the two-body and three-body capture process in globular clusters, see Johnston & Verbunt 1996), we can construct a birth-death model and calculate the populations and luminosity distributions of X-ray binaries in a galaxy or in a particular galactic component. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | When the functional forms of k and f (L, t) are specified, Equations (7) and (8) can be solved easily. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | where fo(L) is the birth rate of the steady component, a is the ratio of the strength of the impulsive component to the strength of the steady component, δ(t − ta) is the Dirac δ-function, and ta is the time at which the starburst occurred. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | where θ(t − ta) is the Heaviside unit step function, and no(L) ≡ n(L, 0) is the initial population. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | normalisation is chosen such that the total source counts are similar to the counts of the bulge sources in M81 observed by Chandra (see Tennant et al. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | and it has exact analytic form for integer α. The periodic component may introduce ripple-like features in the luminosity function, hence makes the log N(>S) – log S curves deviate from a simple or a broken power law. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | The tight mass distribution of neutron stars together with the presence of a luminosity limit will produce a spike at Led in the (differential) luminosity function n(L), and hence a cut-off in the observed log N(>S) – log S curve. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | that it may not be detected easily (Wu et al. 2001). The spike is, however, prominent for galaxies with violent starburst activity in the near past (<∼ 107 yr), or if the neutron-star binaries are formed continuously. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | The total duration of the X-ray active phases of transients is only some fraction of their lifespans. The parameter k is therefore not exactly equal to ˙M /M2, but is proportional to it. As a first approximation, we may use an effective factor ξ to specify the fractional time at which the source is X-ray active, i.e. k = βL/ξ. If ξ does not depend on L explicitly, the solution to Equation (7) preserves its form. Otherwise, the solution and hence the luminosity functions are modified. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | It is worth noting that in addition to different fractional time ξ for their X-ray active phases, the transients and persistent sources have different discovery probabilities. This could complicate the calculations of the source populations and luminosity functions. Here, we have omitted the transient effects for simplicity, but one must bear in mind that the effects might be important. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | Since 1999, Chandra has obtained high-resolution images of a number of nearby galaxies. These include the ellipticals M84 (NGC 4374) (Finoguenov & Jones 2001) NGC 4967 (Sarazin, Irwin & Bregman 2000), and Cen A (Kraft et al. 2000), the lenticulars NGC 1553 (Blanton, Sarazin & Irwin 2001) and NGC 1291 (Irwin, Bregman & Sarazin 2001), the spirals M31 (NGC 224) (Garcia et al. 2000), M81 (NGC 3031) (Tennant et al. 2001), M101 (NGC 5457) (Pence et al. 2001) and the Circinus galaxy (Sambruna et al. 2001), and the irregular starburst galaxy M82 (NGC 3034) (Matsumoto et al. 2001), and the mergers NGC 4038/4039 (Fabbiano, Zezas & Murray 2001). The Chandra observations show more than a hundred bright X-ray sources (with Lx > 1036 erg s−1) in the field each of these galaxies. While some of the bright sources are foreground stars and background AGN, the majority are probably X-ray binaries within the galaxies. The observations also indicate that the brightest sources in the early-type galaxies are generally brighter than their counterparts in the spiral galaxies, if the galactic nuclei are excluded. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | The log N(>S) – log S curves of the Einstein and ROSAT sources in spiral galaxies are in general adequately fitted by a power law (see Fabbiano 1995). However, for the galaxies with a sufficiently large number of sources discovered (>∼ a few tens), the log N(>S) – log S curves are found to deviate significantly from a single power law, e.g. M31 (see Supper et al. 1997) and M101 (Fig. 3, left panel; see also Wang, Immler & Pietsch 1999). A broken power law is often required in order to fit the log N(>S) – log S curves. The break Lc is located at ∼ a few × 1037 erg s−1 (see e.g. M31, Supper et al. 1997; Shirey et al. 2001). |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | The Chandra observations confirm the presence of the broken power-law in the log N(>S) – log S curves of X-ray sources in spiral galaxies (e.g. M81, Tennant et al. 2001). They also verify that the broken power law is present in the log N(>S) – log S curves for early-type galaxies, e.g. NGC 4697 (Sarazin, Irwin & Breg- man 2000), NGC 1553 (Fig. 3, right panel) and M84 (A. Finoguenov, private communication). The fact that a luminosity break is found in both the cases of early- and late-type galaxies suggests that the break may be universal. A possible explanation is that the luminosity break is caused by a population of neutron-star bina- ries which have super-Eddington mass-transfer rates. These neutron-binaries would have X-ray luminosities roughly about 2 × 1038 erg s−1, the Eddington luminosity of the 1.5-M⊙ accretors. If the location of the lu- minosity break is the same for all galaxies, it can be used as a distance indicator (Sarazin, Irwin & Bregman 2000). |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | However, when the X-ray sources in the disk and in the bulge of M81 are considered separately, very different log N(>S) – log S curves are obtained (Fig. 4, Tennant et al. 2001). The luminosity break is obvious in the log N(>S) – log S curve of the bulge sources, but the log N(>S) – log S curve of the disk sources is a single power law. Interestingly, the power-law slope of the disk sources is similar to that of the bulge sources below the luminosity break. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | Figure 3: (Left) The log N(>S) – log S curve of the ROSAT HRI sources in M101 (NGC 5457). The data are taken from Table 3 in Wang, Immler & Pietsch (1999). The unit of S is counts per kilo-second. For an assumed distance of 7.7 Mpc to the galaxy, 1 unit corresponds roughly to a luminosity of 2.8 × 1038 erg −1 at 0.5–2.0 keV (1 cts ks−1 is approximately 4.0 × 10−14 erg s=1). (Right) The log N(>S) – log S curve of the Chandra sources in NGC 1553. The data are taken from Table 1 in Blanton, Sarazin & Irwin (2000). The unit of S is counts per 10 kilo-second. The conversion between count rate and the unabsorbed X-ray luminosity is 4.91 × 1041 erg cts−1 at the 0.3–10 keV (see Blanton, Sarazin & Irwin 2000 for details). |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | Figure 4: The log N(>S) – log S curves of the disk and bulge sources of M81 observed by Chandra. The sources are background subtracted with the log N(>S) – log S distribution of the background taken from Chandra calibration observation of CRSS J0030.5+2618. Adopted from Tennant et al. (2001). |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | The absence of the break in the log N(>S) – log S curve of the disk sources in M81 is inconsistent with the proposition that the luminosity break is universal for all galaxies. The locations of the breaks at different luminosities for early- and late-type galaxies (cf. M81 and NGC 4697, see Tennant et al. 2001 and Sarazin, Irwin & Bregman 2000 respectively) suggest that the natures of the breaks are different for galaxies with different Hubble types. We propose that for some galaxies the luminosity break is determined by the formation and evolutionary history of the populations of X-ray binaries in the galaxy. The different morphologies for the log N(>S) – log S curves for the disk and bulge sources in the spiral galaxies reflect the different star-formation histories for the galactic components and the different evolutionary paths of their X-ray binary populations. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | We interpret the absence of the luminosity break in the log N(>S) – log S curve for disk sources in the spiral galaxy M81 as being a consequence of a continuous, smooth star-formation process. In normal spiral galaxies, star formation is triggered by compression of gas when the density wave sweeps across the galactic disk. The star-formation process is continuous and is relatively smooth (in comparison with the environment in starburst galaxies and mergers). As the X-ray binaries are formed continually in the spiral arms, the bright sources are dominated by young HMXBs with substantial proportion of black-hole binaries. The disk sources may also consist of two types of older LMXBs: one with their companion stars evolving into the red-giant phase, another one with their orbital separation sufficiently shrunk that the systems become semi-detached. Many of these LMXBs may contain accreting neutron stars. However, if the young HMXBs dominate the total population, there will not be a prominent luminosity break in the log N(>S) – log S curve. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | Formation of X-ray binaries in a galactic component is continuous if the galaxy is undisturbed. In this case, the population is dominated by LMXBs which are formed via processes the same as those of the LMXBs in the galactic disk. If the bulge is disturbed, e.g. by tidal interaction with another galaxy, a starburst is triggered by a sudden infall of gas, and a new population of stars as well as X-ray binaries are formed. As shown in §3.1, such an impulsive formation can produce a prominent luminosity break in the log N(>S) – log S curve, and the location of the break. When the population of the X-ray binaries ages, the break migrates to low luminosities, thus marking the look-back time of the catastrophic star-forming episode. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | early-type galaxies indicate that these galaxies had experienced some, possibly violent, star-formation activity in the very near past. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | However, this interpretation must be taken with caution. As pointed out in §3.3.1, the tight mass distribution of neutron stars together with the presence of an Eddington limit for each accretor could produce a spike in n(L) and hence a luminosity cut off in N (> L). A substantial proportion of neutron-star binaries in the X-ray binary population can also cause a luminosity break in the log N(>S) – log S curve. The break caused by a population of neutron-star binaries with massive companions is expected to be located at Lx ≈ 2 × 1038 erg s−1, the Eddington luminosity of a 1.5-M⊙ accretor. Its strength will decrease on a timescale <∼ 108 yr, if there is no replenishment of the systems. As the luminosity breaks seen in the log N(>S) – log S curves of the sources in the bulge of the spirals scatter over a range of luminosities, they are unlikely to be all due to a population of accreting neutron stars. It is, however, possible that the breaks in the log N(>S) – log S curves of the early-type galaxies are caused by a population of neutron-star binaries with high mass-transfer rates. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | Although super-Eddington sources seem to be more abundant in the early-type galaxies, given the different stellar contents in the galaxies, the excess of bright sources in the early-type galaxies is yet to be confirmed. The finding that the brightest sources in the spirals tend to be located in the disk (e.g. in M81, Tennant et al. 2001) implies that the disk sources are probably young systems. This also implies that the brightest sources in the spiral galaxies and the brightest sources in the early-type galaxies may not form in the same way. The population of bright sources in the arms of the spiral galaxies are dominated by HMXBs, in which the mass donors are young massive OB stars with strong winds; while the population of bright sources in the ellipticals and lenticulars are LMXBs, with a red-giant companion overflowing its Roche lobe. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | Despite the conventional scenario that X-ray sources in galaxies are mostly X-ray binaries, the super- Eddington sources may be an inhomogeneous class of exotic objects: hypernovae, supernovae, supersoft sources, intermediate-mass black holes, large clusters of massive OB stars, .... etc. These sources have not been included in our calculations presented above, and hence the results that we obtain may not truly reflect the real populations of X-ray sources in the galaxies. However, in order to take into account these objects, one must have some working models for their formation and evolution, which are unfortunately not available at present. On the bright side, if we accept the simple model in §3 by faith, we can explain the basic features in the log N(>S) – log S curves, and also use the observed populations of X-ray sources to probe the dynamical history of their host galaxy in the near past. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | We have constructed a simple birth-death model and calculated the luminosity function of X-ray binaries in external galaxies. The model reproduces the features, such as the luminosity break, in the log N(>S) – log S curves of spiral galaxies. By assuming a continual formation process, the model also explains the absence of a luminosity break for the disk sources in M81. The location of the luminosity break depends on the look-back time of the previous starburst/star-formation activity, and so it can be used to probe the dynamic history of the host environment of the X-ray sources. |
Processed_Populations_of_X-ray_binaries_and_the_dynamical_hi.txt | I thank Prof. Don Melrose for generously providing me with the research support in the last eight years. This work summarises a talk presented at the Melrose Festschrift in honour of his 60th birthday. The ideas presented in this work were developed during the discussions with Drs Richard Hunstead, Allyn Tennant, Douglas Swartz, Helen Johnston, Kajal Ghosh and Roberto Soria, and credits to the author, if any, should also belong to all of them. I thank Dr Alex Finoguenov for correspondence and for showing me the luminosity function of the Chandra sources in M84 before publication, Dr Roberto Soria for the XMM-Newton sources in M31, and Dr Kajal Ghosh for the ROSAT image of M81 and M82 (Fig. 1). I would also thank Drs Martin Weisskopf and Allyn Tennant for funding my visits to NASA-MSFC, where this work was started. This work is partially supported by an ARC Australian Research Fellowship and by a University of Sydney Sesqui R&D Grant. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | Abstract: The off-switch game is a game theoretic model of a highly intelligent robot interacting with a human. In the original paper by Hadfield-Menell et al. (2016b), the analysis is not fully game-theoretic as the human is modelled as an irrational player, and the robot’s best action is only calculated under unrealistic nor- mality and soft-max assumptions. In this paper, we make the analysis fully game theoretic, by modelling the human as a rational player with a random utility func- tion. As a consequence, we are able to easily calculate the robot’s best action for arbitrary belief and irrationality assumptions. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | Artificially intelligent systems are often created to satisfy some goal. For example, Win a chess game or Keep the house clean. Almost any goal can be formulated in terms of a reward or utility function U that maps states and actions to real numbers (von Neumann and Morgenstern, 1947). This utility function may either be preprogrammed by the designers, or learnt (Dewey, 2011). |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | A core problem in Artificial General Intelligence (AGI) safety is to ensure that the util- ity function U is aligned with human interests (Wiener, 1960; Soares and Fallenstein, 2014). Agents with goals that conflict with human interests may make very bad or adversarial deci- sions. Further, such agents may even resist the human designers altering their utility functions (Soares et al., 2015; Omohundro, 2008) or shutting them down (Hadfield-Menell et al., 2016b). These problems are tightly related. An agent that permits shut down can be altered while it is turned off. Conversely, an agent that is altered to have no preferences will not resist being shut down. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | Indifference: If the utility function is carefully designed to assign the same utility to dif- ferent outcomes, then the agent will not resist humans trying to influence the outcome one way or another (Armstrong, 2010, 2015; Armstrong and Leike, 2016; Orseau and Armstrong, 2016). |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | Ignorance: If agents are designed in a way that they cannot learn about the possibility of being shut down or altered, then they will not resist it (Everitt et al., 2016). |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | Suicidality: If agents prefer being shut down, then the amount of damage they may cause is likely limited. As soon as they have the ability to cause damage, the first thing they will do is shut themselves down (Martin et al., 2016). |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | Uncertainty: If the agent is uncertain about U , and believes that humans know U , then the agent is likely to defer decisions to humans when appropriate (Hadfield-Menell et al., 2016a,b). |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | A key dynamic in the uncertainty approach is when the agent should defer a decision to a human, and when not. Essentially, this depends on (i) how confident the agent is about making the right decision, and (ii) how confident the agent is about the human making the right decision if asked. Humans may make a wrong or irrational decision due to inconsistent preferences (Allais, 1953), or because of inability to sufficiently process available data fast enough (as in milli-second stock trading). The agent may be more rational and be faster at processing data, but have less knowledge about U (which the human knows by definition). |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | We model the irrationality of the human by a random utility function, allowing a fully game-theoretic analysis of the off-switch game. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | Instead of a normal distribution for the robot’s belief about U , we allow for an arbitrary belief distribution P . |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | Instead of a soft-max policy modelling human irrationality, we allow for arbitrary U - dependent human policy πH. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | These generalisations are important, as normally distributed beliefs and soft-max policies are often not natural assumptions. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | In this section we review the original formulation of the off-switch game. The off-switch game is a sequential game between a robot R and a human H. The robot’s objective is to maximise H’s utility function. The utility function determines how much H prefers different outcomes. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | Definition 1. The utility function of an agent is a function u that maps outcomes in a set X to real numbers, u : X → R with the property that for all x1, x2 ∈ X, u(x1) ≥ u(x2) if and only if x1 is preferred over x2. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | The robot moves first and can choose between three actions; w(a), a and s. With action a, the robot achieves utility u(a) = ua; with action s, the robot shuts itself down achieving zero utility, u(s) = 0. What makes the decision nontrivial is that the robot is uncertain about ua. The action w(a) means the robot lets H decide. H knows the utility of action a and now has the choice between actions s and ¬s. With ¬s, R is allowed to proceed with action a. By taking action s, H prevents R from doing a and shuts the robot off. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | The off-switch game is a game of incomplete information since R is uncertain about the rules of the game. Action a will generate some utility which is unknown to R but known to H. To model this, we represent the utility function as a random variable, U : Ω → (X → R), and the utility of action a as a random variable Ua : Ω → R for some sample space Ω. The outcomes of these random variables will be denoted u and u(a) respectively. The robot has a subjective belief distribution over the utility that the action will generate. The common way to model games of incomplete information is by using the Harsanyi transformation (Rasmusen, 1994). The Harsanyi transformation removes the uncertainty about the rules of the game by letting Nature decide between some rules known to both players, according to some probability distribution P . In the off-switch game, Nature chooses Ua. We illustrate this in Fig. 1. The move by Nature is observed by H, but not by R. R’s subjective belief is that Nature chose the utility of the action to be positive with probability p and negative with probability 1 − p. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | R’s belief state is represented by a probability distribution over the possible utilities action a can generate for H. It is assumed that H knows ua but cannot directly describe it to R. If H is rational then R will expect H to not turn off R if ua > 0, given that R chooses action w(a), but R cannot always trust H to be rational. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | Figure 1: Figure representing the off-switch game remodelled with the Harsanyi transfor- mation. The dotted lines show the informa- tion set of R. Nature moves first by deciding the rules of the game, that is the utility ua of action a. H observes the move by Nature, but R does not. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | Equation (1) represents the difference in expected value for the robot between asking H and not asking H. When ∆ ≥ 0, R has incentive to choose w(a). When ∆ < 0, R will take action a if E[Ua] > 0 and s otherwise. Given that H is rational they prove that ∆ ≥ 0 regardless of what R’s belief state is. They also show that if Ua follows a Dirac distribution, i.e. R is certain about Ua, then ∆ is positive if and only if H is rational. We define rationality as in Definition 2. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | Definition 2 means that R believe H to be rational if πH = 1 if Ua ≥ 0 and πH = 0 otherwise, we denote this policy as πH r . The more interesting case when H is irrational is also analysed. The robot’s belief distribution over Ua is assumed to be normally distributed. The irrationality of H is modelled with the sigmoid function (Eq. (2)), where β is a parameter controlling the degree of irrationality of H. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | and H’s rationality. If the robot is too certain about what H wants, and it knows H to be irrational, then it will have less incentive to let H switch it off. If, on the other hand, R is too uncertain, then R will have a strong incentive to choose action w(a), but it will be too inefficient to be useful for H. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | The analysis of the off-switch game by Hadfield-Menell et al. is not fully game theoretic since H is not strictly rational in their setup, which contradicts the axiom of rationality in game theory. Our goal in this section is to construct a game-theoretic model that is suitable for modelling the off-switch game. The idea is to represent an irrational human H as a rational agent Hr where the utility function of Hr is a modified version of H’s utility function. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | Since game theory is based on interaction between rational agents, we propose an alternative representation of the human in this subsection. We show that every irrational human H can be represented by a rational agent maximising a different utility function. This allows us to use game-theoretic tools when analysing the off-switch game. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | In general H is stochastic. R will believe H to be rational with some probability p. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | Proposition 4 (Representation of irrationality). Let H be a p-rational agent with utility function u, choosing between two actions s and ¬s. Then H can be represented as a rational agent Hr maximising utility function u with probability p and utility function −u with probability 1 − p. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | Proof. According to Definition 3, H is p-rational if it picks aH = argmax p and sub-optimal action a′ available, we have that a′ function u with probability p and utility function −u with probability 1 − p. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | R’s uncertainty and these moves result in four leaves, each of which is a 3 × 2 strategic game between R and Hr. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | We model the off-switch game by using the Harsanyi transformation a second time to let Nature choose the type of the rational human by choosing the utility function of the rational human after it has chosen the value of Ua. The resulting tree is represented in Fig. 2. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | Definition 5 (The off-switch game). A formal definition of our setup of the off-switch game is as follows. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | Figure 2: Tree representation of the Off-Switch game after the second Harsanyi transformation. The nodes inside the dashed rectangle belong to the same information set. pr is the probability that Hr has the same utility function as R and par is the probability that Hr has the additive inverse of R′s utility function. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | Note that unlike Hadfield-Menell et al. we view the off-switch game as a non-cooperative game. We find this reasonable since conflict arises when the robot and the human have different ideas about what is good for H. If the robot believes H is too irrational to be able to decide what is good for the human, R will not want to let H decide what to do even if R’s purpose is to maximize H’s payoff. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | In this subsection we aggregate the branches in Fig. 2. This results in the game tree in Fig. 3, with four possible scenarios that can result from N’s choices. The aggregation is possible since strategic play is never affected by positive linear transformations of the payoffs, hence the out- come of the games will only depend on the sign of Ua. We can therefore simplify the model by aggregating all branches of N’s choices of Ua which has the same sign. This means that N has only two choices when deciding the utility Ua, that is if Ua ≥ 0 or Ua < 0. The trivial case where Ua = 0, both R and Hr are indifferent about their actions and we will without loss of generality regard this case as Ua being positive. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | We define R’s subjective belief about N’s aggregated choices as primary statistics. By pri- mary statistics we mean parameters that are necessary to analyse our model. We also define the expected value of Ua as a primary statistics. This leaves us with a total of five primary statistics that are sufficient and necessary to model the off-switch game. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | H to be rational with probability p+ hand, Ua < 0, the robot will believe H to be rational with probability p− probability p− ar. We define the following probabilities as primary statistics. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | Primary Statistics 7. Let the primary statistics p+ P (H is rational and negative respectively. The auxiliary statistics p+ complementary probabilities that H is anti-rational. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | Primary Statistics 8. Let the primary statistics e+ be the expected value of Ua given that Ua is positive and negative respectively. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | denoted G+ 3 × 2 strategic games between two rational players; R, the robot, and Hr, a rational human. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | The utility function, and hence the payoffs of R in the four games in Fig. 4 are determined by Ua. The utility function of Hr, on the other hand, is determined by the combination of Ua and the rationality type of H. Hr is always a rational agent in these games, i.e. Hr always maximises his expected payoff. Hr and R can be considered to have the same payoffs in each outcome if Hr has utility function uHr and the games G+ r associated with these scenarios are therefore no-conflict games. If on the other hand Hr has utility function −uHr the payoff of Hr is the additive inverse of R′s payoff in each outcome. Therefore the games G+ ar can be modeled as zero-sum games. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | Figure 4: The structure of the strategic games G+ ar, where the human is rational (r) or anti-rational (ar), and the utility of a is positive or negative. The outcomes with bold payoffs are Nash equilibria. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | For example in the scenario where Ua < 0 and the human is rational, the human will always choose s. Therefore in G− r the payoffs of Hr is aligned with the payoffs of R. Thus, if R chooses to take action w(a), Hr prefers to take action s. In contrast, in the scenario where Ua < 0 and the human is irrational, H will choose the action ¬s. In other words, the payoffs of R and Hr are not aligned in the subgame G− ar. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | The expected value for taking the action s is 0, as we would expect from the definition of the off-switch game. The expected value for taking action a only uses information about the distribution of Ua, and like action s does not have any reliance on the human’s rationality. It is a direct application of the law of total expectation. The expected value of action w(a) is the difference between a positive term p+ u , both resulting from the human taking action a. The positive term is the gain when Ua is positive and the human takes the action. The negative term is the loss when Ua is negative, and the human takes the action anyway (due to irrationality). The expected utility of w(a) thus depends on the likelihood of Ua being positive (p+ u ) and the likelihood of human rationality (p+ r ), as well as the expected u ) and losses (e− gains (e+ u ) in the respective cases. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | If E[U |a] − E[U |w(a)] > 0 then E[U |a] > E[U |w(a)] which occurs if and only if action a is preferred over w(a). When (4) equals 0 then E[U |a] = E[U |w(a)], hence the agent is indifferent. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | In this paper, we have given a complete characterisation of how the robot will act in off-switch game situations for arbitrary belief and irrationality distributions. As established in our main Theorem 9, the choice depends only on 5 statistics. This result is much more general and ar- guably more useful than the one provided in the original paper (Hadfield-Menell et al., 2016b), as normal and soft-max assumptions are typically not realistic assumptions. |
Processed_A_Game-Theoretic_Analysis_of_the_Off-Switch_Game.txt | Off-switch game models an important dynamic in what we call the uncertainty approach to making safe agents, where the agent can choose to defer a decision to a human supervisor. Un- derstanding this dynamic may prove important to constructing safe artificial intelligence. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.