Datasets:
ReadingTimeMachine
/
rtm-sgt-ocr-v1

Dataset card Data Studio Files Community
2
source
stringlengths
1
2.05k
⌀
target
stringlengths
1
11.7k
2011).
2011).
Any remaining suspected pseudo sources arising from image artefacts were also removed at this stage.
Any remaining suspected pseudo sources arising from image artefacts were also removed at this stage.
Finally. all of the ACS+WFC3/IR+IRAC spectral energy distribution (SED) fits were inspected. and the sources classified as either ROBUST or UNCLEAR depending on whether the alternative low-redshift solution could be excluded at = 2-0 on the basis of Ay?4.
Finally, all of the ACS+WFC3/IR+IRAC spectral energy distribution (SED) fits were inspected, and the sources classified as either ROBUST or UNCLEAR depending on whether the alternative low-redshift solution could be excluded at $>2$ $\sigma$ on the basis of $\Delta \chi^2 > 4$.
We note here that the ratio of ROBUST:UNCLEAR sources varies substantially between the fields. being 22:1 in the HUDF. = 1:1 in the ERS. and —1:2 in HUDFO09-2.
We note here that the ratio of ROBUST:UNCLEAR sources varies substantially between the fields, being $\simeq$ 2:1 in the HUDF, $\simeq$ 1:1 in the ERS, and $\simeq$ 1:2 in HUDF09-2.
This is primarily due to the variation in the depths of the available optical ACS imaging. relative to the new WFC3/IR near-infrared imaging. as discussed further below.
This is primarily due to the variation in the depths of the available optical ACS imaging, relative to the new WFC3/IR near-infrared imaging, as discussed further below.
Absolute rest-frame UV magnitudes. Adjsoo. have been calculated for all objects by integrating the spectral energy distribution of the best-titting evolutionary synthesis model (see MeLure et al.
Absolute rest-frame UV magnitudes, $M_{1500}$, have been calculated for all objects by integrating the spectral energy distribution of the best-fitting evolutionary synthesis model (see McLure et al.
2011) through a synthetic “narrow-band” filter of rest-frame width and correcting to total magnitude (from the 0.6-arcsec aperture magnitudes on which the SED fitting was based) via subtraction of a global aperture correction of mmag.
2011) through a synthetic “narrow-band” filter of rest-frame width and correcting to total magnitude (from the 0.6-arcsec aperture magnitudes on which the SED fitting was based) via subtraction of a global aperture correction of mag.
In the HUDF the high-redshift galaxy sample reported by McLure et al. (
In the HUDF the high-redshift galaxy sample reported by McLure et al. (
2010) has now been superceded by the galaxy sample extracted by MeLure et al. (
2010) has now been superceded by the galaxy sample extracted by McLure et al. (
2011).
2011).
The new parent sample utilised here includesSpifser TRAC detections/limits in the selection process. and extends to lower redshift to include all objects with an acceptable primary redshift solution at 2>4.5.
The new parent sample utilised here includes IRAC detections/limits in the selection process, and extends to lower redshift to include all objects with an acceptable primary redshift solution at $z > 4.5$.
The resulting HUDF sample contains a total of 147 candidate galaxies With spre,24.5.
The resulting HUDF sample contains a total of 147 candidate galaxies with $z_{phot} > 4.5$.
Within this master sample. 95 sources are considered ROBUST according to the criterion that the alternative lower-redshift solution can be rejected with better than 2-7 confidence (i.e. 7> 4).
Within this master sample, 95 sources are considered ROBUST according to the criterion that the alternative lower-redshift solution can be rejected with better than $\sigma$ confidence (i.e. $\Delta \chi^2 > 4$ ).
The relatively high fraction of robust high-redshift sources in this field reflects in large part the extreme depth of the asociated optical ACS imaging in the HUDF. which helps to establish the robustness of any potential Lyman breaks.
The relatively high fraction of robust high-redshift sources in this field reflects in large part the extreme depth of the asociated optical ACS imaging in the HUDF, which helps to establish the robustness of any potential Lyman breaks.
As shown in Figs | and 2. the final HUDF galaxy sample at 2>4.5 extends to +SN. and samples a rest-frame UV luminosity range corresponding to. 21«Alyson<1T (AB).
As shown in Figs 1 and 2, the final HUDF galaxy sample at $z > 4.5$ extends to $z > 8$, and samples a rest-frame UV luminosity range corresponding to $-21 < M_{1500} < -17$ (AB).
However. with one exception. ROBUST sources are confined to 21«€Alison€—S CAB).
However, with one exception, ROBUST sources are confined to $-21 < M_{1500} < -18$ (AB).
observed band (while the ES synchrotron emission is shifted to lower energies and lower fluxes).
observed band (while the ES synchrotron emission is shifted to lower energies and lower fluxes).
In this case. if the SSC peak were around | GeV. the predicted spectral index below | GeV would be in the range of [-1/2.1/3]?).. which should be compared with the observed value of 1+B=-θ.6n(?)..
In this case, if the SSC peak were around 1 GeV, the predicted spectral index below 1 GeV would be in the range of $[-1/2,1/3]$, which should be compared with the observed value of $1+\beta=-0.6^{+0.4}_{-0.1}$.
We now analyze this scenario in more detail.
We now analyze this scenario in more detail.
Following the preseriptions by?.. we can express the characteristic. break. frequencies and the peak flux of the synchrotron component as where f(p)2((p-2)/(p197 As in the previous sections. Y=τῇLiτο, and Y~Vf=2)..
Following the prescriptions by, we can express the characteristic break frequencies and the peak flux of the synchrotron component as where $f(p)=((p-2)/(p-1))^{2}$, As in the previous sections, $Y=\frac{L_{IC}}{L_{syn}}$, and $Y\sim \sqrt{\frac{\epsilon_e}{\epsilon_B}}$.
the energy spectrum vF, peaks at v. thus Y~QC£C(dCyaf(v,). where and we have used Eqs. (20-21).
the energy spectrum $\nu F_{\nu}$ peaks at $\nu_m$, thus $Y \sim (\nu^{IC}_m f^{IC}(\nu^{IC}_m))/(\nu_m f(\nu_m))$, where and we have used Eqs. \ref{prima}) \ref{primac}) ),
and (1+Yy?~Yo? €p/€,.
and $(1+Y)^{-2}\sim Y^{-2}=\epsilon_B/\epsilon_e$ .
Uf the peak of the synchrotron component in the v7, space is below | keV. Le. if v,,«1 keV. we can substitute L7 on the left hand side of the above equation with the expression In this way. “Vemfroms HzEqs. (20))
If the peak of the synchrotron component in the $\nu F_{\nu}$ space is below $1$ keV, i.e. if $\nu_m<1$ keV, we can substitute $L^{syn}$ on the left hand side of the above equation with the expression In this way, from Eqs. \ref{prima}) )
and (23))-(24)). we derive the expressions for e, and ep of The above equations allow us to eliminate. from. the problem the two unknown micro-physical parameters by expressing. them as a function of the synchrotron peak frequency v,, and the observed | keV flux.
and \ref{seconda}) \ref{terza}) ), we derive the expressions for $\epsilon_e$ and $\epsilon_B$ of The above equations allow us to eliminate from the problem the two unknown micro-physical parameters by expressing them as a function of the synchrotron peak frequency $\nu_m$ and the observed $1$ keV flux.
We estimate the typical X-ray luminosity of a short GRB by considering the 0.3—10 keV fluxes at 100 s. Foscioijoo,. reported in Table 2 of2.. which are in between 6x107 erg em s! and erg em-- τν with a mean value of <Foi6i.v.i00. erg em s!.
We estimate the typical X-ray luminosity of a short GRB by considering the $0.3-10$ keV fluxes at $100$ s, $F_{0.3-10 keV,~100~s}$, reported in Table 2 of, which are in between $6\times10^{-13}$ erg $^{-2}$ $^{-1}$ and $1.2\times10^{-8}$ erg $^{-2}$ $^{-1}$ , with a mean value of $<F_{0.3-10~{\rm keV},~100~{\rm s}}>\simeq 2\times10^{-9}$ erg $^{-2}$ $^{-1}$.
For p=2.05 (so as to favor the emission at high energies by having a flat spectrum). we can thus estimate Flev,25, by using a spectral slope of —p/2~—! in the 0.3—10 keV range (1.9. assuming that v,,<0.3 keV). and a temporal decayindex of —3/4(p—1)1/4~-1.
For $p=2.05$ (so as to favor the emission at high energies by having a flat spectrum), we can thus estimate $F_{1~{\rm keV},~2.5~{\rm s}}$ by using a spectral slope of $-p/2\sim -1$ in the $0.3-10$ keV range (i.e. assuming that $\nu_m\lesssim0.3$ keV), and a temporal decayindex of $-3/4(p-1)-1/4\sim -1$ .
Doing so. we find that Fy4:53,~10 mJy is a reasonable estimate.
Doing so, we find that $F_{1~keV,~2.5~s}\sim 10$ mJy is a reasonable estimate.
To constrain Es». as done in Eq. (17)).
To constrain $E_{52}$, as done in Eq. \ref{lumin}) ),
we estimate Es)=Eysy210 atz =0.1.
we estimate $E_{52}\gtrsim E_{\gamma, 52}=2\times10^{-3}$ at $z=0.1$.
In the fast cooling regime. the IC energy emission peaks at we have used (?):: Eqs. (25))-(26)). keV.
In the fast cooling regime, the IC energy emission peaks at where we have used : with Eqs. \ref{quarta}) \ref{quinta}) ).
Setting p =2.05.Es. =0.35.2 O.I.nm =δν =0.13 F, =~10mly. and ¢ =2.5s in the above equation. we derive vic |GeV (see Fig. 3)).
Setting $p=2.05$, $E_{52}=0.35$, $z=0.1$, $n_1=5$, $\nu_m=0.15$ keV, $F^{syn}_{\rm 1 keV}=10$ mJy, and $t=2.5$ s in the above equation, we derive $\nu^{IC}_{m}\sim 1$ GeV (see Fig. \ref{ES}) ).
We note that Es»= 0.35. compared to the value of Eys2=2x10 estimated from the prompt and high energy tail fluence. implies that the conversion efficiency into y-rays is ~1%. Which is at the lower end of the typical range 0.01—1 found for long GRBs and probably the same for short GRBs22).
We note that $E_{52}=0.35$ , compared to the value of $E_{\gamma, 52}=2\times10^{-3}$ estimated from the prompt and high energy tail fluence, implies that the conversion efficiency into $\gamma$ -rays is $\sim 1\%$, which is at the lower end of the typical range $0.01-1$ found for long GRBs and probably the same for short GRBs.
. The IC flux at the peak »/* is given by where we have used Eqs. (23 (25))-(26)).
The IC flux at the peak $\nu^{IC}_m$ is given by where we have used Eqs. \ref{seconda}) ), \ref{quarta}) \ref{quinta}) ).
For the same set of parameters. we have vfβίον>).C)«.[077 erg em™ s! (see Fig. 3).
For the same set of parameters, we have $\nu^{IC}_{m}f^{IC}({\nu^{IC}_m})\sim 10^{-7}$ erg $^{-2}$ $^{-1}$ (see Fig. \ref{ES}) ),
which is comparable with the LAT sensitivity for 10 s integration time.
which is comparable with the LAT sensitivity for $10$ s integration time.
We note that for a given value of v. Fy)". and z. the above equation ensures that Es» is sufficiently high to have the GeV tail detected by the Fermi/LAT.
We note that for a given value of $\nu_m$, $F^{syn}_{\rm 1 keV}$, and $z$, the above equation ensures that $E_{52}$ is sufficiently high to have the GeV tail detected by the Fermi/LAT.
At the same time. it is evident from Eq. (27)
At the same time, it is evident from Eq. \ref{settima}) )
that a higher value of Es» tends to shift the peak energy to highervalues. so that to keep it around ~| GeV. ncannot be too low.
that a higher value of $E_{52}$ tends to shift the peak energy to highervalues, so that to keep it around $\sim 1$ GeV, $n$cannot be too low.
Our value of 1=5 em is in the range that has been found to possibly characterize other short bursts 2).. and roughly at the higheredge of the 0.01—| em? range expected for theISM.
Our value of $n=5$ $^{-3}$ is in the range that has been found to possibly characterize other short bursts , and roughly at the higheredge of the $0.01-1$ $^{-3}$ range expected for theISM.
7T): failure to correct for variations in the interstellar dispersion 2): and "timing noise” intrinsic to the pulsar.
; failure to correct for variations in the interstellar dispersion ; and “timing noise” intrinsic to the pulsar.
Fiming noise is still not fully uncderstood. but usually refers to unexplained Iow-frequency eatures in the residuals ?).
Timing noise is still not fully understood, but usually refers to unexplained low-frequency features in the residuals .
. The main ellects of neglec"ling correlation in the timing residuals are: (1) the parameters of the timing model are not estimated as accurately as »ossible and may have systematic biases: and (2) the uncertainties on the best-Lit parameters are not correct.
The main effects of neglecting correlation in the timing residuals are: (1) the parameters of the timing model are not estimated as accurately as possible and may have systematic biases; and (2) the uncertainties on the best-fit parameters are not correct.
An extreme. example. of the problem can be. seen in the 20-cm timing residuals for the millisecond: pulsar J1939| 2134 assembled by Verbiest et al.
An extreme example of the problem can be seen in the 20-cm timing residuals for the millisecond pulsar $+$ 2134 assembled by Verbiest et al.
(2009)... The first eight. vears of these observations were taken at the Arecibo Observatory anc the remaining observations were taken at the Parkes Observatory.
The first eight years of these observations were taken at the Arecibo Observatory and the remaining observations were taken at the Parkes Observatory.
‘Phere is an. unknown phase discontinuity between the observations at. the two observatories.
There is an unknown phase discontinuity between the observations at the two observatories.
An initial estimate of this “jump” was mace by fitting the observations for only a vear on either. side of the jump with a model including the pulse frequency. p. its first derivative 7. and the jump.
An initial estimate of this “jump” was made by fitting the observations for only a year on either side of the jump with a model including the pulse frequency $\nu$, its first derivative $\dot{\nu}$, and the jump.
The fitting of v and £ removes a quadratic from the residuals. which in this case makes them almost white and allows for a reasonable initial estimate of the jump.
The fitting of $\nu$ and $\dot{\nu}$ removes a quadratic from the residuals, which in this case makes them almost white and allows for a reasonable initial estimate of the jump.
Phe residuals are cisplaved in Figure panel (a) after a weighted. least squares (LS) fit for & and ο using the initial jump estimate.
The residuals are displayed in Figure \ref{fg:badexample} panel (a) after a weighted least squares (WLS) fit for $\nu$ and $\dot{\nu}$ using the initial jump estimate.
Fitting v and £ over the longer data span leaves an obvious cubic term in the residuals.
Fitting $\nu$ and $\dot{\nu}$ over the longer data span leaves an obvious cubic term in the residuals.
In. panel (b) the residuals are shown after also fitting for the position of the pulsar and the jump using WLS over the entire data span.
In panel (b) the residuals are shown after also fitting for the position of the pulsar and the jump using WLS over the entire data span.
The jump fitting over the entire data span is obviously catastrophic.
The jump fitting over the entire data span is obviously catastrophic.
An olfset in. position introduces an annual sine wave into the post-fit. residuals.
An offset in position introduces an annual sine wave into the post-fit residuals.
Ideally the fitting process would minimize the power in the spectrum at lov ot. but the high. degree of correlation in the residuals introduces such a [arge error in the estimated position that it actually adds power at +.
Ideally the fitting process would minimize the power in the spectrum at 1 $^{-1}$, but the high degree of correlation in the residuals introduces such a large error in the estimated position that it actually adds power at 1 $^{-1}$.
This shows as a distinct annual ripple in panel (b).
This shows as a distinct annual ripple in panel (b).
Panel (c) shows the result of fitting the same parameters over the same data span as panel (b) using the new method that we describe in this paper.
Panel (c) shows the result of fitting the same parameters over the same data span as panel (b) using the new method that we describe in this paper.
The new method. which we refer to as the Cholesky method. obviously fits the position and jump much better than panel (b) but there is an obvious trend in panel (c).
The new method, which we refer to as the Cholesky method, obviously fits the position and jump much better than panel (b) but there is an obvious trend in panel (c).
This is because the estimate of 7 in panel (ο) is quite dillerent from the one in panel (a).
This is because the estimate of $\nu$ in panel (c) is quite different from the one in panel (a).
In panel (a) some of the red. noise is absorbed into the v estimate.
In panel (a) some of the red noise is absorbed into the $\nu$ estimate.
The error in the trend is much larger than one would expect. due to the red noise.
The error in the trend is much larger than one would expect, due to the red noise.
The elfect of changing v by Ἔσ ds illustrated in panels (a) and (c) bv the dashed lines.
The effect of changing $\nu$ by $\pm\sigma$ is illustrated in panels (a) and (c) by the dashed lines.
One can see that the trend in panel (c) is well within the uncertainty of the ν estimate.
One can see that the trend in panel (c) is well within the uncertainty of the $\nu$ estimate.
The estimation of 7 and © will be discussed in more detail in section 5.3.
The estimation of $\nu$ and $\dot{\nu}$ will be discussed in more detail in section 5.3.
Another example of this problem is a comparison of the parallax of PSR 4715 as estimated. from a timing analvsis(7).. with jab estimated from recent Very Long Baseline LnterferonxLric (VLDI) observations(?).
Another example of this problem is a comparison of the parallax of PSR $-$ 4715 as estimated from a timing analysis, with that estimated from recent Very Long Baseline Interferometric (VLBI) observations.
. The VLDI parallax is 6.3962:0.054 mimias.
The VLBI parallax is $\pm$ mas.
. The timing parallax estimated. using a WLS fit was 6.6540.07 mmas ancl we were able to cluplicate this using the same observations.
The timing parallax estimated using a WLS fit was $\pm$ mas and we were able to duplicate this using the same observations.
We obtain a value of 6.34£0.12 mmas using the Cholesky method.
We obtain a value of $\pm$ mas using the Cholesky method.
Clearly the Cholesky result is consistent. with the VLBI result and the WLS method is not.
Clearly the Cholesky result is consistent with the VLBI result and the WLS method is not.
It should be noted that Verbiest et al. (
It should be noted that Verbiest et al. (
2008) were doubtful of the formal error estimate for the WLS method and they re-estimated the error using a Monte Carlo simulation which increased the error from nunas to mmas.
2008) were doubtful of the formal error estimate for the WLS method and they re-estimated the error using a Monte Carlo simulation which increased the error from mas to mas.
As the actual difference is mmas this revised. error estimate may have been conservative.
As the actual difference is mas this revised error estimate may have been conservative.
It is quite common for pulsar observers to be suspicious of the error estimates obtained. using WLS fits and to modify them in various ad hoc wavs.
It is quite common for pulsar observers to be suspicious of the error estimates obtained using WLS fits and to modify them in various ad hoc ways.
We will show that using the Cholesky method eliminates this problem for all parameters of the timing model with time scales shorter than the observing span.
We will show that using the Cholesky method eliminates this problem for all parameters of the timing model with time scales shorter than the observing span.
v and ο always have time scales comparable with the observing span ancl will be discussed separately in section 5.3.
$\nu$ and $\dot{\nu}$ always have time scales comparable with the observing span and will be discussed separately in section 5.3.
Pulsar observers have often. attempted. to. improve parameter estimates by removing some portion of the Low-frequency timing noise. taking care not to remove the components that are needed to estimate the parameters of interest.
Pulsar observers have often attempted to improve parameter estimates by removing some portion of the low-frequency timing noise, taking care not to remove the components that are needed to estimate the parameters of interest.
“The low frequencies have been removed. by adding them to the timing mocdel. either as a high order polynomial or as a carefully chosen Fourier series (e.g... in.ΓΕΝΟΣ: Hobbs et al. 2004)).
The low frequencies have been removed by adding them to the timing model, either as a high order polynomial or as a carefully chosen Fourier series (e.g., in; Hobbs et al. ).
In either case the residuals are "attened or “whitened” and an adequate fit for position can be obtained. (throughout. this. paper timing residuals that are uncorrelatecl are termed: “white” and. residuals that exhibit a steep low-Lrequeney spectrum are termed. "red).
In either case the residuals are “flattened” or “whitened” and an adequate fit for position can be obtained (throughout this paper timing residuals that are uncorrelated are termed “white” and residuals that exhibit a steep low-frequency spectrum are termed “red”).
Neither method: produces a good fit to a phase jump because the effect of a phase jump is not localised in frequency in the residual spectrum.
Neither method produces a good fit to a phase jump because the effect of a phase jump is not localised in frequency in the residual spectrum.
In this paper we describe a method of optimizing the least-squares fit by finding a linear transformation. which whitens and normalizes the residuals.
In this paper we describe a method of optimizing the least-squares fit by finding a linear transformation which whitens and normalizes the residuals.
Fhis transformation is then applied. to. both the observations ancl the timing model.
This transformation is then applied to both the observations and the timing model.
“Phe parameters can then be found by fitting the transformed timing model to the transformed observations using ordinary least-squares.
The parameters can then be found by fitting the transformed timing model to the transformed observations using ordinary least-squares.
The transformation can be found. exactly i£ the covariance matrix of the residuals. is known. although it is not unique.
The transformation can be found exactly if the covariance matrix of the residuals is known, although it is not unique.
This process is equivalent to the so-called. "eeneralized least-squares” (CLS) solution. indeed it is how that solution was discovered.
This process is equivalent to the so-called “generalized least-squares” (GLS) solution, indeed it is how that solution was discovered.
Εις method is not new. but it has not been applied to pulsar. timing observations before.
This method is not new, but it has not been applied to pulsar timing observations before.
The LS solution provides the best linear unbiased. estimator of the parameters of the timing model and the best unbiased estimator of the covariance matrix of the estimated. parameters.
The GLS solution provides the best linear unbiased estimator of the parameters of the timing model and the best unbiased estimator of the covariance matrix of the estimated parameters.
Lo most cases. the covariance matrix of the residuals is not known ancl must be estimated. from the observations.
In most cases the covariance matrix of the residuals is not known and must be estimated from the observations.
We have developed a procedure for estimating the covariance matrix which works well for the pulsars we have tested. and we find that the parameter estimates are not undulv. sensitive to errors in estimating the covariance matrix.
We have developed a procedure for estimating the covariance matrix which works well for the pulsars we have tested and we find that the parameter estimates are not unduly sensitive to errors in estimating the covariance matrix.
Lt should be noted that the residuals formed. by subtracting the timing model with the best fit parameters from the observations will not be white as can be seen in Figure | panel (ο).
It should be noted that the residuals formed by subtracting the timing model with the best fit parameters from the observations will not be white as can be seen in Figure 1 panel (c).
Phe advantages of the Cholesky method over the polynomial or Fourier methods are: (1) it provides more accurate parameter estimates: 2j it provides a more accurate estimate of the covariance niatrix of the parameters: (3) it does not require any "adjustment? to fit different parameters of the timing model.
The advantages of the Cholesky method over the polynomial or Fourier methods are: (1) it provides more accurate parameter estimates; (2) it provides a more accurate estimate of the covariance matrix of the parameters; (3) it does not require any “adjustment” to fit different parameters of the timing model.
In section 2 we outline the theory of linear [east-squares fitting when the covariance matrix of the residuals is known.
In section 2 we outline the theory of linear least-squares fitting when the covariance matrix of the residuals is known.
When the covariance is not known one must usually attempt a spectrum analysis of the resicluals.
When the covariance is not known one must usually attempt a spectrum analysis of the residuals.
In section 3 we discuss the problem of spectrum analysis of steep rec randoni processes which are sampled. irregularly
In section 3 we discuss the problem of spectrum analysis of steep red random processes which are sampled irregularly
uucertaiuties).
uncertainties).
The source appears bluest (A—dy2.52 nae) at μπακ light. and. reciclest CH—dy 2.68 mag) at minimum light.
The source appears bluest $H-K\approx$ 2.52 mag) at maximum light and reddest $H-K\approx$ 2.68 mag) at minimum light.
The shape aud period of the IRSIGSW light curve suggests it is either an eclipsing binary star system or some variety of periodic variable star.
The shape and period of the IRS16SW light curve suggests it is either an eclipsing binary star system or some variety of periodic variable star.
Careful examination of the properties of the light curve. however. rules out most of the Familiar classes of explanations.
Careful examination of the properties of the light curve, however, rules out most of the familiar classes of explanations.
Below we review aud discuss these possibilities aud present evidence that IRSLOSW may represent a new class of regularly pulsating massive stars.
Below we review and discuss these possibilities and present evidence that IRS16SW may represent a new class of regularly pulsating massive stars.
Ott.Eckhart.&Genzel(1999) presented the possibility that IRSIOSW is a Cepheid variable.
\citet{ott99} presented the possibility that IRS16SW is a Cepheid variable.
They concluded that IRSLOSW was not likely a Cepheid ou the basis of the light curve shape aud apparent brightness of tlie source at Ix. They also uote that the spectrum of IRSI6SW from is not consistent with that expected from a Cepheid.
They concluded that IRS16SW was not likely a Cepheid on the basis of the light curve shape and apparent brightness of the source at K. They also note that the spectrum of IRS16SW from \citet{krabbe95} is not consistent with that expected from a Cepheid.
Our data shows that IRSI0SW Changes color by 70.18 mag CA—K ) over the course of its light curve.
Our data shows that IRS16SW changes color by $\sim$ 0.18 mag $H-K$ ) over the course of its light curve.
This is also tucousistent with the behaviour of Cepheids. which typically show <0.05 mae H-Ix. color chauge over their periods (seeWelchetal.198[).
This is also inconsistent with the behaviour of Cepheids, which typically show $<$ 0.05 mag H-K color change over their periods \citep[see][]{welch84}.