GPI is a complex instrument and the performance is changing quickly with improvements in data reduction and the instrument wavefront sensor control loops. The performance has been broken down into the following areas:
- Detector Characteristics
- Exposure Times
- Extended Objects
- Limiting magnitudes for instrument and internal wavefront sensors
- Observing Constraints
- Optical throughput
- Satellite Spots
- Spectroscopic Accuracy
- Wavelength accuracy
Our understanding of performance and calibration is a work in progress; these pages will continue to be updated as more on sky calibrations are taken and analyzed.
We find that after correcting for distortion, the average positional residual drops from 0.26 spaxels to 0.04 spaxels.
Astrometric measurements for the purpose of measuring plate scale have been performed in a number of different ways by members of the team. A general comment about astrometry on current data sets is that positions of objects seem to be a function of wavelength in reconstructed cubes. S. Thomas has made several plots of plate scale measurements as a function of wavelength channel. She used the functions find_sat_spots_all.pro and fourier_coreg.pro from the pipeline. Variations may be due to wavelength calibration errors or errors on the positions of the satellite spots, mainly due to low SNR.
As shown, the choice of wavelength for separation measurements can have a significant impact on the derived pixel scale. Currently we are using the approach of averaging all measurements from each wavelength channel. Such an approach could likely be improved by being more selective about which wavelengths to include, since a number of the blue wavelength channels are impacted by low throughput.
Using a different centroiding approach, Q. Konopacky has investigated the separation variation of objects that are independent of the satellite spots (so not relative to the guide star under the coronagraph). She finds for the OMEGA1 Ori B2 and B3 pair (see below) that the typical RMS variation across the full H band is 0.06 spaxels (roughly ~1 mas, see below), with a spread as high as 0.5 spaxels. This could be due to PSF variations, since for the positional measurements PSF fitting was used. A bright spot in the airy ring of θ1Ori B2 impacts the position of B3. An example of the variation seen with this analysis is shown in the figure below. The red line represents the average separation value calculated from all wavelength channels.
Additional analysis and testing of astrometry in spectral cubes is warranted with current and new calibration data. The figure below shows the average separation for θ1Ori B2 + B3 in frames taken on 13 November 2013. Some frames were removed from the analysis due to poor PSF quality.
A weighted average of these data points provided the final separation measurement given in the table below. The large number of frames taken helped to lower the final uncertainty.
R. De Rosa measured centroids using the routine MPFIT in IDL for θ1Ori A and θ1Ori B. He used both unocculted and occulted frames. He notices that positions of components in the cube do tend to shift with wavelength. Overall separation changes may also be related to low SNR on satellite spots. His separations as a function of wavelength are shown in the plots below. Colors denote individual cubes.
The primary targets used to measure the pixel scale for the GPI IFS are θ1Ori B and HR 8799. These targets were chosen because of their well-characterized separation measurements using other instruments.
θ1Ori B was analyzed extensively in Close et al. (2013) using MagAO. By combining their new measurements with those previously obtained over a 15 year time baseline, they were able to determine that the actual separations of components B1 and B4 and B2 and B3 (see figure) change only marginally each year - the primary orbital motion seems to be affect mostly the positional angle. As such, we were able to use the Close et al. (2013) predicted change in separation as a function of time to predict the separations of these objects when observed with GPI on 13 November 2013.
HR 8799 was observed with NIRC2 on Keck on 15 October 2013. Given the long period of the orbits for planets HR 8799c and d, their positions are expected to be the same to within our uncertainties in first light data from 17 November 2013.
Additional targets used for this analysis were θ1Ori A (data taken on 13 November 2013) and the K2+white dwarf pair HD 8049AB (data taken on 16 November 2013). However, neither of these objects have simultaneous measurements from a different instrument/telescope or good orbital estimates to give a sufficiently precise separation measurements. The relative separation of θ1Ori B1+B2 has also been measured by Close et al. (2013), but these components were shown to have a significant change in separation over time, thus decreasing the precision of their estimated separation on 13 November 2013.
Currently, the optical distortion has not been characterized with on sky data. It was characterized pre-shipment in the testing facility at Santa Cruz. Below is a summary of those results. The pre-ship distortion map is currently used in the data pipeline, if the Correct Distortion primitive is included in a given recipe.
On 10 December 2012, IFS data frames were taken with an illuminated square pinhole grid. This grid was rotated such that the pattern was sampled through 360 degrees in 11 frames and at various (slight) offsets. The position of each spot was measured in all frames in order to conduct a comparative analysis in which all other frames were transformed into the reference frames of a “baseline” using only rotation and offset. The difference between the position of each spot in the “baseline” frame and the transformed frames are averaged to produce positional residuals. These residuals are then averaged for each 20x20 spaxel region on the IFS. A map of the residual vectors is shown here (magnified by a factor of 30):
This map represents our zeroth-order understanding of the form of the distortion in the GPI IFS. As listed on the figure, the average vector length is 0.26 spaxels.
The results of these fits are used to correct the frames for distortion in the current DRP. We find that after correcting for distortion, the average positional residual drops from 0.26 spaxels to 0.04 spaxels.
Sky frames were taken from the 13th-17th of November, 10th December (2013) and were reduced using the following primitives:
- Subtract Dark Background
- Load Wavelength Calibration
- Assemble Spectral Datacube
- Interpolate Wavelength Axis
The sky flux was then calculated as the median value within each spectral channel, divided by the integration time per coadd (ITIME keyword).
Y_direct mode skies were obtained on the 13th and 17th. On the 13th the flux was ~0.25 ADU/s, and on the 17th the flux was at a significantly higher level of ~1.75 ADU/s. The only apparent difference between the skies were the exposures times, (~30s on the 13th and 9s on the 17th). Proximity to the Moon can be discounted to explain the high counts on the 17th, as it was more than 20 degrees away at the time. Based on the IR sky background, the sky flux at Y should be similar to or slightly higher than J, suggesting that the measurement on the 13th is incorrect. No Y_coron skies were obtained.
J and H bands
J and H counts in the H-direct mode were consistent over the the Nov run. Adding the lyot mask and apodizer but not occulter didn’t decrease the sky level in either filter. This is surprising and not yet understood given the throughput measurements below. No measurements in the H-coron mode were taken on the 13th, although measurements taken on the 14th in the H-coron mode show a significant drop in flux.
K1 and K2 bands
For the sky frames obtained in K1_coron and K2_coron obsmodes, the flux levels remained roughly the same at <1 ADU/s for K1 and <2.5 ADU/s for K2 on the three different dates. The flux level in the direct obsmode varied significantly between nights.
Average and standard deviation of the median sky count across the band for various obsmode configurations
|Filter||Mode||Date||Background Levels [counts]|
|Y||Y_direct||17th Nov 2013||1.80±0.01|
|J||J_direct||17th Nov 2013||1.62±0.18|
|H||H_coron||13th Nov 2013||0.62 ± 0.06|
|H||H_direct||13th, 15th, 17th Nov 2013||0.62 ± 0.14|
|K1||K1_coron||16th, 17th Nov, 10th Dec 2013||0.42 ± 0.11|
|K1||K1_direct||13th Nov 2013||0.84 ± 0.01|
|K1||K1_direct||15th Nov 2013||7.08 ± 0.39|
|K1||K1_direct||17th Nov 2013||3.02 ± 0.01|
|K2||K2_coron||16th, 17th Nov, 10th Dec 2013||1.20 ± 0.13|
|K2||K2_direct||13th Nov 2013||2.23 ± 0.05|
|K2||K2_direct||15th Nov 2013||8.41 ± 0.51|
|K2||K2_direct||17th Nov 2013||4.12 ± 0.17|
Achieved sensitivity is a function of many parameters, including contrast, inner working distance, brightness of the central star (if any), observing mode (dithered, sky-offset, sky-rotation), and, weather conditions. Here we present graphs on the strongest correlations:
Contrast as function of I-magnitude of the star
- The figure above shows typical GPIES contrasts for a 1-hour sequence (42 x 1 minute exposures) on targets as a function of magnitude
- The contrasts are most strongly correlated with target star magnitude and with atmospheric correlation timescale tau0.
- Contrast at 0.4 arcseconds is determined primarily by residual atmospheric speckles and correlates most strongly with seeing
- Contrast at 0.25 arcseconds is determined primarily by FPM centering and quasi-static errors
- Contrast at 0.8 arcsecods approaches photon-noise limits and hence is a strong function of star magnitude
- There is a significant benefit to having >20 degrees of field rotation to enhance ADI PyKLIP processing
Contrast relation between 1 minute raw image contrast and 1h processed sequence
Image above shows the Correlation between raw 60-second contrasts and final 40-image combined contrasts after PyKLIP processing (contrast expressed in astronomical magnitudes).
Contrast as a function of accumulated field rotation
The image above shows the final pyKlip contrast vs field rotation.
Europa, satellite of Jupiter was observed on November 18, 2013 from 08:42:16 (airmass=1.63) to 08:48:16 (airmass=1.64) under average seeing conditions (DIMM seeing~0.6-0.8) in K1 band, and direct mode (no coronagraph, no apodizer). Eleven data cubes with a true integration time of 8.73 s were processed by M. Perrin using the recipe listed below.
Recipy used to process Europa Data
- Load Wavelength Calibration
- Subtract Dark Background
- Update Spot Shifts for Flexure
- Destripe science image
- Interpolate bad pixels in 2D frame
- Assemble Spectral Datacube
- Interpolate Wavelength Axis
- Divide by Lenslet Flat Field
- Interpolate bad pixels in cube
- Rotate North Up
- Accumulate Images
- Combine 3D datacubes
Ephemeris calculations provided by JPL & IMCCE indicate that the satellite was bright (V~5.6) with an apparent diameter of 0.9508 arcsec and observed with a phase angle of 8.9 deg, leading to an illumination fraction of 99.4%. Europa disk is therefore covering ~10% of the detector in an area almost centered on the detector.
The last 6 images were taken with CCR at low power and visually seem to be of better quality. It is not possible to quantify the resolution on these observations and the gain with CCR at low power since Europa does not display unresolved features necessary to perform a direct measurement.
Europa was not dithered on the set of observations. We estimated the residual jittering to be 0.12 pixel in average by fitting the disk with an ellipsoid and measuring its center for each frame. Figure below shows the astrometric position of Europa (labeled from E0 to E10), plus the residual jittering measured on seven frames of HD1160, a A0 V=7.1 star observed in K1 direct imaging mode on the same night between 01:20 and 01:35 UT (airmass ~1.2, DIMM seeing~0.8 arcsec), label P0 to P5. Because the residual jittering on the “PSF” stars is slightly higher (0.31 pixel in average), we conclude that no residual jittering due to the extended angular size of Europa is apparent. It is likely that the angular resolution on these images of Europa is in fact very close to the angular resolution that we measured on the “PSF”, so 65 +/- 1 mas.
After data processing, the frames in the datacube still display some imperfections visible as 15-20 pixels of low and high intensity on Europa’s disk. Most of these bad pixels are identical on the 11 observations and are probably due to bad pixels on the detector. Additionally, some of the frames in the cube contains an oblique pattern a misalignment between the spectra position in the calibration and the actual position in the detector image. This misalignment results in a poor flux extraction when the cube is assembled.
Pixel Scale from Europa Observations.
We estimated the pixel scale on final assembled cube by fitting the observations of Europa by a sphere illuminated under the same geometry than Europa (Sub-Solar point and Sub-Earth point from the ephemeris) and degraded after convolving it by a gaussian PSF with a FWHM of 65 mas. We generated a grid of solutions varying the Minnaert coefficient which defines the center-to-limb profile from 1.1 to 1.5 (step of 0.02) and the pixel scale from 13 to 15 mas (step of 0.04). The best fit is obtained with a pixel scale estimated to 14.30 ± 0.31.
A 3-color picture (1.95, 1.99,2.18 um) included below shows variegation on Europa’s disk. A comparison with visible observations based on Galileo-Voyager spacecraft suggests an anti-correlation between the dark albedo features in visible of Europa which are bright at ~2 um.
To confirm that these spectroscopic differences are real, hence not instrument artifacts, we extracted the spectra of 3 areas which show significant variations on the complete data set of Europa. The eastern limb is bright at ~1.8um, the bright center area is bright at ~2um and an isolated patch north-west of Europa that could be a young crater. The flux of the extracted spectra is normalized after dividing it by the flux of HD1160 an A0 star without including an airmass correction.
The extracted spectra of these 3 different area on Europa are similar among all the spectro-images within 3%, confirming the stability of the instrument. The spectra of these three regions display identical absorption bands at 1.95, 2.00, 2.05 um which could be of atmospheric origin. It is however clear that spectroscopic variabilities (e.g at lambda>2.1 um) are detectable on Europa disk.
Neptune (JPL ephemeris V=7.9, angular diameter = 2.26 arcsec) was observed on Dec 11 2013 from 00:17 to 00:31. Seven frames with an exposure time of 59.65s were recorded in H direct with an airmass of 1.3 and a DIMM seeing of 0.66 arcsec. This data is still being processed.
GPI limiting magnitudes are determined by several components, the AO WFS (I-band), the LOWFS (H-band), and the IFS (selected filter). In addition the observing conditions add another layer of limits. Thus the brightest of the science object is limited in I band from the AOWFS, in H-band from the LOWFS (not a constraint in DIRECT mode as then no coronographic mask is used and no LOWFS is possible). It should be noted that it is NOT possible to observe without the AOWFS and Coronographic modes are NOT possible to observe without the LOWFS.
|Maximum brightness [mag]||Minimum brightness [mag]|
*Only valid in Coronographic mode, in the Direct and the NRM modes there is no LOWFS and thus no constraint imposed by the LOWFS.
The selected observing mode is strongly affecting the brightness of the target that can be observed with the IFS without saturating in the selected science wavelength by the IFS. The principal four modes (each that can be done in all the filters Y, J, H, K1 and K2) are Coronographic Spectroscopy (the ''-coro'' modes), Coronographic polarization (the ''-coron-pol'' modes), Direct (the ''-direct'') observations with either spectroscopy or polarization, and the NRM (the ''-NRM'') mode observations in either spectroscopy or polarization. Note that this limit is applicable to the relevant science wavelength.
|Observing Mode||Maximum brightness [mag]|
*Only valid in Coronographic mode, in the Direct and NRM modes there is no LOWFS and thus no constraint imposed by the LOWFS.
GPI is a XAO instrument and the stated performance is reached under IQ70 and CC50 conditions. It is possible to use in IQ85, CC70 and CC80 conditions but it should be noted that the performance is highly variable for these conditions and NO guarantee on performance can be given. Observations have been done under both IQ85 and CC80, and taking the limits into account the instrument is safe to use, the main risk is low light levels causing the various WFS systems to guide on noise which could damage the instrument. Note that ANY observations where the effective limit of the AOWFS is passed is NOT allowed.
The given magnitude limits for the AOWFS and the LOWFS (if used) should be modified in the following way for worse than IQ70 CC50 conditions:
|Observing conditions||Decrease of faintness limit [in magnitudes]**|
**Note that the decrease in magnitudes is ONLY applied to the faint end, it does not mean that brighter than normal targets can be observed. This is to make operations safe and avoiding locking on noise in the control loops.
To reach the desired instrument performance GPI is constrained both in the suitable weather conditions and the elevation of the target being observed.
The nominal weather conditions for which the performance data has been specified is IQ70 CC50, please look at Contrast, Limiting magnitudes for instrument and internal wavefront sensors, Optical throughput and Polarimetry for more details on how the performance is affected by observing at other conditions i.e. IQ85, CC70 and CC80.
Note that since performance is not guaranteed for looser constraints then loose constraints are not suitable for planet detections, but rather for extended objects like moons and also for binary searches.
For the AOWFS and the LOWFS to work then the elevation is limited to Zenith distances ≤40°, with a possibility to reach Zenith distances of 50° with a lower performance. The wavefront sensors are designed to reach peak performance on point sources and the performance decreases by observing extended objects, more details can be found at Extended Objects.
It should be noted that a target fainter than the I-band magnitude of 9.0 and Zenith distance >40° is not feasible, note that the 9.0 limit is modified by the weather constraints.
Photometric standards have been used to measure the throughput of the atmosphere+telescope+instrument system based on comparison of GPI datacubes with IRTF Spex spectra. Observations of HD 51956 (=HIP 33657) from the first run in November 2013 were reduced (dark subtraction, bad pixel correction, destriping, basic datacube extraction). The total intensity in each spectral channel of the cube is calculated and compared to the corresponding spectral channel of the low resolution Spex spectrum, using same physical units. The gain of the detector is taken taken into account. The following figure presents throughputs obtained with and without the coronagraphic Lyot stop and apodizers for Y,J,H,K1 and K2-bands.
The dashed lines show the throughput for the coronagraph modes (including apodizers and Lyots) while the solid lines above show the throughput for direct imaging with no apodizer or Lyot stops in the beam.
Independent calculations for verification has been performed on additional targets and the results are very similar.
Line styles the same as the previous plot. The colours indicate the different targets, blue - HD 51956, red - HD 20619, green - HD 8049b. HD 51956 was observed in both unblocked and direct modes, whilst HD 20619 and HD 8049b were only observed in one of the two configurations each. The flux for HD 51956 and HD 20619 was calculated by summing the counts in each of the spectral channels (with the result not significantly changing when a large aperture was used instead). For HD 8049b, an aperture of 3x FWHM was used, with the FWHM measured in each slice. The measurements where corrected by dividing by a Cerro Pachon airmass=1.0, water vapour column=2.3mm atmosphere.
This page details the sensitivity of GPI in differential-polarimetry mode, with focus on sensitivity to extended objects. As opposed to detection of planets, where there is a simple figure of merit (brightness ratio of the point planet to the point star at a given wavelength), disks are extended objects and sensitivity has to be expressed in a way that recognizes that.
Two disks of the same total brightness, which looked identical to IRAS, may be very different from GPI's point of view if in one case the dust is diffuse and spread over a large area and in another there is a sharp HR4796-like ring. Thus this has been modeled in units of either total intensity (assuming a particular geometry) or surface brightness.
GPI's sensitivity to disk-scattered light relies on differential polarimetry to suppress PSF speckles. We show here predicted polarimetric surface brightness sensitivity per spatial element, assuming polarimetry is the only speckle suppression used (no ADI or other PSF subtraction) and given a residual polarized noise floor of 1% of total intensity (see below).
ADI polarimetry or detection of dust disks is very complicated to model and case-specific, it's clearly very dependent on disk geometry, a face-on circular disk would be subtracted by the ADI just like the seeing halo is, in general it's only useful for edge on systems, but for those it does work quite well.
Plotted in solid lines are PSF radial profiles based on current GPI AO simulations. In dotted lines are the derived 5-σ sensitivity levels per resolution element for 100% linearly polarized light. The left axis shows surface brightness in units of fractional flux per spatial element (0.014" GPI lenslet) while the right axis is scaled to show fractional flux per square arcsecond.
Plot of GPI polarization sensitivity in the J band, with an estimated 1h observation.
Plot of GPI polarization sensitivity in the H band, with an estimated 1h observation.
To assess the detectability of a given circumstellar disk requires an assumed polarization fraction derived from dust models, in much the same way that assessing the detectability of a planet from a contrast plot requires an assumed luminosity for that planet's mass derived from evolutionary models. Surface brightness levels for a few known disks are also indicated for comparison at right. For HD 4796 the polarized surface brightness is taken from Hinkley et al. 2009, for HD 15115 from Kalas et al. 2007, and for AU Mic from Fitzgerald et al. 2007.
The achieved speckle-suppression performance depends both on the science camera calibration and also on our ability to calibrate instrumental polarization for the combined GPI+Gemini South system. We conservatively estimate 1% residual post-suppression speckles based on our modeling of instrumental systematics. This level is comparable to polarimetric performance achieved with prior AO polarimeters at Lick (Perrin et al. 2008) and AEOS (Hinkley et al. 2009).
In the polarization mode you don't need to observe PSF reference stars, because the two orthogonal polarizations essentially serve as each other's PSFs, though one can contemplate using some other form of PSF subtraction if you want to obtain a total intensity image as a followup characterization observation. But polarimetry should be most sensitive for detection.
An in depth discussion of the satellite spots can be found in this document Satellite Spot Ratio Recalibration Report by R. deRosa et al. 2019
Coronagraphic observations pose a unique problem for differential spectrophotometry and astrometry of an exoplanet relative to its primary, which is occulted. To overcome this obstacle, GPI includes a square grid in a pupil plane which acts as a two-dimensional amplitude grating. The grid is superimposed on the the pre-occulter pupil apodizer. Diffraction of starlight from this grating injects first-order diffraction spots into the field of view for a given wavelength.
In each wavelength channel in spectral mode, this creates reference spots that we term satellite spots. In this mode the approximate magnitude difference with the star is 9.3±0.5 magnitudes.
In broadband polarimetry mode, the satellite spots become streaks extending radially outward from the location of the star. In this mode the approximate magnitude difference with the star is 9.5±0.5 magnitudes.
These satellite spots preserve the information needed to reconstruct the spectrum and location of the occulted star: the diffraction pattern is centered on the true stellar position and is imprinted with an attenuated version of the stellar spectrum.
The usage of the satellite spots is in general transparent to the user and requires in most cases no adjustment or affect the feasibility of the observations, with the exception of the case of an extremely bright nebula/background emission surrounding the star. Please contact the GPI instrument scientist in this case before submitting a proposal for a feasibility check.
A quick summary of the ratios is shown in the image below.
The spectroscopic accuracy has been evaluated using the white dwarf orbiting HD 8049 discovered by Zurlo et al (2013), believed to have a temperature of 18800+/-2000 K. Observations were performed during the second commissioning run on December 12, 2013. The recipes, calibration files, and reduced cubes used to create the plots in this section are available as part of the first light data release. For this data, zero-point offsets to the wavelength solutions were determined using an Argon arclamp image taken immediately after the exposures. The data were flux calibrated using the spectra extracted from the satellite spots using the 2mass magnitudes combined with a Pickles model of a K2V star. The error bars on the data points in the plots below are determined by normalizing the spectra of the individual satellite spots to a common integrated intensity, then looking at the standard deviation in each wavelength channel. This gives an approximation of the uncertainty of the extracted spectrum but is most certainly an underestimate. Furthermore, the effects of systematic error are not properly included. The effects of systematic error are evident in the extracted spectra of the white dwarf. The following plots compare the extracted spectra in the H, K1 and K2 bands to a 18800 K black body.
The H-band spectrum was extracted out a single 60s image. Overall, the spectral slope is well reproduced, however, systematic error can be seen in both the long and short wavelength extremes (~0.06um) of the band. The large deviation in the first couple data points is probably due to a small error in the wavelength calibration. Because the throughput of the filter is so low in this region, a large correction factor is applied when performing the flux calibration, therefore any small error is greatly amplified. The single blue data point indicates the absolute flux level measured using H-band photometry by Zurlo et al (2013). The blue error bar (upper right in the above figure) indicates the standard deviations in the integrated photometric flux of each satellite spot. It is believed, but not confirmed, that the ratio between mean satellite intensity and the intensity of the central star is constant. However, if this is not the case then this error bar indicates the uncertainty on the flux normalization. Another possible scenario leading to our increased flux level may be a result of an incorrect calibration between the mean peak intensity of the occulted star and the satellite spots. Research into this is on-going.
The K1 spectrum of HD 8049b (above), whose wavelength calibration was performed using the telluric absorption lines at ~2.01 and ~2.06 um also shows evidence of systematic error at shorter wavelengths. However, previously obtained spectra of this object show evidence that it is not a perfect blackbody, but at the moment we do not possess the measured spectrum and cannot do a direct comparison. The extreme deviations are a result of spectral crosstalk at the extreme ends of the bands, therefore this region of the spectrum should be disregarded (at wavelengths shorter than ~1.93um). At this time, the resulting systematic error between 1.93-2.02 remains unclear (assuming it is indeed systematic error). One possible scenario is correction to compensate for flux that is not included in the 3-pixel box extraction algorithm. Because no flat-field image was taken at the position of the science target, no correction could be applied. More thorough extraction algorithms are currently under development that will make such a correction unnecessary. If the observer requires a precise replication of the broad spectral shape, it is recommended to take a flat-field exposure before and/or after the science target. Although no previous K-band photometry exists for HD 8049b, the fact that the flux appears higher in K1 (at ~1.90 um) then H (at 1.8 um) suggests that the current ratio between the satellite and occulted star flux is incorrect by a factor of ~2.
The K2 data had particularly low signal-to-noise in the satellite spot images below 2.15 um and above 2.32 um therefore these regions of the spectrum are subject to large error. Overall, the spectrum does match the slope of the blackbody to within uncertainty. However, we do emphasize that because no systematics are apparent in this plot, they most certainly exist and may vary as a function of field position. Furthermore, the percent error for the K2 data are larger than the other bands and may be larger than the level of systematics.
The following two plots show the simulated GPI Strehl performance as a function of the AO Guide star magnitude, assuming median seeing (IQ50). However, it is unlikely that the performance would improve in IQ20 conditions, because GPI isn't designed to run slowly to take advantage of slow seeing. Further, performance at I=10 would be very sensitive to the read noise on the wavefront sensor. The latter is known to be a function of environment and we do not yet know what it will be on the telescope.
Raw FITS files from the GPI simulations
Use right clock to select the download of the file.
This page is focussed on the wavelength calibration accuracy and the IFS flexure compensation.
Wavelength Calibration Overview
Similar to other near-IR spectrographs, GPI derives its wavelength calibration based on the precisely known emission lines of an arc lamp, specifically using the Xenon and Argon lamps available in GCAL. The GPI data pipeline can use these data to provide wavelength calibrations with current accuracy of better than 0.5% for all bands. As discussed below, flexure within the IFS due to varying orientation with respect to gravity causes small shifts of the spectra on the detector, which must be properly compensated for in order to assemble accurately calibrated datacubes.
Creating Wavelength Solutions
Due to variations across the field of view, a unique wavelength solution is determined for each lenslet for each band (Y, J, H, K1 and K2). The wavelength solution is characterized by the starting positions (x0, y0) of a lenslet at the reference wavelength (λ0), the spectral dispersion (ω) in microns per pixel, and the spectral tilt (θ in radians) with respect to the Y axis. For a given lenslet, the pixel positions as a function of wavelength (x(λ),y(λ) are given as follows:
Wavelength Solution files are written with the extension “_wavecal.fits” in the form of a 3D datacube with 5 slices of size [nlens x nlens] giving information about each lenslet as described below. Note: Due to the field rotation, only a central rotated-square portion of these arrays will have valid values. The outer regions are flagged with the NaN value.
- Slice 1: Y-positions (y0) of spectra (Y=spectral direction) at [λ0]
- Slice 2: X-positions (x0) of spectra at [λ0]
- Slice 3: λ0 [um]
- Slice 4: Dispersion w [um/pixel]
- Slice 5: Tilts of spectra, θ [radians]
Wavelength calibration files can be created from Xenon or Argon arc lamp data (obtained from GCAL) using the “2D Wavelength Solution” primitive. This primitive fits the image of each lenslet individually by simulating an arc lamp lenslet spectrum 2D image at the spectral resolution of the GPI IFS for a given band. Each spectral line is represented by a gaussian PSF at a location predicted by the x0, y0, w and θ values given in a prior wavelength calibration file. The model and detector images are compared using a nonlinear least squares optimization and the values for x0, y0, w and θ are updated for each lenslet.
The figure above shows the Xenon and Argon spectra for Y, J and H bands. Vertical lines show the individual emission lines and the plotted curve shows the spectral response at GPI’s spectral resolution, at the the 37 wavelength channels in each band (in interpolated output cubes which are oversampled relative to Nyquist). The xenon spectrum provides more cleanly separated peaks and is easier to fit. The argon spectrum has many more lines which are strongly blended at GPI’s low resolution; however the much greater brightness of the GCAL Ar lamp (3-10x brighter than Xe) means it is preferred for practical reasons in many cases. The GPI instrument team has worked to ensure that good wavelength calibrations can be derived from either lamp; comparisons of results from the Xe and Ar and further optimizations are ongoing.
Note: The “Quick Wavelength Solution” primitive can also be used to fit a subset of the lenslets and update the x0 and y0 positions only. This is significantly faster than the “2D Wavelength Solution” primitive and is useful to correct for offsets due to flexure (discussed below).
Gravitationally induced flexure within the IFS causes shifts of the lenslet spectra ranging from a small fraction of a pixel to several pixels in magnitude. The figure below shows flexure in the x and y direction as a function of elevation for several epochs of data. During a single epoch, the flexure behavior is relatively repeatable and can be accounted for using the “Update Spot Shifts for Flexure” primitive. Between GPI observing runs when GPI is moved to non-standard orientations, larger shifts of a few pixels can occur.
It is recommended that a snapshot arc lamp image be taken at the elevation of any interesting science targets for the best flexure correction. We recommend two 60 second Argon lamp images in H band. These data can then be reduced using the “Quick Wavelength Solution” primitive in the GPI pipeline. Such data are available for some but not all observations in the GPI early release data.
If a given observation does not have a contemporary arc lamp calibration image, the “Update Spot Shifts for Flexure” primitive will consult a look-up table that will return an estimation of the shifts due to flexure given the value of the ELEVATIO keyword in the image header, relative to whatever the closest-in-time available wavelength calibration in. This primitive also allows for user defined x and y pixel shifts; see below for how to use these.
Symptoms of Imperfect Flexure Compensation
Having an incorrect estimation of the shifts due to flexure will affect both the accuracy of the wavelength solution and the relative brightness of the final science data products. Each wavelength bin is extracted using a three pixel box in the x direction. If the x flexure shift values are incorrect, flux from the source will be lost. By eye, this will result in a strong moire (or checkerboard) pattern. An example of this is in the figure below.
The spectra are dispersed in the y-direction on the detector. Consequently, incorrectly estimating the y-direction flexure shifts will cause the wavelength solution to be off by a value in microns equal to the shift in pixels times the spectral dispersion. If the flux dramatically drops off near one end of the channels in the final datacube, this could indicate that the flexure correction is incorrect, however, this depends strongly on the spectrum of your source.
Adjusting Wavelength Solutions Shifts Manually
The quality of a given wavelength solution for a given science observation can be checked using gpitv’s Plot Wavelength Solution function to overplot the wavelength solution on the 2D detector image. Misalignments of the spectra with their wavelength solution by more than a few tenths of a pixel can be seen by eye. The Plot Wavelength Solution tool allows interactive adjustment of the X and Y shifts. This can be a useful manual approach to deriving an approximately-correct flexure-compensated wavelength solution for a set of science data. Once you have found a set of X and Y shifts that align the spectral traces with the spectra themselves, these can be entered as parameters to the “Update Spot Shifts for Flexure” primitive.
The GPI team is working on updated algorithms that will automatically measure the shifts for each individual science frame. These will be available in a future update of the GPI data pipeline.
Frequently Asked Questions (FAQ) for observing with the Non Redundant Mask in GPI.
Q: Is the NRM commissioned and available?
A: The mode has been partially commissioned. It has been determined that the mode works operationally but there are a few open issues, listed below. This means that the mode is only offered in shared risk mode.
- There is no pipeline reduction of the NRM mode. Observations will only deliver the raw data and the extracted data cubes. Any post-processing is up to the user.
- Due to commisioning being affected by weather, we have no clear performance data, and it was found that in 2015 that the internal vibrations smeared significantly the fringes. Since then we have installed the active dampers on the GPI CCR's, but we have not had the opportunity to use the mode after the active damper install, due to issue with the PPM mask. See "November 12th, 2015" entry on the Status page. We will update the web pages with performance data as soon as we have obtained more observations.
Q: What is Non Redundant Masking?
A: The GPI Non Redundant Mask is a specially designed mask with several apertures (10 hole mask giving 45 baselines, or 36 independent closure-phases) placed in one of the apodizer wheel positions, allowing the use of Aperture Mask Interferometry. Each of the pair apertures is thus forming an optical interferometer. The design of the mask is such that it gives the maximum number of non redundant baseline pairs that each forms fringes with unique spatial frequency in the image plane.
Q: What is the expected performance?
A: Note that all the performance and constraints are based extrapolations of performances achieved at other telescopes, with less advanced AO systems. Currently at the design team for the NRM mask on GPI are expected to soon achieve 5-sigma K-band contrasts of 7.5 magnitudes (1 hour per target). It is estimated that by correcting the dominant systematics this should improve to at least 10 magnitudes with GPI (1 sigma contrast of 2 * 10-5), and a magnitude limit identical to any other mode for the science object, see the wave front sensor limits page, for AOWFS I-band limits.
Q: What is the Seeing constraints when using the NRM? A: The NRM will be used with the GPI AO working, but is less demanding on the Strehl and it is expected that up to IQ70 seeing conditions can be used for K and H band imaging, in J-band it is expected that median seeing (IQ50) is needed to reach the desired performance.
Q: What are the overheads when using the NRM?
A: The method requires larger overheads in the observing as each science observation of one hour must be bracketted by one hour calibration observations, thus a complete sequence is expected to last three hours. Ideally this is done as a Calibrator-Science-Calibrator cycle repeated over the three hours.
Q: Will the NRM have the Cassegrain in fixed or follow mode?
A: Observations will be using the fixed Cassegrain method so that the sky will rotate during the observations.
Q: What is the expected Field of View?
A: With a single observation (no sky rotation), the field of view of NRM is approximately matched to the coronograph size, i.e. a 3 lambda/D radius. i.e. there is a clear boundary between where NRM and standard GPI operate. Beyond 3 lambda/D, the standard GPI observing configuration should do better - the requirement for performance is at 4 lambda/D and although ADI/SDI will barely work at 3 lambda/D, more than 10 magnitudes of contrast should be achievable at that separation. Companions beyond 3 lambda/D will alias back somewhat, but with a little sky rotation (e.g. multiple observations) the masking field of view is increased to ~6 lambda/D.
Q: With the SAM/NRM being inserted into the instrument, will it be more tuned for point source detections or will it also be able to detect extended emission (disks/rings)? i.e what structure sizes will it be most sensitive to?
A: The method is most sensitive to 0.5 to 3 lambda/D structures. Systematics are always higher for visibility amplitudes than closure-phases, so point-symmetric structures (e.g. a disk/ring) are much harder to detect.
Q: What is the product after processing the raw data taken with the IFS?
A: The raw data file is processed to create OIFITS (Optical Interferometer FITS files) and input files for the BSMEM package. There are plans to produce radio-interferometer FITS (UVFITS), as well as a map and a map of binary solutions to model fitting.
Q: What Software can be used to model and/or reconstruct the image?
A: The processed files can be processed by MACIM/BSMEM/MIRA and the model fitting is usually done with custom code f.ex mpfit IDL library.