HESSI Spectroscopy -- First Steps

This is meant as a beginner's guide to HESSI spectroscopy. It will take you through demonstrations of two ways to analyze solar spectra: producing "semi-calibrated" spectra and removing the instrument response more thoroughly with the SPectral EXecutive (SPEX) package.

First, let's produce a semi-calibrated spectrum using only the Graphical User Interface (GUI) to the HESSI software, and using simulated data. Enter solarsoft and type

IDL>hessi

You should get a window that looks like this:



We begin by choosing a flare, which is also the first step for imaging, of course. On the command bar at top, pull down "File" and choose

Select Observation Time Interval
The widget you will get will allow you to select data by raw time, by looking at HESSI's flare list and picking a flare by number, or by looking at the Observing Summary data and picking a time interval from the overall lightcurve. Let's do it the easiest way for now, and pick a flare from the flare list. Click the button that says
Flare Selection...
On the new widget (which gives you choices as to how to sort the flares in the list) let's accept the default and hit
Select Flares
and on the new widget, which shows the list of flares, let's click on flare #4, and then hit
Accept
or
Accept and Close
on each of the last three widgets so that they disappear.

---

Now let's look at spectra. Back on the main widget (pictured above) use the pull-down menu on the top to select:

File ------- Retrieve/Process data ---------- Spectrum...
and you should see the spectroscopy widget:



This lets us make a lot of choices in how to take our spectra. Let's exercise them a bit. We'll leave the top panel (time interval) alone, since we just chose that with the other window (which is more flexible). The second panel covers energy binning. Generally you want to pick energy bins fine enough to distinguish between the different emission models you want to study, but not so fine that the statistical noise is overwhelming. There are several good default choices, or you can construct your own binning by pressing

Define Bins Manually.....
Let's not do that now. Instead, let's look at the defaults by pressing
Show Binning Codes
This output will appear in text back in your original IDL window, not in a widget. Be sure to let it finish scrolling and get your IDL> prompt again before going on with the GUI. We see that our default binning (Binning Code 7; see figure above) is a bit coarse for detailed continuum spectral work (such as separating out thermal, superhot thermal and power-law components, for instance). Let's pick Binning Code 6 instead, which gives us 1-keV bins up to 100 keV (do this with the pull-down menu next to Binning Code on the spectroscopy widget).

The next panel lets us select times. We can create not just a single spectrum but a whole array of them, and when we do full spectroscopy with SPEX (below) we will do so. But for now, let's choose the pulldown menu next to Time Bin Width(s) and set it to 240.000 seconds, so that we have only one time bin.

The next panel lets us select detector segments: front and rear for any or all of the 9 detectors. The document Behind, Beneath and Before HESSI Spectroscopy or "BBB" discusses the detector segments in detail. Basically, the front (F) segments detect most of the hard x-rays up to 100 keV and the rear (R) segments detect most of the gamma-rays above 100 keV. The simulated data have no rear segment events, so we won't bother with those. Instead, let's do something which you'll probably want to do most of the time in real life: remove from consideration the front segment of the detector under Grid 7 (we refer to this detector as "G7" for short). G7 has much poorer energy resolution in its front segment than any of the other detectors (about 3 keV instead of 1 keV FWHM at hard x-ray energies). So hit the button marked

Change...
and on the widget that appears, hit the box next to 7F to unselect this segment, then hit
Accept
to leave the new widget.

Let's look at the spectrum we're getting by pressing the button on the bottom labeled

Plot Spectrum
It should look like this (there may be subtle differences as the calibration gets refined with time):



This is a count spectrum, meaning it's the spectrum of energies that would be reported by the detector. First, notice that there are no events below 5 keV or above 300 keV. These are artificial cutoffs that are part of the simulation. Real data would have at least a few background counts, if nothing else, in these ranges. Now notice that the spectrum is rising from about 5-15 keV. This is not a characteristic of the Sun; it's due to absorption in passive material in front of the detectors (thermal blankets, beryllium detector windows, and most importantly the aluminum attenuators which are often in front of the detectors -- see the BBB document). From here to about 100 keV the spectrum falls roughly as the true solar spectrum does. Above about 100 keV the spectrum has a shoulder. This is not from the flare, but from simulated background. Usually the background spectrum will be very hard, harder than anything solar (and in real life it would continue on past 300 keV).

The rest of this document will be devoted to converting this count spectrum to a photon spectrum, i.e. what's incident on the spacecraft from the Sun. The simplest thing to do is to divide this spectrum by the efficiency of the system as a function of energy. This will remove the effects of absorption from the materials in front. To do this, simply hit the check box next to "Use Semi-calibrated Data" and hit

Plot Spectrum
again. You should now see something like this:



The good news is that the instrumental absorption which caused the spectrum to roll over below 15 keV is gone. The bad news is that the background gets exaggerted: instead of just a slight hump, the spectrum now actually seems to RISE above 100 keV. This is just the background again, but it is amplified (artificially) when we convert to a photon spectrum. If we did a detailed spectral fit on this spectrum we'd probably find that hardening due to this background effect is probably contaminating the spectrum all the way down to 50 keV or so. To do the spectral analysis right, we need to subtract the background BEFORE we remove the instrument response. We will do that next.

"Semicalibrated" spectra can be used with caution for science when the following conditions hold: 1) only the front segments are used, 2) the flare spectrum is at least ten times the background spectrum for all the energies being examined, and 3) only energies below about 75 keV (50 keV to be conservative) are being examined. Above 75 keV, the redistribution of photons to lower energies by Compton scattering becomes more and more significant.

---

Now let's do a real spectral analysis using SPEX. First we'll make some files for SPEX to use. Let's go back and click again on the box next to "Use Semi-calibrated Data" so that the check disappears. Then go back to the time panel and pick a Time Bin Width of 5.000 sec. Now hit

Plot Time History
to see a lightcurve of flux vs. time:



Now we can save a file for SPEX to use. It will contain a time series of spectra, one for each of 52 time bins, with 199 energy bins in each spectrum. We will also ask the GUI to create and save a second file, containing the response of the instrument. Click on the

Write output file...
button. "Create SRM file" is checked; we need to leave it that way so that we will create a Spectral Response Matrix for SPEX to use. Go to the pulldown menu that currently says "2 -- Diagonal terms only" and switch it to "0 -- Full calculation of diagonal, off-diagonal terms".

Each line of the SRM gives the total response of the instrument to photons of a particular energy. The diagonal terms of this matrix represent the total system efficiency for recognizing a photon energy directly. Off-axis elements represent the probability that photons are redistributed to another (generally lower) energy. For hard x-rays, more photons are recorded with their true energy than with a lower energy, but above 100 keV this is reversed. This is why the "semicalibrated" procedure, which simply takes the diagonal elements of the response and divides the count spectrum by them to get a photon spectrum, is inaccurate above 50-100 keV or so, even without considering the issue of background.

Now go ahead and click

Write FITS file
It will offer you a file name for the spectrum file, then another for the SRM files. You may accept these, or substitute something shorter. I will accept what is offered: hsi_spectrum_20000902_035004.fits and hsi_srm_20000902_035004.fits (if something else is offered, be sure to use that when file names are asked for below!). When the box that says "Working..." disappears, you may press
Close
on the "Write Spectrum FITS or GENX file" window.

---

Now let's start SPEX, which is its own command environment within IDL/Solarsoft, by typing

IDL>spex_proc
SPEX will put up two blank windows that we'll fill later. First, we'll tell SPEX that we're using HESSI front-segment data:
SPEX> data,hessi,front
Now we'll tell it to read in the spectral file ("_1file") and response matrix file ("dfile") that we just created:
SPEX> _1file, hsi_spectrum_20000901_035004.fits
SPEX> dfile, hsi_srm_20000901_035004.fits
Note that filenames do not require quotes. To see the data that have been read in, enter
SPEX> graph
And you should see the time history, but this time with several colors representing several energy bands (which are listed on the bottom of the plot).



Now let's create the background spectrum by telling SPEX to accumulate it during intervals before and after the flare when there is nothing going on. This won't always be possible, but when such intervals exist within 10 minutes or so of the flare it's usually the best way to get an accurate background measurement. SPEX can fit a polynomial background model of order N for each energy bin given at least N background time intervals. For this case a constant background (order zero) will do. To get it, we type:

SPEX> back_order,0
SPEX> background
On the pop-up window, hit
Continue by Selecting from Plot Window
and then move the mouse over the window with the active display ("Time History Window"). Push (any) mouse button four times to define each side of two energy ranges:



then move the cursor back to the pop-up window and hit

DONE.
The window should now look something like this:



Now select one or more data intervals for which to analyze the spectra. Each of these intervals should of course be exclusive of your background periods. Type on the command line:

SPEX> select_interval
and on the window that pops up hit
Graphic Input
, then on the next window hit
AS BOUNDARIES FOR CONTIGUOUS INTERVALS
and then on the next pop-up window hit
Continue by Selecting from Plot Window
as was done for the background. Because we selected "contiguous intervals", each mouse click (except the first and last) represents both the beginning and end of an interval. I will do just 3 clicks to make two bins, then move the mouse back out of the window and hit
DONE
on the re-appearing pop-up. Now I have this:



Now SPEX has a background model, an instrument response, and a set of two times in which to accumulate spectra. All that's left to do is define a model and fit to it. We'll do the simplest model, a power law:

SPEX> f_model,f_pow
The full list of available models can be seen by looking at the procedure names in the Solarsoft distribution under packages/xray. To see the spectra in count space before the fitting begins, enter
SPEX> count
and the neglected window comes to life. You will see curves of several colors. The yellow curves are the flare spectra and the orange curves are their backgrounds [MORE THAN ONE?]. The red line represents the input photon spectrum, and the blue lines are the predicted response of the detectors to that spectrum. When the fit is complete, the blue curves should match the yellow, more or less. [BLUE HAS BKG ADDED?] But for now, we have arbitrary values for the power-law parameters. To remedy that, type
SPEX> fit
photon ****Problem:on the count plot (labeled counts/s/cm2/keV), where the bin width changes there is a jump in the orange curve equal to the change in bin widths, and a jump in the blue curve equal to about the sqrt of this, therefore they don't jump together. The yellow curve goes more or less with the blue.