4.1.37. Spln: spline continuum model¶
Sometimes the continuum of an X-ray source may be too complex to model with known physical components. A situation like that may be found in AGN continua, which are a complex superposition of hard power law, soft continuum excess, relativistically broadened and “normal” broad lines with a priori unknown line shape, etc., while in addition a superimposed warm absorber may have well defined narrow absorption lines. In that case it might be useful to fit the continuum with an arbitrary profile in order to get first an accurate description of the absorber, and then after having “removed” the absorber try to understand the underlying continuum spectrum.
For these situations the spln model introduced here is useful. It allows the user to model the continuum within two boundaries and with a cubic spline.
The algorithm works as follows. The user selects the limits and as well as the number of grid points . SPEX then creates a grid with uniform spacing in (see below for details). The spectrum at these grid points is contained in the corresponding array . These have the usual units of photons used throughout SPEX, and is the spectrum emitted at the source. The parameters can be adjusted during the spectral fitting, but , and and thereby remain fixed. At intermediate points between the , the photon spectrum is determined by cubic spline interpolation on the data pairs. We take a natural spline, i.e. at and the second derivative of the spline is zero.
Outside of the range – however, the photon spectrum is put to zero, i.e. no extrapolation is made!
Finally note that we have chosen to define the spline in logarithmic space of , i.e. the of the photon spectrum is fit by a spline. This is done in order to guarantee that the spectrum remains non-negative everywhere. However, the -values listed is the (linear) photon spectrum itself.
There are four different options for the energy grid , indicated by the parameter type:
type=1: is the lower energy in keV, is the upper energy in keV, and the grid is linear in energy in between.
type=2: is the lower energy in keV, is the upper energy in keV, and the grid is logarithmic in energy in between.
type=3: is the lower wavelength in Å, is the upper wavelength in Å, and the grid is linear in wavelength in between.
type=4: is the lower wavelength in Å, is the upper wavelength in Å, and the grid is logarithmic in wavelength in between.
Note that the logarithmic grids can also be used if one wants to keep a fixed velocity resolution (for broadened line features for example). Further, each time that the model is being evaluated, the relevant values of the grid points are evaluated.
Warning
Be aware that if you just set , and
but do not issue the “calc” command or the “fit” command, the
values have not yet been calculated and any listed values
that you get with the par show
command will be wrong. After the first
calculation, they are right.
Warning
If at any time you change one of the parameters , , or , the values will not be appropriate anymore as they correspond to the previous set of values.
The maximum number of grid points that is allowed is 999, for practical reasons. Should you wish to have a larger number, then you must define multiple spln components, each spanning its own (disjunct) – range.
It should be noted, however, that if you take very large, spectral fitting may become slow, in particular if you take your initial guesses of the parameters not too close to the true values. The reason for the slowness is two-fold; first, the computational time for each fitting step is proportional to the number of free parameters (if the number of free parameters is large). The second reason is unavoidable due to our spectral fitting algorithm: our splines are defined in photon spectrum space; if you start for example with the same value for each , the fitting algorithm will start to vary each parameter in turn; if it changes for example parameter by 1, this means a factor of 10; since the neighbouring points (like and however are not adjusted in third step, the photon spectrum has to be drawn as a cubic spline through this very sharp function, and it will show the well-known over-and undershooting at the intermediate x-values between and and between and ; as the data do not show this strong oscillation, will be poor and the fitting algorithm will decide to adjust the parameter only with a tiny amount; the big improvement in would only come if all values of were adjusted simultaneously.
The parameters of the model are:
type
: The parameter type defined above; allowed values are 1–4.
Default value: 1.n
: The number of grid points . Should be at least 2.low
: Lower x-value .upp
: Upper x-value . Take care to take
.x001
: First x-value, by definition equal to .
-values are not allowed to vary (i.e. you may not fit them).x002
: Second x-valuex003
: Third x-value...
: Other x-valuesx999
: last x-value, by definition equal to . If
, replace the 999 by the relevant value (for example, if
, then the last -value is x237).y001
: First y-value. This is a fittable parameter.y002
: Second y-valuey003
: Third y-value...
: Other y-valuesy999
: last y-value. If , replace the 999 by the
relevant value (for example, if , then the last
-value is 237).