1: \documentclass[12pt,preprint]{aastex}
2:
3: \newcommand{\chandra}{{\it Chandra}}
4: \newcommand{\xmm}{{\it XMM-Newton}}
5: \newcommand{\conx}{{\it Constellation-X}}
6: \newcommand{\ee}[1]{\times10^{#1}}
7: \newcommand{\dem}{\mathcal{D}}
8: \newcommand{\ddem}{\mathcal{H}}
9: \newcommand{\mdot}{\dot{M}}
10: \newcommand{\vinf}{v_\infty}
11: \newcommand{\nH}{n_{\rm H}}
12: \newcommand{\kms}{km\,s$^{-1}$}
13: \newcommand{\ergs}{erg\,s$^{-1}$}
14: \newcommand{\thori}{$\theta^1$\,Ori~C}
15: \newcommand{\tnm}[1]{\tablenotemark{#1}}
16: \newcommand{\ud}[2]{$^{+#1}_{-#2}$}
17: \slugcomment{{\Large DRAFT: \today}}
18:
19: \begin{document}
20:
21: \title{Ion-by-Ion Differential Emission Measure Determination of
22: Collisionally Ionized Plasma: I. Method}
23:
24: \shorttitle{Ion-by-Ion DEM Determination: I. Method}
25:
26: \author{Patrick S. Wojdowski and Norbert S. Schulz}
27:
28: \affil{Center for Space Research, Massachusetts Institute of
29: Technology}
30:
31: \email{pswoj@space.mit.edu}
32:
33: \begin{abstract}
34: We describe a technique to derive constraints on the differential
35: emission measure (DEM) distribution, a measure of the temperature
36: distribution, of collisionally ionized hot plasmas from their X-ray
37: emission line spectra. This technique involves fitting spectra using
38: a number of components, each of which is the entire X-ray line
39: emission spectrum for a single ion. It is applicable to
40: high-resolution X-ray spectra of any collisionally ionized plasma and
41: particularly useful for spectra in which the emission lines are
42: broadened and blended such as those of the winds of hot stars. This
43: method does not require that any explicit assumptions about the form
44: of the DEM distribution be made and is easily automated.
45: \end{abstract}
46:
47: \section{Introduction}
48: \label{sec:intro}
49: The X-ray emission for a large class of astrophysical objects is
50: dominated by emission from a hot, thin plasma in which the electrons
51: have a thermal velocity distribution and ion-electron recombination is
52: balanced by collisional ionization --- i.e., collisional ionization
53: equilibrium. This class of astrophysical objects includes both hot
54: and cool stars, clusters of galaxies, and elliptical galaxies as well
55: as some cataclysmic variable systems and active galactic nuclei
56: \citep[see, e.g.,][]{pae03}. A large class of laboratory plasmas
57: %including electric discharges, tokamak plasmas, laser-produced plasmas
58: are also in collisional ionization equilibrium. At temperatures above
59: $\sim10^6$\,K, plasmas in collisional ionization equilibrium emit
60: X-rays in the 1.5--30\,\AA{} (0.4--8\,keV) band. Up to temperatures of
61: $\sim3\ee8$, that X-ray emission is characterized by strong line
62: emission. This line emission varies strongly with temperature changes
63: of a few tenths of a decade or more.
64:
65: The unprecedented ability of the grating spectrometers on the
66: \chandra{} and \xmm{} observatories to resolve this X-ray line
67: emission present new possibilities for measuring temperatures of
68: plasmas in CIE and, therefore, new tests of theories of astrophysical
69: objects. Compilations of X-ray line emissivities have long been
70: available (e.g., \citealt{ray77}, \citealt*{mew85}, \citealt{smi01})
71: and given a detailed model of an astrophysical object, it is, in
72: general, straightforward to compute the emission line spectrum of any
73: optically thin CIE plasma. For the purpose of determining
74: luminosities of most emission lines, the three dimensional temperature
75: and density distributions may be reduced to the differential emission
76: measure (DEM) distribution, a single-valued function of temperature
77: which we define in \S\ref{sec:dem}. However, it is often the case
78: that no explicit theoretical prediction exists for the DEM
79: distribution of objects of interest. Furthermore, even when theory
80: does provide an explicit prediction of the differential emission
81: measure, it is often desirable to make measurements of the
82: differential emission measure without respect to any model.
83: Therefore, in this paper, we develop a method for obtaining
84: constraints on the DEM distribution of a collisionally ionized plasma
85: from its X-ray emission line spectrum without respect to any physical
86: model.
87:
88: Our method consists of fitting the observed spectral data using a
89: model consisting of a continuum plus a number of line emission
90: components, with each line emission component containing all of the
91: lines of a single ion in the observed wavelength band. Since each ion
92: emits only in a specific temperature range, the best fit magnitude for
93: the line emission of a given ion gives a measure of a weighted average
94: of the DEM distribution in that temperature range times the abundance
95: of that element. We plot the constraints for all of the ions as a
96: function of the temperature at which emission from the ion peaks.
97: These plots may be understood as one-dimensional ``images'' of the DEM
98: distributions in that they consist of discrete ``pixels'' (one for
99: each ion) and differ from the true DEM distribution by a convolution
100: which may be understood as a ``temperature-spread'' function. In
101: addition, by fitting the entire X-ray spectrum, rather than attempting
102: to measure fluxes of individual lines, we take advantage of the
103: information available from blended lines. This is particularly
104: important in the analysis of spectra in which the lines are
105: significantly broadened and blended due to insufficient instrumental
106: spectral resolution, Doppler motion of the emitting plasma, or other
107: effects. This method can be automated, facilitating the analysis of
108: large numbers of spectra. This attribute will become increasingly
109: important with the future \conx{} mission which will be able to obtain
110: high-resolution spectra for a large number of objects. While we are
111: unaware of any use of this method exactly as it is described here, it
112: is quite similar to and, in fact, inspired by methods described and
113: applied by \citet{pot63} for the analysis of solar ultraviolet
114: spectra, by \citet{sak99} for the analysis of an X-ray spectrum of the
115: photoionized wind of the high mass X-ray binary Vela~X-1, and by
116: \citet{bri01} and \citet*{beh01} for analyses of the X-ray spectra of
117: the corona of the cool stars HR~1099 and Capella, respectively.
118:
119: In \S\ref{sec:dem} we motivate and describe our method in detail and
120: in \S\ref{sec:simtest}, we apply it to a simulated spectrum of a
121: plasma with a continuous temperature distribution and simulated
122: spectra of several single temperature plasmas. In
123: \S\ref{sec:discuss}, we discuss the possible applications of our
124: method. The X-ray emission of most hot stars originates in supersonic
125: stellar winds, resulting in broadened and blended emission lines.
126: Therefore, our technique is particularly useful in the analysis of
127: these spectra and in a companion paper (hereafter, ``Paper II'') we
128: apply our technique to spectra of nine hot stars.
129:
130: \section{The Differential Emission Measure Distribution and its
131: Determination}
132: \label{sec:dem}
133:
134: In a plasma in CIE, line emission is due primarily to electron-ion
135: collisional excitation. Therefore, at low densities, for a given
136: temperature, line emissivities are proportional to the square of the
137: density. However, at high enough density, ions in some metastable
138: excited states may undergo additional collisional excitation,
139: resulting in the emissivities of some lines depending on higher powers
140: of the density. By definition, in a plasma in CIE, the radiation
141: field is too weak to affect the ionization balance. However, there is
142: a regime in which a plasma may be in CIE but in which the radiation
143: field is strong enough to induce further excitation of ions in
144: metastable excited states. This leads to additional dependence of the
145: line emissivities on the radiation intensity.
146:
147: For simplicity, in \S\ref{sec:low_den}, we define the DEM distribution
148: and describe a method for deriving constraints on it from its X-ray
149: emission spectrum for a plasma in which excited ions do not undergo
150: further excitation. Then, in \S\ref{sec:pump}, we describe a modified
151: version of this method for deriving constraints on the DEM
152: distribution of plasmas in which ions in metastable excited may
153: undergo further excitation.
154:
155: \subsection{Low Density and Radiation Intensity}
156: \label{sec:low_den}
157:
158: In the limit of low density and low radiation intensity, at a given
159: temperature, line emissivities depend on the square of the density.
160: Therefore, for the purpose of determining the emission line spectrum,
161: the three dimensional temperature and density distributions may be
162: reduced to the DEM distribution defined as:
163: \begin{equation}
164: \dem(T)\equiv\frac{dE}{d\log{}T}
165: \label{eqn:dem}
166: \end{equation}
167: where $T$ is the electron temperature and $E$ is the emission measure
168: defined as
169: \begin{equation}
170: E(T)\equiv\int_0^{T} n_e\nH{}dV
171: \end{equation}
172: where $n_e$ is the electron density, $\nH{}$ is the hydrogen atom
173: density and the integration is over that volume where the temperature
174: is less than $T$. In addition, because diffuse plasmas in collisional
175: ionization equilibrium cool radiatively, the cooling of gas is also
176: determined by the DEM distribution. Therefore, constraints on the DEM
177: distribution provide constraints on the overall energetics of the
178: X-ray emitting plasmas.
179:
180: The luminosity of an emission line $i$ from an ion of charge state $z$
181: of element $Z$ may be expressed as
182: \begin{equation}
183: L_{Z,z,i}=A_Z\int \dem(T)P_{Z,z,i}(T)d\log{}T
184: \label{eqn:l}
185: \end{equation}
186: where $A_Z$ is the abundance of element $Z$ relative to solar and
187: $P_{Z,z,i}(T)$ is the line power function. The line
188: power function depends only on atomic physics parameters and the solar
189: abundance of the element. Its variation with temperature is due to
190: the variation of the fractional abundance of the emitting ion as well
191: as the variation of intrinsic collisional excitation rates. Line
192: power functions have been tabulated for a large number of lines.
193: Therefore, from the measurement of a line luminosity, a constraint on
194: the product of the elemental abundance and the DEM distribution can be
195: inferred.
196:
197: In the absence of specific theoretical predictions of DEM
198: distributions, it may be useful to determine, at least approximately,
199: the magnitude and form of the DEM distribution. If we divide
200: Equation~\ref{eqn:l} by $\int{}P_{Z,z,i}(T)d\log{}T$, then it may be
201: seen that the measurement of a line luminosity constrains the product
202: of an elemental abundance and a ``weighted average'' of the DEM
203: distribution. In Figure~\ref{fig:t_deps}, we show the power functions
204: of the lines of two ions.
205: %%% START FIGURE
206: \begin{figure}
207: \begin{tabular}{cc}
208: \includegraphics[angle=-90,width=3.5in]{f1a.eps} &
209: \includegraphics[angle=-90,width=3.5in]{f1b.eps} \\
210: \end{tabular}
211: \caption{Line power functions and summed line power functions for the
212: $L$-shell ion \ion{Fe}{18} and the hydrogen-like ion \ion{Si}{14}.
213: The topmost curve in each panel is the total line power function
214: ($\Theta$, defined by Equation~\ref{eqn:Theta}). All lines from the
215: database are plotted. For \ion{Fe}{18}, the lines can be seen to have
216: nearly the same temperature dependence. For \ion{Si}{14}, a family of
217: line power functions have a temperature dependence significantly
218: different from the others, peaking at a lower temperature. However,
219: the peak values of these line power functions are only a few percent
220: of those of the strongest lines. These figures illustrate
221: the validity of our approximation that the line power functions of
222: most of the lines of a given ion, and all of the strongest lines of
223: that ion, the line power functions have almost the same shape. The
224: sparseness of the temperature grid is evident from the jagged
225: appearance of the curves.}
226: \label{fig:t_deps}
227: \end{figure}
228: %%% END FIGURE
229: The line power data we use in this figure and elsewhere in this paper
230: is from a version of Astrophysical Plasma Emission Database
231: \citep[APED,][]{smi01} based on the publicly released version 1.1 but
232: modified to include the dependence of the line power functions on
233: density (R. Smith, private communication 2002). The grid for the
234: database we use is somewhat sparse: line powers are tabulated for ten
235: temperature values per decade and two density values per decade. It
236: may be seen in the figure that the line power functions are
237: single-peaked and, compared to their peak values, are negligible
238: outside of temperature ranges of about one decade around each peak.
239: Therefore, the weighted average values of the DEM that we obtain from
240: the measurement of line luminosities are, approximately, the average
241: value of the DEM in the vicinity of the peak of that line's power
242: function.
243:
244: Measuring the luminosities of several lines with power functions that
245: peak at different temperatures gives us approximate average values of
246: the DEM distribution in several temperature ranges. However, for many
247: applications, it may be difficult to measure luminosities of
248: individual lines as they may be blended owing either to the finite
249: spectral resolution of the instrument or intrinsic line broadening.
250: In order to avoid the uncertainties related to the measurement of
251: fluxes of individual lines, we take advantage of the fact that the
252: power functions of most of the lines, and all of the strong lines, of
253: an ion have very nearly the same shape. This similarity of line power
254: functions is demonstrated in Figure~\ref{fig:t_deps}. If two lines
255: have power functions that have the same shape but differ by a constant
256: factor, then the ratio of the luminosities of those two lines will
257: differ by that same constant factor regardless of the DEM
258: distribution. Therefore, instead of attempting to measure the fluxes
259: of individual emission lines in a spectrum, we fit the entire spectrum
260: using all of the lines in the database, constraining the line fluxes
261: such that ratios of the line fluxes for any individual ion are fixed.
262: In such a fit, there is only one free line luminosity normalization
263: parameter for each ion. Just as the luminosities of individual lines
264: imply average values of the DEM distribution in temperature ranges
265: defined by the line power functions, so do these normalization
266: parameters. However, because each ion has a distinct set of lines
267: with distinct luminosity ratios, these normalization parameters and
268: the corresponding average values of the DEM distribution can be
269: uniquely constrained, even if lines are significantly blended. For
270: several ions, there is a family of lines, the dielectronic
271: recombination lines, that have power functions that differ
272: significantly from the other lines of the ion. This fact can be seen
273: in Figure~\ref{fig:t_deps} for \ion{Si}{14}.
274: %Furthermore, even neglecting the dielectronic recombination lines,
275: %the line power functions do not have exactly the same shape.
276: However, the dielectronic lines are quite weak. The peak powers of
277: the dielectronic recombination lines are only a few percent of those
278: of strongest lines. Therefore, we proceed with our attempt to derive
279: constraints on DEM distributions by fitting spectra with fixed line
280: luminosity ratios.
281:
282: In order to choose line luminosity ratios for fitting spectra, we
283: define the function $\Theta_{Z,z}$ for each ion as the sum of all of
284: the power functions for all of the lines of that ion:
285: \begin{equation}
286: \Theta_{Z,z}(T)\equiv\sum_iP_{Z,z,i}(T).
287: \label{eqn:Theta}
288: \end{equation}
289: For each function $\Theta$, we define a temperature $T_{\rm p}$ to be
290: the temperature at which it peaks.
291: We then let the line luminosities be
292: given by
293: \begin{equation}
294: L_{Z,z,i}=D_{Z,z}P_{Z,z,i}(T_{{\rm p},Z,z})\Delta\log{}T_{Z,z}
295: \label{eqn:ld}
296: \end{equation}
297: where $D_{Z,z}$ is a variable normalization parameter and
298: \begin{equation}
299: \Delta_{Z,z}\log{}T\equiv\frac{\int\Theta_{Z,z}(T)d\log{}T}{\Theta_{Z,z}(T_{{\rm p},Z,z})}.
300: \label{eqn:dellogt}
301: \end{equation}
302: In Table~\ref{tab:ion_temp} we give values of $T_{{\rm p}}$ and the
303: quantities $\Delta_{-}\log{}T$ and $\Delta_{-}\log{}T$, defined as
304: \begin{eqnarray}
305: \Delta_{-}\log{}T_{Z,z} & \equiv & \frac{\int_0^{T_{{\rm p},Z,z}}
306: \Theta_{Z,z}(T)d\log{}T}{\Theta_{Z,z}(T_{{\rm p},Z,z})},\\
307: \Delta_{+}\log{}T_{Z,z} & \equiv & \frac{\int_{T_{{\rm
308: p},Z,z}}^\infty \Theta_{Z,z}(T)d\log{}T}{\Theta_{Z,z}(T_{{\rm p},Z,z})}
309: \end{eqnarray}
310: for each of
311: the ions we use in our analysis. The quantities $\Delta_{-}\log{}T$ and
312: $\Delta_{+}\log{}T$ provide an indication of the temperature ranges
313: over which the ions have significant line emission.
314: \begin{deluxetable}{lrrr}
315: \tablewidth{0pt}
316: \tablecaption{Temperature Data for Ion Line Power
317: Functions\label{tab:ion_temp}}
318: \tablehead{
319: \colhead{Ion} & \colhead{$T_{\rm p}$(K)} &
320: %\colhead{$\Delta\log{}T$} &
321: \colhead{$\Delta_{-}\log{}T$} &
322: \colhead{$\Delta_{+}\log{}T$}}
323: \startdata
324: \ion{N}{6} & $1.6\ee6$ & 0.20 & 0.34 \\
325: \ion{N}{7} & $2.0\ee6$ & 0.14 & 0.34 \\
326: \ion{O}{7} & $2.0\ee6$ & 0.16 & 0.21 \\
327: \ion{O}{8} & $3.2\ee6$ & 0.18 & 0.34 \\
328: \ion{Ne}{9} & $4.0\ee6$ & 0.22 & 0.19 \\
329: \ion{Ne}{10} & $6.3\ee6$ & 0.23 & 0.35 \\
330: \ion{Mg}{11} & $6.3\ee6$ & 0.22 & 0.22 \\
331: \ion{Mg}{12} & $1.0\ee7$ & 0.20 & 0.40 \\
332: \ion{Si}{13} & $1.0\ee7$ & 0.24 & 0.23 \\
333: \ion{Si}{14} & $1.6\ee7$ & 0.21 & 0.42 \\
334: \ion{S}{15} & $1.6\ee7$ & 0.30 & 0.22 \\
335: \ion{S}{16} & $2.5\ee7$ & 0.24 & 0.43 \\
336: \ion{Ar}{17} & $2.0\ee7$ & 0.25 & 0.28 \\
337: \ion{Ar}{18} & $4.0\ee7$ & 0.28 & 0.42 \\
338: \ion{Ca}{19} & $2.5\ee7$ & 0.24 & 0.33 \\
339: \ion{Ca}{20} & $5.0\ee7$ & 0.23 & 0.48 \\
340: \ion{Fe}{17} & $5.0\ee6$ & 0.18 & 0.21 \\
341: \ion{Fe}{18} & $7.9\ee6$ & 0.21 & 0.10 \\
342: \ion{Fe}{19} & $7.9\ee6$ & 0.11 & 0.15 \\
343: \ion{Fe}{20} & $1.0\ee7$ & 0.13 & 0.11 \\
344: \ion{Fe}{21} & $1.0\ee7$ & 0.08 & 0.18 \\
345: \ion{Fe}{22} & $1.3\ee7$ & 0.11 & 0.15 \\
346: \ion{Fe}{23} & $1.6\ee7$ & 0.16 & 0.17 \\
347: \ion{Fe}{24} & $2.0\ee7$ & 0.18 & 0.32 \\
348: \ion{Fe}{25} & $6.3\ee7$ & 0.33 & 0.33 \\
349: \ion{Fe}{26} & $1.3\ee8$ & 0.28 & 0.48
350: \enddata
351: \end{deluxetable}
352: %We will make use of the quantities $\Delta_{-}\log{}T$ and
353: %$\Delta_{+}\log{}T$ in \S\ref{sec:implementation}, however
354: The fact that several ions have the same value of $T_{\rm p}$ is an
355: artifact of the sparse temperature grid.
356:
357: %The data base we use has line powers as functions of density in the
358: %range $10^6$--$10^{15}$\,cm$^{-3}$ and of temperature in the range
359: %$10^4$--$7.9\ee8$\,K.
360: %In all computations described in this paper,
361: %line power functions are taken to be zero outside that temperature
362: %range.
363: % We first determine $T_{{\rm p},Z,z}$ and $\Delta_{Z,z}\log{}T$
364: %for each ion using $n_e=10^6$\,cm$^{-3}$, the lowest density in the
365: %database.
366:
367: If all of the line power functions of a given ion have exactly the
368: same shape, then the value of $D$ for that ion gives the product of
369: the elemental abundance and the average of the DEM distribution
370: weighted by that ion's function $\theta$ as described above for
371: individual lines. As mentioned above, not all of the line power
372: functions for every ion have the same shape. However, as those line
373: power functions that differ significantly have magnitudes of only
374: approximately 1\% of the value of the strongest line power functions,
375: we expect an error due to the difference of line power functions of no
376: more than about 1\%.
377:
378: \subsection{Line Pumping}
379: \label{sec:pump}
380:
381: %We are not interested here in diagnosing the density distribution of
382: %the X-ray emitting material or the intensity of the radiation field in
383: %it. However, the X-ray emission line spectra of hot stars indicate
384: %significant radiation densities in the emitting plasma. Therefore, a
385: %small modification to the method described in \S\ref{sec:low_den} is
386: %required to fit these spectra and obtain accurate constraints on their
387: %DEM distributions.
388:
389: As we have mentioned before, some ions have metastable excited states
390: that are susceptible to further excitation by collisions with
391: electrons or absorption of photons. The result of this is that the
392: emission in lines resulting from the decay of the metastable state are
393: ``pumped'' into other lines. We first consider pumping by collisional
394: excitation because the effects of the two mechanisms are similar and
395: because data for pumping by collisional excitation is more readily
396: available. For this case, the line emission may still be described by
397: line power functions. However, the line power $P$ is a function of
398: density as well as temperature. Therefore, the emission line spectrum
399: depends on the two-dimensional temperature-density DEM
400: \begin{equation}
401: \ddem(T,n_e)\equiv{}\frac{d^2E}{d\log{}T\,d\log{}n_e}
402: \end{equation}
403: The DEM distribution is related to this quantity by
404: \begin{equation}
405: \dem(T)=\int\ddem{}(T,n_e)d\log{}n_e.
406: \end{equation}
407: If the density dependence of each line power function of a single ion
408: is due to a single upward transition from a single metastable state
409: and the collisional excitation rate for that transition does not
410: change much over the temperature range where that ion emits, then the
411: line power functions for that ion may be well approximated by the form
412: \begin{equation}
413: P_{Z,z,i}(T,n_e)=\Theta_{Z,z}(T)(B_{Z,z,i}+F_{Z,z,i}H_{Z,z}(n_e))
414: \label{eqn:ptd}
415: \end{equation}
416: where $\Theta$ and $H$ are functions that are the
417: same for all of the lines of a given ion and $B$ and
418: $F$ are constant coefficients for each line. Even if the two
419: conditions mentioned above are not satisfied, the line power functions
420: may be well approximated by this form. We discuss the validity of
421: this assumption in \S\ref{sec:accpow} and proceed here with the
422: assumption that the power functions have this form.
423:
424: We define the functions $\Theta$, and the temperatures where they
425: peak, $T_{\rm p}$, as before in the limit of low density and low
426: radiation intensity. With this definition, the functions $H(n_e)$ go
427: to zero as $n_e$ does. If the functions $H$ are continuous, it is
428: possible to show that for any temperature-density DEM distribution
429: $\ddem(T,n_e)$ with an associated DEM distribution
430: $\dem(T)=\int{}\ddem(T,n_e)d\log{}n_e$, there exists, for each ion, a
431: value of the density $n^\prime_{Z,z}$ such that the
432: temperature-density DEM distribution $\ddem_{Z,z}^\prime$ defined as
433: \begin{equation}
434: \ddem_{Z,z}^\prime(T,n_e)\equiv\dem(T)\delta(n_e-n^\prime_{Z,z})
435: \end{equation}
436: produces the same luminosities for all of the emission lines of the
437: ion $Z,z$ as does $\ddem(T,n_e)$. Therefore, we modify
438: the method described above by adopting the variable parameters
439: $n^\prime_{Z,z}$ and let the line luminosities be
440: \begin{equation}
441: L_{Z,z,i}=D_{Z,z}P_{Z,z,i}(T_{{\rm p},Z,z},n^\prime_{Z,z})\Delta_{Z,z}\log{}T
442: \end{equation}
443: where $\Delta_{Z,z}\log{}T$ is defined as before in
444: Equation~\ref{eqn:dellogt}. That is, we fit the entire spectrum using
445: all of the lines in the database as described in \S\ref{sec:low_den}.
446: However, instead of constraining the line flux ratios of individual
447: ions to be fixed, we allow the ratios to vary with a density parameter
448: $n^\prime_{Z,z}$ for that ion. The density parameters for each of the
449: ions are allowed vary independently. Again, $D_{Z,z}$ approximates
450: the product of the elemental abundance and the average of the DEM
451: distribution weighted by $\Theta_{Z,z}(T)$. Even if
452: equation~\ref{eqn:ptd} is not satisfied, this method will result in
453: good fits to the data and accurate constraints on the DEM distribution
454: if the emitting plasma has a single density or if density and
455: temperature are strongly correlated within the plasma.
456:
457: \subsubsection{Accuracy of the Power Function Approximation}
458: \label{sec:accpow}
459:
460: A systematic study of the density dependence of emission lines with
461: wavelengths from 1.2--31\,\AA{} has been undertaken by \citet{smi02}.
462: These authors have made fits to the line powers as functions of
463: density at the constant temperatures $10^6$, $10^{6.5}$, $10^7$, and
464: $10^{7.5}$~K for lines with peak powers exceeding a minimum value and
465: also meeting a criterion for variability with density at each
466: temperature.
467: %These authors also searched for lines meeting these criteria for the
468: %temperatures $10^5$, $10^{5.5}$, and $10^8$\,K but found none.
469: They found the power functions of most of the lines satisfying those
470: criteria ($\sim$90\%) could be approximated well using a function of
471: the form
472: \begin{equation}
473: P(n_e)=c_0+c_1\exp(-n_e/n_1)
474: \label{eqn:pd}
475: \end{equation}
476: where $c_0$, $c_1$, and $n_1$ are fit parameters. If the density
477: dependence for the line powers of all of the lines of an ion have the
478: form of Equation~\ref{eqn:pd}, then Equation~\ref{eqn:ptd} is a valid
479: description of the line power functions if, in the temperature range
480: where line emission is significant,
481: \begin{itemize}
482: \item all of the lines of any one ion have the same value of $n_1$ and
483: \item for any single line, the ratio of the line power at high density
484: ($c_0$) to the value at low density ($c_0+c_1$) and the value of $n_1$
485: do not change with temperature.
486: \end{itemize}
487: We have inspected the results of \citet{smi02} and found that, for all
488: of the ions we use except \ion{Fe}{19}, \ion{Fe}{20}, and
489: \ion{Fe}{21}, most of the lines ($\sim$90\%) are described by
490: Equation~\ref{eqn:pd} and have values of $n_1$ that are very close,
491: having a standard deviation of 0.1 or less in $\log{}n_1$. For those
492: lines, of any ion, that meet the criteria to be fit at more than one
493: temperature, the root mean square (RMS) of the difference between
494: values of $\log{}n_1$ for the same ion at different temperatures is
495: 0.09 and the RMS of the difference between values of
496: $\log(c_0/(c_0+c_1))$ is 0.16. In summary, Equation~\ref{eqn:ptd} is
497: not exactly satisfied for all ions. However, because most line power
498: functions do not depend on density, it is unlikely that this would
499: cause errors greater than a factor of a few. Because
500: Equation~\ref{eqn:ptd} is near to being satisfied for most ions, we
501: expect the actual errors to be much less: not much greater than 10 or
502: 20\%.
503:
504: \subsubsection{Radiation}
505:
506: Because metastable states may be photoexcited, line powers are
507: functions not only of temperature and density but also of the mean
508: radiation intensity at the frequencies of the transitions that affect
509: line emission. That is, the line power may be written
510: $P_{Z,z,i}(T,n_e,J_\nu)$ or
511: $P_{Z,z,i}(T,n_e,J_{\nu_{i1}},J_{\nu_{i2}},...,J_{\nu_{im}})$ where
512: $J_\nu$ is the mean radiation intensity as a function of frequency and
513: $J_{\nu_{ij}}$ are the mean radiation intensities at the frequencies
514: of the transitions affecting line $i$. While this is, in principle, a
515: large number of variables, the number of transitions in which
516: photoexcitation plays a significant role (the value of $m$) is often
517: only one. Furthermore, for a given system, the values of
518: $J_{\nu_{ij}}$ may be a function of a small number of variables. For
519: example, in hot star winds, the radiation intensity due to the stellar
520: photosphere is given by
521: \begin{equation}
522: J_\nu=I_{\star,\nu}W
523: \end{equation}
524: where $I_{\star,\nu}$ is the radiation intensity at the surface of the
525: star and is approximately that of a blackbody with a temperature,
526: depending on the stellar type, of a few tens of thousand K and $W$ is
527: a factor accounting for the geometrical dilution of the stellar
528: radiation with distance from the star. We know of no systematic study
529: (at least, not of the scope of that by \citealt{smi02} for density) of
530: the dependence of line powers on radiation. However, as radiation and
531: density both affect the line power through excitation of metastable
532: stable states, the effects are similar. Therefore, we use the method
533: described above --- we take the line emission from each ion to be
534: determined by the temperature at which its emission peaks and a
535: density which is a free parameter, independent of the density values
536: of the other ions --- and assume that the effects of radiation and
537: density can be replicated approximately by density alone. In Paper II
538: we again address the validity of this approach for plasmas with a
539: significant pumping radiation field.
540:
541: \section{Tests with Simulated Data}
542: \label{sec:simtest}
543:
544: \subsection{Simulations}
545: \label{sec:sims}
546:
547: In order to demonstrate the performance of our model, we tested our
548: method by applying it to a series of simulated spectra. We simulated
549: \chandra{} MEG and HEG spectra for a plasma with a DEM distribution
550: constant with temperature over the range
551: $3.2\ee{5}$--$7.9\ee8$\,K\footnote{actually one component for each of
552: the database's temperature values, with each component having the same
553: emission measure} and zero elsewhere. We also simulated spectra from
554: single-temperature plasmas at the temperatures $3\ee5$, $10^6$,
555: $3\ee6$, $10^7$, $3\ee7$, $10^8$, and $3\ee8$\,K. In Paper II we
556: apply our method to spectra of hot stars. Therefore, for our
557: simulations we adopt the parameters of $\zeta$~Pup, the best-studied
558: hot star in the X-ray band, and of an observation of it with
559: \chandra{} (see Paper II). However, the results of the application of
560: our method to these simulated data indicate the general behavior of
561: our method.
562:
563: To choose a total emission measure for our simulations, we conducted a
564: number of three-temperature fits to the spectrum of $\zeta$~Pup, using
565: an fixed value of $1.0\ee{22}$\,cm$^{-2}$ \citep*[c.f.,][]{ber96} for
566: the interstellar absorption column. These fits resulted in total
567: emission measures in the range (2.4--5.0)$\ee{55}$\,cm$^{-3}$ (for an
568: adopted distance of 450\,kpc, \citealt*{sch97}) depending on whether
569: we tried to fit the nitrogen lines or the oxygen lines (this
570: discrepancy is discussed in more detail in Paper II) and we
571: chose a value of $3.6\ee{55}$\,cm$^{-3}$ for the total emission
572: measure. For the constant DEM distribution, this implies a DEM value
573: of $1.0\ee{55}$\,cm$^{-3}$. For our simulated exposure time, we used
574: 68,598\,s, the exposure time of our observation of $\zeta$~Pup. For
575: all of the simulations, $n_e$ was taken to be
576: $8\times10^{13}$\,cm$^{-3}$. We chose this particular density value
577: because it is large enough so that, like in the actual stellar
578: spectra, the forbidden lines of the helium-like triplets are
579: completely pumped into the intercombination lines. For the simulated
580: line profiles we used the ISIS thermal/turbulent line profile
581: function:
582: \begin{equation}
583: \phi_{Z,z,i}(\lambda)=\frac{1}{\sigma_{Z,z}\lambda_{Z,z,i}\sqrt{2\pi}}
584: \exp\left(\frac{(\lambda/\lambda_{Z,z,i}-(1+v_{\rm r}/c))^2}
585: {2\sigma_{Z,z}^2}\right)
586: \label{eqn:l_profile}
587: \end{equation}
588: where
589: \begin{equation}
590: \sigma_{Z,z}\equiv{}c^{-1}
591: \left(\frac12v_{\rm t}^2+\frac{kT_{{\rm p},Z,z}}{m_Z}\right)^{1/2}
592: \end{equation}
593: where $m_Z$ is the mass of element $Z$, $k$ is Boltzmann's constant.
594: We took $v_{\rm r}$ to be 0 and $v_{\rm t}$ to be 800\,\kms,
595: approximating the line widths observed by \citet{kah01} for
596: $\zeta$~Pup.
597:
598: \subsection{Application of the Method to the Simulated Data}
599: \label{sec:implementation}
600:
601: We fit the simulated spectra using the method described in
602: \S\ref{sec:pump}. In our analysis we do not attempt to use
603: measurements of the continuum to constrain the DEM distribution.
604: However, in order to fit spectra and obtain accurate constraints from
605: the emission lines, it is necessary to account for the continuum. The
606: continuum emission from a collisional plasma is due primarily to
607: bremsstrahlung, though radiative recombination continua and two-photon
608: continua also contribute. For our fits, we use a continuum consisting
609: of three bremsstrahlung components with variable temperatures and
610: normalizations. In all cases, this provides a sufficient empirical
611: representation of the continuum. For the fit model line profiles, we
612: use the same function as for the simulation model line profile
613: (equation~\ref{eqn:l_profile}). However, in the fit, the values of
614: $v_{\rm r}$ and $v_{\rm t}$ are taken to free but to have the same
615: values for all lines. Also, in the fit spectral models, we included
616: the same absorption ($1.0\ee{22}$\,cm$^{-2}$) used for the
617: simulation.
618:
619: Though we use the same atomic database for the simulations and the
620: model we fit to the simulated data, this does not amount to fitting
621: the data with the same model used for the simulation. For the
622: simulations, the line luminosities are given by equation~\ref{eqn:l}
623: with $\dem(T)$ constant or proportioal to $\delta(T-T_{\rm sim})$
624: where $T_{\rm sim}$ is the temperature of the simulated plasma. On
625: the other hand, in the model fit to the simulated data, the line
626: luminosites are given by equation~\ref{eqn:ld}. Unless all of line
627: power functions have the same shape every other line power function of
628: the same ion, the line luminosities of the simulation model and the
629: fit model will necessarily differ. Therefore, the fits to the
630: simulated data test whether or not the differences in line power
631: shapes between lines of individual ions are small enough for our
632: method to be valid.
633:
634: % The decision
635: %as to how many bremsstrahlung components are necessary to obtain a
636: %good fit to the continuum is made somewhat subjectively.
637: %However,
638: %this is the only part of the procedure in which human consideration is
639: %required for each spectrum.
640:
641: We simultaneously fit both the HEG and MEG simulated data to find the
642: best fit values of $D_{Z,z}$ and $n_{Z,z}^\prime$ for each of the ions by
643: minimizing the \citet{cas79} statistic. This statistic, unlike the
644: $\chi^2$ statistic, is valid in the regime where the number of counts
645: in a channel is small such as is the case for several of our data
646: sets. We search for a minimum of the Cash statistic with ISIS, first
647: by using its implementation of the Levenberg-Marquardt algorithm and
648: then using its implementation of the simplex algorithm. We then
649: search for confidence intervals, again using ISIS, on the parameters
650: $v_{\rm r}$, $v_{\rm t}$, and on each of the values of $D_{Z,z}$ and
651: $n_{Z,z}$. In searching for confidence intervals, only the
652: Levenberg-Marquardt algorithm is used. In searching for confidence
653: intervals, a new minimum is frequently found, requiring the process to
654: be restarted. On our workstations (with clock speeds of order
655: 1\,GHz), fitting and finding all of these confidence intervals for one
656: of our data sets generally takes a few weeks.
657:
658: In Figure~\ref{fig:flat} we plot the best fit values of $D$ as a
659: function of $T_{\rm p}$ for each of the ions (indicated by filled
660: circles) and also the simulated data and best fit model spectrum. In
661: the plot of $D$ vs. $T_{\rm p}$, a diamond surrounds each filled
662: circle. The vertical extent of a diamond indicates the statistical
663: error on the best-fit value of $D$ (given by $\Delta{}C=2.706$, where
664: $C$ is the Cash fit statistic --- this is the 90\% confidence region)
665: and the extent of a diamond to the left and right is given by
666: $\Delta_{-}\log{}T$ and $\Delta_{+}\log{}T$, respectively. These
667: temperature ranges are the temperature ranges for which the values of
668: $D_{Z,z}$ represent the approximate average values of the DEM.
669: \begin{figure}
670: \plotone{f2.eps}
671: \caption{In the first panel, the results for the simulated data for a
672: DEM constant in the range $3.2\ee5$--$7.9\ee8$\,K
673: ($\dem=1.0\ee{55}$\,cm$^{-3}$) is plotted. For each ion $Z,z$, we
674: plot a filled circle at ($T_{{\rm p},Z,z}$,$D_{Z,z}$) where $D_{Z,z}$
675: is our best-fit value. The vertical extent of the diamond around a
676: filled circle indicates the confidence region determined for $D_{Z,z}$
677: and the horizontal extent of a diamond is given by $\Delta_{-}T_{Z,z}$
678: and $\Delta_{+}T_{Z,z}$, which is defined in the text and is,
679: approximately, the region over which the ion emits. The ions
680: corresponding to the data points are labeled using Arabic numerals
681: rather than Roman numerals. Because the data points are so close in
682: this plot, we label only a few of them. In the second panel we show
683: the simulated MEG spectrum (black) and the best-fit model obtained
684: with our method (red). We label several bright lines. Though we do
685: not show it here, the HEG spectrum was also used in the fit. For this
686: and all of our simulated spectra, we adopt the absorption column
687: ($1.0\ee{20}$\,cm$^{-2}$), total emission measure
688: ($3.6\ee{55}$\,cm$^{-3}$) emission measure, and line width (800\,\kms)
689: of $\zeta$~Pup. The fact that the data points in the first panel are
690: consistent with $\dem=1.0\ee{55}$\,cm$^{-3}$ and the residuals are no
691: larger than the square root of the number of counts indicates that our
692: method is valid.
693: %\log{}T_{{\rm p},Z,z}-\Delta_{Z,z,{\rm
694: %l}}\log{}T$--$\log{}T_{{\rm p},Z,z}$
695: %are the best-fit values of $C_{Z,z}$ for the
696: %fit to . As
697: %expected, these values are, within errors, equal to the value of
698: %$\dem$. In the second and third panels are the simulated data (black)
699: %and best-fit model (red) for the MEG (second panel) and HEG (third
700: %panel).
701: }
702: \label{fig:flat}
703: \end{figure}
704: While we do not assess the quality of the spectral fit quantitatively,
705: it may be seen that the fit is quite good. The fact that the data
706: points in the first panel are consistent with
707: $\dem=1.0\ee{55}$\,cm$^{-3}$ and the residuals are no larger than the
708: square root of the number of counts indicates that our method is
709: valid. In Figure~\ref{fig:1temp_dems}, we show the plots of $D_{Z,z}$
710: for the simulated spectra of single-temperature plasmas, except for
711: the simulated $3\ee5$\,K plasma spectrum which had only a few counts.
712: \begin{figure}
713: \plotone{f3.eps}
714: \caption{Plots of DEM constraints for spectra of plasmas at the single
715: temperatures $10^6$, $3\ee6$, $10^7$, $3\ee7$, $10^8$, and $3\ee8$ K.
716: The designation ``(ul)'' indicates an upper limit.}
717: \label{fig:1temp_dems}
718: \end{figure}
719: The fits for these spectra, which we do not show, are also quite good.
720: In the first panel of Figure~\ref{fig:flat} and in
721: Figure~\ref{fig:1temp_dems} we use the same temperature range on the
722: horizontal axis and 3 orders of magnitude in DEM on the vertical axis.
723: This allows slopes of lines in the various plots to be compared.
724: Our best-fit values of $v_{\rm r}$ and $v_{\rm t}$ have uncertainties
725: of a few tens of \kms{} and are consistent with the input values of
726: zero and 800\,\kms, respectively.
727:
728: It may be seen from Equation~\ref{eqn:l} that, for a constant DEM
729: distribution, we expect the values of $D_{Z,z}$ to be that constant
730: value of the DEM and, as illustrated in the first panel of
731: Figure~\ref{fig:flat}, within the errors, this is indeed what we
732: find. This, and the good quality of the fits indicates that the fact
733: that not all of the power functions of every ion have the same
734: temperature dependence does not cause our method to be significantly
735: inaccurate. As our simulated plasma has only a single density, this
736: does not test whether or not deviations in the line power functions
737: from the form of Equation~\ref{eqn:ptd} cause significant
738: inaccuracies in our DEM determinations. Using plasmas with
739: distributions of densities would have provided a test of this.
740: However, this would not have provided a test of the accuracy of our
741: method for plasmas with strong ambient radiation fields. Simulating
742: the emission of a plasma with an ambient radiation field is beyond the
743: scope of this work.
744:
745: For each of the single temperature simulations, the determined values
746: of $D_{Z,z}$ show a peak at the temperature of the simulated plasma.
747: However, the sharpness of the peak differs for each of the
748: simulations. These plots may be understood as ``temperature-spread
749: functions'' in analogy with point spread functions in images for use
750: in evaluating our results from observed spectra.
751:
752: \section{Discussion}
753: \label{sec:discuss}
754:
755: We have described a method for deriving constraints on the
756: differential emission measure distribution of a collisionally ionized
757: plasma from its X-ray emission spectrum and displaying these
758: constraints in a way that indicates the DEM distribution. We have
759: designed our technique to account for the line pumping that occurs at
760: high densities and with high radiation intensities. We have applied
761: our method to simulated spectra and demonstrated that, with this
762: method, we can recover simulated DEM distributions in, approximately,
763: the temperature range $10^6$--$3\ee{8}$\,K from their emission line
764: spectra subject to a temperature resolution comparable to the width of
765: the line power functions: a few tenths of a decade.
766:
767: %We have attempted to design our to work
768: %In our analysis, we also derive a characteristic line shift
769: %and width for each spectrum. We intend to post our analysis scripts to the
770: %\chandra{} contributed software exchange
771: %(\url{http://asc.harvard.edu/cont-soft/soft-exchange.html}).
772: We expect that the technique described here will be useful in the
773: analysis of high-resolution spectra of plasmas in collisional
774: ionization equilibrium in several astrophysical contexts. This
775: technique is particularly useful in cases where lines are broadened to
776: the point of being blended. Therefore, we expect this technique to be
777: especially useful in the analysis of the X-ray spectra of hot stars
778: and we apply it to spectra of nine hot stars in Paper II.
779: \citet{muk03} have shown that the X-ray spectra of some cataclysmic
780: variable stars are characteristic of plasmas in collisional ionization
781: equilibrium. Some of those spectra show lines as broad as
782: 500\,\kms{}. Therefore, this technique may be useful in the analysis
783: of those objects. Even for spectra in which lines are not broad and
784: blended, this technique has advantages over others. It does not
785: require fitting of individual lines to determine their fluxes and is
786: therefore easily automated. Also with this technique, it is not
787: necessary to make assumptions or introduce biases about the form of
788: the DEM distributions. Therefore, our technique may also be useful
789: for analysis of high-resolution spectra of the coronae of cool stars
790: obtained with \xmm{} and \chandra. The {\it Astro-E II} observatory
791: will be able to obtain high-resolution spectra of extended objects.
792: Therefore, with the data from that observatory, our technique may be
793: useful in the analysis of the spectra of clusters of galaxies and
794: halos of elliptical galaxies. The fact that this algorithm can be
795: easily automated may be of particular importance when the DEM
796: distributions need to be determined for a large number of spectra.
797: The fact that the anticipated {\it Constellation-X} observatory will
798: have a very large effective area for high-resolution spectroscopy will
799: enable it to obtain spectra for a large number of objects that are too
800: faint to be efficiently observed with the three previously mentioned
801: observatories. Therefore, our technique may be especially useful in
802: the analysis of the large number of high-resolution X-ray spectra that
803: will be obtained with that observatory.
804:
805: \acknowledgements
806:
807: We thank John Houck for assistance implementing our analysis technique
808: in ISIS, Dan Dewey for a careful reading of the manuscript, and the
809: referee, Ehud Behar, for helpful comments. Support for this work was
810: provided by the National Air and Space Administration (NASA) through
811: Chandra Award Number GO0-1119X by the Chandra X-ray Observatory Center
812: (CXC) which is operated for and on behalf of NASA by the Smithsonian
813: Astrophysical Observatory (SAO) under contract NAS8-39073. Support
814: for this work was also provided by NASA through contract NAS8-01129
815: and by the SAO contract SVI-61010 for the CXC.
816:
817: \bibliographystyle{apj}
818: \bibliography{../ion_dem}
819:
820:
821: \end{document}
822: