1: \documentclass[12pt,preprint,xdvi]{aastex}
2: %\documentclass[10pt,namedreferences]{/Users/danalongcope/Library/texmf/tex/misc/SolarPhysicsDWL}
3:
4: \usepackage{psfig}
5:
6: %\setlength{\textwidth}{4.8in}
7:
8: \newcommand{\be}{\begin{equation}}
9: \newcommand{\ee}{\end{equation}}
10: \newcommand\eq{Equation}
11: \newcommand\eqs{Equations}
12: \newcommand\fig{Figure}
13: \newcommand\figs{Figures}
14: \newcommand\etal{{\it et al.}}
15: \newcommand{\bb}[1]{{\mathbf {#1}}}
16: \newcommand{\avg}[1]{{\langle{#1}\rangle}}
17: \newcommand{\half}{\hbox{${1\over2}$}}
18: \newcommand{\rvec}{{\bf r}}
19: \newcommand{\nhat}{{\bf \hat{n}}}
20: \newcommand{\xvec}{{\bf x}}
21: \newcommand{\kvec}{{\bf k}}
22: \newcommand{\bvec}{{\bf B}}
23: \newcommand{\xhat}{{\bf \hat{x}}}
24: \newcommand{\yhat}{{\bf \hat{y}}}
25: \newcommand{\papi}{LBP03}
26:
27: %\bibliographystyle{/Users/danalongcope/stuff/makebst/sphys}
28:
29: \begin{document}
30: %\begin{article}
31: %\begin{opening}
32:
33: \title{The Number of Magnetic Null Points in the Quiet Sun Corona}
34:
35: \author{D.W. Longcope,$^1$ C.E. Parnell$^2$}
36: \affil{1. Department of Physics, Montana State University\\
37: Bozeman, Montana 59717\\
38: 2. School of Mathematics and Statistics,
39: University of St Andrews,\\St Andrews, Fife, UK, KY16 9SS}
40:
41: \keywords{MHD --- Sun: corona --- Sun: magnetic fields}
42:
43: \begin{abstract}
44: The coronal magnetic field above a particular photospheric region
45: will vanish at a certain number
46: of points, called null points. These points can be found
47: directly in a potential field extrapolation or their density
48: can be estimated from Fourier spectrum of the magnetogram.
49: The spectral estimate, which
50: assumes that the extrapolated field is random, homogeneous and
51: has Gaussian statistics, is found here to be relatively accurate for
52: quiet Sun magnetograms from SOHO's MDI. The majority of null points
53: occur at low altitudes, and their
54: distribution is dictated by high wavenumbers in the Fourier spectrum.
55: This portion of the spectrum is affected by Poisson
56: noise, and as many as five-sixths of null points
57: identified from a direct extrapolation can be attributed to noise.
58: The null distribution above 1500 km is found to depend on wavelengths that
59: are reliably measured by MDI in either its low-resolution or
60: high-resolution mode. After correcting the spectrum to remove
61: white noise and compensate for the modulation transfer function
62: we find that a potential field extrapolation contains, on
63: average, one magnetic null point, with altitude greater than
64: 1.5 Mm, above every $322\,{\rm Mm}^2$ patch of quiet Sun. Analysis of 562
65: quiet Sun magnetograms spanning the two latest solar minima shows that
66: the null point density is relatively constant with roughly 10\%
67: day-to-day variation.
68: At heights above 1.5 Mm, the null point density decreases
69: approximately as the inverse cube of height.
70: The photospheric field in the quiet Sun is well approximated as that
71: from discrete elements with mean flux $\avg{|\phi|}=1.0\times10^{19}$ Mx
72: distributed randomly with density $n=0.007\,{\rm Mm}^{-2}$.
73: \end{abstract}
74:
75: \date{Draft: \today}
76:
77: %\end{opening}
78:
79: \section{Introduction}
80:
81: Models of magnetic reconnection, from the very earliest, have assigned
82: particular significance to points where the magnetic field vanishes:
83: magnetic null points \citep{Dungey1958,Sweet1958}.
84: In two-dimensional reconnection
85: models, null points (also called neutral points) are
86: natural locations to change magnetic field line topology. While
87: three dimensions offer alternatives for topological change
88: \citep{Greene1988,Hesse1988}, null
89: points are still natural locations for current intensification
90: \citep{Craig1995,Rickard1996,Galsgaard1997,Pontin2007}
91: and for focusing of magnetosonic waves
92: \citep{Hassam1992,Craig1993,McLaughlin2004}. Dissipation of magnetic
93: energy at magnetic null points is therefore a potential source of
94: heat in the quiet Sun corona. The effectiveness of this dissipation
95: for heating the entire corona will depend, in part, in how null
96: points are distributed throughout its field.
97:
98: % effectiveness of heating by reconnection at coronal null points
99: % would seemingly depend on how commonly such structures are found in
100: % the coronal field.
101:
102: % Jets: common features in X-ray observations. more common w/ Hinode
103: % [Shibata models. Hinode observations]
104: % models involve null point at cusp.
105: % Are null points common enough to explain the ubiquity of jets?
106:
107: Coronal jets are taken as another observational manifestation of
108: coronal null points. Observations in EUV and X-rays show that jets,
109: with their characteristic apex cusp \citep{Shibata1992,Shimojo1996},
110: appear to be more common in coronal holes and the quiet Sun
111: than previously believed \citep{Culhane2007,Cirtain2007}.
112: Theoretical models of jets invoke magnetic reconnection occurring
113: at a null point many megameters above the photospheric surface
114: \citep{Yokoyama1996,MorenoInsertis2007}. In order that such a model apply
115: to the large number of jets observed it would be
116: necessary that magnetic null points are relatively common at the
117: altitudes of observed cusps.
118:
119: % Complicated magnetic fields
120: % field generally believed to get simpler w/ height:
121: % beta decreases, influence of small photospheric structuring
122: % diminishes
123:
124: In many models, coronal null points occur above a photospheric field
125: consisting of one magnetic element completely surrounded
126: by a ring of opposing polarity \citep{Antiochos1998,MorenoInsertis2007}.
127: The null point will persist even if the continuous surrounding ring
128: is broken, but it remains
129: necessary that opposing polarity be found on ``all sides'' of the
130: central element \citep{Seehafer1986,Beveridge2004}. This suggests
131: that coronal null points occur only under special circumstances which
132: are likely to be found infrequently in the actual solar photosphere, and
133: raises once more the question of how common null points might be
134: in the actual coronal magnetic field.
135:
136: % QS models: Schriver & Title, point sources.
137: % LBP03: general theory assuming homogeneous Gaussian field
138: % arising from p-spheric spectrum. Density fell rapidly w/ height.
139: % for flat spectrum, such as from white noise or point sources
140: % \sim z^{-3}.
141:
142: One recent investigation by R\'egnier, Parnell and Haynes (2008)
143: \nocite{Regnier2008} found 80 magnetic null
144: points over a $102\,{\rm Mm}\times116\,{\rm Mm}$ patch of quiet Sun. A
145: potential field was extrapolated onto a rectilinear grid from a
146: magnetogram made by Hinode's NFI instrument. The algorithm of
147: Haynes and Parnell (2007)
148: \nocite{Haynes2007} was then applied to the gridded field to locate all
149: points where an interpolated field would vanish. This revealed that,
150: at least over that region at that time, there was a column of
151: $N_n=6.7\times 10^{-3}$ null points per square megameter, about half
152: above a height of $z=1$ Mm.
153: Close, Parnell and Priest (2004)
154: \nocite{Close2004b} extrapolating from high-resolution MDI
155: magnetograms of a different day found $1.7\times10^{-3}\,{\rm
156: Mm}^{-2}$, almost four times fewer. This discrepancy suggests a
157: potential dependence on instrument and specific date of observation.
158: A more general measurement of null point density, applicable to any
159: time or any region, would require the repetition of this process.
160:
161: Several investigations have estimated the density of magnetic null
162: points in generic quiet Sun coronal fields. Schrijver and Title (2002)
163: \nocite{Schrijver2002}
164: modeled the photospheric field of the quiet Sun as a balanced, random
165: distribution of magnetic point charges. They found that the potential
166: field anchored to these charges vanished at one point
167: in the corona, on average, for every eleven point charges on the
168: photosphere. They estimated that the density of null points
169: diminished exponentially with height.
170:
171: Longcope, Brown and Priest (2003, hereafter \papi)
172: \nocite{Longcope2003b} developed a more general approach to estimating
173: the density of coronal null points in a potential
174: coronal field. They assumed the corona had a random, Gaussian
175: magnetic field, anchored to a statistically homogeneous photospheric
176: field with a known power spectrum $S(k)$. The density of null points
177: at a given height, $\rho_N(z)$, scaled with the inverse cube of the
178: characteristic scale length at that height. This was expressed in
179: terms of integrals of the power spectrum,
180: $S(k)$, as well as on the mean
181: of the vertical field, $\bar{B}_z$. At lower heights it
182: depended on larger wave numbers (smaller scales). A perfectly
183: flat spectrum, such as from ideal white noise or a distribution of
184: point magnetic charges, introduces no length scale of its own. In
185: this case the density decreases with height as
186: \be
187: \rho_N(z) ~=~ {0.040\over z^3} ~~,
188: \ee
189: independent of the strength or other properties of the photospheric
190: field.
191:
192: % show here that this spectral estimate is relatively accurate for
193: % QS. & that QS spectra are relatively invariant. one
194:
195: The spectrum from a magnetogram can be used in the spectral estimate
196: of \papi\ to compute the distribution of nulls above
197: observed regions. Its assumption of randomness and homogeneity are
198: likely to be fairly valid in quiet Sun regions. In this work we quantify the
199: degree to which a potential field extrapolation satisfies the other
200: assumption, that the field components are Gaussian random variables.
201: We also test the estimate against the distribution of null points
202: found directly in an extrapolated field. We find evidence that the
203: spectral estimate gives a reasonable approximation of the number of
204: magnetic null points above a region of quiet Sun, especially above
205: 1,500 km from the magnetogram surface.
206:
207: The spectra of MDI magnetograms reveal several factors which could
208: affect the number of null points found either directly or though
209: spectral estimate. There is a level of white noise which, while
210: reduced by averaging, is present in any magnetogram. We find that
211: removing its contribution from a spectrum reduces the number of null
212: points by as much as 85\%. The defocusing of the telescope, partly on
213: purpose in the low-resolution mode, steepens the spectrum, causing an
214: underestimation of the number of null points. Both of these effects can
215: be removed from the power spectrum, $S(k)$, yielding a spectral
216: estimate of null points less affected by instrumental artifacts.
217:
218: We perform spectral estimates on 562 MDI magnetograms of
219: quiet Sun during the two solar minima it has observed
220: (1996--1998 and 2006--2007). After compensating
221: for the effects mentioned above, we find the number of coronal null
222: points above 1.5 Mm to be fairly constant:
223: $N_n=3.1\times10^{-3}\pm3.0\times10^{-4}\,{\rm Mm}^{-2}$
224: That is, on average, one such null point
225: over every $322\,{\rm Mm}^2$ patch of quiet
226: Sun. When the Sun is completely free of active regions this amounts to
227: about $20,000$ null points above the solar surface.
228:
229: The magnetic spectrum of the quiet Sun is found to be remarkably
230: consistent over the periods investigated. The typical spectrum is
231: consistent with a potential field from magnetic point sources located
232: about 1.4
233: Mm {\em below} the photospheric surface. The variance and kurtosis of
234: the magnetograms can be used to infer the size and areal density of
235: these sources. Assuming them to have an exponential distribution of
236: fluxes we find the sources have mean flux,
237: $\avg{|\phi|}=1.0\times10^{19}$ Mx and areal
238: density $n=0.007\,{\rm Mm}^{-2}$.
239:
240: The paper is structured as follows. In the next section we assess the
241: degree to which potential fields, extrapolated from two particular MDI
242: magnetograms, satisfy the assumptions of the \papi\
243: spectral estimate. We then directly locate all of the null points in
244: extrapolations from these magnetograms and compare the results to the
245: spectral estimate. In \S 3 we apply the spectral
246: estimate to 562 MDI magnetograms spanning solar minimum periods in
247: 1996--1998 and 2006--2007. We find that the number of coronal null
248: points stays at a relatively constant level, with $\sim10\%$
249: day-to-day variation. In \S 4 we show how the generic spectrum of
250: the quiet Sun photosphere may be interpreted as a random
251: super-position of identical magnetic elements. We infer the
252: properties of the elements and their distribution from the Fourier
253: spectrum. Once again, these properties appear relatively constant
254: over time. Finally, we discuss what our findings imply about null
255: points in the actual coronal field above quiet Sun.
256:
257: \section{Distribution of Nulls}
258:
259: \subsection{The Gaussianity of the extrapolated field}
260:
261: The basis for our computation will be magnetograms from regions
262: of quiet Sun, near disk center. Figure \ref{fig:mg} shows two examples
263: of such magnetograms, each made by averaging five one-minute cadence
264: MDI line-of-sight magnetograms.
265: The field of view is $300''\times300''$ centered at disk center. The
266: left and right images are from high-resolution and low-resolution data
267: obtained 1.75 hours apart.
268:
269: \begin{figure}[htp]
270: \centerline{\psfig{file=fig1a.ps,width=2.55in}%
271: \psfig{file=fig1b.ps,width=2.55in}}
272: \caption{Magnetograms of a quiet sun area from 23 Feb.\ 1997, made by
273: averaging five consecutive MDI magnetograms made at one-minute cadence
274: The grey scale shows the line of sight field between $\pm100$ Gauss.
275: (a) High resolution magnetograms and (b) Low resolution magnetograms
276: from 1:45 later.}
277: \label{fig:mg}
278: \end{figure}
279:
280: The spectral estimate of \papi\
281: %\citet[hereafter called \papi]{Longcope2003b}
282: yields the spatial density
283: of null points in a coronal magnetic field satisfying
284: certain conditions. It assumes the coronal field to be potential and
285: to be anchored to a random photospheric field whose statistics are
286: homogeneous and Gaussian. These conditions on the photospheric field
287: (the magnetogram) lead to a coronal field whose components are
288: statistically homogeneous and Gaussian.
289: From the properties of this field
290: \papi\ reports the mean density of null points at any height in the
291: corona.
292:
293: Extrapolating upward
294: from the magnetogram of \fig\ \ref{fig:mg}a to a particular height $z$
295: results in a new field, $B_z(x,y,z)$, whose histogram we find to be
296: nearly Gaussian. Figure
297: \ref{fig:vgp} shows histograms from several heights as solid lines, and
298: the Gaussians best fitting their central portions as dashed curves.
299: The cumulative distribution from a range of values was fitted, and
300: the range was iteratively adjusted until it
301: contained exactly two standard
302: deviations to either side. The standard deviation of this fit,
303: $\sigma_z^{(g)}$, is indicated by triangles and
304: plotted on axes to the left. The
305: standard deviation of the entire field, $\sigma_z^{(m)}$, plotted
306: with a dashed line on the same axis is systematically greater than
307: that of the core.
308:
309: \begin{figure}[htp]
310: %\epsscale{1.0}
311: \psfig{file=fig2.ps,width=5.1in}
312: \caption{Histograms of the field $B_z$ computed by extrapolating the
313: photospheric magnetograms to five different heights. The height, in
314: Mm, of the slice corresponds to the level of the bottom of the
315: histogram on the far left axis. A Gaussian fit to the central
316: region is shown as a dashed line, with triangles at
317: $\pm\sigma_z^{(g)}$. These values are also plotted against a horizontal
318: axis on the left. A solid curve shows the full functional form of
319: $\sigma_z^{(g)}(z)$. A dashed line shows $\sigma_z^{(m)}(z)$ on the
320: same axis (squares occur at heights coincident with the histograms).}
321: \label{fig:vgp}
322: \end{figure}
323:
324: It can be seen in \fig\ \ref{fig:vgp}, that the field $B_z$ has an
325: approximately Gaussian distribution even well above the photospheric
326: level ($z=0$). At the bottom, the Gaussian component has a standard
327: deviation, $\sigma_z^{(g)}\simeq11.3$ Gauss, comparable to the
328: intrinsic noise in a five-minute average of MDI magnetograms
329: \citep{Liu2001}; it decreases with height as the
330: width of the Gaussian narrows. The actual
331: distribution departs from the Gaussian fit principally by an excess
332: (tail) mostly outside $\pm\sigma_z^{(g)}$.
333: Histograms made from the horizontal components of the extrapolated
334: fields at varying heights look very similar to those of the vertical
335: field shown in \fig\ \ref{fig:vgp}.
336:
337: The Gaussian fit to the central core of each
338: distribution can be used to quantify the degree to which the actual field
339: is a Gaussian variable. The amplitude of the Gaussian component ranges from
340: $0.97$ at $z=0$ to $0.81$ higher up (see \fig\
341: \ref{fig:sig}b). That is to say the field is between $81\%$ and $97\%$
342: Gaussian.
343:
344: \begin{figure}[htp]
345: %\epsscale{0.9}
346: \psfig{file=fig3.ps,width=5.1in}
347: \caption{(a) Standard deviations from all three components of the
348: potential extrapolation of the magnetogram \fig\ \ref{fig:mg}a. The
349: solid and dashed curves show $\sigma^{(g)}$ and $\sigma^{(m)}$
350: respectively for $B_x$ (crosses), $B_y$ (triangles) and $B_z$
351: (squares), plotted versus height. The symbols occur at the same
352: heights as the histograms from \fig\ \ref{fig:vgp}. Dotted
353: lines show the standard deviation for vertical
354: (upper) and horizontal (lower) field components extrapolated
355: from white noise.
356: (b) The fraction of each field which is fitted by the Gaussian within
357: the central 4$\sigma^{(g)}$.}
358: \label{fig:sig}
359: \end{figure}
360:
361: While a significant component of the magnetic field has a Gaussian
362: distribution, it is not spatial white noise.
363: Longcope, Brown and Priest (\papi) \nocite{Longcope2003b} show that a
364: potential field extrapolated from a
365: magnetogram with isotropic power spectral density (psd) $S(k)$
366: has vertical field with variance
367: \be
368: \sigma_z^2(z) ~=~ 2\pi \int_0^{\infty} S(k)e^{-2kz}\, k\, dk
369: ~=~ 2\sigma_x^2(z) ~=~ 2\sigma_y^2(z)~~.
370: \label{eq:sig}
371: \ee
372: Spatial white noise is uncorrelated from pixel to pixel and therefore has a
373: flat psd
374: \be
375: S(k)\simeq{\sigma_0^2\over\pi k_c^2}\Theta(k_c-k) ~~,
376: \label{eq:Sk_noise}
377: \ee
378: out to the resolution limit $k_c=2\sqrt{\pi}/\Delta x$ for pixel size
379: $\Delta x$ ($\Theta$ is the Heaviside function). Using this in
380: (\ref{eq:sig}) gives the variance due to spatial white noise alone
381: \be
382: \sigma^2_z(z) ~=~ {\sigma_0^2\over8\pi (z/\Delta x)^2}\,
383: \left[ 1 - (1+4\sqrt{\pi}z/\Delta x )
384: e^{-4\sqrt{\pi}z/\Delta x} \right] ~~.
385: \label{eq:sig_noise}
386: \ee
387: This functional form, plotted as a dotted line for each component of
388: the field in \fig\ \ref{fig:sig}, falls off much faster than either
389: $\sigma^{(g)}$ or $\sigma^{(m)}$. We therefore conclude that the
390: Gaussian component consists of contributions with spatial
391: correlations longer than a single pixel: it is not white noise.
392:
393: It is common practice to assess the degree to which a variable departs
394: from Gaussian by its kurtosis statistic, $K$. The kurtosis is the
395: fourth moment divided by the square of the variance, less three;
396: when the variable is perfectly Gaussian $K=0$.
397: Excesses, relative to a Gaussian, in
398: the tails of the distribution will elevate the fourth moment and lead
399: to a positive kurtosis. Figure \ref{fig:kurt} shows that even when
400: the excess tails account for 3\% of the distribution, they lead to a
401: large kurtosis.
402:
403: \begin{figure}[htp]
404: %\epsscale{0.6}
405: \psfig{file=fig4.ps,width=4.5in}
406: \caption{The kurtosis of the fields for $B_x$ (crosses), $B_y$
407: (triangles) and $B_z$ (squares), plotted versus height. Solid and
408: dashed lines are for the high resolution and low resolution
409: magnetograms respectively. The dotted lines show the results for
410: point magnetic sources, taken from \eq\ (\ref{eq:kurt_mpole}).}
411: \label{fig:kurt}
412: \end{figure}
413:
414: For comparison, a distribution of $n$ point sources per unit surface
415: area has a flat power spectrum with no cut-off
416: ($k_c\to\infty$) whose variance is (\papi)
417: \be
418: \sigma_z^2(z) ~=~{n\avg{\phi^2}\over 8\pi z^2} ~~,
419: \label{eq:sig_mpole}
420: \ee
421: where $\avg{\varphi^2}$ is the variance in the source fluxes.
422: The kurtosis of this same photospheric field is
423: \be
424: K_z ~=~ {4\over 5\pi}\,{\avg{\varphi^4}\over\avg{\varphi^2}^2}\,
425: {1\over nz^2} ~=~ 4K_x ~=~ 4K_y ~~.
426: \label{eq:kurt_mpole}
427: \ee
428: Both of the expressions above diverge at the photosphere ($z\to0$) as a
429: consequence of the very strong fields in the neighborhood of a
430: point source. This divergence is one element leading to the failure
431: of the spectral estimate at low heights, $z<n^{-1/2}$ (\papi).
432:
433: It is evident from \fig\ \ref{fig:kurt} that the
434: kurtosis of the actual field is generally large but
435: remains finite even as $z\to0$. The large magnitude arises from the
436: tails in the pdf corresponding to those pixels of the magnetogram in
437: magnetic elements, where $|B_z|>5\sigma_z(0)\simeq50$ Gauss. These are the
438: excess tails of the distribution function, but they are also the
439: structures which generate field above a few megameters.
440:
441: \subsection{Assessing the accuracy of the spectral estimate}
442:
443: A spectral estimate of the magnetic null point density
444: is found from the isotropic power spectrum,
445: $S(k)$, of the photospheric magnetogram. The spectrum determines the
446: variance, as given by \eq\ (\ref{eq:sig}), and a related quantity
447: \be
448: q^2(z) ~=~ {2\pi \over \sigma_z^2(z)}
449: \int_0^{\infty} S(k)e^{-2kz}\, k^3\, dk ~~.
450: \label{eq:q2}
451: \ee
452: Provided the field has Gaussian statistics the density of null points
453: in it is found to be
454: \be
455: \rho_N(z) ~=~ G(\bar{B}_z/\sigma_z)\, q^3(z) ~~,
456: \label{eq:rho}
457: \ee
458: where $G(\zeta)$ is a dimensionless function defined in \papi.
459: The function depends on the mean vertical field strength $\bar{B}_z$ and also
460: weakly on another moment of the spectrum.
461: Provided the polarities are sufficiently balanced that
462: $\bar{B}_z\ll\sigma_z$ over the relevant range of heights, we find
463: $G\sim 2\times10^{-2}$ for most spectra of quiet Sun magnetograms.
464:
465: We assess the accuracy of the approximation offered by the spectral
466: estimate by direct comparison. Both magnetograms from \fig\
467: \ref{fig:mg} are extrapolated onto a rectilinear grid with periodic
468: lateral boundaries. Between the grid points the magnetic field is
469: defined using tri-linear interpolation. A magnetic null point is the
470: point where all three components of this interpolated field vanish
471: together. These null points are located using the algorithm of
472: Haynes and Parnell (2007).
473: \nocite{Haynes2007} The number of null points found above a given
474: height $z$ is divided by the photospheric area to yield a null-point
475: column density, $N_n(z)$. The right and left axes of \fig\ \ref{fig:nd}b
476: shows the number and column densities found this way, respectively.
477:
478: \begin{figure}[htp]
479: %\epsscale{1.0}
480: \psfig{file=fig5.ps,width=5.1in}
481: \caption{The density (a) and column density (b)
482: of null points in potential fields
483: extrapolated from the MDI magnetograms in \fig\ \ref{fig:mg}. Dark and
484: light curves correspond to the high resolution and low resolution
485: data respectively. The solid curves are computed by finding
486: the nulls on a gridded extrapolation; the dashed curves are the
487: spectral estimates. Symbols appear at heights of one, two, three and
488: four horizontal pixel sizes ($z=\Delta x$,
489: $2\Delta x$, $3\Delta x$ and $4\Delta x$).
490: For panel (b) the left axis reports the column density in nulls per
491: Mm$^2$ above a given height, while the right axis gives the total number
492: over the entire magnetogram.}
493: \label{fig:nd}
494: \end{figure}
495:
496: There are more null points in the high resolution case (dark curves)
497: than in the low resolution (6135 {\em vs.} 664).
498: The majority of the excess null points lie in a layer about one Mm
499: thick. Over heights above one Mm the predicted null columns (dashed
500: curves on \fig\ \ref{fig:nd}b) from the high resolution and low
501: resolution cases are quite similar. The greatest disagreement between
502: the actual null column, $N_n(z)$, and the spectral estimate occurs
503: within this same one Mm bottom layer. Above this height the two estimates
504: are within a factor of two of reality for both resolutions. Since the
505: bottom layer of the high resolution field has the largest kurtosis it
506: is perhaps unsurprising that the greatest
507: disagreement with the spectral estimate occurs there.
508: Perhaps more surprising is
509: that kurtosis as large as 40 (the maximum of the low resolution data)
510: leads to disagreement by not more than a factor of two.
511:
512: \section{The quiet Sun null column in general}
513:
514: \subsection{Comparing nulls from different magnetograms}
515:
516: \label{sec:psd}
517:
518: The extent to which, for levels above one Mm, the spectral estimates
519: of high resolution and low resolution agree can be understood
520: from a comparison of their power spectra $S(k)$, as shown in
521: \fig\ \ref{fig:spec}. A fast Fourier transform (FFT) of the
522: $L_x\times L_y$ magnetogram returns an array of complex
523: coefficients, $\hat{B}_{m,n}$, for the expansion
524: \be
525: B_z(x,y) ~=~ \sum_{m,n}\hat{B}_{m,n}\,
526: \exp[ \,i\,m\Delta k_x x + i\,n\Delta k_y y \,] ~~,
527: \ee
528: where $\Delta k_x=2\pi/L_x$ and $\Delta k_y=2\pi/L_y$. The
529: quiet-Sun magnetograms have very few features at their edges which
530: might suffer discontinuity under periodic repetition so we find it
531: unnecessary to employ any windowing
532: function \citep{Press1986}.
533:
534: \begin{figure}[htp]
535: %\epsscale{1.0}
536: \psfig{file=fig6.ps,width=5.1in}
537: \caption{The isotropized spectra for $300''\times300''$ magnetograms
538: of quiet Sun centered at disk center on 23 Feb.\ 1997. Curves of
539: different colors are $S(k)$ for 30-second MDI exposures at high (green)
540: and low (red) resolution; 5-minute averages at high (magenta) and
541: low (black) resolution; and Kitt Peak magnetogram (blue). The inset
542: is an expansion of the region at low wavenumbers.}
543: \label{fig:spec}
544: \end{figure}
545:
546: The two-dimensional power spectrum at
547: wave number $\kvec=m\Delta k_x\xhat+n\Delta k_y\yhat$ is
548: \be
549: S(\kvec) ~=~ {|\hat{B}_{m,n}|^2\over \Delta k_x\Delta k_y} ~~.
550: \label{eq:Sk}
551: \ee
552: This function is defined only on points of the Fourier-space grid, but
553: approximates a continuous function. The denominator in
554: \eq\ (\ref{eq:Sk}) eliminates from $S(\kvec)$ any dependance on pixel
555: size or field of view.
556:
557: The isotropized power spectrum, $S(k)$, is computed by averaging all values
558: of $S(\kvec)$ within an annulus of Fourier-space.
559: We use annuli of radii extending to a
560: maximum wave number $k_c=2\sqrt{\pi}/\Delta x$,
561: chosen so that the total Fourier-space area, $\pi k_c^2$, matches
562: that occupied by the actual magnetogram.
563: The cut-offs for high and
564: low resolution MDI magnetograms are therefore
565: $8.2$ rad/Mm and $2.5$ rad/Mm respectively since these have
566: pixel sizes of $\Delta x=0.43$ Mm and $\Delta x=1.42$ Mm in
567: this particular magnetogram.
568:
569: The spectrum in \eq\
570: (\ref{eq:Sk}) is normalized so that,
571: \[
572: \avg{B_z^2} ~=~ \sum_{m,n}|\hat{B}_{m,n}|^2\, ~\simeq~
573: \int S(\kvec)\, d^2k ~=~2\pi\int_0^{k_c}S(k)\, k\, dk ~~,
574: \]
575: where the angled brackets are a spatial average.
576: The variance is found from the same expression but
577: excluding from the sum $(m,n)=(0,0)$. The variance at height $z$
578: above the magnetogram plane, \eq\ (\ref{eq:sig}), includes an
579: exponential factor from the potential field extrapolation.
580:
581: Each spectra from an MDI magnetogram in \fig\ \ref{fig:spec} has
582: a range of white noise over its highest wave numbers.
583: We estimate its level, $S_n$, from the median value
584: of $S(k)$ over the highest 15 spectral
585: bins. The noise level is reduced by exactly $1/5$ in the
586: averages strongly suggesting an origin in Poisson counting
587: noise, uncorrelated between frames \citep{Liu2001,DeForest2007}.
588: The flat noise floors contribute
589: $\sigma^2_z(0)=\pi k_c S^2_n = (8.8\,{\rm G})^2$ and
590: $(6.3\,{\rm G})^2$ to the low and high-resolution
591: variances respectively. We noise-correct each spectrum by
592: subtracting from the entire curve
593: the flat noise floor, $S_n$, determined from the top 15
594: spectral bins.
595:
596: Each of the noise-corrected spectra appears to have been
597: steepened by
598: the modulation transfer function (MTF), $M(k)$,
599: of its imaging system \citep{Born1980}, and must be corrected to
600: $S(k)/M^2(k)$. The most conservative correction to the
601: high-resolution spectrum is to assume it to be diffraction-limited and
602: take $M(k)=M_d(k)$, given in an appendix. MDI is intentionally
603: defocussed in its low-resolution mode, leading to an MTF we
604: approximate as
605: $M^2(k) = M_d^2(k)/( 1 + \ell_0^4k^4 )$ where $\ell_0\simeq1.6''$ at
606: this time (see appendix). Performing these corrections restores all
607: four MDI spectra to similar forms, over low wave-numbers,
608: as shown in \fig\ \ref{fig:spec_mtf}.
609: The Kitt Peak spectrum (blue) remains
610: below all others since it has not been divided by its MTF, part of
611: which presumably arises from atmospheric seeing
612: \citep{Hufnagel1964,Abramenko2001}.
613:
614: \begin{figure}[htp]
615: %\epsscale{1.0}
616: \psfig{file=fig7.ps,width=5.1in}
617: \caption{The same spectra from \fig\ \ref{fig:spec} after removal of
618: white noise and correction for the MTF. No corrections are
619: performed on the Kitt Peak spectra (blue) which is retained for
620: reference. Dashed curves show the high resolution spectra
621: before it was corrected for the MTF, but after white noise was removed.}
622: \label{fig:spec_mtf}
623: \end{figure}
624:
625: There is, over the range $k>1.5\,{\rm rad/Mm}$, an excess in
626: high-resolution power, relative to low-resolution; this excess is
627: further amplified by the MTF correction. The excess might
628: be partly an artifact from the granular motions
629: during the brief intervals between different filter-gram exposures
630: \citep{DeForest2007}. Its origin, whatever that may be, is evidently
631: correlated several-fold longer than five minutes,
632: since averaging diminishes it by less than a factor of five.
633: We next show that this power does not significantly affect
634: the null densities at heights where potential field extrapolation is
635: warranted.
636:
637: The spectra from \fig\ \ref{fig:spec} are used in the Laplace
638: transform integrals, (\ref{eq:sig}) and (\ref{eq:q2}), from which are
639: computed the null density $\rho_N(z)$, according to \eq\ (\ref{eq:rho}).
640: This density is integrated downward from some
641: upper height yielding the null column density, $N_n(z)$, plotted in
642: \fig\ \ref{fig:nd2-23}. The exponential factor, $e^{-2kz}$, in each
643: of the spectral integrals means that the null density at a given
644: height $z$ depends principally on the spectrum over wave numbers
645: $k<1/z$. Thus heights above $z=1$ Mm depend mostly on the spectrum
646: below (to the left of) $k=1\,{\rm rad/Mm}$. As discussed above, and
647: evident in \fig\ \ref{fig:spec_mtf}, all MDI spectra are fairly similar
648: over this range. It is therefore natural that all five solid
649: curves in \fig\ \ref{fig:nd2-23} follow similar paths above one Mm.
650:
651: \begin{figure}[htp]
652: %\epsscale{1.0}
653: \psfig{file=fig8.ps,width=5.1in}
654: \caption{Spectral estimates of the null column densities, $N_n(z)$,
655: computed from each of the spectra in \figs\ \ref{fig:spec}
656: and \ref{fig:spec_mtf} using the same color
657: code. Dashed curves are from
658: the original spectra (\fig\ \ref{fig:spec_mtf});
659: solid curves are from spectra
660: from which white noise has been subtracted and which
661: have been divided by the MTF (\fig\ \ref{fig:spec_mtf}).
662: Dotted lines show the curves after noise subtraction without MTF
663: correction.}
664: \label{fig:nd2-23}
665: \end{figure}
666:
667: Noise contributes significant power to the highest wavenumbers,
668: creating a population of artificial null
669: points in an extrapolation.
670: The null column densities after subtraction of this noise,
671: the dotted curves in \fig\
672: \ref{fig:nd2-23}, lie below the corresponding uncorrected
673: (dashed) curves. The fraction of nulls attributable to noise alone is
674: lower in five-minute averages than one-minute magnetograms, since the
675: former have lower noise. The effect is most pronounced in one-minute
676: low-resolution magnetograms, for which $85\%$ of all nulls are due to
677: noise.
678:
679: The small-scale structure revealed in high resolution magnetograms
680: translates into complex fields with high null densities at the
681: lowest level of the potential magnetic field. While potential fields are
682: frequently used to approximate the quiet Sun corona, they are of
683: questionable validity in the
684: lowest layers, where the plasma density and pressure are liable to
685: play their most significant role. Above a height of 1,500 km (1.5
686: Mm) the potential field
687: approximation may be on better footing. It appears from \fig\
688: \ref{fig:nd2-23} that most of the estimates based on potential field
689: extrapolation, are in reasonable agreement above this particular
690: height (at $z=1$ Mm, for example, the high and low resolution
691: estimates appear to disagree to a greater extent). For these reasons
692: we hereafter focus on estimating the total column of null points above
693: $z=1.5$ Mm, denoted $N_n(1.5\,{\rm Mm})$.
694:
695: \subsection{The generic behavior}
696:
697: The foregoing sections have considered different magnetograms of the
698: same region of quiet Sun (23 Feb.\ 1997). In order to generalize our
699: conclusions we perform similar analysis on magnetograms spanning solar
700: minimum from 2006 and 2007. The same central $300''\times300''$ sub-image
701: is extracted from one full-disk (low resolution)
702: daily MDI magnetogram every second day.
703: It is analyzed provided there is no obvious active region
704: field, or preponderance of one polarity, and provided there is less
705: than $8\times 10^{19}$ Mx in pixels above 200 G. This value was
706: selected after some trial and error to eliminate plage regions whose
707: presence would violate our assumption of homogeneity. Separate
708: analyses are performed for one-minute daily magnetograms
709: and for those composed of 5-minute averages.
710:
711: The isotropized spectra from all quiet Sun areas turn out to be very
712: similar to one another, closely resembling the example from the previous
713: section. Figure \ref{fig:med_spec} shows a median spectrum: the
714: the median value, at each wave-number, from over 110
715: quiet Sun spectra of each low-resolution kind.
716: The upper and lower quartile points, bracketing half of all values,
717: are plotted as
718: lighter curves. This clearly shows that the most significant variation
719: occurs at wavenumbers below 0.2 rad/Mm (i.e.\ wavelengths longer
720: than 30 Mm).
721:
722: \begin{figure}[htp]
723: %\epsscale{0.9}
724: \psfig{file=fig9.ps,width=5.1in}
725: \caption{Median spectra from magnetograms in 2006 and 2007.
726: (a) The median spectrum
727: (dark) and the upper and lower quartiles (light curves) from the
728: five-minute averages. (b) The median from 118
729: one-minute magnetograms (upper curve) and 114
730: 5-minute averages (lower curve). The lighter curves show the results
731: of subtracting white noise (lower pair) and then dividing by the MTF (upper
732: pair).}
733: \label{fig:med_spec}
734: \end{figure}
735:
736: Removing the white noise floor by first
737: subtracting it and then extrapolating the noisier portions
738: yields the lower light spectra in \fig\
739: \ref{fig:med_spec}b. This is then divided by the empirically
740: determined MTF, $M^2(k)=M_d^2(k)/(1+\ell_0^4k^4)$, where $\ell_0=1.04''$ as
741: described in the appendix. The results are the two upper light
742: curves.
743:
744: The column density found from these various versions of the median
745: spectra are shown in \fig\ \ref{fig:med_den}. The solid curves of
746: each shade are the result of removing noise and correcting for
747: the MTF. The broken
748: curves above these are from the raw spectra, while the dashed
749: curves below are after noise has been removed
750: but no MTF correction performed.
751: Symbols mark the values of these at $z=1.5$ Mm:
752: $3.0\times10^{-3}$ Mm$^{-2}$ and $2.9\times10^{-3}$
753: Mm$^{-2}$ respectively.
754:
755: \begin{figure}[htp]
756: \psfig{file=fig10.ps,width=5.1in}
757: \caption{The column densities from each of the median spectra in \fig\
758: \ref{fig:med_spec}b. Dark and light curves are for one-minute magnetograms
759: and five-minute averages respectively. Solid curves are from the
760: entire spectra and dashed curves are from the noise-corrected
761: spectra. A dotted curve is an empirical fit.}.
762: \label{fig:med_den}
763: \end{figure}
764:
765: The median spectra are clearly very close to pure exponentials,
766: $S(k)\sim e^{-2kd}$. This functional form suggests the empirical model
767: for the null column density of
768: one-minute magnetograms (dotted line)
769: \be
770: N_n(z) ~\simeq~ {0.021\over (z+d)^2} ~~,
771: \label{eq:LBP_col}
772: \ee
773: where the numerator and $d=1.6$ Mm are found from fitting.
774:
775: The null column above a given height does vary between magnetograms, as
776: shown in \fig\ \ref{fig:smry}. There appear to be
777: random fluctuations superimposed on a much slower variation. We
778: estimated the
779: dispersion about the trend, $2.7\times10^{-4}\,{\rm Mm}^{-2}$
780: for the one-minute magnetograms and $2.5\times10^{-4}\,{\rm Mm}^{-2}$ for
781: five-minute averages, by subtracting a 90-day running mean.
782: Considering the entire data set of one-minute
783: magnetograms, including its slow trend, we find the column of null
784: points above $z=1.5$ Mm to be
785: \[
786: N_n(1.5\, {\rm Mm}) ~=~ (3.1\pm 0.3)\times10^{-3}\, {\rm Mm}^{-2} ~~.
787: \]
788: (The errors quoted above represent rms variation; the error in the mean is far smaller.)
789: The value from five-minute magnetograms is slightly
790: lower: $(2.9\pm 0.3)\times10^{-3}$. We believe this value is lower
791: partly because of motional blurring during the averaging interval,
792: so we adopt the
793: one-minute-magnetogram value as the more accurate of the two.
794:
795: \begin{figure}[htp]
796: %\epsscale{1.0}
797: \psfig{file=fig11.ps,width=5.1in}
798: \caption{The column of null points above $z=1.5$ Mm found from quiet
799: sun magnetograms spanning 2 years of solar minimum.
800: Triangles and crosses denote
801: one-minute and five-minute low-resolution magnetograms respectively;
802: squares are from high-resolution magnetograms. Error
803: bars on the right show one standard deviation of scatter about a
804: 90-day running mean. Solid lines along the bottom show the reported
805: focus position of the MDI telescope.}
806: \label{fig:smry}
807: \end{figure}
808:
809: We believe the slow variation in $N_n(1.5)$ is due to gradual changes
810: in the MTF of the imaging system.
811: The reported focus position of the
812: instrument is shown as a solid curve long the bottom. The focus is
813: periodically changed to compensate for changes in the instrument
814: itself. Our
815: estimate of the MTF was performed through cross-comparison to the high
816: resolution images, plotted as squares, all
817: while the instrument was in focus position 5. This is the MTF by
818: which we are correcting the spectrum, so we believe that the value in
819: the middle of this interval is probably the most accurate.
820: The MTF correction raises the low resolution columns by about 40\%,
821: thereby making them more consistent with high resolution values.
822: Omitting that step gives null columns which could be considered lower
823: bounds on the actual values: medians are
824: $N_n(1.5)=2.2\times10^{-3}\,{\rm Mm}^{-2}$
825: for one-minute
826: and $2.1\times10^{-3}\,{\rm Mm}^{-2}$ for five-minute.
827:
828: A comparable plot from the previous solar minimum, \fig\
829: \ref{fig:smry96}, shows more pronounced evolution on long time scales. This
830: data comes from the very beginning of the SOHO mission, when the
831: instrument was probably changing most rapidly. There is a very
832: large change in
833: Nov. 1997 coincident with the refocussing of the telescope. The
834: later refocussing events in \fig\ \ref{fig:smry} produce much less
835: obvious changes, either because they are more frequent or because
836: there is less dramatic change between them. Due to the
837: larger, and more rapid change we did not attempt an estimate of the
838: MTF. Instead we used the empirical form from the later phase, with
839: $\ell_0=1.6''$ found from fitting the 23 Feb.\ 1997 data described
840: above. This restores the typical values to the same range as they
841: were in the 2006/2007 data set.
842:
843: \begin{figure}[htp]
844: %\epsscale{1.0}
845: \psfig{file=fig12.ps,width=5.1in}
846: \caption{Null columns during the solar minimum from 1996-1998. Plot
847: format is the same as in \fig\ \ref{fig:smry}.}
848: \label{fig:smry96}
849: \end{figure}
850:
851: \section{Modeling the quiet Sun spectrum}
852:
853: \subsection{A synthetic magnetogram}
854:
855: Some insight into the origins of the quiet sun magnetic field may be
856: gleaned from its spectrum. Consider a model photospheric
857: field composed of distinct elements, each characterized by the same
858: normalized shape function, $f(\xvec)$. Elements at locations,
859: $\xvec_j=(x_j,y_j)$,
860: with fluxes, $\phi_j$, will create a vertical field
861: \be
862: B_z(\xvec) ~=~\sum_j\phi_j\,f(\xvec-\xvec_j) ~~,
863: \label{eq:composite}
864: \ee
865: in the photospheric plane.
866: For simplicity let us assume that the fluxes are random,
867: have zero mean, and are uncorrelated:
868: $\avg{\phi_i\phi_j}=\avg{\phi^2}\delta_{ij}$. We also assume that the
869: fluxes are independent of their locations $\xvec_j$ which are distributed
870: randomly with areal density $n$. Under these assumptions the power
871: spectrum of the photospheric field is
872: \be
873: S(\kvec) ~=~ n\avg{\phi^2}\,|\hat{f}(\kvec)|^2 ~~,
874: \ee
875: where $\hat{f}(\kvec)$ is the Fourier transform of the shape function.
876:
877: A purely exponential spectrum can be interpreted as being
878: composed of individual flux elements each with an exponential Fourier
879: transform such as
880: \be
881: f(\xvec) ~=~ {1\over2\pi}{d\over (|\xvec|^2 + d^2)^{3/2}} ~=~
882: \int {e^{-|\kvec|d}\over(2\pi)^2} \, e^{i\kvec\cdot\xvec}d^2 k ~~.
883: \label{eq:shape}
884: \ee
885: This function also happens to be the vertical potential
886: field due to a unit point charge located at $z=-d$; it is the
887: potential field Green's function.
888:
889: The observed spectrum, before MTF-correction, is very close to an
890: exponential, $S(k)\simeq e^{-2kd}$. The exponent is thus directly
891: related to a ``depth'' at which point sources are submerged. On the
892: other hand, the spectrum after correction,
893: $S(k)\simeq e^{-2kd}/M^2(k)$, has an inverse Fourier transform
894: which is not a simple function, and does not admit such a suggestive
895: interpretation. Figure \ref{fig:shape} shows the functions $f(r)$
896: for both the uncorrected (dashed) and MTF-corrected (solid) spectra
897: from 23 Feb.\ 1997. The integrated field (bottom) shows that for
898: each case half of all flux is confined to a similar $r_0\simeq4$
899: Mm circle.
900:
901: \begin{figure}[htp]
902: \psfig{file=fig13.ps,width=4.5in}
903: \caption{The structure of a magnetic element inferred from the power
904: spectrum of a magnetogram. The top curves show the shape functions,
905: $f(r)$, from \eq\ (\ref{eq:shape}), for an uncorrected
906: exponential spectrum with $d=2.2$ Mm (dashed) and for the shape after
907: correction for the low-resolution MTF with $\ell_0=1.6''$ (solid).
908: Symbols on each curve mark $r_*$. The vertical scale is the
909: same for each curve and makes $2\pi\int f(r)\,r\,dr=1$.
910: The bottom panel shows the total fraction of flux with a given radius,
911: and symbols mark the radius of half flux.}
912: \label{fig:shape}
913: \end{figure}
914:
915: The variance of a photospheric field composed of random elements is
916: \be
917: \sigma_p^2 ~=~ n\avg{\phi^2}\int f^2\, d^2x ~=~
918: {n\avg{\phi^2} \over 4\pi r_*^2} ~~,
919: \label{eq:sig_subpole}
920: \ee
921: where $r_*$ characterizes the radius of the shape function, $f(r)$,
922: through the integral of its square. The parameter is defined to match
923: the root-mean-squared radius of a Gaussian shape. The uncorrected
924: shape function, \eq\ (\ref{eq:shape}), yields, $r_*=\sqrt{2}d$, whose
925: introduction into \eq\ (\ref{eq:sig_subpole}) returns the variance of
926: the vertical field at a height $d$
927: above a distribution of point charges, i.e.\ \eq\
928: (\ref{eq:sig_mpole}). The MTF-corrected shape function is
929: notably narrower (solid line in \fig\ \ref{fig:shape}), under the
930: anticipation that it is broadened by the point spread function to
931: produce the broader observed shape (dashed) with its larger radius
932: $r_*$.
933:
934: The kurtosis of the composite field given by expression
935: (\ref{eq:composite}) is
936: \be
937: K_p ~=~ {1\over n}\,{\avg{\phi^4}\over \avg{\phi^2}^2}
938: \,{\chi\over 4\pi r_*^2} ~~,~~ \chi ~=~ (4\pi r_*^2)^3
939: \int f^4\, d^2x ~~,
940: \label{eq:kurt_subpole}
941: \ee
942: where the dimensionless coefficient $\chi$ depends only on the shape
943: of the individual elements, $f(r)$. It is unity for a perfectly flat
944: function and greater as the shape becomes more peaked. For
945: charges submerged at a depth $d$, the value $\chi=32/5$ can be
946: inferred by comparison between expressions (\ref{eq:kurt_mpole}) and
947: (\ref{eq:kurt_subpole}) and use of the relation
948: $r_*=\sqrt{2}d=\sqrt{2}z$.
949:
950: The dimensionless ratio,
951: $\xi=\avg{\phi^4}/\avg{\phi^2}^2$, depends only on the distribution of
952: element amplitudes, and not on their size or shape. Flux amplitudes
953: distributed uniformly
954: or exponentially will be characterized by
955: $\xi=9/5$ or $\xi=6$ respectively. Parnell (2002) \nocite{Parnell2002}
956: found that
957: fluxes of each polarity were distributed with a hybrid between an
958: exponential and power-law known as a Weibull distribution. The
959: particular Weibull distributions reported are characterized by
960: values of $\xi$ between $22$ and $30$. These are larger than for
961: purely exponential distribution since broader wings lead to more common
962: occurrence of large fluxes.
963:
964: The variance of Gaussian white noise, $\sigma_n^2$, adds to expression
965: (\ref{eq:sig_subpole}) to give the total variance in a magnetogram,
966: $\sigma_0^2=\sigma_p^2+\sigma_n^2$. The
967: Gaussian noise has zero kurtosis, so its addition will decrease the
968: kurtosis of the composite, \eq\ (\ref{eq:kurt_subpole}),
969: by the factor $\sigma_p^4/\sigma_0^4$.
970: The areal density of flux elements follows directly from the
971: magnetogram's kurtosis, $K_0$, and properties of the elements
972: \be
973: n ~=~ \left(1 - {\sigma_n^2\over\sigma_0^2}\right)^2
974: {\chi\over 4\pi r_*^2}\,{\xi\over K_0} ~~.
975: \label{eq:n_from_K}
976: \ee
977: Having found the density, expression (\ref{eq:sig_subpole}) can be
978: used to find the root-mean-square (rms) amplitude of the flux,
979: $\avg{\phi^2}^{1/2}$.
980:
981: Using the above reasoning we may model a quiet sun magnetogram as a
982: combination of Gaussian white noise and a random distribution of flux
983: elements. The magnetogram in
984: \fig\ \ref{fig:mg}b, for example, has a power spectrum which,
985: after removal of white noise,
986: fits an exponential $S(k)\sim e^{-2kd}$, with
987: $d\simeq 2.2$ Mm. The magnetogram has a standard deviation
988: $\sigma_0=15.8$ G, only $\sigma_n=8.8$ G of which comes from
989: white noise. Assuming an exponential
990: distribution of flux amplitudes ($\xi=6$) the kurtosis, $K_0=38$, implies a
991: flux element density,
992: $n=0.0039\,{\rm Mm}^{-2}$, from \eq\ (\ref{eq:n_from_K}).
993: Equation (\ref{eq:sig_subpole}) then
994: gives the rms flux, $\avg{\phi^2}^{1/2}=2.3\times 10^{19}$ Mx.
995: The exponential flux distribution and zero mean value can be used
996: to convert the rms to $\avg{|\phi|}=1.6\times10^{19}$ Mx.
997: % slightly larger than values found by
998: % previous investigators \citep{Schrijver1997b,Parnell2002}.
999:
1000: Figure \ref{fig:synth_cmp}a shows a magnetogram synthesized from the
1001: distribution of sources described above. Its parameters were chosen
1002: to resemble the five-minute average low resolution magnetogram of
1003: \ref{fig:mg}b, and the two have indistinguishable power spectra
1004: (\fig\ \ref{fig:synth_cmp}b). The standard deviation and kurtosis of
1005: the synthetic magnetogram, $\sigma_0=16.1$ G and $K_0=39$, are
1006: well matched to the original.
1007:
1008: \begin{figure}[htp]
1009: \centerline{\psfig{file=fig14a.ps,width=2.2in}%
1010: \psfig{file=fig14b.ps,width=2.9in}}
1011: \caption{A magnetogram synthesized from a distribution of point
1012: sources at $d=2.2$ Mm below the photosphere. (a) The magnetogram with
1013: grey scale over $|B_z|<100$ G, exactly as for \ref{fig:mg}b.
1014: (b) The power spectra of this magnetogram (dark) and of \fig\
1015: \ref{fig:mg}b (light).}
1016: \label{fig:synth_cmp}
1017: \end{figure}
1018:
1019: Correcting the spectrum for an MTF with $\ell_0=1.6''$ leads to a
1020: more narrowly peaked element, $f(r)$, whose radius is
1021: $r_*=2.1$ Mm and which has $\chi=13.2$
1022: (compared to $r_*=\sqrt{2}d=3.1$ Mm and
1023: $\chi=6.4$ without correction; see \fig\ \ref{fig:shape}).
1024: These more peaked elements must be distributed with
1025: greater areal density and smaller mean flux
1026: in order to match the observed kurtosis and standard
1027: deviation: $n=0.016\,{\rm Mm}^{-2}$ and
1028: $\avg{|\phi|}=5.6\times10^{18}$ Mx.
1029: The instrument's point spread
1030: function blurs each element reducing the total apparent variance.
1031: The MTF correction thereafter amplifies the observed variance
1032: leading us to the inference of a higher density of fundamental
1033: elements. In this case by a factor of four.
1034:
1035: The same analysis can be performed on all of the spectra analyzed in
1036: the previous section. The MTF correction implies fundamental elements
1037: with a median radius $r_*=1.1$ Mm, too small to observed directly in
1038: low resolution. As a consequence the median element density is quite
1039: large, $n=0.046\,{\rm Mm}^{-2}$, and median flux quite small,
1040: $\avg{|\phi|}=0.22\times10^{19}$ Mx.
1041: These values are outside the range found by direct
1042: identification of flux elements in magnetograms
1043: \citep{Schrijver1997b,Parnell2002,DeForest2007} since the correction for
1044: point spread function is not generally performed. Indeed, our final
1045: results depend quite sensitively on this correction, since
1046: unresolved elements contribute only slightly to the statistics of
1047: the observed magnetic field.
1048:
1049: We repeat our analysis without the MTF correction to obtain an
1050: estimate which is at once more conservative and more consistent with
1051: previous investigation. Figure \ref{fig:phi_smry} shows the
1052: depth (bottom panel), areal density, $n$ (middle panel)
1053: and flux, $\avg{|\phi|}$ (top panel) from low resolution
1054: magnetograms over the 2006-2007 solar minimum.
1055: These exhibit the same slow variation as the
1056: null density, presumably for the same reason: gradual variation of
1057: focus. The MTF-corrected version (not shown) has still more pronounced
1058: variation due to its amplification of the high-wavenumber portion of
1059: each spectrum. The median depth is $d=1.43$ Mm, and
1060: without the MTF correction the median values for areal
1061: density and mean flux are $n=0.007\,{\rm Mm}^{-2}$ and
1062: $\avg{|\phi|}=1.0\times10^{19}$ Mx respectively. These are the value
1063: for which the observations provide direct evidence and are
1064: consistent with studies which directly identify flux elements
1065: \citep{Schrijver1997b,Parnell2002,DeForest2007}.
1066:
1067: \begin{figure}[htp]
1068: \psfig{file=fig15.ps,width=5.1in}
1069: \caption{Parameters for the model magnetograms derived from spectra of
1070: one-minute (triangles) and five-minute (crosses) full-disk magnetograms.
1071: (top) The mean unsigned flux, $\avg{|\phi|}$, in units of $10^{19}$
1072: Mx. (middle) The density of sources in units of Mm$^{-1}$.
1073: (bottom) The depth $d$ at which the sources are placed.}
1074: \label{fig:phi_smry}
1075: \end{figure}
1076:
1077: \subsection{Using the model in a noise-free spectral estimate}
1078:
1079: Submerged point sources alone, without the addition of white noise,
1080: have the same purely exponential power spectrum as the
1081: noise-corrected magnetograms.
1082: It was found in \papi, however, that the
1083: spectral estimate grossly overestimates the column of null points
1084: above a distribution of submerged point sources. This was
1085: attributed to a severe departure from gaussianity within a layer
1086: extending $\Delta z\sim n^{-1/2}$ from the source plane
1087: ($z\le15$ Mm in our case). One
1088: symptom of this departure is the divergence of the kurtosis given by
1089: \eq\ (\ref{eq:kurt_mpole}). The known overestimation for point
1090: sources raises concerns about the accuracy
1091: of the spectral estimate from the noise-corrected magnetogram
1092: spectrum, which has the same spectrum.
1093:
1094: To further explore this comparison
1095: we reduced the noise in the low-resolution 5-minute
1096: averaged magnetogram, \fig\ \ref{fig:mg}b, using a Fourier filter.
1097: Amplitudes of all high-spatial frequency Fourier modes
1098: ($|\kvec|>0.7\,{\rm rad/arcsec}$) were rescaled to fit
1099: a pure exponential, $S(k)\propto e^{-2kd}$ with $d=2.2$, while
1100: retaining their original phases.
1101: The result was a magnetogram (see \fig\ \ref{fig:clean}) which is
1102: visibly cleaner, albeit blurrier than the original
1103: (\fig\ \ref{fig:mg}b). The central
1104: core of its distribution function fits a Gaussian of width
1105: $\sigma^{(g)}_z=7.0$ G. The original, 5-minute averaged magnetogram
1106: had $\sigma^{(g)}_z=10.6$ G of which $k_c\sqrt{\pi S_n}=8.8$ G was
1107: found to be white noise, leaving $\sigma_z=5.9$ G to be be genuine.
1108: Thus it would seem our Fourier filtering removes a substantial amount
1109: of the noise.
1110:
1111: \begin{figure}[ht]
1112: \centerline{\psfig{file=fig16a.ps,width=2.2in}%
1113: \psfig{file=fig16b.ps,width=2.9in}}
1114: \caption{The result of Fourier filtering the low-resolution 5-minute
1115: average magnetogram \fig\ \ref{fig:mg}b. (left) The magnetogram plotted
1116: using the same grey scale as the original. (right) The null densities,
1117: $\rho_N(z)$, for the Fourier-filtered
1118: magnetogram (diamonds) compared to the spectral
1119: estimate (dashed). The null densities from the synthetic
1120: magnetograms created with low and high densities
1121: of submerged point sources are plotted with triangles and squares
1122: respectively.}
1123: \label{fig:clean}
1124: \end{figure}
1125:
1126: The Fourier-filtered magnetogram has, by construction, a psd
1127: similar to a magnetogram synthesized from submerged
1128: point sources (like \fig\ \ref{fig:synth_cmp}a without added
1129: noise). Such a magnetogram was synthesized, extrapolated
1130: onto a grid and all its null points located
1131: using the algorithm of Haynes and Parnell (2007). The result was a
1132: column density $N_n(1.5\,{\rm Mm})=0.3\times10^{-3}\,{\rm Mm}^{-2}$
1133: of null points.
1134: Repeating the procedure for the Fourier filtered magnetogram yielded a
1135: column, $N_n(1.5\,{\rm Mm})=3.6\times10^{-3}\,{\rm Mm}^{-2}$, more
1136: than an order of magnitude greater.
1137: The spectral estimates for both cases (which have similar
1138: spectra) is $N_n(1.5\,{\rm Mm})=2.0\times10^{-3}\,{\rm Mm}^{-2}$; this
1139: is smaller than the results from the previous section, which also
1140: include MCT-correction.
1141:
1142: The spectral estimate exceeds the column above point sources by
1143: about a factor of seven, in agreement with the finding of \papi\
1144: (they find an actual column $N_n(-d)=0.074n=
1145: 0.29\times10^{-3}\,{\rm Mm}^{-2}$).
1146: On the other hand, it {\em underestimates} the column above the
1147: filtered magnetogram by 44\%, in spite of its very large kurtosis:
1148: $K_z=49.0$ (larger than the magnetogram before filtering).
1149:
1150: To explore the significance of kurtosis, a
1151: second synthetic magnetogram was constructed with a greater areal
1152: density ($n=10^{-2}\,{\rm Mm}^{-2}$) of point
1153: sources at the same depth ($d=2.2$)
1154: but with correspondingly smaller flux ($\avg{|\phi|}=0.69\times10^{19}$
1155: Mx) to endow it with the same spectrum as the other two. The kurtosis
1156: of the resulting magnetogram ($K=29$) is lower than either of the
1157: others.
1158: A gridded extrapolation of this field was found to have a column
1159: $N_n(1.5\,{\rm Mm})=1.0\times10^{-3}\,{\rm Mm}^{-2}$. Moreover, the
1160: actual density, $\rho_N(z)$ falls well below the spectral estimate
1161: below $z=n^{-1/2}-d=7.8$ Mm, as shown in the right panel of
1162: \fig\ \ref{fig:clean}. In
1163: contrast, the null density of the Fourier-filtered magnetogram tracks
1164: the spectral estimate, but is slightly above it, over the entire range
1165: of heights. The height at which the null density of the first
1166: synthetic magnetogram becomes accurate, is $z=n^{-1/2}-d=13.8$ Mm,
1167: consistent with \fig\ \ref{fig:clean}b.
1168:
1169: These experiments suggest that the accuracy of the spectral estimate
1170: in predicting the actual null point column, is unaffected by the
1171: absence of white noise. For both the filtered and unfiltered
1172: 5-minute magnetogram it underestimates the column by a similar factor.
1173: This performance is far better than it would be if the
1174: photospheric field consisted of submerged point sources, as its
1175: spectrum suggests. The discrepancy cannot be explained by kurtosis of
1176: the fields, since the spectral estimate is still too large for a
1177: point-source-derived field with smaller kurtosis. It seems that the
1178: spectral estimate owes its success in the quiet Sun to a field
1179: intrinsically more complex than one from point sources.
1180:
1181: \section{Discussion}
1182:
1183: The foregoing has shown that the Fourier spectra of quiet Sun
1184: magnetograms have an invariant basic form, at least over moderately
1185: high wave numbers ($0.2\,{\rm rad/arcsec}<k<1.5\,{\rm rad/arcsec}$).
1186: This spectral range plays the greatest role in
1187: dictating the distribution of magnetic null points in the corona above.
1188: As a consequence we find that the density of magnetic null
1189: points in a potential field extrapolation is about
1190: $N_n(1.5\,{\rm Mm})=3.1\times 10^{-3}\,{\rm Mm}^{-2}$
1191: over any patch of quiet Sun, varying by roughly $10\%$ from
1192: day to day. This value is just more than twice the
1193: number of nulls found by R\'egnier, Parnell and Haynes (2008)
1194: at heights greater
1195: than 1.5 Mm, using Hinode data of much higher resolution. The discrepancy
1196: could possibly arise from the much greater
1197: flux imbalance over that particular magnetogram. The column we find
1198: corresponds to 2 nulls above every 5 hexagonal, 14-Mm-diameter
1199: supergranules \citep{Hagenaar1997} .
1200:
1201: The potential magnetic field extrapolated from a photospheric
1202: magnetogram tends to become spatially smoother at increasing
1203: heights. This is a general property of harmonic functions and their
1204: gradients. One consequence is that the density of null points in such
1205: a field decreases with height and most of the null points are found
1206: close to the surface. Of the null points above $z=1.5$ Mm
1207: roughly $75\%$ are found within the next 3.0 Mm.
1208: We find the spectral estimate of the null point
1209: density in the generic quiet Sun to be
1210: $\rho_n(z)\simeq0.04/(z+d)^{-3}$, where $d\simeq1.4\,{\rm Mm}$. This
1211: simple functional form follows from the approximately exponential
1212: nature of the photospheric spectrum.
1213:
1214: The density at very low heights, say $z<1\,{\rm Mm}$, is dictated by
1215: very small horizontal structure in the photospheric field. This
1216: structure is the most challenging to observe, and its observation is
1217: subject to the most instrumental distortion. Moreover, this lowest
1218: atmospheric layer, the chromosphere, has the greatest plasma pressure
1219: and mass density and is least likely to satisfy the conditions which
1220: would render a magnetic field potential, or even force free.
1221: For these reasons we choose to focus our investigation on null
1222: points above $z=1.5\,{\rm Mm}$. This nominal height has nothing to do
1223: with the empirical depth, $d$, characterizing the exponential slope of
1224: the photospheric spectra. Rather, it arose from our experience with
1225: different instrumentation (MDI low-resolution {\em versus}
1226: high-resolution) and different methods of accommodating instrumental
1227: effects
1228: such as the MTF. We found that the null point density, $\rho_N(z)$,
1229: varied less above $z=1.5\,{\rm Mm}$ due to these instrumental
1230: factors, so we deemed it to be more accurately measured there.
1231:
1232: % density of null points diminishes rapidly with height.
1233: % of nulls above 1.5 Mm, the vast majority are found in the
1234: % next 3 Mm (75%). the actual number depends critically on
1235: % m-gram spectra in range 0.3 rad/Mm - 1 rad/Mm. scales
1236: % are adequately sampled by current instrumentation, MDI, but
1237: % requires careful attention.
1238:
1239: The density and distribution
1240: of null points in the corona is predicted somewhat reliably,
1241: although not perfectly, by spectral estimate of Longcope,
1242: Brown and Priest (2003, \papi). Magnetic null points can be found
1243: directly, and thus with more accuracy, in a potential field
1244: extrapolated from a photospheric
1245: magnetogram. We used this technique to test the spectral estimate,
1246: and found it predicted the shape of the density,
1247: $\rho_N(z)$, reasonably well. The spectral estimate of the total column
1248: was, however, too low by a factor of two to three above one Mm.
1249: If the spectral estimate for every magnetogram is low by the same
1250: factor, then the actual null
1251: density in the quiet Sun would be greater than the value we report.
1252:
1253: This discrepancy is most
1254: likely due to a departure of the actual field from the statistical
1255: properties assumed by the spectral estimate (homogeneity and
1256: Gaussianity). We did not find a satisfactory method of quantifying
1257: these factors. One characteristic in particular, the kurtosis of the
1258: distribution, did not accurately predict when the assumptions were
1259: adequately satisfied. We did find, however, that the conditions
1260: appear to be better satisfied, and the predictions better, for actual
1261: magnetograms than for a random distribution of point magnetic sources.
1262: In the latter case the spectral estimate {\em exceeds}
1263: the number of null points by at least an order of magnitude.
1264:
1265: Although it is more accurate, direct enumeration of magnetic null
1266: points suffers from several disadvantages relative to the spectral
1267: estimate. It is more computationally expensive, so performing the
1268: search on over 500 different grids
1269: would be prohibitive. More significantly, instrumental artifacts
1270: such as noise and MTF are much more difficult to remove from the
1271: extrapolation than to remove from the spectrum. We found that white
1272: noise present in MDI magnetograms can contribute artificial null points
1273: outnumbering the real ones by as many as five-to-one. The ease with
1274: which such factors can be removed the Fourier spectra makes the
1275: spectral estimate a superior one, even given its inaccuracy.
1276:
1277: % Non-potential fields ?
1278:
1279: We found that the spatial spectrum of the quiet Sun magnetic field is
1280: well approximated as an exponential, at least for
1281: $k>0.2\,{\rm rad/Mm}$. This spectrum is consistent with the potential
1282: field from point magnetic sources submerged to a depth of $d\simeq1.4$
1283: Mm. These manifest themselves on the photosphere as circular
1284: concentrations of flux, approximately $r_*=2.5$ Mm in radius. We
1285: are able to infer the size and areal density of these structures,
1286: $\avg{|\phi|}=1.0\times10^{19}$ Mx and
1287: $n=0.007\,{\rm Mm}^{-2}$, solely from the
1288: statistics of the photospheric field. The resolution of
1289: such structures into finer elements, clearly seen in high-resolution
1290: observations, will affect the spectrum only at higher wavenumbers. The
1291: spectral form we observe suggests that such fine scales are
1292: prone to cluster into 2.5 Mm concentrations.
1293:
1294: Along similar lines, the imprint of structuring at still larger
1295: scales, such as by the network, manifests itself at lower
1296: wavenumbers, $k<0.2$ rad/Mm. We find that actual quiet Sun spectra
1297: differ the most over these low wavenumbers, suggesting more pronounced
1298: day-to-day variation at these larger scales. Furthermore, any particular
1299: spectrum departs most from its exponential form over these low
1300: wavenumbers. This can be explained using fundamental elements
1301: distributed non-uniformly across the surface ---
1302: concentrated, for example along network boundaries. This spectral
1303: range dictates the distribution of null points at great heights, where
1304: very few ever occur. Thus the variation has virtually no effect on
1305: the total column we report.
1306:
1307: We performed our analysis on 562 different patches
1308: ($300''\times300''$) of quiet Sun
1309: found at disk center over two different solar minima (1996-1998 and
1310: 2006-2007). The magnetograms were chosen to be free of active
1311: regions, plage or dominance of a single polarity. No other data was
1312: used to determine if the region was, for example, part of a coronal
1313: hole. Given their equatorial locations, however, it seems unlikely
1314: that much coronal hole data was included in our survey. We cannot
1315: therefore say whether the null point density applies to coronal holes
1316: as well as to equatorial quiet Sun.
1317:
1318: These findings place in a clearer context observations and models with
1319: putative relationships to coronal null points. If the quiet Sun
1320: corona is being heated primarily through dissipation at null
1321: points, then the majority of this heat is supplied very low in the
1322: corona. The region between $z=1.5$ Mm and $z=4.5$ Mm contains three
1323: times as many nulls as the whole space above it; it would therefore
1324: probably receive a proportional share of the heating. It remains to
1325: be investigated whether the actual thermal structure of the quiet Sun
1326: corona is consistent with this distribution of heating, and is
1327: therefore consistent with heating by energy dissipation at null
1328: points.
1329:
1330: Along similar lines, if coronal jets were hypothesized to form
1331: anywhere a coronal null point occurs, then our results can be used in
1332: an {\em ab initio}
1333: prediction of their frequency. Moreover, the distribution of their cusp
1334: heights should be somehow influenced by the distribution of null
1335: heights: short jets should out-number tall ones. Actual jets
1336: undoubtedly involve many factors, in addition to magnetic
1337: geometry, which influence the distribution of their properties. Still,
1338: it would be difficult to invoke null points as a
1339: requirement for jets if jets of a given height outnumbered null points
1340: at that level. A survey of coronal jets comprehensive enough to
1341: determine their frequency would thereby cast some light on the
1342: importance of null points in their construction.
1343:
1344: \appendix
1345:
1346: \section{Estimating the modulation transfer function (MTF)}
1347:
1348: To estimate the MTF, consider an ideal imaging system
1349: characterized by the same point spread function at every point on the
1350: image plane (i.e.\ an isoplanatic system).
1351: In this case the power spectral density (psd), $S(k)$, of the image
1352: is the product of the actual psd, $\tilde{S}(k)$, and the square of
1353: the modulation transfer function (MTF), $M(k)$ \citep{Born1980}.
1354: The MTF is the magnitude of the Fourier transform of the normalized point
1355: spread function. Generally, $M(k)$, is a decreasing function of
1356: wave-number, causing the observed psd, $S(k)$,
1357: to be steeper than the actual one, $\tilde{S}(k)$.
1358:
1359: If the same field is observed by MDI in both its
1360: high and low resolution modes the
1361: underlying psd, $\tilde{S}(k)$, will be the same for both images. In
1362: this case
1363: \be
1364: {S_{\rm lo}(k)\over S_{\rm hi}(k)} ~=~
1365: {M^2_{\rm lo}(k)\over M^2_{\rm hi}(k)} ~~.
1366: \label{eq:mtf_rat}
1367: \ee
1368: The MTF of the low resolution images will be dominated by the
1369: defocussing intended to reduce aliasing from
1370: spatial frequencies above the Nyquist limit
1371: $k_{n}=1.6$ rad/arcsec \citep{Scherrer1995}.
1372: It is this contribution we can ascertain by cross comparison.
1373:
1374: One-minute high resolution magnetograms were obtained for 11 different
1375: intervals on five different dates throughout 2007 while MDI was in
1376: focus position 5.
1377: A single psd, $S_{\rm hi}(k)$, was constructed for each time
1378: interval. In cases where several (typically 4) consecutive one-minute
1379: magnetograms were available from a given interval, the individual psd's
1380: were averaged together to produce $S_{\rm hi}(k)$ with lower noise. The
1381: nearest 96-minute full disk low-resolution magnetogram was then used
1382: to compute $S_{\rm lo}(k)$. The high and low resolution magnetograms
1383: were separated by six to 30 minutes (typically 10 minutes)
1384: strengthening the assumption that $\tilde{S}(k)$ is the same for both
1385: observations.
1386:
1387: All psd's were computed as described in section \ref{sec:psd} using
1388: spectral annuli of width $\Delta k=0.05$ rad/arcsec,
1389: and the white noise floor was
1390: subtracted from each psd. The ratio of \eq\ (\ref{eq:mtf_rat}) was
1391: computed over the range $k<1.5$ rad/arcsec where both spectra
1392: were available and fairly free of noise. Full
1393: disk magnetograms have been re-calibrated by the
1394: MDI instrument team while high-resolution magnetograms
1395: have not. To compensate for the different calibration we divided the
1396: ratio by the mean of its lowest 8 spectral bins ($k<0.4$ rad/arcsec).
1397:
1398: A median MTF ratio is constructed by taking the median value of
1399: $S_{\rm lo}(k)/S_{\rm hi}(k)$ from
1400: all 11 time intervals for each spectral bin (see \fig\
1401: \ref{fig:mtf_est}a). It is fairly flat over the long-wavelength range
1402: and decays to low values by $k>1.2$ rad/arcsec. Although the low resolution
1403: magnetograms extend out to $k_c = 1.8$ rad/arcsec, it appears there is little
1404: sensitivity over that last third of that range.
1405:
1406: \begin{figure}[htp]
1407: \centerline{\psfig{file=fig17a.ps,width=2.7in}%
1408: \psfig{file=fig17b.ps,width=2.4in}}
1409: \caption{The MTF and point spread function used for MDI. (a)
1410: The median of MTF estimates from 11 quiet Sun observations.
1411: The dark solid curve is the estimate from the ratio of \eq\
1412: (\ref{eq:mtf_rat}). Dashed and dotted curves are empirical fits.
1413: The triangle is the pre-launch value of the ratio quoted by
1414: Scherrer \etal\ (1995).
1415: (b) The point spread functions for the high
1416: resolution (dashed) and low resolution (solid) based on the MTF
1417: calculation. Sample pixel locations for each resolution are shown
1418: below the curves with crosses.}
1419: \label{fig:mtf_est}
1420: \end{figure}
1421: \nocite{Scherrer1995}
1422:
1423: The estimated MTF ratio
1424: from \fig\ \ref{fig:mtf_est}a was fitted, over the range
1425: $0.1\,{\rm rad/arcsec}<k<1\,{\rm rad/arcsec}$,
1426: by a variety of empirical functions. Dashed and dotted
1427: curves show two of the more successful functions
1428: \begin{eqnarray}
1429: M^2_{\rm lo}/M^2_{\rm hi} &=& {A\over 1 + \ell_0^4 k^4} ~~,
1430: \label{eq:MTF_empirical} \\
1431: M^2_{\rm lo}/M^2_{\rm hi} &=& A' e^{-r_0^2k^2} ~~,
1432: \label{eq:MTF_gauss}
1433: \end{eqnarray}
1434: respectively. The best fits were for parameter values
1435: $\ell_0=1.04''$ and
1436: $r_0=0.88''$. The former curve better fits the flat inner
1437: region and the steeper fall-off beyond $k\simeq0.6$ rad/arcsec.
1438: There is evidence that the actual MTF falls even
1439: faster, perhaps even reaching zero, for $k>1.4$ rad/arcsec. Dividing the
1440: observed spectrum by expression (\ref{eq:MTF_empirical}) is therefore
1441: a conservative correction at the very highest wave numbers.
1442:
1443: While this cross comparison provides the ratio of the two MTFs, it
1444: cannot give any information about either one alone. The most conservative
1445: correction to either spectrum is to assume the high-resolution
1446: mode is diffraction limited and in perfect focus. In this case its
1447: MTF would be \citep{Ghatak1978}
1448: \be
1449: M_{\rm hi}(k) ~=~ M_d(k) ~=~ {2\over\pi}\left[\,
1450: \cos^{-1}\left({k\over k_0}\right) ~-~
1451: {k\over k_0}\sqrt{1 - {k^2\over k_0^2}} \, \right] ~~,
1452: \label{eq:MTF_d}
1453: \ee
1454: where $k_0=2\pi d/\lambda_{\rm Ni} =
1455: 5.63\,{\rm rad/arcsec}$ for a primary mirror of diameter, $d=12.5$ cm
1456: and optical wavelength $\lambda_{\rm Ni}=6768\AA$ \citep{Scherrer1995}.
1457:
1458: \nocite{Scherrer1995}
1459: Scherrer \etal\ (1995) reported that the intentional
1460: defocussing reduced the low resolution MTF at $k=0.8k_n=1.3$ rad/arcsec
1461: to $M=0.17$ from its diffraction-limited value of $M_d=0.47$. That is
1462: to say $M_{\rm lo}/M_{\rm hi}=0.36$, under our assumption that
1463: $M_{\rm hi}=M_d$. The square of this value is
1464: plotted as a triangle in \fig\ \ref{fig:mtf_est}. Our data point is
1465: consistent with this value, but the empirical fits are greater at this
1466: end of the spectrum. Using the empirical fits will then under-correct
1467: the highest wave-numbers.
1468:
1469: \acknowledgements
1470:
1471: We wish to thank Phil Scherrer, Todd Hoeksema, Jesper Schou
1472: and Craig DeForest for helpful discussions concerning MDI. This
1473: work was supported in part by NASA's TR\&T program.
1474:
1475: %\bibliography{/home/dana/tex/inputs/short_abbrevs,/home/dana/tex/full_lib}
1476: %\bibliography{/Users/danalongcope/stuff//long_abbrevs.bib,/Users/danalongcope/stuff/full_lib.bib}
1477:
1478: \begin{thebibliography}{36}
1479: \providecommand{\natexlab}[1]{#1}
1480:
1481: \bibitem[\protect\citeauthoryear{Abramenko \emph{et~al.}}{2001}]{Abramenko2001}
1482: Abramenko, V., Yurchyshyn, V., Wang, H., Goode, P.~R.: 2001,
1483: \emph{Solar~Phys.} \textbf{201}, 225.
1484:
1485: \bibitem[\protect\citeauthoryear{Antiochos}{1998}]{Antiochos1998}
1486: Antiochos, S.~K: 1998, \emph{Astrophys. J} \textbf{502}, L181.
1487:
1488: \bibitem[\protect\citeauthoryear{{Beveridge}, Priest and
1489: Brown}{2004}]{Beveridge2004}
1490: {Beveridge}, C., {Priest}, E.~R., {Brown}, D.~S.: 2004, \emph{Geophysical and
1491: Astrophysical Fluid Dynamics} \textbf{98}, 429.
1492:
1493: \bibitem[\protect\citeauthoryear{Born and Wolf}{1980}]{Born1980}
1494: Born, M., Wolf, E.: 1980, \emph{Principles of Optics: Electromagnetic theory
1495: of propagation and diffraction of light}, Pergamon Press, New York, NY, sixth
1496: edition.
1497:
1498: \bibitem[\protect\citeauthoryear{{Cirtain} \emph{et~al.}}{2007}]{Cirtain2007}
1499: {Cirtain}, J.~W., {Golub}, L., {Lundquist}, L., {van Ballegooijen}, A.,
1500: {Savcheva}, A., {Shimojo}, M., {DeLuca}, E., {Tsuneta}, S., {Sakao}, T.,
1501: {Reeves}, K., {Weber}, M., {Kano}, R., {Narukage}, N., {Shibasaki}, K.:
1502: 2007, \emph{Science} \textbf{318}, 1580.
1503:
1504: \bibitem[\protect\citeauthoryear{Close, Parnell and Priest}{2004}]{Close2004b}
1505: Close, R.~M., Parnell, C.~E., Priest, E.~R.: 2004, \emph{Solar~Phys.}
1506: \textbf{225}, 21.
1507:
1508: \bibitem[\protect\citeauthoryear{{Craig}, Fabling and Henton}{1995}]{Craig1995}
1509: {Craig}, I.~J.~D., {Fabling}, R.~B., {Henton}, S.~M., {Rickard}, G.~J.: 1995,
1510: \emph{Astrophys. J} \textbf{455}, L197.
1511:
1512: \bibitem[\protect\citeauthoryear{Craig and Mc{C}lymont}{1993}]{Craig1993}
1513: Craig, I. J.~D., Mc{C}lymont, A.~N.: 1993, \emph{Astrophys. J} \textbf{405},
1514: 207.
1515:
1516: \bibitem[\protect\citeauthoryear{{Culhane} \emph{et~al.}}{2007}]{Culhane2007}
1517: {Culhane}, L., {Harra}, L.~K., {Baker}, D., {van Driel-Gesztelyi}, L., {Sun},
1518: J., {Doschek}, G.~A., {Brooks}, D.~H., {Lundquist}, L.~L., {Kamio}, S.,
1519: {Young}, P.~R., {Hansteen}, V.~H.: 2007, \emph{Pub. Astron. Soc. Japan} \textbf{59}, 751.
1520:
1521: \bibitem[\protect\citeauthoryear{DeForest \emph{et~al.}}{2007}]{DeForest2007}
1522: DeForest, C.~E., Hagenaar, H.~J., Lamb, D.~A., Parnell, C.~E., Welsch, B.~T.:
1523: 2007, \emph{Astrophys. J} \textbf{666}, 576.
1524:
1525: \bibitem[\protect\citeauthoryear{Dungey}{1958}]{Dungey1958}
1526: Dungey, J.~W.: 1958, \emph{Cosmic Electrodynamics}, Cambridge University Press,
1527: Cambridge, U.K.
1528:
1529: \bibitem[\protect\citeauthoryear{Galsgaard and Nordlund}{1997}]{Galsgaard1997}
1530: Galsgaard, K., Nordlund, A.: 1997, \emph{J Geophys. Res.} \textbf{102}, 231.
1531:
1532: \bibitem[\protect\citeauthoryear{Ghatak and Thyagarajan}{1978}]{Ghatak1978}
1533: Ghatak, A.~J., Thyagarajan, K.: 1978, \emph{Contemporary Optics}, Plenum
1534: Press, New York.
1535:
1536: \bibitem[\protect\citeauthoryear{Greene}{1988}]{Greene1988}
1537: Greene, J.~M.: 1988, \emph{J Geophys. Res.} \textbf{93}, 8583.
1538:
1539: \bibitem[\protect\citeauthoryear{Hagenaar, Schrijver and Title}{1997}]{Hagenaar1997}
1540: Hagenaar, H.~J., Schrijver, C.~J., Title, A.~M.: 1997, \emph{Astrophys. J}
1541: \textbf{481}, 988.
1542:
1543: \bibitem[\protect\citeauthoryear{Hassam}{1992}]{Hassam1992}
1544: Hassam, A.~B.: 1992, \emph{Astrophys. J} \textbf{399}, 159.
1545:
1546: \bibitem[\protect\citeauthoryear{Haynes and Parnell}{2007}]{Haynes2007}
1547: Haynes, A.~L., Parnell, C.~E.: 2007, \emph{Phys.~Plasmas} \textbf{14}, 2107.
1548:
1549: \bibitem[\protect\citeauthoryear{Hesse and Schindler}{1988}]{Hesse1988}
1550: Hesse, M., Schindler, K.: 1988, \emph{J Geophys. Res.} \textbf{93}, 5559.
1551:
1552: \bibitem[\protect\citeauthoryear{Hufnagel and Stanley}{1964}]{Hufnagel1964}
1553: Hufnagel, R.~E., Stanley, N.~R.: 1964, \emph{J. Optical Soc. America} 52--61.
1554:
1555: \bibitem[\protect\citeauthoryear{Liu and Norton}{2001}]{Liu2001}
1556: Liu, Y., Norton, A.~A.: 2001, Mdi measurement errors: The magnetic
1557: perspective, Technical Report SOI Technical Note 01-144, Standford SOI.
1558:
1559: \bibitem[\protect\citeauthoryear{Longcope, Brown and Priest}{2003}]{Longcope2003b}
1560: Longcope, D.~W., Brown, D.~S., Priest, E.~R.: 2003, \emph{Phys.~Plasmas}
1561: \textbf{10}, 3321.
1562:
1563: \bibitem[\protect\citeauthoryear{{McLaughlin} and {Hood}}{2004}]{McLaughlin2004}
1564: {McLaughlin}, J.~A., {Hood}, A.~W.: 2004, \emph{Astron. Astrophy.}
1565: \textbf{420}, 1129.
1566:
1567: \bibitem[\protect\citeauthoryear{{Moreno-Insertis}, Galsgaard and Ugarte-Urra}{2008}]{MorenoInsertis2007}
1568: {Moreno-Insertis}, F., {Galsgaard}, K., {Ugarte-Urra}, I.: 2008,
1569: \emph{Astrophys. J} \textbf{673}, L211.
1570:
1571: \bibitem[\protect\citeauthoryear{Parnell}{2002}]{Parnell2002}
1572: Parnell, C.: 2002, \emph{Mon. Not. Roy. Astron. Soc.} \textbf{335}, 389.
1573:
1574: \bibitem[\protect\citeauthoryear{{Pontin} and {Galsgaard}}{2007}]{Pontin2007}
1575: {Pontin}, D.~I., {Galsgaard}, K.: 2007, \emph{J Geophys. Res.} \textbf{112},
1576: 3103.
1577:
1578: \bibitem[\protect\citeauthoryear{Press \emph{et~al.}}{1986}]{Press1986}
1579: Press, W.~H., Flannery, B.~P., Teukolsky, S.~A., Vetterling, W.~T.: 1986,
1580: \emph{Numerical Recipes: The art of scientific computing}, Cambridge
1581: University Press, Cambridge.
1582:
1583: \bibitem[\protect\citeauthoryear{R{\'e}gnier, Parnell and Haynes}{2008}]{Regnier2008}
1584: R{\'e}gnier, S., Parnell, C.~E., Haynes, A.~L.: 2008, \emph{Astron. Astrophy.}
1585: \textbf{484}, L47.
1586:
1587: \bibitem[\protect\citeauthoryear{{Rickard} and {Titov}}{1996}]{Rickard1996}
1588: {Rickard}, G.~J., {Titov}, V.~S.: 1996, \emph{Astrophys. J} \textbf{472}, 840.
1589:
1590: \bibitem[\protect\citeauthoryear{Scherrer \emph{et~al.}}{1995}]{Scherrer1995}
1591: Scherrer, P.~H., Bogart, R.~S., Bush, R.~I., Hoeksema, J.~T., Kosovichev,
1592: A.~G., Schou, J., Rosenberg, W., Springer, L., Tarbell, T.~D., Title, A.,
1593: Wolfson, C.~J., Zayer, I., {MDI Engineering Team}: 1995, \emph{Solar~Phys.}
1594: \textbf{162}, 129.
1595:
1596: \bibitem[\protect\citeauthoryear{Schrijver and Title}{2002}]{Schrijver2002}
1597: Schrijver, C.~J., Title, A.~M.: 2002, \emph{Solar~Phys.} \textbf{207}, 223.
1598:
1599: \bibitem[\protect\citeauthoryear{Schrijver \emph{et~al.}}{1997}]{Schrijver1997b}
1600: Schrijver, C.~J., Title, A.~M., Hagenaar, H.~J., Shine, R.~A.: 1997,
1601: \emph{Solar~Phys.} \textbf{175}, 329.
1602:
1603: \bibitem[\protect\citeauthoryear{Seehafer}{1986}]{Seehafer1986}
1604: Seehafer, N.: 1986, \emph{Solar~Phys.} \textbf{105}, 223.
1605:
1606: \bibitem[\protect\citeauthoryear{{Shibata} \emph{et~al.}}{1992}]{Shibata1992}
1607: {Shibata}, K., {Ishido}, Y., {Acton}, L.~W., {Strong}, K.~T., {Hirayama}, T.,
1608: {Uchida}, Y., {McAllister}, A.~H., {Matsumoto}, R., {Tsuneta}, S., {Shimizu},
1609: T., {Hara}, H., {Sakurai}, T., {Ichimoto}, K., {Nishino}, Y., {Ogawara},
1610: Y.: 1992, \emph{Pub. Astron. Soc. Japan} \textbf{44}, L173.
1611:
1612: \bibitem[\protect\citeauthoryear{{Shimojo} \emph{et~al.}}{1996}]{Shimojo1996}
1613: {Shimojo}, M., {Hashimoto}, S., {Shibata}, K., {Hirayama}, T., {Hudson}, H.~S.,
1614: {Acton}, L.~W.: 1996, \emph{Pub. Astron. Soc. Japan} \textbf{48}, 123.
1615:
1616: \bibitem[\protect\citeauthoryear{Sweet}{1958}]{Sweet1958}
1617: Sweet, P.~A.: 1958, in B.~Lehnert (ed.), \emph{Electromagnetic Phenomena in
1618: Cosmical Physics}, 123--134, Cambridge University Press, Cambridge, U.K.
1619:
1620: \bibitem[\protect\citeauthoryear{Yokoyama and Shibata}{1996}]{Yokoyama1996}
1621: Yokoyama, T., Shibata, K.: 1996, \emph{PASJ} \textbf{48}, 353.
1622:
1623: \end{thebibliography}
1624:
1625:
1626: %\end{article}
1627: \end{document}
1628: