0807.0442/ms.tex
1: \documentclass[12pt,preprint]{aastex}
2: 
3: \renewcommand{\vec}[1]{\mbox{\boldmath{$#1$}}}
4: 
5: \title{Solar cycle related changes at the base of the convection zone}
6: \author{Charles S. Baldner and Sarbani Basu}
7: \affil{Department of Astronomy, Yale University, P.O. Box 208101, New Haven, CT, 06520-8101}
8: \email{charles.baldner@yale.edu}
9: 
10: \begin{document}
11: \begin{abstract}
12: The frequencies of solar oscillations are known to change with solar 
13: activity.  We use Principal Component Analysis to examine these changes 
14: with high precision.  In addition to the well-documented changes in solar 
15: normal mode oscillations with activity as a function of frequency, which 
16: originate in the surface layers of the Sun, we find a small but 
17: statistically significant change in frequencies with an origin at and 
18: below the base of the convection zone.  We find that at 
19: $r=(0.712^{+0.0097}_{-0.0029})R_\odot$, the change in sound speed is 
20: $\delta c^2 / c^2 = (7.23 \pm 2.08) \times 10^{-5}$ between high and low activity.  
21: This change is very tightly correlated with solar activity.  In addition, 
22: we use the splitting coefficients to examine the latitudinal structure 
23: of these changes.  We find changes in sound speed correlated with 
24: surface activity for $r \gtrsim 0.9R_\odot$.
25: \end{abstract}
26: \keywords{Sun: helioseismology, Sun: activity, Sun: interior}
27: 
28: \section{INTRODUCTION}
29: \label{sec:intro}
30: Normal modes of oscillation of the Sun have provided a 
31: powerful tool to peer into the solar interior.  In particular, 
32: modern experiments, both ground- and space-based, have measured 
33: the intermediate degree global oscillation spectrum with high 
34: precision since the beginning of solar cycle 23.  Accurate 
35: determinations of interior structure and dynamics are now possible 
36: (see, e.g., review by \citealt{C-D02}).  These measurements 
37: contain a wealth of information about the fundamental causes of 
38: solar variability.  
39: 
40: It is generally believed that the seat of the solar dynamo is located 
41: at the base of the convection zone \citep[e.g., review by][]{Charbonneau2005}.  
42: Because helioseismology provides the only direct measurements of this 
43: region of the solar interior, these results can play an important role in 
44: constraining dynamo theories.  In particular, a number of authors have attempted 
45: to use global and local helioseismic techniques to determine limits on 
46: the strength of the magnetic field at the base of the convection zone 
47: \citep[e.g.,][and references therein]{Chouetal03}.  In this paper, we attempt to improve 
48: helioseismic measurements of changes in this region.
49: 
50: Global modes of solar oscillation are described by three numbers that
51: characterize the spherical harmonics that are used to define the horizontal
52: structure of the mode. These are (1) radial order $n$ that related to the number
53: of nodes in the radial direction, (2) the degree $\ell$ that is related to
54: the horizontal wavelength of the mode, and (3) the azimuthal order $m$
55: that defines the number of nodes along the equator.  In a spherically 
56: symmetric star, the $2\ell + 1$ modes of an $(n,\ell)$ multiplet are 
57: degenerate, but effects that break spherical symmetry such as magnetic fields or 
58: rotation lift the degeneracy and results in frequency splittings.  
59: The frequencies $\nu_{n\ell m}$ of the modes within a multiplet can be expressed
60: as an expansion in orthogonal polynomials:
61: \begin{equation}
62: \nu_{n\ell m}
63: = \nu_{n\ell} + \sum_{j=1}^{j_{\rm max}} a_j (n,\ell) \, {\cal P}_j^{(\ell)}(m).
64: \label{eq:acoefs}
65: \end{equation}
66: Early investigators \citep[e.g.,][]{Duvalletal86} commonly used Legendre
67: polynomials, whereas now one often uses the
68: Ritzwoller-Lavely formulation of the Clebsch-Gordan expansion \citep{RL91}
69: where the basis functions are polynomials related
70: to the Clebsch-Gordan coefficients.
71: In either case, the coefficients $a_j$ are referred to as $a$-coefficients 
72: or splitting coefficients.  Solar structure is determined by inverting 
73: the mean frequency $\nu_{n\ell}$, while the odd-order coefficients 
74: $a_1, a_3, \ldots$ depend principally on the rotation rate \citep{Durneyetal88}
75: and reflect the advective, latitudinally symmetric part
76: of the perturbations caused by rotation.  Hence, these
77:  are used to determine the rate of rotation inside the Sun.
78: The even order $a$ coefficients on the
79: other hand result from
80: magnetic fields and asphericities in solar structure,
81: and the second order effects of rotation
82: \citep[e.g.,][]{GT90,DG91}.
83: 
84: Solar oscillation frequencies are known to vary on timescales 
85: related to the solar activity cycle.  This was first suggested 
86: by \citet{WN85} and confirmed soon after by \citet{Elsworthetal90} 
87: and \citet{LW90}.  It was quickly established 
88: that the frequency shifts were strongly correlated with surface activity 
89: \citep[etc.]{Woodardetal91,BB93,Elsworthetal94,Reguloetal94}.  \citet{LW90} 
90: observed that the frequency shifts depended very strongly on mode 
91: frequency $\nu$, and very weakly on degree $\ell$ of the mode, and 
92: \citet{A-Getal92} and \citet{Elsworthetal94} confirmed these results.  
93: These authors concluded that all or most of the physical changes 
94: responsible for the changes in frequency were confined to the 
95: shallow layers of the Sun.  In general, this picture has been 
96: confirmed in more recent studies (e.g., observational 
97: results:  \citealt{HKH99,HKH02,BA00,Verneretal04,DG05}, etc., and theoretical 
98: results:  \citealt{Goldetal91,Balmforthetal96,Lietal03}, etc.).  A change in
99: the second helium ionization zone at $r=0.98R_\odot$, first suggested by \citet{Goldetal91} 
100: and \citet{Gough02}, has been confirmed by \citet{BM04} and 
101: \citet{Verneretal06}.  
102: 
103: The even-order mode splitting parameters sample effects of 
104: structural asphericities on the mode 
105: frequency.  \citet{K88} suggested that they were correlated with 
106: observed changes in surface temperature.  Subsequent work has 
107: shown that the aspherical components of the mode frequencies 
108: are tightly correlated with surface magnetic activity 
109: \citep{HKH99,Antiaetal01}.  This high correlation 
110: lends further credence to the idea that frequency shits are 
111: caused by surface and/or near-surface effects.  This can be tested 
112: directly with high degree modes that sample the near-surface 
113: layers of the Sun.  However, as the degree $l$ increases, global 
114: modes become increasingly hard to measure precisely due to the 
115: decrease in mode lifetimes \citep{Rhodesetal98,R-Setal01,Korzenniketal04}.  
116: The lack of reliable measurements of these modes has led some authors 
117: to use ring diagrams to measure high degree modes and measure changes 
118: in the shallow layers of the solar convection zone.  These studies have 
119: confirmed that structural changes do occur in the near-surface layers of the 
120: Sun \citep{Betal07}.
121: 
122: Direct inversions of changes 
123: in the structure of the solar interior probed by the spherically symmetric global 
124: modes have not yielded any measurable differences in the deep 
125: interior \citep{Basu02,E-Detal02} and there have been 
126: upper limits placed on the changes at the base of the convection 
127: zone \citep{E-Detal02}.  \citet{CS05} and \citet{SC05} have presented 
128: evidence of a possible change in mode frequencies with lower turning 
129: points near the base of the convection zone.
130: 
131: Internal dynamics, on the other hand, show clear and unequivocal 
132: evidence of change over the course of the solar cycle.  
133: In the convection zone, bands of different rotational velocities 
134: (called zonal flows) have been shown to migrate poleward at 
135: high latitudes and equator-ward at low latitudes 
136: \citep{Schou99,Howeetal00,AB00,AB01}.  Temporal variations in dynamics 
137: have been shown to penetrate throughout the entire convection 
138: zone, and even below \citep{Vorontsovetal02,BA06,Howeetal05,Howeetal06}.
139: 
140: One conclusion that can be drawn unequivocally from previous 
141: studies of the changes in solar structure is that any changes 
142: deeper than those in the outermost layers of the Sun are very small, and hence 
143: very difficult to detect through their signatures in oscillation 
144: frequencies.  \citet{CS05} used a smoothing technique 
145: to attempt to remove the effect of surface variations, and 
146: found that the scaled frequency differences showed evidence 
147: of change near the base of the convection zone, but could not 
148: say more about the physical nature of the changes.  Attempts 
149: to invert the frequencies directly have never shown any changes 
150: larger than the inversion errors \citep[e.g.,][]{Basu02,E-Detal02}.  
151: Therefore, although current 
152: helioseismic instruments have determined the solar oscillation 
153: frequencies with tremendous precision, statistical errors in 
154: those frequencies are still too large to make any direct detections 
155: of structure changes in the deep interior.
156: 
157: In this work, we take a somewhat different approach to attempting to 
158: detect changes at the base of the convection.  
159: We use Principal Component Analysis (PCA) to separate the frequency 
160: differences over the last solar cycle into a linear combination of 
161: different time-dependent components.  This has the effect of 
162: decreasing the effects of 
163: measurement errors in the measured helioseismic frequencies, which 
164: allows us to isolate as precisely as possible the changes in frequency 
165: over time.  In section \ref{sec:data}, 
166: we describe the data used, and the methods employed to analyze 
167: them.  In section \ref{sec:results}, we present the results in detail.  
168: Finally, in section \ref{sec:conc} we discuss the significance of 
169: the results and presents our conclusions.
170: 
171: 
172: %It is generally believed that the seat of the solar dynamo is located 
173: %at the base of the convection zone and in the tachocline region 
174: %\citep[see, e.g., review by][]{Charbonneau2005}.  Obviously, since 
175: %helioseismology provides the only direct measurements of the properties 
176: %of this region, these results play an important role in motivating and 
177: %constraining various solar dynamo models.  Although a full discussion of 
178: %the possible implications of finding changes in structure at the 
179: %convection zone base is well beyond the scope of this introduction, we 
180: %can point out a few important open questions that discoveries in 
181: %helioseismology could help solve.  The physical nature of the process(es) 
182: %which generate and amplify the torroidal field which emerges at the 
183: %surfarce is not understood.  One important question is whether or not 
184: %the magnetic field is confined in tachocline region for amplification, 
185: %or whether or not the primary dynamo mechanism takes place in the 
186: %convection zone itself \citep[see][for a discussion of the relative merrits 
187: %of various locations for field generation]{B&S2005}.  Furthermore, a 
188: %measurement of the strength of the magnetic field would provide an 
189: %important clue to the nature of the mechanism limitting field 
190: %amplification.
191: %
192: \section{DATA \& ANALYSIS}
193: \label{sec:data}
194: \subsection{Data}
195: For this work, we use helioseismic global-mode data sets from two 
196: different projects, one from the 
197: Michelson Doppler Imager (MDI) on-board the SOHO spacecraft, and the other 
198: from the Global Oscillations Network Group (GONG).  The MDI mode sets 
199: consist of frequencies and splittings obtained from 72-day long time 
200: series \citep{Schou99}.  We use 54 of these sets, spanning the period from  1996 May 1 
201: to 2007 May 16.  The GONG mode sets are derived from 108-day time series \citep{Hilletal96}.  
202: Although GONG provides sets that overlap in time, we only use non-overlapping 
203: sets in the present work.  We use 40 sets from the period 1995 May 7 to
204: 2007 April 14.  Because the two sets are from completely different 
205: instruments and independent data reduction pipelines, any real solar 
206: signatures should show up in both sets.  The $f$-modes do not sample the 
207: deep interior and are dominated by surface effects, so we exclude them 
208: from our study.  The $n = 1$ modes have larger errors than the higher 
209: order modes, and so we exclude them as well.  The included modes are 
210: low and intermediate degree modes up to $\ell=176$, with order $n$ from 
211: 2 to 16.
212: 
213: As a proxy for total solar activity, we use the 10.7cm radio flux 
214: measurements taken by the Dominion Radio Astrophysical Observatory (DRAO)
215: \footnote{Data can be found at \\
216: http://www.drao-ofr.hia-iha.nrc-cnrc.gc.ca/icarus/www/sol\_home.shtml}.  
217: This measurement has been found to be very tightly correlated with 
218: solar activity \citep[e.g.,][]{Tapping87}.  We average the $F_\mathrm{10.7}$ 
219: measurements over 72-day periods 
220: for comparison with MDI data, and 108-day periods for comparison with GONG 
221: data.  For latitudinal structure in surface activity, we use the surface 
222: magnetic field, taken from SOHO/MDI synoptic maps
223: \footnote{MDI synoptic maps of Carrington rotations can be found at\\
224: http://soi.stanford.edu/magnetic/index6.html}.  
225: The magnetic field strengths are averaged over the same 72-day periods 
226: as the $F_{10.7}$ data, and over the appropriate ranges in latitude.
227: 
228: \subsection{Method}
229: We use Principal Component Analysis to describe the temporal variations of the 
230: weighted frequencies as a small number of uncorrelated basis functions.  The use of 
231: PCA is a common technique in multivariate data analysis to reduce dimensionality 
232: and expose underlying variables (see \citealp[Chapter 2]{MH87}, 
233: for a discussion of its astrophysical applications).  
234: A brief description of the method and discussion of its limitations is 
235: included in the Appendix.  PCA is a technique whereby a set of observations 
236: is expressed as a set of uncorrelated vectors.  
237: The usefulness of the technique arises from 
238: the fact that variation of the data about the first vector is maximal, 
239: and about the second vector, maximal subject to the constraint that it 
240: be orthogonal to the first vector, and so on.  In other words, PCA 
241: provides a very efficient linear representation of a data set, and it 
242: is able to substantially reduce the dimensionality of the data set 
243: without losing any significant information.  
244: 
245: It has been known for many years that the frequencies of the solar global 
246: modes of oscillation change with the solar activity cycle.  With the 
247: arrival of high quality measurements of intermediate degree modes, it 
248: is clear that the amount of frequency shift over the solar cycle is 
249: dependent on the mode.  Each mode has an associated mode inertia $E_{nl}$.  
250: Frequency differences can be scaled by the quantity $Q_{n\ell}=E_{n\ell}/\bar{E}_0(\nu_{n\ell})$, 
251: where $\bar{E}_0(\nu_{n\ell})$ is the inertia of the $\ell = 0$ modes 
252: interpolated to the frequency of the $(n,\ell)$ mode 
253: \citep{C-DB91}.  This scaling accounts for 
254: the fact that the frequencies of modes with lower inertia are changed by a 
255: larger amount than modes with a higher inertia by the same underlying perturbation.  
256: When scaled in this way, the degree-dependence of the frequencies over the 
257: solar cycle largely vanish, and the frequency changes become 
258: slowly varying functions of frequency only \citep[e.g.,][]{Chaplinetal01,Basu02}.  
259: %The frequency shifts are known to be correlated with solar activity 
260: %\citep[e.g.,][etc.]{Woodardetal91,BB93,Elsworthetal94,Bhatnagaretal99,HKH99,Jainetal00}.
261: 
262: Our data points are the scaled frequency differences 
263: $Q_{n\ell}\delta \nu_{n\ell} / \nu_{n\ell}$.  For the MDI observations, there 
264: are 54 total mode sets in our work, which, when one is removed to be used 
265: as the base set, giving us 53 sets of scaled frequency differences.  There are 40 
266: mode sets in our GONG data set, or 39 difference sets.  Because the PCA method 
267: requires a complete covariance matrix (see the Appendix), we 
268: can only include 
269: a mode if it is present in all mode sets.  This dramatically reduces the 
270: amount of usable information, particularly since many excluded modes are 
271: missing in only one or two sets out of 53.  For these modes, it is possible to 
272: interpolate a value for the missing frequency differences.  
273: In this case, because most of the frequency differences for the mode 
274: in question will be actually observed, any errors 
275: introduced through the interpolation will have a negligible effect on the 
276: PCA results.  We tested two interpolation methods --- one a spline interpolation 
277: along ridges in the $\ell$-$\nu$ diagram (interpolated from modes with the 
278: same radial order $n$), and the 
279: other a linear interpolation between neighboring modes in time.  When tested 
280: against existing modes, the interpolation along time proved superior, reproducing 
281: the actual data to better than a factor of 1.2$\sigma$, 
282: while the interpolation along the ridge results had a 2$\sigma$ distribution.  
283: Therefore, only results using the time interpolation are discussed in this paper, but the 
284: PCA results using interpolation over degree were entirely consistent.  Finally, to ensure 
285: that the PCA is robust, Monte Carlo simulations were performed to ensure that 
286: the exclusion of certain modes would not affect the results.  The PCA analysis 
287: of these data appears to be very robust.  Errors in the PCA components were 
288: computed by means of a Monte Carlo simulation.
289: 
290: In addition to the mean frequencies $\nu_{n\ell}$, which contain information about 
291: the spherically symmetric part of the solar interior, we have the even-order splitting 
292: coefficients $a_{2j}(n,\ell)$, which allow us to reproduce mode frequencies as a 
293: function of latitude.  Latitudinal frequencies as a function of colatitude 
294: $\theta$ can be obtained as follows:
295: \begin{equation}\label{eq:lat}
296: \nu_{n\ell}(\theta) = \nu_{n\ell} + \sum_k \frac{\ell a_{2k}(n,\ell)}{\mathcal{Q}_{\ell k}}P_{2k}(\cos \theta),
297: \end{equation}
298: where $\mathcal{Q}_{\ell k}$ are the angular integrals given by
299: \begin{equation}
300: \int_0^{2\pi} d\phi \int_0^\pi \sin \theta d\theta Y_\ell^m(Y_\ell^m)^* P_{2k}(\cos \theta) = \frac{1}{\ell}\mathcal{Q}_{\ell k}\mathcal{P}_{2k}^{(\ell)}(m),
301: \end{equation}
302: and $P_{2k}(\cos \theta)$ are Legendre polynomials of degree $2k$, and the 
303: $\mathcal{P}_{2k}^{(\ell)}$ are the same polynomials as in equation \ref{eq:acoefs}.
304: 
305: Ultimately, we are interested in helioseismic data for what they can 
306: tell us about the solar interior.  To extract this information, we invert 
307: these data sets for the parameters of interest.  We 
308: use two different inversion techniques, and invert for the change in 
309: sound speed relative to the comparison frequency.  The first technique used 
310: is Subtractively Optimized Local Averages \citep[SOLA][]{PT94}.  
311: A description of the implementation used here, and how to select 
312: inversion parameters can be found in \citet{R-Setal99}.  The 
313: second inversion technique is that of Regularized Least Squares (RLS).  The 
314: implementation used here and the selection of inversion parameters have been 
315: discussed in \citet{AB94} and \citet{BT96}.  The use 
316: of these two techniques in tandem is crucial because the techniques are 
317: complimentary in nature \citep{Sekii97}.  Inversions can be trusted if 
318: both inversion techniques return the same results.
319: 
320: The quantities which we invert are appropriately scaled eigenvectors from 
321: the PCA of the frequency differences.  The differences are taken relative to 
322: some fiducial set, usually the first mode set (corresponding to activity minimum 
323: at the beginning of solar cycle 23).  The eigenvectors, $\vec{\xi}_i$, are 
324: normalized so that $\vec{\xi}_i \cdot \vec{\xi}_i = 1$.  Each data set is 
325: the vector $\vec{x}(t)$ that contains the individual mode frequencies 
326: $Q_{n \ell}\delta \nu_{n \ell} / \nu_{n \ell}$.  These data can be 
327: represented as a linear combination of the the eigenvectors with
328: coefficients given by $c_i(t) = \sum_j \xi_{ij}x_j(t)$ (see the Appendix).  
329: The scaled eigenvector $c_i(t)\vec{\xi}_i$, therefore, has a physically reasonable 
330: magnitude, and in all following cases, it is this quantity that we invert.  
331: In general, we choose the coefficient with the largest magnitude, which represents 
332: the largest variation in sound speed.  This coefficient is usually the one associated 
333: with the set at maximum activity.
334: 
335: \section{RESULTS \& DISCUSSION}
336: \label{sec:results}
337: \subsection{Mean frequencies}
338: We have performed the PCA decomposition of the MDI mean frequencies with respect 
339: to the first set (set \#1216, start day 1996 May 1, end day 1996 July 12).  This 
340: is a low activity set.  The first four eigenvectors $\vec{\xi}_1$ --- $\vec{\xi}_4$ 
341: are shown in Fig. \ref{fig:MDIvec}, both as a function of frequency and of the 
342: lower turning point of the modes.  
343: The scaling coefficients for the first four eigenvectors are presented in Fig. 
344: \ref{fig:MDIcoef}.  Also shown in the figure is the difference in the 10.7 cm 
345: radio flux between the first set and the subsequent sets.
346: 
347: When plotted as a function of frequency, the first eigenvector $\vec{\xi}_1$ 
348: appears to be almost entirely due to near-surface effects --- it is seems to 
349: be a slowly varying function of frequency 
350: only.  As a function  of $r_t$, however, a change in the average level 
351: of the frequency differences can be seen below the base of the convection zone 
352: (around $r \approx 0.713R_\odot$ --- marked by a vertical line in 
353: Fig. \ref{fig:MDIvec}).  This implies a time 
354: dependent change near the base of the solar convection zone.  We conclude, therefore, 
355: that there is a statistically significant component of the frequency 
356: variability picked out by $\vec{\xi}_1$ that does not originate at the surface.
357: The coefficients for $\vec{\xi}_1$ are tightly correlated with the 10.7cm 
358: radio flux, a proxy for surface activity; the correlation coefficient is 0.99.  
359: A linear regression fit to the change in radio flux relative to solar minimum, 
360: $\Delta F_\mathrm{10.7}$, gives the relation between the 10.7cm flux and the 
361: coefficients of $\vec{\xi}_1$ as:
362: \begin{equation}
363: \label{eq:f10line}
364: c_1 = (2.53\times 10^{-5} \pm 3.13 \times 10^{-6})\Delta F_\mathrm{10.7} + 3.95\times 10^{-5} \pm 1.6 \times 10^{-4}.
365: \end{equation}
366: Thus, all changes, as manifest by $\vec{\xi}_1$, are tightly correlated with 
367: surface magnetic activity.
368: 
369: The second eigenvector, $\vec{\xi}_2$, also shows variability over the solar cycle.  
370: Comparing it as a function of $\nu$ and of $r_t$, it is clear that the structure is neither 
371: a pure function of frequency nor of lower turning point.  The first four obvious 
372: downturn features correspond to modes of order $n=2$, 3, 4, and 5.  As a 
373: function of $r_t$, it seems clear that the structure is concentrated at or 
374: near the surface.  However, the differences can not be fit by the usual ``surface 
375: term", and further examination reveals that it cannot be fit even by higher-order 
376: surface terms of the form considered in \citet{BV93} and \citet{A95}.  The coefficients for 
377: $\vec{\xi}_2$ offer some hint as to what is going on.  They exhibit no obvious 
378: solar cycle dependency, but rather seem to be a roughly linearly decreasing 
379: function of time.  There does not seem to be any periodicity on a scale of 
380: eleven years or shorter.  Larson \& Schou (2008) have undertaken an in depth study of the
381: systematics in the MDI data reduction pipeline.  The plate scale in the
382: MDI instrument has changed slightly over the course of SOHO's mission,
383: and they have shown that the effect of the resultant
384: error in the measured radius introduces errors that look exactly like
385: the $\vec{\xi}_2$ eigenvector computed in this work.  We do not believe, therefore, 
386: that the features in $\vec{\xi}_2$ are solar in origin, but are rather artifacts 
387: from the MDI data reduction pipeline.  As we will show below, analysis of 
388: GONG data confirms our belief.
389: 
390: The third and fourth eigenvectors, $\vec{\xi}_3$ and $\vec{\xi}_4$ do not exhibit 
391: any significant structure at all.  The vector $\vec{\xi}_3$ shows a slight trend 
392: with frequency, but the scaling coefficients $c_3$ are normally distributed, and 
393: we cannot identify any physical significance in this eigenvector.  The remaining 
394: eigenvectors are statistically consistent with Gaussian noise distributed around zero.
395: We conclude, therefore, that the temporal variation of the MDI frequencies 
396: is dependent on a linear combination of $\vec{\xi}_1$ and $\vec{\xi}_2$ alone.  
397: In Fig. \ref{fig:MDIrecon}, we show two data sets reconstructed from the 
398: first two eigenvectors.  This figure 
399: shows that the PCA decomposition does indeed accurately capture the original data 
400: while significantly reducing the random scatter in the data.  The residuals 
401: normalized by the errors are plotted, and are consistent with Gaussian noise, with 
402: distributions of 1.1$\sigma$ and 0.9$\sigma$ for the two cases.  Having confirmed that 
403: the third and subsequent eigenvectors are Gaussian noise, we do not consider them 
404: further in this paper.  
405: This reduction in noise is important for attempting to invert the small signatures 
406: we are looking at here.
407: 
408: The fact that the PCA is applied to a set of mode sets relative to a single 
409: base set raises the possibility that we are unduly influenced by the choice 
410: of that base set.  We therefore repeat the PCA taking a base set from 
411: halfway up the solar cycle:  MDI set \#2224 (start date: 1999 February 3, end date 
412: 1999 April 16, and an activity level during the 72 day period of $F_\mathrm{10.7}=130.7$ 
413: SFU).  The eigenvectors are consistent with those obtained from the base 
414: \#1216 set insofar as their inner products are close to unity. We conclude, therefore, 
415: that the PCA results are not unduly influenced by the choice of base set.
416: 
417: The mode parameters measured from GONG data are somewhat noisier 
418: than those measured from MDI data.  Nevertheless, a similar 
419: analysis of the GONG data allows us to confirm results obtained from 
420: MDI data.  The first two eigenvectors from the GONG observations 
421: are shown in Fig. \ref{fig:GONGvec}.  As with the MDI data, $\vec{\xi}_1$ plotted 
422: against $r_t$ shows some structure in the deep interior.  The second 
423: eigenvector, $\vec{\xi}_2$, shows no more structure than the MDI $\vec{\xi}_3$ 
424: eigenvector, and the remaining eigenvectors appear to be Gaussian noise, 
425: reinforcing our conclusions that the structure in $\vec{\xi}_2$ from the MDI data 
426: set is instrumental in origin.
427: 
428: We invert the appropriately scaled $\vec{\xi}_1$ eigenvectors to determine the 
429: change in the sound speed as a function of radius.  
430: We show the results of the inversion of the $\vec{\xi}_1$ vector with the 
431: \#1216 base set in Fig. \ref{fig:MDIinv}.  This vector has been scaled by 
432: the coefficient for set \#3160 (start date:  2001 August 27) in order to 
433: give the inversion results physical meaning.  It 
434: is readily apparent that at a depth of $r \approx 0.713R_\odot$, near the 
435: location of the convection zone base, there is a change in the sound speed.  
436: This feature is well matched in both the RLS and the SOLA inversions, 
437: which implies that the feature is actually present in the data.  The depression 
438: in sound speed at the base of the convection zone with increasing activity is 
439: matched with a corresponding enhancement below the convection zone.  
440: We invert the GONG data as well.  The inversion results are shown in Fig. 
441: \ref{fig:GONGinv}.  There is a clear feature at the base of the convection 
442: zone, as seen with MDI data.  The presence of this feature in 
443: the data of an independent instrument with an independent reduction pipeline is 
444: very encouraging --- it strongly implies that the changes implied by the 
445: inversions are present in the Sun itself rather than artifacts in the data.  
446: Figure \ref{fig:MDIinv} shows the difference in sound speed between two extrema 
447: in the solar cycle.  To show how the interior changes with time, in Fig. 
448: \ref{fig:MDIcsq} we show the sound speed inversions at three radii as a function 
449: of $\Delta F_\mathrm{10.7}$.
450: 
451: We confirm, therefore, that the signature in mode frequencies is consistent 
452: with a change in structure at the base of the convection zone.  Other 
453: authors have looked for changes in this region, but have not found any changes.  
454: The change that we have detected, while statistically significant, is 
455: very small, and it is only with the benefit of an entire solar cycle's 
456: worth of high precision observations that we can detect changes at this level.  
457: \citet{BA01} examined the mode frequencies for evidence of a 
458: change in the location of the base of the convection zone.  They did not 
459: detect any change, and the sound speed profile that we find in Fig. \ref{fig:MDIinv} 
460: is very different from the one they expected from a change in the base of 
461: the convection zone.  This implies that, even if the change we are detecting 
462: is thermal in nature, it is unlikely to be related in any way to a change 
463: in the position of the base of the convection zone.
464: 
465: These inversions have been done assuming that the frequency differences are a result 
466: of a change in sound speed only.  It is almost certain, however, given how tightly 
467: correlated this change is with solar activity, that the observed changes are 
468: related in some way to changes in the internal magnetic fields.  What we have really 
469: inverted for, therefore, is a change in the {\em wave} speed.   If we assume 
470: that the entire change is due to a change in the wave propagation speed due the presence of 
471: magnetic fields, in other words that $\delta c^2 / c^2 \approx v_A^2 / c^2$, as in 
472: \citet{Betal04}, we can obtain a value for $B$.  The change at the 
473: base of the convection zone is $\delta c^2 / c^2 = (7.23 \pm 2.08) \times 10^{-5}$, 
474: which implies a magnetic field strength of 290 kG.  This is consistent with 
475: the results of earlier authors --- \citet{GD93} placed an upper limit of 1MG on the 
476: toroidal field at the base of the convection zone, and \citet{Basu97} found that the 
477: magnetic field in this region could not exceed 300kG.  \citet{CS02} found somewhat 
478: stronger fields (400 to 700kG).
479: 
480: 
481: \subsection{Latitudinal changes}
482: The MDI and GONG data sets also contain splitting coefficients.  The even-order 
483: coefficients contain information about the non-spherically symmetric structure 
484: in the solar interior.  Because the surface manifestations of solar activity 
485: are strongly latitudinally dependent, we have used these coefficients to study the 
486: temporal variability of structure at different latitudes.  The frequencies 
487: corresponding to different latitudes are computed using equation \ref{eq:lat}.  
488: The PCA procedure is performed for each latitude as was done with the mean frequencies, 
489: and as usual is done with respect to set \#1216.  
490: The first eigenvector for six different latitudes is shown in Fig. \ref{fig:LATvec}.  
491: When plotted as a function of frequency, the latitudes from the equator to 30$^\circ$ 
492: show a similar frequency dependence as in the case of the mean frequencies in 
493: Fig. \ref{fig:MDIvec}.  When plotted as a function of $r_t$, we see change at 
494: and below the convection zone base.  The higher latitudes show no 
495: structure, and the eigenvectors for these latitudes are consistent with Gaussian 
496: noise.  The scaling coefficients for each latitude as a function of time are 
497: shown in Fig. \ref{fig:LATcoef}, along with the surface magnetic field.  Like 
498: the scaling coefficients for the mean frequencies shown in Fig. \ref{fig:MDIcoef}, 
499: the latitudinal scaling coefficients closely follow the surface activity.
500: 
501: We show the sound speed inversions for the equator, 15$^\circ$, 30$^\circ$, 
502: and 45$^\circ$ in Fig. \ref{fig:LATinv}.  The errors in the eigenvectors are 
503: larger here than for the mean frequencies, in large part because each frequency 
504: is a combination of mean frequency and splitting coefficients, each with their own 
505: errors.  Nevertheless, there are several points of interest 
506: in these inversions.  The first is a clear sound speed change for radii greater than  
507: approximately $r=0.86R_\odot$ at 15$^\circ$ and the equator.  The change seen 
508: in the SOLA inversion results is well matched in this region by the RLS inversion results, 
509: and the significance of the change approaches 2$\sigma$.  There is the 
510: possibility, though less statistically significant, of a change at greater depth, 
511: i.e., approximately $r=0.82R_\odot$.  At 30$^\circ$, a change in sound speed through 
512: the tachocline is seen in the RLS results, but it does not appear to be as clear 
513: in the SOLA results..  It is unclear whether or not this result is statistically 
514: significant.  At latitudes of 45$^\circ$ and higher, the inversion errors become large, 
515: and the inversion results themselves become extremely sensitive to the inversion parameters.  
516: We conclude that there are no structural changes in the solar interior at 
517: or above a latitude of 45$^\circ$ large enough to be present in our data sets.
518: 
519: We have also analyzed directly the change in solar structural asphericity over 
520: the course of the solar cycle, by taking the scaled frequency differences of 
521: high latitudes with respect to the equator.  For this analysis, therefore, we have 
522: 54 mode sets at each latitude.  The difference is equator$-$latitude.  
523: Figure \ref{fig:ASPHvec} shows the 
524: $\vec{\xi}_1$ eigenvectors for five different latitudes with respect to the equator.  
525: Clearly, the signal to noise in these data are worse than either the latitudinal 
526: frequencies or the mean frequencies, but radial structure is discernible in the 
527: eigenvectors for the equator relative to 15$^\circ$, 30$^\circ$, and 45$^\circ$.  
528: In fig. \ref{fig:ASPHcoef}, the scaling coefficients for the asphericity terms 
529: are shown.  There is an evident phase delay in the scaling coefficients of 15$\circ$ 
530: and 30$^\circ$ with respect to the equator.  This is expected from Fig. \ref{fig:LATcoef}.  
531: The scaling coefficients for higher latitudes (45$^\circ$ and 60$^\circ$ with respect 
532: to the equator) show no structure except for an abrupt change from mostly positive 
533: to negative, corresponding closely to the peak of the solar activity cycle.
534: 
535: \section{CONCLUSIONS}
536: \label{sec:conc}
537: We have analyzed the changes in solar oscillation frequencies over the 
538: course of solar cycle 23.  In order to reduce the effects of 
539: measurement errors and detect the faintest signatures of solar 
540: variability possible, we have employed a Principal Component Analysis 
541: of the frequencies.  The mean frequencies are known to vary over the 
542: solar cycle, and this variation is known to be tightly correlated with 
543: surface activity.  We have found this correlation as well.  
544: 
545: In addition to the frequency dependent change which these 
546: earlier authors have detected, we have found a small but statistically 
547: significant change in modes with turning points at or below the convection 
548: zone.  This confirms the result of \citet{CS05}.  This 
549: signature is present in both the MDI and the GONG data sets.  We have 
550: inverted these results to obtain the difference in sound speed, and we 
551: have confirmed that there is a change in solar structure at the base of 
552: the convection zone over the course of the solar cycle.  The measured 
553: change at a radius of $r=(0.712^{+0.0097}_{-0.0029})R_\odot$ is 
554: $\delta c^2 / c^2 = (7.23 \pm 2.08) \times 10^{-5}$, where 
555: the errors in radius are a measure of the resolution of the inversion 
556: taken from the first and third quartile points of the inversion kernel.
557: 
558: We have also used the splitting coefficients to investigate how the 
559: changes in structure vary over latitude.  We have found that the 
560: changes in the solar interior are tightly correlated with the 
561: latitudinal distribution of surface activity.  The most statistically 
562: significant changes detected in the analysis are changes in sound 
563: speed in the outer ten percent (by radius) of the solar interior.
564: 
565: \acknowledgements
566: We wish to thank the anonymous referee for helpful comments which helped 
567: to clarify a number of the discussions in this paper.  We are also grateful 
568: to T.~ Larson and J.~ Schou for their helpful discussion of the anomalous 
569: results in the MDI data.  
570: This work  utilizes data obtained by the Global Oscillation
571: Network Group (GONG) project, managed by the National Solar Observatory,
572: which is operated by AURA, Inc. under a cooperative agreement with the
573: National Science Foundation. The data were acquired by instruments
574: operated by the Big Bear Solar Observatory, High Altitude Observatory,
575: Learmonth Solar Observatory, Udaipur Solar Observatory, Instituto de
576: Astrofisico de Canarias, and Cerro Tololo Inter-American Observatory.
577: This work also utilizes data from the Solar Oscillations
578: Investigation/ Michelson Doppler Imager (SOI/MDI) on the Solar
579: and Heliospheric Observatory (SOHO).  SOHO is a project of
580: international cooperation between ESA and NASA.
581: MDI is supported by NASA grants NAG5-8878 and NAG5-10483
582: to Stanford University. This work is partially supported by NSF grants 
583: ATM 0348837 and ATM 073770 to SB.
584: 
585: 
586: \appendix
587: \section{PRINCIPAL COMPONENT ANALYSIS}
588: \label{sec:PCAappend}
589: We give a brief description of the Principal Component Analysis 
590: technique that follows the one found in \citet{Kendall80}.  This technique assumes 
591: that we begin with a set of observations, each one consisting of a set of data 
592: points (for example, a set of spectra of different objects, or a set of images, or, 
593: as in this paper, sets of mode frequency measurements at different points in time).  
594: Assume we have 
595: $m$ observations, each with $n$ data points.  The $m$ vectors $\vec{x}_i$ each 
596: contain the $n$ data points $x_{ij}$, measured relative to the mean $\bar{\vec{x}}_j$
597: We wish to find a new set of vectors $\vec{c}_i$ that are linearly dependent 
598: on $\vec{x}_i$, and uncorrelated with each other.  In addition, we require that they 
599: have stationary values of their variance.  This condition is imposed to ensure that 
600: most of the variance is accounted for in as few vectors $\vec{c}_i$ as possible.  
601: Alternatively, this condition can be viewed geometrically as a rotation of the basis 
602: vectors such that the variance of the data with respect to each basis vector is maximal.  
603: This is most easily understood in the case of a two dimensional data set, where PCA is 
604: equivalent to a linear fit to the data, and the rotation sets one basis vector parallel 
605: to that fit.  
606: The vectors $\vec{c}_i$ will be given by a linear combination of the original observations:
607: \begin{equation}\label{eq:AP1}
608: \vec{c}_i = \sum_{j = 1}^n \xi_{ji}\vec{x}_j.
609: \end{equation}
610: The vectors $\vec{\xi}_j$ will form a new basis set for the observations.  The variance of 
611: $\vec{c}_i$ is given by
612: \begin{equation}\label{eq:var}
613: \textrm{var}\ \vec{c}_i = \sum_{j=1}^n \sum_{k=1}^n \xi_{ji} \xi_{ki} e_{jk},
614: \end{equation}
615: where $e_{jk}$ is the covariance between $x_j$ and $x_k$.  The covariance matrix is given
616: by
617: \begin{equation}\label{cov}
618: \mathbf{E} = \frac{1}{n}\mathbf{XX}^\mathrm{T},
619: \end{equation}
620: where $\mathbf{X}$ is the matrix $\left( \vec{x}_1, \vec{x}_2, \ldots, \vec{x}_m \right)$.  In order to 
621: get unique solutions for the values of $\xi_{ji}$, we impose a normalization condition:
622: \begin{equation}\label{eq:constraint}
623: \sum_{i=1}^m \xi_{ji}^2 = 1.
624: \end{equation}
625: Finding the stationary values of $\textrm{var}\ \vec{c}_i$ (eq. \ref{eq:var}) with the 
626: condition \ref{eq:constraint} is equivalent to finding the stationary values of:
627: \begin{equation}\label{eq:newvar}
628: \sum_{j=1}^m \sum_{k=1}^m \xi_{ij} \xi_{ik} e_{jk} - \lambda \left( \sum_{j=1}^m \xi_{ij}^2 - 1 \right),
629: \end{equation}
630: where $\lambda$ is some constant.  To find 
631: the stationary values, we differentiate \ref{eq:newvar} by $\xi_{ij}$ and find the roots 
632: of the equation:
633: \begin{equation}
634: \sum_{k=1}^m \xi_{ik} e_{jk} - \lambda \xi_{ij} = 0.
635: \end{equation}
636: This is an eigenvalue problem --- the vector $\vec{\xi}_j$ is an eigenvector of
637: $\mathbf{E}$ and $\lambda$ is the corresponding eigenvalue.  The eigenvectors form an orthonormal 
638: basis set and are called the `principal components' of the data set.  When they are
639: ordered by decreasing eigenvalue $\lambda$, the first principal component will have 
640: the largest possible variance, the second the second largest variance, and so on.  In 
641: our work, we use the singular value decomposition of the covariance matrix to get the 
642: eigenvalues and vectors.  The vectors $\vec{c}_i$ are the scaling coefficients for the 
643: principal components.
644: 
645: The PCA technique has several known weaknesses.  The first is that its simplest
646: implementation requires a completely filled covariance matrix --- in other words
647: there can be no missing data in $\mathbf{X}$.  This is a problem for us  
648: since not all modes are identified
649: in each different observation epoch.  There are generalizations
650: of the technique which allow for large quantities of missing data.  In our case, however,
651: the modes in which we are interested tend to be identified in most of the mode sets, so
652: we can simply interpolate the missing modes from the existing frequency measurements 
653: without any significant effect on the results.  We do not, therefore, use a more 
654: specialized technique.
655: 
656: A second problem with PCA is its sensitivity to outliers.  One often-quoted example 
657: shows a PCA decomposition where the correlation coefficient between the first two 
658: components with one outlier removed is 0.99 \citep[e.g.,][]{Huber81}.  There are 
659: available routines for making PCA more robust, but in our case, because the 
660: dimensionality of the problem is relatively small, we can instead empirically test 
661: the sensitivity of the data set to outliers.  Monte Carlo tests show that our data 
662: set is not prone to errors in the PCA due to outliers.
663: 
664: Principal Component Analysis has been used in a wide variety of astronomical contexts 
665: (see references in \citealp[Chapter2]{MH87}).  In solar physics, PCA has been used for, 
666: among other things, the inversion of Stokes profiles \citep[e.g.,][]{Eydenbergetal05,R-Vetal06}, in 
667: detecting structures in coronal activity \citep[e.g.,][]{Cadavidetal07}, and in helioseismic 
668: rotation inversions \citep[e.g.,][]{E-Detal04}.
669: 
670: 
671: \begin{thebibliography}{}
672: \bibitem[Anguera Gubau et al.(1992)]{A-Getal92} Anguera Gubau, 
673: M., Palle, P.~L., Perez Hernandez, F., Regulo, C., \& Roca Cortes, T.\ 
674: 1992, \aap, 255, 363
675: 
676: \bibitem[Antia(1995)]{A95} Antia, H.~M.\ 1995, \mnras, 274, 
677: 499 
678:  
679: \bibitem[Antia \& Basu(1994)]{AB94} Antia, H.~M., \& Basu, 
680: S.\ 1994, \aaps, 107, 421 
681: 
682: \bibitem[Antia \& Basu(2000)]{AB00} Antia, H.~M., \& Basu, 
683: S.\ 2000, \apj, 541, 442 
684: 
685: \bibitem[Antia \& Basu(2001)]{AB01} Antia, H.~M., \& Basu, 
686: S.\ 2001, \apjl, 559, L67 
687: 
688: \bibitem[Antia et al.(2001)]{Antiaetal01} Antia, H.~M., Basu, S., 
689: Hill, F., Howe, R., Komm, R.~W., \& Schou, J.\ 2001, SOHO 10/GONG 2000 
690: Workshop: Helio- and Asteroseismology at the Dawn of the Millennium, 464, 45 
691: 
692: \bibitem[Bachmann \& Brown(1993)]{BB93} Bachmann, K.~T., 
693: \& Brown, T.~M.\ 1993, \apjl, 411, L45 
694: 
695: \bibitem[Balmforth et al.(1996)]{Balmforthetal96} Balmforth, N.~J., 
696: Gough, D.~O., \& Merryfield, W.~J.\ 1996, \mnras, 278, 437 
697: 
698: \bibitem[Basu(1997)]{Basu97} Basu, S.\ 1997, \mnras, 288, 572 
699: 
700: \bibitem[Basu(2002)]{Basu02} Basu, S.\ 2002, From Solar Min to 
701: Max: Half a Solar Cycle with SOHO, 508, 7 
702: 
703: \bibitem[Basu \& Antia(2000)]{BA00} Basu, S., \& Antia, 
704: H.~M.\ 2000, \solphys, 192, 449 
705: 
706: \bibitem[Basu \& Antia(2001)]{BA01} Basu, S., \& Antia, 
707: H.~M.\ 2001, SOHO 10/GONG 2000 Workshop: Helio- and Asteroseismology 
708: at the Dawn of the Millennium, 464, 297 
709: 
710: \bibitem[Basu \& Antia(2006)]{BA06} Basu, S., \& Antia, 
711: H.~M.\ 2006, Proceedings of SOHO 18/GONG 2006/HELAS I, Beyond the 
712: spherical Sun, 624, 128
713: 
714: \bibitem[Basu et al.(2004)]{Betal04} Basu, S., Antia, H.~M., 
715: \& Bogart, R.~S.\ 2004, \apj, 610, 1157 
716: 
717: \bibitem[Basu et al.(2007)]{Betal07} Basu, S., Antia, H.~M., 
718: \& Bogart, R.~S.\ 2007, \apj, 654, 1146 
719: 
720: \bibitem[Basu \& Mandel(2004)]{BM04} Basu, S., \& Mandel, 
721: A.\ 2004, \apjl, 617, L155 
722: 
723: \bibitem[Basu \& Thompson(1996)]{BT96} Basu, S., \& 
724: Thompson, M.~J.\ 1996, \aap, 305, 631 
725: 
726: %\bibitem[Bhatnagar et al.(1999)]{Bhatnagaretal99} Bhatnagar, A., Jain, 
727: %K., \& Tripathy, S.~C.\ 1999, \apj, 521, 885
728: %
729: %
730: %\bibitem[Brandenburg \& Subramanian(2005)]{B&S2005} Brandenburg, A., 
731: %\& Subramanian, K.\ 2005, \physrep, 417, 1
732: %
733: \bibitem[Brodsky \& Vorontsov(1993)]{BV93} Brodsky, M., \& Vorontsov, 
734: S.~V.\ 1993, \apj, 409, 455 
735: 
736: \bibitem[Cadavid et al.(2007)]{Cadavidetal07} Cadavid, A.~C., 
737: Lawrence, J.~K., \& Ruzmaikin, A.\ 2008, \solphys, 248, 247
738: 
739: \bibitem[Chaplin et al.(2001)]{Chaplinetal01} Chaplin, W.~J., 
740: Elsworth, Y., Isaak, G.~R., New, R., \& Appourchaux, T.\ 2001, SOHO 
741: 10/GONG 2000 Workshop: Helio- and Asteroseismology at the Dawn of the 
742: Millennium, 464, 83 
743: 
744: \bibitem[Charbonneau(2005)]{Charbonneau2005} Charbonneau, P.\ 2005, 
745: Living Reviews in Solar Physics, 2, 2
746: 
747: \bibitem[Chou \& Serebryanskiy(2002)]{CS02} Chou, D.-Y., \& 
748: Serebryanskiy, A.\ 2002, \apjl, 578, L157
749: 
750: \bibitem[Chou \& Serebryanskiy(2005)]{CS05} Chou, D.-Y., 
751: \& Serebryanskiy, A.\ 2005, \apj, 624, 420 
752: 
753: \bibitem[Chou et al.(2003)]{Chouetal03} Chou, D.-Y., Serebryanskiy, 
754: A., \& Sun, M.-T.\ 2003, Space Science Reviews, 107, 35
755: 
756: \bibitem[Christensen-Dalsgaard(2002)]{C-D02} 
757: Christensen-Dalsgaard, J.\ 2002, Reviews of Modern Physics, 74, 1073 
758: 
759: \bibitem[Christensen-Dalsgaard \& Berthomieu(1991)]{C-DB91} 
760: Christensen-Dalsgaard, J., \& Berthomieu, G.\ 1991, Solar Interior and 
761: Atmosphere, 401 
762: 
763: \bibitem[Durney et al.(1988)]{Durneyetal88} Durney, B.~R., Hill, F., 
764: \& Goode, P.~R.\ 1988, \apj, 326, 486 
765: 
766: \bibitem[Duvall et al.(1986)]{Duvalletal86} Duvall, T.~L., Jr., 
767: Harvey, J.~W., \& Pomerantz, M.~A.\ 1986, \nat, 321, 500 
768: 
769: \bibitem[Dziembowski \& Goode(1991)]{DG91} Dziembowski, 
770: W.~A., \& Goode, P.~R.\ 1991, Solar Interior and Atmosphere, 501 
771: 
772: %\bibitem[Dziembowski \& Goode(2002)]{DG02} Dziembowski, 
773: %W.~A., \& Goode, P.~R.\ 2002, IAU Colloq.~185: Radial and Nonradial 
774: %Pulsationsn as Probes of Stellar Physics, 259, 476 
775: 
776: \bibitem[Dziembowski \& Goode(2005)]{DG05} Dziembowski, 
777: W.~A., \& Goode, P.~R.\ 2005, \apj, 625, 548 
778: 
779: \bibitem[Eff-Darwich et al.(2004)]{E-Detal04} Eff-Darwich, A., 
780: Korzennik, S.~G., Jim{\'e}nez-Reyes, S., \& Garc{\'{\i}}a, R.~A.\ 2004, 
781: SOHO 14 Helio- and Asteroseismology: Towards a Golden Future, 559, 420 
782: 
783: \bibitem[Eff-Darwich et al.(2002)]{E-Detal02} Eff-Darwich, A., 
784: Korzennik, S.~G., Jim{\'e}nez-Reyes, S.~J., \& P{\'e}rez Hern{\'a}ndez, 
785: F.\ 2002, \apj, 580, 574 
786: 
787: \bibitem[Elsworth et al.(1994)]{Elsworthetal94} Elsworth, Y., Howe, 
788: R., Isaak, G.~R., McLeod, C.~P., Miller, B.~A., New, R., Speake, C.~C., 
789: \& Wheeler, S.~J.\ 1994, \apj, 434, 801 
790: 
791: \bibitem[Elsworth et al.(1990)]{Elsworthetal90} Elsworth, Y., Howe, 
792: R., Isaak, G.~R., McLeod, C.~P., \& New, R.\ 1990, \nat, 345, 322 
793: 
794: \bibitem[Eydenberg et al.(2005)]{Eydenbergetal05} Eydenberg, M.~S., 
795: Balasubramaniam, K.~S., \& L{\'o}pez Ariste, A.\ 2005, \apj, 619, 1167 
796: 
797: \bibitem[Goldreich et al.(1991)]{Goldetal91} Goldreich, P., 
798: Murray, N., Willette, G., \& Kumar, P.\ 1991, \apj, 370, 752 
799: 
800: \bibitem[Goode \& Dziembowski(1993)]{GD93} Goode, P.~R., 
801: \& Dziembowski, W.~A.\ 1993, GONG 1992.~Seismic Investigation of the Sun 
802: and Stars, 42, 229
803: 
804: \bibitem[Gough(2002)]{Gough02} Gough, D.~O.\ 2002, From Solar 
805: Min to Max: Half a Solar Cycle with SOHO, 508, 577 
806: 
807: \bibitem[Gough \& Thompson(1990)]{GT90} Gough, D.~O., 
808: \& Thompson, M.~J.\ 1990, \mnras, 242, 25 
809: 
810: \bibitem[Hill et al.(1996)]{Hilletal96} Hill, F., et al.\ 1996, 
811: Science, 272, 1292 
812: 
813: \bibitem[Howe et al.(2000)]{Howeetal00} Howe, R., 
814: Christensen-Dalsgaard, J., Hill, F., Komm, R.~W., Larsen, R.~M., Schou, J., 
815: Thompson, M.~J., \& Toomre, J.\ 2000, Science, 287, 2456 
816: 
817: \bibitem[Howe et al.(2005)]{Howeetal05} Howe, R., 
818: Christensen-Dalsgaard, J., Hill, F., Komm, R., Schou, J., 
819: \& Thompson, M.~J.\ 2005, \apj, 634, 1405 
820: 
821: \bibitem[Howe et al.(1999)]{HKH99} Howe, R., Komm, R., 
822: \& Hill, F.\ 1999, \apj, 524, 1084 
823: 
824: \bibitem[Howe et al.(2002)]{HKH02} Howe, R., Komm, R.~W., 
825: \& Hill, F.\ 2002, \apj, 580, 1172 
826: 
827: \bibitem[Howe et al.(2006)]{Howeetal06} Howe, R., Rempel, M., 
828: Christensen-Dalsgaard, J., Hill, F., Komm, R., Larsen, R.~M., Schou, J., 
829: \& Thompson, M.~J.\ 2006, \apj, 649, 1155 
830: 
831: \bibitem[Huber(1981)]{Huber81} Huber, P.~J., 1981, Robust Statistics, 
832: (New York, NY: Wiley)
833: 
834: %\bibitem[Jain et al.(2000)]{Jainetal00} Jain, K., Tripathy, S.~C., 
835: %\& Bhatnagar, A.\ 2000, \apj, 542, 521 
836: 
837: \bibitem[Kendall(1980)]{Kendall80} Kendall, M. 1980, Multivariate 
838: Analysis, (2nd Ed.; New York, NY: Macmillan Publishing Co)
839: 
840: \bibitem[Korzennik et al.(2004)]{Korzenniketal04} Korzennik, S.~G., 
841: Rabello-Soares, M.~C., \& Schou, J.\ 2004, \apj, 602, 481 
842: 
843: \bibitem[Kuhn(1988)]{K88} Kuhn, J.~R.\ 1988, \apjl, 331, 
844: L131 
845: 
846: \bibitem[Larson \& Schou(2008)]{LarsonSchou08} Larson, T., \& Schou, J.\ 2008, 
847: in Helioseismology, Asteroseismology and MHD Connections, J.~ Phys.: Conf.~ Ser.,
848: ed. Gizon, L., in press
849: 
850: \bibitem[Li et al.(2003)]{Lietal03} Li, L.~H., Basu, S., Sofia, 
851: S., Robinson, F.~J., Demarque, P., \& Guenther, D.~B.\ 2003, \apj, 591, 1267 
852: 
853: \bibitem[Libbrecht \& Woodard(1990)]{LW90} Libbrecht, 
854: K.~G., \& Woodard, M.~F.\ 1990, \nat, 345, 779 
855: 
856: \bibitem[Murtagh \& Heck(1987)]{MH87} Murtagh, F., \& Heck, 
857: A.\ 1987, Astrophysics and Space Science Library, 131
858: 
859: \bibitem[Pijpers \& Thompson(1994)]{PT94} Pijpers, F.~P., 
860: \& Thompson, M.~J.\ 1994, \aap, 281, 231 
861: 
862: \bibitem[Rabello-Soares et al.(1999)]{R-Setal99} Rabello-Soares, 
863: M.~C., Basu, S., \& Christensen-Dalsgaard, J.\ 1999, \mnras, 309, 35 
864: 
865: \bibitem[Rabello-Soares et al.(2001)]{R-Setal01} Rabello-Soares, 
866: M.~C., Korzennik, S.~G., \& Schou, J.\ 2001, SOHO 10/GONG 2000 Workshop: 
867: Helio- and Asteroseismology at the Dawn of the Millennium, 464, 129 
868: 
869: \bibitem[R{\'a}mirez V{\'e}lez et al.(2006)]{R-Vetal06} 
870: R{\'a}mirez V{\'e}lez, J.~C., Semel, M., Stift, M.~J., \& Leone, F.\ 2006, 
871: Astronomical Society of the Pacific Conference Series, 358, 405 
872: 
873: \bibitem[Regulo et al.(1994)]{Reguloetal94} Regulo, C., Jimenez, A., 
874: Palle, P.~L., Perez Hernandez, F., \& Roca Cortes, T.\ 1994, \apj, 434, 384 
875: 
876: \bibitem[Rhodes et al.(1998)]{Rhodesetal98} Rhodes, E.~J., Jr., 
877: Reiter, J., Kosovichev, A.~G., Schou, J., \& Scherrer, P.~H.\ 1998, 
878: Structure and Dynamics of the Interior of the Sun and Sun-like Stars, 418, 73 
879: 
880: \bibitem[Ritzwoller \& Lavely(1991)]{RL91} Ritzwoller, 
881: M.~H., \& Lavely, E.~M.\ 1991, \apj, 369, 557 
882: 
883: \bibitem[Schou(1999)]{Schou99} Schou, J.\ 1999, \apjl, 523, 
884: L181 
885: 
886: \bibitem[Sekii(1997)]{Sekii97} Sekii, T.\ 1997, Sounding Solar 
887: and Stellar Interiors, 181, 189 
888: 
889: \bibitem[Serebryanskiy \& Chou(2005)]{SC05} Serebryanskiy, 
890: A., \& Chou, D.-Y.\ 2005, \apj, 633, 1187 
891: 
892: \bibitem[Tapping(1987)]{Tapping87} Tapping, K.~F.\ 1987, \jgr, 
893: 92, 829 
894: 
895: \bibitem[Verner et al.(2004)]{Verneretal04} Verner, G.~A., Chaplin, 
896: W.~J., \& Elsworth, Y.\ 2004, \mnras, 351, 311 
897: 
898: \bibitem[Verner et al.(2006)]{Verneretal06} Verner, G.~A., Chaplin, 
899: W.~J., \& Elsworth, Y.\ 2006, \apjl, 640, L95 
900: 
901: \bibitem[Woodard \& Noyes(1985)]{WN85} Woodard, M.~F., 
902: \& Noyes, R.~W.\ 1985, \nat, 318, 449 
903: 
904: \bibitem[Woodard et al.(1991)]{Woodardetal91} Woodard, M.~F., 
905: Libbrecht, K.~G., Kuhn, J.~R., \& Murray, N.\ 1991, \apjl, 373, L81 
906: 
907: \bibitem[Vorontsov et al.(2002)]{Vorontsovetal02} Vorontsov, S.~V., 
908: Christensen-Dalsgaard, J., Schou, J., Strakhov, V.~N., \& Thompson, 
909: M.~J.\ 2002, Science, 296, 101 
910: \end{thebibliography}
911: 
912: \begin{figure}
913: %\centering
914: %\includegraphics[width=1.\textwidth]{f1}
915: \plotone{f1.ps}
916: \caption{The first four eigenvectors for the MDI data set are shown.  The 
917: base set is \#1216 (1996 May 1).  The left-hand panels show the eigenvectors 
918: as a function of frequency, and the right-hand panels show it as a function 
919: of the lower turning point of the mode.  The vertical line shows the 
920: position of the base of the convection zone.  The vertical axis units are 
921: arbitrary (the vectors are normalized so that $\vec{\xi}_i \cdot \vec{\xi}_i=1$).  
922: The error bars show representative errors calculated using a Monte Carlo 
923: simulation.
924: \label{fig:MDIvec}}
925: \end{figure}
926: 
927: \begin{figure}
928: \plotone{f2.ps}
929: \caption{The scaling coefficients for the first four eigenvectors are shown as 
930: points.  They are shown as a function of time (the start date of the 
931: MDI mode sets).  The dotted line is the change in 10.7cm radio flux with 
932: respect to the beginning of the solar cycle.  The units are Solar Flux 
933: Units (SFU).
934: \label{fig:MDIcoef}}
935: \end{figure}
936: 
937: \begin{figure}
938: \plotone{f3.ps}
939: \caption{To demonstrate that the PCA representation of the data does indeed 
940: capture the actual data, we present two reconstructed data sets and compare 
941: them to the actual data.  Two frequency difference sets are examined:  sets \#3160 
942: (2001 August 27, panel a) and \# 4528 (2005 May 26, panel b), both with respect 
943: to \# 1216.  At left, the frequency differences are plotted (panel (c) for set 
944: \#3160 and panel (d) for set \#4528).  The actual data is shown 
945: in red, and the reconstructed data set using $\vec{\xi}_1$ and $\vec{\xi}_2$ 
946: are overplotted in black.  The normalized residuals are binned and shown 
947: against a Gaussian curve to show that the information which has been 
948: removed is statistically random.
949: \label{fig:MDIrecon}}
950: \end{figure}
951: 
952: \begin{figure}
953: \plotone{f4.ps}
954: \caption{The first two eigenvectors for the GONG data sets plotted as 
955: a function of frequency (the left-hand panels) and as a function of the 
956: lower turning radius (the right-hand panels).  The first eigenvector $\vec{\xi}_1$ 
957: shows a signature similar to the one seen in the MDI data, albeit much 
958: less obviously.  The eigenvector $\vec{\xi}_2$ shows no structure, unlike the 
959: equivalent eigenvector for MDI, implying that the MDI $\vec{\xi}_2$ is an 
960: instrumental artifact.
961: \label{fig:GONGvec}}
962: \end{figure}
963: 
964: \begin{figure}
965: \plotone{f5.ps}
966: \caption{Inversion for sound speed of the $\vec{\xi}_1$ eigenvector.  
967: The top panel shows the inversion of the MDI data.  The lower panel 
968: shows the inversion of the GONG data.  The solid cyan line is the 
969: result from the RLS inversion (the dotted lines are the vertical 
970: error bounds).  The red points are the results from the SOLA 
971: inversion.  The horizontal dashed line is the zero-point.  The 
972: vertical dashed line represents the location of the base of the 
973: convection zone.  At the convection zone base, the MDI inversion 
974: results show a clear depression in sound speed at high activity 
975: (the sense of the inversion is low activity minus high activity) 
976: and an enhancement in the tachocline region.  The depression is 
977: matched in the GONG inversion results.  The location of this feature, 
978: though slightly deeper, is within the horizontal errors of the 
979: MDI result.
980: \label{fig:MDIinv}
981: \label{fig:GONGinv}}
982: \end{figure}
983: 
984: \begin{figure}
985: \plotone{f6.ps}
986: \caption{Change in inferred sound speed as function of activity level 
987: (10.7cm radio flux) is shown for different radii around the base of 
988: the convection zone.  The shaded regions show the errors for each 
989: set of inversions.
990: \label{fig:MDIcsq}}
991: \end{figure}
992: 
993: \begin{figure}
994: \plotone{f7.ps}
995: \caption{The first eigenvector $\vec{\xi}_1(\theta)$ for latitudes from 
996: 0$^\circ$ to 75$^\circ$.  The data are from MDI and are relative to the 
997: \#1216 mode set.  The left-hand panels show the eigenvectors as functions 
998: of frequency, and the right-hand panels as functions of the lower turning 
999: points.
1000: \label{fig:LATvec}}
1001: \end{figure}
1002: 
1003: \begin{figure}
1004: \plotone{f8.ps}
1005: \caption{The scaling coefficients are plotted as a function of time 
1006: and latitude.  The top panel shows the coefficients for each individual 
1007: eigenvector $\vec{\xi}_1(\theta)$.  The bottom panel shows the scaling coefficients 
1008: for all the latitudes as a function of the $\vec{\xi}_1(15^\circ)$.  This shows 
1009: how the changes represented by that eigenvector change as a function of 
1010: both time and latitude.  The average unsigned magnetic flux from MDI 
1011: carrington rotation synoptic maps over each 72-day period is shown in contour.  
1012: The contours are spaced every 52G, with the lowest at 56.5G.  The vertical 
1013: bars in 1998 are gaps in MDI coverage due to spacecraft problems.
1014: \label{fig:LATcoef}}
1015: \end{figure}
1016: 
1017: \begin{figure}
1018: \plotone{f9.ps}
1019: \caption{The inversions of the latitudinal frequencies are shown for 
1020: four different latitudes from the equator to $45^\circ$.  The differences 
1021: are with respect to MDI set \#1216.  As expected from an examination 
1022: of the raw frequencies, there are no discernible features in the inversion 
1023: for $45^\circ$ and above --- the inversions are extremely unstable to the 
1024: choice of inversion parameters.  At $15^\circ$ and the equator, a significant 
1025: (about 2$\sigma$) enhancement in sound speed at high activity is observed 
1026: above approximately $r=0.98R_\odot$.  As before, the vertical dashed line 
1027: indicates the position of the base of the convection zone ($r=0.713R_\odot$).
1028: \label{fig:LATinv}}
1029: \end{figure}
1030: 
1031: \begin{figure}
1032: \plotone{f10.ps}
1033: \caption{The $\vec{\xi}_1$ eigenvectors for the asphericity terms --- equator minus 
1034: latitude --- are shown for five different latitudes.  As usual, the left panels 
1035: show the eigenvectors with respect to frequency, and the right hand panels are 
1036: with respect to lower turning point.
1037: \label{fig:ASPHvec}}
1038: \end{figure}
1039: 
1040: \begin{figure}
1041: \plotone{f11.ps}
1042: \caption{The scaling coefficients for the asphericity eigenvectors are shown.  
1043: The dotted line is the $F_{10.7}$ flux.  The $0^\circ-15^\circ$ and $0^\circ - 30^\circ$ 
1044: both show a phase shift, consistent with the butterfly diagram from Fig. \ref{fig:LATcoef}.  
1045: The higher latitudes show a distinct change from mostly positive to negative at 
1046: high activity.
1047: \label{fig:ASPHcoef}}
1048: \end{figure}
1049: \end{document}
1050: