The accuracy of the center-of-gravity method for measuring velocity and magnetic field strength in the solar photosphere

The accuracy of the center-of-gravity method for measuring velocity and magnetic field strength in the solar photosphere H. Uitenbroek National Solar Observatory/Sacramento Peak 1, P.O. Box 62, Sunspot, NM 88349 huitenbroek@nso.edu ABSTRACT I investigate the accuracy with which the line-of-sight velocity and magnetic field strength in the solar photosphere can be recovered from spatially resolved spectral line profiles with the center of gravity (COG) method. For this purpose theoretical Non-LTE polarized line profiles of a series of Fe I lines were calculated through a two-dimensional slice from a snapshot of a three-dimensional solar magneto-convection simulation. The calculated profiles were analyzed with the COG method for all positions along the slice, and retrieved values of velocity and field strength were compared with actual values at the heights of formation of the lines. The average formation heights of the employed lines range from 60 to almost 400 km above the average photospheric level. The COG method appears reliable for measuring velocities in the lower half of these formation heights, and for measuring field strength over the whole range of heights, for fields up to intermediate strength. Moreover, it is shown that the COG determination is independent of spectral resolution, making it particularly suitable for applications which require high throughput and a correspondingly large spectral bandpass, such as high spatial resolution observations with a large diameter telescope. Finally, the effect of broad angle scattering, which includes a schematic representation of image deterioration through seeing, on the retrieved velocity and field strength was investigated. Subject headings: line: formation polarization methods: numerical radiative transfer magnetic fields Sun: photosphere 1 Operated by the Association of Universities for Research in Astronomy, Inc. (AURA), for the National Science Foundation

2 1. Introduction Our ability to build and verify physical models of the solar atmosphere depends to a large degree on the accuracy with which we can extract relevant information from observed spectra. Unfortunately, these spectroscopic measurements are adversely affected by limitations in observational as well as in data reduction procedures. Observational errors may be introduced by turbulence in the earth atmosphere, imperfections in the telescope or instrumentation, and limited resolution in the spatial, spectral and temporal domains. Once data has been registered further uncertainties may be introduced by inversion procedures which are used to recover physical quantities from the observed spectra. Often inversion procedures are model specific and skew the results in a particular way by either the choice of model, or by limiting the solution space of an inversion, for instance by requiring smoothness and/or monotonicity in a least squares fitting procedure. For this reason model-independent methods are preferred. An excellent overview of different spectroscopic methods for magnetic field determination can be found in Solanki (1993). To compare model calculations with observed quantities it seems natural to strive for the highest resolution in all three domains: spatial, spectral, and temporal. However, despite a high surface brightness, spectroscopic observations of the Sun are photon limited, and cannot be arbitrarily subdivided into increasingly smaller resolution bins. This is especially a problem at high spatial resolution, where the dynamic nature of the solar atmosphere limits exposure times. For instance, at a spatial resolution of 0.05 (close to the theoretical limit in the visible of the proposed 4 m Advanced Technology Solar Telescope, ATST 2 ), the exposure time is limited to only a few seconds before the image is smeared by the dynamics of the solar atmosphere. Assuming a spectral resolution of 100 000, and an efficiency of 10% this results in a signal-to-noise ratio based on photon statistics of only about 10 4. At this spectral resolution line profiles shapes already suffer considerable degradation as it corresponds to a spectral broadening of several km s 1, so that inversion methods that rely on fitting the whole line profile are adversely affected. In a parallel paper Balasubramaniam & Uitenbroek (2003) show that velocity measurements using a line bisector determination require a spectral resolution of at least 10 6 to limit spurious velocity introduced by instrumental broadening to less than 25 m s 1. For observations at the spatial resolution limit of a large diameter telescope it is, therefore, imperative to look for spectral analysis methods that are accurate with a relatively broad wavelength throughput. In this paper the accuracy of so-called centre of gravity determination is investigated. It is shown that the measurement of line-of-sight velocity and magnetic field strength through COG determination is almost insensitive to the 2 http://atst.nso.edu

3 employed spectral resolution, and is therefore an excellent candidate for the measurement of these quantities at high spatial resolution. Recent three-dimensional (magneto-)hydrodynamic simulations of the solar convection (e.g. Stein & Nordlund 1998; Stein & Bercik 2003, and references therein) approach reality, at least by some observational measures. For instance, Asplund et al. (2000a,b) were able to reproduce the observed equivalent width and asymmetry of photospheric Fe I lines very accurately without the use of any freely adjustable parameters like microturbulent broadening. Comparing the detailed results of such simulations with observations is one of the main targets of new high spatial resolution solar telescopes, but because of their large degree of realism these simulations also offer an unique opportunity for evaluating the accuracy of observation and data reduction procedures. Spectra calculated from one or more simulation snapshots can be folded through instrument characteristics and analyzed in the same way as observed spectra. Values obtained through this analysis can then be directly compared with the actual values in the simulation to assess the accuracy of the employed data reduction scheme and/or to estimate the impact of limitations in data quality, such as limited spectral and spatial resolution, on the retrieved values. This is the aim of this paper. The structure of this paper is as follows. Section 2 describes the model calculations that serve as the basis for the presented analysis. The centre of gravity method is discussed in Section 3, and results are discussed in Sections 3 and 4. Conclusions are given in Section 5. 2. Model calculations A state-of-the-art numerical magneto-convection simulation (Bercik et al. 1998; Stein & Bercik 2003) was used to provide a realistic environment to calculate Fe I line Stokes profiles emerging from the solar photosphere. The underlying simulations account for the interaction between convection and the magnetic field through calculations of compressible magnetoconvection. The equations of mass, momentum and energy conservations, and the induction equation are solved. In the particular simulation selected for this paper the physical domain extends 12 Mm in both horizontal directions with a resolution of 96 km, extends 2.5 Mm below the surface, and spans 3 Mm vertically. Periodic boundary conditions were used in the horizontal directions, and transmitting vertical boundary conditions were specified. Radiative contributions to the energy balance were calculated by solving radiative transfer in LTE using a four bin opacity distribution function. For the Non-LTE line transfer calculations a two-dimensional cut through a single threedimensional simulation snapshot was selected. The upper part of this slice, starting at 300

4 km below the photosphere, was interpolated onto a finer vertical grid, logarithmically in densities, and linearly in temperature and velocities. Radiative transfer was solved without accounting for polarization in 1.5-dimensional fashion, treating each of the 125 horizontal grid locations as if it were a one-dimensional plane-parallel atmosphere. Finally, all four Stokes parameters were calculated for all locations using the quasi-parabolic DELO method proposed by Trujillo Bueno (2003, see Socas Navarro et al. 2000 for a first application of this method), with the Non-LTE population numbers resulting from the field-free solution. The coupled equations of statistical equilibrium and radiative transfer were solved for a 23 level, 33 line atomic model of Fe I with a multi-dimensional radiative transfer code (Uitenbroek 1998, 2001) based on the Rybicki & Hummer (1991, 1992) multi-level accelerated lambda iteration (MALI) scheme. All bound-bound and bound-free transitions were solved in detail in the iterative scheme. As an example the calculated spatially resolved I and V profiles for the well-known 630.25 nm line are shown in Figure 1, which clearly shows that the interplay between temperature, density, velocity, and magnetic field yields a wide variety of emergent line profiles. TABLE 1 List of lines studied in detail λ 0 [nm] h FALC [km] h avg [km] g Lande 398.06 39.7 60.5 1.575 630.25 141.7 115.6 2.500 630.15 220.2 186.5 1.667 525.02 286.7 234.3 3.000 516.63 359.9 330.4 1.800 511.04 425.0 396.3 1.575 Of the 33 included lines 6 were selected for closer study on the basis of their average formation heights and magnetic properties. These lines are listed in Table 1 with their central wavelength, formation height in a standard one-dimensional model of the average quiet-sun (FALC, model C of Fontenla et al. 1993), their average formation height in the magnetohydrodynamic slice, and their effective Landé factor. Formation height in this case is defined as the height above the photosphere at which optical depth at the central wavelength of the line is unity. The average formation heights of the lines span about 340 km in the twodimensional slice, ranging from 60 km for the 398.06 nm line up to 396 km for the 511.04 nm line. Figure 2 shows the optical depth unity curves for all six lines listed in Table 1. It is clear that actual formation heights for each individual line vary significantly through the slice, by even more than the range in average formation heights between the different lines. This strong variation in formation height is the main reason why it is difficult to recover

5 4.0 0.6 10 0.4 3.5 Position along slit [Mm] 8 6 4 3.0 2.5 Intensity [10 8 J m 2 s 1 Hz 1 sr 1 ] 0.2 0.0 0.2 Stokes V [10 8 J m 2 s 1 Hz 1 sr 1 ] 0.4 2 2.0 0.6 0 1.5 0.03 0.02 0.01 0.00 0.01 0.02 0.03 λ [nm] 0.03 0.02 0.01 0.00 0.01 0.02 0.03 λ [nm] Fig. 1. The calculated I (left panel) and V profiles (right panel) of the Fe I 630.25 line through the magneto-hydrodynamic slice. Overplotted are the determined centroids of the unpolarized intensity I (white curve, left panel), and the centroids of the left- (right panel, black curve) and righthand (white curve) circularly polarized components I ± V.

6 quantities like velocity and magnetic field strength accurately from observed spectra. Actual values can be recovered faithfully, as will be shown below, but tying those values to a specific height in the atmosphere is considerably more difficult. 600 400 Formation height [km] 200 0 200 511.04 516.63 525.02 630.15 630.25 398.06 0 2 4 6 8 10 12 Position along slit [Mm] Fig. 2. Formation heights of the lines listed in Table 1 as function of position along the magneto-hydrodynamic slice. 3. The centre of gravity method Determination of its centre of gravity (COG) provides an accurate method for measuring the central wavelength position of a spectral line. It is demonstrated below that this determination is independent of spectral resolution and is therefore uniquely suitable for applications which require a relatively large throughput and correspondingly wide spectral bandpass. Thus, the COG method can conveniently be used at high spatial resolution to determine line-of-sight velocity by measuring the wavelength position of the Stokes I profile

7 of a line, and to determine the line-of-sight magnetic field strength by measuring the difference in wavelength position of the right- and lefthand circular polarized components (i.e., I + V and I V, respectively) of the line (Semel 1967, 1970; Rees & Semel 1979; Cauzzi et al. 1993). 3.1. Definition of centre of gravity The centre of gravity wavelength λ COG of a line profile I is defined as the centroid of its residual intensity profile: λ COG = λ (I cont I) dλ/ (I cont I) dλ. (1) From the centre of gravity of the intensity profile I the line-of-sight velocity at the formation height of the line can be determined: v LOS = c(λ 0 λ COG )/λ 0, (2) where c is the speed of light and λ 0 the central wavelength of the line. The line-of-sight strength of the magnetic field (in Tesla = 10 4 Gauss) is then determined through the relation B LOS = λ + λ 2.0 4πmc eg L λ 2 0 (3) (cf. Stenflo 1994, p. 111), where λ ± are the centroids of the right- and left circular polarized line components computed by replacing I in equation (1) with I±V, g L is the line s (effective) Landé factor, and m and e are the electron mass and charge (in MKSA units), respectively. In both cases the velocity and magnetic field strength determined through the COG method represent an average over the contribution function of the line profile, weighted towards the line-center formation height by the choice of the residual intensity as probability distribution function. In the following it is assumed that the derived velocity and field strength are representative of these quantities at a height in the atmosphere corresponding to optical depth unity in the center of the line. 3.2. The COG method is independent of spectral resolution It can be shown analytically that the determination of centroid λ COG is unaffected by spectral broadening, as long as the broadening function is symmetrical, and the wavelength integration spans the domain of this function. Numerical calculations presented in the next

8 section support this finding. The equation for the centroid λ COG (see eq. [1]) can be formally written as: / λ COG = λψ ± (λ)dλ ψ(λ)dλ, (4) where the residual intensity profile ψ I cont I, orψ I cont (I ± V ). In dealing with observations with limited spectral resolution we have to replace ψ with a spectrally broadened profile ψ, which is given by the convolution of ψ with the spectral broadening function f: ψ(λ) = ψ(λ )f(λ λ )dλ. (5) The function f can be either a boxcar, Gaussian, Voigt, or another appropriate function like an Airy function representing the transmission function of a narrow band filter system. Substituting equation (5) into (4) and changing the order of integration between λ and λ in both numerator and denominator yields: / λ COG = ψ(λ ) λf(λ λ )dλdλ ψ(λ )dλ (6) for the centroid λ COG of the convolved profile. The inner integral in the numerator in equation (6) is nothing else than the centroid of the (normalized) broadening function f around λ, which is precisely λ itself, if the broadening function is symmetric, and if the integration covers the whole domain of the broadening function. As a result the centroid λ COG remains invariant under spectral broadening. For illustration purposes let us look at the specific case of a Gaussian broadening function with f given by: f(λ) = 1 σ 2 π e (λ/σ). (7) Then the inner integral in equation (6) reduces to: 1 σ π 1 σ π ( ) λe λ λ 2 σ dλ = (8) ) 2 d(λ λ )+ 1 ( σ λ e λ λ ) 2 σ dλ = λ π ( (λ λ )e λ λ σ (since the first integral is zero), resulting in the equality λ COG = λ COG in equation (6). A similar conclusion was reached by Cauzzi et al. (1993, their figure 1), who concluded that the COG method recovers LOS field strength very accurately, independent of filter bandpass, which they varied from 2.0 to 12.0 pm (FWHM). When they limited the integration interval for determination of the centroids to the inner 12 pm to eliminate the effect of

9 neighboring magnetically sensitive blends they found that the derived field strength was systematically underestimated, and more so for wider bandpasses. This is in good agreement with our conclusion that the integration interval employed for the determination of the centroids should span the whole broadening function, otherwise the result of an integration like that in equation (9) is less than λ leading to underestimate of the centroid λ COG and the derived field strength. Note that the issue of spectral broadening is separate from the issue of spectral sampling. Even though we can afford a fairly wide passband without losing accuracy in the velocity or field strength determination, the broadened profile still needs to be sampled on a fine enough grid so that discretization errors are sufficiently small. 3.3. Comparison with values at actual formation heights Figures 3 and 4 show that the assumption that the line profile represents physical conditions at the formation height of its core wavelength is reasonable for the 630.25 nm line, which forms at an average height of 115.6 km above the photosphere in the magnetohydrodynamic slice (Table 1). The dashed curves in these figures, obtained by determining the COG of the calculated vertically emergent line profiles, agree very well with actual values at optical depth unity (solid curves), proving that the COG method is indeed accurate for velocity and field determination, even in the presence of line asymmetries caused by the velocity field in the solar atmosphere. Figures 5 and 6 show similar plots for the 516.63 line which forms higher up in the atmosphere, at an average of 330 km. The magnetic field is still faithfully recovered, however, the velocity in this line is overestimated, indicating that the COG determination in this case is sensitive to velocity at heights below the formation of the linecenter wavelength. In locations with a field strong enough to split a line completely the COG method underestimates the strength of the field. In these locations the centroid determination of each of the two circularly polarized components is skewed towards the central wavelength of the line by the presence of the unpolarized and unshifted central π component. This effect is illustrated in Figure 7, where the calculated left- and right-hand circular polarized profiles of the 630.25 line are plotted for a location with particularly strong field (at slit position 8.75 Mm, see Figure 4). The derived centroids are marked by the vertical dashed lines, which clearly underestimate the shift in the position of the main absorption troughs. This effect is also noticeable in the curves outlining the determined centroids for the 630.25 line in the right-hand panel of Figure 1.

10 3 2 v τ = 1 v COG 1 v LOS [km] 0 1 2 3 0 2 4 6 8 10 12 Position along slit [Mm] Fig. 3. Comparison of LOS velocity derived through the centre of gravity method from calculated 630.25 nm Stokes I profiles (eq. [2], dashed curves) with the actual longitudinal velocity at optical depth unity in the core of that line.

11 0.25 0.20 B τ = 1 B COG 0.15 B LOS [T] 0.10 0.05 0.00 0 2 4 6 8 10 12 Position along slit [Mm] Fig. 4. Comparison of LOS magnetic field derived from calculated 630.25 nm I ±V profiles with the centre of gravity method (eq. [3], dashed curves) to the actual longitudinal magnetic field component at optical depth unity in that line.

12 3 2 v τ = 1 v COG v LOS [km] 1 0 1 2 0 2 4 6 8 10 12 Position along slit [Mm] Fig. 5. Similar to Figure 3, but for the 516.63 nm line.

13 0.20 B τ = 1 B COG 0.15 B LOS [T] 0.10 0.05 0.00 0 2 4 6 8 10 12 Position along slit [Mm] Fig. 6. Similar to Figure 4, but for the 516.63 nm line.

14 2.0 10 8 Intensity [J m 2 s 1 Hz 1 sr 1 ] 1.8 10 8 1.6 10 8 1.4 10 8 1.2 10 8 630.20 630.22 630.24 630.26 630.28 630.30 Wavelength [nm] Fig. 7. Left- and right-hand circular polarized profiles for a location with strong magnetic field. Note the central π absorption which skews the determination of the centroids (marked by the vertical dashed lines) towards the central wavelength.

15 A comparison between actual velocity and field strength values at line-center formation height, and values determined with the COG method from the computed Stokes profiles for all lines and all positions in the two-dimensional slice is presented in Figures 8. The measured magnetic field values in the right-hand panel show good correspondence with the actual values at the formation height of each line, up to field strengths of about 0.13 T (1300 Gauss). For stronger fields the COG determination underestimates the LOS field due to the complete splitting in the lines as discussed above (cf., Figure 7). Measured velocity (left-hand panel) does not agree well for all lines. The velocity estimates seem to fall into two separate groups. The magnitude of strong downflows in particular, but of upflows as well seems to be underestimated by the COG method, while the slower flows are systematically overestimated (central part of left panel in Fig. 8). In particular, lines that form highest in the atmosphere, namely the 516.63 and 511.04 nm lines, show a false signal, with the COG method indicating higher amplitude velocities than are actually present at the formation heights of these lines. These lines are thus sensitive to velocities that occur deeper than the heights at which line center optical depth is unity. This effect is also clearly visible in Figure 5. The underestimate of high magnitude flows is in part due to the mixing of narrowly 4 2 398.06 630.25 630.15 525.02 516.63 511.04 0.25 0.20 398.06 630.25 630.15 525.02 516.63 511.04 v COG [km s 1 ] 0 B COG [T] 0.15 0.10 2 0.05 4 4 2 0 2 4 v LOS [km s 1 ] 0.00 0.00 0.05 0.10 0.15 0.20 0.25 B LOS [T] Fig. 8. Scatter plot of actual line-of-sight velocity (left panel) and magnetic field strength (right panel) at the actual height of formation of each line (abscissa) against the values determined with the COG method (ordinate), for all lines and positions. confined up and downflows along the line-of-sight. Since these flows (excellent examples are found at a position of 4 Mm along the slice, see Fig. 3) are not vertical, integration along a vertical ray cuts through both negative and vertical flows near their boundaries, as well as through regions with lower amplitude flows higher up in the atmosphere.

16 4 2 398.06 630.25 630.15 525.02 516.63 511.04 0.25 0.20 398.06 630.25 630.15 525.02 516.63 511.04 v COG [km s 1 ] 0 B COG [T] 0.15 0.10 2 0.05 4 4 2 0 2 4 v LOS [km s 1 ] 0.00 0.00 0.05 0.10 0.15 0.20 0.25 B LOS [T] Fig. 9. Scatter plot of actual line-of-sight velocity (left panel) and magnetic field strength (right panel) at the average height of formation of each line (abscissa) against the values determined with the COG method (ordinate), for all lines and positions. 3.4. Comparison with values from average formation heights As mentioned above a major difficulty in retrieving the actual field strength and velocity at a certain location on the solar surface is the problem of estimating the height of formation of the employed spectrum. This specific case refers to the situation when the spatially resolved LOS field and velocity have to be determined without an explicit inversion, or subclassification (i.e., into granule and intergranular lane for instance). Given the strong variation in formation heights that the lines display (Fig. 2) it is to be expected that the scatter between actual and observed values increases if we use this approximate estimate of the formation height, unless the vertical gradients in the measured quantities are small. Figure 9 shows the resulting scatter plots for LOS velocity (left panel) and magnetic field (right panel). From the large scatter in the left plot and the relatively large deviations from the dashed line it is safe to conclude that the relative variations in the vertical velocity gradient are larger than in the magnetic field strength. Below approximately 0.08 T the magnetic field is recovered reliably. For larger field strengths the actual formation height of the lines dips well below the average formation height (see Fig. 2, at the position 8.75 Mm, for instance) towards stronger fields, resulting in a substantial overestimate of the field strength through the COG method with average formation heights. The lowering of the formation height is a result of the relatively low gas pressure and density in high field strength locations, where the magnetic field contributes to balancing the higher gas pressure of the surrounding non-magnetic medium.

17 The COG method underestimates the velocity in locations with strong downflows when sampling the flow at the average formation heights, while other locations are less affected. Again this is the result of the strong lowering of the formation height in the strongest magnetic field locations where the most obvious downflows occur. In the case of velocities however, the flow is slower in deeper locations because of mass flux conservation. Lower in the atmosphere density is higher and the flow speed is correspondingly lower. 3.5. Signal-to-noise considerations The invariance of the COG field measurement to reduced spectral resolution implies that accurate values can be obtained with a relatively wide passband or slit, allowing for shorter exposure times. This means that filter-based instruments can be used, which typically have less spectral resolution than long-slit spectrographs, but which provide a wider range of opportunities for reduction of the degrading effect of the earth atmosphere turbulence on spatial resolution. In slit-based instruments, moreover, the slit width not only determines spectral resolution but also spatial resolution in the direction of the dispersion. Thus, slit width cannot be arbitrarily increased to increase throughput, as this leads to a loss of spatial resolution. Neither can the slit be made too narrow, lest throughput becomes too low and exposure times become longer than necessary to freeze the dynamics of the solar atmosphere. On the other hand, filter-based instruments have the disadvantage that they require step-by-step scanning through the spectral line with small enough spectral sampling to approximate the required wavelength integrations accurately. Changes in seeing that occur on time scales shorter than the step time could induce distorted observed spectral profiles, although correlations between images at successive wavelengths in combination with destretching techniques may be employed to reduce this distortion. 4. Effect of wide-angle scattering 4.1. Analytical formulation We now turn our attention to the influence of spatial (wide angle) scattering on the measurement of the line of sight magnetic field through center of gravity determination in the Fe I 630.25 line. This effect is only investigated for the unbroadened profiles, under the assumption that spectral resolution is infinite. The simulation assumes that the wide angle scattering contributes a constant background to the telescope PSF, which is mathematically equivalent to adding a certain fraction of the spatially averaged spectrum to the spectrum

18 at each spatial location. Since this operation distorts the shape of the individual spectra we may expect that it reduces the reliability of the COG measurement. Let us write the residual line profile ψ obs affected by wide-angle scattering as: ψ obs ± (λ) =(1 α)ψ ± (λ)+αˆψ ± (λ), (9) where ψ and ˆψ are the uncontaminated and spatially averaged residual intensity line profiles, respectively, and α is the fraction of scattered light. Then the centroids for the right- and left-hand circularly polarized profiles are given by: λ obs ± = = λ[(1 α)ψ + α ˆψ]dλ / (1 α)ψ + α ˆψdλ (10) ] / [(1 α)λ ± ψdλ + αˆλ ± ˆψdλ (1 α)ψ + α ˆψdλ = (1 ˆα ± )λ ± +ˆα ±ˆλ± (leaving out some of the ± designations for ψ, ˆψ), with ˆα ± α ˆψ / ± dλ (1 α)ψ ± + α ˆψ ± dλ. (11) Clearly, when ψdλ and ˆψdλ are of similar magnitude ˆα reduces to α, and the observed centroids λ obs are given by the average of the uncontaminated centroids λ ± and centroids ˆλ ± of the spatially averaged profiles, weighted simply by the scattering fraction α. If the intrinsic field strength is small in some location we expect to observe the average field strength at that location, diluted with this scattering fraction. If, however, the surface areas under the curves ψ and ˆψ are different, ˆα α, and the weighting is affected. In the case of strong fields, for instance, the area under ψ is substantially smaller by up to a factor of two (see Fig. 10), resulting in a substantially larger contribution from the average centroids to the observed centroids. The opposite, although to a lesser degree, turns out to be the case for low field-strength locations. 4.2. Numerical experiment As numerical example of the effect of wide-angle scattering we investigate the values of the derived field when 1.0, 2.0, 5.0, and 10.0% of the spatially averaged spectrum is added to the spatially resolved profiles. Figure 10 illustrates the wide variety in the profiles involved. Note that the average field in the whole three-dimensional simulation used for these calculations is high, namely 0.04 T, characteristic more of a plage region than of the

19 2.0 10 8 Weak field Strong field Average field Residual Intensity [J m 2 s 1 Hz 1 sr 1 ] 1.5 10 8 1.0 10 8 5.0 10 9 0 630.20 630.22 630.24 630.26 630.28 630.30 Wavelength [nm] Fig. 10. Examples of the variety in residual line profiles of right- (thin curves) and lefthand (thick curves) circularly polarized line components under different longitudinal field strength, varying from almost zero Tesla (solid curves) to 0.23 Tesla (dashed curves). The profiles of the spatially averaged components are drawn with dot dashed curves. quiet Sun. In the particular slice that was chosen here the magnetic signal derived from the spatially averaged I ± V profiles was 0.044 T. Scatter plots for the derived field strength at all locations are given in Figure 11 for the different levels of scattering contamination. Because of the modified weighting ˆα the relation between observed and intrinsic field is not a simple linear relation with slope 1 α and intercept α ˆB LOS,where ˆBis the average LOS field strength over the spatial sample. The largest deviations between the original field strength and field values derived from the contaminated signal appear in the upper right part of the scatter diagrams, for larger field strengths. This is because the average field centroid gets additional weight in its contribution to the observed centroid (eq. [10]) due to the relatively small area under the strong-field

20 0.02 Scatter = 5.0 % Scatter = 10.0 % 0.01 B LOS [T] 0.00 0.01 0.02 0.02 0.01 Scatter = 1.0 % Scatter = 2.0 % B LOS [T] 0.00 0.01 0.02 0.00 0.05 0.10 0.15 B LOS [T] 0.00 0.05 0.10 0.15 B LOS [T] Fig. 11. Change in derived LOS field strengths for different levels of the scattering background corresponding to 1.0% (lower left), 2.0% (lower right), 5.0% (upper left), and 10.0% (upper right) contamination with the spatially averaged spectrum. residual intensity profile (see previous section). The contamination in locations with small field strength is less obvious because the area under the residual intensity curves is relatively larger in those locations (compare the solid curves in Fig. 10 with the dot dashed curves). 5. Conclusions In Section 3 it is shown analytically that the determination of the central wavelength of a spectral line through the center-of-gravity (COG) method is independent of the spectral resolution of the observed spectrum, as long as the size of the employed wavelength domain spans a significant part of the spectral broadening profile. This makes the COG method a technique of choice in situations where high throughput is required, such as in high spatial

21 resolution observations with short integration times. In such a case a filter based instrument can be employed with a relatively wide bandpass without concern for deterioration of the central wavelength determination. When the telescope point spread function (PSF) has a wide-angle scattering component through which a fraction of the spatially averaged spectrum is added to the resolved spectrum at each location, the fidelity of the COG measurement is reduced. Because of the non-linear averaging of the spectra (see eqs. [10] and [11]) locations with strong fields are more affected by this spatial degradation than locations with weaker field. The six Fe I lines studied in this paper exhibit a large variation in formation height (defined as the height at which optical depth is unity at the central wavelength of the line) through a representative cross section of a magneto-hydrodynamic convection simulation. This poses a challenge to the accurate recovery of physical quantities like line-of-sight velocity and magnetic field strength as a function of height in the atmosphere. The main reason for this is that we do not have a priori knowledge of the formation height of a specific line at a specific location on the solar surface. The correlation between actual velocity and field strength at the formation height, and the values derived through the COG method is reasonably tight (Fig. 8) when the formation height is given. Larger field strengths are underestimated by COG determination because of the presence of the unshifted π component of the Stokes I ± V profiles, which skews the wavelength determination of both oppositely polarized profiles towards line center. Amplitudes of velocities for lines that form higher in the atmosphere tend to be overestimated due to the sensitivity of the COG method to shifts in the wings of these lines which form at greater depths. When the formation heights of spectral lines as a function of position on the surface are not known a priori (unfortunately a more realistic case) we have to resort to an approximate estimate of the height of the atmosphere from which we are recovering information. One such choice is to assume that the line forms at its average formation height (as for instance determined by the type of forward modeling presented here) everywhere. This leads to a considerable widening of the correlation between the actual values of velocity and field strength at these average formation heights and the values determined through the COG method (Fig. 9). In particular, strong fields are now overestimated because the height at which the lines form is overestimated: the actual formation height of the line, which determines the signal in the spectrum, is much lower at higher field strength than the assumed average formation height because of the partial evacuation of the magnetic field concentration. For similar reasons the amplitude of high speed downflows is underestimated. In this case the lower actual formation height samples lower flow speeds at higher density (since mass flux conservation requires the product of density times velocity to be constant).

22 Thus, it cannot be stressed enough that one of the main difficulties in relating a derived observable, like a magnetic field strength or line-of-sight velocity, to a physical value in the atmosphere is the determination of the height in the atmosphere at which this value is measured. In this paper this height of information encoding or formation height was assumed to be the height at which optical depth is unity in the center of the spectral line in question. In a real atmosphere this formation height varies considerably and cannot simply be assumed to be the height determined in one-dimensional plane-parallel modeling (e.g., see Table 1 and Fig. 2), and is not a priori known for each location in the atmosphere. Alternative measures of formation height could be provided by analysis of contribution and response functions. Although these functions suffer from the same limitations of not being known a priori and varying considerably through the atmosphere, they would offer the advantage of providing a better estimate of the height resolution of observations in a given spectral line. This research has made use of NASA s Astrophysics Data System. I am grateful to Bob Stein for providing the magneto-hydrodynamic snapshot on which the calculations presented here are based. REFERENCES Asplund, M., Ludwig, H.-G., Nordlund, Å., & Stein, R. F. 2000a, A&A, 359, 669 Asplund, M., Nordlund, Å., Trampedach, R., & Stein, R. F. 2000b, A&A, 359, 743 Balasubramaniam, K. S., & Uitenbroek, H. 2003, ApJ, in preparation Bercik, D. J., Basu, S., Georgobiani, D., Nordlund, A., & Stein, R. F. 1998, in R. A. Donahue, J. A. Bookbinder (eds.), Cool Stars, Stellar Systems, and the Sun. Tenth Cambridge Workshop, ASP Conf. Ser., Vol. 154, The Astronomical Society of the Pacific, San Francisco, CA Cauzzi, G., Smaldone, L. A., Balasubramaniam, K. S., & Keil, S. L. 1993, Solar Phys., 146, 207 Fontenla, J. M., Avrett, E. H., & Loeser, R. 1993, ApJ, 406, 319 Rees, D. E., & Semel, M. D. 1979, A&A, 74, 1 Rybicki, G. B., & Hummer, D. G. 1991, A&A, 245, 171 Rybicki, G. B., & Hummer, D. G. 1992, A&A, 262, 209

23 Semel, M. D. 1967, Annales d Astrophysique, 30, 513 Semel, M. D. 1970, A&A, 5, 330 Socas Navarro, H., Trujillo Bueno, J., & Ruiz Cobo, B. 2000, ApJ, 530, 977 Solanki, S. K. 1993, Space Science Reviews, 63, 1 Stein, R. F., & Bercik, D. 2003, in A. A. Pevtsov, H. Uitenbroek (eds.), Current Theoretical Models and High Resolution Solar Observations: Preparing for ATST, ASP Conf. Ser., Vol. 286, The Astronomical Society of the Pacific, San Francisco, CA, p. 121 Stein, R. F., & Nordlund, Å. 1998, ApJ, 499, 914 Stenflo, J.-O. 1994, Solar Magnetic Fields. Polarized Radiation Diagnostics, Kluwer Academic Publishers, Dordrecht Trujillo Bueno, J. 2003, in I. Hubeny, D. Mihalas, K. Werner (eds.), Stellar Atmosphere Modeling, ASP Conf. Ser., Vol. 288, The Astronomical Society of the Pacific, San Francisco, CA, in press Uitenbroek, H. 1998, ApJ, 498, 427 Uitenbroek, H. 2001, ApJ, 557, 389 This preprint was prepared with the AAS L A TEX macros v5.0.