Thomson scattering can only produce an E-mode locally since the spherical
harmonics that describe the temperature anisotropy have 
electric parity.
In Figs. 4, 6, and 8,
the 
, m=0,1,2 E-mode patterns are shown in yellow. The B-mode
represents these patterns rotated by 
and are shown in purple
and cannot be generated by scattering.
In this way, the scalars, vectors, and tensors are similar in that scattering
produces a local E-mode only.
However, the pattern of polarization on the sky is not simply this local signature from scattering but is modulated over the last scattering surface by the plane wave spatial dependence of the perturbation (compare Figs. 3 and 10). The modulation changes the amplitude, sign, and angular structure of the polarization but not its nature, e.g. a Q-polarization remains Q. Nonetheless, this modulation generates a B-mode from the local E-mode pattern.
The reason why this occurs is best seen from the local distinction between
E and B-modes. Recall that E-modes have polarization amplitudes that
change parallel or perpendicular to, and B-modes in directions
away from, the polarization direction. On the other hand, plane wave
modulation always changes the polarization amplitude in the direction
or N-S on the sphere.
Whether the resultant pattern possesses E or B-contributions depends on
whether the local polarization has Q or U-contributions.
For scalars, the modulation is of a pure Q-field and thus its E-mode nature is preserved ([Kamionkowski et al.] 1997; [Zaldarriaga & Seljak] 1997). For the vectors, the U-mode dominates the pattern and the modulation is crossed with the polarization direction. Thus vectors generate mainly B-modes for short wavelength fluctuations ([Hu & White] 1997). For the tensors, the comparable Q and U components of the local pattern imply a more comparable distribution of E and B modes at short wavelengths (see Fig. 11a).
These qualitative considerations can be quantified by noting that plane wave
modulation simply represent the addition of angular momentum from the plane
wave ( 
) with the local spin angular dependence.
The result is that plane wave modulation takes the local angular
dependence to higher 
(smaller angles) and splits the signal into E
and B components with ratios which are related to Clebsch-Gordan
coefficients. At short wavelengths, these ratios are B/E=0,6,8/13 in power
for scalars, vectors, and tensors
(see Fig. 11b and [Hu & White] 1997).
The distribution of power in multipole -space is also important.
Due to projection, a single plane wave contributes to a range of angular
scales 
where r is the comoving distance to the last
scattering surface. From Fig. 10, we see that the smallest
angular, largest 
variations occur on lines of sight
or 
though a small amount of power
projects to 
as 
.
The distribution of power in multipole space of Fig. 11b can be
read directly off the local polarization pattern.
In particular, the region near shown in Fig. 11a
determines the behavior of the main contribution to the polarization power
spectrum.
The full power spectrum is of course obtained by summing these plane wave
contributions with weights dependent on the source of the perturbations and
the dynamics of their evolution up to last scattering.
Sharp features in the k-power spectrum will be preserved in the multipole
power spectrum to the extent that the projectors
in Fig. 11b
approximate delta functions.
For scalar E-modes, the sharpness of the projection is enhanced due to strong
Q-contributions near ( 
) that then diminish
as 
.
The same enhancement occurs to a lesser extent for vector
B-modes due to U
near 
and tensor E-modes due to Q there.
On the other hand, a supression occurs for vector E and tensor B-modes due
to the
absence of Q and U at respectively. These considerations have
consequences for the sharpness of features in the polarization power spectrum,
and the generation of asymptotic
``tails'' to the polarization spectrum at low- (see §4.4 and [Hu & White] 1997) .
