Our analysis of the 1997 infrared spectra of Sakurai's Object strongly implies that dust was already present at that time. We note that Duerbeck (2002) did not find evidence for dust in the optical spectra from 1997. On the other hand, the fits by Pavlenko & Duerbeck (2001) to the observed SEDs at optical wavelengths indicate that E_B-V had increased by 0.6 (from 0.7 to 1.3) from April 1997 to August 1998. However, the August 1998 data were best fit using $T_{\rm eff}$ = 5250 $\pm$ 200 K, the same value as for July 1997 in this paper. Between June 1997 and August 1998 there were some variations in the photospheric radiaton, probably caused by mass losses events, evolution of the dusty envelope, and dynamical processes in the photosphere - envelope system (see light curve of Sakurai's Object in Duerbeck 2002). Nevertheless, $T_{\rm eff}$ apparently remained nearly constant during this period.

In 1997, the year of maximum optical brightness of Sakurai's Object, the luminosity was still dominated by optical radiation. At that time "quasi-periodic fluctuations of increasing cycle length and amplitude were superimposed on the general brightness evolution'' (Duerbeck 2002). In general, the effective temperature during such fluctuations does not need to follow changes of luminosity. In fact, it can be anti-correlated, since an increased radius can more than compensate for a lower $T_{\rm eff}$ . On the other hand, a change of radius can change the thermodynamical properties in the radiating region (i.e. photosphere). That may explain the similarity of $T_{\rm eff}$ obtained in this paper for July 1997 and that found in August 1998 by Pavlenko & Duerbeck (2001). The decreased optical brightness of Sakurai's Object in 1998 was mainly caused by development of the dust envelope (Kimeswenger 1999).

One question arises - were the optical and 2 $\mu$ m SED's being affected by the same dust in 1997-1998? The answer is probably yes. As mentioned earlier, the effective temperature remained constant during this time and thus cannot be the cause of the large change in E_B-V. This suggests that the cause of the increase in E_B-V was newly formed dust. The new dust would be expected to have been close to the star and thus quite hot. Indeed the full 1-5 $\mu$ m spectrum from 1998 (e.g., Geballe et al. 2002) shows that the excess peaked close to 3 $\mu$ m, indicating a mean dust temperature close to 1000 K at that time. The dust must have been hotter (and closer) in 1997; this is supported by the data from 1997 (Eyres et al. 1998; Geballe et al. 2002) which show that the continuum flux density decreased monotonically with wavelength in the observed wavelength range, 1-4 $\mu$ m, at that time. Comparison of the 1997 and 1998 SEDs also show that much less dust was present in 1997. Thus we confirm that the first appearance of dust occured in 1997 and that the amount of dust increased through summer 1998, prior to its becoming totally dominant in the latter part of 1998 and since then.

$\begin{figure} \par\includegraphics[width=8.8cm,clip]{MS2440f6.eps}\par\includegraphics[width=8.8cm,clip]{MS2440f6a.eps} \end{figure}$

Figure 6: Top: fits to observed spectrum of of Sakurai's Object on 1997 April 21. Bottom: details of the fits at 1.6-2.0 $\mu$ m; much of the structure at 1.82-1.95 $\mu$ m in the observed spectrum is due to incomplete removal of telluric absorption features. Synthetic spectra were computed for a microturbulent velocity of 6 km s^-1.

$\begin{figure} \par\includegraphics[width=9cm,clip]{MS2440f7new.eps}\par\includegraphics[width=8.8cm,clip]{MS2440f7a.eps} \end{figure}$

Figure 7: Top: fits to observed spectrum of of Sakurai's Object on 1997 July 13. Bottom: details of the fit at 1.6-2.0 $\mu$ m. Synthetic spectra were computed for a microturbulent velocity of 6 km s^-1.

4 Discussion