The paper examines the statistical equilibrium of Na I in stellar atmospher
es with a wide range of parameters: T-eff = 4000-12500 K, log g = 0.0-4.5,
and heavy element content [A] from 0.5 to -4.0. The effect of the "overreco
mbination" of Na I (i.e., excess relative to the equilibrium number density
of Na I) is present over the entire range of parameters considered, and in
creases with T-eff and luminosity. Na I lines are stronger than in the LTE
case, so that non-LTE corrections to the sodium abundance, Delta (NLTE), ar
e negative. Eight Na I lines commonly employed in abundance analyses are us
ed to construct the dependences of the non-LTE corrections on T-eff, log g,
and metallicity. The non-LTE corrections are small only for the Na I lambd
a lambda 615.4, 616.0 nm lines in main-sequence stars: \Delta (NLTE)\ less
than or equal to 0.08 dex. In all other cases, Delta (NLTE) depends strongl
y on T-eff and log g, and a non-LTE treatment must be applied if the sodium
abundance is to be determined with an accuracy no worse than 0.1 dex. The
profiles of solar Na I lines are analyzed in order to empirically refine tw
o types of atomic parameters required for the subsequent analysis of the st
ellar spectra. In the solar atmosphere, inelastic collisions with hydrogen
atoms influence the statistical equilibrium of Na I only weakly, and the cl
assical Unsold formula underestimates the van der Waals constant C-6 The em
pirical correction Delta log C-6 is from 0.6 to 2 for various Na I lines. T
he sodium abundance in the solar atmosphere is determined based on line-pro
file analyses, yielding different results depending on whether the model at
mospheres of Kurucz (log epsilon (Na), = 6.20 +/- 0.02) or Holweger and Mul
ler ( log epsilon (Na) = 6.28 +/- 0.03) are applied. (C) 2000 MAIK "Nauka/I
nterperiodica".