Context : Chemically peculiar ( CP ) stars are unique natural laboratories for investigation of the microscopic diffusion processes of chemical elements .
The element segregation under the influence of gravity and radiation pressure leads to the appearance of strong abundance gradients in the atmospheres of CP stars .
Consequently , the atmospheric temperature-pressure structure of these objects could deviate significantly from the atmospheres of normal stars with homogeneous abundances .

Aims : In this study we performed a self-consistent , empirical model atmosphere study of the brightest rapidly oscillating Ap star \alpha Cir .
We account for chemical stratification in the model atmosphere calculations and assess the importance of non-uniformed vertical element distribution on the model structure , energy distribution and hydrogen line profiles .

Methods : For chemical stratification analysis we use the DDAFit minimization tool in combination with a magnetic spectrum synthesis code .
The model atmospheres with inhomogeneous vertical distributions of elements are calculated with the LLmodels stellar model atmosphere code .

Results : Based on an iterative procedure of the chemical abundance analysis of 52 ions of 35 elements , stratification modeling of 4 elements ( Si , Ca , Cr and Fe ) and subsequent re-calculations of the atmospheric structure , we derived a new model atmosphere of \alpha Cir , which is consistent with the inferred atmospheric chemistry of the star .
We find T _ { eff } = 7500 K , \log g = 4.1 , and demonstrate that chemical stratification has a noticeable impact on the model structure and modifies the formation of the hydrogen Balmer lines .
At the same time , energy distribution appears to be less sensitive to the presence of large abundance gradients .

Conclusions : Our spectroscopically determined T _ { eff } of \alpha Cir agrees with the fundamental effective temperature of this star .
This shows that temperatures inferred in detailed spectroscopic analyses of cool magnetic CP stars are not affected by a large systematic bias .