Over the past year, air quality has become a major priority. The pollutants with human health risk are widespread in indoor and outdoor environments. In residential buildings, capturing those contaminants before their mix with the indoor air is crucial to control the indoor air qual￾ity. It is usually achieved by using an efficient fan. The recent review of Han et al. [1] pointed out the lack of dedicated numerical models able to handle such complex problems. Benchikh Lehocine et al. [2] developed a numerical model using OpenFOAM to investigate the CO2 capture by an axial fan. It accounted for the temperature variations through the Boussinesq approximation and evaluated the capture efficiency for multiple CO2 source configurations over a wide range of rotation rate of the axial fan. The main limitation remains that CO2 can expe￾rience temperature gradients up to 150 K making the Boussinesq approximation no more valid. The present work is an attempt to improve that model by implementing a density-based solver. The dimensions of the experimental room and of the three rectangular entries for the fresh air respect the ASTM E3087-18 standard. The extraction system is equipped with an eleven bladed axial fan (NACA 0012). Four circular CO2 sources are used. A new steady-state density￾based solver based on rhoSimpleFoam has been developed using the OpenFOAM 4.1 libraries and coupled to the k-ω SST model and the multiple reference frame approach. An advection￾diffusion equation is solved for the transport of CO2 considered as a turbulent passive scalar. The unstructured mesh grid is composed of 6 million cell elements. The results show a symmetry of the CO2 field, which remains confined between the 2 CO2 sources and the fan, even at moderate fan’s rotation rate demonstrating the fan efficiency. Tem￾perature displays a similar behavior with moderate temperature regions (293-320 K) along the roof wall of the kitchen and on the right hand side of the fan due to diffusion mechanism. Fi￾nally, the predictions of the present solver compare particularly well with new experimental measurements under the same conditions. The main uncertainty is related to the CO2 concen￾tration measurements at the inlet, which are very sensitive to human presence.