Serious discrepancies have recently been observed between predictions of stellar evolution models in the 0.7-1.1 M_sun mass range and accurately measured properties of binary stars with components in this mass range. We study one of these objects, the eclipsing binary UV Piscium, which is particularly interesting because Popper (1997) derived age estimates for each component which differed by more than a factor of two. In an attempt to solve this significant discrepancy (a difference in age of

more »

... 1 Gyr), we compute a large grid of stellar evolution models with the CESAM code for each component. By fixing the masses to their accurately determined values (relative error smaller than 1% for both stars), we consider a wide range of possible metallicities Z (0.01 to 0.05), and Helium content Y (0.25 to 0.34) uncorrelated to Z. In addition, the mixing length parameter alpha_MLT is left as another free parameter. We obtain a best fit in the T_eff-radius diagram for a common chemical composition (Z, Y)=(0.012, 0.31), but a different MLT parameter alpha_MLT_A = 0.95+-0.12(statistical)+0.30(systematic) and alpha_MLT_B = 0.65+-0.07(stat)+0.10(syst). The apparent age discrepancy found by Popper (1997) disappears with this solution, the components being coeval to within 1%. This suggests that fixing alpha_MLT to its solar value (~1.6), a common hypothesis assumed in most stellar evolutionary models, may not be correct. Secondly, since alpha_MLT is smaller for the less massive component, this suggests that the MLT parameter may decrease with stellar mass, showing yet another shortcoming of the mixing length theory to explain stellar convection. This trend needs further confirmation with other binary stars with accurate data.