Sensitivity analyses of dense cloud chemical models

Context. Because of new telescopes that will dramatically improve our knowledge of the interstellar medium, chemical models will have to be used to simulate the chemistry of many regions with diverse properties. To make these models more robust, it is important to understand their sensitivity to a variety of parameters. Aims. In this article, we report a study of the sensitivity of a chemical model of a cold dense core, with homogeneous and time-independent physical conditions, to variations in the following parameters: initial chemical inventory, gas temperature and density, cosmic- ray ionization rate, chemical reaction rate coefficients, and elemental abundances. Methods. We used a Monte Carlo method to randomly vary individual parameters and groups of parameters within realistic ranges. From the results of the parameter variations, we can quantify the sensitivity of the model to each parameter as a function of time. Our results can be used in principle with observations to constrain some parameters for different cold clouds. We also attempted to use the Monte Carlo approach with all parameters varied collectively. Results. Within the parameter ranges studied, the most critical parameters turn out to be the reaction rate coefficients at times up to 4 x 10(5) yr and elemental abundances at later times. At typical times of best agreement with observation, models are sensitive to both of these parameters. The models are less sensitive to other parameters such as the gas density and temperature. Conclusions. The improvement of models will require that the uncertainties in rate coefficients of important reactions be reduced. As the chemistry becomes better understood and more robust, it should be possible to use model sensitivities concerning other parameters, such as the elemental abundances and the cosmic ray ionization rate, to yield detailed information on cloud properties and history. Nevertheless, at the current stage, we cannot determine the best values of all the parameters simultaneously based on purely observational constraints.