Experimental design
The common garden experiment has been described at greater length elsewhere (Vojtech et al., 2008, Vojtech et al., 2007, HautierVojtech et al., 2018). Briefly, we established monocultures (n=5), all pairwise mixtures (n=10) and the full five-species mixtures (n=1) of five perennial grass species (Poaceae): Alopecurus pratensis L., Anthoxanthum odoratumL., Arrhenatherum elatius (L.) P. Beauv. ex J. Presl & C. Presl,Festuca rubra ssp. commutata Gaud. (= Festuca nigrescens Lam.), Holcus lanatus L. (Lauber and Wagner, 2001). Each species combination was replicated five times for a total of 80 plots. Species were sown at a target density of 1000 seeds m-2 from seeds of individual species. Plants were established in 1 m2 plots on highly fertile soil (Garden humus, Ricoter, Aarberg, Switzerland). The experiment ran from April 2004 to June 2008. Plots were watered daily and regularly weeded throughout the duration of the experiment. During 2005 and 2006, plants were regularly fertilized with an NPK fertilizer corresponding to 15 g m-2 yr-1 of nitrogen to create highly productive conditions. In 2007, we divided the plots into four subplots of 50 x 50 cm (Fig. S1). In half of these subplots, we maintained the initial highly productive conditions by continuously adding the NPK fertilizer. In the other half of the subplots we reduced soil fertility by a combination of the cessation of fertilization and the addition of sucrose (in five applications of 500 g m-2 year-1 during 2007 and two applications of 625 g m-2 in 2008). Addition of a carbon source limits nutrient availability to plants and reduces productivity due to the immobilisation of nitrogen by soil micro-organisms (Killham, 1994) and increases competition between micro-organisms and plants for nitrate and ammonium (Bardgett et al., 2003; Schmidt et al., 1997). Additionally, we crossed the productivity treatments with regular cutting of the canopy structure to create disturbed conditions (HautierVojtech et al., 2018). Calculating daily RGR per species throughout the growing season for the plots that were disturbed was not possible because of the limited number of samples between each cutting event. Here, we therefore analyse only the undisturbed productive and unproductive conditions.