OnReward-FreeReinforcementLearningwith LinearFunctionApproximation