We study the effect of confinement on the mean perimeter of the convex hull of a planar Brownian motion, defined as the minimum convex polygon enclosing the trajectory. We use a minimal model where an infinite reflecting wall confines the walk to its one side. We show that the mean perimeter displays a surprising minimum with respect to the starting distance to the wall and exhibits a nonanalyticity for small distances. In addition, the mean span of the trajectory in a fixed direction θ ∈]0, π/2[, which can be shown to yield the mean perimeter by integration over θ, presents these same two characteristics. This is in striking contrast with the one dimensional case, where the mean span is an increasing analytical function. The non-monotonicity in the 2D case originates from the competition between two antagonistic effects due to the presence of the wall: reduction of the space accessible to the Brownian motion and effective repulsion. How does one characterize the territory covered by a Brownian motion in two dimensions? This question naturally arises in ecology where the trajectory of an animal during the foraging period is well approximated by a Brownian motion [1,2] and one needs to estimate the home range of the animal, i.e., the two dimensional (2D) space over which the animal moves around over a fixed period of time [3]. The most versatile and popular method to characterize the home range consists in drawing the convex hull, i.e., the minimum convex polygon enclosing the trajectory of the animal [4,5]. The size of the home range is then naturally estimated by the mean perimeter or the mean area of the convex hull.For a single planar Brownian motion of duration t and diffusion constant D, the mean perimeter L(t) = √ 16 π D t and the mean area A(t) = π D t were computed exactly in the mathematics literature quite a while back [6][7][8]. Very recently, there has been a growing number of articles both in the physics [9][10][11][12]14] and the mathematics literature [16][17][18][19] generalizing these results in various directions. In particular, adapting Cauchy's formula [22] for closed 2D convex curves to the case of random curves, a general method was recently proposed [9, 10] to compute the mean perimeter and the mean area of the convex hull of any arbitrary stochastic process in 2D. In cases where the process is isotropic in 2D, the mean perimeter and the mean area of its convex hull can be mapped onto computing the extremal statistics of the corresponding one dimensional component process [9,10]. This procedure was then successfully used to compute exactly the mean perimeter and the mean area of a number of isotropic 2D stochastic processes such as N independent Brownian motions [9, 10], random acceleration process [11], branching Brownian motion with absorption with applications to epidemic spread [12] and for anomalous diffusion processes [14].All these results pertain to isotropic stochastic processes in the unconfined 2D geometry. However, in many practical situations, the stochastic process takes plac...