We study the polarization of particles in relativistic heavy-ion collisions at very high energy along the beam direction within a relativistic hydrodynamic framework. We show that this component of the polarization decreases much slower with center-of-mass energy compared to the transverse component, even in the ideal longitudinal boost-invariant scenario with non-fluctuating initial state, and that it can be measured by taking advantage of its quadrupole structure in the transverse momentum plane. In the ideal longitudinal boost-invariant scenario, the polarization is proportional to the gradient of temperature at the hadronization and its measurement can provide important information about the cooling rate of the Quark Gluon Plasma around the critical temperature.Global polarization of hadrons produced in relativistic heavy-ion collisions has been recently observed by the STAR experiment over a center-of-mass energy range between 7.7 and 200 GeV [1,2]. This finding confirms early proposals [3,4], later predictions based on local thermodynamic equilibrium of spin degrees of freedom [5,6] which provides a relation between polarization and relativistic vorticity, and it agrees quantitatively with the hydrodynamic model calculations [7][8][9] to a very good degree of accuracy. In the hydrodynamic framework, the distinctive feature of polarization is its proportionality to the gradients of the combined temperature and velocity fields (see eq. (5)), so that its measurement is a stringent test of the hydrodynamic picture which is distinct and complementary to momentum spectra.The experimental efforts, thus far, focused on the search of the average global polarization of Λ hyperons along the direction of the angular momentum of the plasma. This measurement requires the identification of the reaction plane in peripheral collision as well as its orientation, that is the direction of the total angular momentum vector J. The average global polarization along J is also found to decrease rapidly as a function of centerof-mass energy [1], from few percent at √ s NN = O(10)GeV to few permille at √ s NN = O(100) GeV, in agreement with calculations based on the hydrodynamic model [8][9][10] as well with hybrid approaches [11,12]. In the TeV energy range, at the LHC, the global polarization along J is not seen [13] as it is most likely beyond experimental sensitivity.It is of course desirable to check more -possibly distinctive -predictions of the hydrodynamic model besides the global polarization along J. For instance, in ref. [14], a connection between local vortical structures in eventby-event hydrodynamics and correlation of polarizations of two Λ hyperons in transverse and longitudinal (along the beam line) directions has been studied. In this letter, we argue that in non-central heavy ion collisions, a non-zero longitudinal polarization of Λ with different transverse momenta p T is a more generic effect present in a simple non-fluctuating hydrodynamic picture, and propose to measure it in experiment [15]. As it will be sho...