Subsolidus and melting experiments with a K-doped peridotite (K2O0.64 wt%) were performed at 10--27 GPa in order to investigate the behavior of potassium in the deep mantle, and the element partitioning between Mg-perovskite (Mg-pv) and magnesiowustite (mw). Three new K-rich phases were discovered, K phase I2FeMg4Si5O16> in coexistence with wadsleyite, pyroxene and garnet at 15 GPa, K phase II<(K, Mg)(Al, Si)O4> with CaSiO3-rich perovskite (Ca-pv), ringwoodite and majoritic garnet at 20 GPa, and K phase III <(K, Na, Ca)(Mg, Fe, Cr, Al, Ni, Ti, Mn)4(Si, Al)4O12> in coexistence with Mg-pv, mw, and Ca-pv at 25.5 GPa. These three phases are the main K-bearing phases in the deep peridotite mantle, rather than K-hollandite <KAlSi3O8>. With the exception of pyroxene at relatively low pressure, the solubilities of K2O in other mantle phases, such as wadsleyite, ringwoodite, and garnet are very low (<200 ppm), even under the conditions saturated with one of the K phases. At 25.5 GPa, the contents of K2O in Mg-pv, mw, and Ca-pv are also very low and decrease with decreasing temperature. The K-rich phases should appear in low concentrations in the transition zone and the lower mantle, assuming a pyrolitic mantle composition (K2O,~300 ppm). In contrast to basaltic composition, K-phases in peridotite are the first to melt at pressures up to 27 GPa, and thus K is incompatible in the whole experimental pressure range. The solidus temperature of K-rich peridotite increases drastically from about 1600 ¿C at 20 GPa to 2300 ¿C at 25.5 GPa. Portions of the upper mantle could be partially molten even at the present mantle temperatures. Ascending lower mantle material will partially melt to a very low degree around the transition zone/lower mantle boundary, forming a K-rich melt. This melt would strongly affect the geophysical properties of the transition zone, and may be directly related to the worldwide mantle metasomatism and K-rich magmatism. For partitioning of Fe/Mg, Na2O, NiO, Cr2O3, and MnO between Mg-pv and mw, no evident variation with temperature or pressure was detected. The results show that most of the reported (Mg, Fe)O and MgSiO3 inclusions in diamonds were not in equilibrium. ¿ 2000 American Geophysical Union