The laser-driven ultrafast demagnetization effect is one of the long-standing problems in solid-state physics. The time scale is given not only by the transfer of energy, but also by the transport of angular momentum away from the spin system. Through a double-pulse experiment resembling two-dimensional spectroscopy, we separate the different pathways by their nonlinear properties. We find (a) that the loss of magnetization within 400 fs is not affected by the previous excitations (linear process), and (b) we observe a picosecond demagnetization contribution that is strongly affected by the previous excitations. Our experimental approach is useful not only for studying femtosecond spin dynamics, but can also be adapted to other problems in solid-state dynamics.

Prior to the development of pulsed lasers, one assigned a single local temperature to the lattice, the electron gas, and the spins. With the availability of ultrafast laser sources, one can now drive the temperature of these reservoirs out of equilibrium. Thus, the solid shows new internal degrees of freedom characterized by individual temperatures of the electron gas Te, the lattice Tl and the spins Ts. We demonstrate an analogous behavior in the spin polarization of a ferromagnet in an ultrafast demagnetization experiment: At the Fermi energy, the polarization is reduced faster than at deeper in the valence band. Therefore, on the femtosecond time scale, the magnetization as a macroscopic quantity does not provide the full picture of the spin dynamics: The spin polarization separates into different parts similar to how the single temperature paradigm changed with the development of ultrafast lasers.

The ultrafast demagnetization process allows for the generation of femtosecond spin current pulses. Here, we present a thermodynamic model of the spin current generation process, based on the chemical potential gradients as the driving force for the spin current. We demonstrate that the laser-induced spin current can be estimated by an easy to understand diffusion model.