Sketch of the geometry of the pump–probe experiment at the femtoslicing synchrotron beam line at BESSY. Time-resolved XMCD allows measurement of the ultrafast dynamics of spin and orbital momenta along the quantification axis z parallel to the applied magnetic field. Optical pulses with a central wavelength of λ_pump = 790 nm and a duration of τ_pump = 60 ± 20 fs excite the ferromagnetic films perpendicularly, aligning the electric vector E in the film plane. The density of absorbed laser energy is E_abs = 12 mJ cm⁻². The ellipsoidal shape of -E_PMA illustrates the perpendicular anisotropy of the film. The easy magnetization direction is defined by the largest value of the z-axis projected value of L, (L_z). On applying the external magnetic field H_ext, the spin magnetic moment S aligns parallel to the orbital magnetic moment L along the z axis. A variable delay can be set between the near-infrared pulse and the X-ray probe pulse.

Two XAS spectra (red and black) and the normalized difference spectrum XMCD (line in blue) at the Co L_2,3 edges are displayed for the 15-nm Co_0.5 Pd_0.5 film in normal incidence geometry with a magnetic field of ±4 kOe, collinear with the incident circularly polarized X-rays. The two XAS spectra are obtained by extracting the logarithm of the transmitted X-ray signals recorded by the photodiode. This avoids any saturation effects in the XAS signal (Supplementary Information). Integration of the energy resolved XMCD spectrum (green curve) allows a quantitative determination of the static values (without pump) of the spin and orbital magnetic moments at t < 0: (S_z)_stat = 0.78 ± 0.01  per atom and (L_z)_stat = 0.24 ± 0.01  per atom. A large number of pairs of XAS spectra allow such an accurate determination. a.u., arbitrary units.

a, b, Time-resolved XMCD at the Co L₃ and Co L₂ edges measured on a 15-nm Co_0.5Pd_0.5 alloy film using femtoslicing. The film has an out-of-plane anisotropy. A linear relation connects the two Co L₃ and Co L₂ XMCD signals with the two magnetic moments^{19, 20} L_z and S_z . The dashed lines for Co L₃(t) and Co L₂(t) are calculated from the fitted time evolutions of L_z(t) and S_z(t), for which the resolution broadened double exponential function were used (see Fig. 4a continuous lines and Supplementary Information). c, Time dependence of the electronic Co L₃ XAS intensity recorded at energy E = 780.8 eV. The measured electronic XAS difference (for pumped and non-pumped sample) is related to the electronic changes observed in the valence band of the Co 3d bands when excited with the laser pulse. The pump conditions are identical to those used for the magnetic data (density of absorbed laser energy: E_abs = 12 mJ cm⁻²). The continuous line is a time dependent simulation using the resolution broadened double exponential given in Supplementary Information. The initial rise time τ_e-e = 220 ± 20 fs corresponds to the thermalization of the electrons at the Fermi level following the laser excitation. The subsequent decay time of 360 ± 20 fs is related to the relaxation time of the electrons to the lattice. Error bars of experimental data show standard deviation.

a, Sum rule extracted effective spin and orbital magnetic moments S_z(t) and L_z(t) as a function of the delay time between the laser pump and the X-ray probe. The continuous lines are fits obtained by using a 130-fs FWHM Gaussian function accounting for the time resolution of the experiment (including the X-ray probe and the femtosecond laser pump). The blue dashed line represents the fit to L_z(t) scaled to the value of S_z(t) before laser excitation. The characteristic thermalization times of both spin and orbital magnetic moments are: τ_th(L_z) = 220 ± 20 fs and τ_th(S_z) = 280 ± 20 fs. The fitting procedure and error bars providing the thermalization times are explained in Supplementary Information. The difference between both thermalization times is 60 ± 30 fs, showing a faster decrease of the orbital moment. The maximum relative variation for S_z(t) is –55 ± 3%, whereas for L_z(t) it is –67 ± 6.5%. b, The ratio (L_z/S_z)(t) obtained as a function of the delay time shows that the orbital magnetic moment reduces more than the effective spin magnetic moment during the ultrafast demagnetization process. The black continuous line is the ratio between the two simulations of L_z(t) and S_z(t), showing a relative variation of –29 ± 5%. The red line is the ratio obtained when we take two identical values τ_th(L_z) = τ_th(S_z) = 260 fs. The error bars for L_z(t), S_z(t) and (L_z/S_z)(t) are obtained from the error bars of the time resolved XMCD at the Co L₂ and Co L₃ edges.