of both the central wavelength and bandwidth. In this particular case, the picosecond Raman pump pulse is centered at 793 nm with bandwidth of 1.1 nm and a duration of several picoseconds. Its intensity is kept relatively high (∼1 μJ pulse−1) in order to optimize the amount of Raman scattering produced. The Raman probe pulse is obtained through supercontinuum generation (400–1000 nm) in a sapphire plate, followed by compression in a prism pair wherein only the near‐IR portion of the continuum is retained. In this way, a 80 fs pulse is obtained covering the 870–950 nm region, that corresponds to 1000–2000 cm−1 Raman shifts from the Raman pump. The intensity of the Raman probe is much lower than that of the Raman pump, typically in the range of tens of nJ pulse−1. Finally, the three pulses are made collinear and focused in the same spot (sized a few tens of μm) of the flow cell containing the sample. After traversing the sample, the Raman probe beam is spatially selected, goes through a spectrograph, and analyzed in an array detector.
Figure 3.7 A possible configuration for a time‐resolved stimulated Raman experiment.
Source: Adapted with permission from McCamant et al. [60]. Copyright (2003) American Chemical Society.
Several variants of this general scheme are possible. In some setups, for example, tunable Raman pump and probe pulses are both obtained by NOPAs [62]. In other cases, the fundamental beam from the amplifier was directly used as the Raman probe [63]. In regard to detection, some setups make use of a reference beam [60]: the probe is split in two in order to generate a reference beam, allowing for efficient shot‐to‐shot normalization of white light fluctuations. Experiments have been conducted over a wide range of Raman pump wavelengths, from the near IR to the UV [64]. This wide flexibility also allows to tune the Raman pump for pre‐resonant [63] and resonant [53] Raman process, which provides a convenient route to further enhance the Raman signal, and to single out the Raman contribution from the chromophore of interest.
In regard to data acquisition, several pulse chopping sequences can be used in this type of experiments [52]. By a simple chopper in the Raman pump path, one can acquire the stimulated Raman spectrum from the ratio between Raman‐pumped and Raman‐unpumped spectra cyclically recorded by the array detector. Then, the Raman spectra of the ground state and photoexcited sample can be obtained sequentially, and compared to each other, by shuttering the actinic pump. If the chopper is positioned in the path of the actinic pump, the ratio between chopped and unchopped signals directly records the difference Raman spectrum induced by photoexcitation. More complex schemes where both the actinic and Raman pump are chopped are sometimes used as well [64].
3.5.3 Data Analysis and Interpretation
Data analysis of TRR data involves a series of additional steps which do not have a counterpart in TA, and the details of which depend on the specific pulse sequence used in the acquisition. A typical analysis workflow is illustrated in [65]. If chopping the actinic pump only, one can directly acquire the difference Raman spectrum induced by photoexcitation. The latter will always “sit” over a broad background due to transient absorption changes of the Raman probe beam. Therefore, one always needs to separately measure the TA and subtract it from the recorded signal. The signal after TA subtraction contains the excited state Raman spectrum, ground state Raman bleach contributions, but also some residual contributions of the solvent, due to attenuation of the Raman pump by photoexcited molecules. Therefore, one needs to separately measure and subtract the pure solvent Raman signal, which is rescaled and subtracted in order to get the final difference spectrum. Some specific additional problems which are not encountered in TA may arise both at the time of setup alignment or data processing. For example, the data may contain spurious contributions due to the cross‐phase modulation between Raman pump and probe beams. Several strategies have been proposed to minimize these effects, such as vibrating the Raman pump retroreflector to average out CPM effect [61].
TRR experiments exploiting stimulated Raman scattering are now very well‐established in the toolbox of ultrafast spectroscopies. They are powerful tools capable of following the ultrafast dynamics of photoexcited systems by revealing structurally precise details of the undergoing events through the changes of their vibrational mode pattern. The interpretation of time‐resolved vibrational data follows basically the same principles as ordinary steady state vibrational spectroscopy, where characteristic fingerprint vibrations are highly structure‐specific and, in some frequency regions, can be traced back to specific chemical moieties. The main difference with a steady state Raman experiment is that the vibrational properties change as a result of the electronic redistribution induced by photoexcitation, and of subsequent system dynamics. For example, molecular internal conversion can be pinpointed unambiguously through the appearance and disappearance of the unique fingerprint vibration patterns associated to each of the involved states [66]. On these grounds, TRR methods have been extensively used to interrogate ultrafast photochemical processes such as charge transfer, proton transfer, isomerization, or internal conversion, as demonstrated by many examples in the literature of the last two decades [50, 53, 66, 67].
3.6 Case Studies
3.6.1 Ultrafast Relaxation Dynamics of Molecules in Solution Phase
Transition metal complexes, such as metal–polypyridine complexes, are very interesting for several photochemical and solar energy harvesting applications, such as dye‐sensitized solar cells (DSSCs). The excited state dynamics of these molecules display several phenomena which make them a particularly interesting topic in the “ultrafast” community. One of the typical electronic absorption bands of many metal complexes, named metal‐to‐ligand charge transfer (MLCT), involves an optical charge transfer from the metal to the ligands around it. Excitation within MLCT electronic transitions is the first step toward electron injection in DSSCs, and gives rise to peculiar dynamical behavior which can only be addressed by ultrafast spectroscopic techniques.
It is known [68] that photoexcitation within the singlet 1MLCT state is typically followed by relaxation into a long‐lived triplet 3MLCT state, through an intersystem crossing (ISC) process, which can be surprisingly fast, well in the sub‐picosecond range. Because of such an extremely short 1MLCT fluorescence lifetime, its quantum yield is so small to make it practically invisible in steady state measurements. For the same reason, a direct observation of ISC in real time is very challenging, but can be achieved by FLUC, which allows to directly detect 1MLCT fluorescence and follow its dynamics as a function of time.
An example of this is reported by Cannizzo et al. [34]. The authors of this paper investigated the dynamics of an archetypal metal complex ([Ru(bpy)3]2+) by ultrafast fluorescence upconversion. In particular, they reported for the first time a FLUC experiment where the detection covered a broad wavelength range (440–690 nm) and, simultaneously, a time resolution of ≈110 fs. Photoexciting the metal complex with a very short excitation pulse at 400 nm allows them to follow the dynamics from the earliest stage, thus capturing the whole relaxation of the system.
The signal observed by FLUC [34] can be described as an emission band, peaking at 19 230 cm−1, which is attributed to 1MLCT emission and grows together with the excitation pulse. This band then decays within 200 fs while a weaker band appears at 16 500 cm−1. The latter corresponds to the steady state emission from the triplet state 3MLCT.
The detailed analysis of the temporal kinetics demonstrates that the bluest 1MLCT band rises and decays during the duration of the excitation pulse, and that also the rise of the redder band occurs during the pulse duration, suggesting that the population of the 3MLCT state is driven by an extremely fast ISC which lasts much less than the experimental time resolution. Deconvolution of the instrumental response function suggests an ISC timescale smaller than 30 fs, which implies