Complementary vibrational spectroscopy

Vibrational spectroscopy, comprised of infrared absorption and Raman scattering spectroscopy, is widely used for label-free optical sensing and imaging in various scientific and industrial fields. The two molecular spectroscopy methods are sensitive to different types of vibrations and provide complementary vibrational spectra, but obtaining complete vibrational information with a single spectroscopic device is challenging due to the large wavelength discrepancy between the two methods. Here, we demonstrate simultaneous infrared absorption and Raman scattering spectroscopy that allows us to measure the complete broadband vibrational spectra in the molecular fingerprint region with a single instrument based on an ultrashort pulsed laser. The system is based on dual-modal Fourier-transform spectroscopy enabled by efficient use of nonlinear optical effects. Our proof-of-concept experiment demonstrates rapid, broadband and high spectral resolution measurements of complementary spectra of organic liquids for precise and accurate molecular analysis.

bending) in the IR spectrum 4 in addition to the specific peaks of benzene. Since there are some instrumental noise spikes in the CVS-Raman spectra, we evaluate them by comparing CVS-Raman spectra with and without the sample ( Supplementary Fig. 2).
Supplementary Figure 2| Noise spike assignment in CVS-Raman spectrum. We evaluate the instrumental noise by comparing the obtained spectra with and without sample. The spectrum without sample is normalized with the noise peak at 500 cm -1 . The common peaks around 500 -550 cm -1 , 750 cm -1 , 1,250 cm -1 and 2,500 cm -1 are attributed to the instrumental noise.

Supplementary Note 3: Comparison of multi-modal vibrational spectroscopy
Supplementary Table 1 shows comparison of the spectroscopic specifications of our CVS against the previously demonstrated multi-modal vibrational spectroscopy 5,6 . As clearly seen in the table, our CVS has broader spectral bandwidth spanning over 1,000 cm -1 in the fingerprint region with higher spectral resolution of around a few cm -1 , leading to the larger number of spectral elements than those of the previously demonstrated techniques.
There are advantages which make Fourier-transform spectroscopy (FTS) suitable for realizing the concept of CVS.
First, FTS is known for a molecular spectroscopy technique with broader spectral coverage and higher spectral resolution compared with the other techniques such as grating-based dispersive or frequency-swept spectrometers. This capability originates from the noble interferometric measurement with a single detector. In reality, highresolution and ultra-broadband molecular spectroscopy have been demonstrated with FTS-based systems both for infrared absorption and coherent Raman scattering spectroscopy 7,8 . Second, FTS can gain a speed advantage.
Especially, state-of-the-art FTS techniques such as ultra-rapid-scan FTS and dual-comb spectroscopy (DCS) drastically improve the spectral acquisition rate to for example 10 kHz without losing broadband and high-resolution capability. Third, FTS can provide more accurate frequency axis than the other methods due to the interferometric calibration with a reference continuous-wave laser. The accuracy can be ultimately up to the level of frequency standards when using a fully referenced dual-comb system. In the context of simultaneous measurement of IR absorption and Raman scattering spectra (complementary vibrational spectra), the frequency axes of the infrared and Raman spectra are calibrated at the same accuracy by a common interferometer, providing a big advantage for precise line position comparison of IR/Raman vibrational modes.

Supplementary Table 1| Spectroscopic specifications of CVS and other multi-modal vibrational spectroscopy techniques.
Spectroscopic specifications of CVS are compared with those of the other multi-modal vibrational spectroscopy in terms of spectroscopy method, observable spectral region, spectral bandwidth, spectral resolution and number of spectral elements.

Supplementary Note 4: Retrieval of MIR interferogram
The raw temporal mid-IR (MIR) waveform obtained by CVS contains several components including the linear MIR interferogram. The undesired components come from the additional MIR optical fields generated when the two near-IR (NIR) pulses are temporally overlapped in the nonlinear crystal. The situation is similar to the difference-frequency generation (DFG)-type interferometric (fringe-resolved) autocorrelation 9,10 . Supplementary Fig. 3a shows the AC part of the waveform obtained by a CVS measurement. It consists of several components such as the intensity autocorrelation, high-frequency fringes in addition to the MIR interferogram 9 . Therefore, the Fourier-transformed spectrum shows the corresponding parts as shown in Supplementary Fig. 3c. The lowest frequency part of the spectrum near zero frequency appears due to the slowly-varying intensity autocorrelation, while the spectral components which appear around 12,500 cm -1 and 25,000 cm -1 come from the high frequency fringes corresponding to the fundamental and second harmonic of the NIR pulses. The MIR spectrum appears at the spectral range of around 790 -1,800 cm -1 . Since the MIR spectrum is well separated from the other spectral components in the frequency domain, we can clearly observe the MIR spectrum without the need for correction in our demonstration.
We can extract and retrieve the MIR interferogram by eliminating the unwanted components if necessary. Since the high-frequency spectra are well-separated from the MIR spectrum, low-pass filtering removes the high-frequency fringes. The intensity autocorrelation component can be removed by subtracting a fitted intensity autocorrelation curve (derived by sech 2 pulses) from the raw waveform. The retrieved MIR interferogram and its spectrum are shown in Supplementary Fig. 3b and 3c, respectively. This retrieval would be necessary especially when measuring lower frequency part, for example, less than ~850 cm -1 in our demonstration, where the MIR spectrum may overlap the low frequency spurious spectral component. Retrieved MIR interferogram. c, FFT spectra of the raw temporal waveform (dotted curve in purple) and retrieved MIR interferogram (solid curve in red).

Supplementary Note 5: Sensitivity evaluation of CVS
We evaluate signal-to-noise ratio (SNR) of the CVS spectra of toluene. The SNR of the CVS-Raman signal (evaluated with the line at 1,003 cm -1 ) increases quadratically to the total NIR pulse energy (the ratio of 1 st and 2 nd pulse energy is kept constant at 1:2) and reaches 380 at the pulse energy of 2.8 nJ and the measurement time of 0.2 s ( Supplementary Fig. 4b).
Here, the SNR is evaluated as ratio of a peak intensity of the Raman signal and a standard deviation where no peaks exist. We verify that the dominant noise is the relative intensity noise of the NIR pulses.
The quadratic dependence on the excitation pulse energy can be explained by the cubic dependence of the signal intensity due to the χ (3) nonlinear process ( Supplementary Fig. 4a) and the linear dependence of the relative intensity noise. On the other hand, SNR of the CVS-IR linearly increases to the average power of the MIR pulse and reaches above 150 at the average power of ~ 0.4 µW (estimated from the signal voltage of the detector and its responsivity) and the measurement time of 0.55 s (Supplementary Fig. 4c). The SNR of the CVS-IR spectrum is evaluated by the inverse of a standard deviation of the transmittance spectrum where no peaks exist. The detector noise is dominant in the condition of our demonstration. Note that the SNR of CVS-Raman/IR can be improved by optimizing the system and/or implementing some techniques such as heterodyne detection for CARS measurement. Next, we evaluate detection limit of the CVS by measuring mixture of liquid benzene and chloroform. As shown in Supplementary Fig. 5a, the peak intensity of the CVS-Raman signal at 992 cm -1 shows linear dependence on the concentration, while the CARS process, in principle, is expected to show quadratic dependence. It is known that the heterodyne detection with a local oscillator can lead to linear dependence and we assume our FT-CARS is operated in a condition of the heterodyne detection with leaked NIR pulses through the optical filters. Estimating from the peak intensity and the noise level, the detection limit of benzene is to be ~0.1% with the measurement time of 0.2 s.
Similarly, we measured absorbance of the CVS-IR spectrum with an isolated line of benzene at 1,394 cm -1 and confirm linear dependence on the concentration ( Supplementary Fig. 5b). The detectable absorbance of this relatively weak absorption line is estimated to be ~4×10 -3 , which corresponds to the density detection limit of ~2%, with the measurement time of 0.55 s. The detection limit is to be ~ 0.2% when observing the stronger absorption line at 1,480 cm -1 .
Supplementary Figure 5| SNR of CVS against concentration. a, Peak intensity of CVS-Raman signal at 992 cm -1 against the