Optimized interferometric setup for chiral and achiral ultrafast IR spectroscopy

: We report an actively stabilized interferometer-based set-up for the detection of vibrational circular dichroism (VCD) and optical rotatory dispersion (VORD) with femtosecond laser pulses. Our approach combines and improves elements of several previous measurement strategies, including signal amplification in a crossed polarizer configuration, precise control and modulation of polarization, phase stability, tight focusing, broad-band detection and spectral interferometry. Their importance for static and transient measurements is motivated by a signal analysis based on Jones matrices and response theory. Only depending on the pump-beam polarization, the set-up can selectively detect transient VCD and VORD or transient linear birefringence (LB) and linear dichroism (LD), which usually constitute the dominant artifacts in the chiral measurements. For illustration we present transient LB and LD data of an achiral Rhenium carbonyl complex, detected simultaneously by spectral interferometry, and we analyze residual background signals in the experimental configuration for transient chiral spectroscopy. Abstract: We report an actively stabilized interferometer-based set-up for the detection of vibrational circular dichroism (VCD) and optical rotatory dispersion (VORD) with femtosecond laser pulses. Our approach combines and improves elements of several previous measurement strategies, including signal ampliﬁcation in a crossed polarizer conﬁguration, precise control and modulation of polarization, phase stability, tight focusing, broad-band detection and spectral interferometry. Their importance for static and transient measurements is motivated by a signal analysis based on Jones matrices and response theory. Only depending on the pump-beam polarization, the set-up can selectively detect transient VCD and VORD or transient linear birefringence (LB) and linear dichroism (LD), which usually constitute the dominant artifacts in the chiral measurements. For illustration we present transient LB and LD data of an achiral Rhenium carbonyl complex, detected simultaneously by spectral interferometry, and we analyze residual background signals in the experimental conﬁguration for transient chiral spectroscopy.


Introduction
Vibrational Circular Dichroism (VCD) spectroscopy in combination with DFT calculations is an important analytic tool for determining the configuration of chiral organic molecules; VCD is also commonly used to probe the secondary structure of peptides and proteins and to detect their aggregation. VCD spectroscopy measures the differential absorption (∆A VCD = A L − A R ) of left circular and right circular polarised infrared light, which is opposite for the two enantiomers of a chiral molecule, whose VCD spectra are therefore mirror images. The most widespread approach [1] to detecting VCD is an extension of FTIR spectroscopy where a photo elastic modulator (PEM), placed between the interferometer and the sample, modulates the lightpolarization continuously from left to right circular. The differential interferogram (L minus R) is isolated by lock-in-amplification referenced to the PEM modulation frequency and Fourier transformation yields the VCD spectrum. However, the inherent poor temporal resolution of conventional FTIR spectrometers makes it impossible to resolve fast dynamics. VCD spectra have been recorded in step-scan mode [2,3], where the time resolution would be ultimately limited by the PEM modulation frequency (30-50 kHz), but we are not aware of any transient measurements carried out in this way. In order to overcome the limitations imposed by polarization modulation, Polavarapu and coworkers used a polarization division interferometer and demonstrated time-resolved linear dichroism with microsecond time-resolution [4]. Such an interferometer can also measure VCD with very good signal to noise (S/N) [5]. Because of the low brilliance (light intensity per unit bandwidth and time) of thermal IR sources, however, S/N quickly deteriorates when integration time windows are reduced, making VCD measurements with high time resolution very difficult. Fischer and co-workers [6] have demonstrated that a tunable quantum cascade laser (QCL) can serve as an alternative, much more intense light source for steady state VCD detection. QCL-based monitoring of VCD changes in time has also been reported [7], but these experiments have not yet achieved a time resolution beyond what is already possible with commercial FTIR apparatus [8].
Some of the fundamental limitations encountered with continuous light sources can be overcome when ultrashort mid-IR laser pulses are used instead. Femtosecond IR laser pulses combine a very high peak power with a useful spectral bandwidth of 150-200 cm −1 , and the timeresolution of pump-probe experiments is no longer limited by the light intensity or the detector speed, but only by the duration of the laser pulses. In a first proof-of-principle experiment, our group dispersed femtosecond IR laser pulses in a monochromator and modulated the polarization of the narrow bandwidth output from left to right-handed circular on a shot-to-shot basis with a PEM [9]. In order to minimize artifacts [10], the light was focused through the PEM onto the sample and imaged directly onto a single element MCT detector. Static VCD spectra could be recorded with good signal to noise and, in a pump-probe experiment, changes of the chiral mid-IR signal could be detected for the first time with picosecond resolution [9,11].
Cho and co-workers showed that it is also possible to record static VCD spectra by dispersing femtosecond laser pulses after they have interacted with the sample [12][13][14][15]. In order to avoid artifacts, very high quality linear polarizers were needed, which project out only one polarization component immediately behind the sample, while blocking almost completely the incident light (crossed polarizer set-up). The group used spectral interferometry, employing a replica of the laser beam, to determine simultaneously absorptive VCD and dispersive VORD (vibrational optical rotatory dispersion) spectra. An important advantage of this approach is the possibility to use an array detector to record spectra in a 'single shot' without scanning gratings or time-delays; another benefit is an intrinsic signal to noise enhancement afforded by the crossed-polarizer geometry in combination with the high light intensities [16]. Both advantages can equally be exploited in equivalent single beam measurements using elliptical polarized pulses [16,17].
Despite many efforts, however, the crossed polarizer approach and array detection have not yet been extended to the recording of transient chiral signals in the infrared, although similar experiments have been carried out successfully in the UV-visible spectral range [18][19][20]. In this paper we address a number of experimental difficulties that have, by our experience, so far prevented a straightforward combination of our early single wavelength proof-of-principle time-resolved experiment with signal enhancement and broad band detection techniques. Our most recent experimental set-up, which we present here in detail, is designed to overcome these problems.

Motivation
Figures 1(a) and 1(b) schematically show the two types of set-up which have previously been used for VCD measurements with femtosecond laser pulses, while Fig. 1(c) illustrates our new approach presented in this paper. The purpose of this section is to motivate our modifications, based on a brief summary of the theoretical background, which has been presented previously [13,16]. A detailed description of the new layout follows in the next section.

Measurement principle
An isotropic sample of chiral molecules with thickness L, absorbance εcL = κL/ ln(10), refractive index n = 1 + ρλ /(2π), and small optical rotation δ L/2 = (n L − n R )πL/λ and circular dichroism (ε L − ε R )cL = ηL/ ln(10) is described by the Jones matrix In a crossed polarizer set-up we detect the new, perpendicular field component, created when a linear x-polarized light beam propagating in the z-direction interacts with the sample (Jones vector E 0 0 ). We call this new field component the signal field, which is given by E S is very small and can only be detected interferometrically, by overlap with a so-called local oscillator field E LO = A(ω)e iψ(ω,τ) e y In conventional CD and VCD measurements, we can think of E 0 and E LO ≈ ±i E 0 as the two out of phase components of circular polarized light. In analogy to the differential CD measurement, in a crossed polarizer set-up we measure the change in signal when the sign of either E LO or E 0 (and therefore E S ) is flipped: The y-component of perfectly circular polarized light is phase shifted by exactly π/2 with respect to the x-component and it experiences the same absorption and dispersion as the signal field, so As a result, κ and ρ drop out of the expression in Eq. (4) and we obtain the CD signal S 0 = ηL/ ln(10).  [9]. (b) Crossed polarizer setup for spectral interferometry as introduced by Cho and coworkers [12,13] (c) New setup which combines elements of a and b. Blue circles (±x) and purple and orange arrows (±y) represent the polarization direction and sign of the different fields for four consecutive laser triggers.
In the more general case, however, the phase ψ(ω, τ) and amplitude A(ω) of the local oscillator field is not exactly known. A second measurement must be performed under identical conditions, where the signal field E S is replaced by (a fraction of) the incident field E 0 . In practice, this is done by tilting the analyzer by small angle α 1, which yields the second signal In a final step the full complex signals must be reconstructed from the measured real parts S 0 and S α , i.e. one needs to find the corresponding imaginary parts. This can be achieved by Hilbert or Kramers-Kronig transformation yielding In spectral interferometry, the full complex signals are usually computed by imposing causality. After inverse Fourier transform of the real signals S 0 and S α with respect to ω, all amplitude at negative times is set to zero (multiplication by the Heavyside step function θ (t)) and the resultant signal is Fourier transformed back into frequency domain [21]: The two signals can now be divided which yields the absorptive and dispersive spectra: Multiplication with appropriate scaling factor then yields the circular dichroism and optical rotation spectra: ln(10) S out of phase (12) Note that although both VCD and VORD spectra are obtained simultaneously, the signal processing relies on a Kramers-Kronig transformation in a restricted spectral window, just like a post-measurement determination of the VORD spectrum from a measured VCD spectrum. Based on these equations and the schematic comparison of different laser-based setups in Fig. 1, the new elements in our current approach can be discussed:

The crossed polarizer advantage
The magnitude of the chiral signal S 0 is proportional to the ratio |E 0 |/|A(ω)| or, in other words the ratio of the x-polarized incident field and the y-polarized local oscillator field. With respect to a conventional CD measurement with circular polarized light (or a polarizing interferometer without additional analyzer), where |E 0 | = |A(ω)| this can afford significant signal enhancement. With typical femtosecond laser pulses, E 0 can be made 5-10 times larger than the local oscillator |A(ω)|, which still remains strong enough to nearly saturate the MCT detector. Measurements with thermal light sources, on the other hand, are usually light-limited, meaning that any increase of E 0 at the cost of A(ω) would lead to increased noise levels caused by poor illumination of the detector. Most importantly, after the high-quality analyzer immediately behind the sample, polarization-sensitive optics like a monochromator grating or mirrors can no longer distort the chiral signal by affecting differently the signal and local oscillator fields. It is therefore possible to disperse the broad band femtosecond pulses and record them with an array detector as in conventional time-resolved IR spectroscopy. Spectra can be recorded much more quickly (in principle with a single laser shot) and afford highly correlated noise, which makes it easier to identify small features.

Polarization modulation
Polarization modulation in conventional VCD measurements ideally corresponds to a periodic reversal of either the signal or the local oscillator field, or, in practice, to a λ /(2c) time shift of one polarization component with respect to the other. When these two perpendicular polarized fields propagate as separate beams as in Fig. 1(b), VCD may also be measured by periodically blocking one of them with a chopper [13]. However, in that case the measured signals are only half as large as with a modulator scheme shown in Fig. 1(c). It is much more precise to induce an exact λ /(2c) time shift with a PEM and define the polarization state afterwards with a high quality polarizer than to directly modulate the polarization state in a single beam experiment [10]. In the first case, only the intensity of consecutive pulses must be compared when optimizing the PEM retardation. At the same time the detector sees only the other, ypolarized, unmodulated beam, which is not affected by a possible small PEM-modulation of the beam divergence (see appendix). This is also a big advantage over some of our earlier attempts to realize crossed-polarizer schemes in single beam experiments [10,17].

Phase stability
For a proper signal processing according to Eq. (8), the two signals S 0 and S α must be measured under identical conditions, i.e. the phase ψ(ω, τ) must not change over the course of the two measurements. For example, if the local oscillator is a time-shifted copy of the x-polarized beam, ψ(ω, τ) = (ω − ω 0 )τ and S α exhibits regular fringes in frequency, with a characteristic spacing inversely proportional to the time shift τ. Consequently, τ must be stable within a small fraction of the light period during the measurements. Active beam path length stabilization is required to enable the long data averaging necessary for transient VCD measurements.

Co-propagation of the local oscillator beam
Both Eqs. (4) and (6) contain contributions due to absorption and dispersion, which may change in time or by laser excitation in a time-resolved experiment. However, κL and ρL disappear from the expressions for S 0 and S α if the local oscillator passes through the sample, just as in the case of circular polarized light. The new scheme Fig. 1(c) combines this advantage of the single beam experiment (scheme a) with the independent intensity and polarization control of E S and E LO . Similar considerations apply in transient 2D-IR spectroscopy, where the LO beam must also be guided through the excited sample volume [22].

Beam focusing
In order to reach excitation levels of at least a few percent in pump-probe experiments, the laser beams must be tightly focused to a spot size of approximately 100 µm diameter on the sample. This must be achieved without perturbing the high quality of polarization of the individual beams and thus the extinction of the crossed-polarizer set-up. Tight focusing also allows to exchange the laser-excited sample on a pulse-to-pulse basis in a flow cell.

Setup
A detailed layout of our new setup is shown in Fig. 2. Infrared light from a mid-IR optical parametric amplifier (OPA) is split by a polarizing beam splitter (PBS, holographic wire grid polarizer, Thorlabs WP25H-B, extinction ratio ≈ 10 −2 ) into two parts. The horizontally polarized component is transmitted and vertically polarized light is reflected. The relative strength of transmitted and reflected beams can be adjusted by a half-wave plate (HWP) in front of the beam splitter. The transmitted beam makes a double pass through a ZnSe (zinc selenide) photoelastic modulator (PEM, Hinds Instruments) placed at the focal point of a telescope, which acts as a quarter-wave plate and causes a ±90 • rotation of the polarization. The returning beam is thus reflected by the wire-grid beam splitter and is guided through a Brewster angle germanium wedge polarizer (extinction ratio < 10 −7 ) [23], which defines the x-direction in the laboratory. We call this beam the signal-generating beam.
The second arm of the interferometer comprises a delay line, a ZnSe wedge, and a MgF 2 zero order half-wave plate (HWP, Bernhard Halle Nachf., RZM4.15L), which rotates the polarization of the initially vertically polarized beam. This beam is called local oscillator (LO) beam. The polarization of the LO-beam is cleaned by a free-standing wire grid polarizer (Infraspecs P03, extinction ratio < 10 −4 ) oriented in the y-direction, such that its intensity can be fine-tuned independently by adjusting the wave plate. In order not to destroy the purity of polarization, we use the last wedge of the Brewster angle polarizer in the signal-generating beam to unite signalgenerating and LO-beams. As a result, this interferometer synthesizes pairs of pulses with very pure orthogonal polarization with freely adjustable intensity and delay. They are focused into the sample by a CaF 2 lens of 10 cm focal length and collimated again by an identical lens before passing the analyzer. The analyzer design is identical to that of the pump-beam polarizer and consists of four germanium wedges, placed in a solid mount at Brewster angle (14 • ) as shown in Fig. 2(a). It only transmits the y-polarized LO (E LO ) and the chiral signal (E S ) but completely blocks the x-polarized signal-generating beam (E 0 ). The transmitted light is then dispersed in a monochromator by a 300 lines/mm grating onto a nitrogen cooled MCT array detector (MercuryCadmiumTelluride, Infrared Associates, 2x32 pixels), which is read out on a shot-to-shot basis. The second array of this detector records a fraction of the OPA output which is split off before the interferometer and used as a reference to correct for intensity fluctuations. The PEM is oriented at 45 • with respect to the incoming light polarization and acts as a quarter-wave plate with alternating sign. The femtosecond laser is synchronized to the PEM to ensure that every pulse precisely meets the trough or the crest of the modulation, as described in detail in reference [10]. A double pass through the PEM thus alternately turns the electric field vector of the signal-generating beam into the ±x direction (indicated by ⊕ and in Fig. 2 ). The PEM is placed at the focal point of a telescope to minimize spatial modulation of the beam (see appendix). This and the double pass also allow us to drive the PEM with low voltage, which minimizes electronic noise. The thickness of the ZnSe wedge in the LO beam (1.5875 cm) matches that of the PEM in the pump-beam and introduces similar frequency dispersion (chirp). We observed highly unsymmetrical temporal interferograms between pump and LO beam without the ZnSe wedge in the LO beam path. A helium neon laser beam co-propagates 4 cm above the IR beams, either via the same mirrors or (in the case of the beamsplitters and the telescope) via optics on shared mounts. The He-Ne laser interference serves as a feedback signal to actively stabilize the path difference in the interferometer to within approximately 100 nm via a piezo-mounted mirror in the LO-beam path. In combination with quadrature counting, the HeNe interferogram also permits precise delay tracking and FTIR operation of the set-up, when single element MCT detectors replace the monochromator and array detector and τ is scanned rapidly [24].

Static measurements
In order to test the performance of our setup for static chiral measurements, we recorded spectra for nickel-sparteine complexes [Ni((-)-sparteine)]Cl 2 and [Ni((+)-sparteine)]Cl 2 , dissolved in CDCl 3 at a concentration of 0.7 M. The samples were circulated in a 100 µm thick flow cell to mimic the conditions in a transient measurement.
The top panel of Fig. 3 shows the signals S 0 for the two enantiomers in blue and red, respectively, while the green line represents the solvent background. The enantiomers clearly show opposite signals, while the achiral interferograms S α of the leaked signal-generating beam and the LO in the second panel are almost indistinguishable. The corresponding absorptive (VCD) When the two arms of our interferometer have almost equal length (to within one wavelength of the infrared light), the interferometer should act like a zero order wave plate and can be viewed as a very precise synthesizer of a single femtosecond laser pulse with arbitrary polarization. For example, when the path difference is ±λ /4, it should produce elliptical pulses of opposite handedness, and S 0 should be directly proportional to the absorptive VCD spectrum, just like in our earlier single beam measurements [16]. Likewise, a pathlength difference modulated between 0 and λ /2 should yield linear polarized pulses for a direct measurement of the dispersive VORD spectrum. In Fig. 4 we show a series of signals S 0 , acquired with path differences within one wavelength from zero. Comparison with Fig. 4 immediately shows that none of the spectra are either purely absorptive (VCD) or dispersive (VORD) over the whole spectral window, a direct consequence of the spectral dependence of ψ(ω, τ) caused by different dispersion in the two interferometer arms. For comparison, also in conventional VCD measurements with FTIR spectrometers, the instrument-dependent spectral phase must be known in order to extract the VCD spectrum from the chiral interferogram. In femtosecond experiments with a spectrometer, on the other hand, the spectral phase is eliminated by division with the complex achiral signalS α (Eq. (12)), of which we can only measure the real part S α (Fig. 3(b)). For the Kramers-Kronig transform Eq. (9) we need a sufficiently large delay τ between local oscillator and the signal-generating field in order to minimize the intensity of F −1 {S α } near t = 0 and thus avoid artifacts. The optimal interferometer delay is thus not equivalent to a zero order wave plate but rather corresponds to tens of cycles of the IR fields. We obtained fully reproducible VCD and VORD spectra in the interval 0.2 ps< τ <1 ps. Beyond the upper limit for τ, spectral fringes are no longer sufficiently resolved by our 32 pixel MCT array detector. Concerning the choice of the analyzer tilt angle α for the measurement of S α , one degree turned out to be sufficient for very good signal to noise affording even shorter averaging than for S 0 . Much larger tilt angles should be avoided in order to limit the alignment and light level changes at the detector with respect to the S 0 measurements. Although the intensity ratio of signal-generating and LO beams does not change the amplitudes of the final VCD and VORD spectra, the individual signals S 0 and S α are directly proportional to |E 0 |/|E LO | in the crossed polarizer approach, and signal to noise can be significantly enhanced by increasing this ratio. When doing so, the limiting factor in IR-measurements is almost always the available light intensity (long before for this proportionality is no longer valid [25]). In our case, the poor transmission (≈30%) of the Germanium wedge polarizer, and a further 40% intensity loss inside the interferometer mean that the signal-generating beam can only be 25 times more intense than the LO beam at the sample (|E 0 |/|E LO | ≈ 5). Doubling this factor and thus the signal to noise ratio would require a fourfold increase in laser intensity.

Transient measurements
For our pump-probe experiments we periodically block the visible pump beam at one quarter of the laser frequency and thus measure the intensity at the detector twice for each modulator state (see Fig. 1). This allows us to calculate S 0 (Eq. (4)) and henceS 0 (Eq. (8)) once with the pump laser on and once with the pump laser off. In addition,S α must be determined independently from a measurement with tilted analyzer. We found that pump-induced changes ofS α are usually negligible, meaning that it needs to be measured only once and not separately for each pump-probe delay. In analogy with Eq. (10) we can thus evaluate the transient out of phase and in phase signals: In an ideal setup, where the pump beam polarization is either perfectly parallel or perpendicular to E 0 and E LO , these signals represent, up to the appropriate scaling factors, the transient VCD and ORD spectra respectively. However, the potential distortion of time-resolved circular dichroism signals by pump-induced linear dichroism (LD) and linear birefringence (LB) have already been discussed in the context of the first CD experiments with visible picosecond lasers by Xie and Simon [26,27]. In our description of the experiment on the basis of linear polarized fields and simultaneous consideration of both absorptive and dispersive contributions the very close similarity of the chiral and the achiral transient measurements (and hence the potential danger of misinterpretation) becomes even more evident. Essentially, chiral and the achiral pump-probe experiment differ only in the polarization direction of the excitation pulse with respect to the probe fields. According to non-linear response theory, the transient signal of an isotropic solution is proportional to the orientational average WW XY , where Y is the polarization direction of the LO, which is transmitted by the analyzer, X is the polarization of the signal-generating beam and W is the polarization of the visible excitation beam. The signal is purely chiral if and only if W ≡ X or W ≡ Y (or W ≡ Z). If W has both X and Y components, on the other hand, there are necessarily linear dichroism and linear birefringence contributions. They are strongest when W = X +Y , i.e. when the polarization of the pump pulse is rotated by an angle θ = 45 • with respect to both E 0 and E LO (we neglect the small Z-component introduced by the different propagation directions). The orientational average determining the signal in this case is: The first line in this equation gives rise to the very small, electric dipole-forbidden transient chiral signal, while the second line contains the usually much larger dipole-allowed linear dichroism and linear birefringence contributions (note the identity XY XY + Y XXY = YYYY − XXYY ) [28]. The dominant contributions to the transient signals ∆S out of phase and ∆S in phase for the two limiting cases of pump-beam polarization are summarized in Table 1.
As data acquisition and processing for transient chiral and achiral signals are identical we decided to test the performance of our setup by recording pump-induced LD and LB of the achiral metal carbonyl complex [Re(py)(CO) 3 bipy] in acetonitrile, setting the 400 nm pump pulse polarization to 45 • . Excitation at this wavelength causes an ultrafast metal to ligand electron transfer, which is stabilized in a long-lived triplet state, characterized by a blue-shift of the CO vibrations [29][30][31]. Anisotropy and hence linear dichroism and birefringence are induced by the preferential excitation of molecules whose electronic transition dipole moment is aligned with the pump polarization. The transient signals ∆S in phase (LD) and ∆S out of phase (LB) of the CO stretch vibrations are shown in Figs. 5(b) and 5(c). For comparison, Fig. 5(a) shows the transient LD signal measured in an already established way, by blocking one interferometer arm and changing the orientation of the analyzer between ±β . For β = 45 • this is the conventional LD signal (∆A − ∆A ⊥ ), which is enhanced by a factor cot β for β < 45 • [25,32,33]. The signal decay on a 20 ps timescale is due to rotational diffusion of the complex, which renders the sample isotropic again [29,34].
The good agreement between Figs. 5(a) and 5(b) shows that spectral interferometry can be efficiently used for highly sensitive transient measurements in a crossed-polarizer configuration. The recording and processing of transient LB and LD signals is completely analogous to that of transient VCD and VORD signals, for which the visible pump beam polarization must be set exactly to X. This means that possible LD and LB artifacts on top of a transient chiral signal can be identified in a complementary measurement under almost identical experimental conditions. For example, both transient VCD and LB are out of phase signals and formally correspond to a π/2 phase shift between the LO and the signal-generating beam. They may both resemble the derivative of the transient absorption signal and can thus be easily confounded.
In our achiral test sample, the transient LD and LB signals should completely vanish when the 400 nm polarization is aligned perpendicular to that of the LO. This seems to be the case in Fig. 5(d), showing ∆S out of phase measured with a X-polarized pump beam, where we can no longer identify any signal that decays on the timescale of rotational diffusion. However, we observe a residual signal of the order of 10 −4 OD, approximately 3-4% of the LB amplitude at 45 • pump orientation, which lasts for more than 100 picoseconds and is thus not related to pump-induced anisotropy. We found that this residual signal had a different shape and a larger amplitude when the interferometer delay was changed from 150 fs to 500 fs. This could point to an artifact due to a pump-induced phase change between LO and the signal-generating beam, which we have tried to minimize in our set-up by co-propagating both through the same spot on the sample. In order to test this hypothesis, we have repeated the complete pump-probe scan with a tilted analyzer, recording separateS α for each pump-probe delay and chopper state. However, we found no difference between using these individual reference signals, which should be sensitive to pump-induced phase shifts, or a meanS α for calculating the transient response with the help of Eq. (14). Hence, the origin of the residual signal is not yet fully understood. Such signals must, however, be considered as a possible artifact in any future transient VCD experiment, in addition to pump-induced LB and LD, which can be identified more easily because of rotational diffusion. Reproducibility of a chiral signal for a number of different LO delays may be one possible test criterion, in addition to control measurements with suitable enantiomers.

Conclusions
In conclusion, we have presented an actively phase stabilized interferometer for the generation of orthogonally polarized pairs of IR femtosecond laser pulses for static and transient dichroism and birefringence measurements with multichannel detection. We were guided by the idea to combine the advantage of phase stability of a single beam measurement with the highest possible purity of polarization and its symmetry under fast modulation. By carefully orienting lenses and the sample flow-cell, we managed to maintain these conditions under the tight focusing and sample flow conditions required for time-resolved VCD experiments. We showed that the set-up is suitable for simultaneous transient linear dichroism and birefringence measurements by spectral interferometry with the same signal amplification achieved for acquiring VCD and VORD data. Pump-induced LB is expected to be the dominant background contribution in tran-sient VCD experiments, while LB will mainly perturb transient VORD measurements. Simple rotation of the pump-beam polarization allows us to record and distinguish these signals under quasi identical experimental conditions. As a result of signal enhancement, broad band detection and referencing, transient VCD and VORD experiments with our set-up are no longer limited by signal-to-noise, which could be improved even further with already available higher laser powers and repetition rates. At wavelengths shorter than 5 µm another boost is also possible with more sensitive detector materials such as indium antimonide, as this would allow us to reduce the local oscillator intensity. However, the most important remaining difficulty for transient chiral measurements is the presence of an achiral background signal, which cannot be attributed to either transient LB or LD. We currently investigate the stability and reproducibility of this artifact, which is sensitive to beam alignment, hoping to be able to subtract it from transient data recorded for enantiomers with opposite chiral response.

B. PEM-induced focal spot-size modulations
In the ABCD matrix description, light rays are characterized by their distance from the optical axis r and the angle of inclination θ , Gaussian beams by the radius of curvature R of the phase front and the 1/e 2 -beam radius w 0 . Optical elements are described by 2x2 matrices which act on ray vectors v = (r, θ ) or change the Gaussian beam parameters according to corresponding to and increase of approximately 0.5% at both ±λ /4 retardation in our case. In practice, however, it can be difficult to place the sample exactly at the focal spot of the lens. As illustrated in Fig. 6(c), already 200µm away from the average focus the beam radius can be modulated by more than 5%. b) Modulator in converging beam In CD spectrometers the PEM is often placed in between the collimating optics and the sample at the focus, as shown in Fig. 6(b). Moving the PEM close to the focus can significantly reduce the beam size modulation (Fig. 6(d)), essentially because the numerical aperture of the PEM-lensing is reduced. The optimal position for the PEM in this regard is, of course, at the focus of the collimating lens, or, in the focal plane of a telescope as in the set-up presented in this work.
In linear CD spectroscopy focal spot size modulation may cause offsets due to a different imaging of left and right handed light onto the detector. The effect is more serious when an array detector is used, because different illumination of its pixels can result in apparent spectral shifts or broadening, giving rise to large artifacts near sharp spectral features like absorption bands. This is how we first became aware of the problem in our lab. Finally, in a pump-probe experiment, a small difference in the number of excited molecules can lead to undesired timedependent signals, which may well be of the same order as the chiral response.