13645
2007, 111, 13645-13647 Published on Web 08/28/2007
Nonresonant Background Suppression in Broadband Vibrational Sum-Frequency Generation Spectroscopy Alexei Lagutchev, Selezion A. Hambir, and Dana D. Dlott* School of Chemical Sciences, UniVersity of Illinois at Urbana-Champaign, Chemical and Life Sciences Laboratory, 600 South Mathews AVenue, Urbana, Illinois 61801 ReceiVed: July 10, 2007; In Final Form: August 14, 2007
Suppression of the nonresonant background in vibrational sum-frequency generation (SFG) in the broadband multiplex configuration is achieved using a time-asymmetric pulse, created by passing a femtosecond pulse through a Fabry-Perot e´talon, to temporally discriminate between the faster nonresonant and slower resonant contributions. A mixed time and frequency domain explanation of the SFG process is presented, and spectra with high time resolutions and high degrees of nonresonant background suppression are obtained using selfassembled alkanethiolate monolayers on Au.
In this letter, we describe an improved technique in vibrational sum-frequency generation spectroscopy1 (SFG) that greatly suppresses the nonresonant background. The technique uses a pulse with a time-asymmetric profile to discriminate between nonresonant (NR) and resonant (R) contributions. SFG with one resonant vibrational infrared (IR) pulse and one nonresonant pulse (usually visible or near-IR but historically called “visible”) is growing in popularity because it helps overcome the two biggest issues associated with interface molecular vibrational spectroscopy: sensitivity and selectivity.1 Sensitivity results from the nonlinear coherent process1 employing2 intense 1012 W cm-2 femtosecond laser pulses. Selectivity results from an important property of the second-order susceptibility, χ(2). In the dipole approximation, χ(2) vanishes in centrosymmetric media,1 so in such media SFG signals are selectively generated from surfaces or interfaces.1 One of the biggest problems with SFG, in common with other nonlinear coherent vibrational spectroscopies,3,4 is a NR background that interferes with the detection of molecular vibrational resonances. The NR background problem is especially severe with adsorbates on metal surfaces5 but can be a problem even with dielectrics. The intensity of the SFG signal, ISFG(ω), is given by1,6
ISFG(ω) ∝ [PSFG(ω)]2 ) {[χR(2)(ω) + χ(2)NR(ω)]EIREvis}2 (1) where PSFG ) PR + PNR is the polarization at the sum frequency. The SFG signal results from the square of the sum of complex quantities χ(2)R and χ(2)NR. When |χ(2)NR| > |χ(2)R|, which is frequently the case on metal surfaces, fluctuations in the NR background signal are usually the dominant source of noise. The phase difference between χR and χNR is not known in advance. This phase depends on the nature of the sample and the visible wavelength,7 so the phase is an additional fitting parameter in SFG spectrum analysis.8 In obtaining the number * Corresponding author. E-mail:
[email protected].
10.1021/jp075391j CCC: $37.00
density, N, or the dependence of N on an external field on a surface or interface, χNR can be a significant obstacle. Depending on the relative magnitudes of χR and χNR, ISFG might be proportional to N raised to any power between 1 and 2. However when χNR ) 0, ISFG ∝ N2. For these reasons, the SFG nonresonant background is a hindrance to SFG spectroscopists who wish to eliminate or at least suppress it. Our method is based on the femtosecond broadband multiplex SFG technique (BBSFG).2 A femtosecond laser is used to generate a narrowband picosecond visible pulse by spectral filtering and a broadband IR pulse by pumping an optical parametric amplifier (OPA). The IR and visible pulses are incident on the sample, and the coherent SFG output is detected with a spectrograph and multichannel detector. The spectral envelope of the IR pulse, typically 200 cm-1 fwhm, defines the vibrational wavenumber region probed. The spectral width of the visible pulse, typically 5-15 cm-1, determines the spectroscopic resolution. BBSFG generates an entire SFG spectrum over a range of wavenumbers on a single shot.2 It is useful to think of BBSFG in a mixed time and frequency domain picture6,9,10 in three steps. The sample, taken to consist of an ensemble of two-level systems, is irradiated by an IR pulse, creating a time-dependent IR polarization PIR(t) ) PIRR(t) + PIRNR(t). PIR(t) undergoes a coherent Raman interaction with the visible pulse to create PSFG(t). The coherent pulse radiated by PSFG(t) is Fourier transformed to ISFG(ω) by the spectrograph and square-law detector. Only when both PIR(t) and the visible pulse are simultaneously present can the SFG signal be emitted. Anharmonic effects are not considered in this picture. Although anharmonically red-shifted excited-state transitions are prominent in many IR experiments,4,11 such as IR pump-probe or 2D-IR, those signals depend on interactions with at least two IR pulses, whereas SFG signal generation involves an interaction with just one IR photon. If the IR intensity were too high, then there would be a possibility of multiple excitations on the same molecule that might be coupled by anharmonic interactions, © 2007 American Chemical Society
13646 J. Phys. Chem. C, Vol. 111, No. 37, 2007
Figure 1. (a) Suppression of SFG nonresonant (NR) background with a time-asymmetric visible pulse. A 100-fs IR pulse creates a polarization PIR(t) ) PIRR(t) + PIRNR(t). The NR part, PIRNR(t), has a faster time dependence, which tracks the IR pulse. The resonant part, PIRR(t), due to molecular vibrational transitions, has a slower time dependence. SFG is created during the time that PIR(t) interacts with the time-asymmetric picosecond visible pulse. For clarity, optical-frequency oscillations in the visible pulses are not displayed. Two delay situations involving the time-asymmetric pulse are depicted. Either the visible pulse interacts with both PIRR(t) + PIRNR(t) to generate resonant and NR signals or it is time-delayed beyond PIRNR(t) to suppress the NR contribution, weakening the resonant contributions only slightly. (b) Block diagram of the laser apparatus. Part of the femtosecond pulse is used to generate a broadband infrared (BBIR) pulse in an optical parametric amplifier (OPA), and part is used to generate a time-asymmetric picosecond visible pulse in an e´talon.
leading to an intensity-dependent anharmonic shift of SFG vibrational resonances, but surface damage occurs before this is observed. We now consider SFG of a well-known model system, a selfassembled monolayer (SAM) of octadecylthiolate (ODT) on Au(111).2,5,12 The SFG spectrum of Au-S-(CH2)17-CH3 in the CH-stretch region near 2950 cm-1 has three prominent transitions with fwhm ∼15 cm-1 due to the terminal methyl groups at the SAM/air interface. The methylene transitions are much weaker because the interaction between PIRR(t) and the visible pulses for methylene groups in a highly ordered all-trans SAM structure is forbidden in the dipole approximation.13 Figure 1a depicts the concept of the NR suppression technique. The IR pulse at frequency ω0 is taken to have a Gaussian time profile Gg(t, tp) with fwhm tp ) 100 fs. PIRNR(t) results primarily from Au surface electronic states. Owing to the fast electronic response, the undesirable PIRNR(t) has essentially the same time dependence as the IR pulse. The desired PIRR(t) results from molecular vibrational transitions. PIRR(t) is simply the well-known vibrational free-induction decay (FID)14,15 generated by a resonant femtosecond IR pulse. The FID in Figure 1a was computed using three homogeneously broadened Lorentzian lines having 15 cm-1 fwhm. For Lorentzian lines, fwhm ) (πT2)-1, where T2 is the vibrational dephasing time constant, so T2 ) 0.7 ps. The transition dipole moments, µi, and transition frequencies, ωi, were extracted by fitting the SFG spectra in Figure 3a. The actual FID has contributions from all 37 ODT CH-stretch transitions, but we did not display the methylene contributions that do not interact with both IR and visible pulses to produce SFG. As shown in Figure 1a, the PIRNR(t) contribution to SFG can be suppressed by mixing PIR(t) with an appropriately timedelayed time-asymmetric visible pulse. The time-asymmetric visible pulses are created by spectral filtering of time-symmetric
Letters
Figure 2. Computed and measured femtosecond pulse transmission through an e´talon with gap d. (a) With an input pulse, Pin, with coherence length lc < d, the output, Pout, is a train of phase-locked pulses. In this computation, lc ) 33.2 µm and d ) 67.2 µm. (b) When lc > d, the output is a picosecond pulse that is approximately a singlesided exponential with a nearly Lorentzian spectrum. In this computation, lc ) 33 µm and d ) 11 µm. (c) Measured spectra of femtosecond input pulses and e´talon-filtered picosecond pulses.
femtosecond pulses using an air-spaced Fabry-Perot e´talon16 (vide infra). BBSFG spectra are usually obtained by setting the time delay between IR and visible for maximum ISFG. With timesymmetric visible pulses, this maximum occurs close to the delay time when the pulse peaks are temporally coincident. With a time-asymmetric pulse, the maximum occurs when the visible pulse slightly precedes the IR pulse, as shown in Figure 1a. However, if this visible pulse is delayed relative to IR, by a bit more than one tp, then PIRNR(t) creates little or no NR signal, whereas the PIRR(t) signal is attenuated only slightly, by a factor of ∼(1 - tp/T2). In actual experimental conditions, tp , T2, so the NR signal is much-more-strongly suppressed than the R signal. The time-windowing concept of NR background suppression is an old one, dating back at least to the coherent Raman work17 of W. Kaiser and colleagues in the 1970s. Although it is also possible to suppress NR backgrounds in Fourier-transform9,10 SFG by mathematical signal-processing methods involving apodization of the interferogram near zero delay, all things being equal this Fourier-transform technique cannot achieve the signalto-noise ratios10 obtained here. The key innovation in the present work is the use of Fabry-Perot e´talon filtering to produce a time-asymmetric picosecond visible pulse, as opposed to morecommon methods using grating spectrographs2 or interference filters18 to produce time-symmetric pulses. Time-windowing with a time-symmetric picosecond pulse18 does not achieve a high level of discrimination between PIRR(t) and PIRNR(t). The theory of pulse transmission through an ideal e´talon is well known,19 so only a brief discussion is presented here. The e´talon consists of two parallel partially reflecting mirrors with reflectivity R separated by air gap d. The transmission spectrum of this e´talon is a series of peaks spaced by the free-spectral range FSR ) c/2d. The peak width δν ) FSR/F where the finesse F ) π/(1 - R). In our experiments, an 800-nm femtosecond pulse was incident on such an e´talon tilted slightly to move its transmission peak to the pulse’s central wavelength. The pulse coherence length lc ) tpc is ∼30 µm. The pulse spectral width is greater than δν but less than the FSR, so the filtered pulse spectrum consists of a single peak of width δν (see Figure 2c). In the
Letters
J. Phys. Chem. C, Vol. 111, No. 37, 2007 13647
Figure 3. (a) SFG spectra of octadecylthiolate SAM on polycrystalline Au(111) in the CH-stretch region using ppp polarization, obtained at a 1-kHz repetition rate with 1-s integration time. At t ) 0, the NR background whose shape mimics the IR pulse spectrum is a maximum. The three CH-stretch transitions of the terminal methyl groups are observed as 15 cm-1 fwhm dips against the NR background. Increasing time delay suppresses the NR signal. (b and c) The dashed curves are SFG spectra of an octadecylthiolate SAM on Au. The solid curves are transient SFG spectra, 20 ps after the Au surface was flash-heated to 800 °C as in ref 20. Part c is the same as b except NR suppression is used.
time domain, the e´talon output is a train of pulses that decays exponentially with a cavity lifetime τc ) R2d/c ) (πδν)-1. Figure 2a illustrates such a pulse train in the familiar case of a large gap, when lc < d. Our apparatus functions in the perhaps less-familiar small-gap limit where lc > d. In this case, the pulses overlap significantly. The output pulse becomes smooth, without any high-frequency modulation due to interference between pulses because the pulses in the train are phase-locked by the e´talon. An intensity profile of the picosecond visible pulse computed using 100-fs, 800-nm input pulses, and the parameters of our e´talon is shown in Figure 2b. The transmitted pulse intensity, Itr (t), can be approximately represented as
Itr(t) ) (1 - R)2 Gg(t,tp)* exp(-τc t)
(2)
which is the convolution of the input pulse with a single-sided exponential. The spectrum of the time-asymmetric pulse in the frequency domain shown in Figure 2c, which is the Fourier transform of eq 2, has a line shape that is quite close to Lorentzian. The apparatus used for BBSFG is diagrammed in Figure 1b. A femtosecond fiber laser is amplified at 1 kHz in Ti:sapphire (Quantronix Integra-C 2.0) to produce 2.0-mJ pulses of 120-fs duration at 800 nm. Three-quarters of each pulse pumps an IR OPA (Light Conversion TOPAS-C 800 fs DFG). One-fourth of each pulse is filtered by an e´talon (TecOptics) with air gap d ) 11.1 µm, FSR ) 450 cm-1, F ) 44, to produce a timeasymmetric picosecond pulse with ∆ν ) 10.3 cm-1. The spectra of the input and filtered pulses are shown in Figure 2c. At the sample, the IR pulse energy is 10 µJ, the (1/e2) beam diameter is 200 µm, the visible pulse energy is 20 µJ, and the beam diameter is 300 µm. SFG spectra of ODT/Au using ppp polarization and a 1-s integration time are shown in Figure 3a, as a function of time delay between the IR pulses and the time-asymmetric visible pulses. Time t ) 0 denotes the NR intensity maximum. This well-known spectrum2 consists of a broad feature spanning several hundred cm-1, which represents the IR OPA output pulse spectrum. The three methyl CH-stretch transitions appear as dips
against this NR background because the χNR(ω) and χR(ω) terms are approximately 180° out of phase. As time delay is increased, the NR part decreases. While the NR background decreases, the resonant dips become peaks. The optimum condition for background suppression occurs with our apparatus near t ) 200 fs, where the NR background is suppressed by a factor that is difficult to determine but is certainly >20, while the suppression of the resonant signal is minimal. The use of the NR suppression method to study the vibrational-energy dynamics of molecules on surfaces is illustrated in Figure 3b and c. Recently, our group has studied the flow of heat through long-chain alkanethiolate SAMs initiated by ultrafast flash-heating of the Au substrate to 1100 K.20 When heat from the Au layer traveled along the alkane chains to the terminal methyl groups, the resulting thermal disorder caused the resonant CH-stretch signals to decrease.20 As illustrated in Figure 3b, our measurements of the SFG signal decrease were hindered by the NR background. The NR background made it difficult to determine the extent to which flash-heating reduced the amplitudes of the dips in the spectrum. Fluctuations in the NR background were the dominant noise source. With NR suppression (Figure 3c), determining the amplitude change caused by flash heating is much-more accurate and the signal-to-noise ratio was increased by at least a factor of 20. In conclusion, we have demonstrated a useful method for NR background suppression in BBSFG spectroscopy. Given a femtosecond BBSFG apparatus, it is a simple modification to use an e´talon as the frequency-narrowing element to generate a time-asymmetric visible pulse and to adjust the time delay to optimize NR suppression. Acknowledgment. This material is based upon work supported by the National Science Foundation under award DMR 0504038, the Air Force Office of Scientific Research under award FA9550-06-1-0235, and the U.S. Army Research Office under awards W911NF-04-1-0178, W911NF-05-1-0345, and W911NF-06-1-0171. References and Notes (1) Shen, Y. R. Nature 1989, 337, 519. (2) Richter, L. J.; Petralli-Mallow, T. P.; Stephenson, J. C. Opt. Lett. 1998, 23, 1594. (3) Eesley, G. L. Coherent Raman Spectroscopy; Pergamon: Oxford, 1991. (4) Mukamel, S. Principles of Nonlinear Optical Spectroscopy; Oxford University Press: New York, 1995. (5) Harris, A. L.; Chidsey, C. E. D.; Levinos, N. J.; Loiacono, D. N. Chem. Phys. Lett. 1987, 141, 350. (6) Roke, R.; Kleyn, A. W.; Bonn, M. Chem. Phys. Lett. 2003, 370, 227. (7) Ishibashi, T.; Onishi, H. Chem. Lett. 2004, 33, 1404. (8) Lu, G. Q.; Lagutchev, A.; Dlott, D. D.; Wieckowski, A. Surf. Sci. 2005, 585, 3. (9) McGuire, J. A.; Beck, W.; Wei, X.; Shen, Y. R. Opt. Lett. 1999, 24, 1877. (10) McGuire, J. A.; Shen, Y. R. J. Opt. Soc. Am. B 2006, 23, 363. (11) Wynne, K.; Hochstrasser, R. M. Chem. Phys. 1995, 193, 211. (12) Bain, C. D.; Davies, P. B.; Ong, T. H.; Ward, R. N. Langmuir 1991, 7, 1563. (13) Bain, C. D. J. Chem. Soc., Faraday Trans. 1995, 91, 1281. (14) Brewer, R. G.; Shoemaker, R. L. Phys. ReV. Lett. 1971, 27, 631. (15) Guyot-Sionnest, P. Phys. ReV. Lett. 1991, 66, 1489. (16) Patterson, J. E.; Lagutchev, A. S.; Huang, W.; Dlott, D. D. Phys. ReV. Lett. 2005, 94, 015501. (17) Laubereau, A.; Kaiser, W. ReV. Mod. Phys. 1978, 50, 607. (18) Voges, A. B.; Al-Abadleh, H. A.; Musorrafiti, M. J.; Bertin, P. A.; Nguyen, S. T.; Geiger, F. M. J. Phys. Chem. B 2004, 108, 18675. (19) Born, M.; Wolf, E. Principles of Optics; Cambridge University Press: Cambridge, 1998. (20) Wang, Z.; Carter, J. A.; Lagutchev, A.; Koh, Y. K.; Seong, N.-H.; Cahill, D. G.; Dlott, D. D. Science 2007, 317, 787.
