December 15, 2004 / Vol. 29, No. 24 / OPTICS LETTERS
Coherent anti-Stokes Raman scattering spectral interferometry:
determination of the real and imaginary components
of nonlinear susceptibility x(3)for vibrational microscopy
Conor L. Evans, Eric O. Potma, and X. Sunney Xie
Department of Chemistry and Chemical Biology, Harvard University, 12 Oxford Street, Cambridge, Massachusetts 02138
Received June 10, 2004
We demonstrate coherent anti-Stokes Raman scattering (CARS) heterodyne spectral interferometry for re-
trieval of the real and imaginary components of the third-order nonlinear susceptibility (x?3?) of molecular
vibrations. Extraction of the imaginary component of x?3?allows a straightforward reconstruction of the vi-
brationally resonant signal that is completely free of the electronic nonresonant background and resembles the
spontaneous Raman spectrum.Heterodyne detection offers potential for signal amplification and enhanced
sensitivity for CARS microscopy.© 2004 Optical Society of America
110.0180, 180.3170, 300.6230.
Since its revival in 1999, coherent anti-Stokes Raman
scattering (CARS) microscopy has proved to be a pow-
erful method for sensitive vibrational imaging of bio-
The vibrational contrast in CARS
imaging results from the Raman activityof endogenous
molecules, which circumvents the need for fluorescent
labels.CARS microscopy has been successfully ap-
plied to live cell imaging1and intracellular organelle
tracking2and holds great promise for noninvasive
tissue imaging.However, the sensitivity of CARS mi-
croscopy is limited by a nonresonant background that
is independent of the Raman shift.
the third-order nonlinear susceptibility x?3?, which is
the sum of a vibrationally resonant part x?3?
nonresonant electronic contribution x?3?
generates a nonresonant field that not only introduces
an offset that can be overwhelming at times but also
distorts the resonant spectral profile through interfer-
ence with the resonant field.
to assign CARS spectra based on the wealth of spectral
assignments in the Raman literature.
sensitive detection3and time-resolved CARS4have
been developed to suppress the nonresonant back-
attenuation.To improve the sensitivity and spectral
selectivity of CARS microscopy, an approach is needed
that suppresses the nonresonant background while al-
lowing complete extraction of the resonant vibrational
One solution to this problem lies in taking advan-
tage of a fundamental difference between the resonant
real and imaginary parts.
that the spectrum of the imaginary part of x?3?
Isolating the imaginary component thus
gives only resonant information, free of the nonreso-
nant background.Conventional CARS techniques
measure jx?3?j2, which entangles the real and imagi-
nary components.A direct means of extracting the
imaginary part is heterodyne mixing with a local
CARS arises from
This makes it difficult
nrhas only a real component, x?3?
It was well established
res?, is proportional to the spontaneous Raman
ber of attractive qualities.
comparison with Raman spectral profiles without the
complication of a nonresonant background.
the heterodyne signal is linearly proportional to the
concentration, which permits a straightforward quan-
titative interpretation of images.
mixing provides the possibility of interferometric
CARS signal enhancement by mixing with a strong
local oscillator field.
The potential of CARS interferometry was previ-
ously demonstrated by suppressing the nonresonant
background in the gas phase with narrowband laser
In this Letter we resolve the real and
imaginary parts of x?3?of vibrational bands in the
condensed phase with broadband CARS spectral inter-
ferometry.Although this information is in principle
accessible through time-domain CARS interferome-
extracting accurate phase information from
time-resolved traces is difficult in practice.
interferometry has many advantages over temporal
interferometry, including better signal-to-noise char-
acteristics8and insensitivity to temporal pulse jitter.
In addition, acquisition times can be much shorter
because spectral interferometry requires no scanning
and the amplitude and phase can be extracted through
In our spectral interferometry method, a broadband
CARS field Es?v? from the sample arm interferes with
a broadband CARS local oscillator field Elogenerated
from a nonresonant sample in the local oscillator arm.
The interference appears as a sinusoidal modulation
on the detected spectral intensity:
Heterodyne detection offers a num-
First, it allows a direct
SCARS?v? ? jEloj21 jEs?v?j21 2jEloEs?v?jcos F?v?,
where F?v? ? vt 1 fs?v? 1 finst?v? is the total phase
difference between the arms.
delay between the two arms of the interferometer,
Here t is the temporal
0146-9592/04/242923-03$15.00/0© 2004 Optical Society of America
OPTICS LETTERS / Vol. 29, No. 24 / December 15, 2004
fs?v? is the phase difference introduced by the
sample, and finst?v? is the relative phase delay due to
optical components in the interferometer.
information is encoded in the spectral fringe spacing
and can be extracted from the spectral interfero-
Once fs?v? is known, the real and imaginary
components of x?3??v? can be readily determined.
The experimental setup (Fig. 1) is based on two
follows an excitation scheme similar to that in multi-
plex CARS spectroscopy.10
Raman shift was tuned to the CH-stretch vibrational
region of dodecane, with the picosecond pump laser
set at 712.5 nm and the femtosecond Stokes laser cen-
tered at 897 nm (FWHM 140 cm21).
arm of the Mach–Zehnder interferometer, the pump
and Stokes beams were focused onto a glass–dodecane
interface by use of a high-numerical-aperture lens.
An identical lens was used in the local oscillator
arm to generate the local oscillator field from a glass
coverslip–air interface.Both the CARS signal and
the local oscillator fields were collected in the epi di-
rection,11separated from the pump and Stokes beams
by dichroic mirrors, and combined on a cubic beam
with a spectrometer equipped with a CCD.
integration times of 5 s were used with 30 mW of
average power for each beam at the sample.
Spectrally resolved phase information was extracted
from the interferograms with a method developed
by Lepetit et al.9
Briefly, the CARS beam from the
sample was passed through a quarter-wave plate
and converted into a circular polarized beam.
combination with the local oscillator CARS beam at
the cubic beam splitter, a polarizing beam splitter set
at 45±to the local oscillator beam polarization split
the mixed CARS beam into two orthogonally polarized
components. These two linearly polarized beams
were then spectrally dispersed, and the two separate
interferograms were simultaneously detected by the
CCD. After the homodyne terms were independently
determined, the heterodyne parts were isolated.
Because there is a p?2 phase shift between the two
heterodyne contributions, dividing one by the other
yields the tangent of total spectral phase F?v?.
To isolate fs?v? from the total phase of the sample,
a reference measurement is needed that provides the
spectral reference phase Fref?v? ? vt 1 finst?v?.
Since fs?v? arises only from a resonance, a non-
resonant reference sample should be used.
examined dodecane in the CH-stretch vibrational
region, we used fully deuterated dodecane as the
reference sample to ensure that the material prop-
erties remained nearly identical between the two
measurements. A flow cell was used to keep the
delay between the two arms identical when switching
between the resonant and reference samples.
traction of Fref?v? from F?v? yields fs?v?.
length difference between the arms was found to be
stable during the collection time window.
Figure 2 shows two spectral interferograms ob-
tained from both polarization components that were
collected from a nonresonant sample.
For this experiment the
In the sample
The fringes are
the result of a 1-ps path-length difference between the
arms.This offset results in the linear spectral phase
(vt) that spans the spectral window.
of this spectral interferogram required no scanning,
was shot noise limited, and was completely insensitive
to temporal laser fluctuations.
When the nonresonant sample is replaced by a reso-
nant sample, it is straightforward to ascertain the reso-
nant spectral phase profile.
amplitude and phase of the dodecane CH-stretch
region. The phase shows several maxima.
was a single spectral resonance, a single phase swing
from zero to p centered on that peak would be ex-
pected.In this case, however, there is a cluster of
spectral resonances that interfere with one another,
giving rise to the complex phase profile.
Figure 3(a) shows the
The interferometer consists of a sample arm and a local
oscillator (LO) arm.The length of the sample arm is
adjustable for phase delay control.
dichroic mirror for combining pump and Stokes beams;
DM2, DM3, shortwave-pass dichroic mirrors; SF, spectral
filter (Schott RG695); CB, cubic 50% beam splitter;
SM, stepper motor; PZT, piezoelectric transducer; QWP,
quarter-wave plate; BPF, bandpass filter (600 nm, 40-nm
bandwidth); PB, polarizing beam splitter; Spec, grating
spectrometer; CCD, liquid-nitrogen-cooled charge-coupled
device. Two identical oil immersion lenses are used
(Nikon Plan Apo, 603, N.A. of 1.4).
diagram for picosecond pump and femtosecond Stokes
Schematic of the CARS spectral interferometer.
Inset, CARS energy
ously collected orthogonal polarizations from a nonresonant
sample (glass coverslip) with the arms offset by 1 ps (solid
and dotted curves).The extracted Fref?v? is indicated by
the dashed curve.
CARS spectral interferograms of two simultane-
December 15, 2004 / Vol. 29, No. 24 / OPTICS LETTERS Download full-text
tracted CARS amplitude (solid curve) and phase (dotted
curve) from the spectral interferograms.
spectrum was normalized by the nonresonant CARS
spectrum of the local oscillator arm.
real ?jEs?v?jcos?fs?v?? 2 jEnr?v?j, dotted
imaginary ?jEs?v?jsin?fs?v??, solid curve? parts of the
CARS spectrum. Inset, Raman spectrum of dodecane
in the CH-stretch vibrational range.
measured with a Raman microspectrometer with a 1-min
CARS spectral amplitude and phase of the
vibrationalband of dodecane.(a) Ex-
The spectrum was
The real and imaginary components of x?3?can
readily be calculated from the amplitude and phase
[Fig. 3(b)].The imaginary x?3?spectrum is in good
agreement with the corresponding Raman spectrum,
shown in the inset of Fig. 3(b).
spectrum is purely resonant and free of nonresonant
background contributions, with all the features of the
Raman spectrum reproduced.12
that heterodyne CARS spectral interferometry pro-
vides a means of rapid Raman microspectroscopy,
particularly in regions of the vibrational spectrum,
such as the fingerprint region, that are too congested
to access with conventional CARS spectroscopy.
In conclusion, we have shown that heterodyne CARS
interferometry can be used to extract amplitude and
phase information from x?3?molecular vibrational reso-
nances, leading to complete suppression of the non-
resonant background. Unlike
signals that are complicated by mixing with nonreso-
nant components, the imaginary part of x?3?can be
directly equated to Raman spectra.
the isolated signal comes from the heterodyne term,
it is linear with concentration and lends itself imme-
diately to quantitative analysis.
amplification through heterodyne detection offers
promise for further increasing the sensitivity of CARS
microscopy.As we have shown, CARS heterodyne
interferometry can be readily performed in the epi
direction, which offers prospects for high-sensitivity
CARS tissue imaging.
In addition, since
We thank Brian English for his technical help with
the flow cells.This research was funded by the Na-
tional Science Foundation and the National Institutes
of Health.C. L. Evans acknowledges the National
Science Foundation for a Graduate Research Fellow-
ship.X. S. Xie’s e-mail address is xie@chemistry.
1. J.-X. Cheng and X. S. Xie, J. Phys. Chem. B 108, 827
2. X. L. Nan, J.-X. Cheng, and X. S. Xie, J. Lipid. Res.
44, 2202 (2003).
3. J.-X. Cheng, L. D. Book, and X. S. Xie, Opt. Lett. 26,
4. A. Volkmer, L. D. Book, and X. S. Xie, Appl. Phys. Lett.
80, 1505 (2002).
5. R. W. Hellwarth, Prog. Quantum Electron. 5, 1 (1977).
6. Y. Yacoby and R. Fitzgibbon, J. Appl. Phys. 51, 3072
7. C. Vinegoni, J. S. Bredfeldt, D. L. Marks, and S. A.
Boppart, Opt. Express 12, 331 (2004), http://www.
8. J. F. de Boer, B. Cense, H. Park, M. C. Pierce, G. J.
Tearney, and B. E. Bouma, Opt. Lett. 28, 2067 (2003).
9. L. Lepetit, G. Cheriaux, and M. Joffre, J. Opt. Soc. Am.
B 12, 2467 (1995).
10. J.-X. Cheng, A. Volkmer, L. D. Book, and X. S. Xie,
J. Phys. Chem. B 106, 8493 (2002).
11. A. Volkmer, J.-X. Cheng, and X. S. Xie, Phys. Rev. Lett.
87, 023901 (2001).
12. The subtle differences between the Raman and the
Im?x?3?? spectra might be attributed to Raman depo-