Physicists Bend Raman Light Backwards, Unlocking Same-Side Chemical Imaging

For decades, one of the quiet frustrations of optical microscopy has been a matter of simple geometry. When laser light scatters inside a clear, homogeneous material, the overwhelming majority of the scattered photons travel forward, in the same general direction as the illuminating beam. That physical bias has long forced scientists who want to read the chemical fingerprints of a sample to place a detector on the far side of the specimen. For thick tissues, samples mounted on opaque substrates, or instruments built around optical fibres and spectrometers, that requirement is often impractical or outright impossible. A team at the Institut Fresnel of Aix Marseille University now reports in Nature Photonics that a nonlinear optical trick can break this directional rule, generating coherent Raman photons that emerge directly from the back of a completely featureless medium.

The research, led by Jitendra Nath Acharyya and Sandro Heuke together with Gina Benni, Anne Sentenac and Hervé Rigneault, exploits coherent Stokes Raman scattering, a four-wave mixing process in which pump and Stokes laser fields combine to produce light at a longer, red-shifted wavelength that carries the vibrational signature of the molecules in focus. The key insight is that nonlinear interactions obey a more flexible form of momentum conservation than ordinary linear scattering. In linear optics, redirecting light around a corner requires structure, such as a grating or a refractive-index discontinuity, to supply the missing spatial frequency. In nonlinear wave mixing, the wavevectors of the driving fields themselves combine to produce the output wavevector, so the medium never needs to provide the momentum. The result is that coherent photons can, in effect, be turned around inside a perfectly uniform material.

Whether this redirection is efficient depends critically on the generated wavelength. When the output is longer than the driving wavelengths, as is the case for Stokes Raman signals, backward phase matching becomes genuinely efficient under tight, high-numerical-aperture focusing. The team’s momentum-space simulations show that under high-NA focusing each beam is best described as a broad angular spectrum of plane-wave components, and the χ(3) nonlinear process can combine many of these components in ways that steer the emission into the backward hemisphere. The researchers identified three representative wavevector geometries, which they describe as counterpropagating, copropagating and orthogonal configurations of the interacting Stokes components, each with distinct trade-offs between the steepness of the angles required and the strength of the resulting signal.

The simulations, performed under a deliberately conservative constraint that equalizes the volume-integrated two-photon excitation load across different pupil geometries, pinpointed an especially favourable combination: an azimuthally polarized Stokes beam confined to the outer annulus of the objective pupil, paired with a circularly polarized pump filling the central region. Under this geometry, the backward-emitted signal reaches roughly a quarter of the maximum forward-emitted coherent Stokes power, and fully azimuthal excitation can push the total backward power towards 35 percent, albeit at the cost of an annular refocused mode that is less convenient for imaging. The circular-pump, azimuthal-Stokes combination was therefore selected as the best compromise between backward power and a compact, single-lobed refocused spot.

On the experimental side, the team illuminated samples with a 1,030-nanometre Stokes beam shaped into an azimuthal ring and a pump beam near 790 nanometres, reducing the effective repetition rate to 0.71 megahertz with a pulse-picking scheme to raise the energy per pulse without increasing average power. Detection relied on an InGaAs avalanche photodiode whose effective dark noise was cut by roughly an order of magnitude through boxcar gating. Calibrated against detector specifications, the researchers estimate a directly backscattered coherent Stokes photon flux of approximately seven billion photons per second emitted from the focal volume in sunflower oil, and switching from purely linear excitation to the azimuthal Stokes geometry boosted the signal by about half, matching the polarization dependence predicted by simulation.

A crucial control experiment addressed the obvious sceptical question: could the detected signal simply be forward-generated light that bounced off some surface back into the objective? When the team placed a mirror directly above the sample, the detected photon flux did not measurably increase, supporting a genuinely intrinsic backward origin. Residual forward-generated signal is further suppressed by the small effective collection area of the avalanche detector, roughly three Airy units, which behaves like a confocal pinhole. This stands in sharp contrast to conventional epi-detected coherent anti-Stokes Raman scattering, or CARS, where backward signals in realistic thick samples arise mainly from high spatial frequencies at refractive-index interfaces or from forward-generated light that is multiply scattered on its return path. Such back-reflected CARS cannot be refocused onto a detector plane, which precludes efficient fibre coupling, spectral analysis and accurate characterization of the imaging point spread function.

Because the backward Stokes signal is generated directly at the focus and remains coherent and refocusable, it supports true confocal, same-side collection. The simulations show that the backward field refocuses through the objective and tube lens into a compact, near-Gaussian spot, opening the door to fibre-based and spectrometer-based implementations. The imaging performance is respectable: edge measurements at polymethyl methacrylate interfaces in oil yielded an effective lateral resolution of about 330 nanometres and an axial resolution of roughly 0.92 micrometres, values the authors note should be read as edge-derived effective imaging resolutions that include generation, backward propagation, collection and confocal detection rather than as a direct measure of the nonlinear source alone.

The biological demonstrations are where the approach reveals its practical appeal. Imaging avocado flesh at the lipid resonance of 2,850 wavenumbers, the backward Stokes channel remained robust despite the refractive-index mismatch between lipid droplets and surrounding tissue, and it showed none of the fluorescence background that contaminated the epi-CARS channel even behind a narrowband filter, since detection at around 1,470 nanometres sits well clear of tissue autofluorescence. In human fatty liver tissue, a clinically significant specimen given that hepatic lipid accumulation marks an early and potentially reversible stage of metabolic liver disease, lipid droplets and surrounding tissue were clearly distinguished, with contrast differences relative to epi-CARS consistent with the distinct spatial transfer functions of the two signals. In sheep brain tissue, the method resolved bright ellipsoidal myelin sheaths, the lipid-rich insulating structures whose damage underlies disorders such as multiple sclerosis and Guillain–Barré syndrome, a capability the authors highlight as valuable for demyelination research.

The honest accounting is that backward Stokes emission still delivers fewer photons than forward CARS. The simulations attribute an effective wavelength-scaling exponent of about 2.8 to the extended, micrometre-scale coherent source, yielding a forward CARS-to-Stokes photon-flux ratio of roughly five in the carbon-hydrogen stretch region, and the conservative excitation constraint brings the overall reduction in backward flux to about a factor of twenty relative to forward CARS. Yet detector physics partially rescues the balance: photomultiplier tubes used for visible CARS reach quantum efficiencies near 20 percent, while InGaAs avalanche photodiodes at 1,500 nanometres routinely exceed 80 percent, so the experimentally detected flux differs by only about a factor of five. Measured signal-to-noise ratios of 49 for backward Stokes against 171 for forward CARS at one microsecond integration suggest the current implementation remains detector-noise limited, and the authors point to faster gating, heterodyne detection and nonlinear frequency conversion of the signal into the visible as realistic routes to improvement, alongside objective designs better matched to long wavelengths and steep angles.

The team is careful to position the technique not as a replacement for forward-detected CARS or stimulated Raman scattering when transmission access is available, but as a new contrast mechanism for the many situations where it is not: samples on opaque substrates, thick specimens, tissue biopsies and fibre- or spectrometer-based platforms, where it also removes the need for a separate transmission-side descanned detection path. By turning a nominally dark-field configuration into one that yields bright, efficiently collectable chemical contrast from structurally featureless media, the work converts a fundamental constraint of linear optics into an engineering opportunity, and suggests that same-side, label-free Raman imaging may soon be considerably easier to build than the direction of scattered light once implied.

Subject of Research: Intrinsic backward emission of coherent Stokes Raman photons from homogeneous media for epi-detected coherent Raman microscopy

Article Title: Intrinsic backward emission of coherent Raman photons from homogeneous media

Article References: Acharyya, J. N., Benni, G., Sentenac, A., Rigneault, H., & Heuke, S. (2026). Intrinsic backward emission of coherent Raman photons from homogeneous media. Nature Photonics. https://doi.org/10.1038/s41566-026-01995-z

Image Credits: AI Generated

DOI: 10.1038/s41566-026-01995-z

Keywords: coherent Raman scattering, nonlinear optics, four-wave mixing, Raman microscopy, backward emission, phase matching, biophotonics, CARS, label-free imaging, myelin imaging, InGaAs detectors, epi-detection