For over a century, optical imaging has been governed by a fundamental physical barrier: the diffraction limit. Formulated by Ernst Abbe, this limit dictates that an optical system cannot resolve features smaller than approximately half the wavelength of the light used ($\sim \lambda / 2$). When light passes through a finite aperture, high-frequency spatial information decays exponentially, causing point sources to blur into overlapping patterns known as point spread functions (PSFs).
When two point sources are closer than this characteristic threshold, their individual blurs merge into a single indistinguishable blob. This phenomenon is known as the Rayleigh criterion. From an information-theoretic perspective, this creates the "Rayleigh curse": as the separation between two objects approaches zero, the statistical information available to distinguish them vanishes.
To see past this barrier, modern research shifts the problem from pure optics to a co-design of physics, mathematics, and quantum mechanics. By manipulating the sample, leveraging structural priors, or redesign of the measurement process, researchers can recover subwavelength details that were once considered permanently lost.
Physical and Structural Expansion
One intuitive way to bypass the diffraction limit is to alter the physical scale of the sample or the immediate optical environment before the light ever reaches the far-field lens.
If a microscope cannot resolve two molecules because they are spaced 20 nanometers apart, one can physically increase that distance. In expansion microscopy, a technique advanced by Jinyoung Kang et al., biological tissues are embedded in a polymer gel and physically expanded. This process "decrowds" densely packed molecular structures, pulling them apart uniformly so that their relative spatial distribution can be resolved using standard, diffraction-limited optical microscopes.
Another physical approach targets the root cause of the limit: the exponential decay of high-frequency spatial information, which is carried by near-field "evanescent" waves. Standard lenses only capture far-field propagating waves, losing the fine details. To combat this, Yi Jin demonstrates that incorporating engineered, negative-permittivity metamaterial layers can flatten transmission curves and amplify these decaying evanescent waves. By preserving these high-frequency components, the system allows subwavelength details to propagate into the detectable far field.
Mathematical Priors and Sparse Reconstruction
When modifying the physical sample or hardware is impossible, the diffraction limit can be overcome computationally. This is achieved by treating imaging as an inverse problem and enforcing strict mathematical constraints on what the underlying scene must look like.
If an algorithm assumes a blurry image could be anything, the inversion fails because infinitely many high-frequency configurations can produce the same blurry blob. However, if we introduce a "cheat code"—such as the prior knowledge that the underlying scene is sparse and composed of discrete point sources—the problem becomes highly constrained. Instead of reconstructing an arbitrary continuous image, the algorithm only needs to find the exact coordinates and amplitudes of a few spikes.
[Discrete Spikes] ---> (Optical Blur/Low-Pass Filter) ---> [Overlapping Blobs]
^ |
|================== [Sparse Deconvolution] ================|
(Enforces Sparsity Prior)
Mustafa Mir et al. leverage this concept by combining quantitative phase imaging with sparse deconvolution algorithms to visualize sub-cellular bacterial structures without toxic fluorescent labels. Mathematically, this "grid-free" super-resolution can be formalized using frameworks like the Beurling-LASSO (BLASSO). As proven by Quentin Denoyelle et al., if the underlying measure is positive and sparse, it is theoretically possible to perfectly recover the exact positions of spikes far below the Rayleigh limit. The fundamental constraint shifts from the wavelength of light to the signal-to-noise ratio (SNR); if the SNR scales sufficiently to counteract the blur, mathematical precision can overcome physical diffraction.
To better understand and train models for these inverse problems, Benquan Wang et al. introduced OpticalNet, a benchmark dataset designed to test how effectively deep learning models can reconstruct complex, invisible nanostructures from their diffraction patterns. Furthermore, the mathematical foundations of this inversion have been bridged to machine learning frameworks; Stefan Harmeling et al. demonstrated a direct link between Fraunhofer diffraction and Reproducing Kernel Hilbert Spaces (RKHS), providing a rigorous framework for how kernel methods can map overlapping optical fields back to their sub-diffraction origins.
Quantum and Adaptive Measurement
A third paradigm targets the measurement process itself. Conventional cameras measure light intensity, which discards the phase and quantum correlations of incoming photons. Because intensity profiles flatten out when two sources merge, standard detection cannot lift the Rayleigh curse.
By changing how photons are sorted and measured before they hit a detector, researchers can extract spatial information that traditional sensors destroy. Fanglin Bao et al. developed a quantum-accelerated imaging framework that uses an adaptive information-theoretic loop. Instead of passively collecting light, the system dynamically updates its measurement basis based on incoming photons, maximizing the data extracted per photon and drastically accelerating the resolution of closely spaced point sources.
In some cases, manipulating quantum states yields counterintuitive advantages. Alexander Mikhalychev et al. demonstrated that intentionally omitting or "losing" a photon from an entangled multi-photon state can actually enhance resolution. This intentional loss induces sharp destructive interference patterns in the detection probability. Because these patterns create steep spatial gradients where conventional intensity curves would be flat and featureless, they provide a highly sensitive metric to distinguish closely spaced objects, effectively breaking the classical limits of far-field optical resolution.