Angular Filter Refractometry (AFR)

• Angular Filter Refractometry (AFR) is an optical diagnostic method that converts plasma refraction into patterned images of constant deflection angles.
• It employs an angular filter in the Fourier plane to record sparse contours, enabling reconstruction of line-integrated electron density via fast marching Eikonal solvers.
• Recent advances include improved filter designs and dual-filter strategies that mitigate diffraction artifacts and resolve gradient-sign ambiguities.

Searching arXiv for the cited AFR papers and closely related work. Angular Filter Refractometry (AFR) is an optical diagnostic in which refraction of a probe beam by a plasma is converted into an image of alternating bright and dark bands whose boundaries correspond to contours of constant refraction angle. In the plasma-diagnostic usage emphasized in recent work, AFR measures absolute contours of line-integrated density gradient by placing a filter with alternating opaque and transparent zones in the focal plane of a probe beam; identifying transitions between image bands with specific filter zones allows inference of the underlying line-integrated electron density, although additional analysis is required for full reconstruction (McCluskey et al., 7 Aug 2025). Recent developments include an inversion framework based on a fast marching Eikonal solver for direct 2D recovery of line-integrated electron density and improved filter designs intended to reduce diffraction artifacts and remove sign ambiguities in the density-gradient reconstruction (McCluskey et al., 7 Aug 2025, Heuer et al., 2023).

1. Measurement principle and optical encoding

AFR operates by exploiting the refraction of an optical probe traversing an inhomogeneous plasma. For a probe of vacuum wavelength $\lambda_p$ propagating along the $z$-axis through a plasma with local electron density $n_e(x,y,z)$, the weakly refracting limit $n_e \ll n_{cr}$ yields a phase shift relative to vacuum

$$\phi(x,y) = - \frac{\pi}{\lambda_p n_{cr}} \int_{-\infty}^{\infty} n_e(x,y,z)\,dz,$$

where $n_{cr}=4\pi^2\epsilon_0 m_e c^2/(e^2\lambda_p^2)$ is the plasma critical density at which the refractive index goes to zero (McCluskey et al., 7 Aug 2025). A transverse gradient in $\phi$ produces a local ray-deflection angle

$$\vec{\theta}(x,y)=\frac{\lambda_p}{2\pi}\nabla \phi(x,y),$$

so the magnitude $\theta = |\vec{\theta}|$ measures the transverse gradient of the line-integrated density (McCluskey et al., 7 Aug 2025).

In the imaging system’s Fourier plane, off-axis rays are displaced radially by an amount proportional to their refraction angle $\theta$. An angular filter consisting of concentric opaque and transparent rings admits only rays whose $z$0 lies within selected intervals. The image plane therefore shows alternating bright and dark bands, and the edges between bands correspond to contours in the object plane where $z$1 equals one of the filter transition angles $z$2 (McCluskey et al., 7 Aug 2025).

A closely related description presents the same mapping in geometrical-optics form: after the plasma, a lens of focal length $z$3 maps a local deflection angle $z$4 into a transverse displacement $z$5 in the focal plane, $z$6, while a bull’s-eye angular filter in that plane transmits selected annuli and blocks others. In this formulation, AFR images are described as light and dark fringes of constant $z$7, i.e. constant line-integrated density gradient (Heuer et al., 2023).

This measurement principle implies that AFR is not, by itself, a direct density imager. It encodes contours of the magnitude of the refraction angle, and therefore sparse information about the gradient of the line-integrated density. Reconstruction of the underlying density field requires interpolation, boundary conditions, and inversion.

2. Governing equations and inverse formulation

The inversion strategy developed for AFR is based on the observation that the phase field obeys an Eikonal-type relation. In a general inhomogeneous medium with refractive index $z$8, the optical path $z$9 satisfies

$n_e(x,y,z)$0

If $n_e(x,y,z)$1 is regarded as the “arrival phase” of each ray, then comparison with the deflection relation gives the image-plane equation

$n_e(x,y,z)$2

which is a first-order nonlinear PDE of the same type as the static Hamilton–Jacobi/Eikonal equation (McCluskey et al., 7 Aug 2025).

The plasma dispersion relation provides the link between refractive index and electron density:

$n_e(x,y,z)$3

Integrating along $n_e(x,y,z)$4 and applying the paraxial, small-angle limit leads back to the phase-shift and deflection formulas above (McCluskey et al., 7 Aug 2025).

An alternative 1D presentation writes the line-integrated density inversion in differential form. With object-plane coordinates $n_e(x,y,z)$5 and integration along $n_e(x,y,z)$6,

$n_e(x,y,z)$7

and for deflection along $n_e(x,y,z)$8 only,

$n_e(x,y,z)$9

This yields a 1D integral reconstruction

$n_e \ll n_{cr}$0

with the integration constant fixed from a fringe whose $n_e \ll n_{cr}$1 is known, often by taking $n_e \ll n_{cr}$2 in a presumed zero-plasma region (Heuer et al., 2023).

The distinction between these formulations is consequential. The 1D form is suitable when the geometry is effectively one-dimensional or monotonic. The Eikonal formulation supports direct inversion of AFR data into a full 2D map of line-integrated electron density without assuming symmetry or an analytic density model (McCluskey et al., 7 Aug 2025).

3. Fast marching inversion for 2D line-integrated density

The 2D inversion method presented in recent AFR work uses a fast marching Eikonal solver. On a uniform 2D Cartesian grid with spacings $n_e \ll n_{cr}$3, an entropy-satisfying upwind discretization of $n_e \ll n_{cr}$4 is enforced through

$n_e \ll n_{cr}$5

where, for example,

$n_e \ll n_{cr}$6

This is the discretized basis for the fast marching update (McCluskey et al., 7 Aug 2025).

The fast marching algorithm follows the classical structure associated with Sethian ’96. Grid points on the known boundary $n_e \ll n_{cr}$7, where $n_e \ll n_{cr}$8 is prescribed, are labeled “Known,” and all others are “Unknown.” Unknown points adjacent to Known points form the “Trial” set, and the local quadratic discretization is solved there to obtain tentative values of $n_e \ll n_{cr}$9. The algorithm repeatedly selects the Trial point with the smallest $$\phi(x,y) = - \frac{\pi}{\lambda_p n_{cr}} \int_{-\infty}^{\infty} n_e(x,y,z)\,dz,$$0, relabels it Known, updates its Unknown neighbors to Trial, and recomputes their tentative values until the whole domain is filled (McCluskey et al., 7 Aug 2025).

For AFR inversion, the solver input is a map $$\phi(x,y) = - \frac{\pi}{\lambda_p n_{cr}} \int_{-\infty}^{\infty} n_e(x,y,z)\,dz,$$1 interpolated continuously from the measured contour set $$\phi(x,y) = - \frac{\pi}{\lambda_p n_{cr}} \int_{-\infty}^{\infty} n_e(x,y,z)\,dz,$$2. The solver marches “up” the $$\phi(x,y) = - \frac{\pi}{\lambda_p n_{cr}} \int_{-\infty}^{\infty} n_e(x,y,z)\,dz,$$3 field from a prescribed boundary value, ultimately producing $$\phi(x,y) = - \frac{\pi}{\lambda_p n_{cr}} \int_{-\infty}^{\infty} n_e(x,y,z)\,dz,$$4 over the full domain $$\phi(x,y) = - \frac{\pi}{\lambda_p n_{cr}} \int_{-\infty}^{\infty} n_e(x,y,z)\,dz,$$5 (McCluskey et al., 7 Aug 2025).

This construction is important because AFR directly records only sparse constant-$$\phi(x,y) = - \frac{\pi}{\lambda_p n_{cr}} \int_{-\infty}^{\infty} n_e(x,y,z)\,dz,$$6 contours rather than a dense field. The fast marching formulation provides a mathematically consistent route from those sparse measurements to a full phase map, and thereby to the line-integrated density.

4. Reconstruction workflow and boundary specification

The reconstruction workflow described for AFR inversion comprises six stages (McCluskey et al., 7 Aug 2025).

Stage	Operation	Stated role
1	Image preprocessing	Flat-field the raw AFR image and suppress high-frequency diffraction artifacts
2	Contour extraction	Locate the band edges and assign each curve the corresponding $$\phi(x,y) = - \frac{\pi}{\lambda_p n_{cr}} \int_{-\infty}^{\infty} n_e(x,y,z)\,dz,$$7
3	$$\phi(x,y) = - \frac{\pi}{\lambda_p n_{cr}} \int_{-\infty}^{\infty} n_e(x,y,z)\,dz,$$8-field interpolation	Interpolate between contours, with special handling for branch regions
4	Boundary condition	Prescribe $$\phi(x,y) = - \frac{\pi}{\lambda_p n_{cr}} \int_{-\infty}^{\infty} n_e(x,y,z)\,dz,$$9 on the outermost contour $n_{cr}=4\pi^2\epsilon_0 m_e c^2/(e^2\lambda_p^2)$0
5	Fast marching inversion	Compute $n_{cr}=4\pi^2\epsilon_0 m_e c^2/(e^2\lambda_p^2)$1 throughout $n_{cr}=4\pi^2\epsilon_0 m_e c^2/(e^2\lambda_p^2)$2
6	Post-processing	Convert $n_{cr}=4\pi^2\epsilon_0 m_e c^2/(e^2\lambda_p^2)$3 to physical units and optionally forward-model AFR bands

Contour extraction may be done manually or via edge detection. Outside regions of band branching, the method assumes that $n_{cr}=4\pi^2\epsilon_0 m_e c^2/(e^2\lambda_p^2)$4 varies exponentially between contours and interpolates linearly in $n_{cr}=4\pi^2\epsilon_0 m_e c^2/(e^2\lambda_p^2)$5. In branch regions, such as two-plume overlap, $n_{cr}=4\pi^2\epsilon_0 m_e c^2/(e^2\lambda_p^2)$6 is modeled by an asymmetric hyperbola fitted to the two flanking contours (McCluskey et al., 7 Aug 2025).

A central element is the boundary condition on the outermost contour. The method assumes that locally $n_{cr}=4\pi^2\epsilon_0 m_e c^2/(e^2\lambda_p^2)$7 increases exponentially toward the plasma, so that

$n_{cr}=4\pi^2\epsilon_0 m_e c^2/(e^2\lambda_p^2)$8

From this, the boundary value along $n_{cr}=4\pi^2\epsilon_0 m_e c^2/(e^2\lambda_p^2)$9 is

$\phi$0

with $\phi$1 typically used in a low-density background (McCluskey et al., 7 Aug 2025).

Post-processing converts $\phi$2 into the line-integrated density $\phi$3 and may include forward-modeling of new AFR bands to test consistency against the original measurements (McCluskey et al., 7 Aug 2025). This workflow shows that the inversion depends not only on the numerical solver but also on physically motivated interpolation and boundary estimation. A plausible implication is that inaccuracies in the peripheral $\phi$4 field and boundary data will preferentially affect the outer portions of the reconstruction.

5. Sign degeneracy, diffraction artifacts, and improved filter design

A basic limitation of AFR is that the sign of the density gradient at each transition is degenerate. Each light-dark transition indicates that a ray crossed between annuli of the angular filter, but does not by itself determine whether the deflection angle increased or decreased. For strictly monotonic plasma expansion, one may assume that successive transitions correspond to increasing zone index, but AFR alone cannot decide the $\phi$5 sign of $\phi$6 for non-monotonic profiles such as colliding flows or cavities (Heuer et al., 2023).

Another limitation is diffraction from sharp-edged filter bands. Conventional angular filters use top-hat transmission, with transparent annuli at $\phi$7 and opaque annuli at $\phi$8. The sharp radial discontinuities produce higher-order diffraction peaks that can mimic real fringe contours and complicate data analysis (Heuer et al., 2023). The inversion workflow of the 2D AFR study correspondingly begins by suppressing high-frequency diffraction artifacts during image preprocessing (McCluskey et al., 7 Aug 2025).

To reduce these artifacts, an improved angular-filter design uses a stochastic pixel pattern with a sinusoidal radial profile. The nominal transmission is written as

$\phi$9

or as a piecewise sinusoid that transitions from 0 to 1 to 0 over each annular zone. Because fabrication is binary, the continuous profile is approximated by treating $$\vec{\theta}(x,y)=\frac{\lambda_p}{2\pi}\nabla \phi(x,y),$$0 as a probability and assigning each $$\vec{\theta}(x,y)=\frac{\lambda_p}{2\pi}\nabla \phi(x,y),$$1 pixel to be transparent or opaque stochastically. Averaged over many pixels around each radius, the filter reproduces the sinusoidal profile while suppressing coherent diffraction from pixel edges (Heuer et al., 2023).

A second development uses a pair of angular filters on two branches of the same probe beam to break the density-gradient degeneracy. With complementary AF patterns, the same physical ray lands in different annuli in the two arms, and only one zone-index assignment is consistent with both masks:

$$\vec{\theta}(x,y)=\frac{\lambda_p}{2\pi}\nabla \phi(x,y),$$2

This permits unique recovery of the sign of each step in $$\vec{\theta}(x,y)=\frac{\lambda_p}{2\pi}\nabla \phi(x,y),$$3 (Heuer et al., 2023).

In synthetic and experimental demonstrations, the improved sinusoidal filter eliminated spurious high-order peaks relative to an older top-hat design, diffraction artifacts were reduced by a factor of $$\vec{\theta}(x,y)=\frac{\lambda_p}{2\pi}\nabla \phi(x,y),$$4 in peak amplitude, reconstruction error versus known synthetic $$\vec{\theta}(x,y)=\frac{\lambda_p}{2\pi}\nabla \phi(x,y),$$5 was reported as $$\vec{\theta}(x,y)=\frac{\lambda_p}{2\pi}\nabla \phi(x,y),$$6, and the dual-filter approach uniquely obtained the sign of $$\vec{\theta}(x,y)=\frac{\lambda_p}{2\pi}\nabla \phi(x,y),$$7 even for reversing gradients (Heuer et al., 2023).

6. Validation, experiments, and adjacent refractometric usage

The 2D Eikonal-inversion method was validated first with synthetic data and then with AFR measurements from the OMEGA EP Laser Facility (McCluskey et al., 7 Aug 2025). For a synthetic single plume, an analytic “Angland” model, exponential in $$\vec{\theta}(x,y)=\frac{\lambda_p}{2\pi}\nabla \phi(x,y),$$8 and super-Gaussian in $$\vec{\theta}(x,y)=\frac{\lambda_p}{2\pi}\nabla \phi(x,y),$$9, was used to generate AFR data. Reported error metrics showed typical $\theta = |\vec{\theta}|$0 error in interpolated $\theta = |\vec{\theta}|$1, boundary errors of $\theta = |\vec{\theta}|$2 in reconstructed $\theta = |\vec{\theta}|$3, and interior errors $\theta = |\vec{\theta}|$4 (McCluskey et al., 7 Aug 2025). For a synthetic double plume formed by summing two displaced single-plume profiles and stretching in $\theta = |\vec{\theta}|$5, the branch interpolation via the hyperbola model yielded reconstructed $\theta = |\vec{\theta}|$6 errors that remained $\theta = |\vec{\theta}|$7 in the interaction region (McCluskey et al., 7 Aug 2025).

In OMEGA EP plume experiments, a $\theta = |\vec{\theta}|$8 nm, $\theta = |\vec{\theta}|$9 ps probe propagated along $\theta$0 through plasmas driven by $\theta$1 nm, $\theta$2 ns, $\theta$3 beams on CH foils, with both single- and double-plume geometries studied. Lineouts of $\theta$4 from the fast marching inversion agreed with prior 1D reconstructions to within experimental uncertainties. Forward modeling from the recovered $\theta$5 produced AFR bands whose excellent alignment with the raw image confirmed self-consistency (McCluskey et al., 7 Aug 2025).

The improved-filter study also reported OMEGA-EP measurements using a $\theta$6 ns, $\theta$7 nm probe viewing a laser-ablated Si foil. Experimental lineouts using an older AF3 filter showed extra narrow peaks attributed to diffraction, whereas AF12 removed those peaks; the spatial resolution of the fringe pattern was stated as $\theta$8, limited by camera, pixel size, and refracted-beam divergence (Heuer et al., 2023).

The acronym AFR has also appeared in a different refractometric context associated with near-membrane refractometry by supercritical-angle fluorescence, where the measurement is based on determining the critical-angle ring in the back-focal plane of a microscope. In that implementation, the local refractive index is obtained from $\theta$9, and the method reported $z$00 precision with subcellular resolution on a standard TIRF microscope with a removable Bertrand lens (Brunstein et al., 2016). This usage is conceptually related only in the broad sense that angular filtering in a Fourier or back-focal plane is exploited to infer refractive properties. The plasma-diagnostic AFR literature discussed above concerns line-integrated electron-density gradients and their inversion, rather than near-membrane cellular refractive index.

7. Limitations, misconceptions, and prospective extensions

A frequent misconception is that AFR directly measures density. The reported formulation makes clear that AFR records sparse contours of constant ray-deflection angle $z$01, and therefore absolute contours of line-integrated density gradient; the density itself is obtained only after inversion, interpolation, and specification of boundary data (McCluskey et al., 7 Aug 2025, Heuer et al., 2023).

Another misconception is that a single AFR image necessarily determines the gradient sign. In fact, the sign is degenerate unless monotonicity or some other external constraint is imposed. The dual-filter approach was proposed precisely because AFR alone cannot resolve whether a given transition corresponds to increasing or decreasing deflection angle in non-monotonic profiles (Heuer et al., 2023).

Current discussions of improvement focus on three areas. First, simultaneous interferometry could directly measure $z$02 in the low-density periphery and thereby provide exact Dirichlet data for $z$03 on $z$04 (McCluskey et al., 7 Aug 2025). Second, using two different angular filters in parallel breaks the gradient-sign ambiguity and increases the density of contour sampling; multi-color AFR could likewise provide extra $z$05 contours at different probe wavelengths (McCluskey et al., 7 Aug 2025). Third, combining AFR with 2D shadowgraphy may resolve finer spatial features, such as shocks or instabilities, by correlating $z$06 signatures (McCluskey et al., 7 Aug 2025).

Taken together, these developments position AFR as a diagnostic framework in which optical refraction, angular filtering, and PDE-based inversion are tightly coupled. The recent literature suggests a transition from 1D analysis and forward-fitting techniques toward direct 2D reconstruction with explicit treatment of interpolation, boundary conditions, diffraction suppression, and gradient-sign recovery (McCluskey et al., 7 Aug 2025).