Research
How geometry becomes light.
Three threads — a theory, an object, an experiment — tied by one question: what does it mean for light to follow a geodesic?
One sentence holds the programme together: design the geometry, and light will follow. A refractive index is a metric; a curved surface is a lens; a coordinate transformation is a material. I work on the mathematics that makes those equivalences exact — and on the numerical experiments that keep them honest.
Transformation optics
Maxwell's equations are form-invariant under coordinate transformations. That single fact turns a design problem in materials into a design problem in geometry.
Start with Fermat. In a medium with refractive index $n(\mathbf{r})$, a ray makes the optical path
\[ S=\int n(\mathbf{r})\,ds \]
stationary. That is the statement that rays are geodesics of the optical metric $g_{ij}=n^{2}\delta_{ij}$. Now change coordinates, $\mathbf{r}\mapsto\mathbf{r}'$, and write the same metric in the new frame. Maxwell's equations do not notice the change; the constitutive parameters do.
If $\Lambda$ is the Jacobian of the transformation, the transformed medium is
\[ \varepsilon'=\frac{\Lambda\,\varepsilon\,\Lambda^{T}}{\det\Lambda} \]
\[ \mu'=\frac{\Lambda\,\mu\,\Lambda^{T}}{\det\Lambda}. \]
The physics is unchanged; the material has absorbed the geometry. A region of space can be compressed into a shell, and the interior becomes invisible to the field outside. This is the transformation-optics cloak, proposed independently by Pendry and by Leonhardt in 2006, and demonstrated at microwave frequencies by Schurig and co-workers the same year.
What makes the idea practical is a restriction: if the transformation is conformal in two dimensions, the required medium is isotropic. The price is that only a subset of transformations is available. Much of the field since 2006 has been an argument about that trade — anisotropy for generality, conformality for fabrication, metasurfaces for both.
My interest is the boundary between the two. A cloak that cannot be built is a statement about a metric, not a device. The useful question is not “does an equivalent medium exist?” but “what is the smallest change in the geometry that survives the tolerances of a real surface?”
Geodesic lenses
A spherically symmetric index profile is a surface of revolution in disguise. The disguise is exact: rays in the medium are geodesics on the surface.
For a radially symmetric profile $n=n(r)$, the optical metric is
\[ ds^{2}=n^{2}(r)\left(dr^{2}+r^{2}d\varphi^{2}\right). \]
Look for a surface of revolution with profile $z(\rho)$ whose metric is the same. Matching term by term gives the three equivalence relations
\[ \rho=n(r)\,r,\qquad ds=n(r)\,dr, \]
\[ dz^{2}=ds^{2}-d\rho^{2}. \]
The first relation is the projected radius; the second identifies optical arc length with surface arc length; the third fixes the height. Differentiating $\rho=nr$ gives the master identity $s'(\rho)\,(n+rn')=n$, and the surface follows by quadrature:
\[ z(\rho)=\int_{0}^{\rho}\sqrt{s'(\xi)^{2}-1}\;d\xi. \]
The Luneburg profile makes the geometry unusually explicit. With $n^{2}=2-r^{2}$ the projected radius is $\rho^{2}=1-(1-r^{2})^{2}$, and the master identity gives
\[ s'(\rho)=\frac12\left(1+\frac{1}{\sqrt{1-\rho^{2}}}\right),\qquad s(\rho)=\tfrac12\left(\rho+\arcsin\rho\right). \]
This is Rinehart's surface: the mean of a flat disc, $s=\rho$, and a hemisphere, $s=\arcsin\rho$. It is not a sphere. The optical metric of the Luneburg lens has Gaussian curvature $K=4/(2-r^{2})^{3}$, which is not constant; the lens whose equivalent surface is a sphere is Maxwell's fish-eye, $n=2/(1+r^{2})$, with $K=1$ and great circles for rays. On Rinehart's surface the statement is different and just as sharp: the geodesics that continue a parallel bundle of rays all leave through one point of the rim.
The Morse family generalises this. The profile
\[ n^{2}(r)=2r^{-2+1/B}-r^{-2+2/B},\qquad 0<r\le 1, \]
reduces to Luneburg at $B=\tfrac12$ and to Eaton at $B=1$. My thesis proves that the conformal Morse lens is the $(A=B,\,C=1)$ slice of Cornbleet's general family, with $a=-1/B$, and constructs the equivalent surfaces for the intermediate profiles. The numerical ray traces agree with the surface geodesics to the discretisation error of the integrator. For the whole family the quadrature closes:
\[ s(\rho)=B\left(\rho+\arcsin\rho\right). \]
There is an existence condition hidden in the quadrature: a real surface needs $s'(\rho)\ge1$. Here $s'(0)=2B$, so the family is realisable exactly for $B\ge\tfrac12$, and the Luneburg lens is the borderline case with a smooth apex. For $B>\tfrac12$ the index diverges at the origin, but only as $r^{-1+1/(2B)}$: the optical path to the centre stays finite, the surface stays closed, and what the divergence leaves behind is a conical apex of half-angle $\arcsin(1/2B)$ — $30^{\circ}$ for the Eaton lens. A geodesic with impact parameter $b$ sweeps an azimuth $B\,(\pi+2\arccos b)$ between entering and leaving the rim, which is how one family of surfaces runs from a focusing lens to a retroreflector. The obstruction at the Eaton end is a cone point, not an opening. It is still geometry, not fabrication.
Corrected 10 Nov 2026. An earlier version of this page said that the Luneburg lens is equivalent to a unit sphere and that the Eaton surface fails to close. Both statements were wrong; the paragraphs and Fig. R2 above replace them. The check is in the revision note.
Quantum optics
At the single-photon level, interference is not a wave passing through two slits. It is a statement about what the experiment could have told us.
Send photons through a double slit one at a time. Each photon produces one detection. After a few hundred, the detections trace an interference pattern; after a few thousand, the pattern is unmistakable. Nothing in the apparatus knows in advance where a particular photon will land.
The amplitude at the screen is the sum of the amplitudes from the two slits, and the probability is the square:
\[ I(x)\propto\left|\psi_{1}(x)+\psi_{2}(x)\right|^{2}. \]
For slit separation $d$ and screen distance $L$, the fringe spacing is $\Delta x=\lambda L/d$. The envelope is set by the slit width. The pattern is not a property of a single photon; it is a property of the ensemble, and of the fact that the two paths were indistinguishable.
Make them distinguishable. A which-path detector — a polariser, a path marker, a detector at one slit — gives the experiment information about the route. The interference term disappears, and the pattern becomes the sum of two single-slit envelopes. The visibility $V$ and the path distinguishability $D$ trade off: $V^{2}+D^{2}\le1$. This is complementarity in its quantitative form.
The teaching-lab version is less exotic than the thought experiment and more instructive. Alignment matters; a few microns of slit tilt washes out the fringes. Dark counts add a flat background. The source must be attenuated until the probability of two photons in the apparatus is small. What looks like a clean conceptual story becomes a sequence of experimental decisions — which is exactly why it is worth doing.
The figure is a simulation of that experiment, not a claim of new data. It uses $\lambda=633$ nm, $d=0.12$ mm and $L=1.2$ m; the dots are sampled from the same intensity law the apparatus would obey.
What I want to know next
Three questions that connect the threads — and that I would like to argue about with someone.
- Which profiles admit a real, manufacturable surface? The condition $s'\ge1$ is a first filter, not a design rule: a cone point is a legal surface and a hard thing to machine. I want a classification that includes tolerance and finite aperture.
- Can a conformal design survive fabrication? A computed profile is exact; a machined one is not. The sensitivity of the caustic to surface error is a well-posed numerical question.
- What does a which-path measurement do to the field? The textbook story is about information. A detector is also a physical object with a response time and a back-action, and I would like to understand the bridge.
If any of these is close to your work, write to me. I am especially interested in collaborations that put a computed surface into an optical test.
The shelf behind this page
The papers and books I return to. Links go to the DOI or the arXiv record.
- R. K. Luneburg, Mathematical Theory of Optics. Brown University Press, 1944.
- J. B. Pendry, D. Schurig, and D. R. Smith, “Controlling electromagnetic fields,” Science 312, 1780–1782 (2006). doi
- U. Leonhardt, “Optical conformal mapping,” Science 312, 1777–1780 (2006). doi
- D. Schurig et al., “Metamaterial electromagnetic cloak at microwave frequencies,” Science 314, 977–980 (2006). doi
- R. F. Rinehart, “A solution of the problem of rapid scanning for radar antennae,” J. Appl. Phys. 19, 860–862 (1948). doi
- M. Šarbort and T. Tyc, “Spherical media and geodesic lenses in geometrical optics,” J. Opt. 14, 075705 (2012). doi
- H. Chen, “The Morse lens,” arXiv:2002.03884 (2020). arXiv
- M. Born and E. Wolf, Principles of Optics, 7th ed. Cambridge University Press, 1999.
- B.-G. Englert, “Fringe visibility and which-way information: an inequality,” Phys. Rev. Lett. 77, 2154–2157 (1996). doi
- M. O. Scully and K. Drühl, “Quantum eraser: A proposed photon correlation experiment concerning observation and ‘delayed choice’ in quantum mechanics,” Phys. Rev. A 25, 2208 (1982). doi