research/AMR-005-0003.md

metadata

id: AMR-005-0003
classification: PARTIAL-PROGRESS
wording_corrected: 'no'

AMR-005-0003 — Birkhoff's theorem for Lorentz billiards

Problem (corrected statement if needed)

Source: S. Tabachnikov, "A Baker's Dozen of Problems", Arnold Mathematical Journal 1 (2015), 59–67, DOI 10.1007/s40598-014-0001-3 (Section 3, "Birkhoff's Theorem for Lorentz Billiards"). Original wording, verified verbatim against the published article (publisher HTML and the Springer-final PDF):

The classical Birkhoff theorem states that, for every $n\ge 3$ and $1\le k\le n/2$, the billiard system inside a plane oval has at least two $n$-periodic trajectories with the rotation number $k$. Consider the billiard system inside an oval in the Lorentz plane with the pseudo-Euclidean metric $ds^2=dx^2-dy^2$. Is there an analog of Birkhoff's theorem in this set-up? Billiard trajectories in pseudo-Euclidean space can be of three types: space-like, time-like, and light-like, see Khesin and Tabachnikov (2009) for Lorentz billiards. One would expect separate existence statements for space-like and time-like trajectories. A convex body in $\mathbb{R}^n$ has at least $n$ diameters (2-periodic billiard trajectories). If the ambient space is pseudo-Euclidean, $\mathbb{R}^{p,q}$, then there are at least $p$ space- and at least $q$ time-like diameters (Khesin and Tabachnikov 2009). A lower bound on the number of periodic billiard trajectories in multi-dimensional Euclidean space is obtained in Farber and Tabachnikov (2002). What happens with multi-dimensional pseudo-Euclidean billiards?

The dataset transcription is a faithful condensation of this; no correction needed.

Status / Literature

All references verified against Crossref metadata (DOIs below) or the arXiv API.

• Foundation: pseudo-Riemannian billiards, and the $n=2$ case. B. Khesin, S. Tabachnikov, "Pseudo-Riemannian geodesics and billiards", Adv. Math. 221 (2009), 1364–1396, DOI 10.1016/j.aim.2009.02.010. Develops the symplectic/variational formalism for billiards in pseudo-Euclidean spaces; proves that a convex body in $\mathbb{R}^{p,q}$ has at least $p$ space-like and at least $q$ time-like diameters (2-periodic orbits). In the Lorentz plane this gives one space-like and one time-like 2-periodic trajectory — the first case of the desired Birkhoff analog.
• Euclidean multidimensional benchmark. M. Farber, S. Tabachnikov, "Topology of cyclic configuration spaces and periodic orbits of multi-dimensional billiards", Topology 41 (2002), 553–589, DOI 10.1016/S0040-9383(01)00021-0. Lusternik–Schnirelmann lower bounds for periodic orbits in Euclidean $\mathbb{R}^n$; the pseudo-Euclidean analog is precisely what is being asked for.
• Integrable case (ellipsoids), all dimensions and signatures — SOLVED. V. Dragović, M. Radnović, "Ellipsoidal billiards in pseudo-Euclidean spaces and relativistic quadrics", Adv. Math. 231 (2012), 1173–1201, DOI 10.1016/j.aim.2012.06.004, arXiv:1108.4552. Complete description of periodic billiard trajectories within ellipsoids in $\mathbb{R}^{p,q}$, including light-like ones, via Cayley-type analytic criteria and a "relativistic quadrics" colouring of confocal pencils. See also the same authors' "Minkowski plane, confocal conics, and billiards", Publ. Inst. Math. (Beograd) 94(108) (2013), 17–30.
• Integrable planar case, quantitative. A. K. Adabrah, V. Dragović, M. Radnović, "Periodic Billiards Within Conics in the Minkowski Plane and Akhiezer Polynomials", Regul. Chaotic Dyn. 24 (2019), 464–501, DOI 10.1134/S1560354719050034, arXiv:1906.04911. Explicit existence/counts of periodic trajectories of each causal type inside conics in the Minkowski plane. Thus for ellipses a full "Lorentzian Birkhoff theorem" holds, with separate space-like and time-like statements.
• Related. D. Genin, B. Khesin, S. Tabachnikov, "Geodesics on an ellipsoid in Minkowski space", Enseign. Math. 53 (2007), 307–331 (Poncelet-type theorem for null geodesics; background for item 7 of the same list).
• General ovals, $n\ge 3$ — OPEN. I found no published work proving (or disproving) a Birkhoff-type existence theorem for space-like/time-like $n$-periodic orbits, $n\ge 3$, inside a general oval in the Lorentz plane, nor general existence results for multidimensional pseudo-Euclidean billiards beyond ellipsoids and the $n=2$ case above. The Baker's Dozen article has only 2 citations in Crossref, neither addressing this item; arXiv searches ("Lorentz billiards", "pseudo-Euclidean billiards periodic") return only the integrable-case literature and unrelated "Lorentz gas" channels.

Work done

• Retrieved the original Section 3 text from the published AMJ article (both the HTML and the Springer-final PDF) — the dataset wording is accurate.
• Verified all citations via Crossref (api.crossref.org/works/...) and the arXiv API; caught and corrected a wrong DOI guess for Dragović–Radnović 2012 (correct: 10.1016/j.aim.2012.06.004).
• Searched for post-2015 progress on the general problem (arXiv API, Crossref, web search; attempted Semantic Scholar citation lookup — fetch failed).
• Analyzed the variational problem underlying a possible proof; the rigorous observations below are my own (though presumably known to experts in spirit).

Result

The problem is open in general, with the following state of knowledge and my analysis of the obstruction.

What is known. (i) $n=2$: at least one space-like and one time-like 2-periodic orbit for any Lorentz oval, and $\ge p$ / $\ge q$ diameters in $\mathbb{R}^{p,q}$ (Khesin–Tabachnikov 2009). (ii) Ellipses and ellipsoids: complete Birkhoff-type picture for all $n$ and all causal types, with Cayley-type existence criteria (Dragović–Radnović 2012; Adabrah–Dragović–Radnović 2019). (iii) Multidimensional Euclidean bounds (Farber–Tabachnikov 2002) have no known pseudo-Euclidean counterpart.

My analysis — why the classical proof does not transfer. Let $\gamma$ be a smooth strictly convex oval in $\mathbb{R}^{1,1}$, $ds^2=dx^2-dy^2$.

1. Causal decomposition. The tangent direction map $\gamma\cong S^1\to\mathbb{RP}^1$ has degree 1, so each of the two null directions occurs as a tangent exactly twice: $\gamma$ splits into 4 arcs, two with space-like tangent ($|dy/dx|<1$, top and bottom) and two with time-like tangent (left and right). Each arc has total turning $\pi/2$; by strict convexity the direction of any chord lies strictly between the tangent directions at its endpoints, hence every chord of a closed space-like arc is space-like, and every chord of a time-like arc is time-like. The billiard reflection law is well defined at every interior point of each arc (tangent non-null).

2. The space-like maximum argument collapses. Birkhoff's proof maximizes perimeter over inscribed $n$-gons; the key lemma is that inserting a vertex on the curve strictly increases the perimeter (strict triangle inequality), forcing the maximum to be a genuine $n$-gon. For the Lorentz length $\ell(x,y)=\sqrt{x^2-y^2}$ the Hessian on the space-like cone ${x>|y|}$ is negative semi-definite ($\ell_{xx}=-y^2/\ell^3$, $\ell_{yy}=-x^2/\ell^3$, determinant $0$), so $\ell$ is concave and hence superadditive on the cone: $\ell(u+v)\ge \ell(u)+\ell(v)$. Consequently, inserting a vertex on a space-like arc strictly decreases the Lorentz perimeter, and the maximum of the perimeter over inscribed $n$-gons of a space-like arc is attained on the diagonal stratum — it degenerates to the 2-gon (the diameter). So no space-like $n$-periodic orbit with $n\ge 3$ can be obtained by maximization within an arc: the variational structure genuinely differs from the Euclidean case. (Numerically: $u=(1,\tfrac12)$, $v=(1,-\tfrac12)$ give $\ell(u)+\ell(v)=\sqrt3<2=\ell(u+v)$.)

3. The time-like minimum argument collapses too. Time-like chords in a common causal cone satisfy the reverse triangle inequality, so one should minimize — but the minimum over the compact configuration space is $0$, attained at total collapse; one is forced into minimax/linking arguments on a contractible configuration space, with the functional degenerating on null-chord strata where the reflection law is undefined.

4. What this suggests. A proof of the Lorentzian Birkhoff theorem (if true) must either (a) work with orbits winding around the whole oval, where chords join different arcs and the null-chord strata must be controlled (compactness holds — the inscribed $n$-gon space with fixed rotation number is compact and $\ell$ is continuous — but maximizers may hit null strata), or (b) replace LS-theory on cyclic configuration spaces (Farber–Tabachnikov) by a pseudo-Euclidean Morse theory that accounts for the causal strata. Neither has been carried out in the literature.

What remains

• Main open case: existence of space-like (resp. time-like) $n$-periodic orbits, $n\ge 3$, with given rotation number, for a general (non-ellipsoidal) Lorentz oval. Even the $n=3$ case of a single space-like triangle orbit is unpublished as far as I could verify.
• Multidimensional case: any analog of the Farber–Tabachnikov LS bounds in $\mathbb{R}^{p,q}$ beyond the $n=2$ diameters of Khesin–Tabachnikov.
• Concrete next steps: (1) settle whether a maximum of the Lorentz perimeter over winding $n$-gons can lie on a null-chord stratum — if it always does, the naive analog is false and one must restrict to ovals with additional hypotheses (e.g., ovals whose space-like arcs support a genuine billiard interval exchange); (2) test the question on nearly-elliptical perturbations, where the integrable classification of Dragović–Radnović provides orbits whose persistence could be studied via the twist-map/Poincaré–Birkhoff framework; (3) develop Morse theory for the signed Lorentz-length functional on cyclic configuration spaces with causal stratification.