Global smoothness for relativistic Vlasov–Maxwell

Source and verification

This article explains family 362 of the October 6, 2026 OpenAI collection. Research conclusions are attributed to its manuscripts. The review covers statements, introductions, and selected proof mechanisms; it is not an independent verification of the complete proofs.

Overview

A collisionless plasma transports a particle density through phase space while generating the electromagnetic field that accelerates it. Family 362 claims global smooth evolution for arbitrary smooth admissible data in three dimensions. Its central mechanism estimates the signed impulse along particle trajectories, retaining cancellation before taking absolute values.

🏷️ Preliminaries: phase space and characteristics

Let $f(t,x,v)\ge 0$ be a density on position–momentum space ${\mathbb{R}}^3\times {\mathbb{R}}^3$. Momentum and velocity differ in the relativistic model:

$$q(v)=\sqrt{1+|v{|}^2},\hat{v}=v/q(v).$$

Thus particle speeds are less than one even when momenta are arbitrarily large. In units where the speed of light is one, the one-species Vlasov–Maxwell equations are

$$\begin{array}{rl}& {\partial }_{t}f+\hat{v}\cdot {\nabla }_{x}f\\ & +(E+\hat{v}\times B)\cdot {\nabla }_{v}f=0,\\ & {\partial }_{t}E-\nabla \times B=-j,\\ & {\partial }_{t}B+\nabla \times E=0,\\ & \nabla \cdot E=\rho ,\nabla \cdot B=0,\end{array}$$

where $\rho =\int fdv$ and $j=\int \hat{v}fdv$. There is no neutralizing background. The first equation says that $f$ is constant along characteristics satisfying $X^{\prime }=\hat{V}$ and $V^{\prime }=E(t,X)+\hat{V}\times B(t,X)$.

The divergence constraints need only be prescribed initially. Integrating the transport equation in momentum gives ${\partial }_t\rho +\nabla \cdot j=0$; taking divergences in Maxwell’s equations then shows that both constraint errors are constant in time.

🏷️ The manuscript’s global existence claim

Family 362, as stated in the manuscript

Suppose $f_0\ge 0$ is smooth and compactly supported in phase space. Suppose $E_0,B_0$ are smooth, have bounded derivatives of every order, belong to $L^2({\mathbb{R}}^3)$, and satisfy the two divergence constraints. Then there is a unique global classical solution, smooth on every finite time interval, with compact particle phase-space support on each such interval. No smallness, symmetry, or neutrality condition is imposed. (OpenAI, 2026)

Compact support here concerns the particles. A nonzero total charge produces a field tail, so requiring compactly supported electric fields would change the problem. The momentum bound is allowed to grow with the time interval.

🏷️ Conservation is weaker than continuation

The conserved total energy is

$$\begin{array}{rl}\mathcal{E}(t)=& \int q(v)f(t,x,v)dxdv\\ & +\frac{1}{2}\int (|E{|}^2+|B{|}^2)dx.\end{array}$$

Indeed, along a particle trajectory,

$$\frac{d}{dt}q(V)=\hat{V}\cdot E,$$

because a vector is orthogonal to its cross product with $B$. After integration, particles gain energy at rate $\int j\cdot E$, while the field loses exactly that amount.

This computation controls an integral of $qf$, not the largest momentum in the support. A very small particle mass can carry very large momentum while contributing little total energy. The classical continuation criterion instead requires bounded momentum support on finite time intervals. Passing from energy to that stronger control is the research problem.

🏷️ Signed impulse and the doubling argument

The manuscript estimates

$$\left|{\int }_{t_1}^{t_2}(E(t,X)+\hat{V}\times B(t,X))dt\right|,$$

which equals $|V(t_2)-V(t_1)|$. Placing the absolute value inside the integral would discard the cancellation needed in its argument.

Retarded fields express the force through source particles meeting the receiver’s backward light cone. The proof divides those interactions by momentum, distance, and relative angles. Energy flux and occupation time—the amount of time particles spend in a specified interaction region—give direct estimates. A narrow angular configuration has an additional square-root momentum loss.

The proposed cancellation combines the source-acceleration and nonacceleration terms before integration. A total derivative emerges, together with bulk terms and a term proportional to the receiver’s acceleration. The latter must have a sufficiently small coefficient; simply integrating by parts would not suffice.

The source first uses this identity to limit changes in particle direction. Between direction changes, relative motion crosses certain interaction regions monotonically, improving their occupation bounds. A selection argument then isolates the regions where the direct estimate is too expensive and applies the signed identity there. Summation makes the acceleration coefficient small enough to absorb the earlier loss. Endpoint intervals are restored by a continuity bootstrap.

Finally, the resulting impulse control forces a minimum time for successive momentum doublings. The manuscript proves that these time costs have divergent sum. Infinitely many doublings therefore cannot fit into a finite interval, closing the continuation argument. The energy identity above explains the available budget; it does not establish this signed estimate.

🏷️ Key steps and proof outlines

The estimate that actually closes continuation

Fix a finite time horizon. Let $P$ be a dyadic number at least $1024$ times the largest particle energy on the current classical interval, and set $L=\log (2+P)$. For a receiver whose energy stays between $w/8$ and $8w$, the manuscript’s signed-increment proposition states

$$|V_X(t_2)-V_X(t_1)|\le MP\sqrt{I}+AI\frac{P^{2}L}{w},I=t_2-t_1.$$

The constants $M,A$ depend on the data and horizon, but not on $P$ or the endpoint of the current classical interval. This uniformity is the essential output. A bound with constants depending on the unknown maximum momentum would not close continuation.

The estimate is initially a bootstrap assumption. Energy flux and occupation estimates are proved under it, followed by a strict improvement of the same inequality. Its two terms play different roles: the square-root term controls short-time effects, while the logarithmic linear term will determine the minimum cost of a momentum doubling.

The source-trajectory derivative removes the bad source denominator

Write $u$ for the source velocity, $a$ for the receiver velocity, and $n$ for the unit vector from the retarded source to the receiver. The retarded times satisfy $t=s+r$, where $r$ is their spatial separation. Define

$$d=1-n\cdot u,D=1-n\cdot a,e=1-a\cdot u,k_0=u-\frac{e}{D}n.$$

Both $d$ and $D$ are positive because the velocities are strictly subluminal. Differentiating the light-cone relation gives $dt/ds=d/D$. In source-label measure, the combined acceleration and transport kernels satisfy the exact identity

$$\mathcal{K}=-\frac{d}{ds}\left(\frac{k_0}{rd}\right)+\frac{e}{r^2{D}^2}\left(\frac{n}{q_X^{2}D}-a\right)+\frac{n(a_t\cdot k_0)}{r{D}^2}.$$

The proof is a product-rule calculation, using $|u{|}^2=1-q^{-2}$ and $|a{|}^2=1-q_X^{-2}$. Including the transport kernel is indispensable: it cancels part of the derivative of the cone direction. The residual terms contain neither source acceleration nor inverse powers of $d$. The primitive still contains $d^{-1}$, so the identity transfers the problem to endpoint and cutoff terms; it does not make all singularities disappear.

The receiver derivative obeys

$$a_t=\frac{(\mathrm{Id}-a\otimes a)K}{q_X},K=V_X^{\prime }.$$

Its contribution is therefore linear in the unknown receiver force. The selection and occupation estimates must make its coefficient small enough to absorb the available absolute-force bound.

Why differentiating a selected partition is affordable

The dyadic interaction bins have smooth nonnegative weights ${\beta }_j$ with ${\sum }_j{\beta }_j=1$. Select a fixed set of indices and put $\chi ={\sum }_{j\in \mathcal{S}}{\beta }_j$. For source or geometric derivatives,

$$\partial \chi =-\sum _{j\notin \mathcal{S}}\partial {\beta }_j.$$

Thus these derivative errors live where selected and unselected bins overlap. On that overlap the scales are comparable, allowing the errors to be charged to neighboring bins for which the direct estimate is affordable. One must first factor out their common physical kernel before using the partition identity. The source does not assume the false pointwise bound $|\partial {\beta }_j|\lesssim {\beta }_j$.

In a bin with source energy $p$, angular scales $\theta \asymp \sqrt{d}$, $\varphi \asymp \sqrt{D}$, and distance $h$, the integrated receiver-force coefficient is bounded by

$$\frac{C}{w}\min \left\{h^2(p\theta {)}^3,\frac{1}{p\theta h{\varphi }^2}\right\}.$$

The first bound uses the phase-space density and bin volume; the second uses cone energy flux. Their minimum is what permits summation across distances. The separate direction-change estimate improves occupation at intermediate angles before the final selection is made.

Restoring endpoints without losing the signed estimate

Integration by parts uses a time weight $\psi$ that vanishes at both interval endpoints. Choose ramps of length $\eta I$ and let $\psi =1$ between them. On the left ramp, with $F(t)=V_X(t)-V_X(t_1)$,

$$\int (1-\psi )Kdt=-\int (1-\psi {)}^{\prime }Fdt.$$

The ramp has total variation one, so the signed bootstrap controls this error by $MP\sqrt{\eta I}+A\eta I{P}^{2}L/w$. This avoids integrating $|K|$ on the ramp. Choose $\eta$ first, then $M$, then $A$, and finally the threshold for $P$. The resulting strict improvement extends by continuity to every eligible interval.

At the first time the maximal energy reaches $2^n$, follow a maximizing particle back to its last crossing of $2^{n-1}$. Set $P=1024\cdot 2^n$ and $w=2^n$. Since $q$ is 1-Lipschitz, its momentum changes by at least $2^{n-1}$. The displayed increment bound forces a time cost at least $c/\log (2+1024\cdot 2^n)$. Summing over $n$ diverges. This proves the continuation consequence given the manuscript’s uniform signed estimate; the difficult occupation and selection lemmas are its substantive analytic inputs. (OpenAI, 2026 s “Signed cancellation along the source trajectories,” “Selecting the signed part of the force,” and “Closing the signed-increment bootstrap”)

See Also

• Family 363 treats a collisional kinetic equation in a rough solution class, where uniqueness is a separate issue.
• Family 364 explains how a collisional equation arises from many-particle dynamics.

🏷️ Manuscripts and verification scope

• Global classical solutions of the three-dimensional relativistic Vlasov–Maxwell system — pinned manuscript and companion source.

The family scope page describes selected formalization coverage. The scope page describes global classical existence and uniqueness for the one-species system with the admissible data stated above. No Lean build, axiom audit, or manuscript-to-formal-statement equivalence check was performed for these notes.

OpenAI collection provides prerequisites and the reading sequence; coverage register records completed and pending families.

🐻  OpenAI 2026. Global classical solutions of the three-dimensional relativistic Vlasov–Maxwell system.