Two Boltzmann evolutions with the same data and exact local conservation

← All digests

Family 363 · Partial differential equations
Importance 5/5 · Provisional editorial rating
Sources consulted 2026-10-26

1. The question

The Boltzmann equation describes a dilute gas through a density of particles in position and velocity. Particles travel freely between binary collisions, while each collision conserves momentum and kinetic energy. The entropy inequality supplies an additional direction of time.

Do these requirements determine the gas’s future uniquely from its initial density? For sufficiently controlled data and solutions, many uniqueness theorems answer yes. For the broad class allowed by finite mass, energy and entropy, the issue is more delicate: the density may concentrate strongly in space even when these global quantities are finite.

Family 363 claims nonuniqueness for one such rough datum on a periodic three-dimensional box. The stronger manuscript, M1, constructs two global solutions with exactly the same initial density, both satisfying the entropy inequality and exact local conservation of mass, momentum and kinetic energy.

The initial velocities are bounded, but the density amplitudes become unbounded on smaller and smaller spatial scales. The claim is therefore about weak entropy solutions for singular data, not two smooth classical evolutions from a well-controlled smooth initial gas.

2. Background and essential definitions

Transport plus elastic collisions

The position space is the unit three-torus $\mathbb T^3=\mathbb R^3/\mathbb Z^3$, which imposes periodic boundary conditions. Velocities lie in $\mathbb R^3$. A nonnegative density $F(t,x,v)$ satisfies

$$\partial_tF+v\cdot\nabla_xF=Q(F,F).$$

The left side is free transport. The collision term splits into gain and loss,

$$Q(F,F)=Q^+(F,F)-Q^-(F,F).$$

Gain counts collisions producing velocity $v$; loss counts particles leaving that velocity through collisions.

For another velocity $v_*$ and a unit collision direction $\omega\in S^2$, put

$$a=(v-v_*)\cdot\omega, \qquad v'=v-a\omega, \qquad v_*'=v_*+a\omega.$$

These elastic collision rules preserve $v+v_*$ and $|v|^2+|v_*|^2$. With the manuscript’s hard-sphere normalization,

$$Q(F,F)(v)=\int_{\mathbb R^3\times S^2}|a| \bigl(F(v')F(v_*')-F(v)F(v_*)\bigr)\,dv_*\,d\omega,$$

where every factor has the same $t,x$. The relative normal speed $|a|$ is the hard-sphere collision kernel.

Entropy and renormalization

Finite mass, energy and absolute entropy mean

$$\int_{\mathbb T^3\times\mathbb R^3} F_0\bigl(1+|v|^2+|\log F_0|\bigr)\,dx\,dv<\infty,$$

with $0|\log0|=0$. These integral bounds allow high narrow peaks. Because collisions are quadratic in $F$, they do not automatically make the collision term integrable.

A renormalized solution therefore satisfies equations for bounded-growth transformations of the density. Using the reference Maxwellian

$$M(v)=(2\pi)^{-3/2}e^{-|v|^2/2},$$

the manuscripts test the equation through $M\beta(F/M)$, where $\beta(0)=0$ and $|\beta'(z)|\le C/(1+z)$. This weakens the effect of large amplitudes while retaining the transport and collision structure.

The relative entropy is

$$H_M(F)=\int\bigl(F\log(F/M)-F+M\bigr)\,dx\,dv.$$

The entropy inequality states that this quantity plus accumulated nonnegative Boltzmann collision dissipation does not exceed its initial value. In the mathematical sign convention, entropy decreases; physical thermodynamic entropy has the opposite sign.

Local conservation is stronger than a total bound

The five collision invariants are

$$\psi(v)\in\{1,v_1,v_2,v_3,|v|^2/2\}.$$

Exact local conservation means

$$\partial_t\int F\psi\,dv +\nabla_x\cdot\int vF\psi\,dv=0$$

in distributions, including the prescribed initial traces. These are mass, three momentum components and kinetic energy. They describe balances in each spatial region, not merely a global energy inequality. The energy flux involves a third velocity moment, so its integrability is part of the issue.

3. Development history and previous partial results

DiPerna and Lions established global existence and weak stability for renormalized Boltzmann solutions with finite mass, energy and entropy. Their theory is a foundational existence result, but it does not automatically give uniqueness at that level of regularity. Passing momentum and energy identities through weak limits can also introduce defects, even when local mass conservation survives.

Uniqueness holds in stronger classes. Kaniel and Shinbrot used Maxwellian upper bounds in local theory, while Ukai proved global uniqueness for small periodic perturbations of a Maxwellian in weighted Sobolev spaces. Duan, Huang, Wang and Yang allowed large bounded amplitudes under additional smallness conditions. The unbounded spatial peaks in the present construction place it outside those classes; small relative entropy alone is not enough to invoke their uniqueness theorem.

The spatially homogeneous problem is different. Mischler and Wennberg established uniqueness for conservative hard-sphere evolution with finite mass and energy. Wennberg’s homogeneous nonuniqueness examples for hard potentials permit increasing energy. Family 363 instead uses spatial concentration and, in M1, preserves energy locally.

Recent ill-posedness results also require a distinction. Chen and Holmer, Chen–Shen–Zhang, and Chen–Guo–Shen–Zhang for hard spheres study failures of uniform continuity of solution maps in various rough norms. They compare nearby initial states. Two evolutions from exactly the same datum are a stronger type of failure of well-posedness.

Within this family, M2 develops the concentrating-jet mechanism and obtains nonuniqueness in a global admissible entropy class. M1 retains that mechanism but constructs the global continuation directly, with additional estimates that preserve all local conservation laws. This stronger conservation statement is the principal distinction between the two manuscripts.

4. What the new manuscripts claim

M1: two global solutions with all five local balances

The October manuscript claims one nonnegative datum $F_0$, supported in a bounded velocity ball, with finite mass, energy and absolute entropy, for which there are two distinct global entropy solutions. They differ on a set of positive spacetime–velocity measure, not just on a choice of values on a null set.

Both satisfy the renormalized equation, the full entropy-dissipation inequality and exact local mass, momentum and kinetic-energy conservation with the given initial traces. Both are strongly continuous for all time in $L^1$:

$$F\in C([0,\infty);L^1(\mathbb T^3\times\mathbb R^3)).$$

This means the integral of $|F(t)-F(s)|$ tends to zero as $t\to s$, including at the initial time.

For every finite $T$ and every polynomial weight exponent $k\ge0$, each solution also satisfies

$$\int_0^T\int \langle v\rangle^k \bigl(Q^+(F,F)+Q^-(F,F)\bigr)\,dx\,dv\,dt<\infty,$$

where $\langle v\rangle=\sqrt{1+|v|^2}$. The gains and losses are separately integrable, so the conclusion is not based on canceling two undefined infinite collision terms. This control permits the ordinary distributional equation and the local invariant balances.

M2: the earlier entropy-class construction

M2 claims two global renormalized solutions for one finite-entropy datum with bounded velocity support. They conserve local mass and total momentum, and satisfy global energy and entropy-dissipation inequalities. On a common initial interval they are strongly $L^1$-continuous and have integrable gain and loss terms.

M2 does not require exact local momentum and energy conservation in its admissibility definition. Its initial-interval integrability statement is also narrower than M1’s weighted collision integrability on every finite horizon. These qualifications should not be erased by describing both results simply as “nonuniqueness for Boltzmann.”

The existence statements concern constructed data and solution pairs. They do not assert nonuniqueness for every finite-entropy datum or for a dense class of data.

5. Why the advance matters

Importance: 5/5, conditional on the claims. Nonuniqueness in the large-data entropy theory would be a major boundary on determinism at weak regularity for a central kinetic equation. M1 strengthens that boundary by retaining the local conservation laws built into individual elastic collisions.

The example would show that imposing positivity, entropy dissipation, strong initial traces and exact mass–momentum–energy balances is still insufficient to select one evolution for all rough data. Additional regularity or another selection principle would be needed to recover uniqueness in such a broad class.

This is a statement about the mathematical solution class. It does not claim that the deterministic motion of a specified finite collection of hard spheres has two futures, nor that the well-posed smooth or perturbative Boltzmann theories are invalid.

The global continuation estimate is significant in its own right: it keeps collision products integrable despite an initial density with arbitrarily high spatial amplitudes.

6. The main idea

Concentrate cold jets while keeping total entropy finite

The datum contains narrow velocity distributions near two opposite axial directions. These are called cold jets because their velocity spread is small, not because they have zero kinetic energy.

At a spatial scale $s$, a jet occupies a thin transverse annulus. Its velocity-integrated density can be of order $1/s$, while its total mass is of order $s^2$. The entropy contribution is of order $s^2(1+|\log s|)$. Choosing geometrically decreasing scales makes the total mass and entropy finite despite unbounded amplitudes.

Opposite streams occupy fine alternating spatial stripes. At the original scale, separation suppresses their direct interactions. Under dilation near the accumulating spatial singularity, the stripes average to a medium containing both streams.

The argument assigns auxiliary colors to cold components. Same-color collisions retain their color; interactions between colors or with a designated hot component contribute to that hot component. This is bookkeeping: summing the colored equations gives the original Boltzmann equation, with no extra physical species.

Amplify a seed that disappears from the initial data

One approximation family starts with no extra hot population and remains comparatively dormant. Another starts with a smooth hot seed whose mass, energy and entropy tend to zero as more jet scales are included.

The seed is adjusted until the hot component reaches a prescribed time-dependent weighted bound, or cap. Both approximation families approach the same initial datum. A sufficiently strong gain mechanism can nevertheless make their later limits differ.

A cap hit alone is not enough. Its height might be carried by a set of vanishing mass and disappear in the weak limit. The source therefore analyzes gain kernels to prevent concentration from erasing the event.

Carleman’s representation expresses gain through integrals over velocity planes. Near the velocity axis, two successive gains give the needed nonconcentration. Away from the axis, repeated gains spread the relevant parameters into time, position and velocity. These estimates show that dilation at a cap time produces a nonzero hot limit.

A linear branching limit separates the evolutions

Under the dilation, the cold jets form a two-stream background, while the normalized hot component obeys a linear Boltzmann equation in that background. Strong convergence of suitable velocity averages identifies the cold–hot interaction; weak convergence alone would not justify multiplying the limits.

The linear equation has a branching interpretation in the color bookkeeping: one hot particle colliding with a cold particle produces two hot outputs. Selected finite collision histories yield definite mass growth inside expanding boxes.

If the rescaled observation horizon were infinite, repeated growth would exceed the polynomial bound inherited from the cap. The horizon must therefore remain finite, making the cap scale comparable to the original observation horizon.

The dormant solution has a vanishing normalized limit away from the velocity axis. If the two global solutions coincided at arbitrarily short relevant horizons, comparison of their dilations would force the nonzero hot limit to vanish. This is the proposed separation argument. Branching here concerns the auxiliary labels, not creation of physical particles or energy.

M1 keeps the construction conservative for all time

Periodic space creates a new difficulty: jets eventually encounter their periodic copies. M1 chooses an irrational axis and gives each scale a finite lifetime as a separate cold component. The geometric separation lasts long enough for that schedule.

A surrounding nearly Maxwellian bath supplies a positive collision frequency after a fixed short time. Cold jets then decay, allowing each component to be transferred into a controlled remainder at the end of its lifetime. The decomposition changes, but the total density still solves the same equation.

The global estimate combines small velocity mass, small spatially integrated excess over the Maxwellian, and averaging along free flights. Together these control the nonlinear gain and close a weighted supremum estimate for the remainder, even though no common finite bound holds for all initial cold amplitudes.

The contributions of small scales to weighted collision integrals are summable. This is what lets local momentum and energy conservation, as well as strong time traces, pass to the global limit. M1 therefore does more than attach an arbitrary weak continuation to the early-time nonuniqueness construction.

7. Scope, limitations, and open questions

The equation is the spatially inhomogeneous periodic three-dimensional hard-sphere Boltzmann equation with its stated elastic collision kernel. The theorem is not a homogeneous nonuniqueness result and does not automatically extend to other kernels or the whole-space problem.

The datum has bounded velocity support but unbounded spatial amplitudes. These are not smooth bounded-data counterexamples to classical uniqueness. The construction’s small relative entropy does not put it inside uniqueness classes that also require weighted amplitude control.

M1’s exact local conservation is stronger than global inequalities; M2’s original class does not include all those local identities. Both distinctions remain relevant when comparing to other weak-solution theories.

Questions about additional uniqueness criteria, physical approximation procedures that select one solution, and the range of data exhibiting this behavior remain separate. The two construction mechanisms and their conservation claims are reported as stated, without independent proof verification.

8. Sources and further reading

M1 is the October 5 manuscript with global local-conservation and weighted collision-integrability claims. M2 supplies the earlier concentrating-jet mechanism and a weaker global admissibility class.

Primary background includes DiPerna–Lions on global existence and weak stability, Mischler–Wennberg on conservative homogeneous uniqueness, Duan–Huang–Wang–Yang on large bounded amplitudes, and the hard-sphere instability result. They address different solution classes or different meanings of ill-posedness.

Source manuscripts

• M1. OpenAI, Nonuniqueness with local conservation for the hard-sphere Boltzmann equation (October 5, 2026).
• M2. OpenAI, Nonuniqueness for the periodic hard-sphere Boltzmann equation (September 23, 2026).

Sources consulted on 26 October 2026 at repository revision 1581d4c. Download the manuscript citations.